xref: /titanic_41/usr/src/uts/common/dtrace/dtrace.c (revision 3dd66341647ff4c52307fd8b0cf609bbe9961b4a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2015, Joyent, Inc. All rights reserved.
25  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
26  */
27 
28 /*
29  * DTrace - Dynamic Tracing for Solaris
30  *
31  * This is the implementation of the Solaris Dynamic Tracing framework
32  * (DTrace).  The user-visible interface to DTrace is described at length in
33  * the "Solaris Dynamic Tracing Guide".  The interfaces between the libdtrace
34  * library, the in-kernel DTrace framework, and the DTrace providers are
35  * described in the block comments in the <sys/dtrace.h> header file.  The
36  * internal architecture of DTrace is described in the block comments in the
37  * <sys/dtrace_impl.h> header file.  The comments contained within the DTrace
38  * implementation very much assume mastery of all of these sources; if one has
39  * an unanswered question about the implementation, one should consult them
40  * first.
41  *
42  * The functions here are ordered roughly as follows:
43  *
44  *   - Probe context functions
45  *   - Probe hashing functions
46  *   - Non-probe context utility functions
47  *   - Matching functions
48  *   - Provider-to-Framework API functions
49  *   - Probe management functions
50  *   - DIF object functions
51  *   - Format functions
52  *   - Predicate functions
53  *   - ECB functions
54  *   - Buffer functions
55  *   - Enabling functions
56  *   - DOF functions
57  *   - Anonymous enabling functions
58  *   - Consumer state functions
59  *   - Helper functions
60  *   - Hook functions
61  *   - Driver cookbook functions
62  *
63  * Each group of functions begins with a block comment labelled the "DTrace
64  * [Group] Functions", allowing one to find each block by searching forward
65  * on capital-f functions.
66  */
67 #include <sys/errno.h>
68 #include <sys/stat.h>
69 #include <sys/modctl.h>
70 #include <sys/conf.h>
71 #include <sys/systm.h>
72 #include <sys/ddi.h>
73 #include <sys/sunddi.h>
74 #include <sys/cpuvar.h>
75 #include <sys/kmem.h>
76 #include <sys/strsubr.h>
77 #include <sys/sysmacros.h>
78 #include <sys/dtrace_impl.h>
79 #include <sys/atomic.h>
80 #include <sys/cmn_err.h>
81 #include <sys/mutex_impl.h>
82 #include <sys/rwlock_impl.h>
83 #include <sys/ctf_api.h>
84 #include <sys/panic.h>
85 #include <sys/priv_impl.h>
86 #include <sys/policy.h>
87 #include <sys/cred_impl.h>
88 #include <sys/procfs_isa.h>
89 #include <sys/taskq.h>
90 #include <sys/mkdev.h>
91 #include <sys/kdi.h>
92 #include <sys/zone.h>
93 #include <sys/socket.h>
94 #include <netinet/in.h>
95 #include "strtolctype.h"
96 
97 /*
98  * DTrace Tunable Variables
99  *
100  * The following variables may be tuned by adding a line to /etc/system that
101  * includes both the name of the DTrace module ("dtrace") and the name of the
102  * variable.  For example:
103  *
104  *   set dtrace:dtrace_destructive_disallow = 1
105  *
106  * In general, the only variables that one should be tuning this way are those
107  * that affect system-wide DTrace behavior, and for which the default behavior
108  * is undesirable.  Most of these variables are tunable on a per-consumer
109  * basis using DTrace options, and need not be tuned on a system-wide basis.
110  * When tuning these variables, avoid pathological values; while some attempt
111  * is made to verify the integrity of these variables, they are not considered
112  * part of the supported interface to DTrace, and they are therefore not
113  * checked comprehensively.  Further, these variables should not be tuned
114  * dynamically via "mdb -kw" or other means; they should only be tuned via
115  * /etc/system.
116  */
117 int		dtrace_destructive_disallow = 0;
118 dtrace_optval_t	dtrace_nonroot_maxsize = (16 * 1024 * 1024);
119 size_t		dtrace_difo_maxsize = (256 * 1024);
120 dtrace_optval_t	dtrace_dof_maxsize = (8 * 1024 * 1024);
121 size_t		dtrace_global_maxsize = (16 * 1024);
122 size_t		dtrace_actions_max = (16 * 1024);
123 size_t		dtrace_retain_max = 1024;
124 dtrace_optval_t	dtrace_helper_actions_max = 1024;
125 dtrace_optval_t	dtrace_helper_providers_max = 32;
126 dtrace_optval_t	dtrace_dstate_defsize = (1 * 1024 * 1024);
127 size_t		dtrace_strsize_default = 256;
128 dtrace_optval_t	dtrace_cleanrate_default = 9900990;		/* 101 hz */
129 dtrace_optval_t	dtrace_cleanrate_min = 200000;			/* 5000 hz */
130 dtrace_optval_t	dtrace_cleanrate_max = (uint64_t)60 * NANOSEC;	/* 1/minute */
131 dtrace_optval_t	dtrace_aggrate_default = NANOSEC;		/* 1 hz */
132 dtrace_optval_t	dtrace_statusrate_default = NANOSEC;		/* 1 hz */
133 dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC;	 /* 6/minute */
134 dtrace_optval_t	dtrace_switchrate_default = NANOSEC;		/* 1 hz */
135 dtrace_optval_t	dtrace_nspec_default = 1;
136 dtrace_optval_t	dtrace_specsize_default = 32 * 1024;
137 dtrace_optval_t dtrace_stackframes_default = 20;
138 dtrace_optval_t dtrace_ustackframes_default = 20;
139 dtrace_optval_t dtrace_jstackframes_default = 50;
140 dtrace_optval_t dtrace_jstackstrsize_default = 512;
141 int		dtrace_msgdsize_max = 128;
142 hrtime_t	dtrace_chill_max = MSEC2NSEC(500);		/* 500 ms */
143 hrtime_t	dtrace_chill_interval = NANOSEC;		/* 1000 ms */
144 int		dtrace_devdepth_max = 32;
145 int		dtrace_err_verbose;
146 hrtime_t	dtrace_deadman_interval = NANOSEC;
147 hrtime_t	dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC;
148 hrtime_t	dtrace_deadman_user = (hrtime_t)30 * NANOSEC;
149 hrtime_t	dtrace_unregister_defunct_reap = (hrtime_t)60 * NANOSEC;
150 
151 /*
152  * DTrace External Variables
153  *
154  * As dtrace(7D) is a kernel module, any DTrace variables are obviously
155  * available to DTrace consumers via the backtick (`) syntax.  One of these,
156  * dtrace_zero, is made deliberately so:  it is provided as a source of
157  * well-known, zero-filled memory.  While this variable is not documented,
158  * it is used by some translators as an implementation detail.
159  */
160 const char	dtrace_zero[256] = { 0 };	/* zero-filled memory */
161 
162 /*
163  * DTrace Internal Variables
164  */
165 static dev_info_t	*dtrace_devi;		/* device info */
166 static vmem_t		*dtrace_arena;		/* probe ID arena */
167 static vmem_t		*dtrace_minor;		/* minor number arena */
168 static taskq_t		*dtrace_taskq;		/* task queue */
169 static dtrace_probe_t	**dtrace_probes;	/* array of all probes */
170 static int		dtrace_nprobes;		/* number of probes */
171 static dtrace_provider_t *dtrace_provider;	/* provider list */
172 static dtrace_meta_t	*dtrace_meta_pid;	/* user-land meta provider */
173 static int		dtrace_opens;		/* number of opens */
174 static int		dtrace_helpers;		/* number of helpers */
175 static int		dtrace_getf;		/* number of unpriv getf()s */
176 static void		*dtrace_softstate;	/* softstate pointer */
177 static dtrace_hash_t	*dtrace_bymod;		/* probes hashed by module */
178 static dtrace_hash_t	*dtrace_byfunc;		/* probes hashed by function */
179 static dtrace_hash_t	*dtrace_byname;		/* probes hashed by name */
180 static dtrace_toxrange_t *dtrace_toxrange;	/* toxic range array */
181 static int		dtrace_toxranges;	/* number of toxic ranges */
182 static int		dtrace_toxranges_max;	/* size of toxic range array */
183 static dtrace_anon_t	dtrace_anon;		/* anonymous enabling */
184 static kmem_cache_t	*dtrace_state_cache;	/* cache for dynamic state */
185 static uint64_t		dtrace_vtime_references; /* number of vtimestamp refs */
186 static kthread_t	*dtrace_panicked;	/* panicking thread */
187 static dtrace_ecb_t	*dtrace_ecb_create_cache; /* cached created ECB */
188 static dtrace_genid_t	dtrace_probegen;	/* current probe generation */
189 static dtrace_helpers_t *dtrace_deferred_pid;	/* deferred helper list */
190 static dtrace_enabling_t *dtrace_retained;	/* list of retained enablings */
191 static dtrace_genid_t	dtrace_retained_gen;	/* current retained enab gen */
192 static dtrace_dynvar_t	dtrace_dynhash_sink;	/* end of dynamic hash chains */
193 static int		dtrace_dynvar_failclean; /* dynvars failed to clean */
194 
195 /*
196  * DTrace Locking
197  * DTrace is protected by three (relatively coarse-grained) locks:
198  *
199  * (1) dtrace_lock is required to manipulate essentially any DTrace state,
200  *     including enabling state, probes, ECBs, consumer state, helper state,
201  *     etc.  Importantly, dtrace_lock is _not_ required when in probe context;
202  *     probe context is lock-free -- synchronization is handled via the
203  *     dtrace_sync() cross call mechanism.
204  *
205  * (2) dtrace_provider_lock is required when manipulating provider state, or
206  *     when provider state must be held constant.
207  *
208  * (3) dtrace_meta_lock is required when manipulating meta provider state, or
209  *     when meta provider state must be held constant.
210  *
211  * The lock ordering between these three locks is dtrace_meta_lock before
212  * dtrace_provider_lock before dtrace_lock.  (In particular, there are
213  * several places where dtrace_provider_lock is held by the framework as it
214  * calls into the providers -- which then call back into the framework,
215  * grabbing dtrace_lock.)
216  *
217  * There are two other locks in the mix:  mod_lock and cpu_lock.  With respect
218  * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical
219  * role as a coarse-grained lock; it is acquired before both of these locks.
220  * With respect to dtrace_meta_lock, its behavior is stranger:  cpu_lock must
221  * be acquired _between_ dtrace_meta_lock and any other DTrace locks.
222  * mod_lock is similar with respect to dtrace_provider_lock in that it must be
223  * acquired _between_ dtrace_provider_lock and dtrace_lock.
224  */
225 static kmutex_t		dtrace_lock;		/* probe state lock */
226 static kmutex_t		dtrace_provider_lock;	/* provider state lock */
227 static kmutex_t		dtrace_meta_lock;	/* meta-provider state lock */
228 
229 /*
230  * DTrace Provider Variables
231  *
232  * These are the variables relating to DTrace as a provider (that is, the
233  * provider of the BEGIN, END, and ERROR probes).
234  */
235 static dtrace_pattr_t	dtrace_provider_attr = {
236 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
237 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
238 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
239 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
240 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
241 };
242 
243 static void
dtrace_nullop(void)244 dtrace_nullop(void)
245 {}
246 
247 static int
dtrace_enable_nullop(void)248 dtrace_enable_nullop(void)
249 {
250 	return (0);
251 }
252 
253 static dtrace_pops_t	dtrace_provider_ops = {
254 	(void (*)(void *, const dtrace_probedesc_t *))dtrace_nullop,
255 	(void (*)(void *, struct modctl *))dtrace_nullop,
256 	(int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop,
257 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
258 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
259 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
260 	NULL,
261 	NULL,
262 	NULL,
263 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop
264 };
265 
266 static dtrace_id_t	dtrace_probeid_begin;	/* special BEGIN probe */
267 static dtrace_id_t	dtrace_probeid_end;	/* special END probe */
268 dtrace_id_t		dtrace_probeid_error;	/* special ERROR probe */
269 
270 /*
271  * DTrace Helper Tracing Variables
272  *
273  * These variables should be set dynamically to enable helper tracing.  The
274  * only variables that should be set are dtrace_helptrace_enable (which should
275  * be set to a non-zero value to allocate helper tracing buffers on the next
276  * open of /dev/dtrace) and dtrace_helptrace_disable (which should be set to a
277  * non-zero value to deallocate helper tracing buffers on the next close of
278  * /dev/dtrace).  When (and only when) helper tracing is disabled, the
279  * buffer size may also be set via dtrace_helptrace_bufsize.
280  */
281 int			dtrace_helptrace_enable = 0;
282 int			dtrace_helptrace_disable = 0;
283 int			dtrace_helptrace_bufsize = 16 * 1024 * 1024;
284 uint32_t		dtrace_helptrace_nlocals;
285 static dtrace_helptrace_t *dtrace_helptrace_buffer;
286 static uint32_t		dtrace_helptrace_next = 0;
287 static int		dtrace_helptrace_wrapped = 0;
288 
289 /*
290  * DTrace Error Hashing
291  *
292  * On DEBUG kernels, DTrace will track the errors that has seen in a hash
293  * table.  This is very useful for checking coverage of tests that are
294  * expected to induce DIF or DOF processing errors, and may be useful for
295  * debugging problems in the DIF code generator or in DOF generation .  The
296  * error hash may be examined with the ::dtrace_errhash MDB dcmd.
297  */
298 #ifdef DEBUG
299 static dtrace_errhash_t	dtrace_errhash[DTRACE_ERRHASHSZ];
300 static const char *dtrace_errlast;
301 static kthread_t *dtrace_errthread;
302 static kmutex_t dtrace_errlock;
303 #endif
304 
305 /*
306  * DTrace Macros and Constants
307  *
308  * These are various macros that are useful in various spots in the
309  * implementation, along with a few random constants that have no meaning
310  * outside of the implementation.  There is no real structure to this cpp
311  * mishmash -- but is there ever?
312  */
313 #define	DTRACE_HASHSTR(hash, probe)	\
314 	dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs)))
315 
316 #define	DTRACE_HASHNEXT(hash, probe)	\
317 	(dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs)
318 
319 #define	DTRACE_HASHPREV(hash, probe)	\
320 	(dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs)
321 
322 #define	DTRACE_HASHEQ(hash, lhs, rhs)	\
323 	(strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \
324 	    *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0)
325 
326 #define	DTRACE_AGGHASHSIZE_SLEW		17
327 
328 #define	DTRACE_V4MAPPED_OFFSET		(sizeof (uint32_t) * 3)
329 
330 /*
331  * The key for a thread-local variable consists of the lower 61 bits of the
332  * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL.
333  * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never
334  * equal to a variable identifier.  This is necessary (but not sufficient) to
335  * assure that global associative arrays never collide with thread-local
336  * variables.  To guarantee that they cannot collide, we must also define the
337  * order for keying dynamic variables.  That order is:
338  *
339  *   [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ]
340  *
341  * Because the variable-key and the tls-key are in orthogonal spaces, there is
342  * no way for a global variable key signature to match a thread-local key
343  * signature.
344  */
345 #define	DTRACE_TLS_THRKEY(where) { \
346 	uint_t intr = 0; \
347 	uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \
348 	for (; actv; actv >>= 1) \
349 		intr++; \
350 	ASSERT(intr < (1 << 3)); \
351 	(where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \
352 	    (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
353 }
354 
355 #define	DT_BSWAP_8(x)	((x) & 0xff)
356 #define	DT_BSWAP_16(x)	((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8))
357 #define	DT_BSWAP_32(x)	((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16))
358 #define	DT_BSWAP_64(x)	((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32))
359 
360 #define	DT_MASK_LO 0x00000000FFFFFFFFULL
361 
362 #define	DTRACE_STORE(type, tomax, offset, what) \
363 	*((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what);
364 
365 #ifndef __x86
366 #define	DTRACE_ALIGNCHECK(addr, size, flags)				\
367 	if (addr & (size - 1)) {					\
368 		*flags |= CPU_DTRACE_BADALIGN;				\
369 		cpu_core[CPU->cpu_id].cpuc_dtrace_illval = addr;	\
370 		return (0);						\
371 	}
372 #else
373 #define	DTRACE_ALIGNCHECK(addr, size, flags)
374 #endif
375 
376 /*
377  * Test whether a range of memory starting at testaddr of size testsz falls
378  * within the range of memory described by addr, sz.  We take care to avoid
379  * problems with overflow and underflow of the unsigned quantities, and
380  * disallow all negative sizes.  Ranges of size 0 are allowed.
381  */
382 #define	DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \
383 	((testaddr) - (uintptr_t)(baseaddr) < (basesz) && \
384 	(testaddr) + (testsz) - (uintptr_t)(baseaddr) <= (basesz) && \
385 	(testaddr) + (testsz) >= (testaddr))
386 
387 /*
388  * Test whether alloc_sz bytes will fit in the scratch region.  We isolate
389  * alloc_sz on the righthand side of the comparison in order to avoid overflow
390  * or underflow in the comparison with it.  This is simpler than the INRANGE
391  * check above, because we know that the dtms_scratch_ptr is valid in the
392  * range.  Allocations of size zero are allowed.
393  */
394 #define	DTRACE_INSCRATCH(mstate, alloc_sz) \
395 	((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \
396 	(mstate)->dtms_scratch_ptr >= (alloc_sz))
397 
398 #define	DTRACE_LOADFUNC(bits)						\
399 /*CSTYLED*/								\
400 uint##bits##_t								\
401 dtrace_load##bits(uintptr_t addr)					\
402 {									\
403 	size_t size = bits / NBBY;					\
404 	/*CSTYLED*/							\
405 	uint##bits##_t rval;						\
406 	int i;								\
407 	volatile uint16_t *flags = (volatile uint16_t *)		\
408 	    &cpu_core[CPU->cpu_id].cpuc_dtrace_flags;			\
409 									\
410 	DTRACE_ALIGNCHECK(addr, size, flags);				\
411 									\
412 	for (i = 0; i < dtrace_toxranges; i++) {			\
413 		if (addr >= dtrace_toxrange[i].dtt_limit)		\
414 			continue;					\
415 									\
416 		if (addr + size <= dtrace_toxrange[i].dtt_base)		\
417 			continue;					\
418 									\
419 		/*							\
420 		 * This address falls within a toxic region; return 0.	\
421 		 */							\
422 		*flags |= CPU_DTRACE_BADADDR;				\
423 		cpu_core[CPU->cpu_id].cpuc_dtrace_illval = addr;	\
424 		return (0);						\
425 	}								\
426 									\
427 	*flags |= CPU_DTRACE_NOFAULT;					\
428 	/*CSTYLED*/							\
429 	rval = *((volatile uint##bits##_t *)addr);			\
430 	*flags &= ~CPU_DTRACE_NOFAULT;					\
431 									\
432 	return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0);		\
433 }
434 
435 #ifdef _LP64
436 #define	dtrace_loadptr	dtrace_load64
437 #else
438 #define	dtrace_loadptr	dtrace_load32
439 #endif
440 
441 #define	DTRACE_DYNHASH_FREE	0
442 #define	DTRACE_DYNHASH_SINK	1
443 #define	DTRACE_DYNHASH_VALID	2
444 
445 #define	DTRACE_MATCH_FAIL	-1
446 #define	DTRACE_MATCH_NEXT	0
447 #define	DTRACE_MATCH_DONE	1
448 #define	DTRACE_ANCHORED(probe)	((probe)->dtpr_func[0] != '\0')
449 #define	DTRACE_STATE_ALIGN	64
450 
451 #define	DTRACE_FLAGS2FLT(flags)						\
452 	(((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR :		\
453 	((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP :		\
454 	((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO :		\
455 	((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV :		\
456 	((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV :		\
457 	((flags) & CPU_DTRACE_TUPOFLOW) ?  DTRACEFLT_TUPOFLOW :		\
458 	((flags) & CPU_DTRACE_BADALIGN) ?  DTRACEFLT_BADALIGN :		\
459 	((flags) & CPU_DTRACE_NOSCRATCH) ?  DTRACEFLT_NOSCRATCH :	\
460 	((flags) & CPU_DTRACE_BADSTACK) ?  DTRACEFLT_BADSTACK :		\
461 	DTRACEFLT_UNKNOWN)
462 
463 #define	DTRACEACT_ISSTRING(act)						\
464 	((act)->dta_kind == DTRACEACT_DIFEXPR &&			\
465 	(act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING)
466 
467 static size_t dtrace_strlen(const char *, size_t);
468 static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id);
469 static void dtrace_enabling_provide(dtrace_provider_t *);
470 static int dtrace_enabling_match(dtrace_enabling_t *, int *);
471 static void dtrace_enabling_matchall(void);
472 static void dtrace_enabling_reap(void);
473 static dtrace_state_t *dtrace_anon_grab(void);
474 static uint64_t dtrace_helper(int, dtrace_mstate_t *,
475     dtrace_state_t *, uint64_t, uint64_t);
476 static dtrace_helpers_t *dtrace_helpers_create(proc_t *);
477 static void dtrace_buffer_drop(dtrace_buffer_t *);
478 static int dtrace_buffer_consumed(dtrace_buffer_t *, hrtime_t when);
479 static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t,
480     dtrace_state_t *, dtrace_mstate_t *);
481 static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t,
482     dtrace_optval_t);
483 static int dtrace_ecb_create_enable(dtrace_probe_t *, void *);
484 static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *);
485 static int dtrace_priv_proc(dtrace_state_t *, dtrace_mstate_t *);
486 static void dtrace_getf_barrier(void);
487 
488 /*
489  * DTrace Probe Context Functions
490  *
491  * These functions are called from probe context.  Because probe context is
492  * any context in which C may be called, arbitrarily locks may be held,
493  * interrupts may be disabled, we may be in arbitrary dispatched state, etc.
494  * As a result, functions called from probe context may only call other DTrace
495  * support functions -- they may not interact at all with the system at large.
496  * (Note that the ASSERT macro is made probe-context safe by redefining it in
497  * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary
498  * loads are to be performed from probe context, they _must_ be in terms of
499  * the safe dtrace_load*() variants.
500  *
501  * Some functions in this block are not actually called from probe context;
502  * for these functions, there will be a comment above the function reading
503  * "Note:  not called from probe context."
504  */
505 void
dtrace_panic(const char * format,...)506 dtrace_panic(const char *format, ...)
507 {
508 	va_list alist;
509 
510 	va_start(alist, format);
511 	dtrace_vpanic(format, alist);
512 	va_end(alist);
513 }
514 
515 int
dtrace_assfail(const char * a,const char * f,int l)516 dtrace_assfail(const char *a, const char *f, int l)
517 {
518 	dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l);
519 
520 	/*
521 	 * We just need something here that even the most clever compiler
522 	 * cannot optimize away.
523 	 */
524 	return (a[(uintptr_t)f]);
525 }
526 
527 /*
528  * Atomically increment a specified error counter from probe context.
529  */
530 static void
dtrace_error(uint32_t * counter)531 dtrace_error(uint32_t *counter)
532 {
533 	/*
534 	 * Most counters stored to in probe context are per-CPU counters.
535 	 * However, there are some error conditions that are sufficiently
536 	 * arcane that they don't merit per-CPU storage.  If these counters
537 	 * are incremented concurrently on different CPUs, scalability will be
538 	 * adversely affected -- but we don't expect them to be white-hot in a
539 	 * correctly constructed enabling...
540 	 */
541 	uint32_t oval, nval;
542 
543 	do {
544 		oval = *counter;
545 
546 		if ((nval = oval + 1) == 0) {
547 			/*
548 			 * If the counter would wrap, set it to 1 -- assuring
549 			 * that the counter is never zero when we have seen
550 			 * errors.  (The counter must be 32-bits because we
551 			 * aren't guaranteed a 64-bit compare&swap operation.)
552 			 * To save this code both the infamy of being fingered
553 			 * by a priggish news story and the indignity of being
554 			 * the target of a neo-puritan witch trial, we're
555 			 * carefully avoiding any colorful description of the
556 			 * likelihood of this condition -- but suffice it to
557 			 * say that it is only slightly more likely than the
558 			 * overflow of predicate cache IDs, as discussed in
559 			 * dtrace_predicate_create().
560 			 */
561 			nval = 1;
562 		}
563 	} while (dtrace_cas32(counter, oval, nval) != oval);
564 }
565 
566 /*
567  * Use the DTRACE_LOADFUNC macro to define functions for each of loading a
568  * uint8_t, a uint16_t, a uint32_t and a uint64_t.
569  */
570 DTRACE_LOADFUNC(8)
571 DTRACE_LOADFUNC(16)
572 DTRACE_LOADFUNC(32)
573 DTRACE_LOADFUNC(64)
574 
575 static int
dtrace_inscratch(uintptr_t dest,size_t size,dtrace_mstate_t * mstate)576 dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate)
577 {
578 	if (dest < mstate->dtms_scratch_base)
579 		return (0);
580 
581 	if (dest + size < dest)
582 		return (0);
583 
584 	if (dest + size > mstate->dtms_scratch_ptr)
585 		return (0);
586 
587 	return (1);
588 }
589 
590 static int
dtrace_canstore_statvar(uint64_t addr,size_t sz,dtrace_statvar_t ** svars,int nsvars)591 dtrace_canstore_statvar(uint64_t addr, size_t sz,
592     dtrace_statvar_t **svars, int nsvars)
593 {
594 	int i;
595 
596 	for (i = 0; i < nsvars; i++) {
597 		dtrace_statvar_t *svar = svars[i];
598 
599 		if (svar == NULL || svar->dtsv_size == 0)
600 			continue;
601 
602 		if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size))
603 			return (1);
604 	}
605 
606 	return (0);
607 }
608 
609 /*
610  * Check to see if the address is within a memory region to which a store may
611  * be issued.  This includes the DTrace scratch areas, and any DTrace variable
612  * region.  The caller of dtrace_canstore() is responsible for performing any
613  * alignment checks that are needed before stores are actually executed.
614  */
615 static int
dtrace_canstore(uint64_t addr,size_t sz,dtrace_mstate_t * mstate,dtrace_vstate_t * vstate)616 dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
617     dtrace_vstate_t *vstate)
618 {
619 	/*
620 	 * First, check to see if the address is in scratch space...
621 	 */
622 	if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base,
623 	    mstate->dtms_scratch_size))
624 		return (1);
625 
626 	/*
627 	 * Now check to see if it's a dynamic variable.  This check will pick
628 	 * up both thread-local variables and any global dynamically-allocated
629 	 * variables.
630 	 */
631 	if (DTRACE_INRANGE(addr, sz, vstate->dtvs_dynvars.dtds_base,
632 	    vstate->dtvs_dynvars.dtds_size)) {
633 		dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
634 		uintptr_t base = (uintptr_t)dstate->dtds_base +
635 		    (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t));
636 		uintptr_t chunkoffs;
637 
638 		/*
639 		 * Before we assume that we can store here, we need to make
640 		 * sure that it isn't in our metadata -- storing to our
641 		 * dynamic variable metadata would corrupt our state.  For
642 		 * the range to not include any dynamic variable metadata,
643 		 * it must:
644 		 *
645 		 *	(1) Start above the hash table that is at the base of
646 		 *	the dynamic variable space
647 		 *
648 		 *	(2) Have a starting chunk offset that is beyond the
649 		 *	dtrace_dynvar_t that is at the base of every chunk
650 		 *
651 		 *	(3) Not span a chunk boundary
652 		 *
653 		 */
654 		if (addr < base)
655 			return (0);
656 
657 		chunkoffs = (addr - base) % dstate->dtds_chunksize;
658 
659 		if (chunkoffs < sizeof (dtrace_dynvar_t))
660 			return (0);
661 
662 		if (chunkoffs + sz > dstate->dtds_chunksize)
663 			return (0);
664 
665 		return (1);
666 	}
667 
668 	/*
669 	 * Finally, check the static local and global variables.  These checks
670 	 * take the longest, so we perform them last.
671 	 */
672 	if (dtrace_canstore_statvar(addr, sz,
673 	    vstate->dtvs_locals, vstate->dtvs_nlocals))
674 		return (1);
675 
676 	if (dtrace_canstore_statvar(addr, sz,
677 	    vstate->dtvs_globals, vstate->dtvs_nglobals))
678 		return (1);
679 
680 	return (0);
681 }
682 
683 
684 /*
685  * Convenience routine to check to see if the address is within a memory
686  * region in which a load may be issued given the user's privilege level;
687  * if not, it sets the appropriate error flags and loads 'addr' into the
688  * illegal value slot.
689  *
690  * DTrace subroutines (DIF_SUBR_*) should use this helper to implement
691  * appropriate memory access protection.
692  */
693 static int
dtrace_canload(uint64_t addr,size_t sz,dtrace_mstate_t * mstate,dtrace_vstate_t * vstate)694 dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
695     dtrace_vstate_t *vstate)
696 {
697 	volatile uintptr_t *illval = &cpu_core[CPU->cpu_id].cpuc_dtrace_illval;
698 	file_t *fp;
699 
700 	/*
701 	 * If we hold the privilege to read from kernel memory, then
702 	 * everything is readable.
703 	 */
704 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
705 		return (1);
706 
707 	/*
708 	 * You can obviously read that which you can store.
709 	 */
710 	if (dtrace_canstore(addr, sz, mstate, vstate))
711 		return (1);
712 
713 	/*
714 	 * We're allowed to read from our own string table.
715 	 */
716 	if (DTRACE_INRANGE(addr, sz, mstate->dtms_difo->dtdo_strtab,
717 	    mstate->dtms_difo->dtdo_strlen))
718 		return (1);
719 
720 	if (vstate->dtvs_state != NULL &&
721 	    dtrace_priv_proc(vstate->dtvs_state, mstate)) {
722 		proc_t *p;
723 
724 		/*
725 		 * When we have privileges to the current process, there are
726 		 * several context-related kernel structures that are safe to
727 		 * read, even absent the privilege to read from kernel memory.
728 		 * These reads are safe because these structures contain only
729 		 * state that (1) we're permitted to read, (2) is harmless or
730 		 * (3) contains pointers to additional kernel state that we're
731 		 * not permitted to read (and as such, do not present an
732 		 * opportunity for privilege escalation).  Finally (and
733 		 * critically), because of the nature of their relation with
734 		 * the current thread context, the memory associated with these
735 		 * structures cannot change over the duration of probe context,
736 		 * and it is therefore impossible for this memory to be
737 		 * deallocated and reallocated as something else while it's
738 		 * being operated upon.
739 		 */
740 		if (DTRACE_INRANGE(addr, sz, curthread, sizeof (kthread_t)))
741 			return (1);
742 
743 		if ((p = curthread->t_procp) != NULL && DTRACE_INRANGE(addr,
744 		    sz, curthread->t_procp, sizeof (proc_t))) {
745 			return (1);
746 		}
747 
748 		if (curthread->t_cred != NULL && DTRACE_INRANGE(addr, sz,
749 		    curthread->t_cred, sizeof (cred_t))) {
750 			return (1);
751 		}
752 
753 		if (p != NULL && p->p_pidp != NULL && DTRACE_INRANGE(addr, sz,
754 		    &(p->p_pidp->pid_id), sizeof (pid_t))) {
755 			return (1);
756 		}
757 
758 		if (curthread->t_cpu != NULL && DTRACE_INRANGE(addr, sz,
759 		    curthread->t_cpu, offsetof(cpu_t, cpu_pause_thread))) {
760 			return (1);
761 		}
762 	}
763 
764 	if ((fp = mstate->dtms_getf) != NULL) {
765 		uintptr_t psz = sizeof (void *);
766 		vnode_t *vp;
767 		vnodeops_t *op;
768 
769 		/*
770 		 * When getf() returns a file_t, the enabling is implicitly
771 		 * granted the (transient) right to read the returned file_t
772 		 * as well as the v_path and v_op->vnop_name of the underlying
773 		 * vnode.  These accesses are allowed after a successful
774 		 * getf() because the members that they refer to cannot change
775 		 * once set -- and the barrier logic in the kernel's closef()
776 		 * path assures that the file_t and its referenced vode_t
777 		 * cannot themselves be stale (that is, it impossible for
778 		 * either dtms_getf itself or its f_vnode member to reference
779 		 * freed memory).
780 		 */
781 		if (DTRACE_INRANGE(addr, sz, fp, sizeof (file_t)))
782 			return (1);
783 
784 		if ((vp = fp->f_vnode) != NULL) {
785 			if (DTRACE_INRANGE(addr, sz, &vp->v_path, psz))
786 				return (1);
787 
788 			if (vp->v_path != NULL && DTRACE_INRANGE(addr, sz,
789 			    vp->v_path, strlen(vp->v_path) + 1)) {
790 				return (1);
791 			}
792 
793 			if (DTRACE_INRANGE(addr, sz, &vp->v_op, psz))
794 				return (1);
795 
796 			if ((op = vp->v_op) != NULL &&
797 			    DTRACE_INRANGE(addr, sz, &op->vnop_name, psz)) {
798 				return (1);
799 			}
800 
801 			if (op != NULL && op->vnop_name != NULL &&
802 			    DTRACE_INRANGE(addr, sz, op->vnop_name,
803 			    strlen(op->vnop_name) + 1)) {
804 				return (1);
805 			}
806 		}
807 	}
808 
809 	DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV);
810 	*illval = addr;
811 	return (0);
812 }
813 
814 /*
815  * Convenience routine to check to see if a given string is within a memory
816  * region in which a load may be issued given the user's privilege level;
817  * this exists so that we don't need to issue unnecessary dtrace_strlen()
818  * calls in the event that the user has all privileges.
819  */
820 static int
dtrace_strcanload(uint64_t addr,size_t sz,dtrace_mstate_t * mstate,dtrace_vstate_t * vstate)821 dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
822     dtrace_vstate_t *vstate)
823 {
824 	size_t strsz;
825 
826 	/*
827 	 * If we hold the privilege to read from kernel memory, then
828 	 * everything is readable.
829 	 */
830 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
831 		return (1);
832 
833 	strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz);
834 	if (dtrace_canload(addr, strsz, mstate, vstate))
835 		return (1);
836 
837 	return (0);
838 }
839 
840 /*
841  * Convenience routine to check to see if a given variable is within a memory
842  * region in which a load may be issued given the user's privilege level.
843  */
844 static int
dtrace_vcanload(void * src,dtrace_diftype_t * type,dtrace_mstate_t * mstate,dtrace_vstate_t * vstate)845 dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate,
846     dtrace_vstate_t *vstate)
847 {
848 	size_t sz;
849 	ASSERT(type->dtdt_flags & DIF_TF_BYREF);
850 
851 	/*
852 	 * If we hold the privilege to read from kernel memory, then
853 	 * everything is readable.
854 	 */
855 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
856 		return (1);
857 
858 	if (type->dtdt_kind == DIF_TYPE_STRING)
859 		sz = dtrace_strlen(src,
860 		    vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1;
861 	else
862 		sz = type->dtdt_size;
863 
864 	return (dtrace_canload((uintptr_t)src, sz, mstate, vstate));
865 }
866 
867 /*
868  * Convert a string to a signed integer using safe loads.
869  *
870  * NOTE: This function uses various macros from strtolctype.h to manipulate
871  * digit values, etc -- these have all been checked to ensure they make
872  * no additional function calls.
873  */
874 static int64_t
dtrace_strtoll(char * input,int base,size_t limit)875 dtrace_strtoll(char *input, int base, size_t limit)
876 {
877 	uintptr_t pos = (uintptr_t)input;
878 	int64_t val = 0;
879 	int x;
880 	boolean_t neg = B_FALSE;
881 	char c, cc, ccc;
882 	uintptr_t end = pos + limit;
883 
884 	/*
885 	 * Consume any whitespace preceding digits.
886 	 */
887 	while ((c = dtrace_load8(pos)) == ' ' || c == '\t')
888 		pos++;
889 
890 	/*
891 	 * Handle an explicit sign if one is present.
892 	 */
893 	if (c == '-' || c == '+') {
894 		if (c == '-')
895 			neg = B_TRUE;
896 		c = dtrace_load8(++pos);
897 	}
898 
899 	/*
900 	 * Check for an explicit hexadecimal prefix ("0x" or "0X") and skip it
901 	 * if present.
902 	 */
903 	if (base == 16 && c == '0' && ((cc = dtrace_load8(pos + 1)) == 'x' ||
904 	    cc == 'X') && isxdigit(ccc = dtrace_load8(pos + 2))) {
905 		pos += 2;
906 		c = ccc;
907 	}
908 
909 	/*
910 	 * Read in contiguous digits until the first non-digit character.
911 	 */
912 	for (; pos < end && c != '\0' && lisalnum(c) && (x = DIGIT(c)) < base;
913 	    c = dtrace_load8(++pos))
914 		val = val * base + x;
915 
916 	return (neg ? -val : val);
917 }
918 
919 /*
920  * Compare two strings using safe loads.
921  */
922 static int
dtrace_strncmp(char * s1,char * s2,size_t limit)923 dtrace_strncmp(char *s1, char *s2, size_t limit)
924 {
925 	uint8_t c1, c2;
926 	volatile uint16_t *flags;
927 
928 	if (s1 == s2 || limit == 0)
929 		return (0);
930 
931 	flags = (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
932 
933 	do {
934 		if (s1 == NULL) {
935 			c1 = '\0';
936 		} else {
937 			c1 = dtrace_load8((uintptr_t)s1++);
938 		}
939 
940 		if (s2 == NULL) {
941 			c2 = '\0';
942 		} else {
943 			c2 = dtrace_load8((uintptr_t)s2++);
944 		}
945 
946 		if (c1 != c2)
947 			return (c1 - c2);
948 	} while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT));
949 
950 	return (0);
951 }
952 
953 /*
954  * Compute strlen(s) for a string using safe memory accesses.  The additional
955  * len parameter is used to specify a maximum length to ensure completion.
956  */
957 static size_t
dtrace_strlen(const char * s,size_t lim)958 dtrace_strlen(const char *s, size_t lim)
959 {
960 	uint_t len;
961 
962 	for (len = 0; len != lim; len++) {
963 		if (dtrace_load8((uintptr_t)s++) == '\0')
964 			break;
965 	}
966 
967 	return (len);
968 }
969 
970 /*
971  * Check if an address falls within a toxic region.
972  */
973 static int
dtrace_istoxic(uintptr_t kaddr,size_t size)974 dtrace_istoxic(uintptr_t kaddr, size_t size)
975 {
976 	uintptr_t taddr, tsize;
977 	int i;
978 
979 	for (i = 0; i < dtrace_toxranges; i++) {
980 		taddr = dtrace_toxrange[i].dtt_base;
981 		tsize = dtrace_toxrange[i].dtt_limit - taddr;
982 
983 		if (kaddr - taddr < tsize) {
984 			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
985 			cpu_core[CPU->cpu_id].cpuc_dtrace_illval = kaddr;
986 			return (1);
987 		}
988 
989 		if (taddr - kaddr < size) {
990 			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
991 			cpu_core[CPU->cpu_id].cpuc_dtrace_illval = taddr;
992 			return (1);
993 		}
994 	}
995 
996 	return (0);
997 }
998 
999 /*
1000  * Copy src to dst using safe memory accesses.  The src is assumed to be unsafe
1001  * memory specified by the DIF program.  The dst is assumed to be safe memory
1002  * that we can store to directly because it is managed by DTrace.  As with
1003  * standard bcopy, overlapping copies are handled properly.
1004  */
1005 static void
dtrace_bcopy(const void * src,void * dst,size_t len)1006 dtrace_bcopy(const void *src, void *dst, size_t len)
1007 {
1008 	if (len != 0) {
1009 		uint8_t *s1 = dst;
1010 		const uint8_t *s2 = src;
1011 
1012 		if (s1 <= s2) {
1013 			do {
1014 				*s1++ = dtrace_load8((uintptr_t)s2++);
1015 			} while (--len != 0);
1016 		} else {
1017 			s2 += len;
1018 			s1 += len;
1019 
1020 			do {
1021 				*--s1 = dtrace_load8((uintptr_t)--s2);
1022 			} while (--len != 0);
1023 		}
1024 	}
1025 }
1026 
1027 /*
1028  * Copy src to dst using safe memory accesses, up to either the specified
1029  * length, or the point that a nul byte is encountered.  The src is assumed to
1030  * be unsafe memory specified by the DIF program.  The dst is assumed to be
1031  * safe memory that we can store to directly because it is managed by DTrace.
1032  * Unlike dtrace_bcopy(), overlapping regions are not handled.
1033  */
1034 static void
dtrace_strcpy(const void * src,void * dst,size_t len)1035 dtrace_strcpy(const void *src, void *dst, size_t len)
1036 {
1037 	if (len != 0) {
1038 		uint8_t *s1 = dst, c;
1039 		const uint8_t *s2 = src;
1040 
1041 		do {
1042 			*s1++ = c = dtrace_load8((uintptr_t)s2++);
1043 		} while (--len != 0 && c != '\0');
1044 	}
1045 }
1046 
1047 /*
1048  * Copy src to dst, deriving the size and type from the specified (BYREF)
1049  * variable type.  The src is assumed to be unsafe memory specified by the DIF
1050  * program.  The dst is assumed to be DTrace variable memory that is of the
1051  * specified type; we assume that we can store to directly.
1052  */
1053 static void
dtrace_vcopy(void * src,void * dst,dtrace_diftype_t * type)1054 dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type)
1055 {
1056 	ASSERT(type->dtdt_flags & DIF_TF_BYREF);
1057 
1058 	if (type->dtdt_kind == DIF_TYPE_STRING) {
1059 		dtrace_strcpy(src, dst, type->dtdt_size);
1060 	} else {
1061 		dtrace_bcopy(src, dst, type->dtdt_size);
1062 	}
1063 }
1064 
1065 /*
1066  * Compare s1 to s2 using safe memory accesses.  The s1 data is assumed to be
1067  * unsafe memory specified by the DIF program.  The s2 data is assumed to be
1068  * safe memory that we can access directly because it is managed by DTrace.
1069  */
1070 static int
dtrace_bcmp(const void * s1,const void * s2,size_t len)1071 dtrace_bcmp(const void *s1, const void *s2, size_t len)
1072 {
1073 	volatile uint16_t *flags;
1074 
1075 	flags = (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
1076 
1077 	if (s1 == s2)
1078 		return (0);
1079 
1080 	if (s1 == NULL || s2 == NULL)
1081 		return (1);
1082 
1083 	if (s1 != s2 && len != 0) {
1084 		const uint8_t *ps1 = s1;
1085 		const uint8_t *ps2 = s2;
1086 
1087 		do {
1088 			if (dtrace_load8((uintptr_t)ps1++) != *ps2++)
1089 				return (1);
1090 		} while (--len != 0 && !(*flags & CPU_DTRACE_FAULT));
1091 	}
1092 	return (0);
1093 }
1094 
1095 /*
1096  * Zero the specified region using a simple byte-by-byte loop.  Note that this
1097  * is for safe DTrace-managed memory only.
1098  */
1099 static void
dtrace_bzero(void * dst,size_t len)1100 dtrace_bzero(void *dst, size_t len)
1101 {
1102 	uchar_t *cp;
1103 
1104 	for (cp = dst; len != 0; len--)
1105 		*cp++ = 0;
1106 }
1107 
1108 static void
dtrace_add_128(uint64_t * addend1,uint64_t * addend2,uint64_t * sum)1109 dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum)
1110 {
1111 	uint64_t result[2];
1112 
1113 	result[0] = addend1[0] + addend2[0];
1114 	result[1] = addend1[1] + addend2[1] +
1115 	    (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0);
1116 
1117 	sum[0] = result[0];
1118 	sum[1] = result[1];
1119 }
1120 
1121 /*
1122  * Shift the 128-bit value in a by b. If b is positive, shift left.
1123  * If b is negative, shift right.
1124  */
1125 static void
dtrace_shift_128(uint64_t * a,int b)1126 dtrace_shift_128(uint64_t *a, int b)
1127 {
1128 	uint64_t mask;
1129 
1130 	if (b == 0)
1131 		return;
1132 
1133 	if (b < 0) {
1134 		b = -b;
1135 		if (b >= 64) {
1136 			a[0] = a[1] >> (b - 64);
1137 			a[1] = 0;
1138 		} else {
1139 			a[0] >>= b;
1140 			mask = 1LL << (64 - b);
1141 			mask -= 1;
1142 			a[0] |= ((a[1] & mask) << (64 - b));
1143 			a[1] >>= b;
1144 		}
1145 	} else {
1146 		if (b >= 64) {
1147 			a[1] = a[0] << (b - 64);
1148 			a[0] = 0;
1149 		} else {
1150 			a[1] <<= b;
1151 			mask = a[0] >> (64 - b);
1152 			a[1] |= mask;
1153 			a[0] <<= b;
1154 		}
1155 	}
1156 }
1157 
1158 /*
1159  * The basic idea is to break the 2 64-bit values into 4 32-bit values,
1160  * use native multiplication on those, and then re-combine into the
1161  * resulting 128-bit value.
1162  *
1163  * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) =
1164  *     hi1 * hi2 << 64 +
1165  *     hi1 * lo2 << 32 +
1166  *     hi2 * lo1 << 32 +
1167  *     lo1 * lo2
1168  */
1169 static void
dtrace_multiply_128(uint64_t factor1,uint64_t factor2,uint64_t * product)1170 dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product)
1171 {
1172 	uint64_t hi1, hi2, lo1, lo2;
1173 	uint64_t tmp[2];
1174 
1175 	hi1 = factor1 >> 32;
1176 	hi2 = factor2 >> 32;
1177 
1178 	lo1 = factor1 & DT_MASK_LO;
1179 	lo2 = factor2 & DT_MASK_LO;
1180 
1181 	product[0] = lo1 * lo2;
1182 	product[1] = hi1 * hi2;
1183 
1184 	tmp[0] = hi1 * lo2;
1185 	tmp[1] = 0;
1186 	dtrace_shift_128(tmp, 32);
1187 	dtrace_add_128(product, tmp, product);
1188 
1189 	tmp[0] = hi2 * lo1;
1190 	tmp[1] = 0;
1191 	dtrace_shift_128(tmp, 32);
1192 	dtrace_add_128(product, tmp, product);
1193 }
1194 
1195 /*
1196  * This privilege check should be used by actions and subroutines to
1197  * verify that the user credentials of the process that enabled the
1198  * invoking ECB match the target credentials
1199  */
1200 static int
dtrace_priv_proc_common_user(dtrace_state_t * state)1201 dtrace_priv_proc_common_user(dtrace_state_t *state)
1202 {
1203 	cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1204 
1205 	/*
1206 	 * We should always have a non-NULL state cred here, since if cred
1207 	 * is null (anonymous tracing), we fast-path bypass this routine.
1208 	 */
1209 	ASSERT(s_cr != NULL);
1210 
1211 	if ((cr = CRED()) != NULL &&
1212 	    s_cr->cr_uid == cr->cr_uid &&
1213 	    s_cr->cr_uid == cr->cr_ruid &&
1214 	    s_cr->cr_uid == cr->cr_suid &&
1215 	    s_cr->cr_gid == cr->cr_gid &&
1216 	    s_cr->cr_gid == cr->cr_rgid &&
1217 	    s_cr->cr_gid == cr->cr_sgid)
1218 		return (1);
1219 
1220 	return (0);
1221 }
1222 
1223 /*
1224  * This privilege check should be used by actions and subroutines to
1225  * verify that the zone of the process that enabled the invoking ECB
1226  * matches the target credentials
1227  */
1228 static int
dtrace_priv_proc_common_zone(dtrace_state_t * state)1229 dtrace_priv_proc_common_zone(dtrace_state_t *state)
1230 {
1231 	cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1232 
1233 	/*
1234 	 * We should always have a non-NULL state cred here, since if cred
1235 	 * is null (anonymous tracing), we fast-path bypass this routine.
1236 	 */
1237 	ASSERT(s_cr != NULL);
1238 
1239 	if ((cr = CRED()) != NULL && s_cr->cr_zone == cr->cr_zone)
1240 		return (1);
1241 
1242 	return (0);
1243 }
1244 
1245 /*
1246  * This privilege check should be used by actions and subroutines to
1247  * verify that the process has not setuid or changed credentials.
1248  */
1249 static int
dtrace_priv_proc_common_nocd()1250 dtrace_priv_proc_common_nocd()
1251 {
1252 	proc_t *proc;
1253 
1254 	if ((proc = ttoproc(curthread)) != NULL &&
1255 	    !(proc->p_flag & SNOCD))
1256 		return (1);
1257 
1258 	return (0);
1259 }
1260 
1261 static int
dtrace_priv_proc_destructive(dtrace_state_t * state,dtrace_mstate_t * mstate)1262 dtrace_priv_proc_destructive(dtrace_state_t *state, dtrace_mstate_t *mstate)
1263 {
1264 	int action = state->dts_cred.dcr_action;
1265 
1266 	if (!(mstate->dtms_access & DTRACE_ACCESS_PROC))
1267 		goto bad;
1268 
1269 	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) &&
1270 	    dtrace_priv_proc_common_zone(state) == 0)
1271 		goto bad;
1272 
1273 	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) &&
1274 	    dtrace_priv_proc_common_user(state) == 0)
1275 		goto bad;
1276 
1277 	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) &&
1278 	    dtrace_priv_proc_common_nocd() == 0)
1279 		goto bad;
1280 
1281 	return (1);
1282 
1283 bad:
1284 	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1285 
1286 	return (0);
1287 }
1288 
1289 static int
dtrace_priv_proc_control(dtrace_state_t * state,dtrace_mstate_t * mstate)1290 dtrace_priv_proc_control(dtrace_state_t *state, dtrace_mstate_t *mstate)
1291 {
1292 	if (mstate->dtms_access & DTRACE_ACCESS_PROC) {
1293 		if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL)
1294 			return (1);
1295 
1296 		if (dtrace_priv_proc_common_zone(state) &&
1297 		    dtrace_priv_proc_common_user(state) &&
1298 		    dtrace_priv_proc_common_nocd())
1299 			return (1);
1300 	}
1301 
1302 	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1303 
1304 	return (0);
1305 }
1306 
1307 static int
dtrace_priv_proc(dtrace_state_t * state,dtrace_mstate_t * mstate)1308 dtrace_priv_proc(dtrace_state_t *state, dtrace_mstate_t *mstate)
1309 {
1310 	if ((mstate->dtms_access & DTRACE_ACCESS_PROC) &&
1311 	    (state->dts_cred.dcr_action & DTRACE_CRA_PROC))
1312 		return (1);
1313 
1314 	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1315 
1316 	return (0);
1317 }
1318 
1319 static int
dtrace_priv_kernel(dtrace_state_t * state)1320 dtrace_priv_kernel(dtrace_state_t *state)
1321 {
1322 	if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL)
1323 		return (1);
1324 
1325 	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1326 
1327 	return (0);
1328 }
1329 
1330 static int
dtrace_priv_kernel_destructive(dtrace_state_t * state)1331 dtrace_priv_kernel_destructive(dtrace_state_t *state)
1332 {
1333 	if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE)
1334 		return (1);
1335 
1336 	cpu_core[CPU->cpu_id].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1337 
1338 	return (0);
1339 }
1340 
1341 /*
1342  * Determine if the dte_cond of the specified ECB allows for processing of
1343  * the current probe to continue.  Note that this routine may allow continued
1344  * processing, but with access(es) stripped from the mstate's dtms_access
1345  * field.
1346  */
1347 static int
dtrace_priv_probe(dtrace_state_t * state,dtrace_mstate_t * mstate,dtrace_ecb_t * ecb)1348 dtrace_priv_probe(dtrace_state_t *state, dtrace_mstate_t *mstate,
1349     dtrace_ecb_t *ecb)
1350 {
1351 	dtrace_probe_t *probe = ecb->dte_probe;
1352 	dtrace_provider_t *prov = probe->dtpr_provider;
1353 	dtrace_pops_t *pops = &prov->dtpv_pops;
1354 	int mode = DTRACE_MODE_NOPRIV_DROP;
1355 
1356 	ASSERT(ecb->dte_cond);
1357 
1358 	if (pops->dtps_mode != NULL) {
1359 		mode = pops->dtps_mode(prov->dtpv_arg,
1360 		    probe->dtpr_id, probe->dtpr_arg);
1361 
1362 		ASSERT(mode & (DTRACE_MODE_USER | DTRACE_MODE_KERNEL));
1363 		ASSERT(mode & (DTRACE_MODE_NOPRIV_RESTRICT |
1364 		    DTRACE_MODE_NOPRIV_DROP));
1365 	}
1366 
1367 	/*
1368 	 * If the dte_cond bits indicate that this consumer is only allowed to
1369 	 * see user-mode firings of this probe, check that the probe was fired
1370 	 * while in a user context.  If that's not the case, use the policy
1371 	 * specified by the provider to determine if we drop the probe or
1372 	 * merely restrict operation.
1373 	 */
1374 	if (ecb->dte_cond & DTRACE_COND_USERMODE) {
1375 		ASSERT(mode != DTRACE_MODE_NOPRIV_DROP);
1376 
1377 		if (!(mode & DTRACE_MODE_USER)) {
1378 			if (mode & DTRACE_MODE_NOPRIV_DROP)
1379 				return (0);
1380 
1381 			mstate->dtms_access &= ~DTRACE_ACCESS_ARGS;
1382 		}
1383 	}
1384 
1385 	/*
1386 	 * This is more subtle than it looks. We have to be absolutely certain
1387 	 * that CRED() isn't going to change out from under us so it's only
1388 	 * legit to examine that structure if we're in constrained situations.
1389 	 * Currently, the only times we'll this check is if a non-super-user
1390 	 * has enabled the profile or syscall providers -- providers that
1391 	 * allow visibility of all processes. For the profile case, the check
1392 	 * above will ensure that we're examining a user context.
1393 	 */
1394 	if (ecb->dte_cond & DTRACE_COND_OWNER) {
1395 		cred_t *cr;
1396 		cred_t *s_cr = state->dts_cred.dcr_cred;
1397 		proc_t *proc;
1398 
1399 		ASSERT(s_cr != NULL);
1400 
1401 		if ((cr = CRED()) == NULL ||
1402 		    s_cr->cr_uid != cr->cr_uid ||
1403 		    s_cr->cr_uid != cr->cr_ruid ||
1404 		    s_cr->cr_uid != cr->cr_suid ||
1405 		    s_cr->cr_gid != cr->cr_gid ||
1406 		    s_cr->cr_gid != cr->cr_rgid ||
1407 		    s_cr->cr_gid != cr->cr_sgid ||
1408 		    (proc = ttoproc(curthread)) == NULL ||
1409 		    (proc->p_flag & SNOCD)) {
1410 			if (mode & DTRACE_MODE_NOPRIV_DROP)
1411 				return (0);
1412 
1413 			mstate->dtms_access &= ~DTRACE_ACCESS_PROC;
1414 		}
1415 	}
1416 
1417 	/*
1418 	 * If our dte_cond is set to DTRACE_COND_ZONEOWNER and we are not
1419 	 * in our zone, check to see if our mode policy is to restrict rather
1420 	 * than to drop; if to restrict, strip away both DTRACE_ACCESS_PROC
1421 	 * and DTRACE_ACCESS_ARGS
1422 	 */
1423 	if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) {
1424 		cred_t *cr;
1425 		cred_t *s_cr = state->dts_cred.dcr_cred;
1426 
1427 		ASSERT(s_cr != NULL);
1428 
1429 		if ((cr = CRED()) == NULL ||
1430 		    s_cr->cr_zone->zone_id != cr->cr_zone->zone_id) {
1431 			if (mode & DTRACE_MODE_NOPRIV_DROP)
1432 				return (0);
1433 
1434 			mstate->dtms_access &=
1435 			    ~(DTRACE_ACCESS_PROC | DTRACE_ACCESS_ARGS);
1436 		}
1437 	}
1438 
1439 	/*
1440 	 * By merits of being in this code path at all, we have limited
1441 	 * privileges.  If the provider has indicated that limited privileges
1442 	 * are to denote restricted operation, strip off the ability to access
1443 	 * arguments.
1444 	 */
1445 	if (mode & DTRACE_MODE_LIMITEDPRIV_RESTRICT)
1446 		mstate->dtms_access &= ~DTRACE_ACCESS_ARGS;
1447 
1448 	return (1);
1449 }
1450 
1451 /*
1452  * Note:  not called from probe context.  This function is called
1453  * asynchronously (and at a regular interval) from outside of probe context to
1454  * clean the dirty dynamic variable lists on all CPUs.  Dynamic variable
1455  * cleaning is explained in detail in <sys/dtrace_impl.h>.
1456  */
1457 void
dtrace_dynvar_clean(dtrace_dstate_t * dstate)1458 dtrace_dynvar_clean(dtrace_dstate_t *dstate)
1459 {
1460 	dtrace_dynvar_t *dirty;
1461 	dtrace_dstate_percpu_t *dcpu;
1462 	dtrace_dynvar_t **rinsep;
1463 	int i, j, work = 0;
1464 
1465 	for (i = 0; i < NCPU; i++) {
1466 		dcpu = &dstate->dtds_percpu[i];
1467 		rinsep = &dcpu->dtdsc_rinsing;
1468 
1469 		/*
1470 		 * If the dirty list is NULL, there is no dirty work to do.
1471 		 */
1472 		if (dcpu->dtdsc_dirty == NULL)
1473 			continue;
1474 
1475 		if (dcpu->dtdsc_rinsing != NULL) {
1476 			/*
1477 			 * If the rinsing list is non-NULL, then it is because
1478 			 * this CPU was selected to accept another CPU's
1479 			 * dirty list -- and since that time, dirty buffers
1480 			 * have accumulated.  This is a highly unlikely
1481 			 * condition, but we choose to ignore the dirty
1482 			 * buffers -- they'll be picked up a future cleanse.
1483 			 */
1484 			continue;
1485 		}
1486 
1487 		if (dcpu->dtdsc_clean != NULL) {
1488 			/*
1489 			 * If the clean list is non-NULL, then we're in a
1490 			 * situation where a CPU has done deallocations (we
1491 			 * have a non-NULL dirty list) but no allocations (we
1492 			 * also have a non-NULL clean list).  We can't simply
1493 			 * move the dirty list into the clean list on this
1494 			 * CPU, yet we also don't want to allow this condition
1495 			 * to persist, lest a short clean list prevent a
1496 			 * massive dirty list from being cleaned (which in
1497 			 * turn could lead to otherwise avoidable dynamic
1498 			 * drops).  To deal with this, we look for some CPU
1499 			 * with a NULL clean list, NULL dirty list, and NULL
1500 			 * rinsing list -- and then we borrow this CPU to
1501 			 * rinse our dirty list.
1502 			 */
1503 			for (j = 0; j < NCPU; j++) {
1504 				dtrace_dstate_percpu_t *rinser;
1505 
1506 				rinser = &dstate->dtds_percpu[j];
1507 
1508 				if (rinser->dtdsc_rinsing != NULL)
1509 					continue;
1510 
1511 				if (rinser->dtdsc_dirty != NULL)
1512 					continue;
1513 
1514 				if (rinser->dtdsc_clean != NULL)
1515 					continue;
1516 
1517 				rinsep = &rinser->dtdsc_rinsing;
1518 				break;
1519 			}
1520 
1521 			if (j == NCPU) {
1522 				/*
1523 				 * We were unable to find another CPU that
1524 				 * could accept this dirty list -- we are
1525 				 * therefore unable to clean it now.
1526 				 */
1527 				dtrace_dynvar_failclean++;
1528 				continue;
1529 			}
1530 		}
1531 
1532 		work = 1;
1533 
1534 		/*
1535 		 * Atomically move the dirty list aside.
1536 		 */
1537 		do {
1538 			dirty = dcpu->dtdsc_dirty;
1539 
1540 			/*
1541 			 * Before we zap the dirty list, set the rinsing list.
1542 			 * (This allows for a potential assertion in
1543 			 * dtrace_dynvar():  if a free dynamic variable appears
1544 			 * on a hash chain, either the dirty list or the
1545 			 * rinsing list for some CPU must be non-NULL.)
1546 			 */
1547 			*rinsep = dirty;
1548 			dtrace_membar_producer();
1549 		} while (dtrace_casptr(&dcpu->dtdsc_dirty,
1550 		    dirty, NULL) != dirty);
1551 	}
1552 
1553 	if (!work) {
1554 		/*
1555 		 * We have no work to do; we can simply return.
1556 		 */
1557 		return;
1558 	}
1559 
1560 	dtrace_sync();
1561 
1562 	for (i = 0; i < NCPU; i++) {
1563 		dcpu = &dstate->dtds_percpu[i];
1564 
1565 		if (dcpu->dtdsc_rinsing == NULL)
1566 			continue;
1567 
1568 		/*
1569 		 * We are now guaranteed that no hash chain contains a pointer
1570 		 * into this dirty list; we can make it clean.
1571 		 */
1572 		ASSERT(dcpu->dtdsc_clean == NULL);
1573 		dcpu->dtdsc_clean = dcpu->dtdsc_rinsing;
1574 		dcpu->dtdsc_rinsing = NULL;
1575 	}
1576 
1577 	/*
1578 	 * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make
1579 	 * sure that all CPUs have seen all of the dtdsc_clean pointers.
1580 	 * This prevents a race whereby a CPU incorrectly decides that
1581 	 * the state should be something other than DTRACE_DSTATE_CLEAN
1582 	 * after dtrace_dynvar_clean() has completed.
1583 	 */
1584 	dtrace_sync();
1585 
1586 	dstate->dtds_state = DTRACE_DSTATE_CLEAN;
1587 }
1588 
1589 /*
1590  * Depending on the value of the op parameter, this function looks-up,
1591  * allocates or deallocates an arbitrarily-keyed dynamic variable.  If an
1592  * allocation is requested, this function will return a pointer to a
1593  * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no
1594  * variable can be allocated.  If NULL is returned, the appropriate counter
1595  * will be incremented.
1596  */
1597 dtrace_dynvar_t *
dtrace_dynvar(dtrace_dstate_t * dstate,uint_t nkeys,dtrace_key_t * key,size_t dsize,dtrace_dynvar_op_t op,dtrace_mstate_t * mstate,dtrace_vstate_t * vstate)1598 dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys,
1599     dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op,
1600     dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
1601 {
1602 	uint64_t hashval = DTRACE_DYNHASH_VALID;
1603 	dtrace_dynhash_t *hash = dstate->dtds_hash;
1604 	dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL;
1605 	processorid_t me = CPU->cpu_id, cpu = me;
1606 	dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me];
1607 	size_t bucket, ksize;
1608 	size_t chunksize = dstate->dtds_chunksize;
1609 	uintptr_t kdata, lock, nstate;
1610 	uint_t i;
1611 
1612 	ASSERT(nkeys != 0);
1613 
1614 	/*
1615 	 * Hash the key.  As with aggregations, we use Jenkins' "One-at-a-time"
1616 	 * algorithm.  For the by-value portions, we perform the algorithm in
1617 	 * 16-bit chunks (as opposed to 8-bit chunks).  This speeds things up a
1618 	 * bit, and seems to have only a minute effect on distribution.  For
1619 	 * the by-reference data, we perform "One-at-a-time" iterating (safely)
1620 	 * over each referenced byte.  It's painful to do this, but it's much
1621 	 * better than pathological hash distribution.  The efficacy of the
1622 	 * hashing algorithm (and a comparison with other algorithms) may be
1623 	 * found by running the ::dtrace_dynstat MDB dcmd.
1624 	 */
1625 	for (i = 0; i < nkeys; i++) {
1626 		if (key[i].dttk_size == 0) {
1627 			uint64_t val = key[i].dttk_value;
1628 
1629 			hashval += (val >> 48) & 0xffff;
1630 			hashval += (hashval << 10);
1631 			hashval ^= (hashval >> 6);
1632 
1633 			hashval += (val >> 32) & 0xffff;
1634 			hashval += (hashval << 10);
1635 			hashval ^= (hashval >> 6);
1636 
1637 			hashval += (val >> 16) & 0xffff;
1638 			hashval += (hashval << 10);
1639 			hashval ^= (hashval >> 6);
1640 
1641 			hashval += val & 0xffff;
1642 			hashval += (hashval << 10);
1643 			hashval ^= (hashval >> 6);
1644 		} else {
1645 			/*
1646 			 * This is incredibly painful, but it beats the hell
1647 			 * out of the alternative.
1648 			 */
1649 			uint64_t j, size = key[i].dttk_size;
1650 			uintptr_t base = (uintptr_t)key[i].dttk_value;
1651 
1652 			if (!dtrace_canload(base, size, mstate, vstate))
1653 				break;
1654 
1655 			for (j = 0; j < size; j++) {
1656 				hashval += dtrace_load8(base + j);
1657 				hashval += (hashval << 10);
1658 				hashval ^= (hashval >> 6);
1659 			}
1660 		}
1661 	}
1662 
1663 	if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT))
1664 		return (NULL);
1665 
1666 	hashval += (hashval << 3);
1667 	hashval ^= (hashval >> 11);
1668 	hashval += (hashval << 15);
1669 
1670 	/*
1671 	 * There is a remote chance (ideally, 1 in 2^31) that our hashval
1672 	 * comes out to be one of our two sentinel hash values.  If this
1673 	 * actually happens, we set the hashval to be a value known to be a
1674 	 * non-sentinel value.
1675 	 */
1676 	if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK)
1677 		hashval = DTRACE_DYNHASH_VALID;
1678 
1679 	/*
1680 	 * Yes, it's painful to do a divide here.  If the cycle count becomes
1681 	 * important here, tricks can be pulled to reduce it.  (However, it's
1682 	 * critical that hash collisions be kept to an absolute minimum;
1683 	 * they're much more painful than a divide.)  It's better to have a
1684 	 * solution that generates few collisions and still keeps things
1685 	 * relatively simple.
1686 	 */
1687 	bucket = hashval % dstate->dtds_hashsize;
1688 
1689 	if (op == DTRACE_DYNVAR_DEALLOC) {
1690 		volatile uintptr_t *lockp = &hash[bucket].dtdh_lock;
1691 
1692 		for (;;) {
1693 			while ((lock = *lockp) & 1)
1694 				continue;
1695 
1696 			if (dtrace_casptr((void *)lockp,
1697 			    (void *)lock, (void *)(lock + 1)) == (void *)lock)
1698 				break;
1699 		}
1700 
1701 		dtrace_membar_producer();
1702 	}
1703 
1704 top:
1705 	prev = NULL;
1706 	lock = hash[bucket].dtdh_lock;
1707 
1708 	dtrace_membar_consumer();
1709 
1710 	start = hash[bucket].dtdh_chain;
1711 	ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK ||
1712 	    start->dtdv_hashval != DTRACE_DYNHASH_FREE ||
1713 	    op != DTRACE_DYNVAR_DEALLOC));
1714 
1715 	for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) {
1716 		dtrace_tuple_t *dtuple = &dvar->dtdv_tuple;
1717 		dtrace_key_t *dkey = &dtuple->dtt_key[0];
1718 
1719 		if (dvar->dtdv_hashval != hashval) {
1720 			if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) {
1721 				/*
1722 				 * We've reached the sink, and therefore the
1723 				 * end of the hash chain; we can kick out of
1724 				 * the loop knowing that we have seen a valid
1725 				 * snapshot of state.
1726 				 */
1727 				ASSERT(dvar->dtdv_next == NULL);
1728 				ASSERT(dvar == &dtrace_dynhash_sink);
1729 				break;
1730 			}
1731 
1732 			if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) {
1733 				/*
1734 				 * We've gone off the rails:  somewhere along
1735 				 * the line, one of the members of this hash
1736 				 * chain was deleted.  Note that we could also
1737 				 * detect this by simply letting this loop run
1738 				 * to completion, as we would eventually hit
1739 				 * the end of the dirty list.  However, we
1740 				 * want to avoid running the length of the
1741 				 * dirty list unnecessarily (it might be quite
1742 				 * long), so we catch this as early as
1743 				 * possible by detecting the hash marker.  In
1744 				 * this case, we simply set dvar to NULL and
1745 				 * break; the conditional after the loop will
1746 				 * send us back to top.
1747 				 */
1748 				dvar = NULL;
1749 				break;
1750 			}
1751 
1752 			goto next;
1753 		}
1754 
1755 		if (dtuple->dtt_nkeys != nkeys)
1756 			goto next;
1757 
1758 		for (i = 0; i < nkeys; i++, dkey++) {
1759 			if (dkey->dttk_size != key[i].dttk_size)
1760 				goto next; /* size or type mismatch */
1761 
1762 			if (dkey->dttk_size != 0) {
1763 				if (dtrace_bcmp(
1764 				    (void *)(uintptr_t)key[i].dttk_value,
1765 				    (void *)(uintptr_t)dkey->dttk_value,
1766 				    dkey->dttk_size))
1767 					goto next;
1768 			} else {
1769 				if (dkey->dttk_value != key[i].dttk_value)
1770 					goto next;
1771 			}
1772 		}
1773 
1774 		if (op != DTRACE_DYNVAR_DEALLOC)
1775 			return (dvar);
1776 
1777 		ASSERT(dvar->dtdv_next == NULL ||
1778 		    dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE);
1779 
1780 		if (prev != NULL) {
1781 			ASSERT(hash[bucket].dtdh_chain != dvar);
1782 			ASSERT(start != dvar);
1783 			ASSERT(prev->dtdv_next == dvar);
1784 			prev->dtdv_next = dvar->dtdv_next;
1785 		} else {
1786 			if (dtrace_casptr(&hash[bucket].dtdh_chain,
1787 			    start, dvar->dtdv_next) != start) {
1788 				/*
1789 				 * We have failed to atomically swing the
1790 				 * hash table head pointer, presumably because
1791 				 * of a conflicting allocation on another CPU.
1792 				 * We need to reread the hash chain and try
1793 				 * again.
1794 				 */
1795 				goto top;
1796 			}
1797 		}
1798 
1799 		dtrace_membar_producer();
1800 
1801 		/*
1802 		 * Now set the hash value to indicate that it's free.
1803 		 */
1804 		ASSERT(hash[bucket].dtdh_chain != dvar);
1805 		dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
1806 
1807 		dtrace_membar_producer();
1808 
1809 		/*
1810 		 * Set the next pointer to point at the dirty list, and
1811 		 * atomically swing the dirty pointer to the newly freed dvar.
1812 		 */
1813 		do {
1814 			next = dcpu->dtdsc_dirty;
1815 			dvar->dtdv_next = next;
1816 		} while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next);
1817 
1818 		/*
1819 		 * Finally, unlock this hash bucket.
1820 		 */
1821 		ASSERT(hash[bucket].dtdh_lock == lock);
1822 		ASSERT(lock & 1);
1823 		hash[bucket].dtdh_lock++;
1824 
1825 		return (NULL);
1826 next:
1827 		prev = dvar;
1828 		continue;
1829 	}
1830 
1831 	if (dvar == NULL) {
1832 		/*
1833 		 * If dvar is NULL, it is because we went off the rails:
1834 		 * one of the elements that we traversed in the hash chain
1835 		 * was deleted while we were traversing it.  In this case,
1836 		 * we assert that we aren't doing a dealloc (deallocs lock
1837 		 * the hash bucket to prevent themselves from racing with
1838 		 * one another), and retry the hash chain traversal.
1839 		 */
1840 		ASSERT(op != DTRACE_DYNVAR_DEALLOC);
1841 		goto top;
1842 	}
1843 
1844 	if (op != DTRACE_DYNVAR_ALLOC) {
1845 		/*
1846 		 * If we are not to allocate a new variable, we want to
1847 		 * return NULL now.  Before we return, check that the value
1848 		 * of the lock word hasn't changed.  If it has, we may have
1849 		 * seen an inconsistent snapshot.
1850 		 */
1851 		if (op == DTRACE_DYNVAR_NOALLOC) {
1852 			if (hash[bucket].dtdh_lock != lock)
1853 				goto top;
1854 		} else {
1855 			ASSERT(op == DTRACE_DYNVAR_DEALLOC);
1856 			ASSERT(hash[bucket].dtdh_lock == lock);
1857 			ASSERT(lock & 1);
1858 			hash[bucket].dtdh_lock++;
1859 		}
1860 
1861 		return (NULL);
1862 	}
1863 
1864 	/*
1865 	 * We need to allocate a new dynamic variable.  The size we need is the
1866 	 * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the
1867 	 * size of any auxiliary key data (rounded up to 8-byte alignment) plus
1868 	 * the size of any referred-to data (dsize).  We then round the final
1869 	 * size up to the chunksize for allocation.
1870 	 */
1871 	for (ksize = 0, i = 0; i < nkeys; i++)
1872 		ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
1873 
1874 	/*
1875 	 * This should be pretty much impossible, but could happen if, say,
1876 	 * strange DIF specified the tuple.  Ideally, this should be an
1877 	 * assertion and not an error condition -- but that requires that the
1878 	 * chunksize calculation in dtrace_difo_chunksize() be absolutely
1879 	 * bullet-proof.  (That is, it must not be able to be fooled by
1880 	 * malicious DIF.)  Given the lack of backwards branches in DIF,
1881 	 * solving this would presumably not amount to solving the Halting
1882 	 * Problem -- but it still seems awfully hard.
1883 	 */
1884 	if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) +
1885 	    ksize + dsize > chunksize) {
1886 		dcpu->dtdsc_drops++;
1887 		return (NULL);
1888 	}
1889 
1890 	nstate = DTRACE_DSTATE_EMPTY;
1891 
1892 	do {
1893 retry:
1894 		free = dcpu->dtdsc_free;
1895 
1896 		if (free == NULL) {
1897 			dtrace_dynvar_t *clean = dcpu->dtdsc_clean;
1898 			void *rval;
1899 
1900 			if (clean == NULL) {
1901 				/*
1902 				 * We're out of dynamic variable space on
1903 				 * this CPU.  Unless we have tried all CPUs,
1904 				 * we'll try to allocate from a different
1905 				 * CPU.
1906 				 */
1907 				switch (dstate->dtds_state) {
1908 				case DTRACE_DSTATE_CLEAN: {
1909 					void *sp = &dstate->dtds_state;
1910 
1911 					if (++cpu >= NCPU)
1912 						cpu = 0;
1913 
1914 					if (dcpu->dtdsc_dirty != NULL &&
1915 					    nstate == DTRACE_DSTATE_EMPTY)
1916 						nstate = DTRACE_DSTATE_DIRTY;
1917 
1918 					if (dcpu->dtdsc_rinsing != NULL)
1919 						nstate = DTRACE_DSTATE_RINSING;
1920 
1921 					dcpu = &dstate->dtds_percpu[cpu];
1922 
1923 					if (cpu != me)
1924 						goto retry;
1925 
1926 					(void) dtrace_cas32(sp,
1927 					    DTRACE_DSTATE_CLEAN, nstate);
1928 
1929 					/*
1930 					 * To increment the correct bean
1931 					 * counter, take another lap.
1932 					 */
1933 					goto retry;
1934 				}
1935 
1936 				case DTRACE_DSTATE_DIRTY:
1937 					dcpu->dtdsc_dirty_drops++;
1938 					break;
1939 
1940 				case DTRACE_DSTATE_RINSING:
1941 					dcpu->dtdsc_rinsing_drops++;
1942 					break;
1943 
1944 				case DTRACE_DSTATE_EMPTY:
1945 					dcpu->dtdsc_drops++;
1946 					break;
1947 				}
1948 
1949 				DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP);
1950 				return (NULL);
1951 			}
1952 
1953 			/*
1954 			 * The clean list appears to be non-empty.  We want to
1955 			 * move the clean list to the free list; we start by
1956 			 * moving the clean pointer aside.
1957 			 */
1958 			if (dtrace_casptr(&dcpu->dtdsc_clean,
1959 			    clean, NULL) != clean) {
1960 				/*
1961 				 * We are in one of two situations:
1962 				 *
1963 				 *  (a)	The clean list was switched to the
1964 				 *	free list by another CPU.
1965 				 *
1966 				 *  (b)	The clean list was added to by the
1967 				 *	cleansing cyclic.
1968 				 *
1969 				 * In either of these situations, we can
1970 				 * just reattempt the free list allocation.
1971 				 */
1972 				goto retry;
1973 			}
1974 
1975 			ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE);
1976 
1977 			/*
1978 			 * Now we'll move the clean list to our free list.
1979 			 * It's impossible for this to fail:  the only way
1980 			 * the free list can be updated is through this
1981 			 * code path, and only one CPU can own the clean list.
1982 			 * Thus, it would only be possible for this to fail if
1983 			 * this code were racing with dtrace_dynvar_clean().
1984 			 * (That is, if dtrace_dynvar_clean() updated the clean
1985 			 * list, and we ended up racing to update the free
1986 			 * list.)  This race is prevented by the dtrace_sync()
1987 			 * in dtrace_dynvar_clean() -- which flushes the
1988 			 * owners of the clean lists out before resetting
1989 			 * the clean lists.
