xref: /freebsd/sys/cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c (revision 53accc0452576206d0816f00c4448b10e59419b6)
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  * $FreeBSD$
22  */
23 
24 /*
25  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
26  * Use is subject to license terms.
27  */
28 
29 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 /*
32  * DTrace - Dynamic Tracing for Solaris
33  *
34  * This is the implementation of the Solaris Dynamic Tracing framework
35  * (DTrace).  The user-visible interface to DTrace is described at length in
36  * the "Solaris Dynamic Tracing Guide".  The interfaces between the libdtrace
37  * library, the in-kernel DTrace framework, and the DTrace providers are
38  * described in the block comments in the <sys/dtrace.h> header file.  The
39  * internal architecture of DTrace is described in the block comments in the
40  * <sys/dtrace_impl.h> header file.  The comments contained within the DTrace
41  * implementation very much assume mastery of all of these sources; if one has
42  * an unanswered question about the implementation, one should consult them
43  * first.
44  *
45  * The functions here are ordered roughly as follows:
46  *
47  *   - Probe context functions
48  *   - Probe hashing functions
49  *   - Non-probe context utility functions
50  *   - Matching functions
51  *   - Provider-to-Framework API functions
52  *   - Probe management functions
53  *   - DIF object functions
54  *   - Format functions
55  *   - Predicate functions
56  *   - ECB functions
57  *   - Buffer functions
58  *   - Enabling functions
59  *   - DOF functions
60  *   - Anonymous enabling functions
61  *   - Consumer state functions
62  *   - Helper functions
63  *   - Hook functions
64  *   - Driver cookbook functions
65  *
66  * Each group of functions begins with a block comment labelled the "DTrace
67  * [Group] Functions", allowing one to find each block by searching forward
68  * on capital-f functions.
69  */
70 #include <sys/errno.h>
71 #if !defined(sun)
72 #include <sys/time.h>
73 #endif
74 #include <sys/stat.h>
75 #include <sys/modctl.h>
76 #include <sys/conf.h>
77 #include <sys/systm.h>
78 #if defined(sun)
79 #include <sys/ddi.h>
80 #include <sys/sunddi.h>
81 #endif
82 #include <sys/cpuvar.h>
83 #include <sys/kmem.h>
84 #if defined(sun)
85 #include <sys/strsubr.h>
86 #endif
87 #include <sys/sysmacros.h>
88 #include <sys/dtrace_impl.h>
89 #include <sys/atomic.h>
90 #include <sys/cmn_err.h>
91 #if defined(sun)
92 #include <sys/mutex_impl.h>
93 #include <sys/rwlock_impl.h>
94 #endif
95 #include <sys/ctf_api.h>
96 #if defined(sun)
97 #include <sys/panic.h>
98 #include <sys/priv_impl.h>
99 #endif
100 #include <sys/policy.h>
101 #if defined(sun)
102 #include <sys/cred_impl.h>
103 #include <sys/procfs_isa.h>
104 #endif
105 #include <sys/taskq.h>
106 #if defined(sun)
107 #include <sys/mkdev.h>
108 #include <sys/kdi.h>
109 #endif
110 #include <sys/zone.h>
111 #include <sys/socket.h>
112 #include <netinet/in.h>
113 
114 /* FreeBSD includes: */
115 #if !defined(sun)
116 #include <sys/callout.h>
117 #include <sys/ctype.h>
118 #include <sys/limits.h>
119 #include <sys/kdb.h>
120 #include <sys/kernel.h>
121 #include <sys/malloc.h>
122 #include <sys/sysctl.h>
123 #include <sys/lock.h>
124 #include <sys/mutex.h>
125 #include <sys/rwlock.h>
126 #include <sys/sx.h>
127 #include <sys/dtrace_bsd.h>
128 #include <netinet/in.h>
129 #include "dtrace_cddl.h"
130 #include "dtrace_debug.c"
131 #endif
132 
133 /*
134  * DTrace Tunable Variables
135  *
136  * The following variables may be tuned by adding a line to /etc/system that
137  * includes both the name of the DTrace module ("dtrace") and the name of the
138  * variable.  For example:
139  *
140  *   set dtrace:dtrace_destructive_disallow = 1
141  *
142  * In general, the only variables that one should be tuning this way are those
143  * that affect system-wide DTrace behavior, and for which the default behavior
144  * is undesirable.  Most of these variables are tunable on a per-consumer
145  * basis using DTrace options, and need not be tuned on a system-wide basis.
146  * When tuning these variables, avoid pathological values; while some attempt
147  * is made to verify the integrity of these variables, they are not considered
148  * part of the supported interface to DTrace, and they are therefore not
149  * checked comprehensively.  Further, these variables should not be tuned
150  * dynamically via "mdb -kw" or other means; they should only be tuned via
151  * /etc/system.
152  */
153 int		dtrace_destructive_disallow = 0;
154 dtrace_optval_t	dtrace_nonroot_maxsize = (16 * 1024 * 1024);
155 size_t		dtrace_difo_maxsize = (256 * 1024);
156 dtrace_optval_t	dtrace_dof_maxsize = (256 * 1024);
157 size_t		dtrace_global_maxsize = (16 * 1024);
158 size_t		dtrace_actions_max = (16 * 1024);
159 size_t		dtrace_retain_max = 1024;
160 dtrace_optval_t	dtrace_helper_actions_max = 32;
161 dtrace_optval_t	dtrace_helper_providers_max = 32;
162 dtrace_optval_t	dtrace_dstate_defsize = (1 * 1024 * 1024);
163 size_t		dtrace_strsize_default = 256;
164 dtrace_optval_t	dtrace_cleanrate_default = 9900990;		/* 101 hz */
165 dtrace_optval_t	dtrace_cleanrate_min = 200000;			/* 5000 hz */
166 dtrace_optval_t	dtrace_cleanrate_max = (uint64_t)60 * NANOSEC;	/* 1/minute */
167 dtrace_optval_t	dtrace_aggrate_default = NANOSEC;		/* 1 hz */
168 dtrace_optval_t	dtrace_statusrate_default = NANOSEC;		/* 1 hz */
169 dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC;	 /* 6/minute */
170 dtrace_optval_t	dtrace_switchrate_default = NANOSEC;		/* 1 hz */
171 dtrace_optval_t	dtrace_nspec_default = 1;
172 dtrace_optval_t	dtrace_specsize_default = 32 * 1024;
173 dtrace_optval_t dtrace_stackframes_default = 20;
174 dtrace_optval_t dtrace_ustackframes_default = 20;
175 dtrace_optval_t dtrace_jstackframes_default = 50;
176 dtrace_optval_t dtrace_jstackstrsize_default = 512;
177 int		dtrace_msgdsize_max = 128;
178 hrtime_t	dtrace_chill_max = 500 * (NANOSEC / MILLISEC);	/* 500 ms */
179 hrtime_t	dtrace_chill_interval = NANOSEC;		/* 1000 ms */
180 int		dtrace_devdepth_max = 32;
181 int		dtrace_err_verbose;
182 hrtime_t	dtrace_deadman_interval = NANOSEC;
183 hrtime_t	dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC;
184 hrtime_t	dtrace_deadman_user = (hrtime_t)30 * NANOSEC;
185 
186 /*
187  * DTrace External Variables
188  *
189  * As dtrace(7D) is a kernel module, any DTrace variables are obviously
190  * available to DTrace consumers via the backtick (`) syntax.  One of these,
191  * dtrace_zero, is made deliberately so:  it is provided as a source of
192  * well-known, zero-filled memory.  While this variable is not documented,
193  * it is used by some translators as an implementation detail.
194  */
195 const char	dtrace_zero[256] = { 0 };	/* zero-filled memory */
196 
197 /*
198  * DTrace Internal Variables
199  */
200 #if defined(sun)
201 static dev_info_t	*dtrace_devi;		/* device info */
202 #endif
203 #if defined(sun)
204 static vmem_t		*dtrace_arena;		/* probe ID arena */
205 static vmem_t		*dtrace_minor;		/* minor number arena */
206 static taskq_t		*dtrace_taskq;		/* task queue */
207 #else
208 static struct unrhdr	*dtrace_arena;		/* Probe ID number.     */
209 #endif
210 static dtrace_probe_t	**dtrace_probes;	/* array of all probes */
211 static int		dtrace_nprobes;		/* number of probes */
212 static dtrace_provider_t *dtrace_provider;	/* provider list */
213 static dtrace_meta_t	*dtrace_meta_pid;	/* user-land meta provider */
214 static int		dtrace_opens;		/* number of opens */
215 static int		dtrace_helpers;		/* number of helpers */
216 #if defined(sun)
217 static void		*dtrace_softstate;	/* softstate pointer */
218 #endif
219 static dtrace_hash_t	*dtrace_bymod;		/* probes hashed by module */
220 static dtrace_hash_t	*dtrace_byfunc;		/* probes hashed by function */
221 static dtrace_hash_t	*dtrace_byname;		/* probes hashed by name */
222 static dtrace_toxrange_t *dtrace_toxrange;	/* toxic range array */
223 static int		dtrace_toxranges;	/* number of toxic ranges */
224 static int		dtrace_toxranges_max;	/* size of toxic range array */
225 static dtrace_anon_t	dtrace_anon;		/* anonymous enabling */
226 static kmem_cache_t	*dtrace_state_cache;	/* cache for dynamic state */
227 static uint64_t		dtrace_vtime_references; /* number of vtimestamp refs */
228 static kthread_t	*dtrace_panicked;	/* panicking thread */
229 static dtrace_ecb_t	*dtrace_ecb_create_cache; /* cached created ECB */
230 static dtrace_genid_t	dtrace_probegen;	/* current probe generation */
231 static dtrace_helpers_t *dtrace_deferred_pid;	/* deferred helper list */
232 static dtrace_enabling_t *dtrace_retained;	/* list of retained enablings */
233 static dtrace_dynvar_t	dtrace_dynhash_sink;	/* end of dynamic hash chains */
234 #if !defined(sun)
235 static struct mtx	dtrace_unr_mtx;
236 MTX_SYSINIT(dtrace_unr_mtx, &dtrace_unr_mtx, "Unique resource identifier", MTX_DEF);
237 int		dtrace_in_probe;	/* non-zero if executing a probe */
238 #if defined(__i386__) || defined(__amd64__)
239 uintptr_t	dtrace_in_probe_addr;	/* Address of invop when already in probe */
240 #endif
241 #endif
242 
243 /*
244  * DTrace Locking
245  * DTrace is protected by three (relatively coarse-grained) locks:
246  *
247  * (1) dtrace_lock is required to manipulate essentially any DTrace state,
248  *     including enabling state, probes, ECBs, consumer state, helper state,
249  *     etc.  Importantly, dtrace_lock is _not_ required when in probe context;
250  *     probe context is lock-free -- synchronization is handled via the
251  *     dtrace_sync() cross call mechanism.