1990 			 */
1991 			dcpu = &dstate->dtds_percpu[me];
1992 			rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean);
1993 			ASSERT(rval == NULL);
1994 			goto retry;
1995 		}
1996 
1997 		dvar = free;
1998 		new_free = dvar->dtdv_next;
1999 	} while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free);
2000 
2001 	/*
2002 	 * We have now allocated a new chunk.  We copy the tuple keys into the
2003 	 * tuple array and copy any referenced key data into the data space
2004 	 * following the tuple array.  As we do this, we relocate dttk_value
2005 	 * in the final tuple to point to the key data address in the chunk.
2006 	 */
2007 	kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys];
2008 	dvar->dtdv_data = (void *)(kdata + ksize);
2009 	dvar->dtdv_tuple.dtt_nkeys = nkeys;
2010 
2011 	for (i = 0; i < nkeys; i++) {
2012 		dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i];
2013 		size_t kesize = key[i].dttk_size;
2014 
2015 		if (kesize != 0) {
2016 			dtrace_bcopy(
2017 			    (const void *)(uintptr_t)key[i].dttk_value,
2018 			    (void *)kdata, kesize);
2019 			dkey->dttk_value = kdata;
2020 			kdata += P2ROUNDUP(kesize, sizeof (uint64_t));
2021 		} else {
2022 			dkey->dttk_value = key[i].dttk_value;
2023 		}
2024 
2025 		dkey->dttk_size = kesize;
2026 	}
2027 
2028 	ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE);
2029 	dvar->dtdv_hashval = hashval;
2030 	dvar->dtdv_next = start;
2031 
2032 	if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start)
2033 		return (dvar);
2034 
2035 	/*
2036 	 * The cas has failed.  Either another CPU is adding an element to
2037 	 * this hash chain, or another CPU is deleting an element from this
2038 	 * hash chain.  The simplest way to deal with both of these cases
2039 	 * (though not necessarily the most efficient) is to free our
2040 	 * allocated block and re-attempt it all.  Note that the free is
2041 	 * to the dirty list and _not_ to the free list.  This is to prevent
2042 	 * races with allocators, above.
2043 	 */
2044 	dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
2045 
2046 	dtrace_membar_producer();
2047 
2048 	do {
2049 		free = dcpu->dtdsc_dirty;
2050 		dvar->dtdv_next = free;
2051 	} while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free);
2052 
2053 	goto top;
2054 }
2055 
2056 /*ARGSUSED*/
2057 static void
dtrace_aggregate_min(uint64_t * oval,uint64_t nval,uint64_t arg)2058 dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg)
2059 {
2060 	if ((int64_t)nval < (int64_t)*oval)
2061 		*oval = nval;
2062 }
2063 
2064 /*ARGSUSED*/
2065 static void
dtrace_aggregate_max(uint64_t * oval,uint64_t nval,uint64_t arg)2066 dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg)
2067 {
2068 	if ((int64_t)nval > (int64_t)*oval)
2069 		*oval = nval;
2070 }
2071 
2072 static void
dtrace_aggregate_quantize(uint64_t * quanta,uint64_t nval,uint64_t incr)2073 dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr)
2074 {
2075 	int i, zero = DTRACE_QUANTIZE_ZEROBUCKET;
2076 	int64_t val = (int64_t)nval;
2077 
2078 	if (val < 0) {
2079 		for (i = 0; i < zero; i++) {
2080 			if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) {
2081 				quanta[i] += incr;
2082 				return;
2083 			}
2084 		}
2085 	} else {
2086 		for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) {
2087 			if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) {
2088 				quanta[i - 1] += incr;
2089 				return;
2090 			}
2091 		}
2092 
2093 		quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr;
2094 		return;
2095 	}
2096 
2097 	ASSERT(0);
2098 }
2099 
2100 static void
dtrace_aggregate_lquantize(uint64_t * lquanta,uint64_t nval,uint64_t incr)2101 dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr)
2102 {
2103 	uint64_t arg = *lquanta++;
2104 	int32_t base = DTRACE_LQUANTIZE_BASE(arg);
2105 	uint16_t step = DTRACE_LQUANTIZE_STEP(arg);
2106 	uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg);
2107 	int32_t val = (int32_t)nval, level;
2108 
2109 	ASSERT(step != 0);
2110 	ASSERT(levels != 0);
2111 
2112 	if (val < base) {
2113 		/*
2114 		 * This is an underflow.
2115 		 */
2116 		lquanta[0] += incr;
2117 		return;
2118 	}
2119 
2120 	level = (val - base) / step;
2121 
2122 	if (level < levels) {
2123 		lquanta[level + 1] += incr;
2124 		return;
2125 	}
2126 
2127 	/*
2128 	 * This is an overflow.
2129 	 */
2130 	lquanta[levels + 1] += incr;
2131 }
2132 
2133 static int
dtrace_aggregate_llquantize_bucket(uint16_t factor,uint16_t low,uint16_t high,uint16_t nsteps,int64_t value)2134 dtrace_aggregate_llquantize_bucket(uint16_t factor, uint16_t low,
2135     uint16_t high, uint16_t nsteps, int64_t value)
2136 {
2137 	int64_t this = 1, last, next;
2138 	int base = 1, order;
2139 
2140 	ASSERT(factor <= nsteps);
2141 	ASSERT(nsteps % factor == 0);
2142 
2143 	for (order = 0; order < low; order++)
2144 		this *= factor;
2145 
2146 	/*
2147 	 * If our value is less than our factor taken to the power of the
2148 	 * low order of magnitude, it goes into the zeroth bucket.
2149 	 */
2150 	if (value < (last = this))
2151 		return (0);
2152 
2153 	for (this *= factor; order <= high; order++) {
2154 		int nbuckets = this > nsteps ? nsteps : this;
2155 
2156 		if ((next = this * factor) < this) {
2157 			/*
2158 			 * We should not generally get log/linear quantizations
2159 			 * with a high magnitude that allows 64-bits to
2160 			 * overflow, but we nonetheless protect against this
2161 			 * by explicitly checking for overflow, and clamping
2162 			 * our value accordingly.
2163 			 */
2164 			value = this - 1;
2165 		}
2166 
2167 		if (value < this) {
2168 			/*
2169 			 * If our value lies within this order of magnitude,
2170 			 * determine its position by taking the offset within
2171 			 * the order of magnitude, dividing by the bucket
2172 			 * width, and adding to our (accumulated) base.
2173 			 */
2174 			return (base + (value - last) / (this / nbuckets));
2175 		}
2176 
2177 		base += nbuckets - (nbuckets / factor);
2178 		last = this;
2179 		this = next;
2180 	}
2181 
2182 	/*
2183 	 * Our value is greater than or equal to our factor taken to the
2184 	 * power of one plus the high magnitude -- return the top bucket.
2185 	 */
2186 	return (base);
2187 }
2188 
2189 static void
dtrace_aggregate_llquantize(uint64_t * llquanta,uint64_t nval,uint64_t incr)2190 dtrace_aggregate_llquantize(uint64_t *llquanta, uint64_t nval, uint64_t incr)
2191 {
2192 	uint64_t arg = *llquanta++;
2193 	uint16_t factor = DTRACE_LLQUANTIZE_FACTOR(arg);
2194 	uint16_t low = DTRACE_LLQUANTIZE_LOW(arg);
2195 	uint16_t high = DTRACE_LLQUANTIZE_HIGH(arg);
2196 	uint16_t nsteps = DTRACE_LLQUANTIZE_NSTEP(arg);
2197 
2198 	llquanta[dtrace_aggregate_llquantize_bucket(factor,
2199 	    low, high, nsteps, nval)] += incr;
2200 }
2201 
2202 /*ARGSUSED*/
2203 static void
dtrace_aggregate_avg(uint64_t * data,uint64_t nval,uint64_t arg)2204 dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg)
2205 {
2206 	data[0]++;
2207 	data[1] += nval;
2208 }
2209 
2210 /*ARGSUSED*/
2211 static void
dtrace_aggregate_stddev(uint64_t * data,uint64_t nval,uint64_t arg)2212 dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg)
2213 {
2214 	int64_t snval = (int64_t)nval;
2215 	uint64_t tmp[2];
2216 
2217 	data[0]++;
2218 	data[1] += nval;
2219 
2220 	/*
2221 	 * What we want to say here is:
2222 	 *
2223 	 * data[2] += nval * nval;
2224 	 *
2225 	 * But given that nval is 64-bit, we could easily overflow, so
2226 	 * we do this as 128-bit arithmetic.
2227 	 */
2228 	if (snval < 0)
2229 		snval = -snval;
2230 
2231 	dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp);
2232 	dtrace_add_128(data + 2, tmp, data + 2);
2233 }
2234 
2235 /*ARGSUSED*/
2236 static void
dtrace_aggregate_count(uint64_t * oval,uint64_t nval,uint64_t arg)2237 dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg)
2238 {
2239 	*oval = *oval + 1;
2240 }
2241 
2242 /*ARGSUSED*/
2243 static void
dtrace_aggregate_sum(uint64_t * oval,uint64_t nval,uint64_t arg)2244 dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg)
2245 {
2246 	*oval += nval;
2247 }
2248 
2249 /*
2250  * Aggregate given the tuple in the principal data buffer, and the aggregating
2251  * action denoted by the specified dtrace_aggregation_t.  The aggregation
2252  * buffer is specified as the buf parameter.  This routine does not return
2253  * failure; if there is no space in the aggregation buffer, the data will be
2254  * dropped, and a corresponding counter incremented.
2255  */
2256 static void
dtrace_aggregate(dtrace_aggregation_t * agg,dtrace_buffer_t * dbuf,intptr_t offset,dtrace_buffer_t * buf,uint64_t expr,uint64_t arg)2257 dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf,
2258     intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg)
2259 {
2260 	dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec;
2261 	uint32_t i, ndx, size, fsize;
2262 	uint32_t align = sizeof (uint64_t) - 1;
2263 	dtrace_aggbuffer_t *agb;
2264 	dtrace_aggkey_t *key;
2265 	uint32_t hashval = 0, limit, isstr;
2266 	caddr_t tomax, data, kdata;
2267 	dtrace_actkind_t action;
2268 	dtrace_action_t *act;
2269 	uintptr_t offs;
2270 
2271 	if (buf == NULL)
2272 		return;
2273 
2274 	if (!agg->dtag_hasarg) {
2275 		/*
2276 		 * Currently, only quantize() and lquantize() take additional
2277 		 * arguments, and they have the same semantics:  an increment
2278 		 * value that defaults to 1 when not present.  If additional
2279 		 * aggregating actions take arguments, the setting of the
2280 		 * default argument value will presumably have to become more
2281 		 * sophisticated...
2282 		 */
2283 		arg = 1;
2284 	}
2285 
2286 	action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION;
2287 	size = rec->dtrd_offset - agg->dtag_base;
2288 	fsize = size + rec->dtrd_size;
2289 
2290 	ASSERT(dbuf->dtb_tomax != NULL);
2291 	data = dbuf->dtb_tomax + offset + agg->dtag_base;
2292 
2293 	if ((tomax = buf->dtb_tomax) == NULL) {
2294 		dtrace_buffer_drop(buf);
2295 		return;
2296 	}
2297 
2298 	/*
2299 	 * The metastructure is always at the bottom of the buffer.
2300 	 */
2301 	agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size -
2302 	    sizeof (dtrace_aggbuffer_t));
2303 
2304 	if (buf->dtb_offset == 0) {
2305 		/*
2306 		 * We just kludge up approximately 1/8th of the size to be
2307 		 * buckets.  If this guess ends up being routinely
2308 		 * off-the-mark, we may need to dynamically readjust this
2309 		 * based on past performance.
2310 		 */
2311 		uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t);
2312 
2313 		if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) <
2314 		    (uintptr_t)tomax || hashsize == 0) {
2315 			/*
2316 			 * We've been given a ludicrously small buffer;
2317 			 * increment our drop count and leave.
2318 			 */
2319 			dtrace_buffer_drop(buf);
2320 			return;
2321 		}
2322 
2323 		/*
2324 		 * And now, a pathetic attempt to try to get a an odd (or
2325 		 * perchance, a prime) hash size for better hash distribution.
2326 		 */
2327 		if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3))
2328 			hashsize -= DTRACE_AGGHASHSIZE_SLEW;
2329 
2330 		agb->dtagb_hashsize = hashsize;
2331 		agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb -
2332 		    agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *));
2333 		agb->dtagb_free = (uintptr_t)agb->dtagb_hash;
2334 
2335 		for (i = 0; i < agb->dtagb_hashsize; i++)
2336 			agb->dtagb_hash[i] = NULL;
2337 	}
2338 
2339 	ASSERT(agg->dtag_first != NULL);
2340 	ASSERT(agg->dtag_first->dta_intuple);
2341 
2342 	/*
2343 	 * Calculate the hash value based on the key.  Note that we _don't_
2344 	 * include the aggid in the hashing (but we will store it as part of
2345 	 * the key).  The hashing algorithm is Bob Jenkins' "One-at-a-time"
2346 	 * algorithm: a simple, quick algorithm that has no known funnels, and
2347 	 * gets good distribution in practice.  The efficacy of the hashing
2348 	 * algorithm (and a comparison with other algorithms) may be found by
2349 	 * running the ::dtrace_aggstat MDB dcmd.
2350 	 */
2351 	for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2352 		i = act->dta_rec.dtrd_offset - agg->dtag_base;
2353 		limit = i + act->dta_rec.dtrd_size;
2354 		ASSERT(limit <= size);
2355 		isstr = DTRACEACT_ISSTRING(act);
2356 
2357 		for (; i < limit; i++) {
2358 			hashval += data[i];
2359 			hashval += (hashval << 10);
2360 			hashval ^= (hashval >> 6);
2361 
2362 			if (isstr && data[i] == '\0')
2363 				break;
2364 		}
2365 	}
2366 
2367 	hashval += (hashval << 3);
2368 	hashval ^= (hashval >> 11);
2369 	hashval += (hashval << 15);
2370 
2371 	/*
2372 	 * Yes, the divide here is expensive -- but it's generally the least
2373 	 * of the performance issues given the amount of data that we iterate
2374 	 * over to compute hash values, compare data, etc.
2375 	 */
2376 	ndx = hashval % agb->dtagb_hashsize;
2377 
2378 	for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) {
2379 		ASSERT((caddr_t)key >= tomax);
2380 		ASSERT((caddr_t)key < tomax + buf->dtb_size);
2381 
2382 		if (hashval != key->dtak_hashval || key->dtak_size != size)
2383 			continue;
2384 
2385 		kdata = key->dtak_data;
2386 		ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size);
2387 
2388 		for (act = agg->dtag_first; act->dta_intuple;
2389 		    act = act->dta_next) {
2390 			i = act->dta_rec.dtrd_offset - agg->dtag_base;
2391 			limit = i + act->dta_rec.dtrd_size;
2392 			ASSERT(limit <= size);
2393 			isstr = DTRACEACT_ISSTRING(act);
2394 
2395 			for (; i < limit; i++) {
2396 				if (kdata[i] != data[i])
2397 					goto next;
2398 
2399 				if (isstr && data[i] == '\0')
2400 					break;
2401 			}
2402 		}
2403 
2404 		if (action != key->dtak_action) {
2405 			/*
2406 			 * We are aggregating on the same value in the same
2407 			 * aggregation with two different aggregating actions.
2408 			 * (This should have been picked up in the compiler,
2409 			 * so we may be dealing with errant or devious DIF.)
2410 			 * This is an error condition; we indicate as much,
2411 			 * and return.
2412 			 */
2413 			DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
2414 			return;
2415 		}
2416 
2417 		/*
2418 		 * This is a hit:  we need to apply the aggregator to
2419 		 * the value at this key.
2420 		 */
2421 		agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg);
2422 		return;
2423 next:
2424 		continue;
2425 	}
2426 
2427 	/*
2428 	 * We didn't find it.  We need to allocate some zero-filled space,
2429 	 * link it into the hash table appropriately, and apply the aggregator
2430 	 * to the (zero-filled) value.
2431 	 */
2432 	offs = buf->dtb_offset;
2433 	while (offs & (align - 1))
2434 		offs += sizeof (uint32_t);
2435 
2436 	/*
2437 	 * If we don't have enough room to both allocate a new key _and_
2438 	 * its associated data, increment the drop count and return.
2439 	 */
2440 	if ((uintptr_t)tomax + offs + fsize >
2441 	    agb->dtagb_free - sizeof (dtrace_aggkey_t)) {
2442 		dtrace_buffer_drop(buf);
2443 		return;
2444 	}
2445 
2446 	/*CONSTCOND*/
2447 	ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1)));
2448 	key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t));
2449 	agb->dtagb_free -= sizeof (dtrace_aggkey_t);
2450 
2451 	key->dtak_data = kdata = tomax + offs;
2452 	buf->dtb_offset = offs + fsize;
2453 
2454 	/*
2455 	 * Now copy the data across.
2456 	 */
2457 	*((dtrace_aggid_t *)kdata) = agg->dtag_id;
2458 
2459 	for (i = sizeof (dtrace_aggid_t); i < size; i++)
2460 		kdata[i] = data[i];
2461 
2462 	/*
2463 	 * Because strings are not zeroed out by default, we need to iterate
2464 	 * looking for actions that store strings, and we need to explicitly
2465 	 * pad these strings out with zeroes.
2466 	 */
2467 	for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2468 		int nul;
2469 
2470 		if (!DTRACEACT_ISSTRING(act))
2471 			continue;
2472 
2473 		i = act->dta_rec.dtrd_offset - agg->dtag_base;
2474 		limit = i + act->dta_rec.dtrd_size;
2475 		ASSERT(limit <= size);
2476 
2477 		for (nul = 0; i < limit; i++) {
2478 			if (nul) {
2479 				kdata[i] = '\0';
2480 				continue;
2481 			}
2482 
2483 			if (data[i] != '\0')
2484 				continue;
2485 
2486 			nul = 1;
2487 		}
2488 	}
2489 
2490 	for (i = size; i < fsize; i++)
2491 		kdata[i] = 0;
2492 
2493 	key->dtak_hashval = hashval;
2494 	key->dtak_size = size;
2495 	key->dtak_action = action;
2496 	key->dtak_next = agb->dtagb_hash[ndx];
2497 	agb->dtagb_hash[ndx] = key;
2498 
2499 	/*
2500 	 * Finally, apply the aggregator.
2501 	 */
2502 	*((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial;
2503 	agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg);
2504 }
2505 
2506 /*
2507  * Given consumer state, this routine finds a speculation in the INACTIVE
2508  * state and transitions it into the ACTIVE state.  If there is no speculation
2509  * in the INACTIVE state, 0 is returned.  In this case, no error counter is
2510  * incremented -- it is up to the caller to take appropriate action.
2511  */
2512 static int
dtrace_speculation(dtrace_state_t * state)2513 dtrace_speculation(dtrace_state_t *state)
2514 {
2515 	int i = 0;
2516 	dtrace_speculation_state_t current;
2517 	uint32_t *stat = &state->dts_speculations_unavail, count;
2518 
2519 	while (i < state->dts_nspeculations) {
2520 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2521 
2522 		current = spec->dtsp_state;
2523 
2524 		if (current != DTRACESPEC_INACTIVE) {
2525 			if (current == DTRACESPEC_COMMITTINGMANY ||
2526 			    current == DTRACESPEC_COMMITTING ||
2527 			    current == DTRACESPEC_DISCARDING)
2528 				stat = &state->dts_speculations_busy;
2529 			i++;
2530 			continue;
2531 		}
2532 
2533 		if (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2534 		    current, DTRACESPEC_ACTIVE) == current)
2535 			return (i + 1);
2536 	}
2537 
2538 	/*
2539 	 * We couldn't find a speculation.  If we found as much as a single
2540 	 * busy speculation buffer, we'll attribute this failure as "busy"
2541 	 * instead of "unavail".
2542 	 */
2543 	do {
2544 		count = *stat;
2545 	} while (dtrace_cas32(stat, count, count + 1) != count);
2546 
2547 	return (0);
2548 }
2549 
2550 /*
2551  * This routine commits an active speculation.  If the specified speculation
2552  * is not in a valid state to perform a commit(), this routine will silently do
2553  * nothing.  The state of the specified speculation is transitioned according
2554  * to the state transition diagram outlined in <sys/dtrace_impl.h>
2555  */
2556 static void
dtrace_speculation_commit(dtrace_state_t * state,processorid_t cpu,dtrace_specid_t which)2557 dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu,
2558     dtrace_specid_t which)
2559 {
2560 	dtrace_speculation_t *spec;
2561 	dtrace_buffer_t *src, *dest;
2562 	uintptr_t daddr, saddr, dlimit, slimit;
2563 	dtrace_speculation_state_t current, new;
2564 	intptr_t offs;
2565 	uint64_t timestamp;
2566 
2567 	if (which == 0)
2568 		return;
2569 
2570 	if (which > state->dts_nspeculations) {
2571 		cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2572 		return;
2573 	}
2574 
2575 	spec = &state->dts_speculations[which - 1];
2576 	src = &spec->dtsp_buffer[cpu];
2577 	dest = &state->dts_buffer[cpu];
2578 
2579 	do {
2580 		current = spec->dtsp_state;
2581 
2582 		if (current == DTRACESPEC_COMMITTINGMANY)
2583 			break;
2584 
2585 		switch (current) {
2586 		case DTRACESPEC_INACTIVE:
2587 		case DTRACESPEC_DISCARDING:
2588 			return;
2589 
2590 		case DTRACESPEC_COMMITTING:
2591 			/*
2592 			 * This is only possible if we are (a) commit()'ing
2593 			 * without having done a prior speculate() on this CPU
2594 			 * and (b) racing with another commit() on a different
2595 			 * CPU.  There's nothing to do -- we just assert that
2596 			 * our offset is 0.
2597 			 */
2598 			ASSERT(src->dtb_offset == 0);
2599 			return;
2600 
2601 		case DTRACESPEC_ACTIVE:
2602 			new = DTRACESPEC_COMMITTING;
2603 			break;
2604 
2605 		case DTRACESPEC_ACTIVEONE:
2606 			/*
2607 			 * This speculation is active on one CPU.  If our
2608 			 * buffer offset is non-zero, we know that the one CPU
2609 			 * must be us.  Otherwise, we are committing on a
2610 			 * different CPU from the speculate(), and we must
2611 			 * rely on being asynchronously cleaned.
2612 			 */
2613 			if (src->dtb_offset != 0) {
2614 				new = DTRACESPEC_COMMITTING;
2615 				break;
2616 			}
2617 			/*FALLTHROUGH*/
2618 
2619 		case DTRACESPEC_ACTIVEMANY:
2620 			new = DTRACESPEC_COMMITTINGMANY;
2621 			break;
2622 
2623 		default:
2624 			ASSERT(0);
2625 		}
2626 	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2627 	    current, new) != current);
2628 
2629 	/*
2630 	 * We have set the state to indicate that we are committing this
2631 	 * speculation.  Now reserve the necessary space in the destination
2632 	 * buffer.
2633 	 */
2634 	if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset,
2635 	    sizeof (uint64_t), state, NULL)) < 0) {
2636 		dtrace_buffer_drop(dest);
2637 		goto out;
2638 	}
2639 
2640 	/*
2641 	 * We have sufficient space to copy the speculative buffer into the
2642 	 * primary buffer.  First, modify the speculative buffer, filling
2643 	 * in the timestamp of all entries with the current time.  The data
2644 	 * must have the commit() time rather than the time it was traced,
2645 	 * so that all entries in the primary buffer are in timestamp order.
2646 	 */
2647 	timestamp = dtrace_gethrtime();
2648 	saddr = (uintptr_t)src->dtb_tomax;
2649 	slimit = saddr + src->dtb_offset;
2650 	while (saddr < slimit) {
2651 		size_t size;
2652 		dtrace_rechdr_t *dtrh = (dtrace_rechdr_t *)saddr;
2653 
2654 		if (dtrh->dtrh_epid == DTRACE_EPIDNONE) {
2655 			saddr += sizeof (dtrace_epid_t);
2656 			continue;
2657 		}
2658 		ASSERT3U(dtrh->dtrh_epid, <=, state->dts_necbs);
2659 		size = state->dts_ecbs[dtrh->dtrh_epid - 1]->dte_size;
2660 
2661 		ASSERT3U(saddr + size, <=, slimit);
2662 		ASSERT3U(size, >=, sizeof (dtrace_rechdr_t));
2663 		ASSERT3U(DTRACE_RECORD_LOAD_TIMESTAMP(dtrh), ==, UINT64_MAX);
2664 
2665 		DTRACE_RECORD_STORE_TIMESTAMP(dtrh, timestamp);
2666 
2667 		saddr += size;
2668 	}
2669 
2670 	/*
2671 	 * Copy the buffer across.  (Note that this is a
2672 	 * highly subobtimal bcopy(); in the unlikely event that this becomes
2673 	 * a serious performance issue, a high-performance DTrace-specific
2674 	 * bcopy() should obviously be invented.)
2675 	 */
2676 	daddr = (uintptr_t)dest->dtb_tomax + offs;
2677 	dlimit = daddr + src->dtb_offset;
2678 	saddr = (uintptr_t)src->dtb_tomax;
2679 
2680 	/*
2681 	 * First, the aligned portion.
2682 	 */
2683 	while (dlimit - daddr >= sizeof (uint64_t)) {
2684 		*((uint64_t *)daddr) = *((uint64_t *)saddr);
2685 
2686 		daddr += sizeof (uint64_t);
2687 		saddr += sizeof (uint64_t);
2688 	}
2689 
2690 	/*
2691 	 * Now any left-over bit...
2692 	 */
2693 	while (dlimit - daddr)
2694 		*((uint8_t *)daddr++) = *((uint8_t *)saddr++);
2695 
2696 	/*
2697 	 * Finally, commit the reserved space in the destination buffer.
2698 	 */
2699 	dest->dtb_offset = offs + src->dtb_offset;
2700 
2701 out:
2702 	/*
2703 	 * If we're lucky enough to be the only active CPU on this speculation
2704 	 * buffer, we can just set the state back to DTRACESPEC_INACTIVE.
2705 	 */
2706 	if (current == DTRACESPEC_ACTIVE ||
2707 	    (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) {
2708 		uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state,
2709 		    DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE);
2710 
2711 		ASSERT(rval == DTRACESPEC_COMMITTING);
2712 	}
2713 
2714 	src->dtb_offset = 0;
2715 	src->dtb_xamot_drops += src->dtb_drops;
2716 	src->dtb_drops = 0;
2717 }
2718 
2719 /*
2720  * This routine discards an active speculation.  If the specified speculation
2721  * is not in a valid state to perform a discard(), this routine will silently
2722  * do nothing.  The state of the specified speculation is transitioned
2723  * according to the state transition diagram outlined in <sys/dtrace_impl.h>
2724  */
2725 static void
dtrace_speculation_discard(dtrace_state_t * state,processorid_t cpu,dtrace_specid_t which)2726 dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu,
2727     dtrace_specid_t which)
2728 {
2729 	dtrace_speculation_t *spec;
2730 	dtrace_speculation_state_t current, new;
2731 	dtrace_buffer_t *buf;
2732 
2733 	if (which == 0)
2734 		return;
2735 
2736 	if (which > state->dts_nspeculations) {
2737 		cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2738 		return;
2739 	}
2740 
2741 	spec = &state->dts_speculations[which - 1];
2742 	buf = &spec->dtsp_buffer[cpu];
2743 
2744 	do {
2745 		current = spec->dtsp_state;
2746 
2747 		switch (current) {
2748 		case DTRACESPEC_INACTIVE:
2749 		case DTRACESPEC_COMMITTINGMANY:
2750 		case DTRACESPEC_COMMITTING:
2751 		case DTRACESPEC_DISCARDING:
2752 			return;
2753 
2754 		case DTRACESPEC_ACTIVE:
2755 		case DTRACESPEC_ACTIVEMANY:
2756 			new = DTRACESPEC_DISCARDING;
2757 			break;
2758 
2759 		case DTRACESPEC_ACTIVEONE:
2760 			if (buf->dtb_offset != 0) {
2761 				new = DTRACESPEC_INACTIVE;
2762 			} else {
2763 				new = DTRACESPEC_DISCARDING;
2764 			}
2765 			break;
2766 
2767 		default:
2768 			ASSERT(0);
2769 		}
2770 	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2771 	    current, new) != current);
2772 
2773 	buf->dtb_offset = 0;
2774 	buf->dtb_drops = 0;
2775 }
2776 
2777 /*
2778  * Note:  not called from probe context.  This function is called
2779  * asynchronously from cross call context to clean any speculations that are
2780  * in the COMMITTINGMANY or DISCARDING states.  These speculations may not be
2781  * transitioned back to the INACTIVE state until all CPUs have cleaned the
2782  * speculation.
2783  */
2784 static void
dtrace_speculation_clean_here(dtrace_state_t * state)2785 dtrace_speculation_clean_here(dtrace_state_t *state)
2786 {
2787 	dtrace_icookie_t cookie;
2788 	processorid_t cpu = CPU->cpu_id;
2789 	dtrace_buffer_t *dest = &state->dts_buffer[cpu];
2790 	dtrace_specid_t i;
2791 
2792 	cookie = dtrace_interrupt_disable();
2793 
2794 	if (dest->dtb_tomax == NULL) {
2795 		dtrace_interrupt_enable(cookie);
2796 		return;
2797 	}
2798 
2799 	for (i = 0; i < state->dts_nspeculations; i++) {
2800 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2801 		dtrace_buffer_t *src = &spec->dtsp_buffer[cpu];
2802 
2803 		if (src->dtb_tomax == NULL)
2804 			continue;
2805 
2806 		if (spec->dtsp_state == DTRACESPEC_DISCARDING) {
2807 			src->dtb_offset = 0;
2808 			continue;
2809 		}
2810 
2811 		if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2812 			continue;
2813 
2814 		if (src->dtb_offset == 0)
2815 			continue;
2816 
2817 		dtrace_speculation_commit(state, cpu, i + 1);
2818 	}
2819 
2820 	dtrace_interrupt_enable(cookie);
2821 }
2822 
2823 /*
2824  * Note:  not called from probe context.  This function is called
2825  * asynchronously (and at a regular interval) to clean any speculations that
2826  * are in the COMMITTINGMANY or DISCARDING states.  If it discovers that there
2827  * is work to be done, it cross calls all CPUs to perform that work;
2828  * COMMITMANY and DISCARDING speculations may not be transitioned back to the
2829  * INACTIVE state until they have been cleaned by all CPUs.
2830  */
2831 static void
dtrace_speculation_clean(dtrace_state_t * state)2832 dtrace_speculation_clean(dtrace_state_t *state)
2833 {
2834 	int work = 0, rv;
2835 	dtrace_specid_t i;
2836 
2837 	for (i = 0; i < state->dts_nspeculations; i++) {
2838 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2839 
2840 		ASSERT(!spec->dtsp_cleaning);
2841 
2842 		if (spec->dtsp_state != DTRACESPEC_DISCARDING &&
2843 		    spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2844 			continue;
2845 
2846 		work++;
2847 		spec->dtsp_cleaning = 1;
2848 	}
2849 
2850 	if (!work)
2851 		return;
2852 
2853 	dtrace_xcall(DTRACE_CPUALL,
2854 	    (dtrace_xcall_t)dtrace_speculation_clean_here, state);
2855 
2856 	/*
2857 	 * We now know that all CPUs have committed or discarded their
2858 	 * speculation buffers, as appropriate.  We can now set the state
2859 	 * to inactive.
2860 	 */
2861 	for (i = 0; i < state->dts_nspeculations; i++) {
2862 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2863 		dtrace_speculation_state_t current, new;
2864 
2865 		if (!spec->dtsp_cleaning)
2866 			continue;
2867 
2868 		current = spec->dtsp_state;
2869 		ASSERT(current == DTRACESPEC_DISCARDING ||
2870 		    current == DTRACESPEC_COMMITTINGMANY);
2871 
2872 		new = DTRACESPEC_INACTIVE;
2873 
2874 		rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new);
2875 		ASSERT(rv == current);
2876 		spec->dtsp_cleaning = 0;
2877 	}
2878 }
2879 
2880 /*
2881  * Called as part of a speculate() to get the speculative buffer associated
2882  * with a given speculation.  Returns NULL if the specified speculation is not
2883  * in an ACTIVE state.  If the speculation is in the ACTIVEONE state -- and
2884  * the active CPU is not the specified CPU -- the speculation will be
2885  * atomically transitioned into the ACTIVEMANY state.
2886  */
2887 static dtrace_buffer_t *
dtrace_speculation_buffer(dtrace_state_t * state,processorid_t cpuid,dtrace_specid_t which)2888 dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid,
2889     dtrace_specid_t which)
2890 {
2891 	dtrace_speculation_t *spec;
2892 	dtrace_speculation_state_t current, new;
2893 	dtrace_buffer_t *buf;
2894 
2895 	if (which == 0)
2896 		return (NULL);
2897 
2898 	if (which > state->dts_nspeculations) {
2899 		cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2900 		return (NULL);
2901 	}
2902 
2903 	spec = &state->dts_speculations[which - 1];
2904 	buf = &spec->dtsp_buffer[cpuid];
2905 
2906 	do {
2907 		current = spec->dtsp_state;
2908 
2909 		switch (current) {
2910 		case DTRACESPEC_INACTIVE:
2911 		case DTRACESPEC_COMMITTINGMANY:
2912 		case DTRACESPEC_DISCARDING:
2913 			return (NULL);
2914 
2915 		case DTRACESPEC_COMMITTING:
2916 			ASSERT(buf->dtb_offset == 0);
2917 			return (NULL);
2918 
2919 		case DTRACESPEC_ACTIVEONE:
2920 			/*
2921 			 * This speculation is currently active on one CPU.
2922 			 * Check the offset in the buffer; if it's non-zero,
2923 			 * that CPU must be us (and we leave the state alone).
2924 			 * If it's zero, assume that we're starting on a new
2925 			 * CPU -- and change the state to indicate that the
2926 			 * speculation is active on more than one CPU.
2927 			 */
2928 			if (buf->dtb_offset != 0)
2929 				return (buf);
2930 
2931 			new = DTRACESPEC_ACTIVEMANY;
2932 			break;
2933 
2934 		case DTRACESPEC_ACTIVEMANY:
2935 			return (buf);
2936 
2937 		case DTRACESPEC_ACTIVE:
2938 			new = DTRACESPEC_ACTIVEONE;
2939 			break;
2940 
2941 		default:
2942 			ASSERT(0);
2943 		}
2944 	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2945 	    current, new) != current);
2946 
2947 	ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY);
2948 	return (buf);
2949 }
2950 
2951 /*
2952  * Return a string.  In the event that the user lacks the privilege to access
2953  * arbitrary kernel memory, we copy the string out to scratch memory so that we
2954  * don't fail access checking.
2955  *
2956  * dtrace_dif_variable() uses this routine as a helper for various
2957  * builtin values such as 'execname' and 'probefunc.'
2958  */
2959 uintptr_t
dtrace_dif_varstr(uintptr_t addr,dtrace_state_t * state,dtrace_mstate_t * mstate)2960 dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state,
2961     dtrace_mstate_t *mstate)
2962 {
2963 	uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
2964 	uintptr_t ret;
2965 	size_t strsz;
2966 
2967 	/*
2968 	 * The easy case: this probe is allowed to read all of memory, so
2969 	 * we can just return this as a vanilla pointer.
2970 	 */
2971 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
2972 		return (addr);
2973 
2974 	/*
2975 	 * This is the tougher case: we copy the string in question from
2976 	 * kernel memory into scratch memory and return it that way: this
2977 	 * ensures that we won't trip up when access checking tests the
2978 	 * BYREF return value.
2979 	 */
2980 	strsz = dtrace_strlen((char *)addr, size) + 1;
2981 
2982 	if (mstate->dtms_scratch_ptr + strsz >
2983 	    mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
2984 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
2985 		return (NULL);
2986 	}
2987 
2988 	dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
2989 	    strsz);
2990 	ret = mstate->dtms_scratch_ptr;
2991 	mstate->dtms_scratch_ptr += strsz;
2992 	return (ret);
2993 }
2994 
2995 /*
2996  * This function implements the DIF emulator's variable lookups.  The emulator
2997  * passes a reserved variable identifier and optional built-in array index.
2998  */
2999 static uint64_t
dtrace_dif_variable(dtrace_mstate_t * mstate,dtrace_state_t * state,uint64_t v,uint64_t ndx)3000 dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v,
3001     uint64_t ndx)
3002 {
3003 	/*
3004 	 * If we're accessing one of the uncached arguments, we'll turn this
3005 	 * into a reference in the args array.
3006 	 */
3007 	if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) {
3008 		ndx = v - DIF_VAR_ARG0;
3009 		v = DIF_VAR_ARGS;
3010 	}
3011 
3012 	switch (v) {
3013 	case DIF_VAR_ARGS:
3014 		if (!(mstate->dtms_access & DTRACE_ACCESS_ARGS)) {
3015 			cpu_core[CPU->cpu_id].cpuc_dtrace_flags |=
3016 			    CPU_DTRACE_KPRIV;
3017 			return (0);
3018 		}
3019 
3020 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS);
3021 		if (ndx >= sizeof (mstate->dtms_arg) /
3022 		    sizeof (mstate->dtms_arg[0])) {
3023 			int aframes = mstate->dtms_probe->dtpr_aframes + 2;
3024 			dtrace_provider_t *pv;
3025 			uint64_t val;
3026 
3027 			pv = mstate->dtms_probe->dtpr_provider;
3028 			if (pv->dtpv_pops.dtps_getargval != NULL)
3029 				val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg,
3030 				    mstate->dtms_probe->dtpr_id,
3031 				    mstate->dtms_probe->dtpr_arg, ndx, aframes);
3032 			else
3033 				val = dtrace_getarg(ndx, aframes);
3034 
3035 			/*
3036 			 * This is regrettably required to keep the compiler
3037 			 * from tail-optimizing the call to dtrace_getarg().
3038 			 * The condition always evaluates to true, but the
3039 			 * compiler has no way of figuring that out a priori.
3040 			 * (None of this would be necessary if the compiler
3041 			 * could be relied upon to _always_ tail-optimize
3042 			 * the call to dtrace_getarg() -- but it can't.)
3043 			 */
3044 			if (mstate->dtms_probe != NULL)
3045 				return (val);
3046 
3047 			ASSERT(0);
3048 		}
3049 
3050 		return (mstate->dtms_arg[ndx]);
3051 
3052 	case DIF_VAR_UREGS: {
3053 		klwp_t *lwp;
3054 
3055 		if (!dtrace_priv_proc(state, mstate))
3056 			return (0);
3057 
3058 		if ((lwp = curthread->t_lwp) == NULL) {
3059 			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
3060 			cpu_core[CPU->cpu_id].cpuc_dtrace_illval = NULL;
3061 			return (0);
3062 		}
3063 
3064 		return (dtrace_getreg(lwp->lwp_regs, ndx));
3065 	}
3066 
3067 	case DIF_VAR_VMREGS: {
3068 		uint64_t rval;
3069 
3070 		if (!dtrace_priv_kernel(state))
3071 			return (0);
3072 
3073 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3074 
3075 		rval = dtrace_getvmreg(ndx,
3076 		    &cpu_core[CPU->cpu_id].cpuc_dtrace_flags);
3077 
3078 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3079 
3080 		return (rval);
3081 	}
3082 
3083 	case DIF_VAR_CURTHREAD:
3084 		if (!dtrace_priv_proc(state, mstate))
3085 			return (0);
3086 		return ((uint64_t)(uintptr_t)curthread);
3087 
3088 	case DIF_VAR_TIMESTAMP:
3089 		if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) {
3090 			mstate->dtms_timestamp = dtrace_gethrtime();
3091 			mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP;
3092 		}
3093 		return (mstate->dtms_timestamp);
3094 
3095 	case DIF_VAR_VTIMESTAMP:
3096 		ASSERT(dtrace_vtime_references != 0);
3097 		return (curthread->t_dtrace_vtime);
3098 
3099 	case DIF_VAR_WALLTIMESTAMP:
3100 		if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) {
3101 			mstate->dtms_walltimestamp = dtrace_gethrestime();
3102 			mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP;
3103 		}
3104 		return (mstate->dtms_walltimestamp);
3105 
3106 	case DIF_VAR_IPL:
3107 		if (!dtrace_priv_kernel(state))
3108 			return (0);
3109 		if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) {
3110 			mstate->dtms_ipl = dtrace_getipl();
3111 			mstate->dtms_present |= DTRACE_MSTATE_IPL;
3112 		}
3113 		return (mstate->dtms_ipl);
3114 
3115 	case DIF_VAR_EPID:
3116 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID);
3117 		return (mstate->dtms_epid);
3118 
3119 	case DIF_VAR_ID:
3120 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3121 		return (mstate->dtms_probe->dtpr_id);
3122 
3123 	case DIF_VAR_STACKDEPTH:
3124 		if (!dtrace_priv_kernel(state))
3125 			return (0);
3126 		if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) {
3127 			int aframes = mstate->dtms_probe->dtpr_aframes + 2;
3128 
3129 			mstate->dtms_stackdepth = dtrace_getstackdepth(aframes);
3130 			mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH;
3131 		}
3132 		return (mstate->dtms_stackdepth);
3133 
3134 	case DIF_VAR_USTACKDEPTH:
3135 		if (!dtrace_priv_proc(state, mstate))
3136 			return (0);
3137 		if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) {
3138 			/*
3139 			 * See comment in DIF_VAR_PID.
3140 			 */
3141 			if (DTRACE_ANCHORED(mstate->dtms_probe) &&
3142 			    CPU_ON_INTR(CPU)) {
3143 				mstate->dtms_ustackdepth = 0;
3144 			} else {
3145 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3146 				mstate->dtms_ustackdepth =
3147 				    dtrace_getustackdepth();
3148 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3149 			}
3150 			mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH;
3151 		}
3152 		return (mstate->dtms_ustackdepth);
3153 
3154 	case DIF_VAR_CALLER:
3155 		if (!dtrace_priv_kernel(state))
3156 			return (0);
3157 		if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) {
3158 			int aframes = mstate->dtms_probe->dtpr_aframes + 2;
3159 
3160 			if (!DTRACE_ANCHORED(mstate->dtms_probe)) {
3161 				/*
3162 				 * If this is an unanchored probe, we are
3163 				 * required to go through the slow path:
3164 				 * dtrace_caller() only guarantees correct
3165 				 * results for anchored probes.
3166 				 */
3167 				pc_t caller[2];
3168 
3169 				dtrace_getpcstack(caller, 2, aframes,
3170 				    (uint32_t *)(uintptr_t)mstate->dtms_arg[0]);
3171 				mstate->dtms_caller = caller[1];
3172 			} else if ((mstate->dtms_caller =
3173 			    dtrace_caller(aframes)) == -1) {
3174 				/*
3175 				 * We have failed to do this the quick way;
3176 				 * we must resort to the slower approach of
3177 				 * calling dtrace_getpcstack().
3178 				 */
3179 				pc_t caller;
3180 
3181 				dtrace_getpcstack(&caller, 1, aframes, NULL);
3182 				mstate->dtms_caller = caller;
3183 			}
3184 
3185 			mstate->dtms_present |= DTRACE_MSTATE_CALLER;
3186 		}
3187 		return (mstate->dtms_caller);
3188 
3189 	case DIF_VAR_UCALLER:
3190 		if (!dtrace_priv_proc(state, mstate))
3191 			return (0);
3192 
3193 		if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) {
3194 			uint64_t ustack[3];
3195 
3196 			/*
3197 			 * dtrace_getupcstack() fills in the first uint64_t
3198 			 * with the current PID.  The second uint64_t will
3199 			 * be the program counter at user-level.  The third
3200 			 * uint64_t will contain the caller, which is what
3201 			 * we're after.
3202 			 */
3203 			ustack[2] = NULL;
3204 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3205 			dtrace_getupcstack(ustack, 3);
3206 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3207 			mstate->dtms_ucaller = ustack[2];
3208 			mstate->dtms_present |= DTRACE_MSTATE_UCALLER;
3209 		}
3210 
3211 		return (mstate->dtms_ucaller);
3212 
3213 	case DIF_VAR_PROBEPROV:
3214 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3215 		return (dtrace_dif_varstr(
3216 		    (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name,
3217 		    state, mstate));
3218 
3219 	case DIF_VAR_PROBEMOD:
3220 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3221 		return (dtrace_dif_varstr(
3222 		    (uintptr_t)mstate->dtms_probe->dtpr_mod,
3223 		    state, mstate));
3224 
3225 	case DIF_VAR_PROBEFUNC:
3226 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3227 		return (dtrace_dif_varstr(
3228 		    (uintptr_t)mstate->dtms_probe->dtpr_func,
3229 		    state, mstate));
3230 
3231 	case DIF_VAR_PROBENAME:
3232 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3233 		return (dtrace_dif_varstr(
3234 		    (uintptr_t)mstate->dtms_probe->dtpr_name,
3235 		    state, mstate));
3236 
3237 	case DIF_VAR_PID:
3238 		if (!dtrace_priv_proc(state, mstate))
3239 			return (0);
3240 
3241 		/*
3242 		 * Note that we are assuming that an unanchored probe is
3243 		 * always due to a high-level interrupt.  (And we're assuming
3244 		 * that there is only a single high level interrupt.)
3245 		 */
3246 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3247 			return (pid0.pid_id);
3248 
3249 		/*
3250 		 * It is always safe to dereference one's own t_procp pointer:
3251 		 * it always points to a valid, allocated proc structure.
3252 		 * Further, it is always safe to dereference the p_pidp member
3253 		 * of one's own proc structure.  (These are truisms becuase
3254 		 * threads and processes don't clean up their own state --
3255 		 * they leave that task to whomever reaps them.)
3256 		 */
3257 		return ((uint64_t)curthread->t_procp->p_pidp->pid_id);
3258 
3259 	case DIF_VAR_PPID:
3260 		if (!dtrace_priv_proc(state, mstate))
3261 			return (0);
3262 
3263 		/*
3264 		 * See comment in DIF_VAR_PID.
3265 		 */
3266 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3267 			return (pid0.pid_id);
3268 
3269 		/*
3270 		 * It is always safe to dereference one's own t_procp pointer:
3271 		 * it always points to a valid, allocated proc structure.
3272 		 * (This is true because threads don't clean up their own
3273 		 * state -- they leave that task to whomever reaps them.)
3274 		 */
3275 		return ((uint64_t)curthread->t_procp->p_ppid);
3276 
3277 	case DIF_VAR_TID:
3278 		/*
3279 		 * See comment in DIF_VAR_PID.
3280 		 */
3281 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3282 			return (0);
3283 
3284 		return ((uint64_t)curthread->t_tid);
3285 
3286 	case DIF_VAR_EXECNAME:
3287 		if (!dtrace_priv_proc(state, mstate))
3288 			return (0);
3289 
3290 		/*
3291 		 * See comment in DIF_VAR_PID.
3292 		 */
3293 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3294 			return ((uint64_t)(uintptr_t)p0.p_user.u_comm);
3295 
3296 		/*
3297 		 * It is always safe to dereference one's own t_procp pointer:
3298 		 * it always points to a valid, allocated proc structure.
3299 		 * (This is true because threads don't clean up their own
3300 		 * state -- they leave that task to whomever reaps them.)
3301 		 */
3302 		return (dtrace_dif_varstr(
3303 		    (uintptr_t)curthread->t_procp->p_user.u_comm,
3304 		    state, mstate));
3305 
3306 	case DIF_VAR_ZONENAME:
3307 		if (!dtrace_priv_proc(state, mstate))
3308 			return (0);
3309 
3310 		/*
3311 		 * See comment in DIF_VAR_PID.
3312 		 */
3313 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3314 			return ((uint64_t)(uintptr_t)p0.p_zone->zone_name);
3315 
3316 		/*
3317 		 * It is always safe to dereference one's own t_procp pointer:
3318 		 * it always points to a valid, allocated proc structure.
3319 		 * (This is true because threads don't clean up their own
3320 		 * state -- they leave that task to whomever reaps them.)
3321 		 */
3322 		return (dtrace_dif_varstr(
3323 		    (uintptr_t)curthread->t_procp->p_zone->zone_name,
3324 		    state, mstate));
3325 
3326 	case DIF_VAR_UID:
3327 		if (!dtrace_priv_proc(state, mstate))
3328 			return (0);
3329 
3330 		/*
3331 		 * See comment in DIF_VAR_PID.
3332 		 */
3333 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3334 			return ((uint64_t)p0.p_cred->cr_uid);
3335 
3336 		/*
3337 		 * It is always safe to dereference one's own t_procp pointer:
3338 		 * it always points to a valid, allocated proc structure.
3339 		 * (This is true because threads don't clean up their own
3340 		 * state -- they leave that task to whomever reaps them.)
3341 		 *
3342 		 * Additionally, it is safe to dereference one's own process
3343 		 * credential, since this is never NULL after process birth.
3344 		 */
3345 		return ((uint64_t)curthread->t_procp->p_cred->cr_uid);
3346 
3347 	case DIF_VAR_GID:
3348 		if (!dtrace_priv_proc(state, mstate))
3349 			return (0);
3350 
3351 		/*
3352 		 * See comment in DIF_VAR_PID.
3353 		 */
3354 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3355 			return ((uint64_t)p0.p_cred->cr_gid);
3356 
3357 		/*
3358 		 * It is always safe to dereference one's own t_procp pointer:
3359 		 * it always points to a valid, allocated proc structure.
3360 		 * (This is true because threads don't clean up their own
3361 		 * state -- they leave that task to whomever reaps them.)
3362 		 *
3363 		 * Additionally, it is safe to dereference one's own process
3364 		 * credential, since this is never NULL after process birth.
3365 		 */
3366 		return ((uint64_t)curthread->t_procp->p_cred->cr_gid);
3367 
3368 	case DIF_VAR_ERRNO: {
3369 		klwp_t *lwp;
3370 		if (!dtrace_priv_proc(state, mstate))
3371 			return (0);
3372 
3373 		/*
3374 		 * See comment in DIF_VAR_PID.
3375 		 */
3376 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3377 			return (0);
3378 
3379 		/*
3380 		 * It is always safe to dereference one's own t_lwp pointer in
3381 		 * the event that this pointer is non-NULL.  (This is true
3382 		 * because threads and lwps don't clean up their own state --
3383 		 * they leave that task to whomever reaps them.)
3384 		 */
3385 		if ((lwp = curthread->t_lwp) == NULL)
3386 			return (0);
3387 
3388 		return ((uint64_t)lwp->lwp_errno);
3389 	}
3390 	default:
3391 		DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
3392 		return (0);
3393 	}
3394 }
3395 
3396 
3397 typedef enum dtrace_json_state {
3398 	DTRACE_JSON_REST = 1,
3399 	DTRACE_JSON_OBJECT,
3400 	DTRACE_JSON_STRING,
3401 	DTRACE_JSON_STRING_ESCAPE,
3402 	DTRACE_JSON_STRING_ESCAPE_UNICODE,
3403 	DTRACE_JSON_COLON,
3404 	DTRACE_JSON_COMMA,
3405 	DTRACE_JSON_VALUE,
3406 	DTRACE_JSON_IDENTIFIER,
3407 	DTRACE_JSON_NUMBER,
3408 	DTRACE_JSON_NUMBER_FRAC,
3409 	DTRACE_JSON_NUMBER_EXP,
3410 	DTRACE_JSON_COLLECT_OBJECT
3411 } dtrace_json_state_t;
3412 
3413 /*
3414  * This function possesses just enough knowledge about JSON to extract a single
3415  * value from a JSON string and store it in the scratch buffer.  It is able
3416  * to extract nested object values, and members of arrays by index.
3417  *
3418  * elemlist is a list of JSON keys, stored as packed NUL-terminated strings, to
3419  * be looked up as we descend into the object tree.  e.g.
3420  *
3421  *    foo[0].bar.baz[32] --> "foo" NUL "0" NUL "bar" NUL "baz" NUL "32" NUL
3422  *       with nelems = 5.
3423  *
3424  * The run time of this function must be bounded above by strsize to limit the
3425  * amount of work done in probe context.  As such, it is implemented as a
3426  * simple state machine, reading one character at a time using safe loads
3427  * until we find the requested element, hit a parsing error or run off the
3428  * end of the object or string.
3429  *
3430  * As there is no way for a subroutine to return an error without interrupting
3431  * clause execution, we simply return NULL in the event of a missing key or any
3432  * other error condition.  Each NULL return in this function is commented with
3433  * the error condition it represents -- parsing or otherwise.
3434  *
3435  * The set of states for the state machine closely matches the JSON
3436  * specification (http://json.org/).  Briefly:
3437  *
3438  *   DTRACE_JSON_REST:
3439  *     Skip whitespace until we find either a top-level Object, moving
3440  *     to DTRACE_JSON_OBJECT; or an Array, moving to DTRACE_JSON_VALUE.
3441  *
3442  *   DTRACE_JSON_OBJECT:
3443  *     Locate the next key String in an Object.  Sets a flag to denote
3444  *     the next String as a key string and moves to DTRACE_JSON_STRING.
3445  *
3446  *   DTRACE_JSON_COLON:
3447  *     Skip whitespace until we find the colon that separates key Strings
3448  *     from their values.  Once found, move to DTRACE_JSON_VALUE.
3449  *
3450  *   DTRACE_JSON_VALUE:
3451  *     Detects the type of the next value (String, Number, Identifier, Object
3452  *     or Array) and routes to the states that process that type.  Here we also
3453  *     deal with the element selector list if we are requested to traverse down
3454  *     into the object tree.
3455  *
3456  *   DTRACE_JSON_COMMA:
3457  *     Skip whitespace until we find the comma that separates key-value pairs
3458  *     in Objects (returning to DTRACE_JSON_OBJECT) or values in Arrays
3459  *     (similarly DTRACE_JSON_VALUE).  All following literal value processing
3460  *     states return to this state at the end of their value, unless otherwise
3461  *     noted.
3462  *
3463  *   DTRACE_JSON_NUMBER, DTRACE_JSON_NUMBER_FRAC, DTRACE_JSON_NUMBER_EXP:
3464  *     Processes a Number literal from the JSON, including any exponent
3465  *     component that may be present.  Numbers are returned as strings, which
3466  *     may be passed to strtoll() if an integer is required.
3467  *
3468  *   DTRACE_JSON_IDENTIFIER:
3469  *     Processes a "true", "false" or "null" literal in the JSON.
3470  *
3471  *   DTRACE_JSON_STRING, DTRACE_JSON_STRING_ESCAPE,
3472  *   DTRACE_JSON_STRING_ESCAPE_UNICODE:
3473  *     Processes a String literal from the JSON, whether the String denotes
3474  *     a key, a value or part of a larger Object.  Handles all escape sequences
3475  *     present in the specification, including four-digit unicode characters,
3476  *     but merely includes the escape sequence without converting it to the
3477  *     actual escaped character.  If the String is flagged as a key, we
3478  *     move to DTRACE_JSON_COLON rather than DTRACE_JSON_COMMA.
3479  *
3480  *   DTRACE_JSON_COLLECT_OBJECT:
3481  *     This state collects an entire Object (or Array), correctly handling
3482  *     embedded strings.  If the full element selector list matches this nested
3483  *     object, we return the Object in full as a string.  If not, we use this
3484  *     state to skip to the next value at this level and continue processing.
3485  *
3486  * NOTE: This function uses various macros from strtolctype.h to manipulate
3487  * digit values, etc -- these have all been checked to ensure they make
3488  * no additional function calls.
3489  */
3490 static char *
dtrace_json(uint64_t size,uintptr_t json,char * elemlist,int nelems,char * dest)3491 dtrace_json(uint64_t size, uintptr_t json, char *elemlist, int nelems,
3492     char *dest)
3493 {
3494 	dtrace_json_state_t state = DTRACE_JSON_REST;
3495 	int64_t array_elem = INT64_MIN;
3496 	int64_t array_pos = 0;
3497 	uint8_t escape_unicount = 0;
3498 	boolean_t string_is_key = B_FALSE;
3499 	boolean_t collect_object = B_FALSE;
3500 	boolean_t found_key = B_FALSE;
3501 	boolean_t in_array = B_FALSE;
3502 	uint32_t braces = 0, brackets = 0;
3503 	char *elem = elemlist;
3504 	char *dd = dest;
3505 	uintptr_t cur;
3506 
3507 	for (cur = json; cur < json + size; cur++) {
3508 		char cc = dtrace_load8(cur);
3509 		if (cc == '\0')
3510 			return (NULL);
3511 
3512 		switch (state) {
3513 		case DTRACE_JSON_REST:
3514 			if (isspace(cc))
3515 				break;
3516 
3517 			if (cc == '{') {
3518 				state = DTRACE_JSON_OBJECT;
3519 				break;
3520 			}
3521 
3522 			if (cc == '[') {
3523 				in_array = B_TRUE;
3524 				array_pos = 0;
3525 				array_elem = dtrace_strtoll(elem, 10, size);
3526 				found_key = array_elem == 0 ? B_TRUE : B_FALSE;
3527 				state = DTRACE_JSON_VALUE;
3528 				break;
3529 			}
3530 
3531 			/*
3532 			 * ERROR: expected to find a top-level object or array.
3533 			 */
3534 			return (NULL);
3535 		case DTRACE_JSON_OBJECT:
3536 			if (isspace(cc))
3537 				break;
3538 
3539 			if (cc == '"') {
3540 				state = DTRACE_JSON_STRING;
3541 				string_is_key = B_TRUE;
3542 				break;
3543 			}
3544 
3545 			/*
3546 			 * ERROR: either the object did not start with a key
3547 			 * string, or we've run off the end of the object
3548 			 * without finding the requested key.
3549 			 */
3550 			return (NULL);
3551 		case DTRACE_JSON_STRING:
3552 			if (cc == '\\') {
3553 				*dd++ = '\\';
3554 				state = DTRACE_JSON_STRING_ESCAPE;
3555 				break;
3556 			}
3557 
3558 			if (cc == '"') {
3559 				if (collect_object) {
3560 					/*
3561 					 * We don't reset the dest here, as
3562 					 * the string is part of a larger
3563 					 * object being collected.
3564 					 */
3565 					*dd++ = cc;
3566 					collect_object = B_FALSE;
3567 					state = DTRACE_JSON_COLLECT_OBJECT;
3568 					break;
3569 				}
3570 				*dd = '\0';
3571 				dd = dest; /* reset string buffer */
3572 				if (string_is_key) {
3573 					if (dtrace_strncmp(dest, elem,
3574 					    size) == 0)
3575 						found_key = B_TRUE;
3576 				} else if (found_key) {
3577 					if (nelems > 1) {
3578 						/*
3579 						 * We expected an object, not
3580 						 * this string.
3581 						 */
3582 						return (NULL);
3583 					}
3584 					return (dest);
3585 				}
3586 				state = string_is_key ? DTRACE_JSON_COLON :
3587 				    DTRACE_JSON_COMMA;
3588 				string_is_key = B_FALSE;
3589 				break;
3590 			}
3591 
3592 			*dd++ = cc;
3593 			break;
3594 		case DTRACE_JSON_STRING_ESCAPE:
3595 			*dd++ = cc;
3596 			if (cc == 'u') {
3597 				escape_unicount = 0;
3598 				state = DTRACE_JSON_STRING_ESCAPE_UNICODE;
3599 			} else {
3600 				state = DTRACE_JSON_STRING;
3601 			}
3602 			break;
3603 		case DTRACE_JSON_STRING_ESCAPE_UNICODE:
3604 			if (!isxdigit(cc)) {
3605 				/*
3606 				 * ERROR: invalid unicode escape, expected
3607 				 * four valid hexidecimal digits.
3608 				 */
3609 				return (NULL);
3610 			}
3611 
3612 			*dd++ = cc;
3613 			if (++escape_unicount == 4)
3614 				state = DTRACE_JSON_STRING;
3615 			break;
3616 		case DTRACE_JSON_COLON:
3617 			if (isspace(cc))
3618 				break;
3619 
3620 			if (cc == ':') {
3621 				state = DTRACE_JSON_VALUE;
3622 				break;
3623 			}
3624 
3625 			/*
3626 			 * ERROR: expected a colon.
3627 			 */
3628 			return (NULL);
3629 		case DTRACE_JSON_COMMA:
3630 			if (isspace(cc))
3631 				break;
3632 
3633 			if (cc == ',') {
3634 				if (in_array) {
3635 					state = DTRACE_JSON_VALUE;
3636 					if (++array_pos == array_elem)
3637 						found_key = B_TRUE;
3638 				} else {
3639 					state = DTRACE_JSON_OBJECT;
3640 				}
3641 				break;
3642 			}
3643 
3644 			/*
3645 			 * ERROR: either we hit an unexpected character, or
3646 			 * we reached the end of the object or array without
3647 			 * finding the requested key.
3648 			 */
3649 			return (NULL);
3650 		case DTRACE_JSON_IDENTIFIER:
3651 			if (islower(cc)) {
3652 				*dd++ = cc;
3653 				break;
3654 			}
3655 
3656 			*dd = '\0';
3657 			dd = dest; /* reset string buffer */
3658 
3659 			if (dtrace_strncmp(dest, "true", 5) == 0 ||
3660 			    dtrace_strncmp(dest, "false", 6) == 0 ||
3661 			    dtrace_strncmp(dest, "null", 5) == 0) {
3662 				if (found_key) {
3663 					if (nelems > 1) {
3664 						/*
3665 						 * ERROR: We expected an object,
3666 						 * not this identifier.
3667 						 */
3668 						return (NULL);
3669 					}
3670 					return (dest);
3671 				} else {
3672 					cur--;
3673 					state = DTRACE_JSON_COMMA;
3674 					break;
3675 				}
3676 			}
3677 
3678 			/*
3679 			 * ERROR: we did not recognise the identifier as one
3680 			 * of those in the JSON specification.
3681 			 */
3682 			return (NULL);
3683 		case DTRACE_JSON_NUMBER:
3684 			if (cc == '.') {
3685 				*dd++ = cc;
3686 				state = DTRACE_JSON_NUMBER_FRAC;
3687 				break;
3688 			}
3689 
3690 			if (cc == 'x' || cc == 'X') {
3691 				/*
3692 				 * ERROR: specification explicitly excludes
3693 				 * hexidecimal or octal numbers.
3694 				 */
3695 				return (NULL);
3696 			}
3697 
3698 			/* FALLTHRU */
3699 		case DTRACE_JSON_NUMBER_FRAC:
3700 			if (cc == 'e' || cc == 'E') {
3701 				*dd++ = cc;
3702 				state = DTRACE_JSON_NUMBER_EXP;
3703 				break;
3704 			}
3705 
3706 			if (cc == '+' || cc == '-') {
3707 				/*
3708 				 * ERROR: expect sign as part of exponent only.
3709 				 */
3710 				return (NULL);
3711 			}
3712 			/* FALLTHRU */
3713 		case DTRACE_JSON_NUMBER_EXP:
3714 			if (isdigit(cc) || cc == '+' || cc == '-') {
3715 				*dd++ = cc;
3716 				break;
3717 			}
3718 
3719 			*dd = '\0';
3720 			dd = dest; /* reset string buffer */
3721 			if (found_key) {
3722 				if (nelems > 1) {
3723 					/*
3724 					 * ERROR: We expected an object, not
3725 					 * this number.
3726 					 */
3727 					return (NULL);
3728 				}
3729 				return (dest);
3730 			}
3731 
3732 			cur--;
3733 			state = DTRACE_JSON_COMMA;
3734 			break;
3735 		case DTRACE_JSON_VALUE:
3736 			if (isspace(cc))
3737 				break;
3738 
3739 			if (cc == '{' || cc == '[') {
3740 				if (nelems > 1 && found_key) {
3741 					in_array = cc == '[' ? B_TRUE : B_FALSE;
3742 					/*
3743 					 * If our element selector directs us
3744 					 * to descend into this nested object,
3745 					 * then move to the next selector
3746 					 * element in the list and restart the
3747 					 * state machine.
3748 					 */
3749 					while (*elem != '\0')
3750 						elem++;
3751 					elem++; /* skip the inter-element NUL */
3752 					nelems--;
3753 					dd = dest;
3754 					if (in_array) {
3755 						state = DTRACE_JSON_VALUE;
3756 						array_pos = 0;
3757 						array_elem = dtrace_strtoll(
3758 						    elem, 10, size);
3759 						found_key = array_elem == 0 ?
3760 						    B_TRUE : B_FALSE;
3761 					} else {
3762 						found_key = B_FALSE;
3763 						state = DTRACE_JSON_OBJECT;
3764 					}
3765 					break;
3766 				}
3767 
3768 				/*
3769 				 * Otherwise, we wish to either skip this
3770 				 * nested object or return it in full.
3771 				 */
3772 				if (cc == '[')
3773 					brackets = 1;
3774 				else
3775 					braces = 1;
3776 				*dd++ = cc;
3777 				state = DTRACE_JSON_COLLECT_OBJECT;
3778 				break;
3779 			}
3780 
3781 			if (cc == '"') {
3782 				state = DTRACE_JSON_STRING;
3783 				break;
3784 			}
3785 
3786 			if (islower(cc)) {
3787 				/*
3788 				 * Here we deal with true, false and null.
3789 				 */
3790 				*dd++ = cc;
3791 				state = DTRACE_JSON_IDENTIFIER;
3792 				break;
3793 			}
3794 
3795 			if (cc == '-' || isdigit(cc)) {
3796 				*dd++ = cc;
3797 				state = DTRACE_JSON_NUMBER;
3798 				break;
3799 			}
3800 
3801 			/*
3802 			 * ERROR: unexpected character at start of value.
3803 			 */
3804 			return (NULL);
3805 		case DTRACE_JSON_COLLECT_OBJECT:
3806 			if (cc == '\0')
3807 				/*
3808 				 * ERROR: unexpected end of input.
3809 				 */
3810 				return (NULL);
3811 
3812 			*dd++ = cc;
3813 			if (cc == '"') {
3814 				collect_object = B_TRUE;
3815 				state = DTRACE_JSON_STRING;
3816 				break;
3817 			}
3818 
3819 			if (cc == ']') {
3820 				if (brackets-- == 0) {
3821 					/*
3822 					 * ERROR: unbalanced brackets.
3823 					 */
3824 					return (NULL);
3825 				}
3826 			} else if (cc == '}') {
3827 				if (braces-- == 0) {
3828 					/*
3829 					 * ERROR: unbalanced braces.
3830 					 */
3831 					return (NULL);
3832 				}
3833 			} else if (cc == '{') {
3834 				braces++;
3835 			} else if (cc == '[') {
3836 				brackets++;
3837 			}
3838 
3839 			if (brackets == 0 && braces == 0) {
3840 				if (found_key) {
3841 					*dd = '\0';
3842 					return (dest);
3843 				}
3844 				dd = dest; /* reset string buffer */
3845 				state = DTRACE_JSON_COMMA;
3846 			}
3847 			break;
3848 		}
3849 	}
3850 	return (NULL);
3851 }
3852 
3853 /*
3854  * Emulate the execution of DTrace ID subroutines invoked by the call opcode.
3855  * Notice that we don't bother validating the proper number of arguments or
3856  * their types in the tuple stack.  This isn't needed because all argument
3857  * interpretation is safe because of our load safety -- the worst that can
3858  * happen is that a bogus program can obtain bogus results.
3859  */
3860 static void
dtrace_dif_subr(uint_t subr,uint_t rd,uint64_t * regs,dtrace_key_t * tupregs,int nargs,dtrace_mstate_t * mstate,dtrace_state_t * state)3861 dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs,
3862     dtrace_key_t *tupregs, int nargs,
3863     dtrace_mstate_t *mstate, dtrace_state_t *state)
3864 {
3865 	volatile uint16_t *flags = &cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
3866 	volatile uintptr_t *illval = &cpu_core[CPU->cpu_id].cpuc_dtrace_illval;
3867 	dtrace_vstate_t *vstate = &state->dts_vstate;
3868 
3869 	union {
3870 		mutex_impl_t mi;
3871 		uint64_t mx;
3872 	} m;
3873 
3874 	union {
3875 		krwlock_t ri;
3876 		uintptr_t rw;
3877 	} r;
3878 
3879 	switch (subr) {
3880 	case DIF_SUBR_RAND:
3881 		regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875;
3882 		break;
3883 
3884 	case DIF_SUBR_MUTEX_OWNED:
3885 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3886 		    mstate, vstate)) {
3887 			regs[rd] = NULL;
3888 			break;
3889 		}
3890 
3891 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3892 		if (MUTEX_TYPE_ADAPTIVE(&m.mi))
3893 			regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER;
3894 		else
3895 			regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock);
3896 		break;
3897 
3898 	case DIF_SUBR_MUTEX_OWNER:
3899 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3900 		    mstate, vstate)) {
3901 			regs[rd] = NULL;
3902 			break;
3903 		}
3904 
3905 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3906 		if (MUTEX_TYPE_ADAPTIVE(&m.mi) &&
3907 		    MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER)
3908 			regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi);
3909 		else
3910 			regs[rd] = 0;
3911 		break;
3912 
3913 	case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
3914 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3915 		    mstate, vstate)) {
3916 			regs[rd] = NULL;
3917 			break;
3918 		}
3919 
3920 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3921 		regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi);
3922 		break;
3923 
3924 	case DIF_SUBR_MUTEX_TYPE_SPIN:
3925 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3926 		    mstate, vstate)) {
3927 			regs[rd] = NULL;
3928 			break;
3929 		}
3930 
3931 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3932 		regs[rd] = MUTEX_TYPE_SPIN(&m.mi);
3933 		break;
3934 
3935 	case DIF_SUBR_RW_READ_HELD: {
3936 		uintptr_t tmp;
3937 
3938 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3939 		    mstate, vstate)) {
3940 			regs[rd] = NULL;
3941 			break;
3942 		}
3943 
3944 		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3945 		regs[rd] = _RW_READ_HELD(&r.ri, tmp);
3946 		break;
3947 	}
3948 
3949 	case DIF_SUBR_RW_WRITE_HELD:
3950 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3951 		    mstate, vstate)) {
3952 			regs[rd] = NULL;
3953 			break;
3954 		}
3955 
3956 		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3957 		regs[rd] = _RW_WRITE_HELD(&r.ri);
3958 		break;
3959 
3960 	case DIF_SUBR_RW_ISWRITER:
3961 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3962 		    mstate, vstate)) {
3963 			regs[rd] = NULL;
3964 			break;
3965 		}
3966 
3967 		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3968 		regs[rd] = _RW_ISWRITER(&r.ri);
3969 		break;
3970 
3971 	case DIF_SUBR_BCOPY: {
3972 		/*
3973 		 * We need to be sure that the destination is in the scratch
3974 		 * region -- no other region is allowed.
3975 		 */
3976 		uintptr_t src = tupregs[0].dttk_value;
3977 		uintptr_t dest = tupregs[1].dttk_value;
3978 		size_t size = tupregs[2].dttk_value;
3979 
3980 		if (!dtrace_inscratch(dest, size, mstate)) {
3981 			*flags |= CPU_DTRACE_BADADDR;
3982 			*illval = regs[rd];
3983 			break;
3984 		}
3985 
3986 		if (!dtrace_canload(src, size, mstate, vstate)) {
3987 			regs[rd] = NULL;
3988 			break;
3989 		}
3990 
3991 		dtrace_bcopy((void *)src, (void *)dest, size);
3992 		break;
3993 	}
3994 
3995 	case DIF_SUBR_ALLOCA:
3996 	case DIF_SUBR_COPYIN: {
3997 		uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
3998 		uint64_t size =
3999 		    tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value;
4000 		size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size;
4001 
4002 		/*
4003 		 * This action doesn't require any credential checks since
4004 		 * probes will not activate in user contexts to which the
4005 		 * enabling user does not have permissions.
4006 		 */
4007 
4008 		/*
4009 		 * Rounding up the user allocation size could have overflowed
4010 		 * a large, bogus allocation (like -1ULL) to 0.
4011 		 */
4012 		if (scratch_size < size ||
4013 		    !DTRACE_INSCRATCH(mstate, scratch_size)) {
4014 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4015 			regs[rd] = NULL;
4016 			break;
4017 		}
4018 
4019 		if (subr == DIF_SUBR_COPYIN) {
4020 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
4021 			dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
4022 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
4023 		}
4024 
4025 		mstate->dtms_scratch_ptr += scratch_size;
4026 		regs[rd] = dest;
4027 		break;
4028 	}
4029 
4030 	case DIF_SUBR_COPYINTO: {
4031 		uint64_t size = tupregs[1].dttk_value;
4032 		uintptr_t dest = tupregs[2].dttk_value;
4033 
4034 		/*
4035 		 * This action doesn't require any credential checks since
4036 		 * probes will not activate in user contexts to which the
4037 		 * enabling user does not have permissions.
4038 		 */
4039 		if (!dtrace_inscratch(dest, size, mstate)) {
4040 			*flags |= CPU_DTRACE_BADADDR;
4041 			*illval = regs[rd];
4042 			break;
4043 		}
4044 
4045 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
4046 		dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
4047 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
4048 		break;
4049 	}
4050 
4051 	case DIF_SUBR_COPYINSTR: {
4052 		uintptr_t dest = mstate->dtms_scratch_ptr;
4053 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4054 
4055 		if (nargs > 1 && tupregs[1].dttk_value < size)
4056 			size = tupregs[1].dttk_value + 1;
4057 
4058 		/*
4059 		 * This action doesn't require any credential checks since
4060 		 * probes will not activate in user contexts to which the
4061 		 * enabling user does not have permissions.
4062 		 */
4063 		if (!DTRACE_INSCRATCH(mstate, size)) {
4064 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4065 			regs[rd] = NULL;
4066 			break;
4067 		}
4068 
4069 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
4070 		dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags);
4071 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
4072 
4073 		((char *)dest)[size - 1] = '\0';
4074 		mstate->dtms_scratch_ptr += size;
4075 		regs[rd] = dest;
4076 		break;
4077 	}
4078 
4079 	case DIF_SUBR_MSGSIZE:
4080 	case DIF_SUBR_MSGDSIZE: {
4081 		uintptr_t baddr = tupregs[0].dttk_value, daddr;
4082 		uintptr_t wptr, rptr;
4083 		size_t count = 0;
4084 		int cont = 0;
4085 
4086 		while (baddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4087 
4088 			if (!dtrace_canload(baddr, sizeof (mblk_t), mstate,
4089 			    vstate)) {
4090 				regs[rd] = NULL;
4091 				break;
4092 			}
4093 
4094 			wptr = dtrace_loadptr(baddr +
4095 			    offsetof(mblk_t, b_wptr));
4096 
4097 			rptr = dtrace_loadptr(baddr +
4098 			    offsetof(mblk_t, b_rptr));
4099 
4100 			if (wptr < rptr) {
4101 				*flags |= CPU_DTRACE_BADADDR;
4102 				*illval = tupregs[0].dttk_value;
4103 				break;
4104 			}
4105 
4106 			daddr = dtrace_loadptr(baddr +
4107 			    offsetof(mblk_t, b_datap));
4108 
4109 			baddr = dtrace_loadptr(baddr +
4110 			    offsetof(mblk_t, b_cont));
4111 
4112 			/*
4113 			 * We want to prevent against denial-of-service here,
4114 			 * so we're only going to search the list for
4115 			 * dtrace_msgdsize_max mblks.