252  *
253  * (2) dtrace_provider_lock is required when manipulating provider state, or
254  *     when provider state must be held constant.
255  *
256  * (3) dtrace_meta_lock is required when manipulating meta provider state, or
257  *     when meta provider state must be held constant.
258  *
259  * The lock ordering between these three locks is dtrace_meta_lock before
260  * dtrace_provider_lock before dtrace_lock.  (In particular, there are
261  * several places where dtrace_provider_lock is held by the framework as it
262  * calls into the providers -- which then call back into the framework,
263  * grabbing dtrace_lock.)
264  *
265  * There are two other locks in the mix:  mod_lock and cpu_lock.  With respect
266  * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical
267  * role as a coarse-grained lock; it is acquired before both of these locks.
268  * With respect to dtrace_meta_lock, its behavior is stranger:  cpu_lock must
269  * be acquired _between_ dtrace_meta_lock and any other DTrace locks.
270  * mod_lock is similar with respect to dtrace_provider_lock in that it must be
271  * acquired _between_ dtrace_provider_lock and dtrace_lock.
272  */
273 static kmutex_t		dtrace_lock;		/* probe state lock */
274 static kmutex_t		dtrace_provider_lock;	/* provider state lock */
275 static kmutex_t		dtrace_meta_lock;	/* meta-provider state lock */
276 
277 #if !defined(sun)
278 /* XXX FreeBSD hacks. */
279 static kmutex_t		mod_lock;
280 
281 #define cr_suid		cr_svuid
282 #define cr_sgid		cr_svgid
283 #define	ipaddr_t	in_addr_t
284 #define mod_modname	pathname
285 #define vuprintf	vprintf
286 #define ttoproc(_a)	((_a)->td_proc)
287 #define crgetzoneid(_a)	0
288 #define	NCPU		MAXCPU
289 #define SNOCD		0
290 #define CPU_ON_INTR(_a)	0
291 
292 #define PRIV_EFFECTIVE		(1 << 0)
293 #define PRIV_DTRACE_KERNEL	(1 << 1)
294 #define PRIV_DTRACE_PROC	(1 << 2)
295 #define PRIV_DTRACE_USER	(1 << 3)
296 #define PRIV_PROC_OWNER		(1 << 4)
297 #define PRIV_PROC_ZONE		(1 << 5)
298 #define PRIV_ALL		~0
299 
300 SYSCTL_NODE(_debug, OID_AUTO, dtrace, CTLFLAG_RD, 0, "DTrace Information");
301 #endif
302 
303 #if defined(sun)
304 #define curcpu	CPU->cpu_id
305 #endif
306 
307 
308 /*
309  * DTrace Provider Variables
310  *
311  * These are the variables relating to DTrace as a provider (that is, the
312  * provider of the BEGIN, END, and ERROR probes).
313  */
314 static dtrace_pattr_t	dtrace_provider_attr = {
315 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
316 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
317 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
318 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
319 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
320 };
321 
322 static void
323 dtrace_nullop(void)
324 {}
325 
326 static dtrace_pops_t	dtrace_provider_ops = {
327 	(void (*)(void *, dtrace_probedesc_t *))dtrace_nullop,
328 	(void (*)(void *, modctl_t *))dtrace_nullop,
329 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
330 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
331 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
332 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
333 	NULL,
334 	NULL,
335 	NULL,
336 	(void (*)(void *, dtrace_id_t, void *))dtrace_nullop
337 };
338 
339 static dtrace_id_t	dtrace_probeid_begin;	/* special BEGIN probe */
340 static dtrace_id_t	dtrace_probeid_end;	/* special END probe */
341 dtrace_id_t		dtrace_probeid_error;	/* special ERROR probe */
342 
343 /*
344  * DTrace Helper Tracing Variables
345  */
346 uint32_t dtrace_helptrace_next = 0;
347 uint32_t dtrace_helptrace_nlocals;
348 char	*dtrace_helptrace_buffer;
349 int	dtrace_helptrace_bufsize = 512 * 1024;
350 
351 #ifdef DEBUG
352 int	dtrace_helptrace_enabled = 1;
353 #else
354 int	dtrace_helptrace_enabled = 0;
355 #endif
356 
357 /*
358  * DTrace Error Hashing
359  *
360  * On DEBUG kernels, DTrace will track the errors that has seen in a hash
361  * table.  This is very useful for checking coverage of tests that are
362  * expected to induce DIF or DOF processing errors, and may be useful for
363  * debugging problems in the DIF code generator or in DOF generation .  The
364  * error hash may be examined with the ::dtrace_errhash MDB dcmd.
365  */
366 #ifdef DEBUG
367 static dtrace_errhash_t	dtrace_errhash[DTRACE_ERRHASHSZ];
368 static const char *dtrace_errlast;
369 static kthread_t *dtrace_errthread;
370 static kmutex_t dtrace_errlock;
371 #endif
372 
373 /*
374  * DTrace Macros and Constants
375  *
376  * These are various macros that are useful in various spots in the
377  * implementation, along with a few random constants that have no meaning
378  * outside of the implementation.  There is no real structure to this cpp
379  * mishmash -- but is there ever?
380  */
381 #define	DTRACE_HASHSTR(hash, probe)	\
382 	dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs)))
383 
384 #define	DTRACE_HASHNEXT(hash, probe)	\
385 	(dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs)
386 
387 #define	DTRACE_HASHPREV(hash, probe)	\
388 	(dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs)
389 
390 #define	DTRACE_HASHEQ(hash, lhs, rhs)	\
391 	(strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \
392 	    *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0)
393 
394 #define	DTRACE_AGGHASHSIZE_SLEW		17
395 
396 #define	DTRACE_V4MAPPED_OFFSET		(sizeof (uint32_t) * 3)
397 
398 /*
399  * The key for a thread-local variable consists of the lower 61 bits of the
400  * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL.
401  * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never
402  * equal to a variable identifier.  This is necessary (but not sufficient) to
403  * assure that global associative arrays never collide with thread-local
404  * variables.  To guarantee that they cannot collide, we must also define the
405  * order for keying dynamic variables.  That order is:
406  *
407  *   [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ]
408  *
409  * Because the variable-key and the tls-key are in orthogonal spaces, there is
410  * no way for a global variable key signature to match a thread-local key
411  * signature.
412  */
413 #if defined(sun)
414 #define	DTRACE_TLS_THRKEY(where) { \
415 	uint_t intr = 0; \
416 	uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \
417 	for (; actv; actv >>= 1) \
418 		intr++; \
419 	ASSERT(intr < (1 << 3)); \
420 	(where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \
421 	    (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
422 }
423 #else
424 #define	DTRACE_TLS_THRKEY(where) { \
425 	solaris_cpu_t *_c = &solaris_cpu[curcpu]; \
426 	uint_t intr = 0; \
427 	uint_t actv = _c->cpu_intr_actv; \
428 	for (; actv; actv >>= 1) \
429 		intr++; \
430 	ASSERT(intr < (1 << 3)); \
431 	(where) = ((curthread->td_tid + DIF_VARIABLE_MAX) & \
432 	    (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
433 }
434 #endif
435 
436 #define	DT_BSWAP_8(x)	((x) & 0xff)
437 #define	DT_BSWAP_16(x)	((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8))
438 #define	DT_BSWAP_32(x)	((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16))
439 #define	DT_BSWAP_64(x)	((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32))
440 
441 #define	DT_MASK_LO 0x00000000FFFFFFFFULL
442 
443 #define	DTRACE_STORE(type, tomax, offset, what) \
444 	*((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what);
445 
446 #ifndef __i386
447 #define	DTRACE_ALIGNCHECK(addr, size, flags)				\
448 	if (addr & (size - 1)) {					\
449 		*flags |= CPU_DTRACE_BADALIGN;				\
450 		cpu_core[curcpu].cpuc_dtrace_illval = addr;	\
451 		return (0);						\
452 	}
453 #else
454 #define	DTRACE_ALIGNCHECK(addr, size, flags)
455 #endif
456 
457 /*
458  * Test whether a range of memory starting at testaddr of size testsz falls
459  * within the range of memory described by addr, sz.  We take care to avoid
460  * problems with overflow and underflow of the unsigned quantities, and
461  * disallow all negative sizes.  Ranges of size 0 are allowed.
462  */
463 #define	DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \
464 	((testaddr) - (baseaddr) < (basesz) && \
465 	(testaddr) + (testsz) - (baseaddr) <= (basesz) && \
466 	(testaddr) + (testsz) >= (testaddr))
467 
468 /*
469  * Test whether alloc_sz bytes will fit in the scratch region.  We isolate
470  * alloc_sz on the righthand side of the comparison in order to avoid overflow
471  * or underflow in the comparison with it.  This is simpler than the INRANGE
472  * check above, because we know that the dtms_scratch_ptr is valid in the
473  * range.  Allocations of size zero are allowed.
474  */
475 #define	DTRACE_INSCRATCH(mstate, alloc_sz) \
476 	((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \
477 	(mstate)->dtms_scratch_ptr >= (alloc_sz))
478 
479 #define	DTRACE_LOADFUNC(bits)						\
480 /*CSTYLED*/								\
481 uint##bits##_t								\
482 dtrace_load##bits(uintptr_t addr)					\
483 {									\
484 	size_t size = bits / NBBY;					\
485 	/*CSTYLED*/							\
486 	uint##bits##_t rval;						\
487 	int i;								\
488 	volatile uint16_t *flags = (volatile uint16_t *)		\
489 	    &cpu_core[curcpu].cpuc_dtrace_flags;			\
490 									\
491 	DTRACE_ALIGNCHECK(addr, size, flags);				\
492 									\
493 	for (i = 0; i < dtrace_toxranges; i++) {			\
494 		if (addr >= dtrace_toxrange[i].dtt_limit)		\
495 			continue;					\
496 									\
497 		if (addr + size <= dtrace_toxrange[i].dtt_base)		\
498 			continue;					\
499 									\
500 		/*							\
501 		 * This address falls within a toxic region; return 0.	\
502 		 */							\
503 		*flags |= CPU_DTRACE_BADADDR;				\
504 		cpu_core[curcpu].cpuc_dtrace_illval = addr;		\
505 		return (0);						\
506 	}								\
507 									\
508 	*flags |= CPU_DTRACE_NOFAULT;					\
509 	/*CSTYLED*/							\
510 	rval = *((volatile uint##bits##_t *)addr);			\
511 	*flags &= ~CPU_DTRACE_NOFAULT;					\
512 									\
513 	return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0);		\
514 }
515 
516 #ifdef _LP64
517 #define	dtrace_loadptr	dtrace_load64
518 #else
519 #define	dtrace_loadptr	dtrace_load32
520 #endif
521 
522 #define	DTRACE_DYNHASH_FREE	0
523 #define	DTRACE_DYNHASH_SINK	1
524 #define	DTRACE_DYNHASH_VALID	2
525 
526 #define	DTRACE_MATCH_NEXT	0
527 #define	DTRACE_MATCH_DONE	1
528 #define	DTRACE_ANCHORED(probe)	((probe)->dtpr_func[0] != '\0')
529 #define	DTRACE_STATE_ALIGN	64
530 
531 #define	DTRACE_FLAGS2FLT(flags)						\
532 	(((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR :		\
533 	((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP :		\
534 	((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO :		\
535 	((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV :		\
536 	((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV :		\
537 	((flags) & CPU_DTRACE_TUPOFLOW) ?  DTRACEFLT_TUPOFLOW :		\
538 	((flags) & CPU_DTRACE_BADALIGN) ?  DTRACEFLT_BADALIGN :		\
539 	((flags) & CPU_DTRACE_NOSCRATCH) ?  DTRACEFLT_NOSCRATCH :	\
540 	((flags) & CPU_DTRACE_BADSTACK) ?  DTRACEFLT_BADSTACK :		\
541 	DTRACEFLT_UNKNOWN)
542 
543 #define	DTRACEACT_ISSTRING(act)						\
544 	((act)->dta_kind == DTRACEACT_DIFEXPR &&			\
545 	(act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING)
546 
547 /* Function prototype definitions: */
548 static size_t dtrace_strlen(const char *, size_t);
549 static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id);
550 static void dtrace_enabling_provide(dtrace_provider_t *);
551 static int dtrace_enabling_match(dtrace_enabling_t *, int *);
552 static void dtrace_enabling_matchall(void);
553 static dtrace_state_t *dtrace_anon_grab(void);
554 static uint64_t dtrace_helper(int, dtrace_mstate_t *,
555     dtrace_state_t *, uint64_t, uint64_t);
556 static dtrace_helpers_t *dtrace_helpers_create(proc_t *);
557 static void dtrace_buffer_drop(dtrace_buffer_t *);
558 static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t,
559     dtrace_state_t *, dtrace_mstate_t *);
560 static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t,
561     dtrace_optval_t);
562 static int dtrace_ecb_create_enable(dtrace_probe_t *, void *);
563 static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *);
564 uint16_t dtrace_load16(uintptr_t);
565 uint32_t dtrace_load32(uintptr_t);
566 uint64_t dtrace_load64(uintptr_t);
567 uint8_t dtrace_load8(uintptr_t);
568 void dtrace_dynvar_clean(dtrace_dstate_t *);
569 dtrace_dynvar_t *dtrace_dynvar(dtrace_dstate_t *, uint_t, dtrace_key_t *,
570     size_t, dtrace_dynvar_op_t, dtrace_mstate_t *, dtrace_vstate_t *);
571 uintptr_t dtrace_dif_varstr(uintptr_t, dtrace_state_t *, dtrace_mstate_t *);
572 
573 /*
574  * DTrace Probe Context Functions
575  *
576  * These functions are called from probe context.  Because probe context is
577  * any context in which C may be called, arbitrarily locks may be held,
578  * interrupts may be disabled, we may be in arbitrary dispatched state, etc.
579  * As a result, functions called from probe context may only call other DTrace
580  * support functions -- they may not interact at all with the system at large.
581  * (Note that the ASSERT macro is made probe-context safe by redefining it in
582  * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary
583  * loads are to be performed from probe context, they _must_ be in terms of
584  * the safe dtrace_load*() variants.
585  *
586  * Some functions in this block are not actually called from probe context;
587  * for these functions, there will be a comment above the function reading
588  * "Note:  not called from probe context."
589  */
590 void
591 dtrace_panic(const char *format, ...)
592 {
593 	va_list alist;
594 
595 	va_start(alist, format);
596 	dtrace_vpanic(format, alist);
597 	va_end(alist);
598 }
599 
600 int
601 dtrace_assfail(const char *a, const char *f, int l)
602 {
603 	dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l);
604 
605 	/*
606 	 * We just need something here that even the most clever compiler
607 	 * cannot optimize away.
608 	 */
609 	return (a[(uintptr_t)f]);
610 }
611 
612 /*
613  * Atomically increment a specified error counter from probe context.
614  */
615 static void
616 dtrace_error(uint32_t *counter)
617 {
618 	/*
619 	 * Most counters stored to in probe context are per-CPU counters.
620 	 * However, there are some error conditions that are sufficiently
621 	 * arcane that they don't merit per-CPU storage.  If these counters
622 	 * are incremented concurrently on different CPUs, scalability will be
623 	 * adversely affected -- but we don't expect them to be white-hot in a
624 	 * correctly constructed enabling...
625 	 */
626 	uint32_t oval, nval;
627 
628 	do {
629 		oval = *counter;
630 
631 		if ((nval = oval + 1) == 0) {
632 			/*
633 			 * If the counter would wrap, set it to 1 -- assuring
634 			 * that the counter is never zero when we have seen
635 			 * errors.  (The counter must be 32-bits because we
636 			 * aren't guaranteed a 64-bit compare&swap operation.)
637 			 * To save this code both the infamy of being fingered
638 			 * by a priggish news story and the indignity of being
639 			 * the target of a neo-puritan witch trial, we're
640 			 * carefully avoiding any colorful description of the
641 			 * likelihood of this condition -- but suffice it to
642 			 * say that it is only slightly more likely than the
643 			 * overflow of predicate cache IDs, as discussed in
644 			 * dtrace_predicate_create().
645 			 */
646 			nval = 1;
647 		}
648 	} while (dtrace_cas32(counter, oval, nval) != oval);
649 }
650 
651 /*
652  * Use the DTRACE_LOADFUNC macro to define functions for each of loading a
653  * uint8_t, a uint16_t, a uint32_t and a uint64_t.
654  */
655 DTRACE_LOADFUNC(8)
656 DTRACE_LOADFUNC(16)
657 DTRACE_LOADFUNC(32)
658 DTRACE_LOADFUNC(64)
659 
660 static int
661 dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate)
662 {
663 	if (dest < mstate->dtms_scratch_base)
664 		return (0);
665 
666 	if (dest + size < dest)
667 		return (0);
668 
669 	if (dest + size > mstate->dtms_scratch_ptr)
670 		return (0);
671 
672 	return (1);
673 }
674 
675 static int
676 dtrace_canstore_statvar(uint64_t addr, size_t sz,
677     dtrace_statvar_t **svars, int nsvars)
678 {
679 	int i;
680 
681 	for (i = 0; i < nsvars; i++) {
682 		dtrace_statvar_t *svar = svars[i];
683 
684 		if (svar == NULL || svar->dtsv_size == 0)
685 			continue;
686 
687 		if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size))
688 			return (1);
689 	}
690 
691 	return (0);
692 }
693 
694 /*
695  * Check to see if the address is within a memory region to which a store may
696  * be issued.  This includes the DTrace scratch areas, and any DTrace variable
697  * region.  The caller of dtrace_canstore() is responsible for performing any
698  * alignment checks that are needed before stores are actually executed.
699  */
700 static int
701 dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
702     dtrace_vstate_t *vstate)
703 {
704 	/*
705 	 * First, check to see if the address is in scratch space...
706 	 */
707 	if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base,
708 	    mstate->dtms_scratch_size))
709 		return (1);
710 
711 	/*
712 	 * Now check to see if it's a dynamic variable.  This check will pick
713 	 * up both thread-local variables and any global dynamically-allocated
714 	 * variables.
715 	 */
716 	if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base,
717 	    vstate->dtvs_dynvars.dtds_size)) {
718 		dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
719 		uintptr_t base = (uintptr_t)dstate->dtds_base +
720 		    (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t));
721 		uintptr_t chunkoffs;
722 
723 		/*
724 		 * Before we assume that we can store here, we need to make
725 		 * sure that it isn't in our metadata -- storing to our
726 		 * dynamic variable metadata would corrupt our state.  For
727 		 * the range to not include any dynamic variable metadata,
728 		 * it must:
729 		 *
730 		 *	(1) Start above the hash table that is at the base of
731 		 *	the dynamic variable space
732 		 *
733 		 *	(2) Have a starting chunk offset that is beyond the
734 		 *	dtrace_dynvar_t that is at the base of every chunk
735 		 *
736 		 *	(3) Not span a chunk boundary
737 		 *
738 		 */
739 		if (addr < base)
740 			return (0);
741 
742 		chunkoffs = (addr - base) % dstate->dtds_chunksize;
743 
744 		if (chunkoffs < sizeof (dtrace_dynvar_t))
745 			return (0);
746 
747 		if (chunkoffs + sz > dstate->dtds_chunksize)
748 			return (0);
749 
750 		return (1);
751 	}
752 
753 	/*
754 	 * Finally, check the static local and global variables.  These checks
755 	 * take the longest, so we perform them last.
756 	 */
757 	if (dtrace_canstore_statvar(addr, sz,
758 	    vstate->dtvs_locals, vstate->dtvs_nlocals))
759 		return (1);
760 
761 	if (dtrace_canstore_statvar(addr, sz,
762 	    vstate->dtvs_globals, vstate->dtvs_nglobals))
763 		return (1);
764 
765 	return (0);
766 }
767 
768 
769 /*
770  * Convenience routine to check to see if the address is within a memory
771  * region in which a load may be issued given the user's privilege level;
772  * if not, it sets the appropriate error flags and loads 'addr' into the
773  * illegal value slot.
774  *
775  * DTrace subroutines (DIF_SUBR_*) should use this helper to implement
776  * appropriate memory access protection.
777  */
778 static int
779 dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
780     dtrace_vstate_t *vstate)
781 {
782 	volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
783 
784 	/*
785 	 * If we hold the privilege to read from kernel memory, then
786 	 * everything is readable.
787 	 */
788 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
789 		return (1);
790 
791 	/*
792 	 * You can obviously read that which you can store.
793 	 */
794 	if (dtrace_canstore(addr, sz, mstate, vstate))
795 		return (1);
796 
797 	/*
798 	 * We're allowed to read from our own string table.