4116 			 */
4117 			if (cont++ > dtrace_msgdsize_max) {
4118 				*flags |= CPU_DTRACE_ILLOP;
4119 				break;
4120 			}
4121 
4122 			if (subr == DIF_SUBR_MSGDSIZE) {
4123 				if (dtrace_load8(daddr +
4124 				    offsetof(dblk_t, db_type)) != M_DATA)
4125 					continue;
4126 			}
4127 
4128 			count += wptr - rptr;
4129 		}
4130 
4131 		if (!(*flags & CPU_DTRACE_FAULT))
4132 			regs[rd] = count;
4133 
4134 		break;
4135 	}
4136 
4137 	case DIF_SUBR_PROGENYOF: {
4138 		pid_t pid = tupregs[0].dttk_value;
4139 		proc_t *p;
4140 		int rval = 0;
4141 
4142 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
4143 
4144 		for (p = curthread->t_procp; p != NULL; p = p->p_parent) {
4145 			if (p->p_pidp->pid_id == pid) {
4146 				rval = 1;
4147 				break;
4148 			}
4149 		}
4150 
4151 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
4152 
4153 		regs[rd] = rval;
4154 		break;
4155 	}
4156 
4157 	case DIF_SUBR_SPECULATION:
4158 		regs[rd] = dtrace_speculation(state);
4159 		break;
4160 
4161 	case DIF_SUBR_COPYOUT: {
4162 		uintptr_t kaddr = tupregs[0].dttk_value;
4163 		uintptr_t uaddr = tupregs[1].dttk_value;
4164 		uint64_t size = tupregs[2].dttk_value;
4165 
4166 		if (!dtrace_destructive_disallow &&
4167 		    dtrace_priv_proc_control(state, mstate) &&
4168 		    !dtrace_istoxic(kaddr, size)) {
4169 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
4170 			dtrace_copyout(kaddr, uaddr, size, flags);
4171 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
4172 		}
4173 		break;
4174 	}
4175 
4176 	case DIF_SUBR_COPYOUTSTR: {
4177 		uintptr_t kaddr = tupregs[0].dttk_value;
4178 		uintptr_t uaddr = tupregs[1].dttk_value;
4179 		uint64_t size = tupregs[2].dttk_value;
4180 
4181 		if (!dtrace_destructive_disallow &&
4182 		    dtrace_priv_proc_control(state, mstate) &&
4183 		    !dtrace_istoxic(kaddr, size)) {
4184 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
4185 			dtrace_copyoutstr(kaddr, uaddr, size, flags);
4186 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
4187 		}
4188 		break;
4189 	}
4190 
4191 	case DIF_SUBR_STRLEN: {
4192 		size_t sz;
4193 		uintptr_t addr = (uintptr_t)tupregs[0].dttk_value;
4194 		sz = dtrace_strlen((char *)addr,
4195 		    state->dts_options[DTRACEOPT_STRSIZE]);
4196 
4197 		if (!dtrace_canload(addr, sz + 1, mstate, vstate)) {
4198 			regs[rd] = NULL;
4199 			break;
4200 		}
4201 
4202 		regs[rd] = sz;
4203 
4204 		break;
4205 	}
4206 
4207 	case DIF_SUBR_STRCHR:
4208 	case DIF_SUBR_STRRCHR: {
4209 		/*
4210 		 * We're going to iterate over the string looking for the
4211 		 * specified character.  We will iterate until we have reached
4212 		 * the string length or we have found the character.  If this
4213 		 * is DIF_SUBR_STRRCHR, we will look for the last occurrence
4214 		 * of the specified character instead of the first.
4215 		 */
4216 		uintptr_t saddr = tupregs[0].dttk_value;
4217 		uintptr_t addr = tupregs[0].dttk_value;
4218 		uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE];
4219 		char c, target = (char)tupregs[1].dttk_value;
4220 
4221 		for (regs[rd] = NULL; addr < limit; addr++) {
4222 			if ((c = dtrace_load8(addr)) == target) {
4223 				regs[rd] = addr;
4224 
4225 				if (subr == DIF_SUBR_STRCHR)
4226 					break;
4227 			}
4228 
4229 			if (c == '\0')
4230 				break;
4231 		}
4232 
4233 		if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) {
4234 			regs[rd] = NULL;
4235 			break;
4236 		}
4237 
4238 		break;
4239 	}
4240 
4241 	case DIF_SUBR_STRSTR:
4242 	case DIF_SUBR_INDEX:
4243 	case DIF_SUBR_RINDEX: {
4244 		/*
4245 		 * We're going to iterate over the string looking for the
4246 		 * specified string.  We will iterate until we have reached
4247 		 * the string length or we have found the string.  (Yes, this
4248 		 * is done in the most naive way possible -- but considering
4249 		 * that the string we're searching for is likely to be
4250 		 * relatively short, the complexity of Rabin-Karp or similar
4251 		 * hardly seems merited.)
4252 		 */
4253 		char *addr = (char *)(uintptr_t)tupregs[0].dttk_value;
4254 		char *substr = (char *)(uintptr_t)tupregs[1].dttk_value;
4255 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4256 		size_t len = dtrace_strlen(addr, size);
4257 		size_t sublen = dtrace_strlen(substr, size);
4258 		char *limit = addr + len, *orig = addr;
4259 		int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1;
4260 		int inc = 1;
4261 
4262 		regs[rd] = notfound;
4263 
4264 		if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) {
4265 			regs[rd] = NULL;
4266 			break;
4267 		}
4268 
4269 		if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate,
4270 		    vstate)) {
4271 			regs[rd] = NULL;
4272 			break;
4273 		}
4274 
4275 		/*
4276 		 * strstr() and index()/rindex() have similar semantics if
4277 		 * both strings are the empty string: strstr() returns a
4278 		 * pointer to the (empty) string, and index() and rindex()
4279 		 * both return index 0 (regardless of any position argument).
4280 		 */
4281 		if (sublen == 0 && len == 0) {
4282 			if (subr == DIF_SUBR_STRSTR)
4283 				regs[rd] = (uintptr_t)addr;
4284 			else
4285 				regs[rd] = 0;
4286 			break;
4287 		}
4288 
4289 		if (subr != DIF_SUBR_STRSTR) {
4290 			if (subr == DIF_SUBR_RINDEX) {
4291 				limit = orig - 1;
4292 				addr += len;
4293 				inc = -1;
4294 			}
4295 
4296 			/*
4297 			 * Both index() and rindex() take an optional position
4298 			 * argument that denotes the starting position.
4299 			 */
4300 			if (nargs == 3) {
4301 				int64_t pos = (int64_t)tupregs[2].dttk_value;
4302 
4303 				/*
4304 				 * If the position argument to index() is
4305 				 * negative, Perl implicitly clamps it at
4306 				 * zero.  This semantic is a little surprising
4307 				 * given the special meaning of negative
4308 				 * positions to similar Perl functions like
4309 				 * substr(), but it appears to reflect a
4310 				 * notion that index() can start from a
4311 				 * negative index and increment its way up to
4312 				 * the string.  Given this notion, Perl's
4313 				 * rindex() is at least self-consistent in
4314 				 * that it implicitly clamps positions greater
4315 				 * than the string length to be the string
4316 				 * length.  Where Perl completely loses
4317 				 * coherence, however, is when the specified
4318 				 * substring is the empty string ("").  In
4319 				 * this case, even if the position is
4320 				 * negative, rindex() returns 0 -- and even if
4321 				 * the position is greater than the length,
4322 				 * index() returns the string length.  These
4323 				 * semantics violate the notion that index()
4324 				 * should never return a value less than the
4325 				 * specified position and that rindex() should
4326 				 * never return a value greater than the
4327 				 * specified position.  (One assumes that
4328 				 * these semantics are artifacts of Perl's
4329 				 * implementation and not the results of
4330 				 * deliberate design -- it beggars belief that
4331 				 * even Larry Wall could desire such oddness.)
4332 				 * While in the abstract one would wish for
4333 				 * consistent position semantics across
4334 				 * substr(), index() and rindex() -- or at the
4335 				 * very least self-consistent position
4336 				 * semantics for index() and rindex() -- we
4337 				 * instead opt to keep with the extant Perl
4338 				 * semantics, in all their broken glory.  (Do
4339 				 * we have more desire to maintain Perl's
4340 				 * semantics than Perl does?  Probably.)
4341 				 */
4342 				if (subr == DIF_SUBR_RINDEX) {
4343 					if (pos < 0) {
4344 						if (sublen == 0)
4345 							regs[rd] = 0;
4346 						break;
4347 					}
4348 
4349 					if (pos > len)
4350 						pos = len;
4351 				} else {
4352 					if (pos < 0)
4353 						pos = 0;
4354 
4355 					if (pos >= len) {
4356 						if (sublen == 0)
4357 							regs[rd] = len;
4358 						break;
4359 					}
4360 				}
4361 
4362 				addr = orig + pos;
4363 			}
4364 		}
4365 
4366 		for (regs[rd] = notfound; addr != limit; addr += inc) {
4367 			if (dtrace_strncmp(addr, substr, sublen) == 0) {
4368 				if (subr != DIF_SUBR_STRSTR) {
4369 					/*
4370 					 * As D index() and rindex() are
4371 					 * modeled on Perl (and not on awk),
4372 					 * we return a zero-based (and not a
4373 					 * one-based) index.  (For you Perl
4374 					 * weenies: no, we're not going to add
4375 					 * $[ -- and shouldn't you be at a con
4376 					 * or something?)
4377 					 */
4378 					regs[rd] = (uintptr_t)(addr - orig);
4379 					break;
4380 				}
4381 
4382 				ASSERT(subr == DIF_SUBR_STRSTR);
4383 				regs[rd] = (uintptr_t)addr;
4384 				break;
4385 			}
4386 		}
4387 
4388 		break;
4389 	}
4390 
4391 	case DIF_SUBR_STRTOK: {
4392 		uintptr_t addr = tupregs[0].dttk_value;
4393 		uintptr_t tokaddr = tupregs[1].dttk_value;
4394 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4395 		uintptr_t limit, toklimit = tokaddr + size;
4396 		uint8_t c, tokmap[32];	 /* 256 / 8 */
4397 		char *dest = (char *)mstate->dtms_scratch_ptr;
4398 		int i;
4399 
4400 		/*
4401 		 * Check both the token buffer and (later) the input buffer,
4402 		 * since both could be non-scratch addresses.
4403 		 */
4404 		if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) {
4405 			regs[rd] = NULL;
4406 			break;
4407 		}
4408 
4409 		if (!DTRACE_INSCRATCH(mstate, size)) {
4410 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4411 			regs[rd] = NULL;
4412 			break;
4413 		}
4414 
4415 		if (addr == NULL) {
4416 			/*
4417 			 * If the address specified is NULL, we use our saved
4418 			 * strtok pointer from the mstate.  Note that this
4419 			 * means that the saved strtok pointer is _only_
4420 			 * valid within multiple enablings of the same probe --
4421 			 * it behaves like an implicit clause-local variable.
4422 			 */
4423 			addr = mstate->dtms_strtok;
4424 		} else {
4425 			/*
4426 			 * If the user-specified address is non-NULL we must
4427 			 * access check it.  This is the only time we have
4428 			 * a chance to do so, since this address may reside
4429 			 * in the string table of this clause-- future calls
4430 			 * (when we fetch addr from mstate->dtms_strtok)
4431 			 * would fail this access check.
4432 			 */
4433 			if (!dtrace_strcanload(addr, size, mstate, vstate)) {
4434 				regs[rd] = NULL;
4435 				break;
4436 			}
4437 		}
4438 
4439 		/*
4440 		 * First, zero the token map, and then process the token
4441 		 * string -- setting a bit in the map for every character
4442 		 * found in the token string.
4443 		 */
4444 		for (i = 0; i < sizeof (tokmap); i++)
4445 			tokmap[i] = 0;
4446 
4447 		for (; tokaddr < toklimit; tokaddr++) {
4448 			if ((c = dtrace_load8(tokaddr)) == '\0')
4449 				break;
4450 
4451 			ASSERT((c >> 3) < sizeof (tokmap));
4452 			tokmap[c >> 3] |= (1 << (c & 0x7));
4453 		}
4454 
4455 		for (limit = addr + size; addr < limit; addr++) {
4456 			/*
4457 			 * We're looking for a character that is _not_ contained
4458 			 * in the token string.
4459 			 */
4460 			if ((c = dtrace_load8(addr)) == '\0')
4461 				break;
4462 
4463 			if (!(tokmap[c >> 3] & (1 << (c & 0x7))))
4464 				break;
4465 		}
4466 
4467 		if (c == '\0') {
4468 			/*
4469 			 * We reached the end of the string without finding
4470 			 * any character that was not in the token string.
4471 			 * We return NULL in this case, and we set the saved
4472 			 * address to NULL as well.
4473 			 */
4474 			regs[rd] = NULL;
4475 			mstate->dtms_strtok = NULL;
4476 			break;
4477 		}
4478 
4479 		/*
4480 		 * From here on, we're copying into the destination string.
4481 		 */
4482 		for (i = 0; addr < limit && i < size - 1; addr++) {
4483 			if ((c = dtrace_load8(addr)) == '\0')
4484 				break;
4485 
4486 			if (tokmap[c >> 3] & (1 << (c & 0x7)))
4487 				break;
4488 
4489 			ASSERT(i < size);
4490 			dest[i++] = c;
4491 		}
4492 
4493 		ASSERT(i < size);
4494 		dest[i] = '\0';
4495 		regs[rd] = (uintptr_t)dest;
4496 		mstate->dtms_scratch_ptr += size;
4497 		mstate->dtms_strtok = addr;
4498 		break;
4499 	}
4500 
4501 	case DIF_SUBR_SUBSTR: {
4502 		uintptr_t s = tupregs[0].dttk_value;
4503 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4504 		char *d = (char *)mstate->dtms_scratch_ptr;
4505 		int64_t index = (int64_t)tupregs[1].dttk_value;
4506 		int64_t remaining = (int64_t)tupregs[2].dttk_value;
4507 		size_t len = dtrace_strlen((char *)s, size);
4508 		int64_t i;
4509 
4510 		if (!dtrace_canload(s, len + 1, mstate, vstate)) {
4511 			regs[rd] = NULL;
4512 			break;
4513 		}
4514 
4515 		if (!DTRACE_INSCRATCH(mstate, size)) {
4516 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4517 			regs[rd] = NULL;
4518 			break;
4519 		}
4520 
4521 		if (nargs <= 2)
4522 			remaining = (int64_t)size;
4523 
4524 		if (index < 0) {
4525 			index += len;
4526 
4527 			if (index < 0 && index + remaining > 0) {
4528 				remaining += index;
4529 				index = 0;
4530 			}
4531 		}
4532 
4533 		if (index >= len || index < 0) {
4534 			remaining = 0;
4535 		} else if (remaining < 0) {
4536 			remaining += len - index;
4537 		} else if (index + remaining > size) {
4538 			remaining = size - index;
4539 		}
4540 
4541 		for (i = 0; i < remaining; i++) {
4542 			if ((d[i] = dtrace_load8(s + index + i)) == '\0')
4543 				break;
4544 		}
4545 
4546 		d[i] = '\0';
4547 
4548 		mstate->dtms_scratch_ptr += size;
4549 		regs[rd] = (uintptr_t)d;
4550 		break;
4551 	}
4552 
4553 	case DIF_SUBR_JSON: {
4554 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4555 		uintptr_t json = tupregs[0].dttk_value;
4556 		size_t jsonlen = dtrace_strlen((char *)json, size);
4557 		uintptr_t elem = tupregs[1].dttk_value;
4558 		size_t elemlen = dtrace_strlen((char *)elem, size);
4559 
4560 		char *dest = (char *)mstate->dtms_scratch_ptr;
4561 		char *elemlist = (char *)mstate->dtms_scratch_ptr + jsonlen + 1;
4562 		char *ee = elemlist;
4563 		int nelems = 1;
4564 		uintptr_t cur;
4565 
4566 		if (!dtrace_canload(json, jsonlen + 1, mstate, vstate) ||
4567 		    !dtrace_canload(elem, elemlen + 1, mstate, vstate)) {
4568 			regs[rd] = NULL;
4569 			break;
4570 		}
4571 
4572 		if (!DTRACE_INSCRATCH(mstate, jsonlen + 1 + elemlen + 1)) {
4573 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4574 			regs[rd] = NULL;
4575 			break;
4576 		}
4577 
4578 		/*
4579 		 * Read the element selector and split it up into a packed list
4580 		 * of strings.
4581 		 */
4582 		for (cur = elem; cur < elem + elemlen; cur++) {
4583 			char cc = dtrace_load8(cur);
4584 
4585 			if (cur == elem && cc == '[') {
4586 				/*
4587 				 * If the first element selector key is
4588 				 * actually an array index then ignore the
4589 				 * bracket.
4590 				 */
4591 				continue;
4592 			}
4593 
4594 			if (cc == ']')
4595 				continue;
4596 
4597 			if (cc == '.' || cc == '[') {
4598 				nelems++;
4599 				cc = '\0';
4600 			}
4601 
4602 			*ee++ = cc;
4603 		}
4604 		*ee++ = '\0';
4605 
4606 		if ((regs[rd] = (uintptr_t)dtrace_json(size, json, elemlist,
4607 		    nelems, dest)) != NULL)
4608 			mstate->dtms_scratch_ptr += jsonlen + 1;
4609 		break;
4610 	}
4611 
4612 	case DIF_SUBR_TOUPPER:
4613 	case DIF_SUBR_TOLOWER: {
4614 		uintptr_t s = tupregs[0].dttk_value;
4615 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4616 		char *dest = (char *)mstate->dtms_scratch_ptr, c;
4617 		size_t len = dtrace_strlen((char *)s, size);
4618 		char lower, upper, convert;
4619 		int64_t i;
4620 
4621 		if (subr == DIF_SUBR_TOUPPER) {
4622 			lower = 'a';
4623 			upper = 'z';
4624 			convert = 'A';
4625 		} else {
4626 			lower = 'A';
4627 			upper = 'Z';
4628 			convert = 'a';
4629 		}
4630 
4631 		if (!dtrace_canload(s, len + 1, mstate, vstate)) {
4632 			regs[rd] = NULL;
4633 			break;
4634 		}
4635 
4636 		if (!DTRACE_INSCRATCH(mstate, size)) {
4637 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4638 			regs[rd] = NULL;
4639 			break;
4640 		}
4641 
4642 		for (i = 0; i < size - 1; i++) {
4643 			if ((c = dtrace_load8(s + i)) == '\0')
4644 				break;
4645 
4646 			if (c >= lower && c <= upper)
4647 				c = convert + (c - lower);
4648 
4649 			dest[i] = c;
4650 		}
4651 
4652 		ASSERT(i < size);
4653 		dest[i] = '\0';
4654 		regs[rd] = (uintptr_t)dest;
4655 		mstate->dtms_scratch_ptr += size;
4656 		break;
4657 	}
4658 
4659 case DIF_SUBR_GETMAJOR:
4660 #ifdef _LP64
4661 		regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64;
4662 #else
4663 		regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ;
4664 #endif
4665 		break;
4666 
4667 	case DIF_SUBR_GETMINOR:
4668 #ifdef _LP64
4669 		regs[rd] = tupregs[0].dttk_value & MAXMIN64;
4670 #else
4671 		regs[rd] = tupregs[0].dttk_value & MAXMIN;
4672 #endif
4673 		break;
4674 
4675 	case DIF_SUBR_DDI_PATHNAME: {
4676 		/*
4677 		 * This one is a galactic mess.  We are going to roughly
4678 		 * emulate ddi_pathname(), but it's made more complicated
4679 		 * by the fact that we (a) want to include the minor name and
4680 		 * (b) must proceed iteratively instead of recursively.
4681 		 */
4682 		uintptr_t dest = mstate->dtms_scratch_ptr;
4683 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4684 		char *start = (char *)dest, *end = start + size - 1;
4685 		uintptr_t daddr = tupregs[0].dttk_value;
4686 		int64_t minor = (int64_t)tupregs[1].dttk_value;
4687 		char *s;
4688 		int i, len, depth = 0;
4689 
4690 		/*
4691 		 * Due to all the pointer jumping we do and context we must
4692 		 * rely upon, we just mandate that the user must have kernel
4693 		 * read privileges to use this routine.
4694 		 */
4695 		if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) {
4696 			*flags |= CPU_DTRACE_KPRIV;
4697 			*illval = daddr;
4698 			regs[rd] = NULL;
4699 		}
4700 
4701 		if (!DTRACE_INSCRATCH(mstate, size)) {
4702 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4703 			regs[rd] = NULL;
4704 			break;
4705 		}
4706 
4707 		*end = '\0';
4708 
4709 		/*
4710 		 * We want to have a name for the minor.  In order to do this,
4711 		 * we need to walk the minor list from the devinfo.  We want
4712 		 * to be sure that we don't infinitely walk a circular list,
4713 		 * so we check for circularity by sending a scout pointer
4714 		 * ahead two elements for every element that we iterate over;
4715 		 * if the list is circular, these will ultimately point to the
4716 		 * same element.  You may recognize this little trick as the
4717 		 * answer to a stupid interview question -- one that always
4718 		 * seems to be asked by those who had to have it laboriously
4719 		 * explained to them, and who can't even concisely describe
4720 		 * the conditions under which one would be forced to resort to
4721 		 * this technique.  Needless to say, those conditions are
4722 		 * found here -- and probably only here.  Is this the only use
4723 		 * of this infamous trick in shipping, production code?  If it
4724 		 * isn't, it probably should be...
4725 		 */
4726 		if (minor != -1) {
4727 			uintptr_t maddr = dtrace_loadptr(daddr +
4728 			    offsetof(struct dev_info, devi_minor));
4729 
4730 			uintptr_t next = offsetof(struct ddi_minor_data, next);
4731 			uintptr_t name = offsetof(struct ddi_minor_data,
4732 			    d_minor) + offsetof(struct ddi_minor, name);
4733 			uintptr_t dev = offsetof(struct ddi_minor_data,
4734 			    d_minor) + offsetof(struct ddi_minor, dev);
4735 			uintptr_t scout;
4736 
4737 			if (maddr != NULL)
4738 				scout = dtrace_loadptr(maddr + next);
4739 
4740 			while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4741 				uint64_t m;
4742 #ifdef _LP64
4743 				m = dtrace_load64(maddr + dev) & MAXMIN64;
4744 #else
4745 				m = dtrace_load32(maddr + dev) & MAXMIN;
4746 #endif
4747 				if (m != minor) {
4748 					maddr = dtrace_loadptr(maddr + next);
4749 
4750 					if (scout == NULL)
4751 						continue;
4752 
4753 					scout = dtrace_loadptr(scout + next);
4754 
4755 					if (scout == NULL)
4756 						continue;
4757 
4758 					scout = dtrace_loadptr(scout + next);
4759 
4760 					if (scout == NULL)
4761 						continue;
4762 
4763 					if (scout == maddr) {
4764 						*flags |= CPU_DTRACE_ILLOP;
4765 						break;
4766 					}
4767 
4768 					continue;
4769 				}
4770 
4771 				/*
4772 				 * We have the minor data.  Now we need to
4773 				 * copy the minor's name into the end of the
4774 				 * pathname.
4775 				 */
4776 				s = (char *)dtrace_loadptr(maddr + name);
4777 				len = dtrace_strlen(s, size);
4778 
4779 				if (*flags & CPU_DTRACE_FAULT)
4780 					break;
4781 
4782 				if (len != 0) {
4783 					if ((end -= (len + 1)) < start)
4784 						break;
4785 
4786 					*end = ':';
4787 				}
4788 
4789 				for (i = 1; i <= len; i++)
4790 					end[i] = dtrace_load8((uintptr_t)s++);
4791 				break;
4792 			}
4793 		}
4794 
4795 		while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4796 			ddi_node_state_t devi_state;
4797 
4798 			devi_state = dtrace_load32(daddr +
4799 			    offsetof(struct dev_info, devi_node_state));
4800 
4801 			if (*flags & CPU_DTRACE_FAULT)
4802 				break;
4803 
4804 			if (devi_state >= DS_INITIALIZED) {
4805 				s = (char *)dtrace_loadptr(daddr +
4806 				    offsetof(struct dev_info, devi_addr));
4807 				len = dtrace_strlen(s, size);
4808 
4809 				if (*flags & CPU_DTRACE_FAULT)
4810 					break;
4811 
4812 				if (len != 0) {
4813 					if ((end -= (len + 1)) < start)
4814 						break;
4815 
4816 					*end = '@';
4817 				}
4818 
4819 				for (i = 1; i <= len; i++)
4820 					end[i] = dtrace_load8((uintptr_t)s++);
4821 			}
4822 
4823 			/*
4824 			 * Now for the node name...
4825 			 */
4826 			s = (char *)dtrace_loadptr(daddr +
4827 			    offsetof(struct dev_info, devi_node_name));
4828 
4829 			daddr = dtrace_loadptr(daddr +
4830 			    offsetof(struct dev_info, devi_parent));
4831 
4832 			/*
4833 			 * If our parent is NULL (that is, if we're the root
4834 			 * node), we're going to use the special path
4835 			 * "devices".
4836 			 */
4837 			if (daddr == NULL)
4838 				s = "devices";
4839 
4840 			len = dtrace_strlen(s, size);
4841 			if (*flags & CPU_DTRACE_FAULT)
4842 				break;
4843 
4844 			if ((end -= (len + 1)) < start)
4845 				break;
4846 
4847 			for (i = 1; i <= len; i++)
4848 				end[i] = dtrace_load8((uintptr_t)s++);
4849 			*end = '/';
4850 
4851 			if (depth++ > dtrace_devdepth_max) {
4852 				*flags |= CPU_DTRACE_ILLOP;
4853 				break;
4854 			}
4855 		}
4856 
4857 		if (end < start)
4858 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4859 
4860 		if (daddr == NULL) {
4861 			regs[rd] = (uintptr_t)end;
4862 			mstate->dtms_scratch_ptr += size;
4863 		}
4864 
4865 		break;
4866 	}
4867 
4868 	case DIF_SUBR_STRJOIN: {
4869 		char *d = (char *)mstate->dtms_scratch_ptr;
4870 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4871 		uintptr_t s1 = tupregs[0].dttk_value;
4872 		uintptr_t s2 = tupregs[1].dttk_value;
4873 		int i = 0;
4874 
4875 		if (!dtrace_strcanload(s1, size, mstate, vstate) ||
4876 		    !dtrace_strcanload(s2, size, mstate, vstate)) {
4877 			regs[rd] = NULL;
4878 			break;
4879 		}
4880 
4881 		if (!DTRACE_INSCRATCH(mstate, size)) {
4882 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4883 			regs[rd] = NULL;
4884 			break;
4885 		}
4886 
4887 		for (;;) {
4888 			if (i >= size) {
4889 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4890 				regs[rd] = NULL;
4891 				break;
4892 			}
4893 
4894 			if ((d[i++] = dtrace_load8(s1++)) == '\0') {
4895 				i--;
4896 				break;
4897 			}
4898 		}
4899 
4900 		for (;;) {
4901 			if (i >= size) {
4902 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4903 				regs[rd] = NULL;
4904 				break;
4905 			}
4906 
4907 			if ((d[i++] = dtrace_load8(s2++)) == '\0')
4908 				break;
4909 		}
4910 
4911 		if (i < size) {
4912 			mstate->dtms_scratch_ptr += i;
4913 			regs[rd] = (uintptr_t)d;
4914 		}
4915 
4916 		break;
4917 	}
4918 
4919 	case DIF_SUBR_STRTOLL: {
4920 		uintptr_t s = tupregs[0].dttk_value;
4921 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4922 		int base = 10;
4923 
4924 		if (nargs > 1) {
4925 			if ((base = tupregs[1].dttk_value) <= 1 ||
4926 			    base > ('z' - 'a' + 1) + ('9' - '0' + 1)) {
4927 				*flags |= CPU_DTRACE_ILLOP;
4928 				break;
4929 			}
4930 		}
4931 
4932 		if (!dtrace_strcanload(s, size, mstate, vstate)) {
4933 			regs[rd] = INT64_MIN;
4934 			break;
4935 		}
4936 
4937 		regs[rd] = dtrace_strtoll((char *)s, base, size);
4938 		break;
4939 	}
4940 
4941 	case DIF_SUBR_LLTOSTR: {
4942 		int64_t i = (int64_t)tupregs[0].dttk_value;
4943 		uint64_t val, digit;
4944 		uint64_t size = 65;	/* enough room for 2^64 in binary */
4945 		char *end = (char *)mstate->dtms_scratch_ptr + size - 1;
4946 		int base = 10;
4947 
4948 		if (nargs > 1) {
4949 			if ((base = tupregs[1].dttk_value) <= 1 ||
4950 			    base > ('z' - 'a' + 1) + ('9' - '0' + 1)) {
4951 				*flags |= CPU_DTRACE_ILLOP;
4952 				break;
4953 			}
4954 		}
4955 
4956 		val = (base == 10 && i < 0) ? i * -1 : i;
4957 
4958 		if (!DTRACE_INSCRATCH(mstate, size)) {
4959 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4960 			regs[rd] = NULL;
4961 			break;
4962 		}
4963 
4964 		for (*end-- = '\0'; val; val /= base) {
4965 			if ((digit = val % base) <= '9' - '0') {
4966 				*end-- = '0' + digit;
4967 			} else {
4968 				*end-- = 'a' + (digit - ('9' - '0') - 1);
4969 			}
4970 		}
4971 
4972 		if (i == 0 && base == 16)
4973 			*end-- = '0';
4974 
4975 		if (base == 16)
4976 			*end-- = 'x';
4977 
4978 		if (i == 0 || base == 8 || base == 16)
4979 			*end-- = '0';
4980 
4981 		if (i < 0 && base == 10)
4982 			*end-- = '-';
4983 
4984 		regs[rd] = (uintptr_t)end + 1;
4985 		mstate->dtms_scratch_ptr += size;
4986 		break;
4987 	}
4988 
4989 	case DIF_SUBR_HTONS:
4990 	case DIF_SUBR_NTOHS:
4991 #ifdef _BIG_ENDIAN
4992 		regs[rd] = (uint16_t)tupregs[0].dttk_value;
4993 #else
4994 		regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value);
4995 #endif
4996 		break;
4997 
4998 
4999 	case DIF_SUBR_HTONL:
5000 	case DIF_SUBR_NTOHL:
5001 #ifdef _BIG_ENDIAN
5002 		regs[rd] = (uint32_t)tupregs[0].dttk_value;
5003 #else
5004 		regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value);
5005 #endif
5006 		break;
5007 
5008 
5009 	case DIF_SUBR_HTONLL:
5010 	case DIF_SUBR_NTOHLL:
5011 #ifdef _BIG_ENDIAN
5012 		regs[rd] = (uint64_t)tupregs[0].dttk_value;
5013 #else
5014 		regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value);
5015 #endif
5016 		break;
5017 
5018 
5019 	case DIF_SUBR_DIRNAME:
5020 	case DIF_SUBR_BASENAME: {
5021 		char *dest = (char *)mstate->dtms_scratch_ptr;
5022 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
5023 		uintptr_t src = tupregs[0].dttk_value;
5024 		int i, j, len = dtrace_strlen((char *)src, size);
5025 		int lastbase = -1, firstbase = -1, lastdir = -1;
5026 		int start, end;
5027 
5028 		if (!dtrace_canload(src, len + 1, mstate, vstate)) {
5029 			regs[rd] = NULL;
5030 			break;
5031 		}
5032 
5033 		if (!DTRACE_INSCRATCH(mstate, size)) {
5034 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5035 			regs[rd] = NULL;
5036 			break;
5037 		}
5038 
5039 		/*
5040 		 * The basename and dirname for a zero-length string is
5041 		 * defined to be "."
5042 		 */
5043 		if (len == 0) {
5044 			len = 1;
5045 			src = (uintptr_t)".";
5046 		}
5047 
5048 		/*
5049 		 * Start from the back of the string, moving back toward the
5050 		 * front until we see a character that isn't a slash.  That
5051 		 * character is the last character in the basename.
5052 		 */
5053 		for (i = len - 1; i >= 0; i--) {
5054 			if (dtrace_load8(src + i) != '/')
5055 				break;
5056 		}
5057 
5058 		if (i >= 0)
5059 			lastbase = i;
5060 
5061 		/*
5062 		 * Starting from the last character in the basename, move
5063 		 * towards the front until we find a slash.  The character
5064 		 * that we processed immediately before that is the first
5065 		 * character in the basename.
5066 		 */
5067 		for (; i >= 0; i--) {
5068 			if (dtrace_load8(src + i) == '/')
5069 				break;
5070 		}
5071 
5072 		if (i >= 0)
5073 			firstbase = i + 1;
5074 
5075 		/*
5076 		 * Now keep going until we find a non-slash character.  That
5077 		 * character is the last character in the dirname.
5078 		 */
5079 		for (; i >= 0; i--) {
5080 			if (dtrace_load8(src + i) != '/')
5081 				break;
5082 		}
5083 
5084 		if (i >= 0)
5085 			lastdir = i;
5086 
5087 		ASSERT(!(lastbase == -1 && firstbase != -1));
5088 		ASSERT(!(firstbase == -1 && lastdir != -1));
5089 
5090 		if (lastbase == -1) {
5091 			/*
5092 			 * We didn't find a non-slash character.  We know that
5093 			 * the length is non-zero, so the whole string must be
5094 			 * slashes.  In either the dirname or the basename
5095 			 * case, we return '/'.
5096 			 */
5097 			ASSERT(firstbase == -1);
5098 			firstbase = lastbase = lastdir = 0;
5099 		}
5100 
5101 		if (firstbase == -1) {
5102 			/*
5103 			 * The entire string consists only of a basename
5104 			 * component.  If we're looking for dirname, we need
5105 			 * to change our string to be just "."; if we're
5106 			 * looking for a basename, we'll just set the first
5107 			 * character of the basename to be 0.
5108 			 */
5109 			if (subr == DIF_SUBR_DIRNAME) {
5110 				ASSERT(lastdir == -1);
5111 				src = (uintptr_t)".";
5112 				lastdir = 0;
5113 			} else {
5114 				firstbase = 0;
5115 			}
5116 		}
5117 
5118 		if (subr == DIF_SUBR_DIRNAME) {
5119 			if (lastdir == -1) {
5120 				/*
5121 				 * We know that we have a slash in the name --
5122 				 * or lastdir would be set to 0, above.  And
5123 				 * because lastdir is -1, we know that this
5124 				 * slash must be the first character.  (That
5125 				 * is, the full string must be of the form
5126 				 * "/basename".)  In this case, the last
5127 				 * character of the directory name is 0.
5128 				 */
5129 				lastdir = 0;
5130 			}
5131 
5132 			start = 0;
5133 			end = lastdir;
5134 		} else {
5135 			ASSERT(subr == DIF_SUBR_BASENAME);
5136 			ASSERT(firstbase != -1 && lastbase != -1);
5137 			start = firstbase;
5138 			end = lastbase;
5139 		}
5140 
5141 		for (i = start, j = 0; i <= end && j < size - 1; i++, j++)
5142 			dest[j] = dtrace_load8(src + i);
5143 
5144 		dest[j] = '\0';
5145 		regs[rd] = (uintptr_t)dest;
5146 		mstate->dtms_scratch_ptr += size;
5147 		break;
5148 	}
5149 
5150 	case DIF_SUBR_GETF: {
5151 		uintptr_t fd = tupregs[0].dttk_value;
5152 		uf_info_t *finfo = &curthread->t_procp->p_user.u_finfo;
5153 		file_t *fp;
5154 
5155 		if (!dtrace_priv_proc(state, mstate)) {
5156 			regs[rd] = NULL;
5157 			break;
5158 		}
5159 
5160 		/*
5161 		 * This is safe because fi_nfiles only increases, and the
5162 		 * fi_list array is not freed when the array size doubles.
5163 		 * (See the comment in flist_grow() for details on the
5164 		 * management of the u_finfo structure.)
5165 		 */
5166 		fp = fd < finfo->fi_nfiles ? finfo->fi_list[fd].uf_file : NULL;
5167 
5168 		mstate->dtms_getf = fp;
5169 		regs[rd] = (uintptr_t)fp;
5170 		break;
5171 	}
5172 
5173 	case DIF_SUBR_CLEANPATH: {
5174 		char *dest = (char *)mstate->dtms_scratch_ptr, c;
5175 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
5176 		uintptr_t src = tupregs[0].dttk_value;
5177 		int i = 0, j = 0;
5178 		zone_t *z;
5179 
5180 		if (!dtrace_strcanload(src, size, mstate, vstate)) {
5181 			regs[rd] = NULL;
5182 			break;
5183 		}
5184 
5185 		if (!DTRACE_INSCRATCH(mstate, size)) {
5186 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5187 			regs[rd] = NULL;
5188 			break;
5189 		}
5190 
5191 		/*
5192 		 * Move forward, loading each character.
5193 		 */
5194 		do {
5195 			c = dtrace_load8(src + i++);
5196 next:
5197 			if (j + 5 >= size)	/* 5 = strlen("/..c\0") */
5198 				break;
5199 
5200 			if (c != '/') {
5201 				dest[j++] = c;
5202 				continue;
5203 			}
5204 
5205 			c = dtrace_load8(src + i++);
5206 
5207 			if (c == '/') {
5208 				/*
5209 				 * We have two slashes -- we can just advance
5210 				 * to the next character.
5211 				 */
5212 				goto next;
5213 			}
5214 
5215 			if (c != '.') {
5216 				/*
5217 				 * This is not "." and it's not ".." -- we can
5218 				 * just store the "/" and this character and
5219 				 * drive on.
5220 				 */
5221 				dest[j++] = '/';
5222 				dest[j++] = c;
5223 				continue;
5224 			}
5225 
5226 			c = dtrace_load8(src + i++);
5227 
5228 			if (c == '/') {
5229 				/*
5230 				 * This is a "/./" component.  We're not going
5231 				 * to store anything in the destination buffer;
5232 				 * we're just going to go to the next component.
5233 				 */
5234 				goto next;
5235 			}
5236 
5237 			if (c != '.') {
5238 				/*
5239 				 * This is not ".." -- we can just store the
5240 				 * "/." and this character and continue
5241 				 * processing.
5242 				 */
5243 				dest[j++] = '/';
5244 				dest[j++] = '.';
5245 				dest[j++] = c;
5246 				continue;
5247 			}
5248 
5249 			c = dtrace_load8(src + i++);
5250 
5251 			if (c != '/' && c != '\0') {
5252 				/*
5253 				 * This is not ".." -- it's "..[mumble]".
5254 				 * We'll store the "/.." and this character
5255 				 * and continue processing.
5256 				 */
5257 				dest[j++] = '/';
5258 				dest[j++] = '.';
5259 				dest[j++] = '.';
5260 				dest[j++] = c;
5261 				continue;
5262 			}
5263 
5264 			/*
5265 			 * This is "/../" or "/..\0".  We need to back up
5266 			 * our destination pointer until we find a "/".
5267 			 */
5268 			i--;
5269 			while (j != 0 && dest[--j] != '/')
5270 				continue;
5271 
5272 			if (c == '\0')
5273 				dest[++j] = '/';
5274 		} while (c != '\0');
5275 
5276 		dest[j] = '\0';
5277 
5278 		if (mstate->dtms_getf != NULL &&
5279 		    !(mstate->dtms_access & DTRACE_ACCESS_KERNEL) &&
5280 		    (z = state->dts_cred.dcr_cred->cr_zone) != kcred->cr_zone) {
5281 			/*
5282 			 * If we've done a getf() as a part of this ECB and we
5283 			 * don't have kernel access (and we're not in the global
5284 			 * zone), check if the path we cleaned up begins with
5285 			 * the zone's root path, and trim it off if so.  Note
5286 			 * that this is an output cleanliness issue, not a
5287 			 * security issue: knowing one's zone root path does
5288 			 * not enable privilege escalation.
5289 			 */
5290 			if (strstr(dest, z->zone_rootpath) == dest)
5291 				dest += strlen(z->zone_rootpath) - 1;
5292 		}
5293 
5294 		regs[rd] = (uintptr_t)dest;
5295 		mstate->dtms_scratch_ptr += size;
5296 		break;
5297 	}
5298 
5299 	case DIF_SUBR_INET_NTOA:
5300 	case DIF_SUBR_INET_NTOA6:
5301 	case DIF_SUBR_INET_NTOP: {
5302 		size_t size;
5303 		int af, argi, i;
5304 		char *base, *end;
5305 
5306 		if (subr == DIF_SUBR_INET_NTOP) {
5307 			af = (int)tupregs[0].dttk_value;
5308 			argi = 1;
5309 		} else {
5310 			af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6;
5311 			argi = 0;
5312 		}
5313 
5314 		if (af == AF_INET) {
5315 			ipaddr_t ip4;
5316 			uint8_t *ptr8, val;
5317 
5318 			/*
5319 			 * Safely load the IPv4 address.
5320 			 */
5321 			ip4 = dtrace_load32(tupregs[argi].dttk_value);
5322 
5323 			/*
5324 			 * Check an IPv4 string will fit in scratch.
5325 			 */
5326 			size = INET_ADDRSTRLEN;
5327 			if (!DTRACE_INSCRATCH(mstate, size)) {
5328 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5329 				regs[rd] = NULL;
5330 				break;
5331 			}
5332 			base = (char *)mstate->dtms_scratch_ptr;
5333 			end = (char *)mstate->dtms_scratch_ptr + size - 1;
5334 
5335 			/*
5336 			 * Stringify as a dotted decimal quad.
5337 			 */
5338 			*end-- = '\0';
5339 			ptr8 = (uint8_t *)&ip4;
5340 			for (i = 3; i >= 0; i--) {
5341 				val = ptr8[i];
5342 
5343 				if (val == 0) {
5344 					*end-- = '0';
5345 				} else {
5346 					for (; val; val /= 10) {
5347 						*end-- = '0' + (val % 10);
5348 					}
5349 				}
5350 
5351 				if (i > 0)
5352 					*end-- = '.';
5353 			}
5354 			ASSERT(end + 1 >= base);
5355 
5356 		} else if (af == AF_INET6) {
5357 			struct in6_addr ip6;
5358 			int firstzero, tryzero, numzero, v6end;
5359 			uint16_t val;
5360 			const char digits[] = "0123456789abcdef";
5361 
5362 			/*
5363 			 * Stringify using RFC 1884 convention 2 - 16 bit
5364 			 * hexadecimal values with a zero-run compression.
5365 			 * Lower case hexadecimal digits are used.
5366 			 * 	eg, fe80::214:4fff:fe0b:76c8.
5367 			 * The IPv4 embedded form is returned for inet_ntop,
5368 			 * just the IPv4 string is returned for inet_ntoa6.
5369 			 */
5370 
5371 			/*
5372 			 * Safely load the IPv6 address.
5373 			 */
5374 			dtrace_bcopy(
5375 			    (void *)(uintptr_t)tupregs[argi].dttk_value,
5376 			    (void *)(uintptr_t)&ip6, sizeof (struct in6_addr));
5377 
5378 			/*
5379 			 * Check an IPv6 string will fit in scratch.
5380 			 */
5381 			size = INET6_ADDRSTRLEN;
5382 			if (!DTRACE_INSCRATCH(mstate, size)) {
5383 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5384 				regs[rd] = NULL;
5385 				break;
5386 			}
5387 			base = (char *)mstate->dtms_scratch_ptr;
5388 			end = (char *)mstate->dtms_scratch_ptr + size - 1;
5389 			*end-- = '\0';
5390 
5391 			/*
5392 			 * Find the longest run of 16 bit zero values
5393 			 * for the single allowed zero compression - "::".
5394 			 */
5395 			firstzero = -1;
5396 			tryzero = -1;
5397 			numzero = 1;
5398 			for (i = 0; i < sizeof (struct in6_addr); i++) {
5399 				if (ip6._S6_un._S6_u8[i] == 0 &&
5400 				    tryzero == -1 && i % 2 == 0) {
5401 					tryzero = i;
5402 					continue;
5403 				}
5404 
5405 				if (tryzero != -1 &&
5406 				    (ip6._S6_un._S6_u8[i] != 0 ||
5407 				    i == sizeof (struct in6_addr) - 1)) {
5408 
5409 					if (i - tryzero <= numzero) {
5410 						tryzero = -1;
5411 						continue;
5412 					}
5413 
5414 					firstzero = tryzero;
5415 					numzero = i - i % 2 - tryzero;
5416 					tryzero = -1;
5417 
5418 					if (ip6._S6_un._S6_u8[i] == 0 &&
5419 					    i == sizeof (struct in6_addr) - 1)
5420 						numzero += 2;
5421 				}
5422 			}
5423 			ASSERT(firstzero + numzero <= sizeof (struct in6_addr));
5424 
5425 			/*
5426 			 * Check for an IPv4 embedded address.
5427 			 */
5428 			v6end = sizeof (struct in6_addr) - 2;
5429 			if (IN6_IS_ADDR_V4MAPPED(&ip6) ||
5430 			    IN6_IS_ADDR_V4COMPAT(&ip6)) {
5431 				for (i = sizeof (struct in6_addr) - 1;
5432 				    i >= DTRACE_V4MAPPED_OFFSET; i--) {
5433 					ASSERT(end >= base);
5434 
5435 					val = ip6._S6_un._S6_u8[i];
5436 
5437 					if (val == 0) {
5438 						*end-- = '0';
5439 					} else {
5440 						for (; val; val /= 10) {
5441 							*end-- = '0' + val % 10;
5442 						}
5443 					}
5444 
5445 					if (i > DTRACE_V4MAPPED_OFFSET)
5446 						*end-- = '.';
5447 				}
5448 
5449 				if (subr == DIF_SUBR_INET_NTOA6)
5450 					goto inetout;
5451 
5452 				/*
5453 				 * Set v6end to skip the IPv4 address that
5454 				 * we have already stringified.
5455 				 */
5456 				v6end = 10;
5457 			}
5458 
5459 			/*
5460 			 * Build the IPv6 string by working through the
5461 			 * address in reverse.
5462 			 */
5463 			for (i = v6end; i >= 0; i -= 2) {
5464 				ASSERT(end >= base);
5465 
5466 				if (i == firstzero + numzero - 2) {
5467 					*end-- = ':';
5468 					*end-- = ':';
5469 					i -= numzero - 2;
5470 					continue;
5471 				}
5472 
5473 				if (i < 14 && i != firstzero - 2)
5474 					*end-- = ':';
5475 
5476 				val = (ip6._S6_un._S6_u8[i] << 8) +
5477 				    ip6._S6_un._S6_u8[i + 1];
5478 
5479 				if (val == 0) {
5480 					*end-- = '0';
5481 				} else {
5482 					for (; val; val /= 16) {
5483 						*end-- = digits[val % 16];
5484 					}
5485 				}
5486 			}
5487 			ASSERT(end + 1 >= base);
5488 
5489 		} else {
5490 			/*
5491 			 * The user didn't use AH_INET or AH_INET6.
5492 			 */
5493 			DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
5494 			regs[rd] = NULL;
5495 			break;
5496 		}
5497 
5498 inetout:	regs[rd] = (uintptr_t)end + 1;
5499 		mstate->dtms_scratch_ptr += size;
5500 		break;
5501 	}
5502 
5503 	}
5504 }
5505 
5506 /*
5507  * Emulate the execution of DTrace IR instructions specified by the given
5508  * DIF object.  This function is deliberately void of assertions as all of
5509  * the necessary checks are handled by a call to dtrace_difo_validate().
5510  */
5511 static uint64_t
dtrace_dif_emulate(dtrace_difo_t * difo,dtrace_mstate_t * mstate,dtrace_vstate_t * vstate,dtrace_state_t * state)5512 dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate,
5513     dtrace_vstate_t *vstate, dtrace_state_t *state)
5514 {
5515 	const dif_instr_t *text = difo->dtdo_buf;
5516 	const uint_t textlen = difo->dtdo_len;
5517 	const char *strtab = difo->dtdo_strtab;
5518 	const uint64_t *inttab = difo->dtdo_inttab;
5519 
5520 	uint64_t rval = 0;
5521 	dtrace_statvar_t *svar;
5522 	dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
5523 	dtrace_difv_t *v;
5524 	volatile uint16_t *flags = &cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
5525 	volatile uintptr_t *illval = &cpu_core[CPU->cpu_id].cpuc_dtrace_illval;
5526 
5527 	dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
5528 	uint64_t regs[DIF_DIR_NREGS];
5529 	uint64_t *tmp;
5530 
5531 	uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0;
5532 	int64_t cc_r;
5533 	uint_t pc = 0, id, opc;
5534 	uint8_t ttop = 0;
5535 	dif_instr_t instr;
5536 	uint_t r1, r2, rd;
5537 
5538 	/*
5539 	 * We stash the current DIF object into the machine state: we need it
5540 	 * for subsequent access checking.
5541 	 */
5542 	mstate->dtms_difo = difo;
5543 
5544 	regs[DIF_REG_R0] = 0; 		/* %r0 is fixed at zero */
5545 
5546 	while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) {
5547 		opc = pc;
5548 
5549 		instr = text[pc++];
5550 		r1 = DIF_INSTR_R1(instr);
5551 		r2 = DIF_INSTR_R2(instr);
5552 		rd = DIF_INSTR_RD(instr);
5553 
5554 		switch (DIF_INSTR_OP(instr)) {
5555 		case DIF_OP_OR:
5556 			regs[rd] = regs[r1] | regs[r2];
5557 			break;
5558 		case DIF_OP_XOR:
5559 			regs[rd] = regs[r1] ^ regs[r2];
5560 			break;
5561 		case DIF_OP_AND:
5562 			regs[rd] = regs[r1] & regs[r2];
5563 			break;
5564 		case DIF_OP_SLL:
5565 			regs[rd] = regs[r1] << regs[r2];
5566 			break;
5567 		case DIF_OP_SRL:
5568 			regs[rd] = regs[r1] >> regs[r2];
5569 			break;
5570 		case DIF_OP_SUB:
5571 			regs[rd] = regs[r1] - regs[r2];
5572 			break;
5573 		case DIF_OP_ADD:
5574 			regs[rd] = regs[r1] + regs[r2];
5575 			break;
5576 		case DIF_OP_MUL:
5577 			regs[rd] = regs[r1] * regs[r2];
5578 			break;
5579 		case DIF_OP_SDIV:
5580 			if (regs[r2] == 0) {
5581 				regs[rd] = 0;
5582 				*flags |= CPU_DTRACE_DIVZERO;
5583 			} else {
5584 				regs[rd] = (int64_t)regs[r1] /
5585 				    (int64_t)regs[r2];
5586 			}
5587 			break;
5588 
5589 		case DIF_OP_UDIV:
5590 			if (regs[r2] == 0) {
5591 				regs[rd] = 0;
5592 				*flags |= CPU_DTRACE_DIVZERO;
5593 			} else {
5594 				regs[rd] = regs[r1] / regs[r2];
5595 			}
5596 			break;
5597 
5598 		case DIF_OP_SREM:
5599 			if (regs[r2] == 0) {
5600 				regs[rd] = 0;
5601 				*flags |= CPU_DTRACE_DIVZERO;
5602 			} else {
5603 				regs[rd] = (int64_t)regs[r1] %
5604 				    (int64_t)regs[r2];
5605 			}
5606 			break;
5607 
5608 		case DIF_OP_UREM:
5609 			if (regs[r2] == 0) {
5610 				regs[rd] = 0;
5611 				*flags |= CPU_DTRACE_DIVZERO;
5612 			} else {
5613 				regs[rd] = regs[r1] % regs[r2];
5614 			}
5615 			break;
5616 
5617 		case DIF_OP_NOT:
5618 			regs[rd] = ~regs[r1];
5619 			break;
5620 		case DIF_OP_MOV:
5621 			regs[rd] = regs[r1];
5622 			break;
5623 		case DIF_OP_CMP:
5624 			cc_r = regs[r1] - regs[r2];
5625 			cc_n = cc_r < 0;
5626 			cc_z = cc_r == 0;
5627 			cc_v = 0;
5628 			cc_c = regs[r1] < regs[r2];
5629 			break;
5630 		case DIF_OP_TST:
5631 			cc_n = cc_v = cc_c = 0;
5632 			cc_z = regs[r1] == 0;
5633 			break;
5634 		case DIF_OP_BA:
5635 			pc = DIF_INSTR_LABEL(instr);
5636 			break;
5637 		case DIF_OP_BE:
5638 			if (cc_z)
5639 				pc = DIF_INSTR_LABEL(instr);
5640 			break;
5641 		case DIF_OP_BNE:
5642 			if (cc_z == 0)
5643 				pc = DIF_INSTR_LABEL(instr);
5644 			break;
5645 		case DIF_OP_BG:
5646 			if ((cc_z | (cc_n ^ cc_v)) == 0)
5647 				pc = DIF_INSTR_LABEL(instr);
5648 			break;
5649 		case DIF_OP_BGU:
5650 			if ((cc_c | cc_z) == 0)
5651 				pc = DIF_INSTR_LABEL(instr);
5652 			break;
5653 		case DIF_OP_BGE:
5654 			if ((cc_n ^ cc_v) == 0)
5655 				pc = DIF_INSTR_LABEL(instr);
5656 			break;
5657 		case DIF_OP_BGEU:
5658 			if (cc_c == 0)
5659 				pc = DIF_INSTR_LABEL(instr);
5660 			break;
5661 		case DIF_OP_BL:
5662 			if (cc_n ^ cc_v)
5663 				pc = DIF_INSTR_LABEL(instr);
5664 			break;
5665 		case DIF_OP_BLU:
5666 			if (cc_c)
5667 				pc = DIF_INSTR_LABEL(instr);
5668 			break;
5669 		case DIF_OP_BLE:
5670 			if (cc_z | (cc_n ^ cc_v))
5671 				pc = DIF_INSTR_LABEL(instr);
5672 			break;
5673 		case DIF_OP_BLEU:
5674 			if (cc_c | cc_z)
5675 				pc = DIF_INSTR_LABEL(instr);
5676 			break;
5677 		case DIF_OP_RLDSB:
5678 			if (!dtrace_canload(regs[r1], 1, mstate, vstate))
5679 				break;
5680 			/*FALLTHROUGH*/
5681 		case DIF_OP_LDSB:
5682 			regs[rd] = (int8_t)dtrace_load8(regs[r1]);
5683 			break;
5684 		case DIF_OP_RLDSH:
5685 			if (!dtrace_canload(regs[r1], 2, mstate, vstate))
5686 				break;
5687 			/*FALLTHROUGH*/
5688 		case DIF_OP_LDSH:
5689 			regs[rd] = (int16_t)dtrace_load16(regs[r1]);
5690 			break;
5691 		case DIF_OP_RLDSW:
5692 			if (!dtrace_canload(regs[r1], 4, mstate, vstate))
5693 				break;
5694 			/*FALLTHROUGH*/
5695 		case DIF_OP_LDSW:
5696 			regs[rd] = (int32_t)dtrace_load32(regs[r1]);
5697 			break;
5698 		case DIF_OP_RLDUB:
5699 			if (!dtrace_canload(regs[r1], 1, mstate, vstate))
5700 				break;
5701 			/*FALLTHROUGH*/
5702 		case DIF_OP_LDUB:
5703 			regs[rd] = dtrace_load8(regs[r1]);
5704 			break;
5705 		case DIF_OP_RLDUH:
5706 			if (!dtrace_canload(regs[r1], 2, mstate, vstate))
5707 				break;
5708 			/*FALLTHROUGH*/
5709 		case DIF_OP_LDUH:
5710 			regs[rd] = dtrace_load16(regs[r1]);
5711 			break;
5712 		case DIF_OP_RLDUW:
5713 			if (!dtrace_canload(regs[r1], 4, mstate, vstate))
5714 				break;
5715 			/*FALLTHROUGH*/
5716 		case DIF_OP_LDUW:
5717 			regs[rd] = dtrace_load32(regs[r1]);
5718 			break;
5719 		case DIF_OP_RLDX:
5720 			if (!dtrace_canload(regs[r1], 8, mstate, vstate))
5721 				break;
5722 			/*FALLTHROUGH*/
5723 		case DIF_OP_LDX:
5724 			regs[rd] = dtrace_load64(regs[r1]);
5725 			break;
5726 		case DIF_OP_ULDSB:
5727 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5728 			regs[rd] = (int8_t)
5729 			    dtrace_fuword8((void *)(uintptr_t)regs[r1]);
5730 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5731 			break;
5732 		case DIF_OP_ULDSH:
5733 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5734 			regs[rd] = (int16_t)
5735 			    dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5736 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5737 			break;
5738 		case DIF_OP_ULDSW:
5739 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5740 			regs[rd] = (int32_t)
5741 			    dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5742 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5743 			break;
5744 		case DIF_OP_ULDUB:
5745 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5746 			regs[rd] =
5747 			    dtrace_fuword8((void *)(uintptr_t)regs[r1]);
5748 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5749 			break;
5750 		case DIF_OP_ULDUH:
5751 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5752 			regs[rd] =
5753 			    dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5754 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5755 			break;
5756 		case DIF_OP_ULDUW:
5757 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5758 			regs[rd] =
5759 			    dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5760 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5761 			break;
5762 		case DIF_OP_ULDX:
5763 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5764 			regs[rd] =
5765 			    dtrace_fuword64((void *)(uintptr_t)regs[r1]);
5766 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5767 			break;
5768 		case DIF_OP_RET:
5769 			rval = regs[rd];
5770 			pc = textlen;
5771 			break;
5772 		case DIF_OP_NOP:
5773 			break;
5774 		case DIF_OP_SETX:
5775 			regs[rd] = inttab[DIF_INSTR_INTEGER(instr)];
5776 			break;
5777 		case DIF_OP_SETS:
5778 			regs[rd] = (uint64_t)(uintptr_t)
5779 			    (strtab + DIF_INSTR_STRING(instr));
5780 			break;
5781 		case DIF_OP_SCMP: {
5782 			size_t sz = state->dts_options[DTRACEOPT_STRSIZE];
5783 			uintptr_t s1 = regs[r1];
5784 			uintptr_t s2 = regs[r2];
5785 
5786 			if (s1 != NULL &&
5787 			    !dtrace_strcanload(s1, sz, mstate, vstate))
5788 				break;
5789 			if (s2 != NULL &&
5790 			    !dtrace_strcanload(s2, sz, mstate, vstate))
5791 				break;
5792 
5793 			cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz);
5794 
5795 			cc_n = cc_r < 0;
5796 			cc_z = cc_r == 0;
5797 			cc_v = cc_c = 0;
5798 			break;
5799 		}
5800 		case DIF_OP_LDGA:
5801 			regs[rd] = dtrace_dif_variable(mstate, state,
5802 			    r1, regs[r2]);
5803 			break;
5804 		case DIF_OP_LDGS:
5805 			id = DIF_INSTR_VAR(instr);
5806 
5807 			if (id >= DIF_VAR_OTHER_UBASE) {
5808 				uintptr_t a;
5809 
5810 				id -= DIF_VAR_OTHER_UBASE;
5811 				svar = vstate->dtvs_globals[id];
5812 				ASSERT(svar != NULL);
5813 				v = &svar->dtsv_var;
5814 
5815 				if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) {
5816 					regs[rd] = svar->dtsv_data;
5817 					break;
5818 				}
5819 
5820 				a = (uintptr_t)svar->dtsv_data;
5821 
5822 				if (*(uint8_t *)a == UINT8_MAX) {
5823 					/*
5824 					 * If the 0th byte is set to UINT8_MAX
5825 					 * then this is to be treated as a
5826 					 * reference to a NULL variable.
5827 					 */
5828 					regs[rd] = NULL;
5829 				} else {
5830 					regs[rd] = a + sizeof (uint64_t);
5831 				}
5832 
5833 				break;
5834 			}
5835 
5836 			regs[rd] = dtrace_dif_variable(mstate, state, id, 0);
5837 			break;
5838 
5839 		case DIF_OP_STGS:
5840 			id = DIF_INSTR_VAR(instr);
5841 
5842 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5843 			id -= DIF_VAR_OTHER_UBASE;
5844 
5845 			svar = vstate->dtvs_globals[id];
5846 			ASSERT(svar != NULL);
5847 			v = &svar->dtsv_var;
5848 
5849 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5850 				uintptr_t a = (uintptr_t)svar->dtsv_data;
5851 
5852 				ASSERT(a != NULL);
5853 				ASSERT(svar->dtsv_size != 0);
5854 
5855 				if (regs[rd] == NULL) {
5856 					*(uint8_t *)a = UINT8_MAX;
5857 					break;
5858 				} else {
5859 					*(uint8_t *)a = 0;
5860 					a += sizeof (uint64_t);
5861 				}
5862 				if (!dtrace_vcanload(
5863 				    (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5864 				    mstate, vstate))
5865 					break;
5866 
5867 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
5868 				    (void *)a, &v->dtdv_type);
5869 				break;
5870 			}
5871 
5872 			svar->dtsv_data = regs[rd];
5873 			break;
5874 
5875 		case DIF_OP_LDTA:
5876 			/*
5877 			 * There are no DTrace built-in thread-local arrays at
5878 			 * present.  This opcode is saved for future work.
5879 			 */
5880 			*flags |= CPU_DTRACE_ILLOP;
5881 			regs[rd] = 0;
5882 			break;
5883 
5884 		case DIF_OP_LDLS:
5885 			id = DIF_INSTR_VAR(instr);
5886 
5887 			if (id < DIF_VAR_OTHER_UBASE) {
5888 				/*
5889 				 * For now, this has no meaning.
5890 				 */
5891 				regs[rd] = 0;
5892 				break;
5893 			}
5894 
5895 			id -= DIF_VAR_OTHER_UBASE;
5896 
5897 			ASSERT(id < vstate->dtvs_nlocals);
5898 			ASSERT(vstate->dtvs_locals != NULL);
5899 
5900 			svar = vstate->dtvs_locals[id];
5901 			ASSERT(svar != NULL);
5902 			v = &svar->dtsv_var;
5903 
5904 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5905 				uintptr_t a = (uintptr_t)svar->dtsv_data;
5906 				size_t sz = v->dtdv_type.dtdt_size;
5907 
5908 				sz += sizeof (uint64_t);
5909 				ASSERT(svar->dtsv_size == NCPU * sz);
5910 				a += CPU->cpu_id * sz;
5911 
5912 				if (*(uint8_t *)a == UINT8_MAX) {
5913 					/*
5914 					 * If the 0th byte is set to UINT8_MAX
5915 					 * then this is to be treated as a
5916 					 * reference to a NULL variable.
5917 					 */
5918 					regs[rd] = NULL;
5919 				} else {
5920 					regs[rd] = a + sizeof (uint64_t);
5921 				}
5922 
5923 				break;
5924 			}
5925 
5926 			ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5927 			tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5928 			regs[rd] = tmp[CPU->cpu_id];
5929 			break;
5930 
5931 		case DIF_OP_STLS:
5932 			id = DIF_INSTR_VAR(instr);
5933 
5934 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5935 			id -= DIF_VAR_OTHER_UBASE;
5936 			ASSERT(id < vstate->dtvs_nlocals);
5937 
5938 			ASSERT(vstate->dtvs_locals != NULL);
5939 			svar = vstate->dtvs_locals[id];
5940 			ASSERT(svar != NULL);
5941 			v = &svar->dtsv_var;
5942 
5943 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5944 				uintptr_t a = (uintptr_t)svar->dtsv_data;
5945 				size_t sz = v->dtdv_type.dtdt_size;
5946 
5947 				sz += sizeof (uint64_t);
5948 				ASSERT(svar->dtsv_size == NCPU * sz);
5949 				a += CPU->cpu_id * sz;
5950 
5951 				if (regs[rd] == NULL) {
5952 					*(uint8_t *)a = UINT8_MAX;
5953 					break;
5954 				} else {
5955 					*(uint8_t *)a = 0;
5956 					a += sizeof (uint64_t);
5957 				}
5958 
5959 				if (!dtrace_vcanload(
5960 				    (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5961 				    mstate, vstate))
5962 					break;
5963 
5964 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
5965 				    (void *)a, &v->dtdv_type);
5966 				break;
5967 			}
5968 
5969 			ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5970 			tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5971 			tmp[CPU->cpu_id] = regs[rd];
5972 			break;
5973 
5974 		case DIF_OP_LDTS: {
5975 			dtrace_dynvar_t *dvar;
5976 			dtrace_key_t *key;
5977 
5978 			id = DIF_INSTR_VAR(instr);
5979 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5980 			id -= DIF_VAR_OTHER_UBASE;
5981 			v = &vstate->dtvs_tlocals[id];
5982 
5983 			key = &tupregs[DIF_DTR_NREGS];
5984 			key[0].dttk_value = (uint64_t)id;
5985 			key[0].dttk_size = 0;
5986 			DTRACE_TLS_THRKEY(key[1].dttk_value);
5987 			key[1].dttk_size = 0;
5988 
5989 			dvar = dtrace_dynvar(dstate, 2, key,
5990 			    sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC,
5991 			    mstate, vstate);
5992 
5993 			if (dvar == NULL) {
5994 				regs[rd] = 0;
5995 				break;
5996 			}
5997 
5998 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5999 				regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
6000 			} else {
6001 				regs[rd] = *((uint64_t *)dvar->dtdv_data);
6002 			}
6003 
6004 			break;
6005 		}
6006 
6007 		case DIF_OP_STTS: {
6008 			dtrace_dynvar_t *dvar;
6009 			dtrace_key_t *key;
6010 
6011 			id = DIF_INSTR_VAR(instr);
6012 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
6013 			id -= DIF_VAR_OTHER_UBASE;
6014 
6015 			key = &tupregs[DIF_DTR_NREGS];
6016 			key[0].dttk_value = (uint64_t)id;
6017 			key[0].dttk_size = 0;
6018 			DTRACE_TLS_THRKEY(key[1].dttk_value);
6019 			key[1].dttk_size = 0;
6020 			v = &vstate->dtvs_tlocals[id];
6021 
6022 			dvar = dtrace_dynvar(dstate, 2, key,
6023 			    v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
6024 			    v->dtdv_type.dtdt_size : sizeof (uint64_t),
6025 			    regs[rd] ? DTRACE_DYNVAR_ALLOC :
6026 			    DTRACE_DYNVAR_DEALLOC, mstate, vstate);
6027 
6028 			/*
6029 			 * Given that we're storing to thread-local data,
6030 			 * we need to flush our predicate cache.
6031 			 */
6032 			curthread->t_predcache = NULL;
6033 
6034 			if (dvar == NULL)
6035 				break;
6036 
6037 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
6038 				if (!dtrace_vcanload(
6039 				    (void *)(uintptr_t)regs[rd],
6040 				    &v->dtdv_type, mstate, vstate))
6041 					break;
6042 
6043 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
6044 				    dvar->dtdv_data, &v->dtdv_type);
6045 			} else {
6046 				*((uint64_t *)dvar->dtdv_data) = regs[rd];
6047 			}
6048 
6049 			break;
6050 		}
6051 
6052 		case DIF_OP_SRA:
6053 			regs[rd] = (int64_t)regs[r1] >> regs[r2];
6054 			break;
6055 
6056 		case DIF_OP_CALL:
6057 			dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd,
6058 			    regs, tupregs, ttop, mstate, state);
6059 			break;
6060 
6061 		case DIF_OP_PUSHTR:
6062 			if (ttop == DIF_DTR_NREGS) {
6063 				*flags |= CPU_DTRACE_TUPOFLOW;
6064 				break;
6065 			}
6066 
6067 			if (r1 == DIF_TYPE_STRING) {
6068 				/*
6069 				 * If this is a string type and the size is 0,
6070 				 * we'll use the system-wide default string
6071 				 * size.  Note that we are _not_ looking at
6072 				 * the value of the DTRACEOPT_STRSIZE option;
6073 				 * had this been set, we would expect to have
6074 				 * a non-zero size value in the "pushtr".
6075 				 */
6076 				tupregs[ttop].dttk_size =
6077 				    dtrace_strlen((char *)(uintptr_t)regs[rd],
6078 				    regs[r2] ? regs[r2] :
6079 				    dtrace_strsize_default) + 1;
6080 			} else {
6081 				tupregs[ttop].dttk_size = regs[r2];
6082 			}
6083 
6084 			tupregs[ttop++].dttk_value = regs[rd];
6085 			break;
6086 
6087 		case DIF_OP_PUSHTV:
6088 			if (ttop == DIF_DTR_NREGS) {
6089 				*flags |= CPU_DTRACE_TUPOFLOW;
6090 				break;
6091 			}
6092 
6093 			tupregs[ttop].dttk_value = regs[rd];
6094 			tupregs[ttop++].dttk_size = 0;
6095 			break;
6096 
6097 		case DIF_OP_POPTS:
6098 			if (ttop != 0)
6099 				ttop--;
6100 			break;
6101 
6102 		case DIF_OP_FLUSHTS:
6103 			ttop = 0;
6104 			break;
6105 
6106 		case DIF_OP_LDGAA:
6107 		case DIF_OP_LDTAA: {
6108 			dtrace_dynvar_t *dvar;
6109 			dtrace_key_t *key = tupregs;
6110 			uint_t nkeys = ttop;
6111 
6112 			id = DIF_INSTR_VAR(instr);
6113 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
6114 			id -= DIF_VAR_OTHER_UBASE;
6115 
6116 			key[nkeys].dttk_value = (uint64_t)id;
6117 			key[nkeys++].dttk_size = 0;
6118 
6119 			if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) {
6120 				DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
6121 				key[nkeys++].dttk_size = 0;
6122 				v = &vstate->dtvs_tlocals[id];
6123 			} else {
6124 				v = &vstate->dtvs_globals[id]->dtsv_var;
6125 			}
6126 
6127 			dvar = dtrace_dynvar(dstate, nkeys, key,
6128 			    v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
6129 			    v->dtdv_type.dtdt_size : sizeof (uint64_t),
6130 			    DTRACE_DYNVAR_NOALLOC, mstate, vstate);
6131 
6132 			if (dvar == NULL) {
6133 				regs[rd] = 0;
6134 				break;
6135 			}
6136 
6137 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
6138 				regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
6139 			} else {
6140 				regs[rd] = *((uint64_t *)dvar->dtdv_data);
6141 			}
6142 
6143 			break;
6144 		}
6145 
6146 		case DIF_OP_STGAA:
6147 		case DIF_OP_STTAA: {
6148 			dtrace_dynvar_t *dvar;
6149 			dtrace_key_t *key = tupregs;
6150 			uint_t nkeys = ttop;
6151 
6152 			id = DIF_INSTR_VAR(instr);
6153 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
6154 			id -= DIF_VAR_OTHER_UBASE;
6155 
6156 			key[nkeys].dttk_value = (uint64_t)id;
6157 			key[nkeys++].dttk_size = 0;
6158 
6159 			if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) {
6160 				DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
6161 				key[nkeys++].dttk_size = 0;
6162 				v = &vstate->dtvs_tlocals[id];
6163 			} else {
6164 				v = &vstate->dtvs_globals[id]->dtsv_var;
6165 			}
6166 
6167 			dvar = dtrace_dynvar(dstate, nkeys, key,
6168 			    v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
6169 			    v->dtdv_type.dtdt_size : sizeof (uint64_t),
6170 			    regs[rd] ? DTRACE_DYNVAR_ALLOC :
6171 			    DTRACE_DYNVAR_DEALLOC, mstate, vstate);
6172 
6173 			if (dvar == NULL)
6174 				break;
6175 
6176 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
6177 				if (!dtrace_vcanload(
6178 				    (void *)(uintptr_t)regs[rd], &v->dtdv_type,
6179 				    mstate, vstate))
6180 					break;
6181 
6182 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
6183 				    dvar->dtdv_data, &v->dtdv_type);
6184 			} else {
6185 				*((uint64_t *)dvar->dtdv_data) = regs[rd];
6186 			}
6187 
6188 			break;
6189 		}
6190 
6191 		case DIF_OP_ALLOCS: {
6192 			uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
6193 			size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1];
6194 
6195 			/*
6196 			 * Rounding up the user allocation size could have
6197 			 * overflowed large, bogus allocations (like -1ULL) to
6198 			 * 0.
6199 			 */
6200 			if (size < regs[r1] ||
6201 			    !DTRACE_INSCRATCH(mstate, size)) {
6202 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
6203 				regs[rd] = NULL;
6204 				break;
6205 			}
6206 
6207 			dtrace_bzero((void *) mstate->dtms_scratch_ptr, size);
6208 			mstate->dtms_scratch_ptr += size;
6209 			regs[rd] = ptr;
6210 			break;
6211 		}
6212 
6213 		case DIF_OP_COPYS:
6214 			if (!dtrace_canstore(regs[rd], regs[r2],
6215 			    mstate, vstate)) {
6216 				*flags |= CPU_DTRACE_BADADDR;
6217 				*illval = regs[rd];
6218 				break;
6219 			}
6220 
6221 			if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate))
6222 				break;
6223 
6224 			dtrace_bcopy((void *)(uintptr_t)regs[r1],
6225 			    (void *)(uintptr_t)regs[rd], (size_t)regs[r2]);
6226 			break;
6227 
6228 		case DIF_OP_STB:
6229 			if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) {
6230 				*flags |= CPU_DTRACE_BADADDR;
6231 				*illval = regs[rd];
6232 				break;
6233 			}
6234 			*((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1];
6235 			break;
6236 
6237 		case DIF_OP_STH:
6238 			if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) {
6239 				*flags |= CPU_DTRACE_BADADDR;
6240 				*illval = regs[rd];
6241 				break;
6242 			}
6243 			if (regs[rd] & 1) {
6244 				*flags |= CPU_DTRACE_BADALIGN;
6245 				*illval = regs[rd];
6246 				break;
6247 			}
6248 			*((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1];
6249 			break;
6250 
6251 		case DIF_OP_STW:
6252 			if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) {
6253 				*flags |= CPU_DTRACE_BADADDR;
6254 				*illval = regs[rd];
6255 				break;
6256 			}
6257 			if (regs[rd] & 3) {
6258 				*flags |= CPU_DTRACE_BADALIGN;
6259 				*illval = regs[rd];
6260 				break;
6261 			}
6262 			*((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1];
6263 			break;
6264 
6265 		case DIF_OP_STX:
6266 			if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) {
6267 				*flags |= CPU_DTRACE_BADADDR;
6268 				*illval = regs[rd];
6269 				break;
6270 			}
6271 			if (regs[rd] & 7) {
6272 				*flags |= CPU_DTRACE_BADALIGN;
6273 				*illval = regs[rd];
6274 				break;
6275 			}
6276 			*((uint64_t *)(uintptr_t)regs[rd]) = regs[r1];
6277 			break;
6278 		}
6279 	}
6280 
6281 	if (!(*flags & CPU_DTRACE_FAULT))
6282 		return (rval);
6283 
6284 	mstate->dtms_fltoffs = opc * sizeof (dif_instr_t);
6285 	mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS;
6286 
6287 	return (0);
6288 }
6289 
6290 static void
dtrace_action_breakpoint(dtrace_ecb_t * ecb)6291 dtrace_action_breakpoint(dtrace_ecb_t *ecb)
6292 {
6293 	dtrace_probe_t *probe = ecb->dte_probe;
6294 	dtrace_provider_t *prov = probe->dtpr_provider;
6295 	char c[DTRACE_FULLNAMELEN + 80], *str;
6296 	char *msg = "dtrace: breakpoint action at probe ";
6297 	char *ecbmsg = " (ecb ";
6298 	uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4));
6299 	uintptr_t val = (uintptr_t)ecb;
6300 	int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0;
6301 
6302 	if (dtrace_destructive_disallow)
6303 		return;
6304 
6305 	/*
6306 	 * It's impossible to be taking action on the NULL probe.
6307 	 */
6308 	ASSERT(probe != NULL);
6309 
6310 	/*
6311 	 * This is a poor man's (destitute man's?) sprintf():  we want to
6312 	 * print the provider name, module name, function name and name of
6313 	 * the probe, along with the hex address of the ECB with the breakpoint
6314 	 * action -- all of which we must place in the character buffer by
6315 	 * hand.
6316 	 */
6317 	while (*msg != '\0')
6318 		c[i++] = *msg++;
6319 
6320 	for (str = prov->dtpv_name; *str != '\0'; str++)
6321 		c[i++] = *str;
6322 	c[i++] = ':';
6323 
6324 	for (str = probe->dtpr_mod; *str != '\0'; str++)
6325 		c[i++] = *str;
6326 	c[i++] = ':';
6327 
6328 	for (str = probe->dtpr_func; *str != '\0'; str++)
6329 		c[i++] = *str;
6330 	c[i++] = ':';
6331 
6332 	for (str = probe->dtpr_name; *str != '\0'; str++)
6333 		c[i++] = *str;
6334 
6335 	while (*ecbmsg != '\0')
6336 		c[i++] = *ecbmsg++;
6337 
6338 	while (shift >= 0) {
6339 		mask = (uintptr_t)0xf << shift;
6340 
6341 		if (val >= ((uintptr_t)1 << shift))
6342 			c[i++] = "0123456789abcdef"[(val & mask) >> shift];
6343 		shift -= 4;
6344 	}
6345 
6346 	c[i++] = ')';
6347 	c[i] = '\0';
6348 
6349 	debug_enter(c);
6350 }
6351 
6352 static void
dtrace_action_panic(dtrace_ecb_t * ecb)6353 dtrace_action_panic(dtrace_ecb_t *ecb)
6354 {
6355 	dtrace_probe_t *probe = ecb->dte_probe;
6356 
6357 	/*
6358 	 * It's impossible to be taking action on the NULL probe.
6359 	 */
6360 	ASSERT(probe != NULL);
6361 
6362 	if (dtrace_destructive_disallow)
6363 		return;
6364 
6365 	if (dtrace_panicked != NULL)
6366 		return;
6367 
6368 	if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL)
6369 		return;
6370 
6371 	/*
6372 	 * We won the right to panic.  (We want to be sure that only one
6373 	 * thread calls panic() from dtrace_probe(), and that panic() is
6374 	 * called exactly once.)