799 	 */
800 	if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab,
801 	    mstate->dtms_difo->dtdo_strlen))
802 		return (1);
803 
804 	DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV);
805 	*illval = addr;
806 	return (0);
807 }
808 
809 /*
810  * Convenience routine to check to see if a given string is within a memory
811  * region in which a load may be issued given the user's privilege level;
812  * this exists so that we don't need to issue unnecessary dtrace_strlen()
813  * calls in the event that the user has all privileges.
814  */
815 static int
816 dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
817     dtrace_vstate_t *vstate)
818 {
819 	size_t strsz;
820 
821 	/*
822 	 * If we hold the privilege to read from kernel memory, then
823 	 * everything is readable.
824 	 */
825 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
826 		return (1);
827 
828 	strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz);
829 	if (dtrace_canload(addr, strsz, mstate, vstate))
830 		return (1);
831 
832 	return (0);
833 }
834 
835 /*
836  * Convenience routine to check to see if a given variable is within a memory
837  * region in which a load may be issued given the user's privilege level.
838  */
839 static int
840 dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate,
841     dtrace_vstate_t *vstate)
842 {
843 	size_t sz;
844 	ASSERT(type->dtdt_flags & DIF_TF_BYREF);
845 
846 	/*
847 	 * If we hold the privilege to read from kernel memory, then
848 	 * everything is readable.
849 	 */
850 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
851 		return (1);
852 
853 	if (type->dtdt_kind == DIF_TYPE_STRING)
854 		sz = dtrace_strlen(src,
855 		    vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1;
856 	else
857 		sz = type->dtdt_size;
858 
859 	return (dtrace_canload((uintptr_t)src, sz, mstate, vstate));
860 }
861 
862 /*
863  * Compare two strings using safe loads.
864  */
865 static int
866 dtrace_strncmp(char *s1, char *s2, size_t limit)
867 {
868 	uint8_t c1, c2;
869 	volatile uint16_t *flags;
870 
871 	if (s1 == s2 || limit == 0)
872 		return (0);
873 
874 	flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
875 
876 	do {
877 		if (s1 == NULL) {
878 			c1 = '\0';
879 		} else {
880 			c1 = dtrace_load8((uintptr_t)s1++);
881 		}
882 
883 		if (s2 == NULL) {
884 			c2 = '\0';
885 		} else {
886 			c2 = dtrace_load8((uintptr_t)s2++);
887 		}
888 
889 		if (c1 != c2)
890 			return (c1 - c2);
891 	} while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT));
892 
893 	return (0);
894 }
895 
896 /*
897  * Compute strlen(s) for a string using safe memory accesses.  The additional
898  * len parameter is used to specify a maximum length to ensure completion.
899  */
900 static size_t
901 dtrace_strlen(const char *s, size_t lim)
902 {
903 	uint_t len;
904 
905 	for (len = 0; len != lim; len++) {
906 		if (dtrace_load8((uintptr_t)s++) == '\0')
907 			break;
908 	}
909 
910 	return (len);
911 }
912 
913 /*
914  * Check if an address falls within a toxic region.
915  */
916 static int
917 dtrace_istoxic(uintptr_t kaddr, size_t size)
918 {
919 	uintptr_t taddr, tsize;
920 	int i;
921 
922 	for (i = 0; i < dtrace_toxranges; i++) {
923 		taddr = dtrace_toxrange[i].dtt_base;
924 		tsize = dtrace_toxrange[i].dtt_limit - taddr;
925 
926 		if (kaddr - taddr < tsize) {
927 			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
928 			cpu_core[curcpu].cpuc_dtrace_illval = kaddr;
929 			return (1);
930 		}
931 
932 		if (taddr - kaddr < size) {
933 			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
934 			cpu_core[curcpu].cpuc_dtrace_illval = taddr;
935 			return (1);
936 		}
937 	}
938 
939 	return (0);
940 }
941 
942 /*
943  * Copy src to dst using safe memory accesses.  The src is assumed to be unsafe
944  * memory specified by the DIF program.  The dst is assumed to be safe memory
945  * that we can store to directly because it is managed by DTrace.  As with
946  * standard bcopy, overlapping copies are handled properly.
947  */
948 static void
949 dtrace_bcopy(const void *src, void *dst, size_t len)
950 {
951 	if (len != 0) {
952 		uint8_t *s1 = dst;
953 		const uint8_t *s2 = src;
954 
955 		if (s1 <= s2) {
956 			do {
957 				*s1++ = dtrace_load8((uintptr_t)s2++);
958 			} while (--len != 0);
959 		} else {
960 			s2 += len;
961 			s1 += len;
962 
963 			do {
964 				*--s1 = dtrace_load8((uintptr_t)--s2);
965 			} while (--len != 0);
966 		}
967 	}
968 }
969 
970 /*
971  * Copy src to dst using safe memory accesses, up to either the specified
972  * length, or the point that a nul byte is encountered.  The src is assumed to
973  * be unsafe memory specified by the DIF program.  The dst is assumed to be
974  * safe memory that we can store to directly because it is managed by DTrace.
975  * Unlike dtrace_bcopy(), overlapping regions are not handled.
976  */
977 static void
978 dtrace_strcpy(const void *src, void *dst, size_t len)
979 {
980 	if (len != 0) {
981 		uint8_t *s1 = dst, c;
982 		const uint8_t *s2 = src;
983 
984 		do {
985 			*s1++ = c = dtrace_load8((uintptr_t)s2++);
986 		} while (--len != 0 && c != '\0');
987 	}
988 }
989 
990 /*
991  * Copy src to dst, deriving the size and type from the specified (BYREF)
992  * variable type.  The src is assumed to be unsafe memory specified by the DIF
993  * program.  The dst is assumed to be DTrace variable memory that is of the
994  * specified type; we assume that we can store to directly.
995  */
996 static void
997 dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type)
998 {
999 	ASSERT(type->dtdt_flags & DIF_TF_BYREF);
1000 
1001 	if (type->dtdt_kind == DIF_TYPE_STRING) {
1002 		dtrace_strcpy(src, dst, type->dtdt_size);
1003 	} else {
1004 		dtrace_bcopy(src, dst, type->dtdt_size);
1005 	}
1006 }
1007 
1008 /*
1009  * Compare s1 to s2 using safe memory accesses.  The s1 data is assumed to be
1010  * unsafe memory specified by the DIF program.  The s2 data is assumed to be
1011  * safe memory that we can access directly because it is managed by DTrace.
1012  */
1013 static int
1014 dtrace_bcmp(const void *s1, const void *s2, size_t len)
1015 {
1016 	volatile uint16_t *flags;
1017 
1018 	flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
1019 
1020 	if (s1 == s2)
1021 		return (0);
1022 
1023 	if (s1 == NULL || s2 == NULL)
1024 		return (1);
1025 
1026 	if (s1 != s2 && len != 0) {
1027 		const uint8_t *ps1 = s1;
1028 		const uint8_t *ps2 = s2;
1029 
1030 		do {
1031 			if (dtrace_load8((uintptr_t)ps1++) != *ps2++)
1032 				return (1);
1033 		} while (--len != 0 && !(*flags & CPU_DTRACE_FAULT));
1034 	}
1035 	return (0);
1036 }
1037 
1038 /*
1039  * Zero the specified region using a simple byte-by-byte loop.  Note that this
1040  * is for safe DTrace-managed memory only.
1041  */
1042 static void
1043 dtrace_bzero(void *dst, size_t len)
1044 {
1045 	uchar_t *cp;
1046 
1047 	for (cp = dst; len != 0; len--)
1048 		*cp++ = 0;
1049 }
1050 
1051 static void
1052 dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum)
1053 {
1054 	uint64_t result[2];
1055 
1056 	result[0] = addend1[0] + addend2[0];
1057 	result[1] = addend1[1] + addend2[1] +
1058 	    (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0);
1059 
1060 	sum[0] = result[0];
1061 	sum[1] = result[1];
1062 }
1063 
1064 /*
1065  * Shift the 128-bit value in a by b. If b is positive, shift left.
1066  * If b is negative, shift right.
1067  */
1068 static void
1069 dtrace_shift_128(uint64_t *a, int b)
1070 {
1071 	uint64_t mask;
1072 
1073 	if (b == 0)
1074 		return;
1075 
1076 	if (b < 0) {
1077 		b = -b;
1078 		if (b >= 64) {
1079 			a[0] = a[1] >> (b - 64);
1080 			a[1] = 0;
1081 		} else {
1082 			a[0] >>= b;
1083 			mask = 1LL << (64 - b);
1084 			mask -= 1;
1085 			a[0] |= ((a[1] & mask) << (64 - b));
1086 			a[1] >>= b;
1087 		}
1088 	} else {
1089 		if (b >= 64) {
1090 			a[1] = a[0] << (b - 64);
1091 			a[0] = 0;
1092 		} else {
1093 			a[1] <<= b;
1094 			mask = a[0] >> (64 - b);
1095 			a[1] |= mask;
1096 			a[0] <<= b;
1097 		}
1098 	}
1099 }
1100 
1101 /*
1102  * The basic idea is to break the 2 64-bit values into 4 32-bit values,
1103  * use native multiplication on those, and then re-combine into the
1104  * resulting 128-bit value.
1105  *
1106  * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) =
1107  *     hi1 * hi2 << 64 +
1108  *     hi1 * lo2 << 32 +
1109  *     hi2 * lo1 << 32 +
1110  *     lo1 * lo2
1111  */
1112 static void
1113 dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product)
1114 {
1115 	uint64_t hi1, hi2, lo1, lo2;
1116 	uint64_t tmp[2];
1117 
1118 	hi1 = factor1 >> 32;
1119 	hi2 = factor2 >> 32;
1120 
1121 	lo1 = factor1 & DT_MASK_LO;
1122 	lo2 = factor2 & DT_MASK_LO;
1123 
1124 	product[0] = lo1 * lo2;
1125 	product[1] = hi1 * hi2;
1126 
1127 	tmp[0] = hi1 * lo2;
1128 	tmp[1] = 0;
1129 	dtrace_shift_128(tmp, 32);
1130 	dtrace_add_128(product, tmp, product);
1131 
1132 	tmp[0] = hi2 * lo1;
1133 	tmp[1] = 0;
1134 	dtrace_shift_128(tmp, 32);
1135 	dtrace_add_128(product, tmp, product);
1136 }
1137 
1138 /*
1139  * This privilege check should be used by actions and subroutines to
1140  * verify that the user credentials of the process that enabled the
1141  * invoking ECB match the target credentials
1142  */
1143 static int
1144 dtrace_priv_proc_common_user(dtrace_state_t *state)
1145 {
1146 	cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1147 
1148 	/*
1149 	 * We should always have a non-NULL state cred here, since if cred
1150 	 * is null (anonymous tracing), we fast-path bypass this routine.
1151 	 */
1152 	ASSERT(s_cr != NULL);
1153 
1154 	if ((cr = CRED()) != NULL &&
1155 	    s_cr->cr_uid == cr->cr_uid &&
1156 	    s_cr->cr_uid == cr->cr_ruid &&
1157 	    s_cr->cr_uid == cr->cr_suid &&
1158 	    s_cr->cr_gid == cr->cr_gid &&
1159 	    s_cr->cr_gid == cr->cr_rgid &&
1160 	    s_cr->cr_gid == cr->cr_sgid)
1161 		return (1);
1162 
1163 	return (0);
1164 }
1165 
1166 /*
1167  * This privilege check should be used by actions and subroutines to
1168  * verify that the zone of the process that enabled the invoking ECB
1169  * matches the target credentials
1170  */
1171 static int
1172 dtrace_priv_proc_common_zone(dtrace_state_t *state)
1173 {
1174 #if defined(sun)
1175 	cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1176 
1177 	/*
1178 	 * We should always have a non-NULL state cred here, since if cred
1179 	 * is null (anonymous tracing), we fast-path bypass this routine.
1180 	 */
1181 	ASSERT(s_cr != NULL);
1182 
1183 	if ((cr = CRED()) != NULL &&
1184 	    s_cr->cr_zone == cr->cr_zone)
1185 		return (1);
1186 
1187 	return (0);
1188 #else
1189 	return (1);
1190 #endif
1191 }
1192 
1193 /*
1194  * This privilege check should be used by actions and subroutines to
1195  * verify that the process has not setuid or changed credentials.
1196  */
1197 static int
1198 dtrace_priv_proc_common_nocd(void)
1199 {
1200 	proc_t *proc;
1201 
1202 	if ((proc = ttoproc(curthread)) != NULL &&
1203 	    !(proc->p_flag & SNOCD))
1204 		return (1);
1205 
1206 	return (0);
1207 }
1208 
1209 static int
1210 dtrace_priv_proc_destructive(dtrace_state_t *state)
1211 {
1212 	int action = state->dts_cred.dcr_action;
1213 
1214 	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) &&
1215 	    dtrace_priv_proc_common_zone(state) == 0)
1216 		goto bad;
1217 
1218 	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) &&
1219 	    dtrace_priv_proc_common_user(state) == 0)
1220 		goto bad;
1221 
1222 	if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) &&
1223 	    dtrace_priv_proc_common_nocd() == 0)
1224 		goto bad;
1225 
1226 	return (1);
1227 
1228 bad:
1229 	cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1230 
1231 	return (0);
1232 }
1233 
1234 static int
1235 dtrace_priv_proc_control(dtrace_state_t *state)
1236 {
1237 	if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL)
1238 		return (1);
1239 
1240 	if (dtrace_priv_proc_common_zone(state) &&
1241 	    dtrace_priv_proc_common_user(state) &&
1242 	    dtrace_priv_proc_common_nocd())
1243 		return (1);
1244 
1245 	cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1246 
1247 	return (0);
1248 }
1249 
1250 static int
1251 dtrace_priv_proc(dtrace_state_t *state)
1252 {
1253 	if (state->dts_cred.dcr_action & DTRACE_CRA_PROC)
1254 		return (1);
1255 
1256 	cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1257 
1258 	return (0);
1259 }
1260 
1261 static int
1262 dtrace_priv_kernel(dtrace_state_t *state)
1263 {
1264 	if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL)
1265 		return (1);
1266 
1267 	cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1268 
1269 	return (0);
1270 }
1271 
1272 static int
1273 dtrace_priv_kernel_destructive(dtrace_state_t *state)
1274 {
1275 	if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE)
1276 		return (1);
1277 
1278 	cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1279 
1280 	return (0);
1281 }
1282 
1283 /*
1284  * Note:  not called from probe context.  This function is called
1285  * asynchronously (and at a regular interval) from outside of probe context to
1286  * clean the dirty dynamic variable lists on all CPUs.  Dynamic variable
1287  * cleaning is explained in detail in <sys/dtrace_impl.h>.
1288  */
1289 void
1290 dtrace_dynvar_clean(dtrace_dstate_t *dstate)
1291 {
1292 	dtrace_dynvar_t *dirty;
1293 	dtrace_dstate_percpu_t *dcpu;
1294 	int i, work = 0;
1295 
1296 	for (i = 0; i < NCPU; i++) {
1297 		dcpu = &dstate->dtds_percpu[i];
1298 
1299 		ASSERT(dcpu->dtdsc_rinsing == NULL);
1300 
1301 		/*
1302 		 * If the dirty list is NULL, there is no dirty work to do.
1303 		 */
1304 		if (dcpu->dtdsc_dirty == NULL)
1305 			continue;
1306 
1307 		/*
1308 		 * If the clean list is non-NULL, then we're not going to do
1309 		 * any work for this CPU -- it means that there has not been
1310 		 * a dtrace_dynvar() allocation on this CPU (or from this CPU)
1311 		 * since the last time we cleaned house.
1312 		 */
1313 		if (dcpu->dtdsc_clean != NULL)
1314 			continue;
1315 
1316 		work = 1;
1317 
1318 		/*
1319 		 * Atomically move the dirty list aside.
1320 		 */
1321 		do {
1322 			dirty = dcpu->dtdsc_dirty;
1323 
1324 			/*
1325 			 * Before we zap the dirty list, set the rinsing list.
1326 			 * (This allows for a potential assertion in
1327 			 * dtrace_dynvar():  if a free dynamic variable appears
1328 			 * on a hash chain, either the dirty list or the
1329 			 * rinsing list for some CPU must be non-NULL.)
1330 			 */
1331 			dcpu->dtdsc_rinsing = dirty;
1332 			dtrace_membar_producer();
1333 		} while (dtrace_casptr(&dcpu->dtdsc_dirty,
1334 		    dirty, NULL) != dirty);
1335 	}
1336 
1337 	if (!work) {
1338 		/*
1339 		 * We have no work to do; we can simply return.
1340 		 */
1341 		return;
1342 	}
1343 
1344 	dtrace_sync();
1345 
1346 	for (i = 0; i < NCPU; i++) {
1347 		dcpu = &dstate->dtds_percpu[i];
1348 
1349 		if (dcpu->dtdsc_rinsing == NULL)
1350 			continue;
1351 
1352 		/*
1353 		 * We are now guaranteed that no hash chain contains a pointer
1354 		 * into this dirty list; we can make it clean.
1355 		 */
1356 		ASSERT(dcpu->dtdsc_clean == NULL);
1357 		dcpu->dtdsc_clean = dcpu->dtdsc_rinsing;
1358 		dcpu->dtdsc_rinsing = NULL;
1359 	}
1360 
1361 	/*
1362 	 * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make
1363 	 * sure that all CPUs have seen all of the dtdsc_clean pointers.
1364 	 * This prevents a race whereby a CPU incorrectly decides that
1365 	 * the state should be something other than DTRACE_DSTATE_CLEAN
1366 	 * after dtrace_dynvar_clean() has completed.
1367 	 */
1368 	dtrace_sync();
1369 
1370 	dstate->dtds_state = DTRACE_DSTATE_CLEAN;
1371 }
1372 
1373 /*
1374  * Depending on the value of the op parameter, this function looks-up,
1375  * allocates or deallocates an arbitrarily-keyed dynamic variable.  If an
1376  * allocation is requested, this function will return a pointer to a
1377  * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no
1378  * variable can be allocated.  If NULL is returned, the appropriate counter
1379  * will be incremented.
1380  */
1381 dtrace_dynvar_t *
1382 dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys,
1383     dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op,
1384     dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
1385 {
1386 	uint64_t hashval = DTRACE_DYNHASH_VALID;
1387 	dtrace_dynhash_t *hash = dstate->dtds_hash;
1388 	dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL;
1389 	processorid_t me = curcpu, cpu = me;
1390 	dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me];
1391 	size_t bucket, ksize;
1392 	size_t chunksize = dstate->dtds_chunksize;
1393 	uintptr_t kdata, lock, nstate;
1394 	uint_t i;
1395 
1396 	ASSERT(nkeys != 0);
1397 
1398 	/*
1399 	 * Hash the key.  As with aggregations, we use Jenkins' "One-at-a-time"
1400 	 * algorithm.  For the by-value portions, we perform the algorithm in
1401 	 * 16-bit chunks (as opposed to 8-bit chunks).  This speeds things up a
1402 	 * bit, and seems to have only a minute effect on distribution.  For
1403 	 * the by-reference data, we perform "One-at-a-time" iterating (safely)
1404 	 * over each referenced byte.  It's painful to do this, but it's much
1405 	 * better than pathological hash distribution.  The efficacy of the
1406 	 * hashing algorithm (and a comparison with other algorithms) may be
1407 	 * found by running the ::dtrace_dynstat MDB dcmd.
1408 	 */
1409 	for (i = 0; i < nkeys; i++) {
1410 		if (key[i].dttk_size == 0) {
1411 			uint64_t val = key[i].dttk_value;
1412 
1413 			hashval += (val >> 48) & 0xffff;
1414 			hashval += (hashval << 10);
1415 			hashval ^= (hashval >> 6);
1416 
1417 			hashval += (val >> 32) & 0xffff;
1418 			hashval += (hashval << 10);
1419 			hashval ^= (hashval >> 6);
1420 
1421 			hashval += (val >> 16) & 0xffff;
1422 			hashval += (hashval << 10);
1423 			hashval ^= (hashval >> 6);
1424 
1425 			hashval += val & 0xffff;
1426 			hashval += (hashval << 10);
1427 			hashval ^= (hashval >> 6);
1428 		} else {
1429 			/*
1430 			 * This is incredibly painful, but it beats the hell
1431 			 * out of the alternative.
1432 			 */
1433 			uint64_t j, size = key[i].dttk_size;
1434 			uintptr_t base = (uintptr_t)key[i].dttk_value;
1435 
1436 			if (!dtrace_canload(base, size, mstate, vstate))
1437 				break;
1438 
1439 			for (j = 0; j < size; j++) {
1440 				hashval += dtrace_load8(base + j);
1441 				hashval += (hashval << 10);
1442 				hashval ^= (hashval >> 6);
1443 			}
1444 		}
1445 	}
1446 
1447 	if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT))
1448 		return (NULL);
1449 
1450 	hashval += (hashval << 3);
1451 	hashval ^= (hashval >> 11);
1452 	hashval += (hashval << 15);
1453 
1454 	/*
1455 	 * There is a remote chance (ideally, 1 in 2^31) that our hashval
1456 	 * comes out to be one of our two sentinel hash values.  If this
1457 	 * actually happens, we set the hashval to be a value known to be a
1458 	 * non-sentinel value.