6375 	 */
6376 	dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)",
6377 	    probe->dtpr_provider->dtpv_name, probe->dtpr_mod,
6378 	    probe->dtpr_func, probe->dtpr_name, (void *)ecb);
6379 }
6380 
6381 static void
dtrace_action_raise(uint64_t sig)6382 dtrace_action_raise(uint64_t sig)
6383 {
6384 	if (dtrace_destructive_disallow)
6385 		return;
6386 
6387 	if (sig >= NSIG) {
6388 		DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
6389 		return;
6390 	}
6391 
6392 	/*
6393 	 * raise() has a queue depth of 1 -- we ignore all subsequent
6394 	 * invocations of the raise() action.
6395 	 */
6396 	if (curthread->t_dtrace_sig == 0)
6397 		curthread->t_dtrace_sig = (uint8_t)sig;
6398 
6399 	curthread->t_sig_check = 1;
6400 	aston(curthread);
6401 }
6402 
6403 static void
dtrace_action_stop(void)6404 dtrace_action_stop(void)
6405 {
6406 	if (dtrace_destructive_disallow)
6407 		return;
6408 
6409 	if (!curthread->t_dtrace_stop) {
6410 		curthread->t_dtrace_stop = 1;
6411 		curthread->t_sig_check = 1;
6412 		aston(curthread);
6413 	}
6414 }
6415 
6416 static void
dtrace_action_chill(dtrace_mstate_t * mstate,hrtime_t val)6417 dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val)
6418 {
6419 	hrtime_t now;
6420 	volatile uint16_t *flags;
6421 	cpu_t *cpu = CPU;
6422 
6423 	if (dtrace_destructive_disallow)
6424 		return;
6425 
6426 	flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags;
6427 
6428 	now = dtrace_gethrtime();
6429 
6430 	if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) {
6431 		/*
6432 		 * We need to advance the mark to the current time.
6433 		 */
6434 		cpu->cpu_dtrace_chillmark = now;
6435 		cpu->cpu_dtrace_chilled = 0;
6436 	}
6437 
6438 	/*
6439 	 * Now check to see if the requested chill time would take us over
6440 	 * the maximum amount of time allowed in the chill interval.  (Or
6441 	 * worse, if the calculation itself induces overflow.)
6442 	 */
6443 	if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max ||
6444 	    cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) {
6445 		*flags |= CPU_DTRACE_ILLOP;
6446 		return;
6447 	}
6448 
6449 	while (dtrace_gethrtime() - now < val)
6450 		continue;
6451 
6452 	/*
6453 	 * Normally, we assure that the value of the variable "timestamp" does
6454 	 * not change within an ECB.  The presence of chill() represents an
6455 	 * exception to this rule, however.
6456 	 */
6457 	mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP;
6458 	cpu->cpu_dtrace_chilled += val;
6459 }
6460 
6461 static void
dtrace_action_ustack(dtrace_mstate_t * mstate,dtrace_state_t * state,uint64_t * buf,uint64_t arg)6462 dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state,
6463     uint64_t *buf, uint64_t arg)
6464 {
6465 	int nframes = DTRACE_USTACK_NFRAMES(arg);
6466 	int strsize = DTRACE_USTACK_STRSIZE(arg);
6467 	uint64_t *pcs = &buf[1], *fps;
6468 	char *str = (char *)&pcs[nframes];
6469 	int size, offs = 0, i, j;
6470 	uintptr_t old = mstate->dtms_scratch_ptr, saved;
6471 	uint16_t *flags = &cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
6472 	char *sym;
6473 
6474 	/*
6475 	 * Should be taking a faster path if string space has not been
6476 	 * allocated.
6477 	 */
6478 	ASSERT(strsize != 0);
6479 
6480 	/*
6481 	 * We will first allocate some temporary space for the frame pointers.
6482 	 */
6483 	fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
6484 	size = (uintptr_t)fps - mstate->dtms_scratch_ptr +
6485 	    (nframes * sizeof (uint64_t));
6486 
6487 	if (!DTRACE_INSCRATCH(mstate, size)) {
6488 		/*
6489 		 * Not enough room for our frame pointers -- need to indicate
6490 		 * that we ran out of scratch space.
6491 		 */
6492 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
6493 		return;
6494 	}
6495 
6496 	mstate->dtms_scratch_ptr += size;
6497 	saved = mstate->dtms_scratch_ptr;
6498 
6499 	/*
6500 	 * Now get a stack with both program counters and frame pointers.
6501 	 */
6502 	DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6503 	dtrace_getufpstack(buf, fps, nframes + 1);
6504 	DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6505 
6506 	/*
6507 	 * If that faulted, we're cooked.
6508 	 */
6509 	if (*flags & CPU_DTRACE_FAULT)
6510 		goto out;
6511 
6512 	/*
6513 	 * Now we want to walk up the stack, calling the USTACK helper.  For
6514 	 * each iteration, we restore the scratch pointer.
6515 	 */
6516 	for (i = 0; i < nframes; i++) {
6517 		mstate->dtms_scratch_ptr = saved;
6518 
6519 		if (offs >= strsize)
6520 			break;
6521 
6522 		sym = (char *)(uintptr_t)dtrace_helper(
6523 		    DTRACE_HELPER_ACTION_USTACK,
6524 		    mstate, state, pcs[i], fps[i]);
6525 
6526 		/*
6527 		 * If we faulted while running the helper, we're going to
6528 		 * clear the fault and null out the corresponding string.
6529 		 */
6530 		if (*flags & CPU_DTRACE_FAULT) {
6531 			*flags &= ~CPU_DTRACE_FAULT;
6532 			str[offs++] = '\0';
6533 			continue;
6534 		}
6535 
6536 		if (sym == NULL) {
6537 			str[offs++] = '\0';
6538 			continue;
6539 		}
6540 
6541 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6542 
6543 		/*
6544 		 * Now copy in the string that the helper returned to us.
6545 		 */
6546 		for (j = 0; offs + j < strsize; j++) {
6547 			if ((str[offs + j] = sym[j]) == '\0')
6548 				break;
6549 		}
6550 
6551 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6552 
6553 		offs += j + 1;
6554 	}
6555 
6556 	if (offs >= strsize) {
6557 		/*
6558 		 * If we didn't have room for all of the strings, we don't
6559 		 * abort processing -- this needn't be a fatal error -- but we
6560 		 * still want to increment a counter (dts_stkstroverflows) to
6561 		 * allow this condition to be warned about.  (If this is from
6562 		 * a jstack() action, it is easily tuned via jstackstrsize.)
6563 		 */
6564 		dtrace_error(&state->dts_stkstroverflows);
6565 	}
6566 
6567 	while (offs < strsize)
6568 		str[offs++] = '\0';
6569 
6570 out:
6571 	mstate->dtms_scratch_ptr = old;
6572 }
6573 
6574 static void
dtrace_store_by_ref(dtrace_difo_t * dp,caddr_t tomax,size_t size,size_t * valoffsp,uint64_t * valp,uint64_t end,int intuple,int dtkind)6575 dtrace_store_by_ref(dtrace_difo_t *dp, caddr_t tomax, size_t size,
6576     size_t *valoffsp, uint64_t *valp, uint64_t end, int intuple, int dtkind)
6577 {
6578 	volatile uint16_t *flags;
6579 	uint64_t val = *valp;
6580 	size_t valoffs = *valoffsp;
6581 
6582 	flags = (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
6583 	ASSERT(dtkind == DIF_TF_BYREF || dtkind == DIF_TF_BYUREF);
6584 
6585 	/*
6586 	 * If this is a string, we're going to only load until we find the zero
6587 	 * byte -- after which we'll store zero bytes.
6588 	 */
6589 	if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) {
6590 		char c = '\0' + 1;
6591 		size_t s;
6592 
6593 		for (s = 0; s < size; s++) {
6594 			if (c != '\0' && dtkind == DIF_TF_BYREF) {
6595 				c = dtrace_load8(val++);
6596 			} else if (c != '\0' && dtkind == DIF_TF_BYUREF) {
6597 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6598 				c = dtrace_fuword8((void *)(uintptr_t)val++);
6599 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6600 				if (*flags & CPU_DTRACE_FAULT)
6601 					break;
6602 			}
6603 
6604 			DTRACE_STORE(uint8_t, tomax, valoffs++, c);
6605 
6606 			if (c == '\0' && intuple)
6607 				break;
6608 		}
6609 	} else {
6610 		uint8_t c;
6611 		while (valoffs < end) {
6612 			if (dtkind == DIF_TF_BYREF) {
6613 				c = dtrace_load8(val++);
6614 			} else if (dtkind == DIF_TF_BYUREF) {
6615 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6616 				c = dtrace_fuword8((void *)(uintptr_t)val++);
6617 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6618 				if (*flags & CPU_DTRACE_FAULT)
6619 					break;
6620 			}
6621 
6622 			DTRACE_STORE(uint8_t, tomax,
6623 			    valoffs++, c);
6624 		}
6625 	}
6626 
6627 	*valp = val;
6628 	*valoffsp = valoffs;
6629 }
6630 
6631 /*
6632  * If you're looking for the epicenter of DTrace, you just found it.  This
6633  * is the function called by the provider to fire a probe -- from which all
6634  * subsequent probe-context DTrace activity emanates.
6635  */
6636 void
dtrace_probe(dtrace_id_t id,uintptr_t arg0,uintptr_t arg1,uintptr_t arg2,uintptr_t arg3,uintptr_t arg4)6637 dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1,
6638     uintptr_t arg2, uintptr_t arg3, uintptr_t arg4)
6639 {
6640 	processorid_t cpuid;
6641 	dtrace_icookie_t cookie;
6642 	dtrace_probe_t *probe;
6643 	dtrace_mstate_t mstate;
6644 	dtrace_ecb_t *ecb;
6645 	dtrace_action_t *act;
6646 	intptr_t offs;
6647 	size_t size;
6648 	int vtime, onintr;
6649 	volatile uint16_t *flags;
6650 	hrtime_t now, end;
6651 
6652 	/*
6653 	 * Kick out immediately if this CPU is still being born (in which case
6654 	 * curthread will be set to -1) or the current thread can't allow
6655 	 * probes in its current context.
6656 	 */
6657 	if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE))
6658 		return;
6659 
6660 	cookie = dtrace_interrupt_disable();
6661 	probe = dtrace_probes[id - 1];
6662 	cpuid = CPU->cpu_id;
6663 	onintr = CPU_ON_INTR(CPU);
6664 
6665 	CPU->cpu_dtrace_probes++;
6666 
6667 	if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE &&
6668 	    probe->dtpr_predcache == curthread->t_predcache) {
6669 		/*
6670 		 * We have hit in the predicate cache; we know that
6671 		 * this predicate would evaluate to be false.
6672 		 */
6673 		dtrace_interrupt_enable(cookie);
6674 		return;
6675 	}
6676 
6677 	if (panic_quiesce) {
6678 		/*
6679 		 * We don't trace anything if we're panicking.
6680 		 */
6681 		dtrace_interrupt_enable(cookie);
6682 		return;
6683 	}
6684 
6685 	now = mstate.dtms_timestamp = dtrace_gethrtime();
6686 	mstate.dtms_present |= DTRACE_MSTATE_TIMESTAMP;
6687 	vtime = dtrace_vtime_references != 0;
6688 
6689 	if (vtime && curthread->t_dtrace_start)
6690 		curthread->t_dtrace_vtime += now - curthread->t_dtrace_start;
6691 
6692 	mstate.dtms_difo = NULL;
6693 	mstate.dtms_probe = probe;
6694 	mstate.dtms_strtok = NULL;
6695 	mstate.dtms_arg[0] = arg0;
6696 	mstate.dtms_arg[1] = arg1;
6697 	mstate.dtms_arg[2] = arg2;
6698 	mstate.dtms_arg[3] = arg3;
6699 	mstate.dtms_arg[4] = arg4;
6700 
6701 	flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags;
6702 
6703 	for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
6704 		dtrace_predicate_t *pred = ecb->dte_predicate;
6705 		dtrace_state_t *state = ecb->dte_state;
6706 		dtrace_buffer_t *buf = &state->dts_buffer[cpuid];
6707 		dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid];
6708 		dtrace_vstate_t *vstate = &state->dts_vstate;
6709 		dtrace_provider_t *prov = probe->dtpr_provider;
6710 		uint64_t tracememsize = 0;
6711 		int committed = 0;
6712 		caddr_t tomax;
6713 
6714 		/*
6715 		 * A little subtlety with the following (seemingly innocuous)
6716 		 * declaration of the automatic 'val':  by looking at the
6717 		 * code, you might think that it could be declared in the
6718 		 * action processing loop, below.  (That is, it's only used in
6719 		 * the action processing loop.)  However, it must be declared
6720 		 * out of that scope because in the case of DIF expression
6721 		 * arguments to aggregating actions, one iteration of the
6722 		 * action loop will use the last iteration's value.
6723 		 */
6724 #ifdef lint
6725 		uint64_t val = 0;
6726 #else
6727 		uint64_t val;
6728 #endif
6729 
6730 		mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE;
6731 		mstate.dtms_access = DTRACE_ACCESS_ARGS | DTRACE_ACCESS_PROC;
6732 		mstate.dtms_getf = NULL;
6733 
6734 		*flags &= ~CPU_DTRACE_ERROR;
6735 
6736 		if (prov == dtrace_provider) {
6737 			/*
6738 			 * If dtrace itself is the provider of this probe,
6739 			 * we're only going to continue processing the ECB if
6740 			 * arg0 (the dtrace_state_t) is equal to the ECB's
6741 			 * creating state.  (This prevents disjoint consumers
6742 			 * from seeing one another's metaprobes.)
6743 			 */
6744 			if (arg0 != (uint64_t)(uintptr_t)state)
6745 				continue;
6746 		}
6747 
6748 		if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) {
6749 			/*
6750 			 * We're not currently active.  If our provider isn't
6751 			 * the dtrace pseudo provider, we're not interested.
6752 			 */
6753 			if (prov != dtrace_provider)
6754 				continue;
6755 
6756 			/*
6757 			 * Now we must further check if we are in the BEGIN
6758 			 * probe.  If we are, we will only continue processing
6759 			 * if we're still in WARMUP -- if one BEGIN enabling
6760 			 * has invoked the exit() action, we don't want to
6761 			 * evaluate subsequent BEGIN enablings.
6762 			 */
6763 			if (probe->dtpr_id == dtrace_probeid_begin &&
6764 			    state->dts_activity != DTRACE_ACTIVITY_WARMUP) {
6765 				ASSERT(state->dts_activity ==
6766 				    DTRACE_ACTIVITY_DRAINING);
6767 				continue;
6768 			}
6769 		}
6770 
6771 		if (ecb->dte_cond && !dtrace_priv_probe(state, &mstate, ecb))
6772 			continue;
6773 
6774 		if (now - state->dts_alive > dtrace_deadman_timeout) {
6775 			/*
6776 			 * We seem to be dead.  Unless we (a) have kernel
6777 			 * destructive permissions (b) have explicitly enabled
6778 			 * destructive actions and (c) destructive actions have
6779 			 * not been disabled, we're going to transition into
6780 			 * the KILLED state, from which no further processing
6781 			 * on this state will be performed.
6782 			 */
6783 			if (!dtrace_priv_kernel_destructive(state) ||
6784 			    !state->dts_cred.dcr_destructive ||
6785 			    dtrace_destructive_disallow) {
6786 				void *activity = &state->dts_activity;
6787 				dtrace_activity_t current;
6788 
6789 				do {
6790 					current = state->dts_activity;
6791 				} while (dtrace_cas32(activity, current,
6792 				    DTRACE_ACTIVITY_KILLED) != current);
6793 
6794 				continue;
6795 			}
6796 		}
6797 
6798 		if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed,
6799 		    ecb->dte_alignment, state, &mstate)) < 0)
6800 			continue;
6801 
6802 		tomax = buf->dtb_tomax;
6803 		ASSERT(tomax != NULL);
6804 
6805 		if (ecb->dte_size != 0) {
6806 			dtrace_rechdr_t dtrh;
6807 			if (!(mstate.dtms_present & DTRACE_MSTATE_TIMESTAMP)) {
6808 				mstate.dtms_timestamp = dtrace_gethrtime();
6809 				mstate.dtms_present |= DTRACE_MSTATE_TIMESTAMP;
6810 			}
6811 			ASSERT3U(ecb->dte_size, >=, sizeof (dtrace_rechdr_t));
6812 			dtrh.dtrh_epid = ecb->dte_epid;
6813 			DTRACE_RECORD_STORE_TIMESTAMP(&dtrh,
6814 			    mstate.dtms_timestamp);
6815 			*((dtrace_rechdr_t *)(tomax + offs)) = dtrh;
6816 		}
6817 
6818 		mstate.dtms_epid = ecb->dte_epid;
6819 		mstate.dtms_present |= DTRACE_MSTATE_EPID;
6820 
6821 		if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL)
6822 			mstate.dtms_access |= DTRACE_ACCESS_KERNEL;
6823 
6824 		if (pred != NULL) {
6825 			dtrace_difo_t *dp = pred->dtp_difo;
6826 			int rval;
6827 
6828 			rval = dtrace_dif_emulate(dp, &mstate, vstate, state);
6829 
6830 			if (!(*flags & CPU_DTRACE_ERROR) && !rval) {
6831 				dtrace_cacheid_t cid = probe->dtpr_predcache;
6832 
6833 				if (cid != DTRACE_CACHEIDNONE && !onintr) {
6834 					/*
6835 					 * Update the predicate cache...
6836 					 */
6837 					ASSERT(cid == pred->dtp_cacheid);
6838 					curthread->t_predcache = cid;
6839 				}
6840 
6841 				continue;
6842 			}
6843 		}
6844 
6845 		for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) &&
6846 		    act != NULL; act = act->dta_next) {
6847 			size_t valoffs;
6848 			dtrace_difo_t *dp;
6849 			dtrace_recdesc_t *rec = &act->dta_rec;
6850 
6851 			size = rec->dtrd_size;
6852 			valoffs = offs + rec->dtrd_offset;
6853 
6854 			if (DTRACEACT_ISAGG(act->dta_kind)) {
6855 				uint64_t v = 0xbad;
6856 				dtrace_aggregation_t *agg;
6857 
6858 				agg = (dtrace_aggregation_t *)act;
6859 
6860 				if ((dp = act->dta_difo) != NULL)
6861 					v = dtrace_dif_emulate(dp,
6862 					    &mstate, vstate, state);
6863 
6864 				if (*flags & CPU_DTRACE_ERROR)
6865 					continue;
6866 
6867 				/*
6868 				 * Note that we always pass the expression
6869 				 * value from the previous iteration of the
6870 				 * action loop.  This value will only be used
6871 				 * if there is an expression argument to the
6872 				 * aggregating action, denoted by the
6873 				 * dtag_hasarg field.
6874 				 */
6875 				dtrace_aggregate(agg, buf,
6876 				    offs, aggbuf, v, val);
6877 				continue;
6878 			}
6879 
6880 			switch (act->dta_kind) {
6881 			case DTRACEACT_STOP:
6882 				if (dtrace_priv_proc_destructive(state,
6883 				    &mstate))
6884 					dtrace_action_stop();
6885 				continue;
6886 
6887 			case DTRACEACT_BREAKPOINT:
6888 				if (dtrace_priv_kernel_destructive(state))
6889 					dtrace_action_breakpoint(ecb);
6890 				continue;
6891 
6892 			case DTRACEACT_PANIC:
6893 				if (dtrace_priv_kernel_destructive(state))
6894 					dtrace_action_panic(ecb);
6895 				continue;
6896 
6897 			case DTRACEACT_STACK:
6898 				if (!dtrace_priv_kernel(state))
6899 					continue;
6900 
6901 				dtrace_getpcstack((pc_t *)(tomax + valoffs),
6902 				    size / sizeof (pc_t), probe->dtpr_aframes,
6903 				    DTRACE_ANCHORED(probe) ? NULL :
6904 				    (uint32_t *)arg0);
6905 
6906 				continue;
6907 
6908 			case DTRACEACT_JSTACK:
6909 			case DTRACEACT_USTACK:
6910 				if (!dtrace_priv_proc(state, &mstate))
6911 					continue;
6912 
6913 				/*
6914 				 * See comment in DIF_VAR_PID.
6915 				 */
6916 				if (DTRACE_ANCHORED(mstate.dtms_probe) &&
6917 				    CPU_ON_INTR(CPU)) {
6918 					int depth = DTRACE_USTACK_NFRAMES(
6919 					    rec->dtrd_arg) + 1;
6920 
6921 					dtrace_bzero((void *)(tomax + valoffs),
6922 					    DTRACE_USTACK_STRSIZE(rec->dtrd_arg)
6923 					    + depth * sizeof (uint64_t));
6924 
6925 					continue;
6926 				}
6927 
6928 				if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 &&
6929 				    curproc->p_dtrace_helpers != NULL) {
6930 					/*
6931 					 * This is the slow path -- we have
6932 					 * allocated string space, and we're
6933 					 * getting the stack of a process that
6934 					 * has helpers.  Call into a separate
6935 					 * routine to perform this processing.
6936 					 */
6937 					dtrace_action_ustack(&mstate, state,
6938 					    (uint64_t *)(tomax + valoffs),
6939 					    rec->dtrd_arg);
6940 					continue;
6941 				}
6942 
6943 				/*
6944 				 * Clear the string space, since there's no
6945 				 * helper to do it for us.
6946 				 */
6947 				if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0) {
6948 					int depth = DTRACE_USTACK_NFRAMES(
6949 					    rec->dtrd_arg);
6950 					size_t strsize = DTRACE_USTACK_STRSIZE(
6951 					    rec->dtrd_arg);
6952 					uint64_t *buf = (uint64_t *)(tomax +
6953 					    valoffs);
6954 					void *strspace = &buf[depth + 1];
6955 
6956 					dtrace_bzero(strspace,
6957 					    MIN(depth, strsize));
6958 				}
6959 
6960 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6961 				dtrace_getupcstack((uint64_t *)
6962 				    (tomax + valoffs),
6963 				    DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1);
6964 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6965 				continue;
6966 
6967 			default:
6968 				break;
6969 			}
6970 
6971 			dp = act->dta_difo;
6972 			ASSERT(dp != NULL);
6973 
6974 			val = dtrace_dif_emulate(dp, &mstate, vstate, state);
6975 
6976 			if (*flags & CPU_DTRACE_ERROR)
6977 				continue;
6978 
6979 			switch (act->dta_kind) {
6980 			case DTRACEACT_SPECULATE: {
6981 				dtrace_rechdr_t *dtrh;
6982 
6983 				ASSERT(buf == &state->dts_buffer[cpuid]);
6984 				buf = dtrace_speculation_buffer(state,
6985 				    cpuid, val);
6986 
6987 				if (buf == NULL) {
6988 					*flags |= CPU_DTRACE_DROP;
6989 					continue;
6990 				}
6991 
6992 				offs = dtrace_buffer_reserve(buf,
6993 				    ecb->dte_needed, ecb->dte_alignment,
6994 				    state, NULL);
6995 
6996 				if (offs < 0) {
6997 					*flags |= CPU_DTRACE_DROP;
6998 					continue;
6999 				}
7000 
7001 				tomax = buf->dtb_tomax;
7002 				ASSERT(tomax != NULL);
7003 
7004 				if (ecb->dte_size == 0)
7005 					continue;
7006 
7007 				ASSERT3U(ecb->dte_size, >=,
7008 				    sizeof (dtrace_rechdr_t));
7009 				dtrh = ((void *)(tomax + offs));
7010 				dtrh->dtrh_epid = ecb->dte_epid;
7011 				/*
7012 				 * When the speculation is committed, all of
7013 				 * the records in the speculative buffer will
7014 				 * have their timestamps set to the commit
7015 				 * time.  Until then, it is set to a sentinel
7016 				 * value, for debugability.
7017 				 */
7018 				DTRACE_RECORD_STORE_TIMESTAMP(dtrh, UINT64_MAX);
7019 				continue;
7020 			}
7021 
7022 			case DTRACEACT_CHILL:
7023 				if (dtrace_priv_kernel_destructive(state))
7024 					dtrace_action_chill(&mstate, val);
7025 				continue;
7026 
7027 			case DTRACEACT_RAISE:
7028 				if (dtrace_priv_proc_destructive(state,
7029 				    &mstate))
7030 					dtrace_action_raise(val);
7031 				continue;
7032 
7033 			case DTRACEACT_COMMIT:
7034 				ASSERT(!committed);
7035 
7036 				/*
7037 				 * We need to commit our buffer state.
7038 				 */
7039 				if (ecb->dte_size)
7040 					buf->dtb_offset = offs + ecb->dte_size;
7041 				buf = &state->dts_buffer[cpuid];
7042 				dtrace_speculation_commit(state, cpuid, val);
7043 				committed = 1;
7044 				continue;
7045 
7046 			case DTRACEACT_DISCARD:
7047 				dtrace_speculation_discard(state, cpuid, val);
7048 				continue;
7049 
7050 			case DTRACEACT_DIFEXPR:
7051 			case DTRACEACT_LIBACT:
7052 			case DTRACEACT_PRINTF:
7053 			case DTRACEACT_PRINTA:
7054 			case DTRACEACT_SYSTEM:
7055 			case DTRACEACT_FREOPEN:
7056 			case DTRACEACT_TRACEMEM:
7057 				break;
7058 
7059 			case DTRACEACT_TRACEMEM_DYNSIZE:
7060 				tracememsize = val;
7061 				break;
7062 
7063 			case DTRACEACT_SYM:
7064 			case DTRACEACT_MOD:
7065 				if (!dtrace_priv_kernel(state))
7066 					continue;
7067 				break;
7068 
7069 			case DTRACEACT_USYM:
7070 			case DTRACEACT_UMOD:
7071 			case DTRACEACT_UADDR: {
7072 				struct pid *pid = curthread->t_procp->p_pidp;
7073 
7074 				if (!dtrace_priv_proc(state, &mstate))
7075 					continue;
7076 
7077 				DTRACE_STORE(uint64_t, tomax,
7078 				    valoffs, (uint64_t)pid->pid_id);
7079 				DTRACE_STORE(uint64_t, tomax,
7080 				    valoffs + sizeof (uint64_t), val);
7081 
7082 				continue;
7083 			}
7084 
7085 			case DTRACEACT_EXIT: {
7086 				/*
7087 				 * For the exit action, we are going to attempt
7088 				 * to atomically set our activity to be
7089 				 * draining.  If this fails (either because
7090 				 * another CPU has beat us to the exit action,
7091 				 * or because our current activity is something
7092 				 * other than ACTIVE or WARMUP), we will
7093 				 * continue.  This assures that the exit action
7094 				 * can be successfully recorded at most once
7095 				 * when we're in the ACTIVE state.  If we're
7096 				 * encountering the exit() action while in
7097 				 * COOLDOWN, however, we want to honor the new
7098 				 * status code.  (We know that we're the only
7099 				 * thread in COOLDOWN, so there is no race.)
7100 				 */
7101 				void *activity = &state->dts_activity;
7102 				dtrace_activity_t current = state->dts_activity;
7103 
7104 				if (current == DTRACE_ACTIVITY_COOLDOWN)
7105 					break;
7106 
7107 				if (current != DTRACE_ACTIVITY_WARMUP)
7108 					current = DTRACE_ACTIVITY_ACTIVE;
7109 
7110 				if (dtrace_cas32(activity, current,
7111 				    DTRACE_ACTIVITY_DRAINING) != current) {
7112 					*flags |= CPU_DTRACE_DROP;
7113 					continue;
7114 				}
7115 
7116 				break;
7117 			}
7118 
7119 			default:
7120 				ASSERT(0);
7121 			}
7122 
7123 			if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF ||
7124 			    dp->dtdo_rtype.dtdt_flags & DIF_TF_BYUREF) {
7125 				uintptr_t end = valoffs + size;
7126 
7127 				if (tracememsize != 0 &&
7128 				    valoffs + tracememsize < end) {
7129 					end = valoffs + tracememsize;
7130 					tracememsize = 0;
7131 				}
7132 
7133 				if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF &&
7134 				    !dtrace_vcanload((void *)(uintptr_t)val,
7135 				    &dp->dtdo_rtype, &mstate, vstate))
7136 					continue;
7137 
7138 				dtrace_store_by_ref(dp, tomax, size, &valoffs,
7139 				    &val, end, act->dta_intuple,
7140 				    dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF ?
7141 				    DIF_TF_BYREF: DIF_TF_BYUREF);
7142 				continue;
7143 			}
7144 
7145 			switch (size) {
7146 			case 0:
7147 				break;
7148 
7149 			case sizeof (uint8_t):
7150 				DTRACE_STORE(uint8_t, tomax, valoffs, val);
7151 				break;
7152 			case sizeof (uint16_t):
7153 				DTRACE_STORE(uint16_t, tomax, valoffs, val);
7154 				break;
7155 			case sizeof (uint32_t):
7156 				DTRACE_STORE(uint32_t, tomax, valoffs, val);
7157 				break;
7158 			case sizeof (uint64_t):
7159 				DTRACE_STORE(uint64_t, tomax, valoffs, val);
7160 				break;
7161 			default:
7162 				/*
7163 				 * Any other size should have been returned by
7164 				 * reference, not by value.
7165 				 */
7166 				ASSERT(0);
7167 				break;
7168 			}
7169 		}
7170 
7171 		if (*flags & CPU_DTRACE_DROP)
7172 			continue;
7173 
7174 		if (*flags & CPU_DTRACE_FAULT) {
7175 			int ndx;
7176 			dtrace_action_t *err;
7177 
7178 			buf->dtb_errors++;
7179 
7180 			if (probe->dtpr_id == dtrace_probeid_error) {
7181 				/*
7182 				 * There's nothing we can do -- we had an
7183 				 * error on the error probe.  We bump an
7184 				 * error counter to at least indicate that
7185 				 * this condition happened.
7186 				 */
7187 				dtrace_error(&state->dts_dblerrors);
7188 				continue;
7189 			}
7190 
7191 			if (vtime) {
7192 				/*
7193 				 * Before recursing on dtrace_probe(), we
7194 				 * need to explicitly clear out our start
7195 				 * time to prevent it from being accumulated
7196 				 * into t_dtrace_vtime.
7197 				 */
7198 				curthread->t_dtrace_start = 0;
7199 			}
7200 
7201 			/*
7202 			 * Iterate over the actions to figure out which action
7203 			 * we were processing when we experienced the error.
7204 			 * Note that act points _past_ the faulting action; if
7205 			 * act is ecb->dte_action, the fault was in the
7206 			 * predicate, if it's ecb->dte_action->dta_next it's
7207 			 * in action #1, and so on.
7208 			 */
7209 			for (err = ecb->dte_action, ndx = 0;
7210 			    err != act; err = err->dta_next, ndx++)
7211 				continue;
7212 
7213 			dtrace_probe_error(state, ecb->dte_epid, ndx,
7214 			    (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ?
7215 			    mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags),
7216 			    cpu_core[cpuid].cpuc_dtrace_illval);
7217 
7218 			continue;
7219 		}
7220 
7221 		if (!committed)
7222 			buf->dtb_offset = offs + ecb->dte_size;
7223 	}
7224 
7225 	end = dtrace_gethrtime();
7226 	if (vtime)
7227 		curthread->t_dtrace_start = end;
7228 
7229 	CPU->cpu_dtrace_nsec += end - now;
7230 
7231 	dtrace_interrupt_enable(cookie);
7232 }
7233 
7234 /*
7235  * DTrace Probe Hashing Functions
7236  *
7237  * The functions in this section (and indeed, the functions in remaining
7238  * sections) are not _called_ from probe context.  (Any exceptions to this are
7239  * marked with a "Note:".)  Rather, they are called from elsewhere in the
7240  * DTrace framework to look-up probes in, add probes to and remove probes from
7241  * the DTrace probe hashes.  (Each probe is hashed by each element of the
7242  * probe tuple -- allowing for fast lookups, regardless of what was
7243  * specified.)
7244  */
7245 static uint_t
dtrace_hash_str(char * p)7246 dtrace_hash_str(char *p)
7247 {
7248 	unsigned int g;
7249 	uint_t hval = 0;
7250 
7251 	while (*p) {
7252 		hval = (hval << 4) + *p++;
7253 		if ((g = (hval & 0xf0000000)) != 0)
7254 			hval ^= g >> 24;
7255 		hval &= ~g;
7256 	}
7257 	return (hval);
7258 }
7259 
7260 static dtrace_hash_t *
dtrace_hash_create(uintptr_t stroffs,uintptr_t nextoffs,uintptr_t prevoffs)7261 dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs)
7262 {
7263 	dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP);
7264 
7265 	hash->dth_stroffs = stroffs;
7266 	hash->dth_nextoffs = nextoffs;
7267 	hash->dth_prevoffs = prevoffs;
7268 
7269 	hash->dth_size = 1;
7270 	hash->dth_mask = hash->dth_size - 1;
7271 
7272 	hash->dth_tab = kmem_zalloc(hash->dth_size *
7273 	    sizeof (dtrace_hashbucket_t *), KM_SLEEP);
7274 
7275 	return (hash);
7276 }
7277 
7278 static void
dtrace_hash_destroy(dtrace_hash_t * hash)7279 dtrace_hash_destroy(dtrace_hash_t *hash)
7280 {
7281 #ifdef DEBUG
7282 	int i;
7283 
7284 	for (i = 0; i < hash->dth_size; i++)
7285 		ASSERT(hash->dth_tab[i] == NULL);
7286 #endif
7287 
7288 	kmem_free(hash->dth_tab,
7289 	    hash->dth_size * sizeof (dtrace_hashbucket_t *));
7290 	kmem_free(hash, sizeof (dtrace_hash_t));
7291 }
7292 
7293 static void
dtrace_hash_resize(dtrace_hash_t * hash)7294 dtrace_hash_resize(dtrace_hash_t *hash)
7295 {
7296 	int size = hash->dth_size, i, ndx;
7297 	int new_size = hash->dth_size << 1;
7298 	int new_mask = new_size - 1;
7299 	dtrace_hashbucket_t **new_tab, *bucket, *next;
7300 
7301 	ASSERT((new_size & new_mask) == 0);
7302 
7303 	new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP);
7304 
7305 	for (i = 0; i < size; i++) {
7306 		for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) {
7307 			dtrace_probe_t *probe = bucket->dthb_chain;
7308 
7309 			ASSERT(probe != NULL);
7310 			ndx = DTRACE_HASHSTR(hash, probe) & new_mask;
7311 
7312 			next = bucket->dthb_next;
7313 			bucket->dthb_next = new_tab[ndx];
7314 			new_tab[ndx] = bucket;
7315 		}
7316 	}
7317 
7318 	kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *));
7319 	hash->dth_tab = new_tab;
7320 	hash->dth_size = new_size;
7321 	hash->dth_mask = new_mask;
7322 }
7323 
7324 static void
dtrace_hash_add(dtrace_hash_t * hash,dtrace_probe_t * new)7325 dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new)
7326 {
7327 	int hashval = DTRACE_HASHSTR(hash, new);
7328 	int ndx = hashval & hash->dth_mask;
7329 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
7330 	dtrace_probe_t **nextp, **prevp;
7331 
7332 	for (; bucket != NULL; bucket = bucket->dthb_next) {
7333 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new))
7334 			goto add;
7335 	}
7336 
7337 	if ((hash->dth_nbuckets >> 1) > hash->dth_size) {
7338 		dtrace_hash_resize(hash);
7339 		dtrace_hash_add(hash, new);
7340 		return;
7341 	}
7342 
7343 	bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP);
7344 	bucket->dthb_next = hash->dth_tab[ndx];
7345 	hash->dth_tab[ndx] = bucket;
7346 	hash->dth_nbuckets++;
7347 
7348 add:
7349 	nextp = DTRACE_HASHNEXT(hash, new);
7350 	ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL);
7351 	*nextp = bucket->dthb_chain;
7352 
7353 	if (bucket->dthb_chain != NULL) {
7354 		prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain);
7355 		ASSERT(*prevp == NULL);
7356 		*prevp = new;
7357 	}
7358 
7359 	bucket->dthb_chain = new;
7360 	bucket->dthb_len++;
7361 }
7362 
7363 static dtrace_probe_t *
dtrace_hash_lookup(dtrace_hash_t * hash,dtrace_probe_t * template)7364 dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template)
7365 {
7366 	int hashval = DTRACE_HASHSTR(hash, template);
7367 	int ndx = hashval & hash->dth_mask;
7368 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
7369 
7370 	for (; bucket != NULL; bucket = bucket->dthb_next) {
7371 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
7372 			return (bucket->dthb_chain);
7373 	}
7374 
7375 	return (NULL);
7376 }
7377 
7378 static int
dtrace_hash_collisions(dtrace_hash_t * hash,dtrace_probe_t * template)7379 dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template)
7380 {
7381 	int hashval = DTRACE_HASHSTR(hash, template);
7382 	int ndx = hashval & hash->dth_mask;
7383 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
7384 
7385 	for (; bucket != NULL; bucket = bucket->dthb_next) {
7386 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
7387 			return (bucket->dthb_len);
7388 	}
7389 
7390 	return (NULL);
7391 }
7392 
7393 static void
dtrace_hash_remove(dtrace_hash_t * hash,dtrace_probe_t * probe)7394 dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe)
7395 {
7396 	int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask;
7397 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
7398 
7399 	dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe);
7400 	dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe);
7401 
7402 	/*
7403 	 * Find the bucket that we're removing this probe from.
7404 	 */
7405 	for (; bucket != NULL; bucket = bucket->dthb_next) {
7406 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe))
7407 			break;
7408 	}
7409 
7410 	ASSERT(bucket != NULL);
7411 
7412 	if (*prevp == NULL) {
7413 		if (*nextp == NULL) {
7414 			/*
7415 			 * The removed probe was the only probe on this
7416 			 * bucket; we need to remove the bucket.
7417 			 */
7418 			dtrace_hashbucket_t *b = hash->dth_tab[ndx];
7419 
7420 			ASSERT(bucket->dthb_chain == probe);
7421 			ASSERT(b != NULL);
7422 
7423 			if (b == bucket) {
7424 				hash->dth_tab[ndx] = bucket->dthb_next;
7425 			} else {
7426 				while (b->dthb_next != bucket)
7427 					b = b->dthb_next;
7428 				b->dthb_next = bucket->dthb_next;
7429 			}
7430 
7431 			ASSERT(hash->dth_nbuckets > 0);
7432 			hash->dth_nbuckets--;
7433 			kmem_free(bucket, sizeof (dtrace_hashbucket_t));
7434 			return;
7435 		}
7436 
7437 		bucket->dthb_chain = *nextp;
7438 	} else {
7439 		*(DTRACE_HASHNEXT(hash, *prevp)) = *nextp;
7440 	}
7441 
7442 	if (*nextp != NULL)
7443 		*(DTRACE_HASHPREV(hash, *nextp)) = *prevp;
7444 }
7445 
7446 /*
7447  * DTrace Utility Functions
7448  *
7449  * These are random utility functions that are _not_ called from probe context.
7450  */
7451 static int
dtrace_badattr(const dtrace_attribute_t * a)7452 dtrace_badattr(const dtrace_attribute_t *a)
7453 {
7454 	return (a->dtat_name > DTRACE_STABILITY_MAX ||
7455 	    a->dtat_data > DTRACE_STABILITY_MAX ||
7456 	    a->dtat_class > DTRACE_CLASS_MAX);
7457 }
7458 
7459 /*
7460  * Return a duplicate copy of a string.  If the specified string is NULL,
7461  * this function returns a zero-length string.
7462  */
7463 static char *
dtrace_strdup(const char * str)7464 dtrace_strdup(const char *str)
7465 {
7466 	char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP);
7467 
7468 	if (str != NULL)
7469 		(void) strcpy(new, str);
7470 
7471 	return (new);
7472 }
7473 
7474 #define	DTRACE_ISALPHA(c)	\
7475 	(((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z'))
7476 
7477 static int
dtrace_badname(const char * s)7478 dtrace_badname(const char *s)
7479 {
7480 	char c;
7481 
7482 	if (s == NULL || (c = *s++) == '\0')
7483 		return (0);
7484 
7485 	if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.')
7486 		return (1);
7487 
7488 	while ((c = *s++) != '\0') {
7489 		if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') &&
7490 		    c != '-' && c != '_' && c != '.' && c != '`')
7491 			return (1);
7492 	}
7493 
7494 	return (0);
7495 }
7496 
7497 static void
dtrace_cred2priv(cred_t * cr,uint32_t * privp,uid_t * uidp,zoneid_t * zoneidp)7498 dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp)
7499 {
7500 	uint32_t priv;
7501 
7502 	if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
7503 		/*
7504 		 * For DTRACE_PRIV_ALL, the uid and zoneid don't matter.
7505 		 */
7506 		priv = DTRACE_PRIV_ALL;
7507 	} else {
7508 		*uidp = crgetuid(cr);
7509 		*zoneidp = crgetzoneid(cr);
7510 
7511 		priv = 0;
7512 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE))
7513 			priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER;
7514 		else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE))
7515 			priv |= DTRACE_PRIV_USER;
7516 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE))
7517 			priv |= DTRACE_PRIV_PROC;
7518 		if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
7519 			priv |= DTRACE_PRIV_OWNER;
7520 		if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
7521 			priv |= DTRACE_PRIV_ZONEOWNER;
7522 	}
7523 
7524 	*privp = priv;
7525 }
7526 
7527 #ifdef DTRACE_ERRDEBUG
7528 static void
dtrace_errdebug(const char * str)7529 dtrace_errdebug(const char *str)
7530 {
7531 	int hval = dtrace_hash_str((char *)str) % DTRACE_ERRHASHSZ;
7532 	int occupied = 0;
7533 
7534 	mutex_enter(&dtrace_errlock);
7535 	dtrace_errlast = str;
7536 	dtrace_errthread = curthread;
7537 
7538 	while (occupied++ < DTRACE_ERRHASHSZ) {
7539 		if (dtrace_errhash[hval].dter_msg == str) {
7540 			dtrace_errhash[hval].dter_count++;
7541 			goto out;
7542 		}
7543 
7544 		if (dtrace_errhash[hval].dter_msg != NULL) {
7545 			hval = (hval + 1) % DTRACE_ERRHASHSZ;
7546 			continue;
7547 		}
7548 
7549 		dtrace_errhash[hval].dter_msg = str;
7550 		dtrace_errhash[hval].dter_count = 1;
7551 		goto out;
7552 	}
7553 
7554 	panic("dtrace: undersized error hash");
7555 out:
7556 	mutex_exit(&dtrace_errlock);
7557 }
7558 #endif
7559 
7560 /*
7561  * DTrace Matching Functions
7562  *
7563  * These functions are used to match groups of probes, given some elements of
7564  * a probe tuple, or some globbed expressions for elements of a probe tuple.
7565  */
7566 static int
dtrace_match_priv(const dtrace_probe_t * prp,uint32_t priv,uid_t uid,zoneid_t zoneid)7567 dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid,
7568     zoneid_t zoneid)
7569 {
7570 	if (priv != DTRACE_PRIV_ALL) {
7571 		uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags;
7572 		uint32_t match = priv & ppriv;
7573 
7574 		/*
7575 		 * No PRIV_DTRACE_* privileges...
7576 		 */
7577 		if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER |
7578 		    DTRACE_PRIV_KERNEL)) == 0)
7579 			return (0);
7580 
7581 		/*
7582 		 * No matching bits, but there were bits to match...
7583 		 */
7584 		if (match == 0 && ppriv != 0)
7585 			return (0);
7586 
7587 		/*
7588 		 * Need to have permissions to the process, but don't...
7589 		 */
7590 		if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 &&
7591 		    uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) {
7592 			return (0);
7593 		}
7594 
7595 		/*
7596 		 * Need to be in the same zone unless we possess the
7597 		 * privilege to examine all zones.
7598 		 */
7599 		if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 &&
7600 		    zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) {
7601 			return (0);
7602 		}
7603 	}
7604 
7605 	return (1);
7606 }
7607 
7608 /*
7609  * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which
7610  * consists of input pattern strings and an ops-vector to evaluate them.
7611  * This function returns >0 for match, 0 for no match, and <0 for error.
7612  */
7613 static int
dtrace_match_probe(const dtrace_probe_t * prp,const dtrace_probekey_t * pkp,uint32_t priv,uid_t uid,zoneid_t zoneid)7614 dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp,
7615     uint32_t priv, uid_t uid, zoneid_t zoneid)
7616 {
7617 	dtrace_provider_t *pvp = prp->dtpr_provider;
7618 	int rv;
7619 
7620 	if (pvp->dtpv_defunct)
7621 		return (0);
7622 
7623 	if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0)
7624 		return (rv);
7625 
7626 	if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0)
7627 		return (rv);
7628 
7629 	if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0)
7630 		return (rv);
7631 
7632 	if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0)
7633 		return (rv);
7634 
7635 	if (dtrace_match_priv(prp, priv, uid, zoneid) == 0)
7636 		return (0);
7637 
7638 	return (rv);
7639 }
7640 
7641 /*
7642  * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN)
7643  * interface for matching a glob pattern 'p' to an input string 's'.  Unlike
7644  * libc's version, the kernel version only applies to 8-bit ASCII strings.
7645  * In addition, all of the recursion cases except for '*' matching have been
7646  * unwound.  For '*', we still implement recursive evaluation, but a depth
7647  * counter is maintained and matching is aborted if we recurse too deep.
7648  * The function returns 0 if no match, >0 if match, and <0 if recursion error.
7649  */
7650 static int
dtrace_match_glob(const char * s,const char * p,int depth)7651 dtrace_match_glob(const char *s, const char *p, int depth)
7652 {
7653 	const char *olds;
7654 	char s1, c;
7655 	int gs;
7656 
7657 	if (depth > DTRACE_PROBEKEY_MAXDEPTH)
7658 		return (-1);
7659 
7660 	if (s == NULL)
7661 		s = ""; /* treat NULL as empty string */
7662 
7663 top:
7664 	olds = s;
7665 	s1 = *s++;
7666 
7667 	if (p == NULL)
7668 		return (0);
7669 
7670 	if ((c = *p++) == '\0')
7671 		return (s1 == '\0');
7672 
7673 	switch (c) {
7674 	case '[': {
7675 		int ok = 0, notflag = 0;
7676 		char lc = '\0';
7677 
7678 		if (s1 == '\0')
7679 			return (0);
7680 
7681 		if (*p == '!') {
7682 			notflag = 1;
7683 			p++;
7684 		}
7685 
7686 		if ((c = *p++) == '\0')
7687 			return (0);
7688 
7689 		do {
7690 			if (c == '-' && lc != '\0' && *p != ']') {
7691 				if ((c = *p++) == '\0')
7692 					return (0);
7693 				if (c == '\\' && (c = *p++) == '\0')
7694 					return (0);
7695 
7696 				if (notflag) {
7697 					if (s1 < lc || s1 > c)
7698 						ok++;
7699 					else
7700 						return (0);
7701 				} else if (lc <= s1 && s1 <= c)
7702 					ok++;
7703 
7704 			} else if (c == '\\' && (c = *p++) == '\0')
7705 				return (0);
7706 
7707 			lc = c; /* save left-hand 'c' for next iteration */
7708 
7709 			if (notflag) {
7710 				if (s1 != c)
7711 					ok++;
7712 				else
7713 					return (0);
7714 			} else if (s1 == c)
7715 				ok++;
7716 
7717 			if ((c = *p++) == '\0')
7718 				return (0);
7719 
7720 		} while (c != ']');
7721 
7722 		if (ok)
7723 			goto top;
7724 
7725 		return (0);
7726 	}
7727 
7728 	case '\\':
7729 		if ((c = *p++) == '\0')
7730 			return (0);
7731 		/*FALLTHRU*/
7732 
7733 	default:
7734 		if (c != s1)
7735 			return (0);
7736 		/*FALLTHRU*/
7737 
7738 	case '?':
7739 		if (s1 != '\0')
7740 			goto top;
7741 		return (0);
7742 
7743 	case '*':
7744 		while (*p == '*')
7745 			p++; /* consecutive *'s are identical to a single one */
7746 
7747 		if (*p == '\0')
7748 			return (1);
7749 
7750 		for (s = olds; *s != '\0'; s++) {
7751 			if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0)
7752 				return (gs);
7753 		}
7754 
7755 		return (0);
7756 	}
7757 }
7758 
7759 /*ARGSUSED*/
7760 static int
dtrace_match_string(const char * s,const char * p,int depth)7761 dtrace_match_string(const char *s, const char *p, int depth)
7762 {
7763 	return (s != NULL && strcmp(s, p) == 0);
7764 }
7765 
7766 /*ARGSUSED*/
7767 static int
dtrace_match_nul(const char * s,const char * p,int depth)7768 dtrace_match_nul(const char *s, const char *p, int depth)
7769 {
7770 	return (1); /* always match the empty pattern */
7771 }
7772 
7773 /*ARGSUSED*/
7774 static int
dtrace_match_nonzero(const char * s,const char * p,int depth)7775 dtrace_match_nonzero(const char *s, const char *p, int depth)
7776 {
7777 	return (s != NULL && s[0] != '\0');
7778 }
7779 
7780 static int
dtrace_match(const dtrace_probekey_t * pkp,uint32_t priv,uid_t uid,zoneid_t zoneid,int (* matched)(dtrace_probe_t *,void *),void * arg)7781 dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid,
7782     zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg)
7783 {
7784 	dtrace_probe_t template, *probe;
7785 	dtrace_hash_t *hash = NULL;
7786 	int len, rc, best = INT_MAX, nmatched = 0;
7787 	dtrace_id_t i;
7788 
7789 	ASSERT(MUTEX_HELD(&dtrace_lock));
7790 
7791 	/*
7792 	 * If the probe ID is specified in the key, just lookup by ID and
7793 	 * invoke the match callback once if a matching probe is found.
7794 	 */
7795 	if (pkp->dtpk_id != DTRACE_IDNONE) {
7796 		if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL &&
7797 		    dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) {
7798 			if ((*matched)(probe, arg) == DTRACE_MATCH_FAIL)
7799 				return (DTRACE_MATCH_FAIL);
7800 			nmatched++;
7801 		}
7802 		return (nmatched);
7803 	}
7804 
7805 	template.dtpr_mod = (char *)pkp->dtpk_mod;
7806 	template.dtpr_func = (char *)pkp->dtpk_func;
7807 	template.dtpr_name = (char *)pkp->dtpk_name;
7808 
7809 	/*
7810 	 * We want to find the most distinct of the module name, function
7811 	 * name, and name.  So for each one that is not a glob pattern or
7812 	 * empty string, we perform a lookup in the corresponding hash and
7813 	 * use the hash table with the fewest collisions to do our search.
7814 	 */
7815 	if (pkp->dtpk_mmatch == &dtrace_match_string &&
7816 	    (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) {
7817 		best = len;
7818 		hash = dtrace_bymod;
7819 	}
7820 
7821 	if (pkp->dtpk_fmatch == &dtrace_match_string &&
7822 	    (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) {
7823 		best = len;
7824 		hash = dtrace_byfunc;
7825 	}
7826 
7827 	if (pkp->dtpk_nmatch == &dtrace_match_string &&
7828 	    (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) {
7829 		best = len;
7830 		hash = dtrace_byname;
7831 	}
7832 
7833 	/*
7834 	 * If we did not select a hash table, iterate over every probe and
7835 	 * invoke our callback for each one that matches our input probe key.
7836 	 */
7837 	if (hash == NULL) {
7838 		for (i = 0; i < dtrace_nprobes; i++) {
7839 			if ((probe = dtrace_probes[i]) == NULL ||
7840 			    dtrace_match_probe(probe, pkp, priv, uid,
7841 			    zoneid) <= 0)
7842 				continue;
7843 
7844 			nmatched++;
7845 
7846 			if ((rc = (*matched)(probe, arg)) !=
7847 			    DTRACE_MATCH_NEXT) {
7848 				if (rc == DTRACE_MATCH_FAIL)
7849 					return (DTRACE_MATCH_FAIL);
7850 				break;
7851 			}
7852 		}
7853 
7854 		return (nmatched);
7855 	}
7856 
7857 	/*
7858 	 * If we selected a hash table, iterate over each probe of the same key
7859 	 * name and invoke the callback for every probe that matches the other
7860 	 * attributes of our input probe key.
7861 	 */
7862 	for (probe = dtrace_hash_lookup(hash, &template); probe != NULL;
7863 	    probe = *(DTRACE_HASHNEXT(hash, probe))) {
7864 
7865 		if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0)
7866 			continue;
7867 
7868 		nmatched++;
7869 
7870 		if ((rc = (*matched)(probe, arg)) != DTRACE_MATCH_NEXT) {
7871 			if (rc == DTRACE_MATCH_FAIL)
7872 				return (DTRACE_MATCH_FAIL);
7873 			break;
7874 		}
7875 	}
7876 
7877 	return (nmatched);
7878 }
7879 
7880 /*
7881  * Return the function pointer dtrace_probecmp() should use to compare the
7882  * specified pattern with a string.  For NULL or empty patterns, we select
7883  * dtrace_match_nul().  For glob pattern strings, we use dtrace_match_glob().
7884  * For non-empty non-glob strings, we use dtrace_match_string().
7885  */
7886 static dtrace_probekey_f *
dtrace_probekey_func(const char * p)7887 dtrace_probekey_func(const char *p)
7888 {
7889 	char c;
7890 
7891 	if (p == NULL || *p == '\0')
7892 		return (&dtrace_match_nul);
7893 
7894 	while ((c = *p++) != '\0') {
7895 		if (c == '[' || c == '?' || c == '*' || c == '\\')
7896 			return (&dtrace_match_glob);
7897 	}
7898 
7899 	return (&dtrace_match_string);
7900 }
7901 
7902 /*
7903  * Build a probe comparison key for use with dtrace_match_probe() from the
7904  * given probe description.  By convention, a null key only matches anchored
7905  * probes: if each field is the empty string, reset dtpk_fmatch to
7906  * dtrace_match_nonzero().
7907  */
7908 static void
dtrace_probekey(const dtrace_probedesc_t * pdp,dtrace_probekey_t * pkp)7909 dtrace_probekey(const dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp)
7910 {
7911 	pkp->dtpk_prov = pdp->dtpd_provider;
7912 	pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider);
7913 
7914 	pkp->dtpk_mod = pdp->dtpd_mod;
7915 	pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod);
7916 
7917 	pkp->dtpk_func = pdp->dtpd_func;
7918 	pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func);
7919 
7920 	pkp->dtpk_name = pdp->dtpd_name;
7921 	pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name);
7922 
7923 	pkp->dtpk_id = pdp->dtpd_id;
7924 
7925 	if (pkp->dtpk_id == DTRACE_IDNONE &&
7926 	    pkp->dtpk_pmatch == &dtrace_match_nul &&
7927 	    pkp->dtpk_mmatch == &dtrace_match_nul &&
7928 	    pkp->dtpk_fmatch == &dtrace_match_nul &&
7929 	    pkp->dtpk_nmatch == &dtrace_match_nul)
7930 		pkp->dtpk_fmatch = &dtrace_match_nonzero;
7931 }
7932 
7933 /*
7934  * DTrace Provider-to-Framework API Functions
7935  *
7936  * These functions implement much of the Provider-to-Framework API, as
7937  * described in <sys/dtrace.h>.  The parts of the API not in this section are
7938  * the functions in the API for probe management (found below), and
7939  * dtrace_probe() itself (found above).
7940  */
7941 
7942 /*
7943  * Register the calling provider with the DTrace framework.  This should
7944  * generally be called by DTrace providers in their attach(9E) entry point.
7945  */
7946 int
dtrace_register(const char * name,const dtrace_pattr_t * pap,uint32_t priv,cred_t * cr,const dtrace_pops_t * pops,void * arg,dtrace_provider_id_t * idp)7947 dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv,
7948     cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp)
7949 {
7950 	dtrace_provider_t *provider;
7951 
7952 	if (name == NULL || pap == NULL || pops == NULL || idp == NULL) {
7953 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7954 		    "arguments", name ? name : "<NULL>");
7955 		return (EINVAL);
7956 	}
7957 
7958 	if (name[0] == '\0' || dtrace_badname(name)) {
7959 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7960 		    "provider name", name);
7961 		return (EINVAL);
7962 	}
7963 
7964 	if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) ||
7965 	    pops->dtps_enable == NULL || pops->dtps_disable == NULL ||
7966 	    pops->dtps_destroy == NULL ||
7967 	    ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) {
7968 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7969 		    "provider ops", name);
7970 		return (EINVAL);
7971 	}
7972 
7973 	if (dtrace_badattr(&pap->dtpa_provider) ||
7974 	    dtrace_badattr(&pap->dtpa_mod) ||
7975 	    dtrace_badattr(&pap->dtpa_func) ||
7976 	    dtrace_badattr(&pap->dtpa_name) ||
7977 	    dtrace_badattr(&pap->dtpa_args)) {
7978 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7979 		    "provider attributes", name);
7980 		return (EINVAL);
7981 	}
7982 
7983 	if (priv & ~DTRACE_PRIV_ALL) {
7984 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7985 		    "privilege attributes", name);
7986 		return (EINVAL);
7987 	}
7988 
7989 	if ((priv & DTRACE_PRIV_KERNEL) &&
7990 	    (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) &&
7991 	    pops->dtps_mode == NULL) {
7992 		cmn_err(CE_WARN, "failed to register provider '%s': need "
7993 		    "dtps_mode() op for given privilege attributes", name);
7994 		return (EINVAL);
7995 	}
7996 
7997 	provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP);
7998 	provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
7999 	(void) strcpy(provider->dtpv_name, name);
8000 
8001 	provider->dtpv_attr = *pap;
8002 	provider->dtpv_priv.dtpp_flags = priv;
8003 	if (cr != NULL) {
8004 		provider->dtpv_priv.dtpp_uid = crgetuid(cr);
8005 		provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr);
8006 	}
8007 	provider->dtpv_pops = *pops;
8008 
8009 	if (pops->dtps_provide == NULL) {
8010 		ASSERT(pops->dtps_provide_module != NULL);
8011 		provider->dtpv_pops.dtps_provide =
8012 		    (void (*)(void *, const dtrace_probedesc_t *))dtrace_nullop;
8013 	}
8014 
8015 	if (pops->dtps_provide_module == NULL) {
8016 		ASSERT(pops->dtps_provide != NULL);
8017 		provider->dtpv_pops.dtps_provide_module =
8018 		    (void (*)(void *, struct modctl *))dtrace_nullop;
8019 	}
8020 
8021 	if (pops->dtps_suspend == NULL) {
8022 		ASSERT(pops->dtps_resume == NULL);
8023 		provider->dtpv_pops.dtps_suspend =
8024 		    (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
8025 		provider->dtpv_pops.dtps_resume =
8026 		    (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
8027 	}
8028 
8029 	provider->dtpv_arg = arg;
8030 	*idp = (dtrace_provider_id_t)provider;
8031 
8032 	if (pops == &dtrace_provider_ops) {
8033 		ASSERT(MUTEX_HELD(&dtrace_provider_lock));
8034 		ASSERT(MUTEX_HELD(&dtrace_lock));
8035 		ASSERT(dtrace_anon.dta_enabling == NULL);
8036 
8037 		/*
8038 		 * We make sure that the DTrace provider is at the head of
8039 		 * the provider chain.
8040 		 */
8041 		provider->dtpv_next = dtrace_provider;
8042 		dtrace_provider = provider;
8043 		return (0);
8044 	}
8045 
8046 	mutex_enter(&dtrace_provider_lock);
8047 	mutex_enter(&dtrace_lock);
8048 
8049 	/*
8050 	 * If there is at least one provider registered, we'll add this
8051 	 * provider after the first provider.
8052 	 */
8053 	if (dtrace_provider != NULL) {
8054 		provider->dtpv_next = dtrace_provider->dtpv_next;
8055 		dtrace_provider->dtpv_next = provider;
8056 	} else {
8057 		dtrace_provider = provider;
8058 	}
8059 
8060 	if (dtrace_retained != NULL) {
8061 		dtrace_enabling_provide(provider);
8062 
8063 		/*
8064 		 * Now we need to call dtrace_enabling_matchall() -- which
8065 		 * will acquire cpu_lock and dtrace_lock.  We therefore need
8066 		 * to drop all of our locks before calling into it...
8067 		 */
8068 		mutex_exit(&dtrace_lock);
8069 		mutex_exit(&dtrace_provider_lock);
8070 		dtrace_enabling_matchall();
8071 
8072 		return (0);
8073 	}
8074 
8075 	mutex_exit(&dtrace_lock);
8076 	mutex_exit(&dtrace_provider_lock);
8077 
8078 	return (0);
8079 }
8080 
8081 /*
8082  * Unregister the specified provider from the DTrace framework.  This should
8083  * generally be called by DTrace providers in their detach(9E) entry point.
8084  */
8085 int
dtrace_unregister(dtrace_provider_id_t id)8086 dtrace_unregister(dtrace_provider_id_t id)
8087 {
8088 	dtrace_provider_t *old = (dtrace_provider_t *)id;
8089 	dtrace_provider_t *prev = NULL;
8090 	int i, self = 0, noreap = 0;
8091 	dtrace_probe_t *probe, *first = NULL;
8092 
8093 	if (old->dtpv_pops.dtps_enable ==
8094 	    (int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop) {
8095 		/*
8096 		 * If DTrace itself is the provider, we're called with locks
8097 		 * already held.
8098 		 */
8099 		ASSERT(old == dtrace_provider);
8100 		ASSERT(dtrace_devi != NULL);
8101 		ASSERT(MUTEX_HELD(&dtrace_provider_lock));
8102 		ASSERT(MUTEX_HELD(&dtrace_lock));
8103 		self = 1;
8104 
8105 		if (dtrace_provider->dtpv_next != NULL) {
8106 			/*
8107 			 * There's another provider here; return failure.
8108 			 */
8109 			return (EBUSY);
8110 		}
8111 	} else {
8112 		mutex_enter(&dtrace_provider_lock);
8113 		mutex_enter(&mod_lock);
8114 		mutex_enter(&dtrace_lock);
8115 	}
8116 
8117 	/*
8118 	 * If anyone has /dev/dtrace open, or if there are anonymous enabled
8119 	 * probes, we refuse to let providers slither away, unless this
8120 	 * provider has already been explicitly invalidated.
8121 	 */
8122 	if (!old->dtpv_defunct &&
8123 	    (dtrace_opens || (dtrace_anon.dta_state != NULL &&
8124 	    dtrace_anon.dta_state->dts_necbs > 0))) {
8125 		if (!self) {
8126 			mutex_exit(&dtrace_lock);
8127 			mutex_exit(&mod_lock);
8128 			mutex_exit(&dtrace_provider_lock);
8129 		}
8130 		return (EBUSY);
8131 	}
8132 
8133 	/*
8134 	 * Attempt to destroy the probes associated with this provider.
8135 	 */
8136 	for (i = 0; i < dtrace_nprobes; i++) {
8137 		if ((probe = dtrace_probes[i]) == NULL)
8138 			continue;
8139 
8140 		if (probe->dtpr_provider != old)
8141 			continue;
8142 
8143 		if (probe->dtpr_ecb == NULL)
8144 			continue;
8145 
8146 		/*
8147 		 * If we are trying to unregister a defunct provider, and the
8148 		 * provider was made defunct within the interval dictated by
8149 		 * dtrace_unregister_defunct_reap, we'll (asynchronously)
8150 		 * attempt to reap our enablings.  To denote that the provider
8151 		 * should reattempt to unregister itself at some point in the
8152 		 * future, we will return a differentiable error code (EAGAIN
8153 		 * instead of EBUSY) in this case.
8154 		 */
8155 		if (dtrace_gethrtime() - old->dtpv_defunct >
8156 		    dtrace_unregister_defunct_reap)
8157 			noreap = 1;
8158 
8159 		if (!self) {
8160 			mutex_exit(&dtrace_lock);
8161 			mutex_exit(&mod_lock);
8162 			mutex_exit(&dtrace_provider_lock);
8163 		}
8164 
8165 		if (noreap)
8166 			return (EBUSY);
8167 
8168 		(void) taskq_dispatch(dtrace_taskq,
8169 		    (task_func_t *)dtrace_enabling_reap, NULL, TQ_SLEEP);
8170 
8171 		return (EAGAIN);
8172 	}
8173 
8174 	/*
8175 	 * All of the probes for this provider are disabled; we can safely
8176 	 * remove all of them from their hash chains and from the probe array.
8177 	 */
8178 	for (i = 0; i < dtrace_nprobes; i++) {
8179 		if ((probe = dtrace_probes[i]) == NULL)
8180 			continue;
8181 
8182 		if (probe->dtpr_provider != old)
8183 			continue;
8184 
8185 		dtrace_probes[i] = NULL;
8186 
8187 		dtrace_hash_remove(dtrace_bymod, probe);
8188 		dtrace_hash_remove(dtrace_byfunc, probe);
8189 		dtrace_hash_remove(dtrace_byname, probe);
8190 
8191 		if (first == NULL) {
8192 			first = probe;
8193 			probe->dtpr_nextmod = NULL;
8194 		} else {
8195 			probe->dtpr_nextmod = first;
8196 			first = probe;
8197 		}
8198 	}
8199 
8200 	/*
8201 	 * The provider's probes have been removed from the hash chains and
8202 	 * from the probe array.  Now issue a dtrace_sync() to be sure that
8203 	 * everyone has cleared out from any probe array processing.
8204 	 */
8205 	dtrace_sync();
8206 
8207 	for (probe = first; probe != NULL; probe = first) {
8208 		first = probe->dtpr_nextmod;
8209 
8210 		old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id,
8211 		    probe->dtpr_arg);
8212 		kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
8213 		kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
8214 		kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
8215 		vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1);
8216 		kmem_free(probe, sizeof (dtrace_probe_t));
8217 	}
8218 
8219 	if ((prev = dtrace_provider) == old) {
8220 		ASSERT(self || dtrace_devi == NULL);
8221 		ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL);
8222 		dtrace_provider = old->dtpv_next;
8223 	} else {
8224 		while (prev != NULL && prev->dtpv_next != old)
8225 			prev = prev->dtpv_next;
8226 
8227 		if (prev == NULL) {
8228 			panic("attempt to unregister non-existent "
8229 			    "dtrace provider %p\n", (void *)id);
8230 		}
8231 
8232 		prev->dtpv_next = old->dtpv_next;
8233 	}
8234 
8235 	if (!self) {
8236 		mutex_exit(&dtrace_lock);
8237 		mutex_exit(&mod_lock);
8238 		mutex_exit(&dtrace_provider_lock);
8239 	}
8240 
8241 	kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1);
8242 	kmem_free(old, sizeof (dtrace_provider_t));
8243 
8244 	return (0);
8245 }
8246 
8247 /*
8248  * Invalidate the specified provider.  All subsequent probe lookups for the
8249  * specified provider will fail, but its probes will not be removed.
8250  */
8251 void
dtrace_invalidate(dtrace_provider_id_t id)8252 dtrace_invalidate(dtrace_provider_id_t id)
8253 {
8254 	dtrace_provider_t *pvp = (dtrace_provider_t *)id;
8255 
8256 	ASSERT(pvp->dtpv_pops.dtps_enable !=
8257 	    (int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop);
8258 
8259 	mutex_enter(&dtrace_provider_lock);
8260 	mutex_enter(&dtrace_lock);
8261 
8262 	pvp->dtpv_defunct = dtrace_gethrtime();
8263 
8264 	mutex_exit(&dtrace_lock);
8265 	mutex_exit(&dtrace_provider_lock);
8266 }
8267 
8268 /*
8269  * Indicate whether or not DTrace has attached.
8270  */
8271 int
dtrace_attached(void)8272 dtrace_attached(void)
8273 {
8274 	/*
8275 	 * dtrace_provider will be non-NULL iff the DTrace driver has
8276 	 * attached.  (It's non-NULL because DTrace is always itself a
8277 	 * provider.)
8278 	 */
8279 	return (dtrace_provider != NULL);
8280 }
8281 
8282 /*
8283  * Remove all the unenabled probes for the given provider.  This function is
8284  * not unlike dtrace_unregister(), except that it doesn't remove the provider
8285  * -- just as many of its associated probes as it can.
8286  */
8287 int
dtrace_condense(dtrace_provider_id_t id)8288 dtrace_condense(dtrace_provider_id_t id)
8289 {
8290 	dtrace_provider_t *prov = (dtrace_provider_t *)id;
8291 	int i;
8292 	dtrace_probe_t *probe;
8293 
8294 	/*
8295 	 * Make sure this isn't the dtrace provider itself.
8296 	 */
8297 	ASSERT(prov->dtpv_pops.dtps_enable !=
8298 	    (int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop);
8299 
8300 	mutex_enter(&dtrace_provider_lock);
8301 	mutex_enter(&dtrace_lock);
8302 
8303 	/*
8304 	 * Attempt to destroy the probes associated with this provider.
8305 	 */
8306 	for (i = 0; i < dtrace_nprobes; i++) {
8307 		if ((probe = dtrace_probes[i]) == NULL)
8308 			continue;
8309 
8310 		if (probe->dtpr_provider != prov)
8311 			continue;
8312 
8313 		if (probe->dtpr_ecb != NULL)
8314 			continue;
8315 
8316 		dtrace_probes[i] = NULL;
8317 
8318 		dtrace_hash_remove(dtrace_bymod, probe);
8319 		dtrace_hash_remove(dtrace_byfunc, probe);
8320 		dtrace_hash_remove(dtrace_byname, probe);
8321 
8322 		prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1,
8323 		    probe->dtpr_arg);
8324 		kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
8325 		kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
8326 		kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
8327 		kmem_free(probe, sizeof (dtrace_probe_t));
8328 		vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1);
8329 	}
8330 
8331 	mutex_exit(&dtrace_lock);
8332 	mutex_exit(&dtrace_provider_lock);
8333 
8334 	return (0);
8335 }
8336 
8337 /*
8338  * DTrace Probe Management Functions
8339  *
8340  * The functions in this section perform the DTrace probe management,
8341  * including functions to create probes, look-up probes, and call into the
8342  * providers to request that probes be provided.  Some of these functions are
8343  * in the Provider-to-Framework API; these functions can be identified by the
8344  * fact that they are not declared "static".
8345  */
8346 
8347 /*
8348  * Create a probe with the specified module name, function name, and name.
8349  */
8350 dtrace_id_t
dtrace_probe_create(dtrace_provider_id_t prov,const char * mod,const char * func,const char * name,int aframes,void * arg)8351 dtrace_probe_create(dtrace_provider_id_t prov, const char *mod,
8352     const char *func, const char *name, int aframes, void *arg)
8353 {
8354 	dtrace_probe_t *probe, **probes;
8355 	dtrace_provider_t *provider = (dtrace_provider_t *)prov;
8356 	dtrace_id_t id;
8357 
8358 	if (provider == dtrace_provider) {
8359 		ASSERT(MUTEX_HELD(&dtrace_lock));
8360 	} else {
8361 		mutex_enter(&dtrace_lock);
8362 	}
8363 
8364 	id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1,
8365 	    VM_BESTFIT | VM_SLEEP);
8366 	probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP);
8367 
8368 	probe->dtpr_id = id;
8369 	probe->dtpr_gen = dtrace_probegen++;
8370 	probe->dtpr_mod = dtrace_strdup(mod);
8371 	probe->dtpr_func = dtrace_strdup(func);
8372 	probe->dtpr_name = dtrace_strdup(name);
8373 	probe->dtpr_arg = arg;
8374 	probe->dtpr_aframes = aframes;
8375 	probe->dtpr_provider = provider;
8376 
8377 	dtrace_hash_add(dtrace_bymod, probe);
8378 	dtrace_hash_add(dtrace_byfunc, probe);
8379 	dtrace_hash_add(dtrace_byname, probe);
8380 
8381 	if (id - 1 >= dtrace_nprobes) {
8382 		size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *);
8383 		size_t nsize = osize << 1;
8384 
8385 		if (nsize == 0) {
8386 			ASSERT(osize == 0);
8387 			ASSERT(dtrace_probes == NULL);
8388 			nsize = sizeof (dtrace_probe_t *);
8389 		}
8390 
8391 		probes = kmem_zalloc(nsize, KM_SLEEP);
8392 
8393 		if (dtrace_probes == NULL) {
8394 			ASSERT(osize == 0);
8395 			dtrace_probes = probes;
8396 			dtrace_nprobes = 1;
8397 		} else {
8398 			dtrace_probe_t **oprobes = dtrace_probes;
8399 
8400 			bcopy(oprobes, probes, osize);
8401 			dtrace_membar_producer();
8402 			dtrace_probes = probes;
8403 
8404 			dtrace_sync();
8405 
8406 			/*
8407 			 * All CPUs are now seeing the new probes array; we can
8408 			 * safely free the old array.
8409 			 */
8410 			kmem_free(oprobes, osize);
8411 			dtrace_nprobes <<= 1;
8412 		}
8413 
8414 		ASSERT(id - 1 < dtrace_nprobes);
8415 	}
8416 
8417 	ASSERT(dtrace_probes[id - 1] == NULL);
8418 	dtrace_probes[id - 1] = probe;
8419 
8420 	if (provider != dtrace_provider)
8421 		mutex_exit(&dtrace_lock);
8422 
8423 	return (id);
8424 }
8425 
8426 static dtrace_probe_t *
dtrace_probe_lookup_id(dtrace_id_t id)8427 dtrace_probe_lookup_id(dtrace_id_t id)
8428 {
8429 	ASSERT(MUTEX_HELD(&dtrace_lock));
8430 
8431 	if (id == 0 || id > dtrace_nprobes)
8432 		return (NULL);
8433 
8434 	return (dtrace_probes[id - 1]);
8435 }
8436 
8437 static int
dtrace_probe_lookup_match(dtrace_probe_t * probe,void * arg)8438 dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg)
8439 {
8440 	*((dtrace_id_t *)arg) = probe->dtpr_id;
8441 
8442 	return (DTRACE_MATCH_DONE);
8443 }
8444 
8445 /*
8446  * Look up a probe based on provider and one or more of module name, function
8447  * name and probe name.
8448  */
8449 dtrace_id_t
dtrace_probe_lookup(dtrace_provider_id_t prid,const char * mod,const char * func,const char * name)8450 dtrace_probe_lookup(dtrace_provider_id_t prid, const char *mod,
8451     const char *func, const char *name)
8452 {
8453 	dtrace_probekey_t pkey;
8454 	dtrace_id_t id;
8455 	int match;
8456 
8457 	pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name;
8458 	pkey.dtpk_pmatch = &dtrace_match_string;
8459 	pkey.dtpk_mod = mod;
8460 	pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul;
8461 	pkey.dtpk_func = func;
8462 	pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul;
8463 	pkey.dtpk_name = name;
8464 	pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul;
8465 	pkey.dtpk_id = DTRACE_IDNONE;
8466 
8467 	mutex_enter(&dtrace_lock);
8468 	match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0,
8469 	    dtrace_probe_lookup_match, &id);
8470 	mutex_exit(&dtrace_lock);
8471 
8472 	ASSERT(match == 1 || match == 0);
8473 	return (match ? id : 0);
8474 }
8475 
8476 /*
8477  * Returns the probe argument associated with the specified probe.
8478  */
8479 void *
dtrace_probe_arg(dtrace_provider_id_t id,dtrace_id_t pid)8480 dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid)
8481 {
8482 	dtrace_probe_t *probe;
8483 	void *rval = NULL;
8484 
8485 	mutex_enter(&dtrace_lock);
8486 
8487 	if ((probe = dtrace_probe_lookup_id(pid)) != NULL &&
8488 	    probe->dtpr_provider == (dtrace_provider_t *)id)
8489 		rval = probe->dtpr_arg;
8490 
8491 	mutex_exit(&dtrace_lock);
8492 
8493 	return (rval);
8494 }
8495 
8496 /*
8497  * Copy a probe into a probe description.
8498  */
8499 static void
dtrace_probe_description(const dtrace_probe_t * prp,dtrace_probedesc_t * pdp)8500 dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp)
8501 {
8502 	bzero(pdp, sizeof (dtrace_probedesc_t));
8503 	pdp->dtpd_id = prp->dtpr_id;
8504 
8505 	(void) strncpy(pdp->dtpd_provider,
8506 	    prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1);
8507 
8508 	(void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1);
8509 	(void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1);
8510 	(void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1);
8511 }
8512 
8513 /*
8514  * Called to indicate that a probe -- or probes -- should be provided by a
8515  * specfied provider.  If the specified description is NULL, the provider will
8516  * be told to provide all of its probes.  (This is done whenever a new
8517  * consumer comes along, or whenever a retained enabling is to be matched.) If
8518  * the specified description is non-NULL, the provider is given the
8519  * opportunity to dynamically provide the specified probe, allowing providers
8520  * to support the creation of probes on-the-fly.  (So-called _autocreated_
8521  * probes.)  If the provider is NULL, the operations will be applied to all
8522  * providers; if the provider is non-NULL the operations will only be applied
8523  * to the specified provider.  The dtrace_provider_lock must be held, and the
8524  * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation
8525  * will need to grab the dtrace_lock when it reenters the framework through
8526  * dtrace_probe_lookup(), dtrace_probe_create(), etc.