1459 	 */
1460 	if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK)
1461 		hashval = DTRACE_DYNHASH_VALID;
1462 
1463 	/*
1464 	 * Yes, it's painful to do a divide here.  If the cycle count becomes
1465 	 * important here, tricks can be pulled to reduce it.  (However, it's
1466 	 * critical that hash collisions be kept to an absolute minimum;
1467 	 * they're much more painful than a divide.)  It's better to have a
1468 	 * solution that generates few collisions and still keeps things
1469 	 * relatively simple.
1470 	 */
1471 	bucket = hashval % dstate->dtds_hashsize;
1472 
1473 	if (op == DTRACE_DYNVAR_DEALLOC) {
1474 		volatile uintptr_t *lockp = &hash[bucket].dtdh_lock;
1475 
1476 		for (;;) {
1477 			while ((lock = *lockp) & 1)
1478 				continue;
1479 
1480 			if (dtrace_casptr((volatile void *)lockp,
1481 			    (volatile void *)lock, (volatile void *)(lock + 1)) == (void *)lock)
1482 				break;
1483 		}
1484 
1485 		dtrace_membar_producer();
1486 	}
1487 
1488 top:
1489 	prev = NULL;
1490 	lock = hash[bucket].dtdh_lock;
1491 
1492 	dtrace_membar_consumer();
1493 
1494 	start = hash[bucket].dtdh_chain;
1495 	ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK ||
1496 	    start->dtdv_hashval != DTRACE_DYNHASH_FREE ||
1497 	    op != DTRACE_DYNVAR_DEALLOC));
1498 
1499 	for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) {
1500 		dtrace_tuple_t *dtuple = &dvar->dtdv_tuple;
1501 		dtrace_key_t *dkey = &dtuple->dtt_key[0];
1502 
1503 		if (dvar->dtdv_hashval != hashval) {
1504 			if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) {
1505 				/*
1506 				 * We've reached the sink, and therefore the
1507 				 * end of the hash chain; we can kick out of
1508 				 * the loop knowing that we have seen a valid
1509 				 * snapshot of state.
1510 				 */
1511 				ASSERT(dvar->dtdv_next == NULL);
1512 				ASSERT(dvar == &dtrace_dynhash_sink);
1513 				break;
1514 			}
1515 
1516 			if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) {
1517 				/*
1518 				 * We've gone off the rails:  somewhere along
1519 				 * the line, one of the members of this hash
1520 				 * chain was deleted.  Note that we could also
1521 				 * detect this by simply letting this loop run
1522 				 * to completion, as we would eventually hit
1523 				 * the end of the dirty list.  However, we
1524 				 * want to avoid running the length of the
1525 				 * dirty list unnecessarily (it might be quite
1526 				 * long), so we catch this as early as
1527 				 * possible by detecting the hash marker.  In
1528 				 * this case, we simply set dvar to NULL and
1529 				 * break; the conditional after the loop will
1530 				 * send us back to top.
1531 				 */
1532 				dvar = NULL;
1533 				break;
1534 			}
1535 
1536 			goto next;
1537 		}
1538 
1539 		if (dtuple->dtt_nkeys != nkeys)
1540 			goto next;
1541 
1542 		for (i = 0; i < nkeys; i++, dkey++) {
1543 			if (dkey->dttk_size != key[i].dttk_size)
1544 				goto next; /* size or type mismatch */
1545 
1546 			if (dkey->dttk_size != 0) {
1547 				if (dtrace_bcmp(
1548 				    (void *)(uintptr_t)key[i].dttk_value,
1549 				    (void *)(uintptr_t)dkey->dttk_value,
1550 				    dkey->dttk_size))
1551 					goto next;
1552 			} else {
1553 				if (dkey->dttk_value != key[i].dttk_value)
1554 					goto next;
1555 			}
1556 		}
1557 
1558 		if (op != DTRACE_DYNVAR_DEALLOC)
1559 			return (dvar);
1560 
1561 		ASSERT(dvar->dtdv_next == NULL ||
1562 		    dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE);
1563 
1564 		if (prev != NULL) {
1565 			ASSERT(hash[bucket].dtdh_chain != dvar);
1566 			ASSERT(start != dvar);
1567 			ASSERT(prev->dtdv_next == dvar);
1568 			prev->dtdv_next = dvar->dtdv_next;
1569 		} else {
1570 			if (dtrace_casptr(&hash[bucket].dtdh_chain,
1571 			    start, dvar->dtdv_next) != start) {
1572 				/*
1573 				 * We have failed to atomically swing the
1574 				 * hash table head pointer, presumably because
1575 				 * of a conflicting allocation on another CPU.
1576 				 * We need to reread the hash chain and try
1577 				 * again.
1578 				 */
1579 				goto top;
1580 			}
1581 		}
1582 
1583 		dtrace_membar_producer();
1584 
1585 		/*
1586 		 * Now set the hash value to indicate that it's free.
1587 		 */
1588 		ASSERT(hash[bucket].dtdh_chain != dvar);
1589 		dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
1590 
1591 		dtrace_membar_producer();
1592 
1593 		/*
1594 		 * Set the next pointer to point at the dirty list, and
1595 		 * atomically swing the dirty pointer to the newly freed dvar.
1596 		 */
1597 		do {
1598 			next = dcpu->dtdsc_dirty;
1599 			dvar->dtdv_next = next;
1600 		} while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next);
1601 
1602 		/*
1603 		 * Finally, unlock this hash bucket.
1604 		 */
1605 		ASSERT(hash[bucket].dtdh_lock == lock);
1606 		ASSERT(lock & 1);
1607 		hash[bucket].dtdh_lock++;
1608 
1609 		return (NULL);
1610 next:
1611 		prev = dvar;
1612 		continue;
1613 	}
1614 
1615 	if (dvar == NULL) {
1616 		/*
1617 		 * If dvar is NULL, it is because we went off the rails:
1618 		 * one of the elements that we traversed in the hash chain
1619 		 * was deleted while we were traversing it.  In this case,
1620 		 * we assert that we aren't doing a dealloc (deallocs lock
1621 		 * the hash bucket to prevent themselves from racing with
1622 		 * one another), and retry the hash chain traversal.
1623 		 */
1624 		ASSERT(op != DTRACE_DYNVAR_DEALLOC);
1625 		goto top;
1626 	}
1627 
1628 	if (op != DTRACE_DYNVAR_ALLOC) {
1629 		/*
1630 		 * If we are not to allocate a new variable, we want to
1631 		 * return NULL now.  Before we return, check that the value
1632 		 * of the lock word hasn't changed.  If it has, we may have
1633 		 * seen an inconsistent snapshot.
1634 		 */
1635 		if (op == DTRACE_DYNVAR_NOALLOC) {
1636 			if (hash[bucket].dtdh_lock != lock)
1637 				goto top;
1638 		} else {
1639 			ASSERT(op == DTRACE_DYNVAR_DEALLOC);
1640 			ASSERT(hash[bucket].dtdh_lock == lock);
1641 			ASSERT(lock & 1);
1642 			hash[bucket].dtdh_lock++;
1643 		}
1644 
1645 		return (NULL);
1646 	}
1647 
1648 	/*
1649 	 * We need to allocate a new dynamic variable.  The size we need is the
1650 	 * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the
1651 	 * size of any auxiliary key data (rounded up to 8-byte alignment) plus
1652 	 * the size of any referred-to data (dsize).  We then round the final
1653 	 * size up to the chunksize for allocation.
1654 	 */
1655 	for (ksize = 0, i = 0; i < nkeys; i++)
1656 		ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
1657 
1658 	/*
1659 	 * This should be pretty much impossible, but could happen if, say,
1660 	 * strange DIF specified the tuple.  Ideally, this should be an
1661 	 * assertion and not an error condition -- but that requires that the
1662 	 * chunksize calculation in dtrace_difo_chunksize() be absolutely
1663 	 * bullet-proof.  (That is, it must not be able to be fooled by
1664 	 * malicious DIF.)  Given the lack of backwards branches in DIF,
1665 	 * solving this would presumably not amount to solving the Halting
1666 	 * Problem -- but it still seems awfully hard.
1667 	 */
1668 	if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) +
1669 	    ksize + dsize > chunksize) {
1670 		dcpu->dtdsc_drops++;
1671 		return (NULL);
1672 	}
1673 
1674 	nstate = DTRACE_DSTATE_EMPTY;
1675 
1676 	do {
1677 retry:
1678 		free = dcpu->dtdsc_free;
1679 
1680 		if (free == NULL) {
1681 			dtrace_dynvar_t *clean = dcpu->dtdsc_clean;
1682 			void *rval;
1683 
1684 			if (clean == NULL) {
1685 				/*
1686 				 * We're out of dynamic variable space on
1687 				 * this CPU.  Unless we have tried all CPUs,
1688 				 * we'll try to allocate from a different
1689 				 * CPU.
1690 				 */
1691 				switch (dstate->dtds_state) {
1692 				case DTRACE_DSTATE_CLEAN: {
1693 					void *sp = &dstate->dtds_state;
1694 
1695 					if (++cpu >= NCPU)
1696 						cpu = 0;
1697 
1698 					if (dcpu->dtdsc_dirty != NULL &&
1699 					    nstate == DTRACE_DSTATE_EMPTY)
1700 						nstate = DTRACE_DSTATE_DIRTY;
1701 
1702 					if (dcpu->dtdsc_rinsing != NULL)
1703 						nstate = DTRACE_DSTATE_RINSING;
1704 
1705 					dcpu = &dstate->dtds_percpu[cpu];
1706 
1707 					if (cpu != me)
1708 						goto retry;
1709 
1710 					(void) dtrace_cas32(sp,
1711 					    DTRACE_DSTATE_CLEAN, nstate);
1712 
1713 					/*
1714 					 * To increment the correct bean
1715 					 * counter, take another lap.
1716 					 */
1717 					goto retry;
1718 				}
1719 
1720 				case DTRACE_DSTATE_DIRTY:
1721 					dcpu->dtdsc_dirty_drops++;
1722 					break;
1723 
1724 				case DTRACE_DSTATE_RINSING:
1725 					dcpu->dtdsc_rinsing_drops++;
1726 					break;
1727 
1728 				case DTRACE_DSTATE_EMPTY:
1729 					dcpu->dtdsc_drops++;
1730 					break;
1731 				}
1732 
1733 				DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP);
1734 				return (NULL);
1735 			}
1736 
1737 			/*
1738 			 * The clean list appears to be non-empty.  We want to
1739 			 * move the clean list to the free list; we start by
1740 			 * moving the clean pointer aside.
1741 			 */
1742 			if (dtrace_casptr(&dcpu->dtdsc_clean,
1743 			    clean, NULL) != clean) {
1744 				/*
1745 				 * We are in one of two situations:
1746 				 *
1747 				 *  (a)	The clean list was switched to the
1748 				 *	free list by another CPU.
1749 				 *
1750 				 *  (b)	The clean list was added to by the
1751 				 *	cleansing cyclic.
1752 				 *
1753 				 * In either of these situations, we can
1754 				 * just reattempt the free list allocation.
1755 				 */
1756 				goto retry;
1757 			}
1758 
1759 			ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE);
1760 
1761 			/*
1762 			 * Now we'll move the clean list to the free list.
1763 			 * It's impossible for this to fail:  the only way
1764 			 * the free list can be updated is through this
1765 			 * code path, and only one CPU can own the clean list.
1766 			 * Thus, it would only be possible for this to fail if
1767 			 * this code were racing with dtrace_dynvar_clean().
1768 			 * (That is, if dtrace_dynvar_clean() updated the clean
1769 			 * list, and we ended up racing to update the free
1770 			 * list.)  This race is prevented by the dtrace_sync()
1771 			 * in dtrace_dynvar_clean() -- which flushes the
1772 			 * owners of the clean lists out before resetting
1773 			 * the clean lists.
1774 			 */
1775 			rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean);
1776 			ASSERT(rval == NULL);
1777 			goto retry;
1778 		}
1779 
1780 		dvar = free;
1781 		new_free = dvar->dtdv_next;
1782 	} while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free);
1783 
1784 	/*
1785 	 * We have now allocated a new chunk.  We copy the tuple keys into the
1786 	 * tuple array and copy any referenced key data into the data space
1787 	 * following the tuple array.  As we do this, we relocate dttk_value
1788 	 * in the final tuple to point to the key data address in the chunk.
1789 	 */
1790 	kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys];
1791 	dvar->dtdv_data = (void *)(kdata + ksize);
1792 	dvar->dtdv_tuple.dtt_nkeys = nkeys;
1793 
1794 	for (i = 0; i < nkeys; i++) {
1795 		dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i];
1796 		size_t kesize = key[i].dttk_size;
1797 
1798 		if (kesize != 0) {
1799 			dtrace_bcopy(
1800 			    (const void *)(uintptr_t)key[i].dttk_value,
1801 			    (void *)kdata, kesize);
1802 			dkey->dttk_value = kdata;
1803 			kdata += P2ROUNDUP(kesize, sizeof (uint64_t));
1804 		} else {
1805 			dkey->dttk_value = key[i].dttk_value;
1806 		}
1807 
1808 		dkey->dttk_size = kesize;
1809 	}
1810 
1811 	ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE);
1812 	dvar->dtdv_hashval = hashval;
1813 	dvar->dtdv_next = start;
1814 
1815 	if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start)
1816 		return (dvar);
1817 
1818 	/*
1819 	 * The cas has failed.  Either another CPU is adding an element to
1820 	 * this hash chain, or another CPU is deleting an element from this
1821 	 * hash chain.  The simplest way to deal with both of these cases
1822 	 * (though not necessarily the most efficient) is to free our
1823 	 * allocated block and tail-call ourselves.  Note that the free is
1824 	 * to the dirty list and _not_ to the free list.  This is to prevent
1825 	 * races with allocators, above.
1826 	 */
1827 	dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
1828 
1829 	dtrace_membar_producer();
1830 
1831 	do {
1832 		free = dcpu->dtdsc_dirty;
1833 		dvar->dtdv_next = free;
1834 	} while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free);
1835 
1836 	return (dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate));
1837 }
1838 
1839 /*ARGSUSED*/
1840 static void
1841 dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg)
1842 {
1843 	if ((int64_t)nval < (int64_t)*oval)
1844 		*oval = nval;
1845 }
1846 
1847 /*ARGSUSED*/
1848 static void
1849 dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg)
1850 {
1851 	if ((int64_t)nval > (int64_t)*oval)
1852 		*oval = nval;
1853 }
1854 
1855 static void
1856 dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr)
1857 {
1858 	int i, zero = DTRACE_QUANTIZE_ZEROBUCKET;
1859 	int64_t val = (int64_t)nval;
1860 
1861 	if (val < 0) {
1862 		for (i = 0; i < zero; i++) {
1863 			if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) {
1864 				quanta[i] += incr;
1865 				return;
1866 			}
1867 		}
1868 	} else {
1869 		for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) {
1870 			if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) {
1871 				quanta[i - 1] += incr;
1872 				return;
1873 			}
1874 		}
1875 
1876 		quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr;
1877 		return;
1878 	}
1879 
1880 	ASSERT(0);
1881 }
1882 
1883 static void
1884 dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr)
1885 {
1886 	uint64_t arg = *lquanta++;
1887 	int32_t base = DTRACE_LQUANTIZE_BASE(arg);
1888 	uint16_t step = DTRACE_LQUANTIZE_STEP(arg);
1889 	uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg);
1890 	int32_t val = (int32_t)nval, level;
1891 
1892 	ASSERT(step != 0);
1893 	ASSERT(levels != 0);
1894 
1895 	if (val < base) {
1896 		/*
1897 		 * This is an underflow.
1898 		 */
1899 		lquanta[0] += incr;
1900 		return;
1901 	}
1902 
1903 	level = (val - base) / step;
1904 
1905 	if (level < levels) {
1906 		lquanta[level + 1] += incr;
1907 		return;
1908 	}
1909 
1910 	/*
1911 	 * This is an overflow.
1912 	 */
1913 	lquanta[levels + 1] += incr;
1914 }
1915 
1916 /*ARGSUSED*/
1917 static void
1918 dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg)
1919 {
1920 	data[0]++;
1921 	data[1] += nval;
1922 }
1923 
1924 /*ARGSUSED*/
1925 static void
1926 dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg)
1927 {
1928 	int64_t snval = (int64_t)nval;
1929 	uint64_t tmp[2];
1930 
1931 	data[0]++;
1932 	data[1] += nval;
1933 
1934 	/*
1935 	 * What we want to say here is:
1936 	 *
1937 	 * data[2] += nval * nval;
1938 	 *
1939 	 * But given that nval is 64-bit, we could easily overflow, so
1940 	 * we do this as 128-bit arithmetic.
1941 	 */
1942 	if (snval < 0)
1943 		snval = -snval;
1944 
1945 	dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp);
1946 	dtrace_add_128(data + 2, tmp, data + 2);
1947 }
1948 
1949 /*ARGSUSED*/
1950 static void
1951 dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg)
1952 {
1953 	*oval = *oval + 1;
1954 }
1955 
1956 /*ARGSUSED*/
1957 static void
1958 dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg)
1959 {
1960 	*oval += nval;
1961 }
1962 
1963 /*
1964  * Aggregate given the tuple in the principal data buffer, and the aggregating
1965  * action denoted by the specified dtrace_aggregation_t.  The aggregation
1966  * buffer is specified as the buf parameter.  This routine does not return
1967  * failure; if there is no space in the aggregation buffer, the data will be
1968  * dropped, and a corresponding counter incremented.
1969  */
1970 static void
1971 dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf,
1972     intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg)
1973 {
1974 	dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec;
1975 	uint32_t i, ndx, size, fsize;
1976 	uint32_t align = sizeof (uint64_t) - 1;
1977 	dtrace_aggbuffer_t *agb;
1978 	dtrace_aggkey_t *key;
1979 	uint32_t hashval = 0, limit, isstr;
1980 	caddr_t tomax, data, kdata;
1981 	dtrace_actkind_t action;
1982 	dtrace_action_t *act;
1983 	uintptr_t offs;
1984 
1985 	if (buf == NULL)
1986 		return;
1987 
1988 	if (!agg->dtag_hasarg) {
1989 		/*
1990 		 * Currently, only quantize() and lquantize() take additional
1991 		 * arguments, and they have the same semantics:  an increment
1992 		 * value that defaults to 1 when not present.  If additional
1993 		 * aggregating actions take arguments, the setting of the
1994 		 * default argument value will presumably have to become more
1995 		 * sophisticated...
1996 		 */
1997 		arg = 1;
1998 	}
1999 
2000 	action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION;
2001 	size = rec->dtrd_offset - agg->dtag_base;
2002 	fsize = size + rec->dtrd_size;
2003 
2004 	ASSERT(dbuf->dtb_tomax != NULL);
2005 	data = dbuf->dtb_tomax + offset + agg->dtag_base;
2006 
2007 	if ((tomax = buf->dtb_tomax) == NULL) {
2008 		dtrace_buffer_drop(buf);
2009 		return;
2010 	}
2011 
2012 	/*
2013 	 * The metastructure is always at the bottom of the buffer.
2014 	 */
2015 	agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size -
2016 	    sizeof (dtrace_aggbuffer_t));
2017 
2018 	if (buf->dtb_offset == 0) {
2019 		/*
2020 		 * We just kludge up approximately 1/8th of the size to be
2021 		 * buckets.  If this guess ends up being routinely
2022 		 * off-the-mark, we may need to dynamically readjust this
2023 		 * based on past performance.
2024 		 */
2025 		uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t);
2026 
2027 		if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) <
2028 		    (uintptr_t)tomax || hashsize == 0) {
2029 			/*
2030 			 * We've been given a ludicrously small buffer;
2031 			 * increment our drop count and leave.
2032 			 */
2033 			dtrace_buffer_drop(buf);
2034 			return;
2035 		}
2036 
2037 		/*
2038 		 * And now, a pathetic attempt to try to get a an odd (or
2039 		 * perchance, a prime) hash size for better hash distribution.