8527  */
8528 static void
dtrace_probe_provide(dtrace_probedesc_t * desc,dtrace_provider_t * prv)8529 dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv)
8530 {
8531 	struct modctl *ctl;
8532 	int all = 0;
8533 
8534 	ASSERT(MUTEX_HELD(&dtrace_provider_lock));
8535 
8536 	if (prv == NULL) {
8537 		all = 1;
8538 		prv = dtrace_provider;
8539 	}
8540 
8541 	do {
8542 		/*
8543 		 * First, call the blanket provide operation.
8544 		 */
8545 		prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc);
8546 
8547 		/*
8548 		 * Now call the per-module provide operation.  We will grab
8549 		 * mod_lock to prevent the list from being modified.  Note
8550 		 * that this also prevents the mod_busy bits from changing.
8551 		 * (mod_busy can only be changed with mod_lock held.)
8552 		 */
8553 		mutex_enter(&mod_lock);
8554 
8555 		ctl = &modules;
8556 		do {
8557 			if (ctl->mod_busy || ctl->mod_mp == NULL)
8558 				continue;
8559 
8560 			prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
8561 
8562 		} while ((ctl = ctl->mod_next) != &modules);
8563 
8564 		mutex_exit(&mod_lock);
8565 	} while (all && (prv = prv->dtpv_next) != NULL);
8566 }
8567 
8568 /*
8569  * Iterate over each probe, and call the Framework-to-Provider API function
8570  * denoted by offs.
8571  */
8572 static void
dtrace_probe_foreach(uintptr_t offs)8573 dtrace_probe_foreach(uintptr_t offs)
8574 {
8575 	dtrace_provider_t *prov;
8576 	void (*func)(void *, dtrace_id_t, void *);
8577 	dtrace_probe_t *probe;
8578 	dtrace_icookie_t cookie;
8579 	int i;
8580 
8581 	/*
8582 	 * We disable interrupts to walk through the probe array.  This is
8583 	 * safe -- the dtrace_sync() in dtrace_unregister() assures that we
8584 	 * won't see stale data.
8585 	 */
8586 	cookie = dtrace_interrupt_disable();
8587 
8588 	for (i = 0; i < dtrace_nprobes; i++) {
8589 		if ((probe = dtrace_probes[i]) == NULL)
8590 			continue;
8591 
8592 		if (probe->dtpr_ecb == NULL) {
8593 			/*
8594 			 * This probe isn't enabled -- don't call the function.
8595 			 */
8596 			continue;
8597 		}
8598 
8599 		prov = probe->dtpr_provider;
8600 		func = *((void(**)(void *, dtrace_id_t, void *))
8601 		    ((uintptr_t)&prov->dtpv_pops + offs));
8602 
8603 		func(prov->dtpv_arg, i + 1, probe->dtpr_arg);
8604 	}
8605 
8606 	dtrace_interrupt_enable(cookie);
8607 }
8608 
8609 static int
dtrace_probe_enable(const dtrace_probedesc_t * desc,dtrace_enabling_t * enab)8610 dtrace_probe_enable(const dtrace_probedesc_t *desc, dtrace_enabling_t *enab)
8611 {
8612 	dtrace_probekey_t pkey;
8613 	uint32_t priv;
8614 	uid_t uid;
8615 	zoneid_t zoneid;
8616 
8617 	ASSERT(MUTEX_HELD(&dtrace_lock));
8618 	dtrace_ecb_create_cache = NULL;
8619 
8620 	if (desc == NULL) {
8621 		/*
8622 		 * If we're passed a NULL description, we're being asked to
8623 		 * create an ECB with a NULL probe.
8624 		 */
8625 		(void) dtrace_ecb_create_enable(NULL, enab);
8626 		return (0);
8627 	}
8628 
8629 	dtrace_probekey(desc, &pkey);
8630 	dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred,
8631 	    &priv, &uid, &zoneid);
8632 
8633 	return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable,
8634 	    enab));
8635 }
8636 
8637 /*
8638  * DTrace Helper Provider Functions
8639  */
8640 static void
dtrace_dofattr2attr(dtrace_attribute_t * attr,const dof_attr_t dofattr)8641 dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr)
8642 {
8643 	attr->dtat_name = DOF_ATTR_NAME(dofattr);
8644 	attr->dtat_data = DOF_ATTR_DATA(dofattr);
8645 	attr->dtat_class = DOF_ATTR_CLASS(dofattr);
8646 }
8647 
8648 static void
dtrace_dofprov2hprov(dtrace_helper_provdesc_t * hprov,const dof_provider_t * dofprov,char * strtab)8649 dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov,
8650     const dof_provider_t *dofprov, char *strtab)
8651 {
8652 	hprov->dthpv_provname = strtab + dofprov->dofpv_name;
8653 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider,
8654 	    dofprov->dofpv_provattr);
8655 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod,
8656 	    dofprov->dofpv_modattr);
8657 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func,
8658 	    dofprov->dofpv_funcattr);
8659 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name,
8660 	    dofprov->dofpv_nameattr);
8661 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args,
8662 	    dofprov->dofpv_argsattr);
8663 }
8664 
8665 static void
dtrace_helper_provide_one(dof_helper_t * dhp,dof_sec_t * sec,pid_t pid)8666 dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8667 {
8668 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8669 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8670 	dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
8671 	dof_provider_t *provider;
8672 	dof_probe_t *probe;
8673 	uint32_t *off, *enoff;
8674 	uint8_t *arg;
8675 	char *strtab;
8676 	uint_t i, nprobes;
8677 	dtrace_helper_provdesc_t dhpv;
8678 	dtrace_helper_probedesc_t dhpb;
8679 	dtrace_meta_t *meta = dtrace_meta_pid;
8680 	dtrace_mops_t *mops = &meta->dtm_mops;
8681 	void *parg;
8682 
8683 	provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8684 	str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8685 	    provider->dofpv_strtab * dof->dofh_secsize);
8686 	prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8687 	    provider->dofpv_probes * dof->dofh_secsize);
8688 	arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8689 	    provider->dofpv_prargs * dof->dofh_secsize);
8690 	off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8691 	    provider->dofpv_proffs * dof->dofh_secsize);
8692 
8693 	strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8694 	off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset);
8695 	arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
8696 	enoff = NULL;
8697 
8698 	/*
8699 	 * See dtrace_helper_provider_validate().
8700 	 */
8701 	if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
8702 	    provider->dofpv_prenoffs != DOF_SECT_NONE) {
8703 		enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8704 		    provider->dofpv_prenoffs * dof->dofh_secsize);
8705 		enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset);
8706 	}
8707 
8708 	nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
8709 
8710 	/*
8711 	 * Create the provider.
8712 	 */
8713 	dtrace_dofprov2hprov(&dhpv, provider, strtab);
8714 
8715 	if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL)
8716 		return;
8717 
8718 	meta->dtm_count++;
8719 
8720 	/*
8721 	 * Create the probes.
8722 	 */
8723 	for (i = 0; i < nprobes; i++) {
8724 		probe = (dof_probe_t *)(uintptr_t)(daddr +
8725 		    prb_sec->dofs_offset + i * prb_sec->dofs_entsize);
8726 
8727 		dhpb.dthpb_mod = dhp->dofhp_mod;
8728 		dhpb.dthpb_func = strtab + probe->dofpr_func;
8729 		dhpb.dthpb_name = strtab + probe->dofpr_name;
8730 		dhpb.dthpb_base = probe->dofpr_addr;
8731 		dhpb.dthpb_offs = off + probe->dofpr_offidx;
8732 		dhpb.dthpb_noffs = probe->dofpr_noffs;
8733 		if (enoff != NULL) {
8734 			dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx;
8735 			dhpb.dthpb_nenoffs = probe->dofpr_nenoffs;
8736 		} else {
8737 			dhpb.dthpb_enoffs = NULL;
8738 			dhpb.dthpb_nenoffs = 0;
8739 		}
8740 		dhpb.dthpb_args = arg + probe->dofpr_argidx;
8741 		dhpb.dthpb_nargc = probe->dofpr_nargc;
8742 		dhpb.dthpb_xargc = probe->dofpr_xargc;
8743 		dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv;
8744 		dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv;
8745 
8746 		mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb);
8747 	}
8748 }
8749 
8750 static void
dtrace_helper_provide(dof_helper_t * dhp,pid_t pid)8751 dtrace_helper_provide(dof_helper_t *dhp, pid_t pid)
8752 {
8753 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8754 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8755 	int i;
8756 
8757 	ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8758 
8759 	for (i = 0; i < dof->dofh_secnum; i++) {
8760 		dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8761 		    dof->dofh_secoff + i * dof->dofh_secsize);
8762 
8763 		if (sec->dofs_type != DOF_SECT_PROVIDER)
8764 			continue;
8765 
8766 		dtrace_helper_provide_one(dhp, sec, pid);
8767 	}
8768 
8769 	/*
8770 	 * We may have just created probes, so we must now rematch against
8771 	 * any retained enablings.  Note that this call will acquire both
8772 	 * cpu_lock and dtrace_lock; the fact that we are holding
8773 	 * dtrace_meta_lock now is what defines the ordering with respect to
8774 	 * these three locks.
8775 	 */
8776 	dtrace_enabling_matchall();
8777 }
8778 
8779 static void
dtrace_helper_provider_remove_one(dof_helper_t * dhp,dof_sec_t * sec,pid_t pid)8780 dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8781 {
8782 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8783 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8784 	dof_sec_t *str_sec;
8785 	dof_provider_t *provider;
8786 	char *strtab;
8787 	dtrace_helper_provdesc_t dhpv;
8788 	dtrace_meta_t *meta = dtrace_meta_pid;
8789 	dtrace_mops_t *mops = &meta->dtm_mops;
8790 
8791 	provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8792 	str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8793 	    provider->dofpv_strtab * dof->dofh_secsize);
8794 
8795 	strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8796 
8797 	/*
8798 	 * Create the provider.
8799 	 */
8800 	dtrace_dofprov2hprov(&dhpv, provider, strtab);
8801 
8802 	mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid);
8803 
8804 	meta->dtm_count--;
8805 }
8806 
8807 static void
dtrace_helper_provider_remove(dof_helper_t * dhp,pid_t pid)8808 dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid)
8809 {
8810 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8811 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8812 	int i;
8813 
8814 	ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8815 
8816 	for (i = 0; i < dof->dofh_secnum; i++) {
8817 		dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8818 		    dof->dofh_secoff + i * dof->dofh_secsize);
8819 
8820 		if (sec->dofs_type != DOF_SECT_PROVIDER)
8821 			continue;
8822 
8823 		dtrace_helper_provider_remove_one(dhp, sec, pid);
8824 	}
8825 }
8826 
8827 /*
8828  * DTrace Meta Provider-to-Framework API Functions
8829  *
8830  * These functions implement the Meta Provider-to-Framework API, as described
8831  * in <sys/dtrace.h>.
8832  */
8833 int
dtrace_meta_register(const char * name,const dtrace_mops_t * mops,void * arg,dtrace_meta_provider_id_t * idp)8834 dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg,
8835     dtrace_meta_provider_id_t *idp)
8836 {
8837 	dtrace_meta_t *meta;
8838 	dtrace_helpers_t *help, *next;
8839 	int i;
8840 
8841 	*idp = DTRACE_METAPROVNONE;
8842 
8843 	/*
8844 	 * We strictly don't need the name, but we hold onto it for
8845 	 * debuggability. All hail error queues!
8846 	 */
8847 	if (name == NULL) {
8848 		cmn_err(CE_WARN, "failed to register meta-provider: "
8849 		    "invalid name");
8850 		return (EINVAL);
8851 	}
8852 
8853 	if (mops == NULL ||
8854 	    mops->dtms_create_probe == NULL ||
8855 	    mops->dtms_provide_pid == NULL ||
8856 	    mops->dtms_remove_pid == NULL) {
8857 		cmn_err(CE_WARN, "failed to register meta-register %s: "
8858 		    "invalid ops", name);
8859 		return (EINVAL);
8860 	}
8861 
8862 	meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP);
8863 	meta->dtm_mops = *mops;
8864 	meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
8865 	(void) strcpy(meta->dtm_name, name);
8866 	meta->dtm_arg = arg;
8867 
8868 	mutex_enter(&dtrace_meta_lock);
8869 	mutex_enter(&dtrace_lock);
8870 
8871 	if (dtrace_meta_pid != NULL) {
8872 		mutex_exit(&dtrace_lock);
8873 		mutex_exit(&dtrace_meta_lock);
8874 		cmn_err(CE_WARN, "failed to register meta-register %s: "
8875 		    "user-land meta-provider exists", name);
8876 		kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1);
8877 		kmem_free(meta, sizeof (dtrace_meta_t));
8878 		return (EINVAL);
8879 	}
8880 
8881 	dtrace_meta_pid = meta;
8882 	*idp = (dtrace_meta_provider_id_t)meta;
8883 
8884 	/*
8885 	 * If there are providers and probes ready to go, pass them
8886 	 * off to the new meta provider now.
8887 	 */
8888 
8889 	help = dtrace_deferred_pid;
8890 	dtrace_deferred_pid = NULL;
8891 
8892 	mutex_exit(&dtrace_lock);
8893 
8894 	while (help != NULL) {
8895 		for (i = 0; i < help->dthps_nprovs; i++) {
8896 			dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
8897 			    help->dthps_pid);
8898 		}
8899 
8900 		next = help->dthps_next;
8901 		help->dthps_next = NULL;
8902 		help->dthps_prev = NULL;
8903 		help->dthps_deferred = 0;
8904 		help = next;
8905 	}
8906 
8907 	mutex_exit(&dtrace_meta_lock);
8908 
8909 	return (0);
8910 }
8911 
8912 int
dtrace_meta_unregister(dtrace_meta_provider_id_t id)8913 dtrace_meta_unregister(dtrace_meta_provider_id_t id)
8914 {
8915 	dtrace_meta_t **pp, *old = (dtrace_meta_t *)id;
8916 
8917 	mutex_enter(&dtrace_meta_lock);
8918 	mutex_enter(&dtrace_lock);
8919 
8920 	if (old == dtrace_meta_pid) {
8921 		pp = &dtrace_meta_pid;
8922 	} else {
8923 		panic("attempt to unregister non-existent "
8924 		    "dtrace meta-provider %p\n", (void *)old);
8925 	}
8926 
8927 	if (old->dtm_count != 0) {
8928 		mutex_exit(&dtrace_lock);
8929 		mutex_exit(&dtrace_meta_lock);
8930 		return (EBUSY);
8931 	}
8932 
8933 	*pp = NULL;
8934 
8935 	mutex_exit(&dtrace_lock);
8936 	mutex_exit(&dtrace_meta_lock);
8937 
8938 	kmem_free(old->dtm_name, strlen(old->dtm_name) + 1);
8939 	kmem_free(old, sizeof (dtrace_meta_t));
8940 
8941 	return (0);
8942 }
8943 
8944 
8945 /*
8946  * DTrace DIF Object Functions
8947  */
8948 static int
dtrace_difo_err(uint_t pc,const char * format,...)8949 dtrace_difo_err(uint_t pc, const char *format, ...)
8950 {
8951 	if (dtrace_err_verbose) {
8952 		va_list alist;
8953 
8954 		(void) uprintf("dtrace DIF object error: [%u]: ", pc);
8955 		va_start(alist, format);
8956 		(void) vuprintf(format, alist);
8957 		va_end(alist);
8958 	}
8959 
8960 #ifdef DTRACE_ERRDEBUG
8961 	dtrace_errdebug(format);
8962 #endif
8963 	return (1);
8964 }
8965 
8966 /*
8967  * Validate a DTrace DIF object by checking the IR instructions.  The following
8968  * rules are currently enforced by dtrace_difo_validate():
8969  *
8970  * 1. Each instruction must have a valid opcode
8971  * 2. Each register, string, variable, or subroutine reference must be valid
8972  * 3. No instruction can modify register %r0 (must be zero)
8973  * 4. All instruction reserved bits must be set to zero
8974  * 5. The last instruction must be a "ret" instruction
8975  * 6. All branch targets must reference a valid instruction _after_ the branch
8976  */
8977 static int
dtrace_difo_validate(dtrace_difo_t * dp,dtrace_vstate_t * vstate,uint_t nregs,cred_t * cr)8978 dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs,
8979     cred_t *cr)
8980 {
8981 	int err = 0, i;
8982 	int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
8983 	int kcheckload;
8984 	uint_t pc;
8985 
8986 	kcheckload = cr == NULL ||
8987 	    (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0;
8988 
8989 	dp->dtdo_destructive = 0;
8990 
8991 	for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) {
8992 		dif_instr_t instr = dp->dtdo_buf[pc];
8993 
8994 		uint_t r1 = DIF_INSTR_R1(instr);
8995 		uint_t r2 = DIF_INSTR_R2(instr);
8996 		uint_t rd = DIF_INSTR_RD(instr);
8997 		uint_t rs = DIF_INSTR_RS(instr);
8998 		uint_t label = DIF_INSTR_LABEL(instr);
8999 		uint_t v = DIF_INSTR_VAR(instr);
9000 		uint_t subr = DIF_INSTR_SUBR(instr);
9001 		uint_t type = DIF_INSTR_TYPE(instr);
9002 		uint_t op = DIF_INSTR_OP(instr);
9003 
9004 		switch (op) {
9005 		case DIF_OP_OR:
9006 		case DIF_OP_XOR:
9007 		case DIF_OP_AND:
9008 		case DIF_OP_SLL:
9009 		case DIF_OP_SRL:
9010 		case DIF_OP_SRA:
9011 		case DIF_OP_SUB:
9012 		case DIF_OP_ADD:
9013 		case DIF_OP_MUL:
9014 		case DIF_OP_SDIV:
9015 		case DIF_OP_UDIV:
9016 		case DIF_OP_SREM:
9017 		case DIF_OP_UREM:
9018 		case DIF_OP_COPYS:
9019 			if (r1 >= nregs)
9020 				err += efunc(pc, "invalid register %u\n", r1);
9021 			if (r2 >= nregs)
9022 				err += efunc(pc, "invalid register %u\n", r2);
9023 			if (rd >= nregs)
9024 				err += efunc(pc, "invalid register %u\n", rd);
9025 			if (rd == 0)
9026 				err += efunc(pc, "cannot write to %r0\n");
9027 			break;
9028 		case DIF_OP_NOT:
9029 		case DIF_OP_MOV:
9030 		case DIF_OP_ALLOCS:
9031 			if (r1 >= nregs)
9032 				err += efunc(pc, "invalid register %u\n", r1);
9033 			if (r2 != 0)
9034 				err += efunc(pc, "non-zero reserved bits\n");
9035 			if (rd >= nregs)
9036 				err += efunc(pc, "invalid register %u\n", rd);
9037 			if (rd == 0)
9038 				err += efunc(pc, "cannot write to %r0\n");
9039 			break;
9040 		case DIF_OP_LDSB:
9041 		case DIF_OP_LDSH:
9042 		case DIF_OP_LDSW:
9043 		case DIF_OP_LDUB:
9044 		case DIF_OP_LDUH:
9045 		case DIF_OP_LDUW:
9046 		case DIF_OP_LDX:
9047 			if (r1 >= nregs)
9048 				err += efunc(pc, "invalid register %u\n", r1);
9049 			if (r2 != 0)
9050 				err += efunc(pc, "non-zero reserved bits\n");
9051 			if (rd >= nregs)
9052 				err += efunc(pc, "invalid register %u\n", rd);
9053 			if (rd == 0)
9054 				err += efunc(pc, "cannot write to %r0\n");
9055 			if (kcheckload)
9056 				dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op +
9057 				    DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd);
9058 			break;
9059 		case DIF_OP_RLDSB:
9060 		case DIF_OP_RLDSH:
9061 		case DIF_OP_RLDSW:
9062 		case DIF_OP_RLDUB:
9063 		case DIF_OP_RLDUH:
9064 		case DIF_OP_RLDUW:
9065 		case DIF_OP_RLDX:
9066 			if (r1 >= nregs)
9067 				err += efunc(pc, "invalid register %u\n", r1);
9068 			if (r2 != 0)
9069 				err += efunc(pc, "non-zero reserved bits\n");
9070 			if (rd >= nregs)
9071 				err += efunc(pc, "invalid register %u\n", rd);
9072 			if (rd == 0)
9073 				err += efunc(pc, "cannot write to %r0\n");
9074 			break;
9075 		case DIF_OP_ULDSB:
9076 		case DIF_OP_ULDSH:
9077 		case DIF_OP_ULDSW:
9078 		case DIF_OP_ULDUB:
9079 		case DIF_OP_ULDUH:
9080 		case DIF_OP_ULDUW:
9081 		case DIF_OP_ULDX:
9082 			if (r1 >= nregs)
9083 				err += efunc(pc, "invalid register %u\n", r1);
9084 			if (r2 != 0)
9085 				err += efunc(pc, "non-zero reserved bits\n");
9086 			if (rd >= nregs)
9087 				err += efunc(pc, "invalid register %u\n", rd);
9088 			if (rd == 0)
9089 				err += efunc(pc, "cannot write to %r0\n");
9090 			break;
9091 		case DIF_OP_STB:
9092 		case DIF_OP_STH:
9093 		case DIF_OP_STW:
9094 		case DIF_OP_STX:
9095 			if (r1 >= nregs)
9096 				err += efunc(pc, "invalid register %u\n", r1);
9097 			if (r2 != 0)
9098 				err += efunc(pc, "non-zero reserved bits\n");
9099 			if (rd >= nregs)
9100 				err += efunc(pc, "invalid register %u\n", rd);
9101 			if (rd == 0)
9102 				err += efunc(pc, "cannot write to 0 address\n");
9103 			break;
9104 		case DIF_OP_CMP:
9105 		case DIF_OP_SCMP:
9106 			if (r1 >= nregs)
9107 				err += efunc(pc, "invalid register %u\n", r1);
9108 			if (r2 >= nregs)
9109 				err += efunc(pc, "invalid register %u\n", r2);
9110 			if (rd != 0)
9111 				err += efunc(pc, "non-zero reserved bits\n");
9112 			break;
9113 		case DIF_OP_TST:
9114 			if (r1 >= nregs)
9115 				err += efunc(pc, "invalid register %u\n", r1);
9116 			if (r2 != 0 || rd != 0)
9117 				err += efunc(pc, "non-zero reserved bits\n");
9118 			break;
9119 		case DIF_OP_BA:
9120 		case DIF_OP_BE:
9121 		case DIF_OP_BNE:
9122 		case DIF_OP_BG:
9123 		case DIF_OP_BGU:
9124 		case DIF_OP_BGE:
9125 		case DIF_OP_BGEU:
9126 		case DIF_OP_BL:
9127 		case DIF_OP_BLU:
9128 		case DIF_OP_BLE:
9129 		case DIF_OP_BLEU:
9130 			if (label >= dp->dtdo_len) {
9131 				err += efunc(pc, "invalid branch target %u\n",
9132 				    label);
9133 			}
9134 			if (label <= pc) {
9135 				err += efunc(pc, "backward branch to %u\n",
9136 				    label);
9137 			}
9138 			break;
9139 		case DIF_OP_RET:
9140 			if (r1 != 0 || r2 != 0)
9141 				err += efunc(pc, "non-zero reserved bits\n");
9142 			if (rd >= nregs)
9143 				err += efunc(pc, "invalid register %u\n", rd);
9144 			break;
9145 		case DIF_OP_NOP:
9146 		case DIF_OP_POPTS:
9147 		case DIF_OP_FLUSHTS:
9148 			if (r1 != 0 || r2 != 0 || rd != 0)
9149 				err += efunc(pc, "non-zero reserved bits\n");
9150 			break;
9151 		case DIF_OP_SETX:
9152 			if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) {
9153 				err += efunc(pc, "invalid integer ref %u\n",
9154 				    DIF_INSTR_INTEGER(instr));
9155 			}
9156 			if (rd >= nregs)
9157 				err += efunc(pc, "invalid register %u\n", rd);
9158 			if (rd == 0)
9159 				err += efunc(pc, "cannot write to %r0\n");
9160 			break;
9161 		case DIF_OP_SETS:
9162 			if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) {
9163 				err += efunc(pc, "invalid string ref %u\n",
9164 				    DIF_INSTR_STRING(instr));
9165 			}
9166 			if (rd >= nregs)
9167 				err += efunc(pc, "invalid register %u\n", rd);
9168 			if (rd == 0)
9169 				err += efunc(pc, "cannot write to %r0\n");
9170 			break;
9171 		case DIF_OP_LDGA:
9172 		case DIF_OP_LDTA:
9173 			if (r1 > DIF_VAR_ARRAY_MAX)
9174 				err += efunc(pc, "invalid array %u\n", r1);
9175 			if (r2 >= nregs)
9176 				err += efunc(pc, "invalid register %u\n", r2);
9177 			if (rd >= nregs)
9178 				err += efunc(pc, "invalid register %u\n", rd);
9179 			if (rd == 0)
9180 				err += efunc(pc, "cannot write to %r0\n");
9181 			break;
9182 		case DIF_OP_LDGS:
9183 		case DIF_OP_LDTS:
9184 		case DIF_OP_LDLS:
9185 		case DIF_OP_LDGAA:
9186 		case DIF_OP_LDTAA:
9187 			if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX)
9188 				err += efunc(pc, "invalid variable %u\n", v);
9189 			if (rd >= nregs)
9190 				err += efunc(pc, "invalid register %u\n", rd);
9191 			if (rd == 0)
9192 				err += efunc(pc, "cannot write to %r0\n");
9193 			break;
9194 		case DIF_OP_STGS:
9195 		case DIF_OP_STTS:
9196 		case DIF_OP_STLS:
9197 		case DIF_OP_STGAA:
9198 		case DIF_OP_STTAA:
9199 			if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX)
9200 				err += efunc(pc, "invalid variable %u\n", v);
9201 			if (rs >= nregs)
9202 				err += efunc(pc, "invalid register %u\n", rd);
9203 			break;
9204 		case DIF_OP_CALL:
9205 			if (subr > DIF_SUBR_MAX)
9206 				err += efunc(pc, "invalid subr %u\n", subr);
9207 			if (rd >= nregs)
9208 				err += efunc(pc, "invalid register %u\n", rd);
9209 			if (rd == 0)
9210 				err += efunc(pc, "cannot write to %r0\n");
9211 
9212 			if (subr == DIF_SUBR_COPYOUT ||
9213 			    subr == DIF_SUBR_COPYOUTSTR) {
9214 				dp->dtdo_destructive = 1;
9215 			}
9216 
9217 			if (subr == DIF_SUBR_GETF) {
9218 				/*
9219 				 * If we have a getf() we need to record that
9220 				 * in our state.  Note that our state can be
9221 				 * NULL if this is a helper -- but in that
9222 				 * case, the call to getf() is itself illegal,
9223 				 * and will be caught (slightly later) when
9224 				 * the helper is validated.
9225 				 */
9226 				if (vstate->dtvs_state != NULL)
9227 					vstate->dtvs_state->dts_getf++;
9228 			}
9229 
9230 			break;
9231 		case DIF_OP_PUSHTR:
9232 			if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF)
9233 				err += efunc(pc, "invalid ref type %u\n", type);
9234 			if (r2 >= nregs)
9235 				err += efunc(pc, "invalid register %u\n", r2);
9236 			if (rs >= nregs)
9237 				err += efunc(pc, "invalid register %u\n", rs);
9238 			break;
9239 		case DIF_OP_PUSHTV:
9240 			if (type != DIF_TYPE_CTF)
9241 				err += efunc(pc, "invalid val type %u\n", type);
9242 			if (r2 >= nregs)
9243 				err += efunc(pc, "invalid register %u\n", r2);
9244 			if (rs >= nregs)
9245 				err += efunc(pc, "invalid register %u\n", rs);
9246 			break;
9247 		default:
9248 			err += efunc(pc, "invalid opcode %u\n",
9249 			    DIF_INSTR_OP(instr));
9250 		}
9251 	}
9252 
9253 	if (dp->dtdo_len != 0 &&
9254 	    DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) {
9255 		err += efunc(dp->dtdo_len - 1,
9256 		    "expected 'ret' as last DIF instruction\n");
9257 	}
9258 
9259 	if (!(dp->dtdo_rtype.dtdt_flags & (DIF_TF_BYREF | DIF_TF_BYUREF))) {
9260 		/*
9261 		 * If we're not returning by reference, the size must be either
9262 		 * 0 or the size of one of the base types.
9263 		 */
9264 		switch (dp->dtdo_rtype.dtdt_size) {
9265 		case 0:
9266 		case sizeof (uint8_t):
9267 		case sizeof (uint16_t):
9268 		case sizeof (uint32_t):
9269 		case sizeof (uint64_t):
9270 			break;
9271 
9272 		default:
9273 			err += efunc(dp->dtdo_len - 1, "bad return size\n");
9274 		}
9275 	}
9276 
9277 	for (i = 0; i < dp->dtdo_varlen && err == 0; i++) {
9278 		dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL;
9279 		dtrace_diftype_t *vt, *et;
9280 		uint_t id, ndx;
9281 
9282 		if (v->dtdv_scope != DIFV_SCOPE_GLOBAL &&
9283 		    v->dtdv_scope != DIFV_SCOPE_THREAD &&
9284 		    v->dtdv_scope != DIFV_SCOPE_LOCAL) {
9285 			err += efunc(i, "unrecognized variable scope %d\n",
9286 			    v->dtdv_scope);
9287 			break;
9288 		}
9289 
9290 		if (v->dtdv_kind != DIFV_KIND_ARRAY &&
9291 		    v->dtdv_kind != DIFV_KIND_SCALAR) {
9292 			err += efunc(i, "unrecognized variable type %d\n",
9293 			    v->dtdv_kind);
9294 			break;
9295 		}
9296 
9297 		if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) {
9298 			err += efunc(i, "%d exceeds variable id limit\n", id);
9299 			break;
9300 		}
9301 
9302 		if (id < DIF_VAR_OTHER_UBASE)
9303 			continue;
9304 
9305 		/*
9306 		 * For user-defined variables, we need to check that this
9307 		 * definition is identical to any previous definition that we
9308 		 * encountered.
9309 		 */
9310 		ndx = id - DIF_VAR_OTHER_UBASE;
9311 
9312 		switch (v->dtdv_scope) {
9313 		case DIFV_SCOPE_GLOBAL:
9314 			if (ndx < vstate->dtvs_nglobals) {
9315 				dtrace_statvar_t *svar;
9316 
9317 				if ((svar = vstate->dtvs_globals[ndx]) != NULL)
9318 					existing = &svar->dtsv_var;
9319 			}
9320 
9321 			break;
9322 
9323 		case DIFV_SCOPE_THREAD:
9324 			if (ndx < vstate->dtvs_ntlocals)
9325 				existing = &vstate->dtvs_tlocals[ndx];
9326 			break;
9327 
9328 		case DIFV_SCOPE_LOCAL:
9329 			if (ndx < vstate->dtvs_nlocals) {
9330 				dtrace_statvar_t *svar;
9331 
9332 				if ((svar = vstate->dtvs_locals[ndx]) != NULL)
9333 					existing = &svar->dtsv_var;
9334 			}
9335 
9336 			break;
9337 		}
9338 
9339 		vt = &v->dtdv_type;
9340 
9341 		if (vt->dtdt_flags & DIF_TF_BYREF) {
9342 			if (vt->dtdt_size == 0) {
9343 				err += efunc(i, "zero-sized variable\n");
9344 				break;
9345 			}
9346 
9347 			if (v->dtdv_scope == DIFV_SCOPE_GLOBAL &&
9348 			    vt->dtdt_size > dtrace_global_maxsize) {
9349 				err += efunc(i, "oversized by-ref global\n");
9350 				break;
9351 			}
9352 		}
9353 
9354 		if (existing == NULL || existing->dtdv_id == 0)
9355 			continue;
9356 
9357 		ASSERT(existing->dtdv_id == v->dtdv_id);
9358 		ASSERT(existing->dtdv_scope == v->dtdv_scope);
9359 
9360 		if (existing->dtdv_kind != v->dtdv_kind)
9361 			err += efunc(i, "%d changed variable kind\n", id);
9362 
9363 		et = &existing->dtdv_type;
9364 
9365 		if (vt->dtdt_flags != et->dtdt_flags) {
9366 			err += efunc(i, "%d changed variable type flags\n", id);
9367 			break;
9368 		}
9369 
9370 		if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) {
9371 			err += efunc(i, "%d changed variable type size\n", id);
9372 			break;
9373 		}
9374 	}
9375 
9376 	return (err);
9377 }
9378 
9379 /*
9380  * Validate a DTrace DIF object that it is to be used as a helper.  Helpers
9381  * are much more constrained than normal DIFOs.  Specifically, they may
9382  * not:
9383  *
9384  * 1. Make calls to subroutines other than copyin(), copyinstr() or
9385  *    miscellaneous string routines
9386  * 2. Access DTrace variables other than the args[] array, and the
9387  *    curthread, pid, ppid, tid, execname, zonename, uid and gid variables.
9388  * 3. Have thread-local variables.
9389  * 4. Have dynamic variables.
9390  */
9391 static int
dtrace_difo_validate_helper(dtrace_difo_t * dp)9392 dtrace_difo_validate_helper(dtrace_difo_t *dp)
9393 {
9394 	int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
9395 	int err = 0;
9396 	uint_t pc;
9397 
9398 	for (pc = 0; pc < dp->dtdo_len; pc++) {
9399 		dif_instr_t instr = dp->dtdo_buf[pc];
9400 
9401 		uint_t v = DIF_INSTR_VAR(instr);
9402 		uint_t subr = DIF_INSTR_SUBR(instr);
9403 		uint_t op = DIF_INSTR_OP(instr);
9404 
9405 		switch (op) {
9406 		case DIF_OP_OR:
9407 		case DIF_OP_XOR:
9408 		case DIF_OP_AND:
9409 		case DIF_OP_SLL:
9410 		case DIF_OP_SRL:
9411 		case DIF_OP_SRA:
9412 		case DIF_OP_SUB:
9413 		case DIF_OP_ADD:
9414 		case DIF_OP_MUL:
9415 		case DIF_OP_SDIV:
9416 		case DIF_OP_UDIV:
9417 		case DIF_OP_SREM:
9418 		case DIF_OP_UREM:
9419 		case DIF_OP_COPYS:
9420 		case DIF_OP_NOT:
9421 		case DIF_OP_MOV:
9422 		case DIF_OP_RLDSB:
9423 		case DIF_OP_RLDSH:
9424 		case DIF_OP_RLDSW:
9425 		case DIF_OP_RLDUB:
9426 		case DIF_OP_RLDUH:
9427 		case DIF_OP_RLDUW:
9428 		case DIF_OP_RLDX:
9429 		case DIF_OP_ULDSB:
9430 		case DIF_OP_ULDSH:
9431 		case DIF_OP_ULDSW:
9432 		case DIF_OP_ULDUB:
9433 		case DIF_OP_ULDUH:
9434 		case DIF_OP_ULDUW:
9435 		case DIF_OP_ULDX:
9436 		case DIF_OP_STB:
9437 		case DIF_OP_STH:
9438 		case DIF_OP_STW:
9439 		case DIF_OP_STX:
9440 		case DIF_OP_ALLOCS:
9441 		case DIF_OP_CMP:
9442 		case DIF_OP_SCMP:
9443 		case DIF_OP_TST:
9444 		case DIF_OP_BA:
9445 		case DIF_OP_BE:
9446 		case DIF_OP_BNE:
9447 		case DIF_OP_BG:
9448 		case DIF_OP_BGU:
9449 		case DIF_OP_BGE:
9450 		case DIF_OP_BGEU:
9451 		case DIF_OP_BL:
9452 		case DIF_OP_BLU:
9453 		case DIF_OP_BLE:
9454 		case DIF_OP_BLEU:
9455 		case DIF_OP_RET:
9456 		case DIF_OP_NOP:
9457 		case DIF_OP_POPTS:
9458 		case DIF_OP_FLUSHTS:
9459 		case DIF_OP_SETX:
9460 		case DIF_OP_SETS:
9461 		case DIF_OP_LDGA:
9462 		case DIF_OP_LDLS:
9463 		case DIF_OP_STGS:
9464 		case DIF_OP_STLS:
9465 		case DIF_OP_PUSHTR:
9466 		case DIF_OP_PUSHTV:
9467 			break;
9468 
9469 		case DIF_OP_LDGS:
9470 			if (v >= DIF_VAR_OTHER_UBASE)
9471 				break;
9472 
9473 			if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9)
9474 				break;
9475 
9476 			if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID ||
9477 			    v == DIF_VAR_PPID || v == DIF_VAR_TID ||
9478 			    v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME ||
9479 			    v == DIF_VAR_UID || v == DIF_VAR_GID)
9480 				break;
9481 
9482 			err += efunc(pc, "illegal variable %u\n", v);
9483 			break;
9484 
9485 		case DIF_OP_LDTA:
9486 		case DIF_OP_LDTS:
9487 		case DIF_OP_LDGAA:
9488 		case DIF_OP_LDTAA:
9489 			err += efunc(pc, "illegal dynamic variable load\n");
9490 			break;
9491 
9492 		case DIF_OP_STTS:
9493 		case DIF_OP_STGAA:
9494 		case DIF_OP_STTAA:
9495 			err += efunc(pc, "illegal dynamic variable store\n");
9496 			break;
9497 
9498 		case DIF_OP_CALL:
9499 			if (subr == DIF_SUBR_ALLOCA ||
9500 			    subr == DIF_SUBR_BCOPY ||
9501 			    subr == DIF_SUBR_COPYIN ||
9502 			    subr == DIF_SUBR_COPYINTO ||
9503 			    subr == DIF_SUBR_COPYINSTR ||
9504 			    subr == DIF_SUBR_INDEX ||
9505 			    subr == DIF_SUBR_INET_NTOA ||
9506 			    subr == DIF_SUBR_INET_NTOA6 ||
9507 			    subr == DIF_SUBR_INET_NTOP ||
9508 			    subr == DIF_SUBR_JSON ||
9509 			    subr == DIF_SUBR_LLTOSTR ||
9510 			    subr == DIF_SUBR_STRTOLL ||
9511 			    subr == DIF_SUBR_RINDEX ||
9512 			    subr == DIF_SUBR_STRCHR ||
9513 			    subr == DIF_SUBR_STRJOIN ||
9514 			    subr == DIF_SUBR_STRRCHR ||
9515 			    subr == DIF_SUBR_STRSTR ||
9516 			    subr == DIF_SUBR_HTONS ||
9517 			    subr == DIF_SUBR_HTONL ||
9518 			    subr == DIF_SUBR_HTONLL ||
9519 			    subr == DIF_SUBR_NTOHS ||
9520 			    subr == DIF_SUBR_NTOHL ||
9521 			    subr == DIF_SUBR_NTOHLL)
9522 				break;
9523 
9524 			err += efunc(pc, "invalid subr %u\n", subr);
9525 			break;
9526 
9527 		default:
9528 			err += efunc(pc, "invalid opcode %u\n",
9529 			    DIF_INSTR_OP(instr));
9530 		}
9531 	}
9532 
9533 	return (err);
9534 }
9535 
9536 /*
9537  * Returns 1 if the expression in the DIF object can be cached on a per-thread
9538  * basis; 0 if not.
9539  */
9540 static int
dtrace_difo_cacheable(dtrace_difo_t * dp)9541 dtrace_difo_cacheable(dtrace_difo_t *dp)
9542 {
9543 	int i;
9544 
9545 	if (dp == NULL)
9546 		return (0);
9547 
9548 	for (i = 0; i < dp->dtdo_varlen; i++) {
9549 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9550 
9551 		if (v->dtdv_scope != DIFV_SCOPE_GLOBAL)
9552 			continue;
9553 
9554 		switch (v->dtdv_id) {
9555 		case DIF_VAR_CURTHREAD:
9556 		case DIF_VAR_PID:
9557 		case DIF_VAR_TID:
9558 		case DIF_VAR_EXECNAME:
9559 		case DIF_VAR_ZONENAME:
9560 			break;
9561 
9562 		default:
9563 			return (0);
9564 		}
9565 	}
9566 
9567 	/*
9568 	 * This DIF object may be cacheable.  Now we need to look for any
9569 	 * array loading instructions, any memory loading instructions, or
9570 	 * any stores to thread-local variables.
9571 	 */
9572 	for (i = 0; i < dp->dtdo_len; i++) {
9573 		uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]);
9574 
9575 		if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) ||
9576 		    (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) ||
9577 		    (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) ||
9578 		    op == DIF_OP_LDGA || op == DIF_OP_STTS)
9579 			return (0);
9580 	}
9581 
9582 	return (1);
9583 }
9584 
9585 static void
dtrace_difo_hold(dtrace_difo_t * dp)9586 dtrace_difo_hold(dtrace_difo_t *dp)
9587 {
9588 	int i;
9589 
9590 	ASSERT(MUTEX_HELD(&dtrace_lock));
9591 
9592 	dp->dtdo_refcnt++;
9593 	ASSERT(dp->dtdo_refcnt != 0);
9594 
9595 	/*
9596 	 * We need to check this DIF object for references to the variable
9597 	 * DIF_VAR_VTIMESTAMP.
9598 	 */
9599 	for (i = 0; i < dp->dtdo_varlen; i++) {
9600 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9601 
9602 		if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
9603 			continue;
9604 
9605 		if (dtrace_vtime_references++ == 0)
9606 			dtrace_vtime_enable();
9607 	}
9608 }
9609 
9610 /*
9611  * This routine calculates the dynamic variable chunksize for a given DIF
9612  * object.  The calculation is not fool-proof, and can probably be tricked by
9613  * malicious DIF -- but it works for all compiler-generated DIF.  Because this
9614  * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail
9615  * if a dynamic variable size exceeds the chunksize.
9616  */
9617 static void
dtrace_difo_chunksize(dtrace_difo_t * dp,dtrace_vstate_t * vstate)9618 dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9619 {
9620 	uint64_t sval;
9621 	dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
9622 	const dif_instr_t *text = dp->dtdo_buf;
9623 	uint_t pc, srd = 0;
9624 	uint_t ttop = 0;
9625 	size_t size, ksize;
9626 	uint_t id, i;
9627 
9628 	for (pc = 0; pc < dp->dtdo_len; pc++) {
9629 		dif_instr_t instr = text[pc];
9630 		uint_t op = DIF_INSTR_OP(instr);
9631 		uint_t rd = DIF_INSTR_RD(instr);
9632 		uint_t r1 = DIF_INSTR_R1(instr);
9633 		uint_t nkeys = 0;
9634 		uchar_t scope;
9635 
9636 		dtrace_key_t *key = tupregs;
9637 
9638 		switch (op) {
9639 		case DIF_OP_SETX:
9640 			sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)];
9641 			srd = rd;
9642 			continue;
9643 
9644 		case DIF_OP_STTS:
9645 			key = &tupregs[DIF_DTR_NREGS];
9646 			key[0].dttk_size = 0;
9647 			key[1].dttk_size = 0;
9648 			nkeys = 2;
9649 			scope = DIFV_SCOPE_THREAD;
9650 			break;
9651 
9652 		case DIF_OP_STGAA:
9653 		case DIF_OP_STTAA:
9654 			nkeys = ttop;
9655 
9656 			if (DIF_INSTR_OP(instr) == DIF_OP_STTAA)
9657 				key[nkeys++].dttk_size = 0;
9658 
9659 			key[nkeys++].dttk_size = 0;
9660 
9661 			if (op == DIF_OP_STTAA) {
9662 				scope = DIFV_SCOPE_THREAD;
9663 			} else {
9664 				scope = DIFV_SCOPE_GLOBAL;
9665 			}
9666 
9667 			break;
9668 
9669 		case DIF_OP_PUSHTR:
9670 			if (ttop == DIF_DTR_NREGS)
9671 				return;
9672 
9673 			if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) {
9674 				/*
9675 				 * If the register for the size of the "pushtr"
9676 				 * is %r0 (or the value is 0) and the type is
9677 				 * a string, we'll use the system-wide default
9678 				 * string size.
9679 				 */
9680 				tupregs[ttop++].dttk_size =
9681 				    dtrace_strsize_default;
9682 			} else {
9683 				if (srd == 0)
9684 					return;
9685 
9686 				tupregs[ttop++].dttk_size = sval;
9687 			}
9688 
9689 			break;
9690 
9691 		case DIF_OP_PUSHTV:
9692 			if (ttop == DIF_DTR_NREGS)
9693 				return;
9694 
9695 			tupregs[ttop++].dttk_size = 0;
9696 			break;
9697 
9698 		case DIF_OP_FLUSHTS:
9699 			ttop = 0;
9700 			break;
9701 
9702 		case DIF_OP_POPTS:
9703 			if (ttop != 0)
9704 				ttop--;
9705 			break;
9706 		}
9707 
9708 		sval = 0;
9709 		srd = 0;
9710 
9711 		if (nkeys == 0)
9712 			continue;
9713 
9714 		/*
9715 		 * We have a dynamic variable allocation; calculate its size.
9716 		 */
9717 		for (ksize = 0, i = 0; i < nkeys; i++)
9718 			ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
9719 
9720 		size = sizeof (dtrace_dynvar_t);
9721 		size += sizeof (dtrace_key_t) * (nkeys - 1);
9722 		size += ksize;
9723 
9724 		/*
9725 		 * Now we need to determine the size of the stored data.
9726 		 */
9727 		id = DIF_INSTR_VAR(instr);
9728 
9729 		for (i = 0; i < dp->dtdo_varlen; i++) {
9730 			dtrace_difv_t *v = &dp->dtdo_vartab[i];
9731 
9732 			if (v->dtdv_id == id && v->dtdv_scope == scope) {
9733 				size += v->dtdv_type.dtdt_size;
9734 				break;
9735 			}
9736 		}
9737 
9738 		if (i == dp->dtdo_varlen)
9739 			return;
9740 
9741 		/*
9742 		 * We have the size.  If this is larger than the chunk size
9743 		 * for our dynamic variable state, reset the chunk size.
9744 		 */
9745 		size = P2ROUNDUP(size, sizeof (uint64_t));
9746 
9747 		if (size > vstate->dtvs_dynvars.dtds_chunksize)
9748 			vstate->dtvs_dynvars.dtds_chunksize = size;
9749 	}
9750 }
9751 
9752 static void
dtrace_difo_init(dtrace_difo_t * dp,dtrace_vstate_t * vstate)9753 dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9754 {
9755 	int i, oldsvars, osz, nsz, otlocals, ntlocals;
9756 	uint_t id;
9757 
9758 	ASSERT(MUTEX_HELD(&dtrace_lock));
9759 	ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0);
9760 
9761 	for (i = 0; i < dp->dtdo_varlen; i++) {
9762 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9763 		dtrace_statvar_t *svar, ***svarp;
9764 		size_t dsize = 0;
9765 		uint8_t scope = v->dtdv_scope;
9766 		int *np;
9767 
9768 		if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9769 			continue;
9770 
9771 		id -= DIF_VAR_OTHER_UBASE;
9772 
9773 		switch (scope) {
9774 		case DIFV_SCOPE_THREAD:
9775 			while (id >= (otlocals = vstate->dtvs_ntlocals)) {
9776 				dtrace_difv_t *tlocals;
9777 
9778 				if ((ntlocals = (otlocals << 1)) == 0)
9779 					ntlocals = 1;
9780 
9781 				osz = otlocals * sizeof (dtrace_difv_t);
9782 				nsz = ntlocals * sizeof (dtrace_difv_t);
9783 
9784 				tlocals = kmem_zalloc(nsz, KM_SLEEP);
9785 
9786 				if (osz != 0) {
9787 					bcopy(vstate->dtvs_tlocals,
9788 					    tlocals, osz);
9789 					kmem_free(vstate->dtvs_tlocals, osz);
9790 				}
9791 
9792 				vstate->dtvs_tlocals = tlocals;
9793 				vstate->dtvs_ntlocals = ntlocals;
9794 			}
9795 
9796 			vstate->dtvs_tlocals[id] = *v;
9797 			continue;
9798 
9799 		case DIFV_SCOPE_LOCAL:
9800 			np = &vstate->dtvs_nlocals;
9801 			svarp = &vstate->dtvs_locals;
9802 
9803 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9804 				dsize = NCPU * (v->dtdv_type.dtdt_size +
9805 				    sizeof (uint64_t));
9806 			else
9807 				dsize = NCPU * sizeof (uint64_t);
9808 
9809 			break;
9810 
9811 		case DIFV_SCOPE_GLOBAL:
9812 			np = &vstate->dtvs_nglobals;
9813 			svarp = &vstate->dtvs_globals;
9814 
9815 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9816 				dsize = v->dtdv_type.dtdt_size +
9817 				    sizeof (uint64_t);
9818 
9819 			break;
9820 
9821 		default:
9822 			ASSERT(0);
9823 		}
9824 
9825 		while (id >= (oldsvars = *np)) {
9826 			dtrace_statvar_t **statics;
9827 			int newsvars, oldsize, newsize;
9828 
9829 			if ((newsvars = (oldsvars << 1)) == 0)
9830 				newsvars = 1;
9831 
9832 			oldsize = oldsvars * sizeof (dtrace_statvar_t *);
9833 			newsize = newsvars * sizeof (dtrace_statvar_t *);
9834 
9835 			statics = kmem_zalloc(newsize, KM_SLEEP);
9836 
9837 			if (oldsize != 0) {
9838 				bcopy(*svarp, statics, oldsize);
9839 				kmem_free(*svarp, oldsize);
9840 			}
9841 
9842 			*svarp = statics;
9843 			*np = newsvars;
9844 		}
9845 
9846 		if ((svar = (*svarp)[id]) == NULL) {
9847 			svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP);
9848 			svar->dtsv_var = *v;
9849 
9850 			if ((svar->dtsv_size = dsize) != 0) {
9851 				svar->dtsv_data = (uint64_t)(uintptr_t)
9852 				    kmem_zalloc(dsize, KM_SLEEP);
9853 			}
9854 
9855 			(*svarp)[id] = svar;
9856 		}
9857 
9858 		svar->dtsv_refcnt++;
9859 	}
9860 
9861 	dtrace_difo_chunksize(dp, vstate);
9862 	dtrace_difo_hold(dp);
9863 }
9864 
9865 static dtrace_difo_t *
dtrace_difo_duplicate(dtrace_difo_t * dp,dtrace_vstate_t * vstate)9866 dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9867 {
9868 	dtrace_difo_t *new;
9869 	size_t sz;
9870 
9871 	ASSERT(dp->dtdo_buf != NULL);
9872 	ASSERT(dp->dtdo_refcnt != 0);
9873 
9874 	new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
9875 
9876 	ASSERT(dp->dtdo_buf != NULL);
9877 	sz = dp->dtdo_len * sizeof (dif_instr_t);
9878 	new->dtdo_buf = kmem_alloc(sz, KM_SLEEP);
9879 	bcopy(dp->dtdo_buf, new->dtdo_buf, sz);
9880 	new->dtdo_len = dp->dtdo_len;
9881 
9882 	if (dp->dtdo_strtab != NULL) {
9883 		ASSERT(dp->dtdo_strlen != 0);
9884 		new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP);
9885 		bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen);
9886 		new->dtdo_strlen = dp->dtdo_strlen;
9887 	}
9888 
9889 	if (dp->dtdo_inttab != NULL) {
9890 		ASSERT(dp->dtdo_intlen != 0);
9891 		sz = dp->dtdo_intlen * sizeof (uint64_t);
9892 		new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP);
9893 		bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz);
9894 		new->dtdo_intlen = dp->dtdo_intlen;
9895 	}
9896 
9897 	if (dp->dtdo_vartab != NULL) {
9898 		ASSERT(dp->dtdo_varlen != 0);
9899 		sz = dp->dtdo_varlen * sizeof (dtrace_difv_t);
9900 		new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP);
9901 		bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz);
9902 		new->dtdo_varlen = dp->dtdo_varlen;
9903 	}
9904 
9905 	dtrace_difo_init(new, vstate);
9906 	return (new);
9907 }
9908 
9909 static void
dtrace_difo_destroy(dtrace_difo_t * dp,dtrace_vstate_t * vstate)9910 dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9911 {
9912 	int i;
9913 
9914 	ASSERT(dp->dtdo_refcnt == 0);
9915 
9916 	for (i = 0; i < dp->dtdo_varlen; i++) {
9917 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9918 		dtrace_statvar_t *svar, **svarp;
9919 		uint_t id;
9920 		uint8_t scope = v->dtdv_scope;
9921 		int *np;
9922 
9923 		switch (scope) {
9924 		case DIFV_SCOPE_THREAD:
9925 			continue;
9926 
9927 		case DIFV_SCOPE_LOCAL:
9928 			np = &vstate->dtvs_nlocals;
9929 			svarp = vstate->dtvs_locals;
9930 			break;
9931 
9932 		case DIFV_SCOPE_GLOBAL:
9933 			np = &vstate->dtvs_nglobals;
9934 			svarp = vstate->dtvs_globals;
9935 			break;
9936 
9937 		default:
9938 			ASSERT(0);
9939 		}
9940 
9941 		if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9942 			continue;
9943 
9944 		id -= DIF_VAR_OTHER_UBASE;
9945 		ASSERT(id < *np);
9946 
9947 		svar = svarp[id];
9948 		ASSERT(svar != NULL);
9949 		ASSERT(svar->dtsv_refcnt > 0);
9950 
9951 		if (--svar->dtsv_refcnt > 0)
9952 			continue;
9953 
9954 		if (svar->dtsv_size != 0) {
9955 			ASSERT(svar->dtsv_data != NULL);
9956 			kmem_free((void *)(uintptr_t)svar->dtsv_data,
9957 			    svar->dtsv_size);
9958 		}
9959 
9960 		kmem_free(svar, sizeof (dtrace_statvar_t));
9961 		svarp[id] = NULL;
9962 	}
9963 
9964 	kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
9965 	kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
9966 	kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
9967 	kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
9968 
9969 	kmem_free(dp, sizeof (dtrace_difo_t));
9970 }
9971 
9972 static void
dtrace_difo_release(dtrace_difo_t * dp,dtrace_vstate_t * vstate)9973 dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9974 {
9975 	int i;
9976 
9977 	ASSERT(MUTEX_HELD(&dtrace_lock));
9978 	ASSERT(dp->dtdo_refcnt != 0);
9979 
9980 	for (i = 0; i < dp->dtdo_varlen; i++) {
9981 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9982 
9983 		if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
9984 			continue;
9985 
9986 		ASSERT(dtrace_vtime_references > 0);
9987 		if (--dtrace_vtime_references == 0)
9988 			dtrace_vtime_disable();
9989 	}
9990 
9991 	if (--dp->dtdo_refcnt == 0)
9992 		dtrace_difo_destroy(dp, vstate);
9993 }
9994 
9995 /*
9996  * DTrace Format Functions
9997  */
9998 static uint16_t
dtrace_format_add(dtrace_state_t * state,char * str)9999 dtrace_format_add(dtrace_state_t *state, char *str)
10000 {
10001 	char *fmt, **new;
10002 	uint16_t ndx, len = strlen(str) + 1;
10003 
10004 	fmt = kmem_zalloc(len, KM_SLEEP);
10005 	bcopy(str, fmt, len);
10006 
10007 	for (ndx = 0; ndx < state->dts_nformats; ndx++) {
10008 		if (state->dts_formats[ndx] == NULL) {
10009 			state->dts_formats[ndx] = fmt;
10010 			return (ndx + 1);
10011 		}
10012 	}
10013 
10014 	if (state->dts_nformats == USHRT_MAX) {
10015 		/*
10016 		 * This is only likely if a denial-of-service attack is being
10017 		 * attempted.  As such, it's okay to fail silently here.
10018 		 */
10019 		kmem_free(fmt, len);
10020 		return (0);
10021 	}
10022 
10023 	/*
10024 	 * For simplicity, we always resize the formats array to be exactly the
10025 	 * number of formats.
10026 	 */
10027 	ndx = state->dts_nformats++;
10028 	new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP);
10029 
10030 	if (state->dts_formats != NULL) {
10031 		ASSERT(ndx != 0);
10032 		bcopy(state->dts_formats, new, ndx * sizeof (char *));
10033 		kmem_free(state->dts_formats, ndx * sizeof (char *));
10034 	}
10035 
10036 	state->dts_formats = new;
10037 	state->dts_formats[ndx] = fmt;
10038 
10039 	return (ndx + 1);
10040 }
10041 
10042 static void
dtrace_format_remove(dtrace_state_t * state,uint16_t format)10043 dtrace_format_remove(dtrace_state_t *state, uint16_t format)
10044 {
10045 	char *fmt;
10046 
10047 	ASSERT(state->dts_formats != NULL);
10048 	ASSERT(format <= state->dts_nformats);
10049 	ASSERT(state->dts_formats[format - 1] != NULL);
10050 
10051 	fmt = state->dts_formats[format - 1];
10052 	kmem_free(fmt, strlen(fmt) + 1);
10053 	state->dts_formats[format - 1] = NULL;
10054 }
10055 
10056 static void
dtrace_format_destroy(dtrace_state_t * state)10057 dtrace_format_destroy(dtrace_state_t *state)
10058 {
10059 	int i;
10060 
10061 	if (state->dts_nformats == 0) {
10062 		ASSERT(state->dts_formats == NULL);
10063 		return;
10064 	}
10065 
10066 	ASSERT(state->dts_formats != NULL);
10067 
10068 	for (i = 0; i < state->dts_nformats; i++) {
10069 		char *fmt = state->dts_formats[i];
10070 
10071 		if (fmt == NULL)
10072 			continue;
10073 
10074 		kmem_free(fmt, strlen(fmt) + 1);
10075 	}
10076 
10077 	kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *));
10078 	state->dts_nformats = 0;
10079 	state->dts_formats = NULL;
10080 }
10081 
10082 /*
10083  * DTrace Predicate Functions
10084  */
10085 static dtrace_predicate_t *
dtrace_predicate_create(dtrace_difo_t * dp)10086 dtrace_predicate_create(dtrace_difo_t *dp)
10087 {
10088 	dtrace_predicate_t *pred;
10089 
10090 	ASSERT(MUTEX_HELD(&dtrace_lock));
10091 	ASSERT(dp->dtdo_refcnt != 0);
10092 
10093 	pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP);
10094 	pred->dtp_difo = dp;
10095 	pred->dtp_refcnt = 1;
10096 
10097 	if (!dtrace_difo_cacheable(dp))
10098 		return (pred);
10099 
10100 	if (dtrace_predcache_id == DTRACE_CACHEIDNONE) {
10101 		/*
10102 		 * This is only theoretically possible -- we have had 2^32
10103 		 * cacheable predicates on this machine.  We cannot allow any
10104 		 * more predicates to become cacheable:  as unlikely as it is,
10105 		 * there may be a thread caching a (now stale) predicate cache
10106 		 * ID. (N.B.: the temptation is being successfully resisted to
10107 		 * have this cmn_err() "Holy shit -- we executed this code!")
10108 		 */
10109 		return (pred);
10110 	}
10111 
10112 	pred->dtp_cacheid = dtrace_predcache_id++;
10113 
10114 	return (pred);
10115 }
10116 
10117 static void
dtrace_predicate_hold(dtrace_predicate_t * pred)10118 dtrace_predicate_hold(dtrace_predicate_t *pred)
10119 {
10120 	ASSERT(MUTEX_HELD(&dtrace_lock));
10121 	ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0);
10122 	ASSERT(pred->dtp_refcnt > 0);
10123 
10124 	pred->dtp_refcnt++;
10125 }
10126 
10127 static void
dtrace_predicate_release(dtrace_predicate_t * pred,dtrace_vstate_t * vstate)10128 dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate)
10129 {
10130 	dtrace_difo_t *dp = pred->dtp_difo;
10131 
10132 	ASSERT(MUTEX_HELD(&dtrace_lock));
10133 	ASSERT(dp != NULL && dp->dtdo_refcnt != 0);
10134 	ASSERT(pred->dtp_refcnt > 0);
10135 
10136 	if (--pred->dtp_refcnt == 0) {
10137 		dtrace_difo_release(pred->dtp_difo, vstate);
10138 		kmem_free(pred, sizeof (dtrace_predicate_t));
10139 	}
10140 }
10141 
10142 /*
10143  * DTrace Action Description Functions
10144  */
10145 static dtrace_actdesc_t *
dtrace_actdesc_create(dtrace_actkind_t kind,uint32_t ntuple,uint64_t uarg,uint64_t arg)10146 dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple,
10147     uint64_t uarg, uint64_t arg)
10148 {
10149 	dtrace_actdesc_t *act;
10150 
10151 	ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL &&
10152 	    arg >= KERNELBASE) || (arg == NULL && kind == DTRACEACT_PRINTA));
10153 
10154 	act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP);
10155 	act->dtad_kind = kind;
10156 	act->dtad_ntuple = ntuple;
10157 	act->dtad_uarg = uarg;
10158 	act->dtad_arg = arg;
10159 	act->dtad_refcnt = 1;
10160 
10161 	return (act);
10162 }
10163 
10164 static void
dtrace_actdesc_hold(dtrace_actdesc_t * act)10165 dtrace_actdesc_hold(dtrace_actdesc_t *act)
10166 {
10167 	ASSERT(act->dtad_refcnt >= 1);
10168 	act->dtad_refcnt++;
10169 }
10170 
10171 static void
dtrace_actdesc_release(dtrace_actdesc_t * act,dtrace_vstate_t * vstate)10172 dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate)
10173 {
10174 	dtrace_actkind_t kind = act->dtad_kind;
10175 	dtrace_difo_t *dp;
10176 
10177 	ASSERT(act->dtad_refcnt >= 1);
10178 
10179 	if (--act->dtad_refcnt != 0)
10180 		return;
10181 
10182 	if ((dp = act->dtad_difo) != NULL)
10183 		dtrace_difo_release(dp, vstate);
10184 
10185 	if (DTRACEACT_ISPRINTFLIKE(kind)) {
10186 		char *str = (char *)(uintptr_t)act->dtad_arg;
10187 
10188 		ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) ||
10189 		    (str == NULL && act->dtad_kind == DTRACEACT_PRINTA));
10190 
10191 		if (str != NULL)
10192 			kmem_free(str, strlen(str) + 1);
10193 	}
10194 
10195 	kmem_free(act, sizeof (dtrace_actdesc_t));
10196 }
10197 
10198 /*
10199  * DTrace ECB Functions
10200  */
10201 static dtrace_ecb_t *
dtrace_ecb_add(dtrace_state_t * state,dtrace_probe_t * probe)10202 dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe)
10203 {
10204 	dtrace_ecb_t *ecb;
10205 	dtrace_epid_t epid;
10206 
10207 	ASSERT(MUTEX_HELD(&dtrace_lock));
10208 
10209 	ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP);
10210 	ecb->dte_predicate = NULL;
10211 	ecb->dte_probe = probe;
10212 
10213 	/*
10214 	 * The default size is the size of the default action: recording
10215 	 * the header.
10216 	 */
10217 	ecb->dte_size = ecb->dte_needed = sizeof (dtrace_rechdr_t);
10218 	ecb->dte_alignment = sizeof (dtrace_epid_t);
10219 
10220 	epid = state->dts_epid++;
10221 
10222 	if (epid - 1 >= state->dts_necbs) {
10223 		dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs;
10224 		int necbs = state->dts_necbs << 1;
10225 
10226 		ASSERT(epid == state->dts_necbs + 1);
10227 
10228 		if (necbs == 0) {
10229 			ASSERT(oecbs == NULL);
10230 			necbs = 1;
10231 		}
10232 
10233 		ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP);
10234 
10235 		if (oecbs != NULL)
10236 			bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs));
10237 
10238 		dtrace_membar_producer();
10239 		state->dts_ecbs = ecbs;
10240 
10241 		if (oecbs != NULL) {
10242 			/*
10243 			 * If this state is active, we must dtrace_sync()
10244 			 * before we can free the old dts_ecbs array:  we're
10245 			 * coming in hot, and there may be active ring
10246 			 * buffer processing (which indexes into the dts_ecbs
10247 			 * array) on another CPU.
10248 			 */
10249 			if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
10250 				dtrace_sync();
10251 
10252 			kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs));
10253 		}
10254 
10255 		dtrace_membar_producer();
10256 		state->dts_necbs = necbs;
10257 	}
10258 
10259 	ecb->dte_state = state;
10260 
10261 	ASSERT(state->dts_ecbs[epid - 1] == NULL);
10262 	dtrace_membar_producer();
10263 	state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb;
10264 
10265 	return (ecb);
10266 }
10267 
10268 static int
dtrace_ecb_enable(dtrace_ecb_t * ecb)10269 dtrace_ecb_enable(dtrace_ecb_t *ecb)
10270 {
10271 	dtrace_probe_t *probe = ecb->dte_probe;
10272 
10273 	ASSERT(MUTEX_HELD(&cpu_lock));
10274 	ASSERT(MUTEX_HELD(&dtrace_lock));
10275 	ASSERT(ecb->dte_next == NULL);
10276 
10277 	if (probe == NULL) {
10278 		/*
10279 		 * This is the NULL probe -- there's nothing to do.
10280 		 */
10281 		return (0);
10282 	}
10283 
10284 	if (probe->dtpr_ecb == NULL) {
10285 		dtrace_provider_t *prov = probe->dtpr_provider;
10286 
10287 		/*
10288 		 * We're the first ECB on this probe.
10289 		 */
10290 		probe->dtpr_ecb = probe->dtpr_ecb_last = ecb;
10291 
10292 		if (ecb->dte_predicate != NULL)
10293 			probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid;
10294 
10295 		return (prov->dtpv_pops.dtps_enable(prov->dtpv_arg,
10296 		    probe->dtpr_id, probe->dtpr_arg));
10297 	} else {
10298 		/*
10299 		 * This probe is already active.  Swing the last pointer to
10300 		 * point to the new ECB, and issue a dtrace_sync() to assure
10301 		 * that all CPUs have seen the change.
10302 		 */
10303 		ASSERT(probe->dtpr_ecb_last != NULL);
10304 		probe->dtpr_ecb_last->dte_next = ecb;
10305 		probe->dtpr_ecb_last = ecb;
10306 		probe->dtpr_predcache = 0;
10307 
10308 		dtrace_sync();
10309 		return (0);
10310 	}
10311 }
10312 
10313 static void
dtrace_ecb_resize(dtrace_ecb_t * ecb)10314 dtrace_ecb_resize(dtrace_ecb_t *ecb)
10315 {
10316 	dtrace_action_t *act;
10317 	uint32_t curneeded = UINT32_MAX;
10318 	uint32_t aggbase = UINT32_MAX;
10319 
10320 	/*
10321 	 * If we record anything, we always record the dtrace_rechdr_t.  (And
10322 	 * we always record it first.)
10323 	 */
10324 	ecb->dte_size = sizeof (dtrace_rechdr_t);
10325 	ecb->dte_alignment = sizeof (dtrace_epid_t);
10326 
10327 	for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
10328 		dtrace_recdesc_t *rec = &act->dta_rec;
10329 		ASSERT(rec->dtrd_size > 0 || rec->dtrd_alignment == 1);
10330 
10331 		ecb->dte_alignment = MAX(ecb->dte_alignment,
10332 		    rec->dtrd_alignment);
10333 
10334 		if (DTRACEACT_ISAGG(act->dta_kind)) {
10335 			dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
10336 
10337 			ASSERT(rec->dtrd_size != 0);
10338 			ASSERT(agg->dtag_first != NULL);
10339 			ASSERT(act->dta_prev->dta_intuple);
10340 			ASSERT(aggbase != UINT32_MAX);
10341 			ASSERT(curneeded != UINT32_MAX);
10342 
10343 			agg->dtag_base = aggbase;
10344 
10345 			curneeded = P2ROUNDUP(curneeded, rec->dtrd_alignment);
10346 			rec->dtrd_offset = curneeded;
10347 			curneeded += rec->dtrd_size;
10348 			ecb->dte_needed = MAX(ecb->dte_needed, curneeded);
10349 
10350 			aggbase = UINT32_MAX;
10351 			curneeded = UINT32_MAX;
10352 		} else if (act->dta_intuple) {
10353 			if (curneeded == UINT32_MAX) {
10354 				/*
10355 				 * This is the first record in a tuple.  Align
10356 				 * curneeded to be at offset 4 in an 8-byte
10357 				 * aligned block.
10358 				 */
10359 				ASSERT(act->dta_prev == NULL ||
10360 				    !act->dta_prev->dta_intuple);
10361 				ASSERT3U(aggbase, ==, UINT32_MAX);
10362 				curneeded = P2PHASEUP(ecb->dte_size,
10363 				    sizeof (uint64_t), sizeof (dtrace_aggid_t));
10364 
10365 				aggbase = curneeded - sizeof (dtrace_aggid_t);
10366 				ASSERT(IS_P2ALIGNED(aggbase,
10367 				    sizeof (uint64_t)));
10368 			}
10369 			curneeded = P2ROUNDUP(curneeded, rec->dtrd_alignment);
10370 			rec->dtrd_offset = curneeded;
10371 			curneeded += rec->dtrd_size;
10372 		} else {
10373 			/* tuples must be followed by an aggregation */
10374 			ASSERT(act->dta_prev == NULL ||
10375 			    !act->dta_prev->dta_intuple);
10376 
10377 			ecb->dte_size = P2ROUNDUP(ecb->dte_size,
10378 			    rec->dtrd_alignment);
10379 			rec->dtrd_offset = ecb->dte_size;
10380 			ecb->dte_size += rec->dtrd_size;
10381 			ecb->dte_needed = MAX(ecb->dte_needed, ecb->dte_size);
10382 		}
10383 	}
10384 
10385 	if ((act = ecb->dte_action) != NULL &&
10386 	    !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) &&
10387 	    ecb->dte_size == sizeof (dtrace_rechdr_t)) {
10388 		/*
10389 		 * If the size is still sizeof (dtrace_rechdr_t), then all
10390 		 * actions store no data; set the size to 0.
10391 		 */
10392 		ecb->dte_size = 0;
10393 	}
10394 
10395 	ecb->dte_size = P2ROUNDUP(ecb->dte_size, sizeof (dtrace_epid_t));
10396 	ecb->dte_needed = P2ROUNDUP(ecb->dte_needed, (sizeof (dtrace_epid_t)));
10397 	ecb->dte_state->dts_needed = MAX(ecb->dte_state->dts_needed,
10398 	    ecb->dte_needed);
10399 }
10400 
10401 static dtrace_action_t *
dtrace_ecb_aggregation_create(dtrace_ecb_t * ecb,dtrace_actdesc_t * desc)10402 dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
10403 {
10404 	dtrace_aggregation_t *agg;
10405 	size_t size = sizeof (uint64_t);
10406 	int ntuple = desc->dtad_ntuple;
10407 	dtrace_action_t *act;
10408 	dtrace_recdesc_t *frec;
10409 	dtrace_aggid_t aggid;
10410 	dtrace_state_t *state = ecb->dte_state;
10411 
10412 	agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP);
10413 	agg->dtag_ecb = ecb;
10414 
10415 	ASSERT(DTRACEACT_ISAGG(desc->dtad_kind));
10416 
10417 	switch (desc->dtad_kind) {
10418 	case DTRACEAGG_MIN:
10419 		agg->dtag_initial = INT64_MAX;
10420 		agg->dtag_aggregate = dtrace_aggregate_min;
10421 		break;
10422 
10423 	case DTRACEAGG_MAX:
10424 		agg->dtag_initial = INT64_MIN;
10425 		agg->dtag_aggregate = dtrace_aggregate_max;
10426 		break;
10427 
10428 	case DTRACEAGG_COUNT:
10429 		agg->dtag_aggregate = dtrace_aggregate_count;
10430 		break;
10431 
10432 	case DTRACEAGG_QUANTIZE:
10433 		agg->dtag_aggregate = dtrace_aggregate_quantize;
10434 		size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) *
10435 		    sizeof (uint64_t);
10436 		break;
10437 
10438 	case DTRACEAGG_LQUANTIZE: {
10439 		uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg);
10440 		uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg);
10441 
10442 		agg->dtag_initial = desc->dtad_arg;
10443 		agg->dtag_aggregate = dtrace_aggregate_lquantize;
10444 
10445 		if (step == 0 || levels == 0)
10446 			goto err;
10447 
10448 		size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t);
10449 		break;
10450 	}
10451 
10452 	case DTRACEAGG_LLQUANTIZE: {
10453 		uint16_t factor = DTRACE_LLQUANTIZE_FACTOR(desc->dtad_arg);
10454 		uint16_t low = DTRACE_LLQUANTIZE_LOW(desc->dtad_arg);
10455 		uint16_t high = DTRACE_LLQUANTIZE_HIGH(desc->dtad_arg);
10456 		uint16_t nsteps = DTRACE_LLQUANTIZE_NSTEP(desc->dtad_arg);
10457 		int64_t v;
10458 
10459 		agg->dtag_initial = desc->dtad_arg;
10460 		agg->dtag_aggregate = dtrace_aggregate_llquantize;
10461 
10462 		if (factor < 2 || low >= high || nsteps < factor)
10463 			goto err;
10464 
10465 		/*
10466 		 * Now check that the number of steps evenly divides a power
10467 		 * of the factor.  (This assures both integer bucket size and
10468 		 * linearity within each magnitude.)
10469 		 */
10470 		for (v = factor; v < nsteps; v *= factor)
10471 			continue;
10472 
10473 		if ((v % nsteps) || (nsteps % factor))
10474 			goto err;
10475 
10476 		size = (dtrace_aggregate_llquantize_bucket(factor,
10477 		    low, high, nsteps, INT64_MAX) + 2) * sizeof (uint64_t);
10478 		break;
10479 	}
10480 
10481 	case DTRACEAGG_AVG:
10482 		agg->dtag_aggregate = dtrace_aggregate_avg;
10483 		size = sizeof (uint64_t) * 2;
10484 		break;
10485 
10486 	case DTRACEAGG_STDDEV:
10487 		agg->dtag_aggregate = dtrace_aggregate_stddev;
10488 		size = sizeof (uint64_t) * 4;
10489 		break;
10490 
10491 	case DTRACEAGG_SUM:
10492 		agg->dtag_aggregate = dtrace_aggregate_sum;
10493 		break;
10494 
10495 	default:
10496 		goto err;
10497 	}
10498 
10499 	agg->dtag_action.dta_rec.dtrd_size = size;
10500 
10501 	if (ntuple == 0)
10502 		goto err;
10503 
10504 	/*
10505 	 * We must make sure that we have enough actions for the n-tuple.
10506 	 */
10507 	for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) {
10508 		if (DTRACEACT_ISAGG(act->dta_kind))
10509 			break;
10510 
10511 		if (--ntuple == 0) {
10512 			/*
10513 			 * This is the action with which our n-tuple begins.
10514 			 */
10515 			agg->dtag_first = act;
10516 			goto success;
10517 		}
10518 	}
10519 
10520 	/*
10521 	 * This n-tuple is short by ntuple elements.  Return failure.
10522 	 */
10523 	ASSERT(ntuple != 0);
10524 err:
10525 	kmem_free(agg, sizeof (dtrace_aggregation_t));
10526 	return (NULL);
10527 
10528 success:
10529 	/*
10530 	 * If the last action in the tuple has a size of zero, it's actually
10531 	 * an expression argument for the aggregating action.
10532 	 */
10533 	ASSERT(ecb->dte_action_last != NULL);
10534 	act = ecb->dte_action_last;
10535 
10536 	if (act->dta_kind == DTRACEACT_DIFEXPR) {
10537 		ASSERT(act->dta_difo != NULL);
10538 
10539 		if (act->dta_difo->dtdo_rtype.dtdt_size == 0)
10540 			agg->dtag_hasarg = 1;
10541 	}
10542 
10543 	/*
10544 	 * We need to allocate an id for this aggregation.
10545 	 */
10546 	aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1,
10547 	    VM_BESTFIT | VM_SLEEP);
10548 
10549 	if (aggid - 1 >= state->dts_naggregations) {
10550 		dtrace_aggregation_t **oaggs = state->dts_aggregations;
10551 		dtrace_aggregation_t **aggs;
10552 		int naggs = state->dts_naggregations << 1;
10553 		int onaggs = state->dts_naggregations;
10554 
10555 		ASSERT(aggid == state->dts_naggregations + 1);
10556 
10557 		if (naggs == 0) {
10558 			ASSERT(oaggs == NULL);
10559 			naggs = 1;
10560 		}
10561 
10562 		aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP);
10563 
10564 		if (oaggs != NULL) {
10565 			bcopy(oaggs, aggs, onaggs * sizeof (*aggs));
10566 			kmem_free(oaggs, onaggs * sizeof (*aggs));
10567 		}
10568 
10569 		state->dts_aggregations = aggs;
10570 		state->dts_naggregations = naggs;
10571 	}
10572 
10573 	ASSERT(state->dts_aggregations[aggid - 1] == NULL);
10574 	state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg;
10575 
10576 	frec = &agg->dtag_first->dta_rec;
10577 	if (frec->dtrd_alignment < sizeof (dtrace_aggid_t))
10578 		frec->dtrd_alignment = sizeof (dtrace_aggid_t);
10579 
10580 	for (act = agg->dtag_first; act != NULL; act = act->dta_next) {
10581 		ASSERT(!act->dta_intuple);
10582 		act->dta_intuple = 1;
10583 	}
10584 
10585 	return (&agg->dtag_action);
10586 }
10587 
10588 static void
dtrace_ecb_aggregation_destroy(dtrace_ecb_t * ecb,dtrace_action_t * act)10589 dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act)
10590 {
10591 	dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
10592 	dtrace_state_t *state = ecb->dte_state;
10593 	dtrace_aggid_t aggid = agg->dtag_id;
10594 
10595 	ASSERT(DTRACEACT_ISAGG(act->dta_kind));
10596 	vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1);
10597 
10598 	ASSERT(state->dts_aggregations[aggid - 1] == agg);
10599 	state->dts_aggregations[aggid - 1] = NULL;
10600 
10601 	kmem_free(agg, sizeof (dtrace_aggregation_t));
10602 }
10603 
10604 static int
dtrace_ecb_action_add(dtrace_ecb_t * ecb,dtrace_actdesc_t * desc)10605 dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
10606 {
10607 	dtrace_action_t *action, *last;
10608 	dtrace_difo_t *dp = desc->dtad_difo;
10609 	uint32_t size = 0, align = sizeof (uint8_t), mask;
10610 	uint16_t format = 0;
10611 	dtrace_recdesc_t *rec;
10612 	dtrace_state_t *state = ecb->dte_state;
10613 	dtrace_optval_t *opt = state->dts_options, nframes, strsize;
10614 	uint64_t arg = desc->dtad_arg;
10615 
10616 	ASSERT(MUTEX_HELD(&dtrace_lock));
10617 	ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1);
10618 
10619 	if (DTRACEACT_ISAGG(desc->dtad_kind)) {
10620 		/*
10621 		 * If this is an aggregating action, there must be neither
10622 		 * a speculate nor a commit on the action chain.
10623 		 */
10624 		dtrace_action_t *act;
10625 
10626 		for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
10627 			if (act->dta_kind == DTRACEACT_COMMIT)
10628 				return (EINVAL);
10629 
10630 			if (act->dta_kind == DTRACEACT_SPECULATE)
10631 				return (EINVAL);
10632 		}
10633 
10634 		action = dtrace_ecb_aggregation_create(ecb, desc);
10635 
10636 		if (action == NULL)
10637 			return (EINVAL);
10638 	} else {
10639 		if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) ||
10640 		    (desc->dtad_kind == DTRACEACT_DIFEXPR &&
10641 		    dp != NULL && dp->dtdo_destructive)) {
10642 			state->dts_destructive = 1;
10643 		}
10644 
10645 		switch (desc->dtad_kind) {
10646 		case DTRACEACT_PRINTF:
10647 		case DTRACEACT_PRINTA:
10648 		case DTRACEACT_SYSTEM:
10649 		case DTRACEACT_FREOPEN:
10650 		case DTRACEACT_DIFEXPR:
10651 			/*
10652 			 * We know that our arg is a string -- turn it into a
10653 			 * format.
10654 			 */
10655 			if (arg == NULL) {
10656 				ASSERT(desc->dtad_kind == DTRACEACT_PRINTA ||
10657 				    desc->dtad_kind == DTRACEACT_DIFEXPR);
10658 				format = 0;
10659 			} else {
10660 				ASSERT(arg != NULL);
10661 				ASSERT(arg > KERNELBASE);
10662 				format = dtrace_format_add(state,
10663 				    (char *)(uintptr_t)arg);
10664 			}
10665 
10666 			/*FALLTHROUGH*/
10667 		case DTRACEACT_LIBACT:
10668 		case DTRACEACT_TRACEMEM:
10669 		case DTRACEACT_TRACEMEM_DYNSIZE:
10670 			if (dp == NULL)
10671 				return (EINVAL);
10672 
10673 			if ((size = dp->dtdo_rtype.dtdt_size) != 0)
10674 				break;
10675 
10676 			if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) {
10677 				if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10678 					return (EINVAL);
10679 
10680 				size = opt[DTRACEOPT_STRSIZE];
10681 			}
10682 
10683 			break;
10684 
10685 		case DTRACEACT_STACK:
10686 			if ((nframes = arg) == 0) {
10687 				nframes = opt[DTRACEOPT_STACKFRAMES];
10688 				ASSERT(nframes > 0);
10689 				arg = nframes;
10690 			}
10691 
10692 			size = nframes * sizeof (pc_t);
10693 			break;
10694 
10695 		case DTRACEACT_JSTACK:
10696 			if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0)
10697 				strsize = opt[DTRACEOPT_JSTACKSTRSIZE];
10698 
10699 			if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0)
10700 				nframes = opt[DTRACEOPT_JSTACKFRAMES];
10701 
10702 			arg = DTRACE_USTACK_ARG(nframes, strsize);
10703 
10704 			/*FALLTHROUGH*/
10705 		case DTRACEACT_USTACK:
10706 			if (desc->dtad_kind != DTRACEACT_JSTACK &&
10707 			    (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) {
10708 				strsize = DTRACE_USTACK_STRSIZE(arg);
10709 				nframes = opt[DTRACEOPT_USTACKFRAMES];
10710 				ASSERT(nframes > 0);
10711 				arg = DTRACE_USTACK_ARG(nframes, strsize);
10712 			}
10713 
10714 			/*
10715 			 * Save a slot for the pid.
10716 			 */
10717 			size = (nframes + 1) * sizeof (uint64_t);
10718 			size += DTRACE_USTACK_STRSIZE(arg);
10719 			size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t)));
10720 
10721 			break;
10722 
10723 		case DTRACEACT_SYM:
10724 		case DTRACEACT_MOD:
10725 			if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) !=
10726 			    sizeof (uint64_t)) ||
10727 			    (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10728 				return (EINVAL);
10729 			break;
10730 
10731 		case DTRACEACT_USYM:
10732 		case DTRACEACT_UMOD:
10733 		case DTRACEACT_UADDR:
10734 			if (dp == NULL ||
10735 			    (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) ||
10736 			    (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10737 				return (EINVAL);
10738 
10739 			/*
10740 			 * We have a slot for the pid, plus a slot for the
10741 			 * argument.  To keep things simple (aligned with
10742 			 * bitness-neutral sizing), we store each as a 64-bit
10743 			 * quantity.
10744 			 */
10745 			size = 2 * sizeof (uint64_t);
10746 			break;
10747 
10748 		case DTRACEACT_STOP:
10749 		case DTRACEACT_BREAKPOINT:
10750 		case DTRACEACT_PANIC:
10751 			break;
10752 
10753 		case DTRACEACT_CHILL:
10754 		case DTRACEACT_DISCARD:
10755 		case DTRACEACT_RAISE:
10756 			if (dp == NULL)
10757 				return (EINVAL);
10758 			break;
10759 
10760 		case DTRACEACT_EXIT:
10761 			if (dp == NULL ||
10762 			    (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) ||
10763 			    (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10764 				return (EINVAL);
10765 			break;
10766 
10767 		case DTRACEACT_SPECULATE:
10768 			if (ecb->dte_size > sizeof (dtrace_rechdr_t))
10769 				return (EINVAL);
10770 
10771 			if (dp == NULL)
10772 				return (EINVAL);
10773 
10774 			state->dts_speculates = 1;
10775 			break;
10776 
10777 		case DTRACEACT_COMMIT: {
10778 			dtrace_action_t *act = ecb->dte_action;
10779 
10780 			for (; act != NULL; act = act->dta_next) {
10781 				if (act->dta_kind == DTRACEACT_COMMIT)
10782 					return (EINVAL);
10783 			}
10784 
10785 			if (dp == NULL)
10786 				return (EINVAL);
10787 			break;
10788 		}
10789 
10790 		default:
10791 			return (EINVAL);
10792 		}
10793 
10794 		if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) {
10795 			/*
10796 			 * If this is a data-storing action or a speculate,
10797 			 * we must be sure that there isn't a commit on the
10798 			 * action chain.
10799 			 */
10800 			dtrace_action_t *act = ecb->dte_action;
10801 
10802 			for (; act != NULL; act = act->dta_next) {
10803 				if (act->dta_kind == DTRACEACT_COMMIT)
10804 					return (EINVAL);
10805 			}
10806 		}
10807 
10808 		action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP);
10809 		action->dta_rec.dtrd_size = size;
10810 	}
10811 
10812 	action->dta_refcnt = 1;
10813 	rec = &action->dta_rec;
10814 	size = rec->dtrd_size;
10815 
10816 	for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) {
10817 		if (!(size & mask)) {
10818 			align = mask + 1;
10819 			break;
10820 		}
10821 	}
10822 
10823 	action->dta_kind = desc->dtad_kind;
10824 
10825 	if ((action->dta_difo = dp) != NULL)
10826 		dtrace_difo_hold(dp);
10827 
10828 	rec->dtrd_action = action->dta_kind;
10829 	rec->dtrd_arg = arg;
10830 	rec->dtrd_uarg = desc->dtad_uarg;
10831 	rec->dtrd_alignment = (uint16_t)align;
10832 	rec->dtrd_format = format;
10833 
10834 	if ((last = ecb->dte_action_last) != NULL) {
10835 		ASSERT(ecb->dte_action != NULL);
10836 		action->dta_prev = last;
10837 		last->dta_next = action;
10838 	} else {
10839 		ASSERT(ecb->dte_action == NULL);
10840 		ecb->dte_action = action;
10841 	}
10842 
10843 	ecb->dte_action_last = action;
10844 
10845 	return (0);
10846 }
10847 
10848 static void
dtrace_ecb_action_remove(dtrace_ecb_t * ecb)10849 dtrace_ecb_action_remove(dtrace_ecb_t *ecb)
10850 {
10851 	dtrace_action_t *act = ecb->dte_action, *next;
10852 	dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate;
10853 	dtrace_difo_t *dp;
10854 	uint16_t format;
10855 
10856 	if (act != NULL && act->dta_refcnt > 1) {
10857 		ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1);
10858 		act->dta_refcnt--;
10859 	} else {
10860 		for (; act != NULL; act = next) {
10861 			next = act->dta_next;
10862 			ASSERT(next != NULL || act == ecb->dte_action_last);
10863 			ASSERT(act->dta_refcnt == 1);
10864 
10865 			if ((format = act->dta_rec.dtrd_format) != 0)
10866 				dtrace_format_remove(ecb->dte_state, format);
10867 
10868 			if ((dp = act->dta_difo) != NULL)
10869 				dtrace_difo_release(dp, vstate);
10870 
10871 			if (DTRACEACT_ISAGG(act->dta_kind)) {
10872 				dtrace_ecb_aggregation_destroy(ecb, act);
10873 			} else {
10874 				kmem_free(act, sizeof (dtrace_action_t));
10875 			}
10876 		}
10877 	}
10878 
10879 	ecb->dte_action = NULL;
10880 	ecb->dte_action_last = NULL;
10881 	ecb->dte_size = 0;
10882 }
10883 
10884 static void
dtrace_ecb_disable(dtrace_ecb_t * ecb)10885 dtrace_ecb_disable(dtrace_ecb_t *ecb)
10886 {
10887 	/*
10888 	 * We disable the ECB by removing it from its probe.
10889 	 */
10890 	dtrace_ecb_t *pecb, *prev = NULL;
10891 	dtrace_probe_t *probe = ecb->dte_probe;
10892 
10893 	ASSERT(MUTEX_HELD(&dtrace_lock));
10894 
10895 	if (probe == NULL) {
10896 		/*
10897 		 * This is the NULL probe; there is nothing to disable.
10898 		 */
10899 		return;
10900 	}
10901 
10902 	for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) {
10903 		if (pecb == ecb)
10904 			break;
10905 		prev = pecb;
10906 	}
10907 
10908 	ASSERT(pecb != NULL);
10909 
10910 	if (prev == NULL) {
10911 		probe->dtpr_ecb = ecb->dte_next;
10912 	} else {
10913 		prev->dte_next = ecb->dte_next;
10914 	}
10915 
10916 	if (ecb == probe->dtpr_ecb_last) {
10917 		ASSERT(ecb->dte_next == NULL);
10918 		probe->dtpr_ecb_last = prev;
10919 	}
10920 
10921 	/*
10922 	 * The ECB has been disconnected from the probe; now sync to assure
10923 	 * that all CPUs have seen the change before returning.
10924 	 */
10925 	dtrace_sync();
10926 
10927 	if (probe->dtpr_ecb == NULL) {
10928 		/*
10929 		 * That was the last ECB on the probe; clear the predicate
10930 		 * cache ID for the probe, disable it and sync one more time
10931 		 * to assure that we'll never hit it again.
10932 		 */
10933 		dtrace_provider_t *prov = probe->dtpr_provider;
10934 
10935 		ASSERT(ecb->dte_next == NULL);
10936 		ASSERT(probe->dtpr_ecb_last == NULL);
10937 		probe->dtpr_predcache = DTRACE_CACHEIDNONE;
10938 		prov->dtpv_pops.dtps_disable(prov->dtpv_arg,
10939 		    probe->dtpr_id, probe->dtpr_arg);
10940 		dtrace_sync();
10941 	} else {
10942 		/*
10943 		 * There is at least one ECB remaining on the probe.  If there
10944 		 * is _exactly_ one, set the probe's predicate cache ID to be
10945 		 * the predicate cache ID of the remaining ECB.
10946 		 */
10947 		ASSERT(probe->dtpr_ecb_last != NULL);
10948 		ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE);
10949 
10950 		if (probe->dtpr_ecb == probe->dtpr_ecb_last) {
10951 			dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate;
10952 
10953 			ASSERT(probe->dtpr_ecb->dte_next == NULL);
10954 
10955 			if (p != NULL)
10956 				probe->dtpr_predcache = p->dtp_cacheid;
10957 		}
10958 
10959 		ecb->dte_next = NULL;
10960 	}
10961 }
10962 
10963 static void
dtrace_ecb_destroy(dtrace_ecb_t * ecb)10964 dtrace_ecb_destroy(dtrace_ecb_t *ecb)
10965 {
10966 	dtrace_state_t *state = ecb->dte_state;
10967 	dtrace_vstate_t *vstate = &state->dts_vstate;
10968 	dtrace_predicate_t *pred;
10969 	dtrace_epid_t epid = ecb->dte_epid;
10970 
10971 	ASSERT(MUTEX_HELD(&dtrace_lock));
10972 	ASSERT(ecb->dte_next == NULL);
10973 	ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb);
10974 
10975 	if ((pred = ecb->dte_predicate) != NULL)
10976 		dtrace_predicate_release(pred, vstate);
10977 
10978 	dtrace_ecb_action_remove(ecb);
10979 
10980 	ASSERT(state->dts_ecbs[epid - 1] == ecb);
10981 	state->dts_ecbs[epid - 1] = NULL;
10982 
10983 	kmem_free(ecb, sizeof (dtrace_ecb_t));
10984 }
10985 
10986 static dtrace_ecb_t *
dtrace_ecb_create(dtrace_state_t * state,dtrace_probe_t * probe,dtrace_enabling_t * enab)10987 dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe,
10988     dtrace_enabling_t *enab)
10989 {
10990 	dtrace_ecb_t *ecb;
10991 	dtrace_predicate_t *pred;
10992 	dtrace_actdesc_t *act;
10993 	dtrace_provider_t *prov;
10994 	dtrace_ecbdesc_t *desc = enab->dten_current;
10995 
10996 	ASSERT(MUTEX_HELD(&dtrace_lock));
10997 	ASSERT(state != NULL);
10998 
10999 	ecb = dtrace_ecb_add(state, probe);
11000 	ecb->dte_uarg = desc->dted_uarg;
11001 
11002 	if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) {
11003 		dtrace_predicate_hold(pred);
11004 		ecb->dte_predicate = pred;
11005 	}
11006 
11007 	if (probe != NULL) {
11008 		/*
11009 		 * If the provider shows more leg than the consumer is old
11010 		 * enough to see, we need to enable the appropriate implicit
11011 		 * predicate bits to prevent the ecb from activating at
11012 		 * revealing times.
11013 		 *
11014 		 * Providers specifying DTRACE_PRIV_USER at register time
11015 		 * are stating that they need the /proc-style privilege
11016 		 * model to be enforced, and this is what DTRACE_COND_OWNER
11017 		 * and DTRACE_COND_ZONEOWNER will then do at probe time.
11018 		 */
11019 		prov = probe->dtpr_provider;
11020 		if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) &&
11021 		    (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
11022 			ecb->dte_cond |= DTRACE_COND_OWNER;
11023 
11024 		if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) &&
11025 		    (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
11026 			ecb->dte_cond |= DTRACE_COND_ZONEOWNER;
11027 
11028 		/*
11029 		 * If the provider shows us kernel innards and the user
11030 		 * is lacking sufficient privilege, enable the
11031 		 * DTRACE_COND_USERMODE implicit predicate.
11032 		 */
11033 		if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) &&
11034 		    (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL))
11035 			ecb->dte_cond |= DTRACE_COND_USERMODE;
11036 	}
11037 
11038 	if (dtrace_ecb_create_cache != NULL) {
11039 		/*
11040 		 * If we have a cached ecb, we'll use its action list instead
11041 		 * of creating our own (saving both time and space).
11042 		 */
11043 		dtrace_ecb_t *cached = dtrace_ecb_create_cache;
11044 		dtrace_action_t *act = cached->dte_action;
11045 
11046 		if (act != NULL) {
11047 			ASSERT(act->dta_refcnt > 0);
11048 			act->dta_refcnt++;
11049 			ecb->dte_action = act;
11050 			ecb->dte_action_last = cached->dte_action_last;
11051 			ecb->dte_needed = cached->dte_needed;
11052 			ecb->dte_size = cached->dte_size;
11053 			ecb->dte_alignment = cached->dte_alignment;
11054 		}
11055 
11056 		return (ecb);
11057 	}
11058 
11059 	for (act = desc->dted_action; act != NULL; act = act->dtad_next) {
11060 		if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) {
11061 			dtrace_ecb_destroy(ecb);
11062 			return (NULL);
11063 		}
11064 	}
11065 
11066 	dtrace_ecb_resize(ecb);
11067 
11068 	return (dtrace_ecb_create_cache = ecb);
11069 }
11070 
11071 static int
dtrace_ecb_create_enable(dtrace_probe_t * probe,void * arg)11072 dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg)
11073 {
11074 	dtrace_ecb_t *ecb;
11075 	dtrace_enabling_t *enab = arg;
11076 	dtrace_state_t *state = enab->dten_vstate->dtvs_state;
11077 
11078 	ASSERT(state != NULL);
11079 
11080 	if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) {
11081 		/*
11082 		 * This probe was created in a generation for which this
11083 		 * enabling has previously created ECBs; we don't want to
11084 		 * enable it again, so just kick out.
11085 		 */
11086 		return (DTRACE_MATCH_NEXT);
11087 	}
11088 
11089 	if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL)
11090 		return (DTRACE_MATCH_DONE);
11091 
11092 	if (dtrace_ecb_enable(ecb) < 0)
11093 		return (DTRACE_MATCH_FAIL);
11094 
11095 	return (DTRACE_MATCH_NEXT);
11096 }
11097 
11098 static dtrace_ecb_t *
dtrace_epid2ecb(dtrace_state_t * state,dtrace_epid_t id)11099 dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id)
11100 {
11101 	dtrace_ecb_t *ecb;
11102 
11103 	ASSERT(MUTEX_HELD(&dtrace_lock));
11104 
11105 	if (id == 0 || id > state->dts_necbs)
11106 		return (NULL);
11107 
11108 	ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL);
11109 	ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id);
11110 
11111 	return (state->dts_ecbs[id - 1]);
11112 }
11113 
11114 static dtrace_aggregation_t *
dtrace_aggid2agg(dtrace_state_t * state,dtrace_aggid_t id)11115 dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id)
11116 {
11117 	dtrace_aggregation_t *agg;
11118 
11119 	ASSERT(MUTEX_HELD(&dtrace_lock));
11120 
11121 	if (id == 0 || id > state->dts_naggregations)
11122 		return (NULL);
11123 
11124 	ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL);
11125 	ASSERT((agg = state->dts_aggregations[id - 1]) == NULL ||
11126 	    agg->dtag_id == id);
11127 
11128 	return (state->dts_aggregations[id - 1]);
11129 }
11130 
11131 /*
11132  * DTrace Buffer Functions
11133  *
11134  * The following functions manipulate DTrace buffers.  Most of these functions
11135  * are called in the context of establishing or processing consumer state;
11136  * exceptions are explicitly noted.
11137  */
11138 
11139 /*
11140  * Note:  called from cross call context.  This function switches the two
11141  * buffers on a given CPU.  The atomicity of this operation is assured by
11142  * disabling interrupts while the actual switch takes place; the disabling of
11143  * interrupts serializes the execution with any execution of dtrace_probe() on
11144  * the same CPU.
11145  */
11146 static void
dtrace_buffer_switch(dtrace_buffer_t * buf)11147 dtrace_buffer_switch(dtrace_buffer_t *buf)
11148 {
11149 	caddr_t tomax = buf->dtb_tomax;
11150 	caddr_t xamot = buf->dtb_xamot;
11151 	dtrace_icookie_t cookie;
11152 	hrtime_t now;
11153 
11154 	ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
11155 	ASSERT(!(buf->dtb_flags & DTRACEBUF_RING));
11156 
11157 	cookie = dtrace_interrupt_disable();
11158 	now = dtrace_gethrtime();
11159 	buf->dtb_tomax = xamot;
11160 	buf->dtb_xamot = tomax;
11161 	buf->dtb_xamot_drops = buf->dtb_drops;
11162 	buf->dtb_xamot_offset = buf->dtb_offset;
11163 	buf->dtb_xamot_errors = buf->dtb_errors;
11164 	buf->dtb_xamot_flags = buf->dtb_flags;
11165 	buf->dtb_offset = 0;
11166 	buf->dtb_drops = 0;
11167 	buf->dtb_errors = 0;
11168 	buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED);
11169 	buf->dtb_interval = now - buf->dtb_switched;
11170 	buf->dtb_switched = now;
11171 	dtrace_interrupt_enable(cookie);
11172 }
11173 
11174 /*
11175  * Note:  called from cross call context.  This function activates a buffer
11176  * on a CPU.  As with dtrace_buffer_switch(), the atomicity of the operation
11177  * is guaranteed by the disabling of interrupts.
11178  */
11179 static void
dtrace_buffer_activate(dtrace_state_t * state)11180 dtrace_buffer_activate(dtrace_state_t *state)
11181 {
11182 	dtrace_buffer_t *buf;
11183 	dtrace_icookie_t cookie = dtrace_interrupt_disable();
11184 
11185 	buf = &state->dts_buffer[CPU->cpu_id];
11186 
11187 	if (buf->dtb_tomax != NULL) {
11188 		/*
11189 		 * We might like to assert that the buffer is marked inactive,
11190 		 * but this isn't necessarily true:  the buffer for the CPU
11191 		 * that processes the BEGIN probe has its buffer activated
11192 		 * manually.  In this case, we take the (harmless) action
11193 		 * re-clearing the bit INACTIVE bit.
11194 		 */
11195 		buf->dtb_flags &= ~DTRACEBUF_INACTIVE;
11196 	}
11197 
11198 	dtrace_interrupt_enable(cookie);
11199 }
11200 
11201 static int
dtrace_buffer_alloc(dtrace_buffer_t * bufs,size_t size,int flags,processorid_t cpu,int * factor)11202 dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags,
11203     processorid_t cpu, int *factor)
11204 {
11205 	cpu_t *cp;
11206 	dtrace_buffer_t *buf;
11207 	int allocated = 0, desired = 0;
11208 
11209 	ASSERT(MUTEX_HELD(&cpu_lock));
11210 	ASSERT(MUTEX_HELD(&dtrace_lock));
11211 
11212 	*factor = 1;
11213 
11214 	if (size > dtrace_nonroot_maxsize &&
11215 	    !PRIV_POLICY_CHOICE(CRED(), PRIV_ALL, B_FALSE))
11216 		return (EFBIG);
11217 
11218 	cp = cpu_list;
11219 
11220 	do {
11221 		if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id)
11222 			continue;
11223 
11224 		buf = &bufs[cp->cpu_id];
11225 
11226 		/*
11227 		 * If there is already a buffer allocated for this CPU, it
11228 		 * is only possible that this is a DR event.  In this case,
11229 		 * the buffer size must match our specified size.
11230 		 */
11231 		if (buf->dtb_tomax != NULL) {
11232 			ASSERT(buf->dtb_size == size);
11233 			continue;
11234 		}
11235 
11236 		ASSERT(buf->dtb_xamot == NULL);
11237 
11238 		if ((buf->dtb_tomax = kmem_zalloc(size,
11239 		    KM_NOSLEEP | KM_NORMALPRI)) == NULL)
11240 			goto err;
11241 
11242 		buf->dtb_size = size;
11243 		buf->dtb_flags = flags;
11244 		buf->dtb_offset = 0;
11245 		buf->dtb_drops = 0;
11246 
11247 		if (flags & DTRACEBUF_NOSWITCH)
11248 			continue;
11249 
11250 		if ((buf->dtb_xamot = kmem_zalloc(size,
11251 		    KM_NOSLEEP | KM_NORMALPRI)) == NULL)
11252 			goto err;
11253 	} while ((cp = cp->cpu_next) != cpu_list);
11254 
11255 	return (0);
11256 
11257 err:
11258 	cp = cpu_list;
11259 
11260 	do {
11261 		if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id)
11262 			continue;
11263 
11264 		buf = &bufs[cp->cpu_id];
11265 		desired += 2;
11266 
11267 		if (buf->dtb_xamot != NULL) {
11268 			ASSERT(buf->dtb_tomax != NULL);
11269 			ASSERT(buf->dtb_size == size);
11270 			kmem_free(buf->dtb_xamot, size);
11271 			allocated++;
11272 		}
11273 
11274 		if (buf->dtb_tomax != NULL) {
11275 			ASSERT(buf->dtb_size == size);
11276 			kmem_free(buf->dtb_tomax, size);
11277 			allocated++;
11278 		}
11279 
11280 		buf->dtb_tomax = NULL;
11281 		buf->dtb_xamot = NULL;
11282 		buf->dtb_size = 0;
11283 	} while ((cp = cp->cpu_next) != cpu_list);
11284 
11285 	*factor = desired / (allocated > 0 ? allocated : 1);
11286 
11287 	return (ENOMEM);
11288 }
11289 
11290 /*
11291  * Note:  called from probe context.  This function just increments the drop
11292  * count on a buffer.  It has been made a function to allow for the
11293  * possibility of understanding the source of mysterious drop counts.  (A
11294  * problem for which one may be particularly disappointed that DTrace cannot
11295  * be used to understand DTrace.)
11296  */
11297 static void
dtrace_buffer_drop(dtrace_buffer_t * buf)11298 dtrace_buffer_drop(dtrace_buffer_t *buf)
11299 {
11300 	buf->dtb_drops++;
11301 }
11302 
11303 /*
11304  * Note:  called from probe context.  This function is called to reserve space
11305  * in a buffer.  If mstate is non-NULL, sets the scratch base and size in the
11306  * mstate.  Returns the new offset in the buffer, or a negative value if an
11307  * error has occurred.
11308  */
11309 static intptr_t
dtrace_buffer_reserve(dtrace_buffer_t * buf,size_t needed,size_t align,dtrace_state_t * state,dtrace_mstate_t * mstate)11310 dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align,
11311     dtrace_state_t *state, dtrace_mstate_t *mstate)
11312 {
11313 	intptr_t offs = buf->dtb_offset, soffs;
11314 	intptr_t woffs;
11315 	caddr_t tomax;
11316 	size_t total;
11317 
11318 	if (buf->dtb_flags & DTRACEBUF_INACTIVE)
11319 		return (-1);
11320 
11321 	if ((tomax = buf->dtb_tomax) == NULL) {
11322 		dtrace_buffer_drop(buf);
11323 		return (-1);
11324 	}
11325 
11326 	if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) {
11327 		while (offs & (align - 1)) {
11328 			/*
11329 			 * Assert that our alignment is off by a number which
11330 			 * is itself sizeof (uint32_t) aligned.
11331 			 */
11332 			ASSERT(!((align - (offs & (align - 1))) &
11333 			    (sizeof (uint32_t) - 1)));
11334 			DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
11335 			offs += sizeof (uint32_t);
11336 		}
11337 
11338 		if ((soffs = offs + needed) > buf->dtb_size) {
11339 			dtrace_buffer_drop(buf);
11340 			return (-1);
11341 		}
11342 
11343 		if (mstate == NULL)
11344 			return (offs);
11345 
11346 		mstate->dtms_scratch_base = (uintptr_t)tomax + soffs;
11347 		mstate->dtms_scratch_size = buf->dtb_size - soffs;
11348 		mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
11349 
11350 		return (offs);
11351 	}
11352 
11353 	if (buf->dtb_flags & DTRACEBUF_FILL) {
11354 		if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN &&
11355 		    (buf->dtb_flags & DTRACEBUF_FULL))
11356 			return (-1);
11357 		goto out;
11358 	}
11359 
11360 	total = needed + (offs & (align - 1));
11361 
11362 	/*
11363 	 * For a ring buffer, life is quite a bit more complicated.  Before
11364 	 * we can store any padding, we need to adjust our wrapping offset.
11365 	 * (If we've never before wrapped or we're not about to, no adjustment
11366 	 * is required.)
11367 	 */
11368 	if ((buf->dtb_flags & DTRACEBUF_WRAPPED) ||
11369 	    offs + total > buf->dtb_size) {
11370 		woffs = buf->dtb_xamot_offset;
11371 
11372 		if (offs + total > buf->dtb_size) {
11373 			/*
11374 			 * We can't fit in the end of the buffer.  First, a
11375 			 * sanity check that we can fit in the buffer at all.
11376 			 */
11377 			if (total > buf->dtb_size) {
11378 				dtrace_buffer_drop(buf);
11379 				return (-1);
11380 			}
11381 
11382 			/*
11383 			 * We're going to be storing at the top of the buffer,
11384 			 * so now we need to deal with the wrapped offset.  We
11385 			 * only reset our wrapped offset to 0 if it is
11386 			 * currently greater than the current offset.  If it
11387 			 * is less than the current offset, it is because a
11388 			 * previous allocation induced a wrap -- but the
11389 			 * allocation didn't subsequently take the space due
11390 			 * to an error or false predicate evaluation.  In this
11391 			 * case, we'll just leave the wrapped offset alone: if
11392 			 * the wrapped offset hasn't been advanced far enough
11393 			 * for this allocation, it will be adjusted in the
11394 			 * lower loop.
11395 			 */
11396 			if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
11397 				if (woffs >= offs)
11398 					woffs = 0;
11399 			} else {
11400 				woffs = 0;
11401 			}
11402 
11403 			/*
11404 			 * Now we know that we're going to be storing to the
11405 			 * top of the buffer and that there is room for us
11406 			 * there.  We need to clear the buffer from the current
11407 			 * offset to the end (there may be old gunk there).
11408 			 */
11409 			while (offs < buf->dtb_size)
11410 				tomax[offs++] = 0;
11411 
11412 			/*
11413 			 * We need to set our offset to zero.  And because we
11414 			 * are wrapping, we need to set the bit indicating as
11415 			 * much.  We can also adjust our needed space back
11416 			 * down to the space required by the ECB -- we know
11417 			 * that the top of the buffer is aligned.
11418 			 */
11419 			offs = 0;
11420 			total = needed;
11421 			buf->dtb_flags |= DTRACEBUF_WRAPPED;
11422 		} else {
11423 			/*
11424 			 * There is room for us in the buffer, so we simply
11425 			 * need to check the wrapped offset.
11426 			 */
11427 			if (woffs < offs) {
11428 				/*
11429 				 * The wrapped offset is less than the offset.
11430 				 * This can happen if we allocated buffer space
11431 				 * that induced a wrap, but then we didn't
11432 				 * subsequently take the space due to an error
11433 				 * or false predicate evaluation.  This is
11434 				 * okay; we know that _this_ allocation isn't
11435 				 * going to induce a wrap.  We still can't
11436 				 * reset the wrapped offset to be zero,
11437 				 * however: the space may have been trashed in
11438 				 * the previous failed probe attempt.  But at
11439 				 * least the wrapped offset doesn't need to
11440 				 * be adjusted at all...
11441 				 */
11442 				goto out;
11443 			}
11444 		}
11445 
11446 		while (offs + total > woffs) {
11447 			dtrace_epid_t epid = *(uint32_t *)(tomax + woffs);
11448 			size_t size;
11449 
11450 			if (epid == DTRACE_EPIDNONE) {
11451 				size = sizeof (uint32_t);
11452 			} else {
11453 				ASSERT3U(epid, <=, state->dts_necbs);
11454 				ASSERT(state->dts_ecbs[epid - 1] != NULL);
11455 
11456 				size = state->dts_ecbs[epid - 1]->dte_size;
11457 			}
11458 
11459 			ASSERT(woffs + size <= buf->dtb_size);
11460 			ASSERT(size != 0);
11461 
11462 			if (woffs + size == buf->dtb_size) {
11463 				/*
11464 				 * We've reached the end of the buffer; we want
11465 				 * to set the wrapped offset to 0 and break
11466 				 * out.  However, if the offs is 0, then we're
11467 				 * in a strange edge-condition:  the amount of
11468 				 * space that we want to reserve plus the size
11469 				 * of the record that we're overwriting is
11470 				 * greater than the size of the buffer.  This
11471 				 * is problematic because if we reserve the
11472 				 * space but subsequently don't consume it (due
11473 				 * to a failed predicate or error) the wrapped
11474 				 * offset will be 0 -- yet the EPID at offset 0
11475 				 * will not be committed.  This situation is
11476 				 * relatively easy to deal with:  if we're in
11477 				 * this case, the buffer is indistinguishable
11478 				 * from one that hasn't wrapped; we need only
11479 				 * finish the job by clearing the wrapped bit,
11480 				 * explicitly setting the offset to be 0, and
11481 				 * zero'ing out the old data in the buffer.
11482 				 */
11483 				if (offs == 0) {
11484 					buf->dtb_flags &= ~DTRACEBUF_WRAPPED;
11485 					buf->dtb_offset = 0;
11486 					woffs = total;
11487 
11488 					while (woffs < buf->dtb_size)
11489 						tomax[woffs++] = 0;
11490 				}
11491 
11492 				woffs = 0;
11493 				break;
11494 			}
11495 
11496 			woffs += size;
11497 		}
11498 
11499 		/*
11500 		 * We have a wrapped offset.  It may be that the wrapped offset
11501 		 * has become zero -- that's okay.
11502 		 */
11503 		buf->dtb_xamot_offset = woffs;
11504 	}
11505 
11506 out:
11507 	/*
11508 	 * Now we can plow the buffer with any necessary padding.
11509 	 */
11510 	while (offs & (align - 1)) {
11511 		/*
11512 		 * Assert that our alignment is off by a number which
11513 		 * is itself sizeof (uint32_t) aligned.
11514 		 */
11515 		ASSERT(!((align - (offs & (align - 1))) &
11516 		    (sizeof (uint32_t) - 1)));
11517 		DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
11518 		offs += sizeof (uint32_t);
11519 	}
11520 
11521 	if (buf->dtb_flags & DTRACEBUF_FILL) {
11522 		if (offs + needed > buf->dtb_size - state->dts_reserve) {
11523 			buf->dtb_flags |= DTRACEBUF_FULL;
11524 			return (-1);
11525 		}
11526 	}
11527 
11528 	if (mstate == NULL)
11529 		return (offs);
11530 
11531 	/*
11532 	 * For ring buffers and fill buffers, the scratch space is always
11533 	 * the inactive buffer.
11534 	 */
11535 	mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot;
11536 	mstate->dtms_scratch_size = buf->dtb_size;
11537 	mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
11538 
11539 	return (offs);
11540 }
11541 
11542 static void
dtrace_buffer_polish(dtrace_buffer_t * buf)11543 dtrace_buffer_polish(dtrace_buffer_t *buf)
11544 {
11545 	ASSERT(buf->dtb_flags & DTRACEBUF_RING);
11546 	ASSERT(MUTEX_HELD(&dtrace_lock));
11547 
11548 	if (!(buf->dtb_flags & DTRACEBUF_WRAPPED))
11549 		return;
11550 
11551 	/*
11552 	 * We need to polish the ring buffer.  There are three cases:
11553 	 *
11554 	 * - The first (and presumably most common) is that there is no gap
11555 	 *   between the buffer offset and the wrapped offset.  In this case,
11556 	 *   there is nothing in the buffer that isn't valid data; we can
11557 	 *   mark the buffer as polished and return.
11558 	 *
11559 	 * - The second (less common than the first but still more common
11560 	 *   than the third) is that there is a gap between the buffer offset
11561 	 *   and the wrapped offset, and the wrapped offset is larger than the
11562 	 *   buffer offset.  This can happen because of an alignment issue, or
11563 	 *   can happen because of a call to dtrace_buffer_reserve() that
11564 	 *   didn't subsequently consume the buffer space.  In this case,
11565 	 *   we need to zero the data from the buffer offset to the wrapped
11566 	 *   offset.
11567 	 *
11568 	 * - The third (and least common) is that there is a gap between the
11569 	 *   buffer offset and the wrapped offset, but the wrapped offset is
11570 	 *   _less_ than the buffer offset.  This can only happen because a
11571 	 *   call to dtrace_buffer_reserve() induced a wrap, but the space
11572 	 *   was not subsequently consumed.  In this case, we need to zero the
11573 	 *   space from the offset to the end of the buffer _and_ from the
11574 	 *   top of the buffer to the wrapped offset.
11575 	 */
11576 	if (buf->dtb_offset < buf->dtb_xamot_offset) {
11577 		bzero(buf->dtb_tomax + buf->dtb_offset,
11578 		    buf->dtb_xamot_offset - buf->dtb_offset);
11579 	}
11580 
11581 	if (buf->dtb_offset > buf->dtb_xamot_offset) {
11582 		bzero(buf->dtb_tomax + buf->dtb_offset,
11583 		    buf->dtb_size - buf->dtb_offset);
11584 		bzero(buf->dtb_tomax, buf->dtb_xamot_offset);
11585 	}
11586 }
11587 
11588 /*
11589  * This routine determines if data generated at the specified time has likely
11590  * been entirely consumed at user-level.  This routine is called to determine
11591  * if an ECB on a defunct probe (but for an active enabling) can be safely
11592  * disabled and destroyed.
11593  */
11594 static int
dtrace_buffer_consumed(dtrace_buffer_t * bufs,hrtime_t when)11595 dtrace_buffer_consumed(dtrace_buffer_t *bufs, hrtime_t when)
11596 {
11597 	int i;
11598 
11599 	for (i = 0; i < NCPU; i++) {
11600 		dtrace_buffer_t *buf = &bufs[i];
11601 
11602 		if (buf->dtb_size == 0)
11603 			continue;
11604 
11605 		if (buf->dtb_flags & DTRACEBUF_RING)
11606 			return (0);
11607 
11608 		if (!buf->dtb_switched && buf->dtb_offset != 0)
11609 			return (0);
11610 
11611 		if (buf->dtb_switched - buf->dtb_interval < when)
11612 			return (0);
11613 	}
11614 
11615 	return (1);
11616 }
11617 
11618 static void
dtrace_buffer_free(dtrace_buffer_t * bufs)11619 dtrace_buffer_free(dtrace_buffer_t *bufs)
11620 {
11621 	int i;
11622 
11623 	for (i = 0; i < NCPU; i++) {
11624 		dtrace_buffer_t *buf = &bufs[i];
11625 
11626 		if (buf->dtb_tomax == NULL) {
11627 			ASSERT(buf->dtb_xamot == NULL);
11628 			ASSERT(buf->dtb_size == 0);
11629 			continue;
11630 		}
11631 
11632 		if (buf->dtb_xamot != NULL) {
11633 			ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
11634 			kmem_free(buf->dtb_xamot, buf->dtb_size);
11635 		}
11636 
11637 		kmem_free(buf->dtb_tomax, buf->dtb_size);
11638 		buf->dtb_size = 0;
11639 		buf->dtb_tomax = NULL;
11640 		buf->dtb_xamot = NULL;
11641 	}
11642 }
11643 
11644 /*
11645  * DTrace Enabling Functions
11646  */
11647 static dtrace_enabling_t *
dtrace_enabling_create(dtrace_vstate_t * vstate)11648 dtrace_enabling_create(dtrace_vstate_t *vstate)
11649 {
11650 	dtrace_enabling_t *enab;
11651 
11652 	enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP);
11653 	enab->dten_vstate = vstate;
11654 
11655 	return (enab);
11656 }
11657 
11658 static void
dtrace_enabling_add(dtrace_enabling_t * enab,dtrace_ecbdesc_t * ecb)11659 dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb)
11660 {
11661 	dtrace_ecbdesc_t **ndesc;
11662 	size_t osize, nsize;
11663 
11664 	/*
11665 	 * We can't add to enablings after we've enabled them, or after we've
11666 	 * retained them.
11667 	 */
11668 	ASSERT(enab->dten_probegen == 0);
11669 	ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11670 
11671 	if (enab->dten_ndesc < enab->dten_maxdesc) {
11672 		enab->dten_desc[enab->dten_ndesc++] = ecb;
11673 		return;
11674 	}
11675 
11676 	osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11677 
11678 	if (enab->dten_maxdesc == 0) {
11679 		enab->dten_maxdesc = 1;
11680 	} else {
11681 		enab->dten_maxdesc <<= 1;
11682 	}
11683 
11684 	ASSERT(enab->dten_ndesc < enab->dten_maxdesc);
11685 
11686 	nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11687 	ndesc = kmem_zalloc(nsize, KM_SLEEP);
11688 	bcopy(enab->dten_desc, ndesc, osize);
11689 	kmem_free(enab->dten_desc, osize);
11690 
11691 	enab->dten_desc = ndesc;
11692 	enab->dten_desc[enab->dten_ndesc++] = ecb;
11693 }
11694 
11695 static void
dtrace_enabling_addlike(dtrace_enabling_t * enab,dtrace_ecbdesc_t * ecb,dtrace_probedesc_t * pd)11696 dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb,
11697     dtrace_probedesc_t *pd)
11698 {
11699 	dtrace_ecbdesc_t *new;
11700 	dtrace_predicate_t *pred;
11701 	dtrace_actdesc_t *act;
11702 
11703 	/*
11704 	 * We're going to create a new ECB description that matches the
11705 	 * specified ECB in every way, but has the specified probe description.
11706 	 */
11707 	new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
11708 
11709 	if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL)
11710 		dtrace_predicate_hold(pred);
11711 
11712 	for (act = ecb->dted_action; act != NULL; act = act->dtad_next)
11713 		dtrace_actdesc_hold(act);
11714 
11715 	new->dted_action = ecb->dted_action;
11716 	new->dted_pred = ecb->dted_pred;
11717 	new->dted_probe = *pd;
11718 	new->dted_uarg = ecb->dted_uarg;
11719 
11720 	dtrace_enabling_add(enab, new);
11721 }
11722 
11723 static void
dtrace_enabling_dump(dtrace_enabling_t * enab)11724 dtrace_enabling_dump(dtrace_enabling_t *enab)
11725 {
11726 	int i;
11727 
11728 	for (i = 0; i < enab->dten_ndesc; i++) {
11729 		dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe;
11730 
11731 		cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i,
11732 		    desc->dtpd_provider, desc->dtpd_mod,
11733 		    desc->dtpd_func, desc->dtpd_name);
11734 	}
11735 }
11736 
11737 static void
dtrace_enabling_destroy(dtrace_enabling_t * enab)11738 dtrace_enabling_destroy(dtrace_enabling_t *enab)
11739 {
11740 	int i;
11741 	dtrace_ecbdesc_t *ep;
11742 	dtrace_vstate_t *vstate = enab->dten_vstate;
11743 
11744 	ASSERT(MUTEX_HELD(&dtrace_lock));
11745 
11746 	for (i = 0; i < enab->dten_ndesc; i++) {
11747 		dtrace_actdesc_t *act, *next;
11748 		dtrace_predicate_t *pred;
11749 
11750 		ep = enab->dten_desc[i];
11751 
11752 		if ((pred = ep->dted_pred.dtpdd_predicate) != NULL)
11753 			dtrace_predicate_release(pred, vstate);
11754 
11755 		for (act = ep->dted_action; act != NULL; act = next) {
11756 			next = act->dtad_next;
11757 			dtrace_actdesc_release(act, vstate);
11758 		}
11759 
11760 		kmem_free(ep, sizeof (dtrace_ecbdesc_t));
11761 	}
11762 
11763 	kmem_free(enab->dten_desc,
11764 	    enab->dten_maxdesc * sizeof (dtrace_enabling_t *));
11765 
11766 	/*
11767 	 * If this was a retained enabling, decrement the dts_nretained count
11768 	 * and take it off of the dtrace_retained list.
11769 	 */
11770 	if (enab->dten_prev != NULL || enab->dten_next != NULL ||
11771 	    dtrace_retained == enab) {
11772 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
11773 		ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0);
11774 		enab->dten_vstate->dtvs_state->dts_nretained--;
11775 		dtrace_retained_gen++;
11776 	}
11777 
11778 	if (enab->dten_prev == NULL) {
11779 		if (dtrace_retained == enab) {
11780 			dtrace_retained = enab->dten_next;
11781 
11782 			if (dtrace_retained != NULL)
11783 				dtrace_retained->dten_prev = NULL;
11784 		}
11785 	} else {
11786 		ASSERT(enab != dtrace_retained);
11787 		ASSERT(dtrace_retained != NULL);
11788 		enab->dten_prev->dten_next = enab->dten_next;
11789 	}
11790 
11791 	if (enab->dten_next != NULL) {
11792 		ASSERT(dtrace_retained != NULL);
11793 		enab->dten_next->dten_prev = enab->dten_prev;
11794 	}
11795 
11796 	kmem_free(enab, sizeof (dtrace_enabling_t));
11797 }
11798 
11799 static int
dtrace_enabling_retain(dtrace_enabling_t * enab)11800 dtrace_enabling_retain(dtrace_enabling_t *enab)
11801 {
11802 	dtrace_state_t *state;
11803 
11804 	ASSERT(MUTEX_HELD(&dtrace_lock));
11805 	ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11806 	ASSERT(enab->dten_vstate != NULL);
11807 
11808 	state = enab->dten_vstate->dtvs_state;
11809 	ASSERT(state != NULL);
11810 
11811 	/*
11812 	 * We only allow each state to retain dtrace_retain_max enablings.
11813 	 */
11814 	if (state->dts_nretained >= dtrace_retain_max)
11815 		return (ENOSPC);
11816 
11817 	state->dts_nretained++;
11818 	dtrace_retained_gen++;
11819 
11820 	if (dtrace_retained == NULL) {
11821 		dtrace_retained = enab;
11822 		return (0);
11823 	}
11824 
11825 	enab->dten_next = dtrace_retained;
11826 	dtrace_retained->dten_prev = enab;
11827 	dtrace_retained = enab;
11828 
11829 	return (0);
11830 }
11831 
11832 static int
dtrace_enabling_replicate(dtrace_state_t * state,dtrace_probedesc_t * match,dtrace_probedesc_t * create)11833 dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match,
11834     dtrace_probedesc_t *create)
11835 {
11836 	dtrace_enabling_t *new, *enab;
11837 	int found = 0, err = ENOENT;
11838 
11839 	ASSERT(MUTEX_HELD(&dtrace_lock));
11840 	ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN);
11841 	ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN);
11842 	ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN);
11843 	ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN);
11844 
11845 	new = dtrace_enabling_create(&state->dts_vstate);
11846 
11847 	/*
11848 	 * Iterate over all retained enablings, looking for enablings that
11849 	 * match the specified state.
11850 	 */
11851 	for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11852 		int i;
11853 
11854 		/*
11855 		 * dtvs_state can only be NULL for helper enablings -- and
11856 		 * helper enablings can't be retained.
11857 		 */
11858 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
11859 
11860 		if (enab->dten_vstate->dtvs_state != state)
11861 			continue;
11862 
11863 		/*
11864 		 * Now iterate over each probe description; we're looking for
11865 		 * an exact match to the specified probe description.
11866 		 */
11867 		for (i = 0; i < enab->dten_ndesc; i++) {
11868 			dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11869 			dtrace_probedesc_t *pd = &ep->dted_probe;
11870 
11871 			if (strcmp(pd->dtpd_provider, match->dtpd_provider))
11872 				continue;
11873 
11874 			if (strcmp(pd->dtpd_mod, match->dtpd_mod))
11875 				continue;
11876 
11877 			if (strcmp(pd->dtpd_func, match->dtpd_func))
11878 				continue;
11879 
11880 			if (strcmp(pd->dtpd_name, match->dtpd_name))
11881 				continue;
11882 
11883 			/*
11884 			 * We have a winning probe!  Add it to our growing
11885 			 * enabling.
11886 			 */
11887 			found = 1;
11888 			dtrace_enabling_addlike(new, ep, create);
11889 		}
11890 	}
11891 
11892 	if (!found || (err = dtrace_enabling_retain(new)) != 0) {
11893 		dtrace_enabling_destroy(new);
11894 		return (err);
11895 	}
11896 
11897 	return (0);
11898 }
11899 
11900 static void
dtrace_enabling_retract(dtrace_state_t * state)11901 dtrace_enabling_retract(dtrace_state_t *state)
11902 {
11903 	dtrace_enabling_t *enab, *next;
11904 
11905 	ASSERT(MUTEX_HELD(&dtrace_lock));
11906 
11907 	/*
11908 	 * Iterate over all retained enablings, destroy the enablings retained
11909 	 * for the specified state.
11910 	 */
11911 	for (enab = dtrace_retained; enab != NULL; enab = next) {
11912 		next = enab->dten_next;
11913 
11914 		/*
11915 		 * dtvs_state can only be NULL for helper enablings -- and
11916 		 * helper enablings can't be retained.
11917 		 */
11918 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
11919 
11920 		if (enab->dten_vstate->dtvs_state == state) {
11921 			ASSERT(state->dts_nretained > 0);
11922 			dtrace_enabling_destroy(enab);
11923 		}
11924 	}
11925 
11926 	ASSERT(state->dts_nretained == 0);
11927 }
11928 
11929 static int
dtrace_enabling_match(dtrace_enabling_t * enab,int * nmatched)11930 dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched)
11931 {
11932 	int i = 0;
11933 	int total_matched = 0, matched = 0;
11934 
11935 	ASSERT(MUTEX_HELD(&cpu_lock));
11936 	ASSERT(MUTEX_HELD(&dtrace_lock));
11937 
11938 	for (i = 0; i < enab->dten_ndesc; i++) {
11939 		dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11940 
11941 		enab->dten_current = ep;
11942 		enab->dten_error = 0;
11943 
11944 		/*
11945 		 * If a provider failed to enable a probe then get out and
11946 		 * let the consumer know we failed.
11947 		 */
11948 		if ((matched = dtrace_probe_enable(&ep->dted_probe, enab)) < 0)
11949 			return (EBUSY);
11950 
11951 		total_matched += matched;
11952 
11953 		if (enab->dten_error != 0) {
11954 			/*
11955 			 * If we get an error half-way through enabling the
11956 			 * probes, we kick out -- perhaps with some number of
11957 			 * them enabled.  Leaving enabled probes enabled may
11958 			 * be slightly confusing for user-level, but we expect
11959 			 * that no one will attempt to actually drive on in
11960 			 * the face of such errors.  If this is an anonymous
11961 			 * enabling (indicated with a NULL nmatched pointer),
11962 			 * we cmn_err() a message.  We aren't expecting to
11963 			 * get such an error -- such as it can exist at all,
11964 			 * it would be a result of corrupted DOF in the driver
11965 			 * properties.
11966 			 */
11967 			if (nmatched == NULL) {
11968 				cmn_err(CE_WARN, "dtrace_enabling_match() "
11969 				    "error on %p: %d", (void *)ep,
11970 				    enab->dten_error);
11971 			}
11972 
11973 			return (enab->dten_error);
11974 		}
11975 	}
11976 
11977 	enab->dten_probegen = dtrace_probegen;
11978 	if (nmatched != NULL)
11979 		*nmatched = total_matched;
11980 
11981 	return (0);
11982 }
11983 
11984 static void
dtrace_enabling_matchall(void)11985 dtrace_enabling_matchall(void)
11986 {
11987 	dtrace_enabling_t *enab;
11988 
11989 	mutex_enter(&cpu_lock);
11990 	mutex_enter(&dtrace_lock);
11991 
11992 	/*
11993 	 * Iterate over all retained enablings to see if any probes match
11994 	 * against them.  We only perform this operation on enablings for which
11995 	 * we have sufficient permissions by virtue of being in the global zone
11996 	 * or in the same zone as the DTrace client.  Because we can be called
11997 	 * after dtrace_detach() has been called, we cannot assert that there
11998 	 * are retained enablings.  We can safely load from dtrace_retained,
11999 	 * however:  the taskq_destroy() at the end of dtrace_detach() will
12000 	 * block pending our completion.
12001 	 */
12002 	for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
12003 		dtrace_cred_t *dcr = &enab->dten_vstate->dtvs_state->dts_cred;
12004 		cred_t *cr = dcr->dcr_cred;
12005 		zoneid_t zone = cr != NULL ? crgetzoneid(cr) : 0;
12006 
12007 		if ((dcr->dcr_visible & DTRACE_CRV_ALLZONE) || (cr != NULL &&
12008 		    (zone == GLOBAL_ZONEID || getzoneid() == zone)))
12009 			(void) dtrace_enabling_match(enab, NULL);
12010 	}
12011 
12012 	mutex_exit(&dtrace_lock);
12013 	mutex_exit(&cpu_lock);
12014 }
12015 
12016 /*
12017  * If an enabling is to be enabled without having matched probes (that is, if
12018  * dtrace_state_go() is to be called on the underlying dtrace_state_t), the
12019  * enabling must be _primed_ by creating an ECB for every ECB description.
12020  * This must be done to assure that we know the number of speculations, the
12021  * number of aggregations, the minimum buffer size needed, etc. before we
12022  * transition out of DTRACE_ACTIVITY_INACTIVE.  To do this without actually
12023  * enabling any probes, we create ECBs for every ECB decription, but with a
12024  * NULL probe -- which is exactly what this function does.
12025  */
12026 static void
dtrace_enabling_prime(dtrace_state_t * state)12027 dtrace_enabling_prime(dtrace_state_t *state)
12028 {
12029 	dtrace_enabling_t *enab;
12030 	int i;
12031 
12032 	for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
12033 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
12034 
12035 		if (enab->dten_vstate->dtvs_state != state)
12036 			continue;
12037 
12038 		/*
12039 		 * We don't want to prime an enabling more than once, lest
12040 		 * we allow a malicious user to induce resource exhaustion.
12041 		 * (The ECBs that result from priming an enabling aren't
12042 		 * leaked -- but they also aren't deallocated until the
12043 		 * consumer state is destroyed.)
12044 		 */
12045 		if (enab->dten_primed)
12046 			continue;
12047 
12048 		for (i = 0; i < enab->dten_ndesc; i++) {
12049 			enab->dten_current = enab->dten_desc[i];
12050 			(void) dtrace_probe_enable(NULL, enab);
12051 		}
12052 
12053 		enab->dten_primed = 1;
12054 	}
12055 }
12056 
12057 /*
12058  * Called to indicate that probes should be provided due to retained
12059  * enablings.  This is implemented in terms of dtrace_probe_provide(), but it
12060  * must take an initial lap through the enabling calling the dtps_provide()
12061  * entry point explicitly to allow for autocreated probes.
12062  */
12063 static void
dtrace_enabling_provide(dtrace_provider_t * prv)12064 dtrace_enabling_provide(dtrace_provider_t *prv)
12065 {
12066 	int i, all = 0;
12067 	dtrace_probedesc_t desc;
12068 	dtrace_genid_t gen;
12069 
12070 	ASSERT(MUTEX_HELD(&dtrace_lock));
12071 	ASSERT(MUTEX_HELD(&dtrace_provider_lock));
12072 
12073 	if (prv == NULL) {
12074 		all = 1;
12075 		prv = dtrace_provider;
12076 	}
12077 
12078 	do {
12079 		dtrace_enabling_t *enab;
12080 		void *parg = prv->dtpv_arg;
12081 
12082 retry:
12083 		gen = dtrace_retained_gen;
12084 		for (enab = dtrace_retained; enab != NULL;
12085 		    enab = enab->dten_next) {
12086 			for (i = 0; i < enab->dten_ndesc; i++) {
12087 				desc = enab->dten_desc[i]->dted_probe;
12088 				mutex_exit(&dtrace_lock);
12089 				prv->dtpv_pops.dtps_provide(parg, &desc);
12090 				mutex_enter(&dtrace_lock);
12091 				/*
12092 				 * Process the retained enablings again if
12093 				 * they have changed while we weren't holding
12094 				 * dtrace_lock.
12095 				 */
12096 				if (gen != dtrace_retained_gen)
12097 					goto retry;
12098 			}
12099 		}
12100 	} while (all && (prv = prv->dtpv_next) != NULL);
12101 
12102 	mutex_exit(&dtrace_lock);
12103 	dtrace_probe_provide(NULL, all ? NULL : prv);
12104 	mutex_enter(&dtrace_lock);
12105 }
12106 
12107 /*
12108  * Called to reap ECBs that are attached to probes from defunct providers.
12109  */
12110 static void
dtrace_enabling_reap(void)12111 dtrace_enabling_reap(void)
12112 {
12113 	dtrace_provider_t *prov;
12114 	dtrace_probe_t *probe;
12115 	dtrace_ecb_t *ecb;
12116 	hrtime_t when;
12117 	int i;
12118 
12119 	mutex_enter(&cpu_lock);
12120 	mutex_enter(&dtrace_lock);
12121 
12122 	for (i = 0; i < dtrace_nprobes; i++) {
12123 		if ((probe = dtrace_probes[i]) == NULL)
12124 			continue;
12125 
12126 		if (probe->dtpr_ecb == NULL)
12127 			continue;
12128 
12129 		prov = probe->dtpr_provider;
12130 
12131 		if ((when = prov->dtpv_defunct) == 0)
12132 			continue;
12133 
12134 		/*
12135 		 * We have ECBs on a defunct provider:  we want to reap these
12136 		 * ECBs to allow the provider to unregister.  The destruction
12137 		 * of these ECBs must be done carefully:  if we destroy the ECB
12138 		 * and the consumer later wishes to consume an EPID that
12139 		 * corresponds to the destroyed ECB (and if the EPID metadata
12140 		 * has not been previously consumed), the consumer will abort
12141 		 * processing on the unknown EPID.  To reduce (but not, sadly,
12142 		 * eliminate) the possibility of this, we will only destroy an
12143 		 * ECB for a defunct provider if, for the state that
12144 		 * corresponds to the ECB:
12145 		 *
12146 		 *  (a)	There is no speculative tracing (which can effectively
12147 		 *	cache an EPID for an arbitrary amount of time).
12148 		 *
12149 		 *  (b)	The principal buffers have been switched twice since the
12150 		 *	provider became defunct.
12151 		 *
12152 		 *  (c)	The aggregation buffers are of zero size or have been
12153 		 *	switched twice since the provider became defunct.
12154 		 *
12155 		 * We use dts_speculates to determine (a) and call a function
12156 		 * (dtrace_buffer_consumed()) to determine (b) and (c).  Note
12157 		 * that as soon as we've been unable to destroy one of the ECBs
12158 		 * associated with the probe, we quit trying -- reaping is only
12159 		 * fruitful in as much as we can destroy all ECBs associated
12160 		 * with the defunct provider's probes.
12161 		 */
12162 		while ((ecb = probe->dtpr_ecb) != NULL) {
12163 			dtrace_state_t *state = ecb->dte_state;
12164 			dtrace_buffer_t *buf = state->dts_buffer;
12165 			dtrace_buffer_t *aggbuf = state->dts_aggbuffer;
12166 
12167 			if (state->dts_speculates)
12168 				break;
12169 
12170 			if (!dtrace_buffer_consumed(buf, when))
12171 				break;
12172 
12173 			if (!dtrace_buffer_consumed(aggbuf, when))
12174 				break;
12175 
12176 			dtrace_ecb_disable(ecb);
12177 			ASSERT(probe->dtpr_ecb != ecb);
12178 			dtrace_ecb_destroy(ecb);
12179 		}
12180 	}
12181 
12182 	mutex_exit(&dtrace_lock);
12183 	mutex_exit(&cpu_lock);
12184 }
12185 
12186 /*
12187  * DTrace DOF Functions
12188  */
12189 /*ARGSUSED*/
12190 static void
dtrace_dof_error(dof_hdr_t * dof,const char * str)12191 dtrace_dof_error(dof_hdr_t *dof, const char *str)
12192 {
12193 	if (dtrace_err_verbose)
12194 		cmn_err(CE_WARN, "failed to process DOF: %s", str);
12195 
12196 #ifdef DTRACE_ERRDEBUG
12197 	dtrace_errdebug(str);
12198 #endif
12199 }
12200 
12201 /*
12202  * Create DOF out of a currently enabled state.  Right now, we only create
12203  * DOF containing the run-time options -- but this could be expanded to create
12204  * complete DOF representing the enabled state.
12205  */
12206 static dof_hdr_t *
dtrace_dof_create(dtrace_state_t * state)12207 dtrace_dof_create(dtrace_state_t *state)
12208 {
12209 	dof_hdr_t *dof;
12210 	dof_sec_t *sec;
12211 	dof_optdesc_t *opt;
12212 	int i, len = sizeof (dof_hdr_t) +
12213 	    roundup(sizeof (dof_sec_t), sizeof (uint64_t)) +
12214 	    sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
12215 
12216 	ASSERT(MUTEX_HELD(&dtrace_lock));
12217 
12218 	dof = kmem_zalloc(len, KM_SLEEP);
12219 	dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0;
12220 	dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1;
12221 	dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2;
12222 	dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3;
12223 
12224 	dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE;
12225 	dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE;
12226 	dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION;
12227 	dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION;
12228 	dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS;
12229 	dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS;
12230 
12231 	dof->dofh_flags = 0;
12232 	dof->dofh_hdrsize = sizeof (dof_hdr_t);
12233 	dof->dofh_secsize = sizeof (dof_sec_t);
12234 	dof->dofh_secnum = 1;	/* only DOF_SECT_OPTDESC */
12235 	dof->dofh_secoff = sizeof (dof_hdr_t);
12236 	dof->dofh_loadsz = len;
12237 	dof->dofh_filesz = len;
12238 	dof->dofh_pad = 0;
12239 
12240 	/*
12241 	 * Fill in the option section header...
12242 	 */
12243 	sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t));
12244 	sec->dofs_type = DOF_SECT_OPTDESC;
12245 	sec->dofs_align = sizeof (uint64_t);
12246 	sec->dofs_flags = DOF_SECF_LOAD;
12247 	sec->dofs_entsize = sizeof (dof_optdesc_t);
12248 
12249 	opt = (dof_optdesc_t *)((uintptr_t)sec +
12250 	    roundup(sizeof (dof_sec_t), sizeof (uint64_t)));
12251 
12252 	sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof;
12253 	sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
12254 
12255 	for (i = 0; i < DTRACEOPT_MAX; i++) {
12256 		opt[i].dofo_option = i;
12257 		opt[i].dofo_strtab = DOF_SECIDX_NONE;
12258 		opt[i].dofo_value = state->dts_options[i];
12259 	}
12260 
12261 	return (dof);
12262 }
12263 
12264 static dof_hdr_t *
dtrace_dof_copyin(uintptr_t uarg,int * errp)12265 dtrace_dof_copyin(uintptr_t uarg, int *errp)
12266 {
12267 	dof_hdr_t hdr, *dof;
12268 
12269 	ASSERT(!MUTEX_HELD(&dtrace_lock));
12270 
12271 	/*
12272 	 * First, we're going to copyin() the sizeof (dof_hdr_t).
12273 	 */
12274 	if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) {
12275 		dtrace_dof_error(NULL, "failed to copyin DOF header");
12276 		*errp = EFAULT;
12277 		return (NULL);
12278 	}
12279 
12280 	/*
12281 	 * Now we'll allocate the entire DOF and copy it in -- provided
12282 	 * that the length isn't outrageous.
12283 	 */
12284 	if (hdr.dofh_loadsz >= dtrace_dof_maxsize) {
12285 		dtrace_dof_error(&hdr, "load size exceeds maximum");
12286 		*errp = E2BIG;
12287 		return (NULL);
12288 	}
12289 
12290 	if (hdr.dofh_loadsz < sizeof (hdr)) {
12291 		dtrace_dof_error(&hdr, "invalid load size");
12292 		*errp = EINVAL;
12293 		return (NULL);
12294 	}
12295 
12296 	dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP);
12297 
12298 	if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0 ||
12299 	    dof->dofh_loadsz != hdr.dofh_loadsz) {
12300 		kmem_free(dof, hdr.dofh_loadsz);
12301 		*errp = EFAULT;
12302 		return (NULL);
12303 	}
12304 
12305 	return (dof);
12306 }
12307 
12308 static dof_hdr_t *
dtrace_dof_property(const char * name)12309 dtrace_dof_property(const char *name)
12310 {
12311 	uchar_t *buf;
12312 	uint64_t loadsz;
12313 	unsigned int len, i;
12314 	dof_hdr_t *dof;
12315 
12316 	/*
12317 	 * Unfortunately, array of values in .conf files are always (and
12318 	 * only) interpreted to be integer arrays.  We must read our DOF
12319 	 * as an integer array, and then squeeze it into a byte array.
12320 	 */
12321 	if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0,
12322 	    (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS)
12323 		return (NULL);
12324 
12325 	for (i = 0; i < len; i++)
12326 		buf[i] = (uchar_t)(((int *)buf)[i]);
12327 
12328 	if (len < sizeof (dof_hdr_t)) {
12329 		ddi_prop_free(buf);
12330 		dtrace_dof_error(NULL, "truncated header");
12331 		return (NULL);
12332 	}
12333 
12334 	if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) {
12335 		ddi_prop_free(buf);
12336 		dtrace_dof_error(NULL, "truncated DOF");
12337 		return (NULL);
12338 	}
12339 
12340 	if (loadsz >= dtrace_dof_maxsize) {
12341 		ddi_prop_free(buf);
12342 		dtrace_dof_error(NULL, "oversized DOF");
12343 		return (NULL);
12344 	}
12345 
12346 	dof = kmem_alloc(loadsz, KM_SLEEP);
12347 	bcopy(buf, dof, loadsz);
12348 	ddi_prop_free(buf);
12349 
12350 	return (dof);
12351 }
12352 
12353 static void
dtrace_dof_destroy(dof_hdr_t * dof)12354 dtrace_dof_destroy(dof_hdr_t *dof)
12355 {
12356 	kmem_free(dof, dof->dofh_loadsz);
12357 }
12358 
12359 /*
12360  * Return the dof_sec_t pointer corresponding to a given section index.  If the
12361  * index is not valid, dtrace_dof_error() is called and NULL is returned.  If
12362  * a type other than DOF_SECT_NONE is specified, the header is checked against
12363  * this type and NULL is returned if the types do not match.
12364  */
12365 static dof_sec_t *
dtrace_dof_sect(dof_hdr_t * dof,uint32_t type,dof_secidx_t i)12366 dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i)
12367 {
12368 	dof_sec_t *sec = (dof_sec_t *)(uintptr_t)
12369 	    ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize);
12370 
12371 	if (i >= dof->dofh_secnum) {
12372 		dtrace_dof_error(dof, "referenced section index is invalid");
12373 		return (NULL);
12374 	}
12375 
12376 	if (!(sec->dofs_flags & DOF_SECF_LOAD)) {
12377 		dtrace_dof_error(dof, "referenced section is not loadable");
12378 		return (NULL);
12379 	}
12380 
12381 	if (type != DOF_SECT_NONE && type != sec->dofs_type) {
12382 		dtrace_dof_error(dof, "referenced section is the wrong type");
12383 		return (NULL);
12384 	}
12385 
12386 	return (sec);
12387 }
12388 
12389 static dtrace_probedesc_t *
dtrace_dof_probedesc(dof_hdr_t * dof,dof_sec_t * sec,dtrace_probedesc_t * desc)12390 dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc)
12391 {
12392 	dof_probedesc_t *probe;
12393 	dof_sec_t *strtab;
12394 	uintptr_t daddr = (uintptr_t)dof;
12395 	uintptr_t str;
12396 	size_t size;
12397 
12398 	if (sec->dofs_type != DOF_SECT_PROBEDESC) {
12399 		dtrace_dof_error(dof, "invalid probe section");
12400 		return (NULL);
12401 	}
12402 
12403 	if (sec->dofs_align != sizeof (dof_secidx_t)) {
12404 		dtrace_dof_error(dof, "bad alignment in probe description");
12405 		return (NULL);
12406 	}
12407 
12408 	if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) {
12409 		dtrace_dof_error(dof, "truncated probe description");
12410 		return (NULL);
12411 	}
12412 
12413 	probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset);
12414 	strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab);
12415 
12416 	if (strtab == NULL)
12417 		return (NULL);
12418 
12419 	str = daddr + strtab->dofs_offset;
12420 	size = strtab->dofs_size;
12421 
12422 	if (probe->dofp_provider >= strtab->dofs_size) {
12423 		dtrace_dof_error(dof, "corrupt probe provider");
12424 		return (NULL);
12425 	}
12426 
12427 	(void) strncpy(desc->dtpd_provider,
12428 	    (char *)(str + probe->dofp_provider),
12429 	    MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider));
12430 
12431 	if (probe->dofp_mod >= strtab->dofs_size) {
12432 		dtrace_dof_error(dof, "corrupt probe module");
12433 		return (NULL);
12434 	}
12435 
12436 	(void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod),
12437 	    MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod));
12438 
12439 	if (probe->dofp_func >= strtab->dofs_size) {
12440 		dtrace_dof_error(dof, "corrupt probe function");
12441 		return (NULL);
12442 	}
12443 
12444 	(void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func),
12445 	    MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func));
12446 
12447 	if (probe->dofp_name >= strtab->dofs_size) {
12448 		dtrace_dof_error(dof, "corrupt probe name");
12449 		return (NULL);
12450 	}
12451 
12452 	(void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name),
12453 	    MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name));
12454 
12455 	return (desc);
12456 }
12457 
12458 static dtrace_difo_t *
dtrace_dof_difo(dof_hdr_t * dof,dof_sec_t * sec,dtrace_vstate_t * vstate,cred_t * cr)12459 dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12460     cred_t *cr)
12461 {
12462 	dtrace_difo_t *dp;
12463 	size_t ttl = 0;
12464 	dof_difohdr_t *dofd;
12465 	uintptr_t daddr = (uintptr_t)dof;
12466 	size_t max = dtrace_difo_maxsize;
12467 	int i, l, n;
12468 
12469 	static const struct {
12470 		int section;
12471 		int bufoffs;
12472 		int lenoffs;
12473 		int entsize;
12474 		int align;
12475 		const char *msg;
12476 	} difo[] = {
12477 		{ DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf),
12478 		offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t),
12479 		sizeof (dif_instr_t), "multiple DIF sections" },
12480 
12481 		{ DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab),
12482 		offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t),
12483 		sizeof (uint64_t), "multiple integer tables" },
12484 
12485 		{ DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab),
12486 		offsetof(dtrace_difo_t, dtdo_strlen), 0,
12487 		sizeof (char), "multiple string tables" },
12488 
12489 		{ DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab),
12490 		offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t),
12491 		sizeof (uint_t), "multiple variable tables" },
12492 
12493 		{ DOF_SECT_NONE, 0, 0, 0, NULL }
12494 	};
12495 
12496 	if (sec->dofs_type != DOF_SECT_DIFOHDR) {
12497 		dtrace_dof_error(dof, "invalid DIFO header section");
12498 		return (NULL);
12499 	}
12500 
12501 	if (sec->dofs_align != sizeof (dof_secidx_t)) {
12502 		dtrace_dof_error(dof, "bad alignment in DIFO header");
12503 		return (NULL);
12504 	}
12505 
12506 	if (sec->dofs_size < sizeof (dof_difohdr_t) ||
12507 	    sec->dofs_size % sizeof (dof_secidx_t)) {
12508 		dtrace_dof_error(dof, "bad size in DIFO header");
12509 		return (NULL);
12510 	}
12511 
12512 	dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
12513 	n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1;
12514 
12515 	dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
12516 	dp->dtdo_rtype = dofd->dofd_rtype;
12517 
12518 	for (l = 0; l < n; l++) {
12519 		dof_sec_t *subsec;
12520 		void **bufp;
12521 		uint32_t *lenp;
12522 
12523 		if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE,
12524 		    dofd->dofd_links[l])) == NULL)
12525 			goto err; /* invalid section link */
12526 
12527 		if (ttl + subsec->dofs_size > max) {
12528 			dtrace_dof_error(dof, "exceeds maximum size");
12529 			goto err;
12530 		}
12531 
12532 		ttl += subsec->dofs_size;
12533 
12534 		for (i = 0; difo[i].section != DOF_SECT_NONE; i++) {
12535 			if (subsec->dofs_type != difo[i].section)
12536 				continue;
12537 
12538 			if (!(subsec->dofs_flags & DOF_SECF_LOAD)) {
12539 				dtrace_dof_error(dof, "section not loaded");
12540 				goto err;
12541 			}
12542 
12543 			if (subsec->dofs_align != difo[i].align) {
12544 				dtrace_dof_error(dof, "bad alignment");
12545 				goto err;
12546 			}
12547 
12548 			bufp = (void **)((uintptr_t)dp + difo[i].bufoffs);
12549 			lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs);
12550 
12551 			if (*bufp != NULL) {
12552 				dtrace_dof_error(dof, difo[i].msg);
12553 				goto err;
12554 			}
12555 
12556 			if (difo[i].entsize != subsec->dofs_entsize) {
12557 				dtrace_dof_error(dof, "entry size mismatch");
12558 				goto err;
12559 			}
12560 
12561 			if (subsec->dofs_entsize != 0 &&
12562 			    (subsec->dofs_size % subsec->dofs_entsize) != 0) {
12563 				dtrace_dof_error(dof, "corrupt entry size");
12564 				goto err;
12565 			}
12566 
12567 			*lenp = subsec->dofs_size;
12568 			*bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP);
12569 			bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset),
12570 			    *bufp, subsec->dofs_size);
12571 
12572 			if (subsec->dofs_entsize != 0)
12573 				*lenp /= subsec->dofs_entsize;
12574 
12575 			break;
12576 		}
12577 
12578 		/*
12579 		 * If we encounter a loadable DIFO sub-section that is not
12580 		 * known to us, assume this is a broken program and fail.
12581 		 */
12582 		if (difo[i].section == DOF_SECT_NONE &&
12583 		    (subsec->dofs_flags & DOF_SECF_LOAD)) {
12584 			dtrace_dof_error(dof, "unrecognized DIFO subsection");
12585 			goto err;
12586 		}
12587 	}
12588 
12589 	if (dp->dtdo_buf == NULL) {
12590 		/*
12591 		 * We can't have a DIF object without DIF text.
12592 		 */
12593 		dtrace_dof_error(dof, "missing DIF text");
12594 		goto err;
12595 	}
12596 
12597 	/*
12598 	 * Before we validate the DIF object, run through the variable table
12599 	 * looking for the strings -- if any of their size are under, we'll set
12600 	 * their size to be the system-wide default string size.  Note that
12601 	 * this should _not_ happen if the "strsize" option has been set --
12602 	 * in this case, the compiler should have set the size to reflect the
12603 	 * setting of the option.
12604 	 */
12605 	for (i = 0; i < dp->dtdo_varlen; i++) {
12606 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
12607 		dtrace_diftype_t *t = &v->dtdv_type;
12608 
12609 		if (v->dtdv_id < DIF_VAR_OTHER_UBASE)
12610 			continue;
12611 
12612 		if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0)
12613 			t->dtdt_size = dtrace_strsize_default;
12614 	}
12615 
12616 	if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0)
12617 		goto err;
12618 
12619 	dtrace_difo_init(dp, vstate);
12620 	return (dp);
12621 
12622 err:
12623 	kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
12624 	kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
12625 	kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
12626 	kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
12627 
12628 	kmem_free(dp, sizeof (dtrace_difo_t));
12629 	return (NULL);
12630 }
12631 
12632 static dtrace_predicate_t *
dtrace_dof_predicate(dof_hdr_t * dof,dof_sec_t * sec,dtrace_vstate_t * vstate,cred_t * cr)12633 dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12634     cred_t *cr)
12635 {
12636 	dtrace_difo_t *dp;
12637 
12638 	if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL)
12639 		return (NULL);
12640 
12641 	return (dtrace_predicate_create(dp));
12642 }
12643 
12644 static dtrace_actdesc_t *
dtrace_dof_actdesc(dof_hdr_t * dof,dof_sec_t * sec,dtrace_vstate_t * vstate,cred_t * cr)12645 dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12646     cred_t *cr)
12647 {
12648 	dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next;
12649 	dof_actdesc_t *desc;
12650 	dof_sec_t *difosec;
12651 	size_t offs;
12652 	uintptr_t daddr = (uintptr_t)dof;
12653 	uint64_t arg;
12654 	dtrace_actkind_t kind;
12655 
12656 	if (sec->dofs_type != DOF_SECT_ACTDESC) {
12657 		dtrace_dof_error(dof, "invalid action section");
12658 		return (NULL);
12659 	}
12660 
12661 	if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) {
12662 		dtrace_dof_error(dof, "truncated action description");
12663 		return (NULL);
12664 	}
12665 
12666 	if (sec->dofs_align != sizeof (uint64_t)) {
12667 		dtrace_dof_error(dof, "bad alignment in action description");
12668 		return (NULL);
12669 	}
12670 
12671 	if (sec->dofs_size < sec->dofs_entsize) {
12672 		dtrace_dof_error(dof, "section entry size exceeds total size");
12673 		return (NULL);
12674 	}
12675 
12676 	if (sec->dofs_entsize != sizeof (dof_actdesc_t)) {
12677 		dtrace_dof_error(dof, "bad entry size in action description");
12678 		return (NULL);
12679 	}
12680 
12681 	if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) {
12682 		dtrace_dof_error(dof, "actions exceed dtrace_actions_max");
12683 		return (NULL);
12684 	}
12685 
12686 	for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) {
12687 		desc = (dof_actdesc_t *)(daddr +
12688 		    (uintptr_t)sec->dofs_offset + offs);
12689 		kind = (dtrace_actkind_t)desc->dofa_kind;
12690 
12691 		if ((DTRACEACT_ISPRINTFLIKE(kind) &&
12692 		    (kind != DTRACEACT_PRINTA ||
12693 		    desc->dofa_strtab != DOF_SECIDX_NONE)) ||
12694 		    (kind == DTRACEACT_DIFEXPR &&
12695 		    desc->dofa_strtab != DOF_SECIDX_NONE)) {
12696 			dof_sec_t *strtab;
12697 			char *str, *fmt;
12698 			uint64_t i;
12699 
12700 			/*
12701 			 * The argument to these actions is an index into the
12702 			 * DOF string table.  For printf()-like actions, this
12703 			 * is the format string.  For print(), this is the
12704 			 * CTF type of the expression result.
12705 			 */
12706 			if ((strtab = dtrace_dof_sect(dof,
12707 			    DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL)
12708 				goto err;
12709 
12710 			str = (char *)((uintptr_t)dof +
12711 			    (uintptr_t)strtab->dofs_offset);
12712 
12713 			for (i = desc->dofa_arg; i < strtab->dofs_size; i++) {
12714 				if (str[i] == '\0')
12715 					break;
12716 			}
12717 
12718 			if (i >= strtab->dofs_size) {
12719 				dtrace_dof_error(dof, "bogus format string");
12720 				goto err;
12721 			}
12722 
12723 			if (i == desc->dofa_arg) {
12724 				dtrace_dof_error(dof, "empty format string");
12725 				goto err;
12726 			}
12727 
12728 			i -= desc->dofa_arg;
12729 			fmt = kmem_alloc(i + 1, KM_SLEEP);
12730 			bcopy(&str[desc->dofa_arg], fmt, i + 1);
12731 			arg = (uint64_t)(uintptr_t)fmt;
12732 		} else {
12733 			if (kind == DTRACEACT_PRINTA) {
12734 				ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE);
12735 				arg = 0;
12736 			} else {
12737 				arg = desc->dofa_arg;
12738 			}
12739 		}
12740 
12741 		act = dtrace_actdesc_create(kind, desc->dofa_ntuple,
12742 		    desc->dofa_uarg, arg);
12743 
12744 		if (last != NULL) {
12745 			last->dtad_next = act;
12746 		} else {
12747 			first = act;
12748 		}
12749 
12750 		last = act;
12751 
12752 		if (desc->dofa_difo == DOF_SECIDX_NONE)
12753 			continue;
12754 
12755 		if ((difosec = dtrace_dof_sect(dof,
12756 		    DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL)
12757 			goto err;
12758 
12759 		act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr);
12760 
12761 		if (act->dtad_difo == NULL)
12762 			goto err;
12763 	}
12764 
12765 	ASSERT(first != NULL);
12766 	return (first);
12767 
12768 err:
12769 	for (act = first; act != NULL; act = next) {
12770 		next = act->dtad_next;
12771 		dtrace_actdesc_release(act, vstate);
12772 	}
12773 
12774 	return (NULL);
12775 }
12776 
12777 static dtrace_ecbdesc_t *
dtrace_dof_ecbdesc(dof_hdr_t * dof,dof_sec_t * sec,dtrace_vstate_t * vstate,cred_t * cr)12778 dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12779     cred_t *cr)
12780 {
12781 	dtrace_ecbdesc_t *ep;
12782 	dof_ecbdesc_t *ecb;
12783 	dtrace_probedesc_t *desc;
12784 	dtrace_predicate_t *pred = NULL;
12785 
12786 	if (sec->dofs_size < sizeof (dof_ecbdesc_t)) {
12787 		dtrace_dof_error(dof, "truncated ECB description");
12788 		return (NULL);
12789 	}
12790 
12791 	if (sec->dofs_align != sizeof (uint64_t)) {
12792 		dtrace_dof_error(dof, "bad alignment in ECB description");
12793 		return (NULL);
12794 	}
12795 
12796 	ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset);
12797 	sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes);
12798 
12799 	if (sec == NULL)
12800 		return (NULL);
12801 
12802 	ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
12803 	ep->dted_uarg = ecb->dofe_uarg;
12804 	desc = &ep->dted_probe;
12805 
12806 	if (dtrace_dof_probedesc(dof, sec, desc) == NULL)
12807 		goto err;
12808 
12809 	if (ecb->dofe_pred != DOF_SECIDX_NONE) {
12810 		if ((sec = dtrace_dof_sect(dof,
12811 		    DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL)
12812 			goto err;
12813 
12814 		if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL)
12815 			goto err;
12816 
12817 		ep->dted_pred.dtpdd_predicate = pred;
12818 	}
12819 
12820 	if (ecb->dofe_actions != DOF_SECIDX_NONE) {
12821 		if ((sec = dtrace_dof_sect(dof,
12822 		    DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL)
12823 			goto err;
12824 
12825 		ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr);
12826 
12827 		if (ep->dted_action == NULL)
12828 			goto err;
12829 	}
12830 
12831 	return (ep);
12832 
12833 err:
12834 	if (pred != NULL)
12835 		dtrace_predicate_release(pred, vstate);
12836 	kmem_free(ep, sizeof (dtrace_ecbdesc_t));
12837 	return (NULL);
12838 }
12839 
12840 /*
12841  * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the
12842  * specified DOF.  At present, this amounts to simply adding 'ubase' to the
12843  * site of any user SETX relocations to account for load object base address.
12844  * In the future, if we need other relocations, this function can be extended.
12845  */
12846 static int
dtrace_dof_relocate(dof_hdr_t * dof,dof_sec_t * sec,uint64_t ubase)12847 dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
12848 {
12849 	uintptr_t daddr = (uintptr_t)dof;
12850 	dof_relohdr_t *dofr =
12851 	    (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
12852 	dof_sec_t *ss, *rs, *ts;
12853 	dof_relodesc_t *r;
12854 	uint_t i, n;
12855 
12856 	if (sec->dofs_size < sizeof (dof_relohdr_t) ||
12857 	    sec->dofs_align != sizeof (dof_secidx_t)) {
12858 		dtrace_dof_error(dof, "invalid relocation header");
12859 		return (-1);
12860 	}
12861 
12862 	ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab);
12863 	rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec);
12864 	ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec);
12865 
12866 	if (ss == NULL || rs == NULL || ts == NULL)
12867 		return (-1); /* dtrace_dof_error() has been called already */
12868 
12869 	if (rs->dofs_entsize < sizeof (dof_relodesc_t) ||
12870 	    rs->dofs_align != sizeof (uint64_t)) {
12871 		dtrace_dof_error(dof, "invalid relocation section");
12872 		return (-1);
12873 	}
12874 
12875 	r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset);
12876 	n = rs->dofs_size / rs->dofs_entsize;
12877 
12878 	for (i = 0; i < n; i++) {
12879 		uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset;
12880 
12881 		switch (r->dofr_type) {
12882 		case DOF_RELO_NONE:
12883 			break;
12884 		case DOF_RELO_SETX:
12885 			if (r->dofr_offset >= ts->dofs_size || r->dofr_offset +
12886 			    sizeof (uint64_t) > ts->dofs_size) {
12887 				dtrace_dof_error(dof, "bad relocation offset");
12888 				return (-1);
12889 			}
12890 
12891 			if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) {
12892 				dtrace_dof_error(dof, "misaligned setx relo");
12893 				return (-1);
12894 			}
12895 
12896 			*(uint64_t *)taddr += ubase;
12897 			break;
12898 		default:
12899 			dtrace_dof_error(dof, "invalid relocation type");
12900 			return (-1);
12901 		}
12902 
12903 		r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize);
12904 	}
12905 
12906 	return (0);
12907 }
12908 
12909 /*
12910  * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated
12911  * header:  it should be at the front of a memory region that is at least
12912  * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in
12913  * size.  It need not be validated in any other way.
12914  */
12915 static int
dtrace_dof_slurp(dof_hdr_t * dof,dtrace_vstate_t * vstate,cred_t * cr,dtrace_enabling_t ** enabp,uint64_t ubase,int noprobes)12916 dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr,
12917     dtrace_enabling_t **enabp, uint64_t ubase, int noprobes)
12918 {
12919 	uint64_t len = dof->dofh_loadsz, seclen;
12920 	uintptr_t daddr = (uintptr_t)dof;
12921 	dtrace_ecbdesc_t *ep;
12922 	dtrace_enabling_t *enab;
12923 	uint_t i;
12924 
12925 	ASSERT(MUTEX_HELD(&dtrace_lock));
12926 	ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t));
12927 
12928 	/*
12929 	 * Check the DOF header identification bytes.  In addition to checking
12930 	 * valid settings, we also verify that unused bits/bytes are zeroed so
12931 	 * we can use them later without fear of regressing existing binaries.
12932 	 */
12933 	if (bcmp(&dof->dofh_ident[DOF_ID_MAG0],
12934 	    DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) {
12935 		dtrace_dof_error(dof, "DOF magic string mismatch");
12936 		return (-1);
12937 	}
12938 
12939 	if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 &&
12940 	    dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) {
12941 		dtrace_dof_error(dof, "DOF has invalid data model");
12942 		return (-1);
12943 	}
12944 
12945 	if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) {
12946 		dtrace_dof_error(dof, "DOF encoding mismatch");
12947 		return (-1);
12948 	}
12949 
12950 	if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
12951 	    dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) {
12952 		dtrace_dof_error(dof, "DOF version mismatch");
12953 		return (-1);
12954 	}
12955 
12956 	if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) {
12957 		dtrace_dof_error(dof, "DOF uses unsupported instruction set");
12958 		return (-1);
12959 	}
12960 
12961 	if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) {
12962 		dtrace_dof_error(dof, "DOF uses too many integer registers");
12963 		return (-1);
12964 	}
12965 
12966 	if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) {
12967 		dtrace_dof_error(dof, "DOF uses too many tuple registers");
12968 		return (-1);
12969 	}
12970 
12971 	for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) {
12972 		if (dof->dofh_ident[i] != 0) {
12973 			dtrace_dof_error(dof, "DOF has invalid ident byte set");
12974 			return (-1);
12975 		}
12976 	}
12977 
12978 	if (dof->dofh_flags & ~DOF_FL_VALID) {
12979 		dtrace_dof_error(dof, "DOF has invalid flag bits set");
12980 		return (-1);
12981 	}
12982 
12983 	if (dof->dofh_secsize == 0) {
12984 		dtrace_dof_error(dof, "zero section header size");
12985 		return (-1);
12986 	}
12987 
12988 	/*
12989 	 * Check that the section headers don't exceed the amount of DOF
12990 	 * data.  Note that we cast the section size and number of sections
12991 	 * to uint64_t's to prevent possible overflow in the multiplication.
12992 	 */
12993 	seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize;
12994 
12995 	if (dof->dofh_secoff > len || seclen > len ||
12996 	    dof->dofh_secoff + seclen > len) {
12997 		dtrace_dof_error(dof, "truncated section headers");
12998 		return (-1);
12999 	}
13000 
13001 	if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) {
13002 		dtrace_dof_error(dof, "misaligned section headers");
13003 		return (-1);
13004 	}
13005 
13006 	if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) {
13007 		dtrace_dof_error(dof, "misaligned section size");
13008 		return (-1);
13009 	}
13010 
13011 	/*
13012 	 * Take an initial pass through the section headers to be sure that
13013 	 * the headers don't have stray offsets.  If the 'noprobes' flag is
13014 	 * set, do not permit sections relating to providers, probes, or args.
13015 	 */
13016 	for (i = 0; i < dof->dofh_secnum; i++) {
13017 		dof_sec_t *sec = (dof_sec_t *)(daddr +
13018 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
13019 
13020 		if (noprobes) {
13021 			switch (sec->dofs_type) {
13022 			case DOF_SECT_PROVIDER:
13023 			case DOF_SECT_PROBES:
13024 			case DOF_SECT_PRARGS:
13025 			case DOF_SECT_PROFFS:
13026 				dtrace_dof_error(dof, "illegal sections "
13027 				    "for enabling");
13028 				return (-1);
13029 			}
13030 		}
13031 
13032 		if (DOF_SEC_ISLOADABLE(sec->dofs_type) &&
13033 		    !(sec->dofs_flags & DOF_SECF_LOAD)) {
13034 			dtrace_dof_error(dof, "loadable section with load "
13035 			    "flag unset");
13036 			return (-1);
13037 		}
13038 
13039 		if (!(sec->dofs_flags & DOF_SECF_LOAD))
13040 			continue; /* just ignore non-loadable sections */
13041 
13042 		if (!ISP2(sec->dofs_align)) {
13043 			dtrace_dof_error(dof, "bad section alignment");
13044 			return (-1);
13045 		}
13046 
13047 		if (sec->dofs_offset & (sec->dofs_align - 1)) {
13048 			dtrace_dof_error(dof, "misaligned section");
13049 			return (-1);
13050 		}
13051 
13052 		if (sec->dofs_offset > len || sec->dofs_size > len ||
13053 		    sec->dofs_offset + sec->dofs_size > len) {
13054 			dtrace_dof_error(dof, "corrupt section header");
13055 			return (-1);
13056 		}
13057 
13058 		if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr +
13059 		    sec->dofs_offset + sec->dofs_size - 1) != '\0') {
13060 			dtrace_dof_error(dof, "non-terminating string table");
13061 			return (-1);
13062 		}
13063 	}
13064 
13065 	/*
13066 	 * Take a second pass through the sections and locate and perform any
13067 	 * relocations that are present.  We do this after the first pass to
13068 	 * be sure that all sections have had their headers validated.
13069 	 */
13070 	for (i = 0; i < dof->dofh_secnum; i++) {
13071 		dof_sec_t *sec = (dof_sec_t *)(daddr +
13072 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
13073 
13074 		if (!(sec->dofs_flags & DOF_SECF_LOAD))
13075 			continue; /* skip sections that are not loadable */
13076 
13077 		switch (sec->dofs_type) {
13078 		case DOF_SECT_URELHDR:
13079 			if (dtrace_dof_relocate(dof, sec, ubase) != 0)
13080 				return (-1);
13081 			break;
13082 		}
13083 	}
13084 
13085 	if ((enab = *enabp) == NULL)
13086 		enab = *enabp = dtrace_enabling_create(vstate);
13087 
13088 	for (i = 0; i < dof->dofh_secnum; i++) {
13089 		dof_sec_t *sec = (dof_sec_t *)(daddr +
13090 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
13091 
13092 		if (sec->dofs_type != DOF_SECT_ECBDESC)
13093 			continue;
13094 
13095 		if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) {
13096 			dtrace_enabling_destroy(enab);
13097 			*enabp = NULL;
13098 			return (-1);
13099 		}
13100 
13101 		dtrace_enabling_add(enab, ep);
13102 	}
13103 
13104 	return (0);
13105 }
13106 
13107 /*
13108  * Process DOF for any options.  This routine assumes that the DOF has been
13109  * at least processed by dtrace_dof_slurp().
13110  */
13111 static int
dtrace_dof_options(dof_hdr_t * dof,dtrace_state_t * state)13112 dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state)
13113 {
13114 	int i, rval;
13115 	uint32_t entsize;
13116 	size_t offs;
13117 	dof_optdesc_t *desc;
13118 
13119 	for (i = 0; i < dof->dofh_secnum; i++) {
13120 		dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof +
13121 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
13122 
13123 		if (sec->dofs_type != DOF_SECT_OPTDESC)
13124 			continue;
13125 
13126 		if (sec->dofs_align != sizeof (uint64_t)) {
13127 			dtrace_dof_error(dof, "bad alignment in "
13128 			    "option description");
13129 			return (EINVAL);
13130 		}
13131 
13132 		if ((entsize = sec->dofs_entsize) == 0) {
13133 			dtrace_dof_error(dof, "zeroed option entry size");
13134 			return (EINVAL);
13135 		}
13136 
13137 		if (entsize < sizeof (dof_optdesc_t)) {
13138 			dtrace_dof_error(dof, "bad option entry size");
13139 			return (EINVAL);
13140 		}
13141 
13142 		for (offs = 0; offs < sec->dofs_size; offs += entsize) {
13143 			desc = (dof_optdesc_t *)((uintptr_t)dof +
13144 			    (uintptr_t)sec->dofs_offset + offs);
13145 
13146 			if (desc->dofo_strtab != DOF_SECIDX_NONE) {
13147 				dtrace_dof_error(dof, "non-zero option string");
13148 				return (EINVAL);
13149 			}
13150 
13151 			if (desc->dofo_value == DTRACEOPT_UNSET) {
13152 				dtrace_dof_error(dof, "unset option");
13153 				return (EINVAL);
13154 			}
13155 
13156 			if ((rval = dtrace_state_option(state,
13157 			    desc->dofo_option, desc->dofo_value)) != 0) {
13158 				dtrace_dof_error(dof, "rejected option");
13159 				return (rval);
13160 			}
13161 		}
13162 	}
13163 
13164 	return (0);
13165 }
13166 
13167 /*
13168  * DTrace Consumer State Functions
13169  */
13170 int
dtrace_dstate_init(dtrace_dstate_t * dstate,size_t size)13171 dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size)
13172 {
13173 	size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize;
13174 	void *base;
13175 	uintptr_t limit;
13176 	dtrace_dynvar_t *dvar, *next, *start;
13177 	int i;
13178 
13179 	ASSERT(MUTEX_HELD(&dtrace_lock));
13180 	ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL);
13181 
13182 	bzero(dstate, sizeof (dtrace_dstate_t));
13183 
13184 	if ((dstate->dtds_chunksize = chunksize) == 0)
13185 		dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE;
13186 
13187 	if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)))
13188 		size = min;
13189 
13190 	if ((base = kmem_zalloc(size, KM_NOSLEEP | KM_NORMALPRI)) == NULL)
13191 		return (ENOMEM);
13192 
13193 	dstate->dtds_size = size;
13194 	dstate->dtds_base = base;
13195 	dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP);
13196 	bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t));
13197 
13198 	hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t));
13199 
13200 	if (hashsize != 1 && (hashsize & 1))
13201 		hashsize--;
13202 
13203 	dstate->dtds_hashsize = hashsize;
13204 	dstate->dtds_hash = dstate->dtds_base;
13205 
13206 	/*
13207 	 * Set all of our hash buckets to point to the single sink, and (if
13208 	 * it hasn't already been set), set the sink's hash value to be the
13209 	 * sink sentinel value.  The sink is needed for dynamic variable
13210 	 * lookups to know that they have iterated over an entire, valid hash
13211 	 * chain.
13212 	 */
13213 	for (i = 0; i < hashsize; i++)
13214 		dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink;
13215 
13216 	if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK)
13217 		dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK;
13218 
13219 	/*
13220 	 * Determine number of active CPUs.  Divide free list evenly among
13221 	 * active CPUs.
13222 	 */
13223 	start = (dtrace_dynvar_t *)
13224 	    ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t));
13225 	limit = (uintptr_t)base + size;
13226 
13227 	maxper = (limit - (uintptr_t)start) / NCPU;
13228 	maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize;
13229 
13230 	for (i = 0; i < NCPU; i++) {
13231 		dstate->dtds_percpu[i].dtdsc_free = dvar = start;
13232 
13233 		/*
13234 		 * If we don't even have enough chunks to make it once through
13235 		 * NCPUs, we're just going to allocate everything to the first
13236 		 * CPU.  And if we're on the last CPU, we're going to allocate
13237 		 * whatever is left over.  In either case, we set the limit to
13238 		 * be the limit of the dynamic variable space.
13239 		 */
13240 		if (maxper == 0 || i == NCPU - 1) {
13241 			limit = (uintptr_t)base + size;
13242 			start = NULL;
13243 		} else {
13244 			limit = (uintptr_t)start + maxper;
13245 			start = (dtrace_dynvar_t *)limit;
13246 		}
13247 
13248 		ASSERT(limit <= (uintptr_t)base + size);
13249 
13250 		for (;;) {
13251 			next = (dtrace_dynvar_t *)((uintptr_t)dvar +
13252 			    dstate->dtds_chunksize);
13253 
13254 			if ((uintptr_t)next + dstate->dtds_chunksize >= limit)
13255 				break;
13256 
13257 			dvar->dtdv_next = next;
13258 			dvar = next;
13259 		}
13260 
13261 		if (maxper == 0)
13262 			break;
13263 	}
13264 
13265 	return (0);
13266 }
13267 
13268 void
dtrace_dstate_fini(dtrace_dstate_t * dstate)13269 dtrace_dstate_fini(dtrace_dstate_t *dstate)
13270 {
13271 	ASSERT(MUTEX_HELD(&cpu_lock));
13272 
13273 	if (dstate->dtds_base == NULL)
13274 		return;
13275 
13276 	kmem_free(dstate->dtds_base, dstate->dtds_size);
13277 	kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu);
13278 }
13279 
13280 static void
dtrace_vstate_fini(dtrace_vstate_t * vstate)13281 dtrace_vstate_fini(dtrace_vstate_t *vstate)
13282 {
13283 	/*
13284 	 * Logical XOR, where are you?
13285 	 */
13286 	ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL));
13287 
13288 	if (vstate->dtvs_nglobals > 0) {
13289 		kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals *
13290 		    sizeof (dtrace_statvar_t *));
13291 	}
13292 
13293 	if (vstate->dtvs_ntlocals > 0) {
13294 		kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals *
13295 		    sizeof (dtrace_difv_t));
13296 	}
13297 
13298 	ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL));
13299 
13300 	if (vstate->dtvs_nlocals > 0) {
13301 		kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals *
13302 		    sizeof (dtrace_statvar_t *));
13303 	}
13304 }
13305 
13306 static void
dtrace_state_clean(dtrace_state_t * state)13307 dtrace_state_clean(dtrace_state_t *state)
13308 {
13309 	if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE)
13310 		return;
13311 
13312 	dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars);
13313 	dtrace_speculation_clean(state);
13314 }
13315 
13316 static void
dtrace_state_deadman(dtrace_state_t * state)13317 dtrace_state_deadman(dtrace_state_t *state)
13318 {
13319 	hrtime_t now;
13320 
13321 	dtrace_sync();
13322 
13323 	now = dtrace_gethrtime();
13324 
13325 	if (state != dtrace_anon.dta_state &&
13326 	    now - state->dts_laststatus >= dtrace_deadman_user)
13327 		return;
13328 
13329 	/*
13330 	 * We must be sure that dts_alive never appears to be less than the
13331 	 * value upon entry to dtrace_state_deadman(), and because we lack a
13332 	 * dtrace_cas64(), we cannot store to it atomically.  We thus instead
13333 	 * store INT64_MAX to it, followed by a memory barrier, followed by
13334 	 * the new value.  This assures that dts_alive never appears to be
13335 	 * less than its true value, regardless of the order in which the
13336 	 * stores to the underlying storage are issued.
13337 	 */
13338 	state->dts_alive = INT64_MAX;
13339 	dtrace_membar_producer();
13340 	state->dts_alive = now;
13341 }
13342 
13343 dtrace_state_t *
dtrace_state_create(dev_t * devp,cred_t * cr)13344 dtrace_state_create(dev_t *devp, cred_t *cr)
13345 {
13346 	minor_t minor;
13347 	major_t major;
13348 	char c[30];
13349 	dtrace_state_t *state;
13350 	dtrace_optval_t *opt;
13351 	int bufsize = NCPU * sizeof (dtrace_buffer_t), i;
13352 
13353 	ASSERT(MUTEX_HELD(&dtrace_lock));
13354 	ASSERT(MUTEX_HELD(&cpu_lock));
13355 
13356 	minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1,
13357 	    VM_BESTFIT | VM_SLEEP);
13358 
13359 	if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) {
13360 		vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
13361 		return (NULL);
13362 	}
13363 
13364 	state = ddi_get_soft_state(dtrace_softstate, minor);
13365 	state->dts_epid = DTRACE_EPIDNONE + 1;
13366 
13367 	(void) snprintf(c, sizeof (c), "dtrace_aggid_%d", minor);
13368 	state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1,
13369 	    NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
13370 
13371 	if (devp != NULL) {
13372 		major = getemajor(*devp);
13373 	} else {
13374 		major = ddi_driver_major(dtrace_devi);
13375 	}
13376 
13377 	state->dts_dev = makedevice(major, minor);
13378 
13379 	if (devp != NULL)
13380 		*devp = state->dts_dev;
13381 
13382 	/*
13383 	 * We allocate NCPU buffers.  On the one hand, this can be quite
13384 	 * a bit of memory per instance (nearly 36K on a Starcat).  On the
13385 	 * other hand, it saves an additional memory reference in the probe
13386 	 * path.
13387 	 */
13388 	state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP);
13389 	state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP);
13390 	state->dts_cleaner = CYCLIC_NONE;
13391 	state->dts_deadman = CYCLIC_NONE;
13392 	state->dts_vstate.dtvs_state = state;
13393 
13394 	for (i = 0; i < DTRACEOPT_MAX; i++)
13395 		state->dts_options[i] = DTRACEOPT_UNSET;
13396 
13397 	/*
13398 	 * Set the default options.
13399 	 */
13400 	opt = state->dts_options;
13401 	opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH;
13402 	opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO;
13403 	opt[DTRACEOPT_NSPEC] = dtrace_nspec_default;
13404 	opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default;
13405 	opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL;
13406 	opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default;
13407 	opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default;
13408 	opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default;
13409 	opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default;
13410 	opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default;
13411 	opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default;
13412 	opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default;
13413 	opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default;
13414 	opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default;
13415 
13416 	state->dts_activity = DTRACE_ACTIVITY_INACTIVE;
13417 
13418 	/*
13419 	 * Depending on the user credentials, we set flag bits which alter probe
13420 	 * visibility or the amount of destructiveness allowed.  In the case of
13421 	 * actual anonymous tracing, or the possession of all privileges, all of
13422 	 * the normal checks are bypassed.
13423 	 */
13424 	if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
13425 		state->dts_cred.dcr_visible = DTRACE_CRV_ALL;
13426 		state->dts_cred.dcr_action = DTRACE_CRA_ALL;
13427 	} else {
13428 		/*
13429 		 * Set up the credentials for this instantiation.  We take a
13430 		 * hold on the credential to prevent it from disappearing on
13431 		 * us; this in turn prevents the zone_t referenced by this
13432 		 * credential from disappearing.  This means that we can
13433 		 * examine the credential and the zone from probe context.
13434 		 */
13435 		crhold(cr);
13436 		state->dts_cred.dcr_cred = cr;
13437 
13438 		/*
13439 		 * CRA_PROC means "we have *some* privilege for dtrace" and
13440 		 * unlocks the use of variables like pid, zonename, etc.
13441 		 */
13442 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) ||
13443 		    PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
13444 			state->dts_cred.dcr_action |= DTRACE_CRA_PROC;
13445 		}
13446 
13447 		/*
13448 		 * dtrace_user allows use of syscall and profile providers.
13449 		 * If the user also has proc_owner and/or proc_zone, we
13450 		 * extend the scope to include additional visibility and
13451 		 * destructive power.
13452 		 */
13453 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) {
13454 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) {
13455 				state->dts_cred.dcr_visible |=
13456 				    DTRACE_CRV_ALLPROC;
13457 
13458 				state->dts_cred.dcr_action |=
13459 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13460 			}
13461 
13462 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) {
13463 				state->dts_cred.dcr_visible |=
13464 				    DTRACE_CRV_ALLZONE;
13465 
13466 				state->dts_cred.dcr_action |=
13467 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13468 			}
13469 
13470 			/*
13471 			 * If we have all privs in whatever zone this is,
13472 			 * we can do destructive things to processes which
13473 			 * have altered credentials.
13474 			 */
13475 			if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
13476 			    cr->cr_zone->zone_privset)) {
13477 				state->dts_cred.dcr_action |=
13478 				    DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
13479 			}
13480 		}
13481 
13482 		/*
13483 		 * Holding the dtrace_kernel privilege also implies that
13484 		 * the user has the dtrace_user privilege from a visibility
13485 		 * perspective.  But without further privileges, some
13486 		 * destructive actions are not available.
13487 		 */
13488 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) {
13489 			/*
13490 			 * Make all probes in all zones visible.  However,
13491 			 * this doesn't mean that all actions become available
13492 			 * to all zones.
13493 			 */
13494 			state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL |
13495 			    DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE;
13496 
13497 			state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL |
13498 			    DTRACE_CRA_PROC;
13499 			/*
13500 			 * Holding proc_owner means that destructive actions
13501 			 * for *this* zone are allowed.
13502 			 */
13503 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
13504 				state->dts_cred.dcr_action |=
13505 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13506 
13507 			/*
13508 			 * Holding proc_zone means that destructive actions
13509 			 * for this user/group ID in all zones is allowed.
13510 			 */
13511 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
13512 				state->dts_cred.dcr_action |=
13513 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13514 
13515 			/*
13516 			 * If we have all privs in whatever zone this is,
13517 			 * we can do destructive things to processes which
13518 			 * have altered credentials.
13519 			 */
13520 			if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
13521 			    cr->cr_zone->zone_privset)) {
13522 				state->dts_cred.dcr_action |=
13523 				    DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
13524 			}
13525 		}
13526 
13527 		/*
13528 		 * Holding the dtrace_proc privilege gives control over fasttrap
13529 		 * and pid providers.  We need to grant wider destructive
13530 		 * privileges in the event that the user has proc_owner and/or
13531 		 * proc_zone.
13532 		 */
13533 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
13534 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
13535 				state->dts_cred.dcr_action |=
13536 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13537 
13538 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
13539 				state->dts_cred.dcr_action |=
13540 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13541 		}
13542 	}
13543 
13544 	return (state);
13545 }
13546 
13547 static int
dtrace_state_buffer(dtrace_state_t * state,dtrace_buffer_t * buf,int which)13548 dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which)
13549 {
13550 	dtrace_optval_t *opt = state->dts_options, size;
13551 	processorid_t cpu;
13552 	int flags = 0, rval, factor, divisor = 1;
13553 
13554 	ASSERT(MUTEX_HELD(&dtrace_lock));
13555 	ASSERT(MUTEX_HELD(&cpu_lock));
13556 	ASSERT(which < DTRACEOPT_MAX);
13557 	ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE ||
13558 	    (state == dtrace_anon.dta_state &&
13559 	    state->dts_activity == DTRACE_ACTIVITY_ACTIVE));
13560 
13561 	if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0)
13562 		return (0);
13563 
13564 	if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET)
13565 		cpu = opt[DTRACEOPT_CPU];
13566 
13567 	if (which == DTRACEOPT_SPECSIZE)
13568 		flags |= DTRACEBUF_NOSWITCH;
13569 
13570 	if (which == DTRACEOPT_BUFSIZE) {
13571 		if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING)
13572 			flags |= DTRACEBUF_RING;
13573 
13574 		if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL)
13575 			flags |= DTRACEBUF_FILL;
13576 
13577 		if (state != dtrace_anon.dta_state ||
13578 		    state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
13579 			flags |= DTRACEBUF_INACTIVE;
13580 	}
13581 
13582 	for (size = opt[which]; size >= sizeof (uint64_t); size /= divisor) {
13583 		/*
13584 		 * The size must be 8-byte aligned.  If the size is not 8-byte
13585 		 * aligned, drop it down by the difference.
13586 		 */
13587 		if (size & (sizeof (uint64_t) - 1))
13588 			size -= size & (sizeof (uint64_t) - 1);
13589 
13590 		if (size < state->dts_reserve) {
13591 			/*
13592 			 * Buffers always must be large enough to accommodate
13593 			 * their prereserved space.  We return E2BIG instead
13594 			 * of ENOMEM in this case to allow for user-level
13595 			 * software to differentiate the cases.
13596 			 */
13597 			return (E2BIG);
13598 		}
13599 
13600 		rval = dtrace_buffer_alloc(buf, size, flags, cpu, &factor);
13601 
13602 		if (rval != ENOMEM) {
13603 			opt[which] = size;
13604 			return (rval);
13605 		}
13606 
13607 		if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13608 			return (rval);
13609 
13610 		for (divisor = 2; divisor < factor; divisor <<= 1)
13611 			continue;
13612 	}
13613 
13614 	return (ENOMEM);
13615 }
13616 
13617 static int
dtrace_state_buffers(dtrace_state_t * state)13618 dtrace_state_buffers(dtrace_state_t *state)
13619 {
13620 	dtrace_speculation_t *spec = state->dts_speculations;
13621 	int rval, i;
13622 
13623 	if ((rval = dtrace_state_buffer(state, state->dts_buffer,
13624 	    DTRACEOPT_BUFSIZE)) != 0)
13625 		return (rval);
13626 
13627 	if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer,
13628 	    DTRACEOPT_AGGSIZE)) != 0)
13629 		return (rval);
13630 
13631 	for (i = 0; i < state->dts_nspeculations; i++) {
13632 		if ((rval = dtrace_state_buffer(state,
13633 		    spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0)
13634 			return (rval);
13635 	}
13636 
13637 	return (0);
13638 }
13639 
13640 static void
dtrace_state_prereserve(dtrace_state_t * state)13641 dtrace_state_prereserve(dtrace_state_t *state)
13642 {
13643 	dtrace_ecb_t *ecb;
13644 	dtrace_probe_t *probe;
13645 
13646 	state->dts_reserve = 0;
13647 
13648 	if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL)
13649 		return;
13650 
13651 	/*
13652 	 * If our buffer policy is a "fill" buffer policy, we need to set the
13653 	 * prereserved space to be the space required by the END probes.
13654 	 */
13655 	probe = dtrace_probes[dtrace_probeid_end - 1];
13656 	ASSERT(probe != NULL);
13657 
13658 	for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
13659 		if (ecb->dte_state != state)
13660 			continue;
13661 
13662 		state->dts_reserve += ecb->dte_needed + ecb->dte_alignment;
13663 	}
13664 }
13665 
13666 static int
dtrace_state_go(dtrace_state_t * state,processorid_t * cpu)13667 dtrace_state_go(dtrace_state_t *state, processorid_t *cpu)
13668 {
13669 	dtrace_optval_t *opt = state->dts_options, sz, nspec;
13670 	dtrace_speculation_t *spec;
13671 	dtrace_buffer_t *buf;
13672 	cyc_handler_t hdlr;
13673 	cyc_time_t when;
13674 	int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t);
13675 	dtrace_icookie_t cookie;
13676 
13677 	mutex_enter(&cpu_lock);
13678 	mutex_enter(&dtrace_lock);
13679 
13680 	if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
13681 		rval = EBUSY;
13682 		goto out;
13683 	}
13684 
13685 	/*
13686 	 * Before we can perform any checks, we must prime all of the
13687 	 * retained enablings that correspond to this state.
13688 	 */
13689 	dtrace_enabling_prime(state);
13690 
13691 	if (state->dts_destructive && !state->dts_cred.dcr_destructive) {
13692 		rval = EACCES;
13693 		goto out;
13694 	}
13695 
13696 	dtrace_state_prereserve(state);
13697 
13698 	/*
13699 	 * Now we want to do is try to allocate our speculations.
13700 	 * We do not automatically resize the number of speculations; if
13701 	 * this fails, we will fail the operation.
13702 	 */
13703 	nspec = opt[DTRACEOPT_NSPEC];
13704 	ASSERT(nspec != DTRACEOPT_UNSET);
13705 
13706 	if (nspec > INT_MAX) {
13707 		rval = ENOMEM;
13708 		goto out;
13709 	}
13710 
13711 	spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t),
13712 	    KM_NOSLEEP | KM_NORMALPRI);
13713 
13714 	if (spec == NULL) {
13715 		rval = ENOMEM;
13716 		goto out;
13717 	}
13718 
13719 	state->dts_speculations = spec;
13720 	state->dts_nspeculations = (int)nspec;
13721 
13722 	for (i = 0; i < nspec; i++) {
13723 		if ((buf = kmem_zalloc(bufsize,
13724 		    KM_NOSLEEP | KM_NORMALPRI)) == NULL) {
13725 			rval = ENOMEM;
13726 			goto err;
13727 		}
13728 
13729 		spec[i].dtsp_buffer = buf;
13730 	}
13731 
13732 	if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) {
13733 		if (dtrace_anon.dta_state == NULL) {
13734 			rval = ENOENT;
13735 			goto out;
13736 		}
13737 
13738 		if (state->dts_necbs != 0) {
13739 			rval = EALREADY;
13740 			goto out;
13741 		}
13742 
13743 		state->dts_anon = dtrace_anon_grab();
13744 		ASSERT(state->dts_anon != NULL);
13745 		state = state->dts_anon;
13746 
13747 		/*
13748 		 * We want "grabanon" to be set in the grabbed state, so we'll
13749 		 * copy that option value from the grabbing state into the
13750 		 * grabbed state.
13751 		 */
13752 		state->dts_options[DTRACEOPT_GRABANON] =
13753 		    opt[DTRACEOPT_GRABANON];
13754 
13755 		*cpu = dtrace_anon.dta_beganon;
13756 
13757 		/*
13758 		 * If the anonymous state is active (as it almost certainly
13759 		 * is if the anonymous enabling ultimately matched anything),
13760 		 * we don't allow any further option processing -- but we
13761 		 * don't return failure.
13762 		 */
13763 		if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
13764 			goto out;
13765 	}
13766 
13767 	if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET &&
13768 	    opt[DTRACEOPT_AGGSIZE] != 0) {
13769 		if (state->dts_aggregations == NULL) {
13770 			/*
13771 			 * We're not going to create an aggregation buffer
13772 			 * because we don't have any ECBs that contain
13773 			 * aggregations -- set this option to 0.
13774 			 */
13775 			opt[DTRACEOPT_AGGSIZE] = 0;
13776 		} else {
13777 			/*
13778 			 * If we have an aggregation buffer, we must also have
13779 			 * a buffer to use as scratch.
13780 			 */
13781 			if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET ||
13782 			    opt[DTRACEOPT_BUFSIZE] < state->dts_needed) {
13783 				opt[DTRACEOPT_BUFSIZE] = state->dts_needed;
13784 			}
13785 		}
13786 	}
13787 
13788 	if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET &&
13789 	    opt[DTRACEOPT_SPECSIZE] != 0) {
13790 		if (!state->dts_speculates) {
13791 			/*
13792 			 * We're not going to create speculation buffers
13793 			 * because we don't have any ECBs that actually
13794 			 * speculate -- set the speculation size to 0.
13795 			 */
13796 			opt[DTRACEOPT_SPECSIZE] = 0;
13797 		}
13798 	}
13799 
13800 	/*
13801 	 * The bare minimum size for any buffer that we're actually going to
13802 	 * do anything to is sizeof (uint64_t).
13803 	 */
13804 	sz = sizeof (uint64_t);
13805 
13806 	if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) ||
13807 	    (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) ||
13808 	    (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) {
13809 		/*
13810 		 * A buffer size has been explicitly set to 0 (or to a size
13811 		 * that will be adjusted to 0) and we need the space -- we
13812 		 * need to return failure.  We return ENOSPC to differentiate
13813 		 * it from failing to allocate a buffer due to failure to meet
13814 		 * the reserve (for which we return E2BIG).
13815 		 */
13816 		rval = ENOSPC;
13817 		goto out;
13818 	}
13819 
13820 	if ((rval = dtrace_state_buffers(state)) != 0)
13821 		goto err;
13822 
13823 	if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET)
13824 		sz = dtrace_dstate_defsize;
13825 
13826 	do {
13827 		rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz);
13828 
13829 		if (rval == 0)
13830 			break;
13831 
13832 		if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13833 			goto err;
13834 	} while (sz >>= 1);
13835 
13836 	opt[DTRACEOPT_DYNVARSIZE] = sz;
13837 
13838 	if (rval != 0)
13839 		goto err;
13840 
13841 	if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max)
13842 		opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max;
13843 
13844 	if (opt[DTRACEOPT_CLEANRATE] == 0)
13845 		opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13846 
13847 	if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min)
13848 		opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min;
13849 
13850 	if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max)
13851 		opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13852 
13853 	hdlr.cyh_func = (cyc_func_t)dtrace_state_clean;
13854 	hdlr.cyh_arg = state;
13855 	hdlr.cyh_level = CY_LOW_LEVEL;
13856 
13857 	when.cyt_when = 0;
13858 	when.cyt_interval = opt[DTRACEOPT_CLEANRATE];
13859 
13860 	state->dts_cleaner = cyclic_add(&hdlr, &when);
13861 
13862 	hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman;
13863 	hdlr.cyh_arg = state;
13864 	hdlr.cyh_level = CY_LOW_LEVEL;
13865 
13866 	when.cyt_when = 0;
13867 	when.cyt_interval = dtrace_deadman_interval;
13868 
13869 	state->dts_alive = state->dts_laststatus = dtrace_gethrtime();
13870 	state->dts_deadman = cyclic_add(&hdlr, &when);
13871 
13872 	state->dts_activity = DTRACE_ACTIVITY_WARMUP;
13873 
13874 	if (state->dts_getf != 0 &&
13875 	    !(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL)) {
13876 		/*
13877 		 * We don't have kernel privs but we have at least one call
13878 		 * to getf(); we need to bump our zone's count, and (if
13879 		 * this is the first enabling to have an unprivileged call
13880 		 * to getf()) we need to hook into closef().
13881 		 */
13882 		state->dts_cred.dcr_cred->cr_zone->zone_dtrace_getf++;
13883 
13884 		if (dtrace_getf++ == 0) {
13885 			ASSERT(dtrace_closef == NULL);
13886 			dtrace_closef = dtrace_getf_barrier;
13887 		}
13888 	}
13889 
13890 	/*
13891 	 * Now it's time to actually fire the BEGIN probe.  We need to disable
13892 	 * interrupts here both to record the CPU on which we fired the BEGIN
13893 	 * probe (the data from this CPU will be processed first at user
13894 	 * level) and to manually activate the buffer for this CPU.
13895 	 */
13896 	cookie = dtrace_interrupt_disable();
13897 	*cpu = CPU->cpu_id;
13898 	ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE);
13899 	state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE;
13900 
13901 	dtrace_probe(dtrace_probeid_begin,
13902 	    (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
13903 	dtrace_interrupt_enable(cookie);
13904 	/*
13905 	 * We may have had an exit action from a BEGIN probe; only change our
13906 	 * state to ACTIVE if we're still in WARMUP.
13907 	 */
13908 	ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP ||
13909 	    state->dts_activity == DTRACE_ACTIVITY_DRAINING);
13910 
13911 	if (state->dts_activity == DTRACE_ACTIVITY_WARMUP)
13912 		state->dts_activity = DTRACE_ACTIVITY_ACTIVE;
13913 
13914 	/*
13915 	 * Regardless of whether or not now we're in ACTIVE or DRAINING, we
13916 	 * want each CPU to transition its principal buffer out of the
13917 	 * INACTIVE state.  Doing this assures that no CPU will suddenly begin
13918 	 * processing an ECB halfway down a probe's ECB chain; all CPUs will
13919 	 * atomically transition from processing none of a state's ECBs to
13920 	 * processing all of them.
13921 	 */
13922 	dtrace_xcall(DTRACE_CPUALL,
13923 	    (dtrace_xcall_t)dtrace_buffer_activate, state);
13924 	goto out;
13925 
13926 err:
13927 	dtrace_buffer_free(state->dts_buffer);
13928 	dtrace_buffer_free(state->dts_aggbuffer);
13929 
13930 	if ((nspec = state->dts_nspeculations) == 0) {
13931 		ASSERT(state->dts_speculations == NULL);
13932 		goto out;
13933 	}
13934 
13935 	spec = state->dts_speculations;
13936 	ASSERT(spec != NULL);
13937 
13938 	for (i = 0; i < state->dts_nspeculations; i++) {
13939 		if ((buf = spec[i].dtsp_buffer) == NULL)
13940 			break;
13941 
13942 		dtrace_buffer_free(buf);
13943 		kmem_free(buf, bufsize);
13944 	}
13945 
13946 	kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
13947 	state->dts_nspeculations = 0;
13948 	state->dts_speculations = NULL;
13949 
13950 out:
13951 	mutex_exit(&dtrace_lock);
13952 	mutex_exit(&cpu_lock);
13953 
13954 	return (rval);
13955 }
13956 
13957 static int
dtrace_state_stop(dtrace_state_t * state,processorid_t * cpu)13958 dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu)
13959 {
13960 	dtrace_icookie_t cookie;
13961 
13962 	ASSERT(MUTEX_HELD(&dtrace_lock));
13963 
13964 	if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE &&
13965 	    state->dts_activity != DTRACE_ACTIVITY_DRAINING)
13966 		return (EINVAL);
13967 
13968 	/*
13969 	 * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync
13970 	 * to be sure that every CPU has seen it.  See below for the details
13971 	 * on why this is done.
13972 	 */
13973 	state->dts_activity = DTRACE_ACTIVITY_DRAINING;
13974 	dtrace_sync();
13975 
13976 	/*
13977 	 * By this point, it is impossible for any CPU to be still processing
13978 	 * with DTRACE_ACTIVITY_ACTIVE.  We can thus set our activity to
13979 	 * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any
13980 	 * other CPU in dtrace_buffer_reserve().  This allows dtrace_probe()
13981 	 * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN
13982 	 * iff we're in the END probe.
13983 	 */
13984 	state->dts_activity = DTRACE_ACTIVITY_COOLDOWN;
13985 	dtrace_sync();
13986 	ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN);
13987 
13988 	/*
13989 	 * Finally, we can release the reserve and call the END probe.  We
13990 	 * disable interrupts across calling the END probe to allow us to
13991 	 * return the CPU on which we actually called the END probe.  This
13992 	 * allows user-land to be sure that this CPU's principal buffer is
13993 	 * processed last.
13994 	 */
13995 	state->dts_reserve = 0;
13996 
13997 	cookie = dtrace_interrupt_disable();
13998 	*cpu = CPU->cpu_id;
13999 	dtrace_probe(dtrace_probeid_end,
14000 	    (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
14001 	dtrace_interrupt_enable(cookie);
14002 
14003 	state->dts_activity = DTRACE_ACTIVITY_STOPPED;
14004 	dtrace_sync();
14005 
14006 	if (state->dts_getf != 0 &&
14007 	    !(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL)) {
14008 		/*
14009 		 * We don't have kernel privs but we have at least one call
14010 		 * to getf(); we need to lower our zone's count, and (if
14011 		 * this is the last enabling to have an unprivileged call
14012 		 * to getf()) we need to clear the closef() hook.
14013 		 */
14014 		ASSERT(state->dts_cred.dcr_cred->cr_zone->zone_dtrace_getf > 0);
14015 		ASSERT(dtrace_closef == dtrace_getf_barrier);
14016 		ASSERT(dtrace_getf > 0);
14017 
14018 		state->dts_cred.dcr_cred->cr_zone->zone_dtrace_getf--;
14019 
14020 		if (--dtrace_getf == 0)
14021 			dtrace_closef = NULL;
14022 	}
14023 
14024 	return (0);
14025 }
14026 
14027 static int
dtrace_state_option(dtrace_state_t * state,dtrace_optid_t option,dtrace_optval_t val)14028 dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option,
14029     dtrace_optval_t val)
14030 {
14031 	ASSERT(MUTEX_HELD(&dtrace_lock));
14032 
14033 	if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
14034 		return (EBUSY);
14035 
14036 	if (option >= DTRACEOPT_MAX)
14037 		return (EINVAL);
14038 
14039 	if (option != DTRACEOPT_CPU && val < 0)
14040 		return (EINVAL);
14041 
14042 	switch (option) {
14043 	case DTRACEOPT_DESTRUCTIVE:
14044 		if (dtrace_destructive_disallow)
14045 			return (EACCES);
14046 
14047 		state->dts_cred.dcr_destructive = 1;
14048 		break;
14049 
14050 	case DTRACEOPT_BUFSIZE:
14051 	case DTRACEOPT_DYNVARSIZE:
14052 	case DTRACEOPT_AGGSIZE:
14053 	case DTRACEOPT_SPECSIZE:
14054 	case DTRACEOPT_STRSIZE:
14055 		if (val < 0)
14056 			return (EINVAL);
14057 
14058 		if (val >= LONG_MAX) {
14059 			/*
14060 			 * If this is an otherwise negative value, set it to
14061 			 * the highest multiple of 128m less than LONG_MAX.
14062 			 * Technically, we're adjusting the size without
14063 			 * regard to the buffer resizing policy, but in fact,
14064 			 * this has no effect -- if we set the buffer size to
14065 			 * ~LONG_MAX and the buffer policy is ultimately set to
14066 			 * be "manual", the buffer allocation is guaranteed to
14067 			 * fail, if only because the allocation requires two
14068 			 * buffers.  (We set the the size to the highest
14069 			 * multiple of 128m because it ensures that the size
14070 			 * will remain a multiple of a megabyte when
14071 			 * repeatedly halved -- all the way down to 15m.)
14072 			 */
14073 			val = LONG_MAX - (1 << 27) + 1;
14074 		}
14075 	}
14076 
14077 	state->dts_options[option] = val;
14078 
14079 	return (0);
14080 }
14081 
14082 static void
dtrace_state_destroy(dtrace_state_t * state)14083 dtrace_state_destroy(dtrace_state_t *state)
14084 {
14085 	dtrace_ecb_t *ecb;
14086 	dtrace_vstate_t *vstate = &state->dts_vstate;
14087 	minor_t minor = getminor(state->dts_dev);
14088 	int i, bufsize = NCPU * sizeof (dtrace_buffer_t);
14089 	dtrace_speculation_t *spec = state->dts_speculations;
14090 	int nspec = state->dts_nspeculations;
14091 	uint32_t match;
14092 
14093 	ASSERT(MUTEX_HELD(&dtrace_lock));
14094 	ASSERT(MUTEX_HELD(&cpu_lock));
14095 
14096 	/*
14097 	 * First, retract any retained enablings for this state.
14098 	 */
14099 	dtrace_enabling_retract(state);
14100 	ASSERT(state->dts_nretained == 0);
14101 
14102 	if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE ||
14103 	    state->dts_activity == DTRACE_ACTIVITY_DRAINING) {
14104 		/*
14105 		 * We have managed to come into dtrace_state_destroy() on a
14106 		 * hot enabling -- almost certainly because of a disorderly
14107 		 * shutdown of a consumer.  (That is, a consumer that is
14108 		 * exiting without having called dtrace_stop().) In this case,
14109 		 * we're going to set our activity to be KILLED, and then
14110 		 * issue a sync to be sure that everyone is out of probe
14111 		 * context before we start blowing away ECBs.
14112 		 */
14113 		state->dts_activity = DTRACE_ACTIVITY_KILLED;
14114 		dtrace_sync();
14115 	}
14116 
14117 	/*
14118 	 * Release the credential hold we took in dtrace_state_create().
14119 	 */
14120 	if (state->dts_cred.dcr_cred != NULL)
14121 		crfree(state->dts_cred.dcr_cred);
14122 
14123 	/*
14124 	 * Now we can safely disable and destroy any enabled probes.  Because
14125 	 * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress
14126 	 * (especially if they're all enabled), we take two passes through the
14127 	 * ECBs:  in the first, we disable just DTRACE_PRIV_KERNEL probes, and
14128 	 * in the second we disable whatever is left over.
14129 	 */
14130 	for (match = DTRACE_PRIV_KERNEL; ; match = 0) {
14131 		for (i = 0; i < state->dts_necbs; i++) {
14132 			if ((ecb = state->dts_ecbs[i]) == NULL)
14133 				continue;
14134 
14135 			if (match && ecb->dte_probe != NULL) {
14136 				dtrace_probe_t *probe = ecb->dte_probe;
14137 				dtrace_provider_t *prov = probe->dtpr_provider;
14138 
14139 				if (!(prov->dtpv_priv.dtpp_flags & match))
14140 					continue;
14141 			}
14142 
14143 			dtrace_ecb_disable(ecb);
14144 			dtrace_ecb_destroy(ecb);
14145 		}
14146 
14147 		if (!match)
14148 			break;
14149 	}
14150 
14151 	/*
14152 	 * Before we free the buffers, perform one more sync to assure that
14153 	 * every CPU is out of probe context.
14154 	 */
14155 	dtrace_sync();
14156 
14157 	dtrace_buffer_free(state->dts_buffer);
14158 	dtrace_buffer_free(state->dts_aggbuffer);
14159 
14160 	for (i = 0; i < nspec; i++)
14161 		dtrace_buffer_free(spec[i].dtsp_buffer);
14162 
14163 	if (state->dts_cleaner != CYCLIC_NONE)
14164 		cyclic_remove(state->dts_cleaner);
14165 
14166 	if (state->dts_deadman != CYCLIC_NONE)
14167 		cyclic_remove(state->dts_deadman);
14168 
14169 	dtrace_dstate_fini(&vstate->dtvs_dynvars);
14170 	dtrace_vstate_fini(vstate);
14171 	kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *));
14172 
14173 	if (state->dts_aggregations != NULL) {
14174 #ifdef DEBUG
14175 		for (i = 0; i < state->dts_naggregations; i++)
14176 			ASSERT(state->dts_aggregations[i] == NULL);
14177 #endif
14178 		ASSERT(state->dts_naggregations > 0);
14179 		kmem_free(state->dts_aggregations,
14180 		    state->dts_naggregations * sizeof (dtrace_aggregation_t *));
14181 	}
14182 
14183 	kmem_free(state->dts_buffer, bufsize);
14184 	kmem_free(state->dts_aggbuffer, bufsize);
14185 
14186 	for (i = 0; i < nspec; i++)
14187 		kmem_free(spec[i].dtsp_buffer, bufsize);
14188 
14189 	kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
14190 
14191 	dtrace_format_destroy(state);
14192 
14193 	vmem_destroy(state->dts_aggid_arena);
14194 	ddi_soft_state_free(dtrace_softstate, minor);
14195 	vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
14196 }
14197 
14198 /*
14199  * DTrace Anonymous Enabling Functions
14200  */
14201 static dtrace_state_t *
dtrace_anon_grab(void)14202 dtrace_anon_grab(void)
14203 {
14204 	dtrace_state_t *state;
14205 
14206 	ASSERT(MUTEX_HELD(&dtrace_lock));
14207 
14208 	if ((state = dtrace_anon.dta_state) == NULL) {
14209 		ASSERT(dtrace_anon.dta_enabling == NULL);
14210 		return (NULL);
14211 	}
14212 
14213 	ASSERT(dtrace_anon.dta_enabling != NULL);
14214 	ASSERT(dtrace_retained != NULL);
14215 
14216 	dtrace_enabling_destroy(dtrace_anon.dta_enabling);
14217 	dtrace_anon.dta_enabling = NULL;
14218 	dtrace_anon.dta_state = NULL;
14219 
14220 	return (state);
14221 }
14222 
14223 static void
dtrace_anon_property(void)14224 dtrace_anon_property(void)
14225 {
14226 	int i, rv;
14227 	dtrace_state_t *state;
14228 	dof_hdr_t *dof;
14229 	char c[32];		/* enough for "dof-data-" + digits */
14230 
14231 	ASSERT(MUTEX_HELD(&dtrace_lock));
14232 	ASSERT(MUTEX_HELD(&cpu_lock));
14233 
14234 	for (i = 0; ; i++) {
14235 		(void) snprintf(c, sizeof (c), "dof-data-%d", i);
14236 
14237 		dtrace_err_verbose = 1;
14238 
14239 		if ((dof = dtrace_dof_property(c)) == NULL) {
14240 			dtrace_err_verbose = 0;
14241 			break;
14242 		}
14243 
14244 		/*
14245 		 * We want to create anonymous state, so we need to transition
14246 		 * the kernel debugger to indicate that DTrace is active.  If
14247 		 * this fails (e.g. because the debugger has modified text in
14248 		 * some way), we won't continue with the processing.
14249 		 */
14250 		if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
14251 			cmn_err(CE_NOTE, "kernel debugger active; anonymous "
14252 			    "enabling ignored.");
14253 			dtrace_dof_destroy(dof);
14254 			break;
14255 		}
14256 
14257 		/*
14258 		 * If we haven't allocated an anonymous state, we'll do so now.
14259 		 */
14260 		if ((state = dtrace_anon.dta_state) == NULL) {
14261 			state = dtrace_state_create(NULL, NULL);
14262 			dtrace_anon.dta_state = state;
14263 
14264 			if (state == NULL) {
14265 				/*
14266 				 * This basically shouldn't happen:  the only
14267 				 * failure mode from dtrace_state_create() is a
14268 				 * failure of ddi_soft_state_zalloc() that
14269 				 * itself should never happen.  Still, the
14270 				 * interface allows for a failure mode, and
14271 				 * we want to fail as gracefully as possible:
14272 				 * we'll emit an error message and cease
14273 				 * processing anonymous state in this case.
14274 				 */
14275 				cmn_err(CE_WARN, "failed to create "
14276 				    "anonymous state");
14277 				dtrace_dof_destroy(dof);
14278 				break;
14279 			}
14280 		}
14281 
14282 		rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(),
14283 		    &dtrace_anon.dta_enabling, 0, B_TRUE);
14284 
14285 		if (rv == 0)
14286 			rv = dtrace_dof_options(dof, state);
14287 
14288 		dtrace_err_verbose = 0;
14289 		dtrace_dof_destroy(dof);
14290 
14291 		if (rv != 0) {
14292 			/*
14293 			 * This is malformed DOF; chuck any anonymous state
14294 			 * that we created.
14295 			 */
14296 			ASSERT(dtrace_anon.dta_enabling == NULL);
14297 			dtrace_state_destroy(state);
14298 			dtrace_anon.dta_state = NULL;
14299 			break;
14300 		}
14301 
14302 		ASSERT(dtrace_anon.dta_enabling != NULL);
14303 	}
14304 
14305 	if (dtrace_anon.dta_enabling != NULL) {
14306 		int rval;
14307 
14308 		/*
14309 		 * dtrace_enabling_retain() can only fail because we are
14310 		 * trying to retain more enablings than are allowed -- but
14311 		 * we only have one anonymous enabling, and we are guaranteed
14312 		 * to be allowed at least one retained enabling; we assert
14313 		 * that dtrace_enabling_retain() returns success.
14314 		 */
14315 		rval = dtrace_enabling_retain(dtrace_anon.dta_enabling);
14316 		ASSERT(rval == 0);
14317 
14318 		dtrace_enabling_dump(dtrace_anon.dta_enabling);
14319 	}
14320 }
14321 
14322 /*
14323  * DTrace Helper Functions
14324  */
14325 static void
dtrace_helper_trace(dtrace_helper_action_t * helper,dtrace_mstate_t * mstate,dtrace_vstate_t * vstate,int where)14326 dtrace_helper_trace(dtrace_helper_action_t *helper,
14327     dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where)
14328 {
14329 	uint32_t size, next, nnext, i;
14330 	dtrace_helptrace_t *ent, *buffer;
14331 	uint16_t flags = cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
14332 
14333 	if ((buffer = dtrace_helptrace_buffer) == NULL)
14334 		return;
14335 
14336 	ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals);
14337 
14338 	/*
14339 	 * What would a tracing framework be without its own tracing
14340 	 * framework?  (Well, a hell of a lot simpler, for starters...)
14341 	 */
14342 	size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals *
14343 	    sizeof (uint64_t) - sizeof (uint64_t);
14344 
14345 	/*
14346 	 * Iterate until we can allocate a slot in the trace buffer.
14347 	 */
14348 	do {
14349 		next = dtrace_helptrace_next;
14350 
14351 		if (next + size < dtrace_helptrace_bufsize) {
14352 			nnext = next + size;
14353 		} else {
14354 			nnext = size;
14355 		}
14356 	} while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next);
14357 
14358 	/*
14359 	 * We have our slot; fill it in.
14360 	 */
14361 	if (nnext == size) {
14362 		dtrace_helptrace_wrapped++;
14363 		next = 0;
14364 	}
14365 
14366 	ent = (dtrace_helptrace_t *)((uintptr_t)buffer + next);
14367 	ent->dtht_helper = helper;
14368 	ent->dtht_where = where;
14369 	ent->dtht_nlocals = vstate->dtvs_nlocals;
14370 
14371 	ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ?
14372 	    mstate->dtms_fltoffs : -1;
14373 	ent->dtht_fault = DTRACE_FLAGS2FLT(flags);
14374 	ent->dtht_illval = cpu_core[CPU->cpu_id].cpuc_dtrace_illval;
14375 
14376 	for (i = 0; i < vstate->dtvs_nlocals; i++) {
14377 		dtrace_statvar_t *svar;
14378 
14379 		if ((svar = vstate->dtvs_locals[i]) == NULL)
14380 			continue;
14381 
14382 		ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t));
14383 		ent->dtht_locals[i] =
14384 		    ((uint64_t *)(uintptr_t)svar->dtsv_data)[CPU->cpu_id];
14385 	}
14386 }
14387 
14388 static uint64_t
dtrace_helper(int which,dtrace_mstate_t * mstate,dtrace_state_t * state,uint64_t arg0,uint64_t arg1)14389 dtrace_helper(int which, dtrace_mstate_t *mstate,
14390     dtrace_state_t *state, uint64_t arg0, uint64_t arg1)
14391 {
14392 	uint16_t *flags = &cpu_core[CPU->cpu_id].cpuc_dtrace_flags;
14393 	uint64_t sarg0 = mstate->dtms_arg[0];
14394 	uint64_t sarg1 = mstate->dtms_arg[1];
14395 	uint64_t rval;
14396 	dtrace_helpers_t *helpers = curproc->p_dtrace_helpers;
14397 	dtrace_helper_action_t *helper;
14398 	dtrace_vstate_t *vstate;
14399 	dtrace_difo_t *pred;
14400 	int i, trace = dtrace_helptrace_buffer != NULL;
14401 
14402 	ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS);
14403 
14404 	if (helpers == NULL)
14405 		return (0);
14406 
14407 	if ((helper = helpers->dthps_actions[which]) == NULL)
14408 		return (0);
14409 
14410 	vstate = &helpers->dthps_vstate;
14411 	mstate->dtms_arg[0] = arg0;
14412 	mstate->dtms_arg[1] = arg1;
14413 
14414 	/*
14415 	 * Now iterate over each helper.  If its predicate evaluates to 'true',
14416 	 * we'll call the corresponding actions.  Note that the below calls
14417 	 * to dtrace_dif_emulate() may set faults in machine state.  This is
14418 	 * okay:  our caller (the outer dtrace_dif_emulate()) will simply plow
14419 	 * the stored DIF offset with its own (which is the desired behavior).
14420 	 * Also, note the calls to dtrace_dif_emulate() may allocate scratch
14421 	 * from machine state; this is okay, too.
14422 	 */
14423 	for (; helper != NULL; helper = helper->dtha_next) {
14424 		if ((pred = helper->dtha_predicate) != NULL) {
14425 			if (trace)
14426 				dtrace_helper_trace(helper, mstate, vstate, 0);
14427 
14428 			if (!dtrace_dif_emulate(pred, mstate, vstate, state))
14429 				goto next;
14430 
14431 			if (*flags & CPU_DTRACE_FAULT)
14432 				goto err;
14433 		}
14434 
14435 		for (i = 0; i < helper->dtha_nactions; i++) {
14436 			if (trace)
14437 				dtrace_helper_trace(helper,
14438 				    mstate, vstate, i + 1);
14439 
14440 			rval = dtrace_dif_emulate(helper->dtha_actions[i],
14441 			    mstate, vstate, state);
14442 
14443 			if (*flags & CPU_DTRACE_FAULT)
14444 				goto err;
14445 		}
14446 
14447 next:
14448 		if (trace)
14449 			dtrace_helper_trace(helper, mstate, vstate,
14450 			    DTRACE_HELPTRACE_NEXT);
14451 	}
14452 
14453 	if (trace)
14454 		dtrace_helper_trace(helper, mstate, vstate,
14455 		    DTRACE_HELPTRACE_DONE);
14456 
14457 	/*
14458 	 * Restore the arg0 that we saved upon entry.
14459 	 */
14460 	mstate->dtms_arg[0] = sarg0;
14461 	mstate->dtms_arg[1] = sarg1;
14462 
14463 	return (rval);
14464 
14465 err:
14466 	if (trace)
14467 		dtrace_helper_trace(helper, mstate, vstate,
14468 		    DTRACE_HELPTRACE_ERR);
14469 
14470 	/*
14471 	 * Restore the arg0 that we saved upon entry.
14472 	 */
14473 	mstate->dtms_arg[0] = sarg0;
14474 	mstate->dtms_arg[1] = sarg1;
14475 
14476 	return (NULL);
14477 }
14478 
14479 static void
dtrace_helper_action_destroy(dtrace_helper_action_t * helper,dtrace_vstate_t * vstate)14480 dtrace_helper_action_destroy(dtrace_helper_action_t *helper,
14481     dtrace_vstate_t *vstate)
14482 {
14483 	int i;
14484 
14485 	if (helper->dtha_predicate != NULL)
14486 		dtrace_difo_release(helper->dtha_predicate, vstate);
14487 
14488 	for (i = 0; i < helper->dtha_nactions; i++) {
14489 		ASSERT(helper->dtha_actions[i] != NULL);
14490 		dtrace_difo_release(helper->dtha_actions[i], vstate);
14491 	}
14492 
14493 	kmem_free(helper->dtha_actions,
14494 	    helper->dtha_nactions * sizeof (dtrace_difo_t *));
14495 	kmem_free(helper, sizeof (dtrace_helper_action_t));
14496 }
14497 
14498 static int
dtrace_helper_destroygen(int gen)14499 dtrace_helper_destroygen(int gen)
14500 {
14501 	proc_t *p = curproc;
14502 	dtrace_helpers_t *help = p->p_dtrace_helpers;
14503 	dtrace_vstate_t *vstate;
14504 	int i;
14505 
14506 	ASSERT(MUTEX_HELD(&dtrace_lock));
14507 
14508 	if (help == NULL || gen > help->dthps_generation)
14509 		return (EINVAL);
14510 
14511 	vstate = &help->dthps_vstate;
14512 
14513 	for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
14514 		dtrace_helper_action_t *last = NULL, *h, *next;
14515 
14516 		for (h = help->dthps_actions[i]; h != NULL; h = next) {
14517 			next = h->dtha_next;
14518 
14519 			if (h->dtha_generation == gen) {
14520 				if (last != NULL) {
14521 					last->dtha_next = next;
14522 				} else {
14523 					help->dthps_actions[i] = next;
14524 				}
14525 
14526 				dtrace_helper_action_destroy(h, vstate);
14527 			} else {
14528 				last = h;
14529 			}
14530 		}
14531 	}
14532 
14533 	/*
14534 	 * Interate until we've cleared out all helper providers with the
14535 	 * given generation number.
14536 	 */
14537 	for (;;) {
14538 		dtrace_helper_provider_t *prov;
14539 
14540 		/*
14541 		 * Look for a helper provider with the right generation. We
14542 		 * have to start back at the beginning of the list each time
14543 		 * because we drop dtrace_lock. It's unlikely that we'll make
14544 		 * more than two passes.
14545 		 */
14546 		for (i = 0; i < help->dthps_nprovs; i++) {
14547 			prov = help->dthps_provs[i];
14548 
14549 			if (prov->dthp_generation == gen)
14550 				break;
14551 		}
14552 
14553 		/*
14554 		 * If there were no matches, we're done.
14555 		 */
14556 		if (i == help->dthps_nprovs)
14557 			break;
14558 
14559 		/*
14560 		 * Move the last helper provider into this slot.
14561 		 */
14562 		help->dthps_nprovs--;
14563 		help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs];
14564 		help->dthps_provs[help->dthps_nprovs] = NULL;
14565 
14566 		mutex_exit(&dtrace_lock);
14567 
14568 		/*
14569 		 * If we have a meta provider, remove this helper provider.
14570 		 */
14571 		mutex_enter(&dtrace_meta_lock);
14572 		if (dtrace_meta_pid != NULL) {
14573 			ASSERT(dtrace_deferred_pid == NULL);
14574 			dtrace_helper_provider_remove(&prov->dthp_prov,
14575 			    p->p_pid);
14576 		}
14577 		mutex_exit(&dtrace_meta_lock);
14578 
14579 		dtrace_helper_provider_destroy(prov);
14580 
14581 		mutex_enter(&dtrace_lock);
14582 	}
14583 
14584 	return (0);
14585 }
14586 
14587 static int
dtrace_helper_validate(dtrace_helper_action_t * helper)14588 dtrace_helper_validate(dtrace_helper_action_t *helper)
14589 {
14590 	int err = 0, i;
14591 	dtrace_difo_t *dp;
14592 
14593 	if ((dp = helper->dtha_predicate) != NULL)
14594 		err += dtrace_difo_validate_helper(dp);
14595 
14596 	for (i = 0; i < helper->dtha_nactions; i++)
14597 		err += dtrace_difo_validate_helper(helper->dtha_actions[i]);
14598 
14599 	return (err == 0);
14600 }
14601 
14602 static int
dtrace_helper_action_add(int which,dtrace_ecbdesc_t * ep)14603 dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep)
14604 {
14605 	dtrace_helpers_t *help;
14606 	dtrace_helper_action_t *helper, *last;
14607 	dtrace_actdesc_t *act;
14608 	dtrace_vstate_t *vstate;
14609 	dtrace_predicate_t *pred;
14610 	int count = 0, nactions = 0, i;
14611 
14612 	if (which < 0 || which >= DTRACE_NHELPER_ACTIONS)
14613 		return (EINVAL);
14614 
14615 	help = curproc->p_dtrace_helpers;
14616 	last = help->dthps_actions[which];
14617 	vstate = &help->dthps_vstate;
14618 
14619 	for (count = 0; last != NULL; last = last->dtha_next) {
14620 		count++;
14621 		if (last->dtha_next == NULL)
14622 			break;
14623 	}
14624 
14625 	/*
14626 	 * If we already have dtrace_helper_actions_max helper actions for this
14627 	 * helper action type, we'll refuse to add a new one.
14628 	 */
14629 	if (count >= dtrace_helper_actions_max)
14630 		return (ENOSPC);
14631 
14632 	helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP);
14633 	helper->dtha_generation = help->dthps_generation;
14634 
14635 	if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) {
14636 		ASSERT(pred->dtp_difo != NULL);
14637 		dtrace_difo_hold(pred->dtp_difo);
14638 		helper->dtha_predicate = pred->dtp_difo;
14639 	}
14640 
14641 	for (act = ep->dted_action; act != NULL; act = act->dtad_next) {
14642 		if (act->dtad_kind != DTRACEACT_DIFEXPR)
14643 			goto err;
14644 
14645 		if (act->dtad_difo == NULL)
14646 			goto err;
14647 
14648 		nactions++;
14649 	}
14650 
14651 	helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) *
14652 	    (helper->dtha_nactions = nactions), KM_SLEEP);
14653 
14654 	for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) {
14655 		dtrace_difo_hold(act->dtad_difo);
14656 		helper->dtha_actions[i++] = act->dtad_difo;
14657 	}
14658 
14659 	if (!dtrace_helper_validate(helper))
14660 		goto err;
14661 
14662 	if (last == NULL) {
14663 		help->dthps_actions[which] = helper;
14664 	} else {
14665 		last->dtha_next = helper;
14666 	}
14667 
14668 	if (vstate->dtvs_nlocals > dtrace_helptrace_nlocals) {
14669 		dtrace_helptrace_nlocals = vstate->dtvs_nlocals;
14670 		dtrace_helptrace_next = 0;
14671 	}
14672 
14673 	return (0);
14674 err:
14675 	dtrace_helper_action_destroy(helper, vstate);
14676 	return (EINVAL);
14677 }
14678 
14679 static void
dtrace_helper_provider_register(proc_t * p,dtrace_helpers_t * help,dof_helper_t * dofhp)14680 dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help,
14681     dof_helper_t *dofhp)
14682 {
14683 	ASSERT(MUTEX_NOT_HELD(&dtrace_lock));
14684 
14685 	mutex_enter(&dtrace_meta_lock);
14686 	mutex_enter(&dtrace_lock);
14687 
14688 	if (!dtrace_attached() || dtrace_meta_pid == NULL) {
14689 		/*
14690 		 * If the dtrace module is loaded but not attached, or if
14691 		 * there aren't isn't a meta provider registered to deal with
14692 		 * these provider descriptions, we need to postpone creating
14693 		 * the actual providers until later.
14694 		 */
14695 
14696 		if (help->dthps_next == NULL && help->dthps_prev == NULL &&
14697 		    dtrace_deferred_pid != help) {
14698 			help->dthps_deferred = 1;
14699 			help->dthps_pid = p->p_pid;
14700 			help->dthps_next = dtrace_deferred_pid;
14701 			help->dthps_prev = NULL;
14702 			if (dtrace_deferred_pid != NULL)
14703 				dtrace_deferred_pid->dthps_prev = help;
14704 			dtrace_deferred_pid = help;
14705 		}
14706 
14707 		mutex_exit(&dtrace_lock);
14708 
14709 	} else if (dofhp != NULL) {
14710 		/*
14711 		 * If the dtrace module is loaded and we have a particular
14712 		 * helper provider description, pass that off to the
14713 		 * meta provider.
14714 		 */
14715 
14716 		mutex_exit(&dtrace_lock);
14717 
14718 		dtrace_helper_provide(dofhp, p->p_pid);
14719 
14720 	} else {
14721 		/*
14722 		 * Otherwise, just pass all the helper provider descriptions
14723 		 * off to the meta provider.
14724 		 */
14725 
14726 		int i;
14727 		mutex_exit(&dtrace_lock);
14728 
14729 		for (i = 0; i < help->dthps_nprovs; i++) {
14730 			dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
14731 			    p->p_pid);
14732 		}
14733 	}
14734 
14735 	mutex_exit(&dtrace_meta_lock);
14736 }
14737 
14738 static int
dtrace_helper_provider_add(dof_helper_t * dofhp,int gen)14739 dtrace_helper_provider_add(dof_helper_t *dofhp, int gen)
14740 {
14741 	dtrace_helpers_t *help;
14742 	dtrace_helper_provider_t *hprov, **tmp_provs;
14743 	uint_t tmp_maxprovs, i;
14744 
14745 	ASSERT(MUTEX_HELD(&dtrace_lock));
14746 
14747 	help = curproc->p_dtrace_helpers;
14748 	ASSERT(help != NULL);
14749 
14750 	/*
14751 	 * If we already have dtrace_helper_providers_max helper providers,
14752 	 * we're refuse to add a new one.
14753 	 */
14754 	if (help->dthps_nprovs >= dtrace_helper_providers_max)
14755 		return (ENOSPC);
14756 
14757 	/*
14758 	 * Check to make sure this isn't a duplicate.
14759 	 */
14760 	for (i = 0; i < help->dthps_nprovs; i++) {
14761 		if (dofhp->dofhp_dof ==
14762 		    help->dthps_provs[i]->dthp_prov.dofhp_dof)
14763 			return (EALREADY);
14764 	}
14765 
14766 	hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP);
14767 	hprov->dthp_prov = *dofhp;
14768 	hprov->dthp_ref = 1;
14769 	hprov->dthp_generation = gen;
14770 
14771 	/*
14772 	 * Allocate a bigger table for helper providers if it's already full.
14773 	 */
14774 	if (help->dthps_maxprovs == help->dthps_nprovs) {
14775 		tmp_maxprovs = help->dthps_maxprovs;
14776 		tmp_provs = help->dthps_provs;
14777 
14778 		if (help->dthps_maxprovs == 0)
14779 			help->dthps_maxprovs = 2;
14780 		else
14781 			help->dthps_maxprovs *= 2;
14782 		if (help->dthps_maxprovs > dtrace_helper_providers_max)
14783 			help->dthps_maxprovs = dtrace_helper_providers_max;
14784 
14785 		ASSERT(tmp_maxprovs < help->dthps_maxprovs);
14786 
14787 		help->dthps_provs = kmem_zalloc(help->dthps_maxprovs *
14788 		    sizeof (dtrace_helper_provider_t *), KM_SLEEP);
14789 
14790 		if (tmp_provs != NULL) {
14791 			bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs *
14792 			    sizeof (dtrace_helper_provider_t *));
14793 			kmem_free(tmp_provs, tmp_maxprovs *
14794 			    sizeof (dtrace_helper_provider_t *));
14795 		}
14796 	}
14797 
14798 	help->dthps_provs[help->dthps_nprovs] = hprov;
14799 	help->dthps_nprovs++;
14800 
14801 	return (0);
14802 }
14803 
14804 static void
dtrace_helper_provider_destroy(dtrace_helper_provider_t * hprov)14805 dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov)
14806 {
14807 	mutex_enter(&dtrace_lock);
14808 
14809 	if (--hprov->dthp_ref == 0) {
14810 		dof_hdr_t *dof;
14811 		mutex_exit(&dtrace_lock);
14812 		dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof;
14813 		dtrace_dof_destroy(dof);
14814 		kmem_free(hprov, sizeof (dtrace_helper_provider_t));
14815 	} else {
14816 		mutex_exit(&dtrace_lock);
14817 	}
14818 }
14819 
14820 static int
dtrace_helper_provider_validate(dof_hdr_t * dof,dof_sec_t * sec)14821 dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec)
14822 {
14823 	uintptr_t daddr = (uintptr_t)dof;
14824 	dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
14825 	dof_provider_t *provider;
14826 	dof_probe_t *probe;
14827 	uint8_t *arg;
14828 	char *strtab, *typestr;
14829 	dof_stridx_t typeidx;
14830 	size_t typesz;
14831 	uint_t nprobes, j, k;
14832 
14833 	ASSERT(sec->dofs_type == DOF_SECT_PROVIDER);
14834 
14835 	if (sec->dofs_offset & (sizeof (uint_t) - 1)) {
14836 		dtrace_dof_error(dof, "misaligned section offset");
14837 		return (-1);
14838 	}
14839 
14840 	/*
14841 	 * The section needs to be large enough to contain the DOF provider
14842 	 * structure appropriate for the given version.
14843 	 */
14844 	if (sec->dofs_size <
14845 	    ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ?
14846 	    offsetof(dof_provider_t, dofpv_prenoffs) :
14847 	    sizeof (dof_provider_t))) {
14848 		dtrace_dof_error(dof, "provider section too small");
14849 		return (-1);
14850 	}
14851 
14852 	provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
14853 	str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab);
14854 	prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes);
14855 	arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs);
14856 	off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs);
14857 
14858 	if (str_sec == NULL || prb_sec == NULL ||
14859 	    arg_sec == NULL || off_sec == NULL)
14860 		return (-1);
14861 
14862 	enoff_sec = NULL;
14863 
14864 	if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
14865 	    provider->dofpv_prenoffs != DOF_SECT_NONE &&
14866 	    (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS,
14867 	    provider->dofpv_prenoffs)) == NULL)
14868 		return (-1);
14869 
14870 	strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
14871 
14872 	if (provider->dofpv_name >= str_sec->dofs_size ||
14873 	    strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) {
14874 		dtrace_dof_error(dof, "invalid provider name");
14875 		return (-1);
14876 	}
14877 
14878 	if (prb_sec->dofs_entsize == 0 ||
14879 	    prb_sec->dofs_entsize > prb_sec->dofs_size) {
14880 		dtrace_dof_error(dof, "invalid entry size");
14881 		return (-1);
14882 	}
14883 
14884 	if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) {
14885 		dtrace_dof_error(dof, "misaligned entry size");
14886 		return (-1);
14887 	}
14888 
14889 	if (off_sec->dofs_entsize != sizeof (uint32_t)) {
14890 		dtrace_dof_error(dof, "invalid entry size");
14891 		return (-1);
14892 	}
14893 
14894 	if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) {
14895 		dtrace_dof_error(dof, "misaligned section offset");
14896 		return (-1);
14897 	}
14898 
14899 	if (arg_sec->dofs_entsize != sizeof (uint8_t)) {
14900 		dtrace_dof_error(dof, "invalid entry size");
14901 		return (-1);
14902 	}
14903 
14904 	arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
14905 
14906 	nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
14907 
14908 	/*
14909 	 * Take a pass through the probes to check for errors.
14910 	 */
14911 	for (j = 0; j < nprobes; j++) {
14912 		probe = (dof_probe_t *)(uintptr_t)(daddr +
14913 		    prb_sec->dofs_offset + j * prb_sec->dofs_entsize);
14914 
14915 		if (probe->dofpr_func >= str_sec->dofs_size) {
14916 			dtrace_dof_error(dof, "invalid function name");
14917 			return (-1);
14918 		}
14919 
14920 		if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) {
14921 			dtrace_dof_error(dof, "function name too long");
14922 			return (-1);
14923 		}
14924 
14925 		if (probe->dofpr_name >= str_sec->dofs_size ||
14926 		    strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) {
14927 			dtrace_dof_error(dof, "invalid probe name");
14928 			return (-1);
14929 		}
14930 
14931 		/*
14932 		 * The offset count must not wrap the index, and the offsets
14933 		 * must also not overflow the section's data.
14934 		 */
14935 		if (probe->dofpr_offidx + probe->dofpr_noffs <
14936 		    probe->dofpr_offidx ||
14937 		    (probe->dofpr_offidx + probe->dofpr_noffs) *
14938 		    off_sec->dofs_entsize > off_sec->dofs_size) {
14939 			dtrace_dof_error(dof, "invalid probe offset");
14940 			return (-1);
14941 		}
14942 
14943 		if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) {
14944 			/*
14945 			 * If there's no is-enabled offset section, make sure
14946 			 * there aren't any is-enabled offsets. Otherwise
14947 			 * perform the same checks as for probe offsets
14948 			 * (immediately above).
14949 			 */
14950 			if (enoff_sec == NULL) {
14951 				if (probe->dofpr_enoffidx != 0 ||
14952 				    probe->dofpr_nenoffs != 0) {
14953 					dtrace_dof_error(dof, "is-enabled "
14954 					    "offsets with null section");
14955 					return (-1);
14956 				}
14957 			} else if (probe->dofpr_enoffidx +
14958 			    probe->dofpr_nenoffs < probe->dofpr_enoffidx ||
14959 			    (probe->dofpr_enoffidx + probe->dofpr_nenoffs) *
14960 			    enoff_sec->dofs_entsize > enoff_sec->dofs_size) {
14961 				dtrace_dof_error(dof, "invalid is-enabled "
14962 				    "offset");
14963 				return (-1);
14964 			}
14965 
14966 			if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) {
14967 				dtrace_dof_error(dof, "zero probe and "
14968 				    "is-enabled offsets");
14969 				return (-1);
14970 			}
14971 		} else if (probe->dofpr_noffs == 0) {
14972 			dtrace_dof_error(dof, "zero probe offsets");
14973 			return (-1);
14974 		}
14975 
14976 		if (probe->dofpr_argidx + probe->dofpr_xargc <
14977 		    probe->dofpr_argidx ||
14978 		    (probe->dofpr_argidx + probe->dofpr_xargc) *
14979 		    arg_sec->dofs_entsize > arg_sec->dofs_size) {
14980 			dtrace_dof_error(dof, "invalid args");
14981 			return (-1);
14982 		}
14983 
14984 		typeidx = probe->dofpr_nargv;
14985 		typestr = strtab + probe->dofpr_nargv;
14986 		for (k = 0; k < probe->dofpr_nargc; k++) {
14987 			if (typeidx >= str_sec->dofs_size) {
14988 				dtrace_dof_error(dof, "bad "
14989 				    "native argument type");
14990 				return (-1);
14991 			}
14992 
14993 			typesz = strlen(typestr) + 1;
14994 			if (typesz > DTRACE_ARGTYPELEN) {
14995 				dtrace_dof_error(dof, "native "
14996 				    "argument type too long");
14997 				return (-1);
14998 			}
14999 			typeidx += typesz;
15000 			typestr += typesz;
15001 		}
15002 
15003 		typeidx = probe->dofpr_xargv;
15004 		typestr = strtab + probe->dofpr_xargv;
15005 		for (k = 0; k < probe->dofpr_xargc; k++) {
15006 			if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) {
15007 				dtrace_dof_error(dof, "bad "
15008 				    "native argument index");
15009 				return (-1);
15010 			}
15011 
15012 			if (typeidx >= str_sec->dofs_size) {
15013 				dtrace_dof_error(dof, "bad "
15014 				    "translated argument type");
15015 				return (-1);
15016 			}
15017 
15018 			typesz = strlen(typestr) + 1;
15019 			if (typesz > DTRACE_ARGTYPELEN) {
15020 				dtrace_dof_error(dof, "translated argument "
15021 				    "type too long");
15022 				return (-1);
15023 			}
15024 
15025 			typeidx += typesz;
15026 			typestr += typesz;
15027 		}
15028 	}
15029 
15030 	return (0);
15031 }
15032 
15033 static int
dtrace_helper_slurp(dof_hdr_t * dof,dof_helper_t * dhp)15034 dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp)
15035 {
15036 	dtrace_helpers_t *help;
15037 	dtrace_vstate_t *vstate;
15038 	dtrace_enabling_t *enab = NULL;
15039 	int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1;
15040 	uintptr_t daddr = (uintptr_t)dof;
15041 
15042 	ASSERT(MUTEX_HELD(&dtrace_lock));
15043 
15044 	if ((help = curproc->p_dtrace_helpers) == NULL)
15045 		help = dtrace_helpers_create(curproc);
15046 
15047 	vstate = &help->dthps_vstate;
15048 
15049 	if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab,
15050 	    dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) {
15051 		dtrace_dof_destroy(dof);
15052 		return (rv);
15053 	}
15054 
15055 	/*
15056 	 * Look for helper providers and validate their descriptions.
15057 	 */
15058 	if (dhp != NULL) {
15059 		for (i = 0; i < dof->dofh_secnum; i++) {
15060 			dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
15061 			    dof->dofh_secoff + i * dof->dofh_secsize);
15062 
15063 			if (sec->dofs_type != DOF_SECT_PROVIDER)
15064 				continue;
15065 
15066 			if (dtrace_helper_provider_validate(dof, sec) != 0) {
15067 				dtrace_enabling_destroy(enab);
15068 				dtrace_dof_destroy(dof);
15069 				return (-1);
15070 			}
15071 
15072 			nprovs++;
15073 		}
15074 	}
15075 
15076 	/*
15077 	 * Now we need to walk through the ECB descriptions in the enabling.
15078 	 */
15079 	for (i = 0; i < enab->dten_ndesc; i++) {
15080 		dtrace_ecbdesc_t *ep = enab->dten_desc[i];
15081 		dtrace_probedesc_t *desc = &ep->dted_probe;
15082 
15083 		if (strcmp(desc->dtpd_provider, "dtrace") != 0)
15084 			continue;
15085 
15086 		if (strcmp(desc->dtpd_mod, "helper") != 0)
15087 			continue;
15088 
15089 		if (strcmp(desc->dtpd_func, "ustack") != 0)
15090 			continue;
15091 
15092 		if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK,
15093 		    ep)) != 0) {
15094 			/*
15095 			 * Adding this helper action failed -- we are now going
15096 			 * to rip out the entire generation and return failure.
15097 			 */
15098 			(void) dtrace_helper_destroygen(help->dthps_generation);
15099 			dtrace_enabling_destroy(enab);
15100 			dtrace_dof_destroy(dof);
15101 			return (-1);
15102 		}
15103 
15104 		nhelpers++;
15105 	}
15106 
15107 	if (nhelpers < enab->dten_ndesc)
15108 		dtrace_dof_error(dof, "unmatched helpers");
15109 
15110 	gen = help->dthps_generation++;
15111 	dtrace_enabling_destroy(enab);
15112 
15113 	if (dhp != NULL && nprovs > 0) {
15114 		dhp->dofhp_dof = (uint64_t)(uintptr_t)dof;
15115 		if (dtrace_helper_provider_add(dhp, gen) == 0) {
15116 			mutex_exit(&dtrace_lock);
15117 			dtrace_helper_provider_register(curproc, help, dhp);
15118 			mutex_enter(&dtrace_lock);
15119 
15120 			destroy = 0;
15121 		}
15122 	}
15123 
15124 	if (destroy)
15125 		dtrace_dof_destroy(dof);
15126 
15127 	return (gen);
15128 }
15129 
15130 static dtrace_helpers_t *
dtrace_helpers_create(proc_t * p)15131 dtrace_helpers_create(proc_t *p)
15132 {
15133 	dtrace_helpers_t *help;
15134 
15135 	ASSERT(MUTEX_HELD(&dtrace_lock));
15136 	ASSERT(p->p_dtrace_helpers == NULL);
15137 
15138 	help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP);
15139 	help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) *
15140 	    DTRACE_NHELPER_ACTIONS, KM_SLEEP);
15141 
15142 	p->p_dtrace_helpers = help;
15143 	dtrace_helpers++;
15144 
15145 	return (help);
15146 }
15147 
15148 static void
dtrace_helpers_destroy(void)15149 dtrace_helpers_destroy(void)
15150 {
15151 	dtrace_helpers_t *help;
15152 	dtrace_vstate_t *vstate;
15153 	proc_t *p = curproc;
15154 	int i;
15155 
15156 	mutex_enter(&dtrace_lock);
15157 
15158 	ASSERT(p->p_dtrace_helpers != NULL);
15159 	ASSERT(dtrace_helpers > 0);
15160 
15161 	help = p->p_dtrace_helpers;
15162 	vstate = &help->dthps_vstate;
15163 
15164 	/*
15165 	 * We're now going to lose the help from this process.
15166 	 */
15167 	p->p_dtrace_helpers = NULL;
15168 	dtrace_sync();
15169 
15170 	/*
15171 	 * Destory the helper actions.
15172 	 */
15173 	for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
15174 		dtrace_helper_action_t *h, *next;
15175 
15176 		for (h = help->dthps_actions[i]; h != NULL; h = next) {
15177 			next = h->dtha_next;
15178 			dtrace_helper_action_destroy(h, vstate);
15179 			h = next;
15180 		}
15181 	}
15182 
15183 	mutex_exit(&dtrace_lock);
15184 
15185 	/*
15186 	 * Destroy the helper providers.
15187 	 */
15188 	if (help->dthps_maxprovs > 0) {
15189 		mutex_enter(&dtrace_meta_lock);
15190 		if (dtrace_meta_pid != NULL) {
15191 			ASSERT(dtrace_deferred_pid == NULL);
15192 
15193 			for (i = 0; i < help->dthps_nprovs; i++) {
15194 				dtrace_helper_provider_remove(
15195 				    &help->dthps_provs[i]->dthp_prov, p->p_pid);
15196 			}
15197 		} else {
15198 			mutex_enter(&dtrace_lock);
15199 			ASSERT(help->dthps_deferred == 0 ||
15200 			    help->dthps_next != NULL ||
15201 			    help->dthps_prev != NULL ||
15202 			    help == dtrace_deferred_pid);
15203 
15204 			/*
15205 			 * Remove the helper from the deferred list.
15206 			 */
15207 			if (help->dthps_next != NULL)
15208 				help->dthps_next->dthps_prev = help->dthps_prev;
15209 			if (help->dthps_prev != NULL)
15210 				help->dthps_prev->dthps_next = help->dthps_next;
15211 			if (dtrace_deferred_pid == help) {
15212 				dtrace_deferred_pid = help->dthps_next;
15213 				ASSERT(help->dthps_prev == NULL);
15214 			}
15215 
15216 			mutex_exit(&dtrace_lock);
15217 		}
15218 
15219 		mutex_exit(&dtrace_meta_lock);
15220 
15221 		for (i = 0; i < help->dthps_nprovs; i++) {
15222 			dtrace_helper_provider_destroy(help->dthps_provs[i]);
15223 		}
15224 
15225 		kmem_free(help->dthps_provs, help->dthps_maxprovs *
15226 		    sizeof (dtrace_helper_provider_t *));
15227 	}
15228 
15229 	mutex_enter(&dtrace_lock);
15230 
15231 	dtrace_vstate_fini(&help->dthps_vstate);
15232 	kmem_free(help->dthps_actions,
15233 	    sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS);
15234 	kmem_free(help, sizeof (dtrace_helpers_t));
15235 
15236 	--dtrace_helpers;
15237 	mutex_exit(&dtrace_lock);
15238 }
15239 
15240 static void
dtrace_helpers_duplicate(proc_t * from,proc_t * to)15241 dtrace_helpers_duplicate(proc_t *from, proc_t *to)
15242 {
15243 	dtrace_helpers_t *help, *newhelp;
15244 	dtrace_helper_action_t *helper, *new, *last;
15245 	dtrace_difo_t *dp;
15246 	dtrace_vstate_t *vstate;
15247 	int i, j, sz, hasprovs = 0;
15248 
15249 	mutex_enter(&dtrace_lock);
15250 	ASSERT(from->p_dtrace_helpers != NULL);
15251 	ASSERT(dtrace_helpers > 0);
15252 
15253 	help = from->p_dtrace_helpers;
15254 	newhelp = dtrace_helpers_create(to);
15255 	ASSERT(to->p_dtrace_helpers != NULL);
15256 
15257 	newhelp->dthps_generation = help->dthps_generation;
15258 	vstate = &newhelp->dthps_vstate;
15259 
15260 	/*
15261 	 * Duplicate the helper actions.
15262 	 */
15263 	for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
15264 		if ((helper = help->dthps_actions[i]) == NULL)
15265 			continue;
15266 
15267 		for (last = NULL; helper != NULL; helper = helper->dtha_next) {
15268 			new = kmem_zalloc(sizeof (dtrace_helper_action_t),
15269 			    KM_SLEEP);
15270 			new->dtha_generation = helper->dtha_generation;
15271 
15272 			if ((dp = helper->dtha_predicate) != NULL) {
15273 				dp = dtrace_difo_duplicate(dp, vstate);
15274 				new->dtha_predicate = dp;
15275 			}
15276 
15277 			new->dtha_nactions = helper->dtha_nactions;
15278 			sz = sizeof (dtrace_difo_t *) * new->dtha_nactions;
15279 			new->dtha_actions = kmem_alloc(sz, KM_SLEEP);
15280 
15281 			for (j = 0; j < new->dtha_nactions; j++) {
15282 				dtrace_difo_t *dp = helper->dtha_actions[j];
15283 
15284 				ASSERT(dp != NULL);
15285 				dp = dtrace_difo_duplicate(dp, vstate);
15286 				new->dtha_actions[j] = dp;
15287 			}
15288 
15289 			if (last != NULL) {
15290 				last->dtha_next = new;
15291 			} else {
15292 				newhelp->dthps_actions[i] = new;
15293 			}
15294 
15295 			last = new;
15296 		}
15297 	}
15298 
15299 	/*
15300 	 * Duplicate the helper providers and register them with the
15301 	 * DTrace framework.
15302 	 */
15303 	if (help->dthps_nprovs > 0) {
15304 		newhelp->dthps_nprovs = help->dthps_nprovs;
15305 		newhelp->dthps_maxprovs = help->dthps_nprovs;
15306 		newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs *
15307 		    sizeof (dtrace_helper_provider_t *), KM_SLEEP);
15308 		for (i = 0; i < newhelp->dthps_nprovs; i++) {
15309 			newhelp->dthps_provs[i] = help->dthps_provs[i];
15310 			newhelp->dthps_provs[i]->dthp_ref++;
15311 		}
15312 
15313 		hasprovs = 1;
15314 	}
15315 
15316 	mutex_exit(&dtrace_lock);
15317 
15318 	if (hasprovs)
15319 		dtrace_helper_provider_register(to, newhelp, NULL);
15320 }
15321 
15322 /*
15323  * DTrace Hook Functions
15324  */
15325 static void
dtrace_module_loaded(struct modctl * ctl)15326 dtrace_module_loaded(struct modctl *ctl)
15327 {
15328 	dtrace_provider_t *prv;
15329 
15330 	mutex_enter(&dtrace_provider_lock);
15331 	mutex_enter(&mod_lock);
15332 
15333 	ASSERT(ctl->mod_busy);
15334 
15335 	/*
15336 	 * We're going to call each providers per-module provide operation
15337 	 * specifying only this module.
15338 	 */
15339 	for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next)
15340 		prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
15341 
15342 	mutex_exit(&mod_lock);
15343 	mutex_exit(&dtrace_provider_lock);
15344 
15345 	/*
15346 	 * If we have any retained enablings, we need to match against them.
15347 	 * Enabling probes requires that cpu_lock be held, and we cannot hold
15348 	 * cpu_lock here -- it is legal for cpu_lock to be held when loading a
15349 	 * module.  (In particular, this happens when loading scheduling
15350 	 * classes.)  So if we have any retained enablings, we need to dispatch
15351 	 * our task queue to do the match for us.
15352 	 */
15353 	mutex_enter(&dtrace_lock);
15354 
15355 	if (dtrace_retained == NULL) {
15356 		mutex_exit(&dtrace_lock);
15357 		return;
15358 	}
15359 
15360 	(void) taskq_dispatch(dtrace_taskq,
15361 	    (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP);
15362 
15363 	mutex_exit(&dtrace_lock);
15364 
15365 	/*
15366 	 * And now, for a little heuristic sleaze:  in general, we want to
15367 	 * match modules as soon as they load.  However, we cannot guarantee
15368 	 * this, because it would lead us to the lock ordering violation
15369 	 * outlined above.  The common case, of course, is that cpu_lock is
15370 	 * _not_ held -- so we delay here for a clock tick, hoping that that's
15371 	 * long enough for the task queue to do its work.  If it's not, it's
15372 	 * not a serious problem -- it just means that the module that we
15373 	 * just loaded may not be immediately instrumentable.
15374 	 */
15375 	delay(1);
15376 }
15377 
15378 static void
dtrace_module_unloaded(struct modctl * ctl)15379 dtrace_module_unloaded(struct modctl *ctl)
15380 {
15381 	dtrace_probe_t template, *probe, *first, *next;
15382 	dtrace_provider_t *prov;
15383 
15384 	template.dtpr_mod = ctl->mod_modname;
15385 
15386 	mutex_enter(&dtrace_provider_lock);
15387 	mutex_enter(&mod_lock);
15388 	mutex_enter(&dtrace_lock);
15389 
15390 	if (dtrace_bymod == NULL) {
15391 		/*
15392 		 * The DTrace module is loaded (obviously) but not attached;
15393 		 * we don't have any work to do.
15394 		 */
15395 		mutex_exit(&dtrace_provider_lock);
15396 		mutex_exit(&mod_lock);
15397 		mutex_exit(&dtrace_lock);
15398 		return;
15399 	}
15400 
15401 	for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template);
15402 	    probe != NULL; probe = probe->dtpr_nextmod) {
15403 		if (probe->dtpr_ecb != NULL) {
15404 			mutex_exit(&dtrace_provider_lock);
15405 			mutex_exit(&mod_lock);
15406 			mutex_exit(&dtrace_lock);
15407 
15408 			/*
15409 			 * This shouldn't _actually_ be possible -- we're
15410 			 * unloading a module that has an enabled probe in it.
15411 			 * (It's normally up to the provider to make sure that
15412 			 * this can't happen.)  However, because dtps_enable()
15413 			 * doesn't have a failure mode, there can be an
15414 			 * enable/unload race.  Upshot:  we don't want to
15415 			 * assert, but we're not going to disable the
15416 			 * probe, either.
15417 			 */
15418 			if (dtrace_err_verbose) {
15419 				cmn_err(CE_WARN, "unloaded module '%s' had "
15420 				    "enabled probes", ctl->mod_modname);
15421 			}
15422 
15423 			return;
15424 		}
15425 	}
15426 
15427 	probe = first;
15428 
15429 	for (first = NULL; probe != NULL; probe = next) {
15430 		ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe);
15431 
15432 		dtrace_probes[probe->dtpr_id - 1] = NULL;
15433 
15434 		next = probe->dtpr_nextmod;
15435 		dtrace_hash_remove(dtrace_bymod, probe);
15436 		dtrace_hash_remove(dtrace_byfunc, probe);
15437 		dtrace_hash_remove(dtrace_byname, probe);
15438 
15439 		if (first == NULL) {
15440 			first = probe;
15441 			probe->dtpr_nextmod = NULL;
15442 		} else {
15443 			probe->dtpr_nextmod = first;
15444 			first = probe;
15445 		}
15446 	}
15447 
15448 	/*
15449 	 * We've removed all of the module's probes from the hash chains and
15450 	 * from the probe array.  Now issue a dtrace_sync() to be sure that
15451 	 * everyone has cleared out from any probe array processing.
15452 	 */
15453 	dtrace_sync();
15454 
15455 	for (probe = first; probe != NULL; probe = first) {
15456 		first = probe->dtpr_nextmod;
15457 		prov = probe->dtpr_provider;
15458 		prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id,
15459 		    probe->dtpr_arg);
15460 		kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
15461 		kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
15462 		kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
15463 		vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1);
15464 		kmem_free(probe, sizeof (dtrace_probe_t));
15465 	}
15466 
15467 	mutex_exit(&dtrace_lock);
15468 	mutex_exit(&mod_lock);
15469 	mutex_exit(&dtrace_provider_lock);
15470 }
15471 
15472 void
dtrace_suspend(void)15473 dtrace_suspend(void)
15474 {
15475 	dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend));
15476 }
15477 
15478 void
dtrace_resume(void)15479 dtrace_resume(void)
15480 {
15481 	dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume));
15482 }
15483 
15484 static int
dtrace_cpu_setup(cpu_setup_t what,processorid_t cpu)15485 dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu)
15486 {
15487 	ASSERT(MUTEX_HELD(&cpu_lock));
15488 	mutex_enter(&dtrace_lock);
15489 
15490 	switch (what) {
15491 	case CPU_CONFIG: {
15492 		dtrace_state_t *state;
15493 		dtrace_optval_t *opt, rs, c;
15494 
15495 		/*
15496 		 * For now, we only allocate a new buffer for anonymous state.
15497 		 */
15498 		if ((state = dtrace_anon.dta_state) == NULL)
15499 			break;
15500 
15501 		if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
15502 			break;
15503 
15504 		opt = state->dts_options;
15505 		c = opt[DTRACEOPT_CPU];
15506 
15507 		if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu)
15508 			break;
15509 
15510 		/*
15511 		 * Regardless of what the actual policy is, we're going to
15512 		 * temporarily set our resize policy to be manual.  We're
15513 		 * also going to temporarily set our CPU option to denote
15514 		 * the newly configured CPU.
15515 		 */
15516 		rs = opt[DTRACEOPT_BUFRESIZE];
15517 		opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL;
15518 		opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu;
15519 
15520 		(void) dtrace_state_buffers(state);
15521 
15522 		opt[DTRACEOPT_BUFRESIZE] = rs;
15523 		opt[DTRACEOPT_CPU] = c;
15524 
15525 		break;
15526 	}
15527 
15528 	case CPU_UNCONFIG:
15529 		/*
15530 		 * We don't free the buffer in the CPU_UNCONFIG case.  (The
15531 		 * buffer will be freed when the consumer exits.)
15532 		 */
15533 		break;
15534 
15535 	default:
15536 		break;
15537 	}
15538 
15539 	mutex_exit(&dtrace_lock);
15540 	return (0);
15541 }
15542 
15543 static void
dtrace_cpu_setup_initial(processorid_t cpu)15544 dtrace_cpu_setup_initial(processorid_t cpu)
15545 {
15546 	(void) dtrace_cpu_setup(CPU_CONFIG, cpu);
15547 }
15548 
15549 static void
dtrace_toxrange_add(uintptr_t base,uintptr_t limit)15550 dtrace_toxrange_add(uintptr_t base, uintptr_t limit)
15551 {
15552 	if (dtrace_toxranges >= dtrace_toxranges_max) {
15553 		int osize, nsize;
15554 		dtrace_toxrange_t *range;
15555 
15556 		osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15557 
15558 		if (osize == 0) {
15559 			ASSERT(dtrace_toxrange == NULL);
15560 			ASSERT(dtrace_toxranges_max == 0);
15561 			dtrace_toxranges_max = 1;
15562 		} else {
15563 			dtrace_toxranges_max <<= 1;
15564 		}
15565 
15566 		nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15567 		range = kmem_zalloc(nsize, KM_SLEEP);
15568 
15569 		if (dtrace_toxrange != NULL) {
15570 			ASSERT(osize != 0);
15571 			bcopy(dtrace_toxrange, range, osize);
15572 			kmem_free(dtrace_toxrange, osize);
15573 		}
15574 
15575 		dtrace_toxrange = range;
15576 	}
15577 
15578 	ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == NULL);
15579 	ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == NULL);
15580 
15581 	dtrace_toxrange[dtrace_toxranges].dtt_base = base;
15582 	dtrace_toxrange[dtrace_toxranges].dtt_limit = limit;
15583 	dtrace_toxranges++;
15584 }
15585 
15586 static void
dtrace_getf_barrier()15587 dtrace_getf_barrier()
15588 {
15589 	/*
15590 	 * When we have unprivileged (that is, non-DTRACE_CRV_KERNEL) enablings
15591 	 * that contain calls to getf(), this routine will be called on every
15592 	 * closef() before either the underlying vnode is released or the
15593 	 * file_t itself is freed.  By the time we are here, it is essential
15594 	 * that the file_t can no longer be accessed from a call to getf()
15595 	 * in probe context -- that assures that a dtrace_sync() can be used
15596 	 * to clear out any enablings referring to the old structures.
15597 	 */
15598 	if (curthread->t_procp->p_zone->zone_dtrace_getf != 0 ||
15599 	    kcred->cr_zone->zone_dtrace_getf != 0)
15600 		dtrace_sync();
15601 }
15602 
15603 /*
15604  * DTrace Driver Cookbook Functions
15605  */
15606 /*ARGSUSED*/
15607 static int
dtrace_attach(dev_info_t * devi,ddi_attach_cmd_t cmd)15608 dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
15609 {
15610 	dtrace_provider_id_t id;
15611 	dtrace_state_t *state = NULL;
15612 	dtrace_enabling_t *enab;
15613 
15614 	mutex_enter(&cpu_lock);
15615 	mutex_enter(&dtrace_provider_lock);
15616 	mutex_enter(&dtrace_lock);
15617 
15618 	if (ddi_soft_state_init(&dtrace_softstate,
15619 	    sizeof (dtrace_state_t), 0) != 0) {
15620 		cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state");
15621 		mutex_exit(&cpu_lock);
15622 		mutex_exit(&dtrace_provider_lock);
15623 		mutex_exit(&dtrace_lock);
15624 		return (DDI_FAILURE);
15625 	}
15626 
15627 	if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR,
15628 	    DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE ||
15629 	    ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR,
15630 	    DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) {
15631 		cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes");
15632 		ddi_remove_minor_node(devi, NULL);
15633 		ddi_soft_state_fini(&dtrace_softstate);
15634 		mutex_exit(&cpu_lock);
15635 		mutex_exit(&dtrace_provider_lock);
15636 		mutex_exit(&dtrace_lock);
15637 		return (DDI_FAILURE);
15638 	}
15639 
15640 	ddi_report_dev(devi);
15641 	dtrace_devi = devi;
15642 
15643 	dtrace_modload = dtrace_module_loaded;
15644 	dtrace_modunload = dtrace_module_unloaded;
15645 	dtrace_cpu_init = dtrace_cpu_setup_initial;
15646 	dtrace_helpers_cleanup = dtrace_helpers_destroy;
15647 	dtrace_helpers_fork = dtrace_helpers_duplicate;
15648 	dtrace_cpustart_init = dtrace_suspend;
15649 	dtrace_cpustart_fini = dtrace_resume;
15650 	dtrace_debugger_init = dtrace_suspend;
15651 	dtrace_debugger_fini = dtrace_resume;
15652 
15653 	register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
15654 
15655 	ASSERT(MUTEX_HELD(&cpu_lock));
15656 
15657 	dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1,
15658 	    NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
15659 	dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE,
15660 	    UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0,
15661 	    VM_SLEEP | VMC_IDENTIFIER);
15662 	dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri,
15663 	    1, INT_MAX, 0);
15664 
15665 	dtrace_state_cache = kmem_cache_create("dtrace_state_cache",
15666 	    sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN,
15667 	    NULL, NULL, NULL, NULL, NULL, 0);
15668 
15669 	ASSERT(MUTEX_HELD(&cpu_lock));
15670 	dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod),
15671 	    offsetof(dtrace_probe_t, dtpr_nextmod),
15672 	    offsetof(dtrace_probe_t, dtpr_prevmod));
15673 
15674 	dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func),
15675 	    offsetof(dtrace_probe_t, dtpr_nextfunc),
15676 	    offsetof(dtrace_probe_t, dtpr_prevfunc));
15677 
15678 	dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name),
15679 	    offsetof(dtrace_probe_t, dtpr_nextname),
15680 	    offsetof(dtrace_probe_t, dtpr_prevname));
15681 
15682 	if (dtrace_retain_max < 1) {
15683 		cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; "
15684 		    "setting to 1", dtrace_retain_max);
15685 		dtrace_retain_max = 1;
15686 	}
15687 
15688 	/*
15689 	 * Now discover our toxic ranges.
15690 	 */
15691 	dtrace_toxic_ranges(dtrace_toxrange_add);
15692 
15693 	/*
15694 	 * Before we register ourselves as a provider to our own framework,
15695 	 * we would like to assert that dtrace_provider is NULL -- but that's
15696 	 * not true if we were loaded as a dependency of a DTrace provider.
15697 	 * Once we've registered, we can assert that dtrace_provider is our
15698 	 * pseudo provider.
15699 	 */
15700 	(void) dtrace_register("dtrace", &dtrace_provider_attr,
15701 	    DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id);
15702 
15703 	ASSERT(dtrace_provider != NULL);
15704 	ASSERT((dtrace_provider_id_t)dtrace_provider == id);
15705 
15706 	dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t)
15707 	    dtrace_provider, NULL, NULL, "BEGIN", 0, NULL);
15708 	dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t)
15709 	    dtrace_provider, NULL, NULL, "END", 0, NULL);
15710 	dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t)
15711 	    dtrace_provider, NULL, NULL, "ERROR", 1, NULL);
15712 
15713 	dtrace_anon_property();
15714 	mutex_exit(&cpu_lock);
15715 
15716 	/*
15717 	 * If there are already providers, we must ask them to provide their
15718 	 * probes, and then match any anonymous enabling against them.  Note
15719 	 * that there should be no other retained enablings at this time:
15720 	 * the only retained enablings at this time should be the anonymous
15721 	 * enabling.
15722 	 */
15723 	if (dtrace_anon.dta_enabling != NULL) {
15724 		ASSERT(dtrace_retained == dtrace_anon.dta_enabling);
15725 
15726 		dtrace_enabling_provide(NULL);
15727 		state = dtrace_anon.dta_state;
15728 
15729 		/*
15730 		 * We couldn't hold cpu_lock across the above call to
15731 		 * dtrace_enabling_provide(), but we must hold it to actually
15732 		 * enable the probes.  We have to drop all of our locks, pick
15733 		 * up cpu_lock, and regain our locks before matching the
15734 		 * retained anonymous enabling.
15735 		 */
15736 		mutex_exit(&dtrace_lock);
15737 		mutex_exit(&dtrace_provider_lock);
15738 
15739 		mutex_enter(&cpu_lock);
15740 		mutex_enter(&dtrace_provider_lock);
15741 		mutex_enter(&dtrace_lock);
15742 
15743 		if ((enab = dtrace_anon.dta_enabling) != NULL)
15744 			(void) dtrace_enabling_match(enab, NULL);
15745 
15746 		mutex_exit(&cpu_lock);
15747 	}
15748 
15749 	mutex_exit(&dtrace_lock);
15750 	mutex_exit(&dtrace_provider_lock);
15751 
15752 	if (state != NULL) {
15753 		/*
15754 		 * If we created any anonymous state, set it going now.
15755 		 */
15756 		(void) dtrace_state_go(state, &dtrace_anon.dta_beganon);
15757 	}
15758 
15759 	return (DDI_SUCCESS);
15760 }
15761 
15762 /*ARGSUSED*/
15763 static int
dtrace_open(dev_t * devp,int flag,int otyp,cred_t * cred_p)15764 dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
15765 {
15766 	dtrace_state_t *state;
15767 	uint32_t priv;
15768 	uid_t uid;
15769 	zoneid_t zoneid;
15770 
15771 	if (getminor(*devp) == DTRACEMNRN_HELPER)
15772 		return (0);
15773 
15774 	/*
15775 	 * If this wasn't an open with the "helper" minor, then it must be
15776 	 * the "dtrace" minor.
15777 	 */
15778 	if (getminor(*devp) != DTRACEMNRN_DTRACE)
15779 		return (ENXIO);
15780 
15781 	/*
15782 	 * If no DTRACE_PRIV_* bits are set in the credential, then the
15783 	 * caller lacks sufficient permission to do anything with DTrace.
15784 	 */
15785 	dtrace_cred2priv(cred_p, &priv, &uid, &zoneid);
15786 	if (priv == DTRACE_PRIV_NONE)
15787 		return (EACCES);
15788 
15789 	/*
15790 	 * Ask all providers to provide all their probes.
15791 	 */
15792 	mutex_enter(&dtrace_provider_lock);
15793 	dtrace_probe_provide(NULL, NULL);
15794 	mutex_exit(&dtrace_provider_lock);
15795 
15796 	mutex_enter(&cpu_lock);
15797 	mutex_enter(&dtrace_lock);
15798 	dtrace_opens++;
15799 	dtrace_membar_producer();
15800 
15801 	/*
15802 	 * If the kernel debugger is active (that is, if the kernel debugger
15803 	 * modified text in some way), we won't allow the open.
15804 	 */
15805 	if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
15806 		dtrace_opens--;
15807 		mutex_exit(&cpu_lock);
15808 		mutex_exit(&dtrace_lock);
15809 		return (EBUSY);
15810 	}
15811 
15812 	if (dtrace_helptrace_enable && dtrace_helptrace_buffer == NULL) {
15813 		/*
15814 		 * If DTrace helper tracing is enabled, we need to allocate the
15815 		 * trace buffer and initialize the values.
15816 		 */
15817 		dtrace_helptrace_buffer =
15818 		    kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP);
15819 		dtrace_helptrace_next = 0;
15820 		dtrace_helptrace_wrapped = 0;
15821 		dtrace_helptrace_enable = 0;
15822 	}
15823 
15824 	state = dtrace_state_create(devp, cred_p);
15825 	mutex_exit(&cpu_lock);
15826 
15827 	if (state == NULL) {
15828 		if (--dtrace_opens == 0 && dtrace_anon.dta_enabling == NULL)
15829 			(void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15830 		mutex_exit(&dtrace_lock);
15831 		return (EAGAIN);
15832 	}
15833 
15834 	mutex_exit(&dtrace_lock);
15835 
15836 	return (0);
15837 }
15838 
15839 /*ARGSUSED*/
15840 static int
dtrace_close(dev_t dev,int flag,int otyp,cred_t * cred_p)15841 dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
15842 {
15843 	minor_t minor = getminor(dev);
15844 	dtrace_state_t *state;
15845 	dtrace_helptrace_t *buf = NULL;
15846 
15847 	if (minor == DTRACEMNRN_HELPER)
15848 		return (0);
15849 
15850 	state = ddi_get_soft_state(dtrace_softstate, minor);
15851 
15852 	mutex_enter(&cpu_lock);
15853 	mutex_enter(&dtrace_lock);
15854 
15855 	if (state->dts_anon) {
15856 		/*
15857 		 * There is anonymous state. Destroy that first.
15858 		 */
15859 		ASSERT(dtrace_anon.dta_state == NULL);
15860 		dtrace_state_destroy(state->dts_anon);
15861 	}
15862 
15863 	if (dtrace_helptrace_disable) {
15864 		/*
15865 		 * If we have been told to disable helper tracing, set the
15866 		 * buffer to NULL before calling into dtrace_state_destroy();
15867 		 * we take advantage of its dtrace_sync() to know that no
15868 		 * CPU is in probe context with enabled helper tracing
15869 		 * after it returns.
15870 		 */
15871 		buf = dtrace_helptrace_buffer;
15872 		dtrace_helptrace_buffer = NULL;
15873 	}
15874 
15875 	dtrace_state_destroy(state);
15876 	ASSERT(dtrace_opens > 0);
15877 
15878 	/*
15879 	 * Only relinquish control of the kernel debugger interface when there
15880 	 * are no consumers and no anonymous enablings.
15881 	 */
15882 	if (--dtrace_opens == 0 && dtrace_anon.dta_enabling == NULL)
15883 		(void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15884 
15885 	if (buf != NULL) {
15886 		kmem_free(buf, dtrace_helptrace_bufsize);
15887 		dtrace_helptrace_disable = 0;
15888 	}
15889 
15890 	mutex_exit(&dtrace_lock);
15891 	mutex_exit(&cpu_lock);
15892 
15893 	return (0);
15894 }
15895 
15896 /*ARGSUSED*/
15897 static int
dtrace_ioctl_helper(int cmd,intptr_t arg,int * rv)15898 dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv)
15899 {
15900 	int rval;
15901 	dof_helper_t help, *dhp = NULL;
15902 
15903 	switch (cmd) {
15904 	case DTRACEHIOC_ADDDOF:
15905 		if (copyin((void *)arg, &help, sizeof (help)) != 0) {
15906 			dtrace_dof_error(NULL, "failed to copyin DOF helper");
15907 			return (EFAULT);
15908 		}
15909 
15910 		dhp = &help;
15911 		arg = (intptr_t)help.dofhp_dof;
15912 		/*FALLTHROUGH*/
15913 
15914 	case DTRACEHIOC_ADD: {
15915 		dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval);
15916 
15917 		if (dof == NULL)
15918 			return (rval);
15919 
15920 		mutex_enter(&dtrace_lock);
15921 
15922 		/*
15923 		 * dtrace_helper_slurp() takes responsibility for the dof --
15924 		 * it may free it now or it may save it and free it later.
15925 		 */
15926 		if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) {
15927 			*rv = rval;
15928 			rval = 0;
15929 		} else {
15930 			rval = EINVAL;
15931 		}
15932 
15933 		mutex_exit(&dtrace_lock);
15934 		return (rval);
15935 	}
15936 
15937 	case DTRACEHIOC_REMOVE: {
15938 		mutex_enter(&dtrace_lock);
15939 		rval = dtrace_helper_destroygen(arg);
15940 		mutex_exit(&dtrace_lock);
15941 
15942 		return (rval);
15943 	}
15944 
15945 	default:
15946 		break;
15947 	}
15948 
15949 	return (ENOTTY);
15950 }
15951 
15952 /*ARGSUSED*/
15953 static int
dtrace_ioctl(dev_t dev,int cmd,intptr_t arg,int md,cred_t * cr,int * rv)15954 dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
15955 {
15956 	minor_t minor = getminor(dev);
15957 	dtrace_state_t *state;
15958 	int rval;
15959 
15960 	if (minor == DTRACEMNRN_HELPER)
15961 		return (dtrace_ioctl_helper(cmd, arg, rv));
15962 
15963 	state = ddi_get_soft_state(dtrace_softstate, minor);
15964 
15965 	if (state->dts_anon) {
15966 		ASSERT(dtrace_anon.dta_state == NULL);
15967 		state = state->dts_anon;
15968 	}
15969 
15970 	switch (cmd) {
15971 	case DTRACEIOC_PROVIDER: {
15972 		dtrace_providerdesc_t pvd;
15973 		dtrace_provider_t *pvp;
15974 
15975 		if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0)
15976 			return (EFAULT);
15977 
15978 		pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0';
15979 		mutex_enter(&dtrace_provider_lock);
15980 
15981 		for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) {
15982 			if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0)
15983 				break;
15984 		}
15985 
15986 		mutex_exit(&dtrace_provider_lock);
15987 
15988 		if (pvp == NULL)
15989 			return (ESRCH);
15990 
15991 		bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t));
15992 		bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t));
15993 		if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0)
15994 			return (EFAULT);
15995 
15996 		return (0);
15997 	}
15998 
15999 	case DTRACEIOC_EPROBE: {
16000 		dtrace_eprobedesc_t epdesc;
16001 		dtrace_ecb_t *ecb;
16002 		dtrace_action_t *act;
16003 		void *buf;
16004 		size_t size;
16005 		uintptr_t dest;
16006 		int nrecs;
16007 
16008 		if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0)
16009 			return (EFAULT);
16010 
16011 		mutex_enter(&dtrace_lock);
16012 
16013 		if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) {
16014 			mutex_exit(&dtrace_lock);
16015 			return (EINVAL);
16016 		}
16017 
16018 		if (ecb->dte_probe == NULL) {
16019 			mutex_exit(&dtrace_lock);
16020 			return (EINVAL);
16021 		}
16022 
16023 		epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id;
16024 		epdesc.dtepd_uarg = ecb->dte_uarg;
16025 		epdesc.dtepd_size = ecb->dte_size;
16026 
16027 		nrecs = epdesc.dtepd_nrecs;
16028 		epdesc.dtepd_nrecs = 0;
16029 		for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
16030 			if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
16031 				continue;
16032 
16033 			epdesc.dtepd_nrecs++;
16034 		}
16035 
16036 		/*
16037 		 * Now that we have the size, we need to allocate a temporary
16038 		 * buffer in which to store the complete description.  We need
16039 		 * the temporary buffer to be able to drop dtrace_lock()
16040 		 * across the copyout(), below.
16041 		 */
16042 		size = sizeof (dtrace_eprobedesc_t) +
16043 		    (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t));
16044 
16045 		buf = kmem_alloc(size, KM_SLEEP);
16046 		dest = (uintptr_t)buf;
16047 
16048 		bcopy(&epdesc, (void *)dest, sizeof (epdesc));
16049 		dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]);
16050 
16051 		for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
16052 			if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
16053 				continue;
16054 
16055 			if (nrecs-- == 0)
16056 				break;
16057 
16058 			bcopy(&act->dta_rec, (void *)dest,
16059 			    sizeof (dtrace_recdesc_t));
16060 			dest += sizeof (dtrace_recdesc_t);
16061 		}
16062 
16063 		mutex_exit(&dtrace_lock);
16064 
16065 		if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
16066 			kmem_free(buf, size);
16067 			return (EFAULT);
16068 		}
16069 
16070 		kmem_free(buf, size);
16071 		return (0);
16072 	}
16073 
16074 	case DTRACEIOC_AGGDESC: {
16075 		dtrace_aggdesc_t aggdesc;
16076 		dtrace_action_t *act;
16077 		dtrace_aggregation_t *agg;
16078 		int nrecs;
16079 		uint32_t offs;
16080 		dtrace_recdesc_t *lrec;
16081 		void *buf;
16082 		size_t size;
16083 		uintptr_t dest;
16084 
16085 		if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0)
16086 			return (EFAULT);
16087 
16088 		mutex_enter(&dtrace_lock);
16089 
16090 		if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) {
16091 			mutex_exit(&dtrace_lock);
16092 			return (EINVAL);
16093 		}
16094 
16095 		aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid;
16096 
16097 		nrecs = aggdesc.dtagd_nrecs;
16098 		aggdesc.dtagd_nrecs = 0;
16099 
16100 		offs = agg->dtag_base;
16101 		lrec = &agg->dtag_action.dta_rec;
16102 		aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs;
16103 
16104 		for (act = agg->dtag_first; ; act = act->dta_next) {
16105 			ASSERT(act->dta_intuple ||
16106 			    DTRACEACT_ISAGG(act->dta_kind));
16107 
16108 			/*
16109 			 * If this action has a record size of zero, it
16110 			 * denotes an argument to the aggregating action.
16111 			 * Because the presence of this record doesn't (or
16112 			 * shouldn't) affect the way the data is interpreted,
16113 			 * we don't copy it out to save user-level the
16114 			 * confusion of dealing with a zero-length record.
16115 			 */
16116 			if (act->dta_rec.dtrd_size == 0) {
16117 				ASSERT(agg->dtag_hasarg);
16118 				continue;
16119 			}
16120 
16121 			aggdesc.dtagd_nrecs++;
16122 
16123 			if (act == &agg->dtag_action)
16124 				break;
16125 		}
16126 
16127 		/*
16128 		 * Now that we have the size, we need to allocate a temporary
16129 		 * buffer in which to store the complete description.  We need
16130 		 * the temporary buffer to be able to drop dtrace_lock()
16131 		 * across the copyout(), below.
16132 		 */
16133 		size = sizeof (dtrace_aggdesc_t) +
16134 		    (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t));
16135 
16136 		buf = kmem_alloc(size, KM_SLEEP);
16137 		dest = (uintptr_t)buf;
16138 
16139 		bcopy(&aggdesc, (void *)dest, sizeof (aggdesc));
16140 		dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]);
16141 
16142 		for (act = agg->dtag_first; ; act = act->dta_next) {
16143 			dtrace_recdesc_t rec = act->dta_rec;
16144 
16145 			/*
16146 			 * See the comment in the above loop for why we pass
16147 			 * over zero-length records.
16148 			 */
16149 			if (rec.dtrd_size == 0) {
16150 				ASSERT(agg->dtag_hasarg);
16151 				continue;
16152 			}
16153 
16154 			if (nrecs-- == 0)
16155 				break;
16156 
16157 			rec.dtrd_offset -= offs;
16158 			bcopy(&rec, (void *)dest, sizeof (rec));
16159 			dest += sizeof (dtrace_recdesc_t);
16160 
16161 			if (act == &agg->dtag_action)
16162 				break;
16163 		}
16164 
16165 		mutex_exit(&dtrace_lock);
16166 
16167 		if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
16168 			kmem_free(buf, size);
16169 			return (EFAULT);
16170 		}
16171 
16172 		kmem_free(buf, size);
16173 		return (0);
16174 	}
16175 
16176 	case DTRACEIOC_ENABLE: {
16177 		dof_hdr_t *dof;
16178 		dtrace_enabling_t *enab = NULL;
16179 		dtrace_vstate_t *vstate;
16180 		int err = 0;
16181 
16182 		*rv = 0;
16183 
16184 		/*
16185 		 * If a NULL argument has been passed, we take this as our
16186 		 * cue to reevaluate our enablings.
16187 		 */
16188 		if (arg == NULL) {
16189 			dtrace_enabling_matchall();
16190 
16191 			return (0);
16192 		}
16193 
16194 		if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL)
16195 			return (rval);
16196 
16197 		mutex_enter(&cpu_lock);
16198 		mutex_enter(&dtrace_lock);
16199 		vstate = &state->dts_vstate;
16200 
16201 		if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
16202 			mutex_exit(&dtrace_lock);
16203 			mutex_exit(&cpu_lock);
16204 			dtrace_dof_destroy(dof);
16205 			return (EBUSY);
16206 		}
16207 
16208 		if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) {
16209 			mutex_exit(&dtrace_lock);
16210 			mutex_exit(&cpu_lock);
16211 			dtrace_dof_destroy(dof);
16212 			return (EINVAL);
16213 		}
16214 
16215 		if ((rval = dtrace_dof_options(dof, state)) != 0) {
16216 			dtrace_enabling_destroy(enab);
16217 			mutex_exit(&dtrace_lock);
16218 			mutex_exit(&cpu_lock);
16219 			dtrace_dof_destroy(dof);
16220 			return (rval);
16221 		}
16222 
16223 		if ((err = dtrace_enabling_match(enab, rv)) == 0) {
16224 			err = dtrace_enabling_retain(enab);
16225 		} else {
16226 			dtrace_enabling_destroy(enab);
16227 		}
16228 
16229 		mutex_exit(&cpu_lock);
16230 		mutex_exit(&dtrace_lock);
16231 		dtrace_dof_destroy(dof);
16232 
16233 		return (err);
16234 	}
16235 
16236 	case DTRACEIOC_REPLICATE: {
16237 		dtrace_repldesc_t desc;
16238 		dtrace_probedesc_t *match = &desc.dtrpd_match;
16239 		dtrace_probedesc_t *create = &desc.dtrpd_create;
16240 		int err;
16241 
16242 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16243 			return (EFAULT);
16244 
16245 		match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
16246 		match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
16247 		match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
16248 		match->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
16249 
16250 		create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
16251 		create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
16252 		create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
16253 		create->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
16254 
16255 		mutex_enter(&dtrace_lock);
16256 		err = dtrace_enabling_replicate(state, match, create);
16257 		mutex_exit(&dtrace_lock);
16258 
16259 		return (err);
16260 	}
16261 
16262 	case DTRACEIOC_PROBEMATCH:
16263 	case DTRACEIOC_PROBES: {
16264 		dtrace_probe_t *probe = NULL;
16265 		dtrace_probedesc_t desc;
16266 		dtrace_probekey_t pkey;
16267 		dtrace_id_t i;
16268 		int m = 0;
16269 		uint32_t priv;
16270 		uid_t uid;
16271 		zoneid_t zoneid;
16272 
16273 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16274 			return (EFAULT);
16275 
16276 		desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
16277 		desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
16278 		desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
16279 		desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0';
16280 
16281 		/*
16282 		 * Before we attempt to match this probe, we want to give
16283 		 * all providers the opportunity to provide it.
16284 		 */
16285 		if (desc.dtpd_id == DTRACE_IDNONE) {
16286 			mutex_enter(&dtrace_provider_lock);
16287 			dtrace_probe_provide(&desc, NULL);
16288 			mutex_exit(&dtrace_provider_lock);
16289 			desc.dtpd_id++;
16290 		}
16291 
16292 		if (cmd == DTRACEIOC_PROBEMATCH)  {
16293 			dtrace_probekey(&desc, &pkey);
16294 			pkey.dtpk_id = DTRACE_IDNONE;
16295 		}
16296 
16297 		dtrace_cred2priv(cr, &priv, &uid, &zoneid);
16298 
16299 		mutex_enter(&dtrace_lock);
16300 
16301 		if (cmd == DTRACEIOC_PROBEMATCH) {
16302 			for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
16303 				if ((probe = dtrace_probes[i - 1]) != NULL &&
16304 				    (m = dtrace_match_probe(probe, &pkey,
16305 				    priv, uid, zoneid)) != 0)
16306 					break;
16307 			}
16308 
16309 			if (m < 0) {
16310 				mutex_exit(&dtrace_lock);
16311 				return (EINVAL);
16312 			}
16313 
16314 		} else {
16315 			for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
16316 				if ((probe = dtrace_probes[i - 1]) != NULL &&
16317 				    dtrace_match_priv(probe, priv, uid, zoneid))
16318 					break;
16319 			}
16320 		}
16321 
16322 		if (probe == NULL) {
16323 			mutex_exit(&dtrace_lock);
16324 			return (ESRCH);
16325 		}
16326 
16327 		dtrace_probe_description(probe, &desc);
16328 		mutex_exit(&dtrace_lock);
16329 
16330 		if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16331 			return (EFAULT);
16332 
16333 		return (0);
16334 	}
16335 
16336 	case DTRACEIOC_PROBEARG: {
16337 		dtrace_argdesc_t desc;
16338 		dtrace_probe_t *probe;
16339 		dtrace_provider_t *prov;
16340 
16341 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16342 			return (EFAULT);
16343 
16344 		if (desc.dtargd_id == DTRACE_IDNONE)
16345 			return (EINVAL);
16346 
16347 		if (desc.dtargd_ndx == DTRACE_ARGNONE)
16348 			return (EINVAL);
16349 
16350 		mutex_enter(&dtrace_provider_lock);
16351 		mutex_enter(&mod_lock);
16352 		mutex_enter(&dtrace_lock);
16353 
16354 		if (desc.dtargd_id > dtrace_nprobes) {
16355 			mutex_exit(&dtrace_lock);
16356 			mutex_exit(&mod_lock);
16357 			mutex_exit(&dtrace_provider_lock);
16358 			return (EINVAL);
16359 		}
16360 
16361 		if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) {
16362 			mutex_exit(&dtrace_lock);
16363 			mutex_exit(&mod_lock);
16364 			mutex_exit(&dtrace_provider_lock);
16365 			return (EINVAL);
16366 		}
16367 
16368 		mutex_exit(&dtrace_lock);
16369 
16370 		prov = probe->dtpr_provider;
16371 
16372 		if (prov->dtpv_pops.dtps_getargdesc == NULL) {
16373 			/*
16374 			 * There isn't any typed information for this probe.
16375 			 * Set the argument number to DTRACE_ARGNONE.
16376 			 */
16377 			desc.dtargd_ndx = DTRACE_ARGNONE;
16378 		} else {
16379 			desc.dtargd_native[0] = '\0';
16380 			desc.dtargd_xlate[0] = '\0';
16381 			desc.dtargd_mapping = desc.dtargd_ndx;
16382 
16383 			prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg,
16384 			    probe->dtpr_id, probe->dtpr_arg, &desc);
16385 		}
16386 
16387 		mutex_exit(&mod_lock);
16388 		mutex_exit(&dtrace_provider_lock);
16389 
16390 		if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16391 			return (EFAULT);
16392 
16393 		return (0);
16394 	}
16395 
16396 	case DTRACEIOC_GO: {
16397 		processorid_t cpuid;
16398 		rval = dtrace_state_go(state, &cpuid);
16399 
16400 		if (rval != 0)
16401 			return (rval);
16402 
16403 		if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
16404 			return (EFAULT);
16405 
16406 		return (0);
16407 	}
16408 
16409 	case DTRACEIOC_STOP: {
16410 		processorid_t cpuid;
16411 
16412 		mutex_enter(&dtrace_lock);
16413 		rval = dtrace_state_stop(state, &cpuid);
16414 		mutex_exit(&dtrace_lock);
16415 
16416 		if (rval != 0)
16417 			return (rval);
16418 
16419 		if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
16420 			return (EFAULT);
16421 
16422 		return (0);
16423 	}
16424 
16425 	case DTRACEIOC_DOFGET: {
16426 		dof_hdr_t hdr, *dof;
16427 		uint64_t len;
16428 
16429 		if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0)
16430 			return (EFAULT);
16431 
16432 		mutex_enter(&dtrace_lock);
16433 		dof = dtrace_dof_create(state);
16434 		mutex_exit(&dtrace_lock);
16435 
16436 		len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz);
16437 		rval = copyout(dof, (void *)arg, len);
16438 		dtrace_dof_destroy(dof);
16439 
16440 		return (rval == 0 ? 0 : EFAULT);
16441 	}
16442 
16443 	case DTRACEIOC_AGGSNAP:
16444 	case DTRACEIOC_BUFSNAP: {
16445 		dtrace_bufdesc_t desc;
16446 		caddr_t cached;
16447 		dtrace_buffer_t *buf;
16448 
16449 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16450 			return (EFAULT);
16451 
16452 		if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU)
16453 			return (EINVAL);
16454 
16455 		mutex_enter(&dtrace_lock);
16456 
16457 		if (cmd == DTRACEIOC_BUFSNAP) {
16458 			buf = &state->dts_buffer[desc.dtbd_cpu];
16459 		} else {
16460 			buf = &state->dts_aggbuffer[desc.dtbd_cpu];
16461 		}
16462 
16463 		if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) {
16464 			size_t sz = buf->dtb_offset;
16465 
16466 			if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) {
16467 				mutex_exit(&dtrace_lock);
16468 				return (EBUSY);
16469 			}
16470 
16471 			/*
16472 			 * If this buffer has already been consumed, we're
16473 			 * going to indicate that there's nothing left here
16474 			 * to consume.
16475 			 */
16476 			if (buf->dtb_flags & DTRACEBUF_CONSUMED) {
16477 				mutex_exit(&dtrace_lock);
16478 
16479 				desc.dtbd_size = 0;
16480 				desc.dtbd_drops = 0;
16481 				desc.dtbd_errors = 0;
16482 				desc.dtbd_oldest = 0;
16483 				sz = sizeof (desc);
16484 
16485 				if (copyout(&desc, (void *)arg, sz) != 0)
16486 					return (EFAULT);
16487 
16488 				return (0);
16489 			}
16490 
16491 			/*
16492 			 * If this is a ring buffer that has wrapped, we want
16493 			 * to copy the whole thing out.
16494 			 */
16495 			if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
16496 				dtrace_buffer_polish(buf);
16497 				sz = buf->dtb_size;
16498 			}
16499 
16500 			if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) {
16501 				mutex_exit(&dtrace_lock);
16502 				return (EFAULT);
16503 			}
16504 
16505 			desc.dtbd_size = sz;
16506 			desc.dtbd_drops = buf->dtb_drops;
16507 			desc.dtbd_errors = buf->dtb_errors;
16508 			desc.dtbd_oldest = buf->dtb_xamot_offset;
16509 			desc.dtbd_timestamp = dtrace_gethrtime();
16510 
16511 			mutex_exit(&dtrace_lock);
16512 
16513 			if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16514 				return (EFAULT);
16515 
16516 			buf->dtb_flags |= DTRACEBUF_CONSUMED;
16517 
16518 			return (0);
16519 		}
16520 
16521 		if (buf->dtb_tomax == NULL) {
16522 			ASSERT(buf->dtb_xamot == NULL);
16523 			mutex_exit(&dtrace_lock);
16524 			return (ENOENT);
16525 		}
16526 
16527 		cached = buf->dtb_tomax;
16528 		ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
16529 
16530 		dtrace_xcall(desc.dtbd_cpu,
16531 		    (dtrace_xcall_t)dtrace_buffer_switch, buf);
16532 
16533 		state->dts_errors += buf->dtb_xamot_errors;
16534 
16535 		/*
16536 		 * If the buffers did not actually switch, then the cross call
16537 		 * did not take place -- presumably because the given CPU is
16538 		 * not in the ready set.  If this is the case, we'll return
16539 		 * ENOENT.
16540 		 */
16541 		if (buf->dtb_tomax == cached) {
16542 			ASSERT(buf->dtb_xamot != cached);
16543 			mutex_exit(&dtrace_lock);
16544 			return (ENOENT);
16545 		}
16546 
16547 		ASSERT(cached == buf->dtb_xamot);
16548 
16549 		/*
16550 		 * We have our snapshot; now copy it out.
16551 		 */
16552 		if (copyout(buf->dtb_xamot, desc.dtbd_data,
16553 		    buf->dtb_xamot_offset) != 0) {
16554 			mutex_exit(&dtrace_lock);
16555 			return (EFAULT);
16556 		}
16557 
16558 		desc.dtbd_size = buf->dtb_xamot_offset;
16559 		desc.dtbd_drops = buf->dtb_xamot_drops;
16560 		desc.dtbd_errors = buf->dtb_xamot_errors;
16561 		desc.dtbd_oldest = 0;
16562 		desc.dtbd_timestamp = buf->dtb_switched;
16563 
16564 		mutex_exit(&dtrace_lock);
16565 
16566 		/*
16567 		 * Finally, copy out the buffer description.
16568 		 */
16569 		if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16570 			return (EFAULT);
16571 
16572 		return (0);
16573 	}
16574 
16575 	case DTRACEIOC_CONF: {
16576 		dtrace_conf_t conf;
16577 
16578 		bzero(&conf, sizeof (conf));
16579 		conf.dtc_difversion = DIF_VERSION;
16580 		conf.dtc_difintregs = DIF_DIR_NREGS;
16581 		conf.dtc_diftupregs = DIF_DTR_NREGS;
16582 		conf.dtc_ctfmodel = CTF_MODEL_NATIVE;
16583 
16584 		if (copyout(&conf, (void *)arg, sizeof (conf)) != 0)
16585 			return (EFAULT);
16586 
16587 		return (0);
16588 	}
16589 
16590 	case DTRACEIOC_STATUS: {
16591 		dtrace_status_t stat;
16592 		dtrace_dstate_t *dstate;
16593 		int i, j;
16594 		uint64_t nerrs;
16595 
16596 		/*
16597 		 * See the comment in dtrace_state_deadman() for the reason
16598 		 * for setting dts_laststatus to INT64_MAX before setting
16599 		 * it to the correct value.
16600 		 */
16601 		state->dts_laststatus = INT64_MAX;
16602 		dtrace_membar_producer();
16603 		state->dts_laststatus = dtrace_gethrtime();
16604 
16605 		bzero(&stat, sizeof (stat));
16606 
16607 		mutex_enter(&dtrace_lock);
16608 
16609 		if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) {
16610 			mutex_exit(&dtrace_lock);
16611 			return (ENOENT);
16612 		}
16613 
16614 		if (state->dts_activity == DTRACE_ACTIVITY_DRAINING)
16615 			stat.dtst_exiting = 1;
16616 
16617 		nerrs = state->dts_errors;
16618 		dstate = &state->dts_vstate.dtvs_dynvars;
16619 
16620 		for (i = 0; i < NCPU; i++) {
16621 			dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i];
16622 
16623 			stat.dtst_dyndrops += dcpu->dtdsc_drops;
16624 			stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops;
16625 			stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops;
16626 
16627 			if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL)
16628 				stat.dtst_filled++;
16629 
16630 			nerrs += state->dts_buffer[i].dtb_errors;
16631 
16632 			for (j = 0; j < state->dts_nspeculations; j++) {
16633 				dtrace_speculation_t *spec;
16634 				dtrace_buffer_t *buf;
16635 
16636 				spec = &state->dts_speculations[j];
16637 				buf = &spec->dtsp_buffer[i];
16638 				stat.dtst_specdrops += buf->dtb_xamot_drops;
16639 			}
16640 		}
16641 
16642 		stat.dtst_specdrops_busy = state->dts_speculations_busy;
16643 		stat.dtst_specdrops_unavail = state->dts_speculations_unavail;
16644 		stat.dtst_stkstroverflows = state->dts_stkstroverflows;
16645 		stat.dtst_dblerrors = state->dts_dblerrors;
16646 		stat.dtst_killed =
16647 		    (state->dts_activity == DTRACE_ACTIVITY_KILLED);
16648 		stat.dtst_errors = nerrs;
16649 
16650 		mutex_exit(&dtrace_lock);
16651 
16652 		if (copyout(&stat, (void *)arg, sizeof (stat)) != 0)
16653 			return (EFAULT);
16654 
16655 		return (0);
16656 	}
16657 
16658 	case DTRACEIOC_FORMAT: {
16659 		dtrace_fmtdesc_t fmt;
16660 		char *str;
16661 		int len;
16662 
16663 		if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0)
16664 			return (EFAULT);
16665 
16666 		mutex_enter(&dtrace_lock);
16667 
16668 		if (fmt.dtfd_format == 0 ||
16669 		    fmt.dtfd_format > state->dts_nformats) {
16670 			mutex_exit(&dtrace_lock);
16671 			return (EINVAL);
16672 		}
16673 
16674 		/*
16675 		 * Format strings are allocated contiguously and they are
16676 		 * never freed; if a format index is less than the number
16677 		 * of formats, we can assert that the format map is non-NULL
16678 		 * and that the format for the specified index is non-NULL.
16679 		 */
16680 		ASSERT(state->dts_formats != NULL);
16681 		str = state->dts_formats[fmt.dtfd_format - 1];
16682 		ASSERT(str != NULL);
16683 
16684 		len = strlen(str) + 1;
16685 
16686 		if (len > fmt.dtfd_length) {
16687 			fmt.dtfd_length = len;
16688 
16689 			if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) {
16690 				mutex_exit(&dtrace_lock);
16691 				return (EINVAL);
16692 			}
16693 		} else {
16694 			if (copyout(str, fmt.dtfd_string, len) != 0) {
16695 				mutex_exit(&dtrace_lock);
16696 				return (EINVAL);
16697 			}
16698 		}
16699 
16700 		mutex_exit(&dtrace_lock);
16701 		return (0);
16702 	}
16703 
16704 	default:
16705 		break;
16706 	}
16707 
16708 	return (ENOTTY);
16709 }
16710 
16711 /*ARGSUSED*/
16712 static int
dtrace_detach(dev_info_t * dip,ddi_detach_cmd_t cmd)16713 dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
16714 {
16715 	dtrace_state_t *state;
16716 
16717 	switch (cmd) {
16718 	case DDI_DETACH:
16719 		break;
16720 
16721 	case DDI_SUSPEND:
16722 		return (DDI_SUCCESS);
16723 
16724 	default:
16725 		return (DDI_FAILURE);
16726 	}
16727 
16728 	mutex_enter(&cpu_lock);
16729 	mutex_enter(&dtrace_provider_lock);
16730 	mutex_enter(&dtrace_lock);
16731 
16732 	ASSERT(dtrace_opens == 0);
16733 
16734 	if (dtrace_helpers > 0) {
16735 		mutex_exit(&dtrace_provider_lock);
16736 		mutex_exit(&dtrace_lock);
16737 		mutex_exit(&cpu_lock);
16738 		return (DDI_FAILURE);
16739 	}
16740 
16741 	if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) {
16742 		mutex_exit(&dtrace_provider_lock);
16743 		mutex_exit(&dtrace_lock);
16744 		mutex_exit(&cpu_lock);
16745 		return (DDI_FAILURE);
16746 	}
16747 
16748 	dtrace_provider = NULL;
16749 
16750 	if ((state = dtrace_anon_grab()) != NULL) {
16751 		/*
16752 		 * If there were ECBs on this state, the provider should
16753 		 * have not been allowed to detach; assert that there is
16754 		 * none.
16755 		 */
16756 		ASSERT(state->dts_necbs == 0);
16757 		dtrace_state_destroy(state);
16758 
16759 		/*
16760 		 * If we're being detached with anonymous state, we need to
16761 		 * indicate to the kernel debugger that DTrace is now inactive.
16762 		 */
16763 		(void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
16764 	}
16765 
16766 	bzero(&dtrace_anon, sizeof (dtrace_anon_t));
16767 	unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
16768 	dtrace_cpu_init = NULL;
16769 	dtrace_helpers_cleanup = NULL;
16770 	dtrace_helpers_fork = NULL;
16771 	dtrace_cpustart_init = NULL;
16772 	dtrace_cpustart_fini = NULL;
16773 	dtrace_debugger_init = NULL;
16774 	dtrace_debugger_fini = NULL;
16775 	dtrace_modload = NULL;
16776 	dtrace_modunload = NULL;
16777 
16778 	ASSERT(dtrace_getf == 0);
16779 	ASSERT(dtrace_closef == NULL);
16780 
16781 	mutex_exit(&cpu_lock);
16782 
16783 	kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *));
16784 	dtrace_probes = NULL;
16785 	dtrace_nprobes = 0;
16786 
16787 	dtrace_hash_destroy(dtrace_bymod);
16788 	dtrace_hash_destroy(dtrace_byfunc);
16789 	dtrace_hash_destroy(dtrace_byname);
16790 	dtrace_bymod = NULL;
16791 	dtrace_byfunc = NULL;
16792 	dtrace_byname = NULL;
16793 
16794 	kmem_cache_destroy(dtrace_state_cache);
16795 	vmem_destroy(dtrace_minor);
16796 	vmem_destroy(dtrace_arena);
16797 
16798 	if (dtrace_toxrange != NULL) {
16799 		kmem_free(dtrace_toxrange,
16800 		    dtrace_toxranges_max * sizeof (dtrace_toxrange_t));
16801 		dtrace_toxrange = NULL;
16802 		dtrace_toxranges = 0;
16803 		dtrace_toxranges_max = 0;
16804 	}
16805 
16806 	ddi_remove_minor_node(dtrace_devi, NULL);
16807 	dtrace_devi = NULL;
16808 
16809 	ddi_soft_state_fini(&dtrace_softstate);
16810 
16811 	ASSERT(dtrace_vtime_references == 0);
16812 	ASSERT(dtrace_opens == 0);
16813 	ASSERT(dtrace_retained == NULL);
16814 
16815 	mutex_exit(&dtrace_lock);
16816 	mutex_exit(&dtrace_provider_lock);
16817 
16818 	/*
16819 	 * We don't destroy the task queue until after we have dropped our
16820 	 * locks (taskq_destroy() may block on running tasks).  To prevent
16821 	 * attempting to do work after we have effectively detached but before
16822 	 * the task queue has been destroyed, all tasks dispatched via the
16823 	 * task queue must check that DTrace is still attached before
16824 	 * performing any operation.
16825 	 */
16826 	taskq_destroy(dtrace_taskq);
16827 	dtrace_taskq = NULL;
16828 
16829 	return (DDI_SUCCESS);
16830 }
16831 
16832 /*ARGSUSED*/
16833 static int
dtrace_info(dev_info_t * dip,ddi_info_cmd_t infocmd,void * arg,void ** result)16834 dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
16835 {
16836 	int error;
16837 
16838 	switch (infocmd) {
16839 	case DDI_INFO_DEVT2DEVINFO:
16840 		*result = (void *)dtrace_devi;
16841 		error = DDI_SUCCESS;
16842 		break;
16843 	case DDI_INFO_DEVT2INSTANCE:
16844 		*result = (void *)0;
16845 		error = DDI_SUCCESS;
16846 		break;
16847 	default:
16848 		error = DDI_FAILURE;
16849 	}
16850 	return (error);
16851 }
16852 
16853 static struct cb_ops dtrace_cb_ops = {
16854 	dtrace_open,		/* open */
16855 	dtrace_close,		/* close */
16856 	nulldev,		/* strategy */
16857 	nulldev,		/* print */
16858 	nodev,			/* dump */
16859 	nodev,			/* read */
16860 	nodev,			/* write */
16861 	dtrace_ioctl,		/* ioctl */
16862 	nodev,			/* devmap */
16863 	nodev,			/* mmap */
16864 	nodev,			/* segmap */
16865 	nochpoll,		/* poll */
16866 	ddi_prop_op,		/* cb_prop_op */
16867 	0,			/* streamtab  */
16868 	D_NEW | D_MP		/* Driver compatibility flag */
16869 };
16870 
16871 static struct dev_ops dtrace_ops = {
16872 	DEVO_REV,		/* devo_rev */
16873 	0,			/* refcnt */
16874 	dtrace_info,		/* get_dev_info */
16875 	nulldev,		/* identify */
16876 	nulldev,		/* probe */
16877 	dtrace_attach,		/* attach */
16878 	dtrace_detach,		/* detach */
16879 	nodev,			/* reset */
16880 	&dtrace_cb_ops,		/* driver operations */
16881 	NULL,			/* bus operations */
16882 	nodev,			/* dev power */
16883 	ddi_quiesce_not_needed,		/* quiesce */
16884 };
16885 
16886 static struct modldrv modldrv = {
16887 	&mod_driverops,		/* module type (this is a pseudo driver) */
16888 	"Dynamic Tracing",	/* name of module */
16889 	&dtrace_ops,		/* driver ops */
16890 };
16891 
16892 static struct modlinkage modlinkage = {
16893 	MODREV_1,
16894 	(void *)&modldrv,
16895 	NULL
16896 };
16897 
16898 int
_init(void)16899 _init(void)
16900 {
16901 	return (mod_install(&modlinkage));
16902 }
16903 
16904 int
_info(struct modinfo * modinfop)16905 _info(struct modinfo *modinfop)
16906 {
16907 	return (mod_info(&modlinkage, modinfop));
16908 }
16909 
16910 int
_fini(void)16911 _fini(void)
16912 {
16913 	return (mod_remove(&modlinkage));
16914 }
16915