2040 		 */
2041 		if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3))
2042 			hashsize -= DTRACE_AGGHASHSIZE_SLEW;
2043 
2044 		agb->dtagb_hashsize = hashsize;
2045 		agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb -
2046 		    agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *));
2047 		agb->dtagb_free = (uintptr_t)agb->dtagb_hash;
2048 
2049 		for (i = 0; i < agb->dtagb_hashsize; i++)
2050 			agb->dtagb_hash[i] = NULL;
2051 	}
2052 
2053 	ASSERT(agg->dtag_first != NULL);
2054 	ASSERT(agg->dtag_first->dta_intuple);
2055 
2056 	/*
2057 	 * Calculate the hash value based on the key.  Note that we _don't_
2058 	 * include the aggid in the hashing (but we will store it as part of
2059 	 * the key).  The hashing algorithm is Bob Jenkins' "One-at-a-time"
2060 	 * algorithm: a simple, quick algorithm that has no known funnels, and
2061 	 * gets good distribution in practice.  The efficacy of the hashing
2062 	 * algorithm (and a comparison with other algorithms) may be found by
2063 	 * running the ::dtrace_aggstat MDB dcmd.
2064 	 */
2065 	for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2066 		i = act->dta_rec.dtrd_offset - agg->dtag_base;
2067 		limit = i + act->dta_rec.dtrd_size;
2068 		ASSERT(limit <= size);
2069 		isstr = DTRACEACT_ISSTRING(act);
2070 
2071 		for (; i < limit; i++) {
2072 			hashval += data[i];
2073 			hashval += (hashval << 10);
2074 			hashval ^= (hashval >> 6);
2075 
2076 			if (isstr && data[i] == '\0')
2077 				break;
2078 		}
2079 	}
2080 
2081 	hashval += (hashval << 3);
2082 	hashval ^= (hashval >> 11);
2083 	hashval += (hashval << 15);
2084 
2085 	/*
2086 	 * Yes, the divide here is expensive -- but it's generally the least
2087 	 * of the performance issues given the amount of data that we iterate
2088 	 * over to compute hash values, compare data, etc.
2089 	 */
2090 	ndx = hashval % agb->dtagb_hashsize;
2091 
2092 	for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) {
2093 		ASSERT((caddr_t)key >= tomax);
2094 		ASSERT((caddr_t)key < tomax + buf->dtb_size);
2095 
2096 		if (hashval != key->dtak_hashval || key->dtak_size != size)
2097 			continue;
2098 
2099 		kdata = key->dtak_data;
2100 		ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size);
2101 
2102 		for (act = agg->dtag_first; act->dta_intuple;
2103 		    act = act->dta_next) {
2104 			i = act->dta_rec.dtrd_offset - agg->dtag_base;
2105 			limit = i + act->dta_rec.dtrd_size;
2106 			ASSERT(limit <= size);
2107 			isstr = DTRACEACT_ISSTRING(act);
2108 
2109 			for (; i < limit; i++) {
2110 				if (kdata[i] != data[i])
2111 					goto next;
2112 
2113 				if (isstr && data[i] == '\0')
2114 					break;
2115 			}
2116 		}
2117 
2118 		if (action != key->dtak_action) {
2119 			/*
2120 			 * We are aggregating on the same value in the same
2121 			 * aggregation with two different aggregating actions.
2122 			 * (This should have been picked up in the compiler,
2123 			 * so we may be dealing with errant or devious DIF.)
2124 			 * This is an error condition; we indicate as much,
2125 			 * and return.
2126 			 */
2127 			DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
2128 			return;
2129 		}
2130 
2131 		/*
2132 		 * This is a hit:  we need to apply the aggregator to
2133 		 * the value at this key.
2134 		 */
2135 		agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg);
2136 		return;
2137 next:
2138 		continue;
2139 	}
2140 
2141 	/*
2142 	 * We didn't find it.  We need to allocate some zero-filled space,
2143 	 * link it into the hash table appropriately, and apply the aggregator
2144 	 * to the (zero-filled) value.
2145 	 */
2146 	offs = buf->dtb_offset;
2147 	while (offs & (align - 1))
2148 		offs += sizeof (uint32_t);
2149 
2150 	/*
2151 	 * If we don't have enough room to both allocate a new key _and_
2152 	 * its associated data, increment the drop count and return.
2153 	 */
2154 	if ((uintptr_t)tomax + offs + fsize >
2155 	    agb->dtagb_free - sizeof (dtrace_aggkey_t)) {
2156 		dtrace_buffer_drop(buf);
2157 		return;
2158 	}
2159 
2160 	/*CONSTCOND*/
2161 	ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1)));
2162 	key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t));
2163 	agb->dtagb_free -= sizeof (dtrace_aggkey_t);
2164 
2165 	key->dtak_data = kdata = tomax + offs;
2166 	buf->dtb_offset = offs + fsize;
2167 
2168 	/*
2169 	 * Now copy the data across.
2170 	 */
2171 	*((dtrace_aggid_t *)kdata) = agg->dtag_id;
2172 
2173 	for (i = sizeof (dtrace_aggid_t); i < size; i++)
2174 		kdata[i] = data[i];
2175 
2176 	/*
2177 	 * Because strings are not zeroed out by default, we need to iterate
2178 	 * looking for actions that store strings, and we need to explicitly
2179 	 * pad these strings out with zeroes.
2180 	 */
2181 	for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2182 		int nul;
2183 
2184 		if (!DTRACEACT_ISSTRING(act))
2185 			continue;
2186 
2187 		i = act->dta_rec.dtrd_offset - agg->dtag_base;
2188 		limit = i + act->dta_rec.dtrd_size;
2189 		ASSERT(limit <= size);
2190 
2191 		for (nul = 0; i < limit; i++) {
2192 			if (nul) {
2193 				kdata[i] = '\0';
2194 				continue;
2195 			}
2196 
2197 			if (data[i] != '\0')
2198 				continue;
2199 
2200 			nul = 1;
2201 		}
2202 	}
2203 
2204 	for (i = size; i < fsize; i++)
2205 		kdata[i] = 0;
2206 
2207 	key->dtak_hashval = hashval;
2208 	key->dtak_size = size;
2209 	key->dtak_action = action;
2210 	key->dtak_next = agb->dtagb_hash[ndx];
2211 	agb->dtagb_hash[ndx] = key;
2212 
2213 	/*
2214 	 * Finally, apply the aggregator.
2215 	 */
2216 	*((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial;
2217 	agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg);
2218 }
2219 
2220 /*
2221  * Given consumer state, this routine finds a speculation in the INACTIVE
2222  * state and transitions it into the ACTIVE state.  If there is no speculation
2223  * in the INACTIVE state, 0 is returned.  In this case, no error counter is
2224  * incremented -- it is up to the caller to take appropriate action.
2225  */
2226 static int
2227 dtrace_speculation(dtrace_state_t *state)
2228 {
2229 	int i = 0;
2230 	dtrace_speculation_state_t current;
2231 	uint32_t *stat = &state->dts_speculations_unavail, count;
2232 
2233 	while (i < state->dts_nspeculations) {
2234 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2235 
2236 		current = spec->dtsp_state;
2237 
2238 		if (current != DTRACESPEC_INACTIVE) {
2239 			if (current == DTRACESPEC_COMMITTINGMANY ||
2240 			    current == DTRACESPEC_COMMITTING ||
2241 			    current == DTRACESPEC_DISCARDING)
2242 				stat = &state->dts_speculations_busy;
2243 			i++;
2244 			continue;
2245 		}
2246 
2247 		if (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2248 		    current, DTRACESPEC_ACTIVE) == current)
2249 			return (i + 1);
2250 	}
2251 
2252 	/*
2253 	 * We couldn't find a speculation.  If we found as much as a single
2254 	 * busy speculation buffer, we'll attribute this failure as "busy"
2255 	 * instead of "unavail".
2256 	 */
2257 	do {
2258 		count = *stat;
2259 	} while (dtrace_cas32(stat, count, count + 1) != count);
2260 
2261 	return (0);
2262 }
2263 
2264 /*
2265  * This routine commits an active speculation.  If the specified speculation
2266  * is not in a valid state to perform a commit(), this routine will silently do
2267  * nothing.  The state of the specified speculation is transitioned according
2268  * to the state transition diagram outlined in <sys/dtrace_impl.h>
2269  */
2270 static void
2271 dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu,
2272     dtrace_specid_t which)
2273 {
2274 	dtrace_speculation_t *spec;
2275 	dtrace_buffer_t *src, *dest;
2276 	uintptr_t daddr, saddr, dlimit;
2277 	dtrace_speculation_state_t current, new = 0;
2278 	intptr_t offs;
2279 
2280 	if (which == 0)
2281 		return;
2282 
2283 	if (which > state->dts_nspeculations) {
2284 		cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2285 		return;
2286 	}
2287 
2288 	spec = &state->dts_speculations[which - 1];
2289 	src = &spec->dtsp_buffer[cpu];
2290 	dest = &state->dts_buffer[cpu];
2291 
2292 	do {
2293 		current = spec->dtsp_state;
2294 
2295 		if (current == DTRACESPEC_COMMITTINGMANY)
2296 			break;
2297 
2298 		switch (current) {
2299 		case DTRACESPEC_INACTIVE:
2300 		case DTRACESPEC_DISCARDING:
2301 			return;
2302 
2303 		case DTRACESPEC_COMMITTING:
2304 			/*
2305 			 * This is only possible if we are (a) commit()'ing
2306 			 * without having done a prior speculate() on this CPU
2307 			 * and (b) racing with another commit() on a different
2308 			 * CPU.  There's nothing to do -- we just assert that
2309 			 * our offset is 0.
2310 			 */
2311 			ASSERT(src->dtb_offset == 0);
2312 			return;
2313 
2314 		case DTRACESPEC_ACTIVE:
2315 			new = DTRACESPEC_COMMITTING;
2316 			break;
2317 
2318 		case DTRACESPEC_ACTIVEONE:
2319 			/*
2320 			 * This speculation is active on one CPU.  If our
2321 			 * buffer offset is non-zero, we know that the one CPU
2322 			 * must be us.  Otherwise, we are committing on a
2323 			 * different CPU from the speculate(), and we must
2324 			 * rely on being asynchronously cleaned.
2325 			 */
2326 			if (src->dtb_offset != 0) {
2327 				new = DTRACESPEC_COMMITTING;
2328 				break;
2329 			}
2330 			/*FALLTHROUGH*/
2331 
2332 		case DTRACESPEC_ACTIVEMANY:
2333 			new = DTRACESPEC_COMMITTINGMANY;
2334 			break;
2335 
2336 		default:
2337 			ASSERT(0);
2338 		}
2339 	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2340 	    current, new) != current);
2341 
2342 	/*
2343 	 * We have set the state to indicate that we are committing this
2344 	 * speculation.  Now reserve the necessary space in the destination
2345 	 * buffer.
2346 	 */
2347 	if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset,
2348 	    sizeof (uint64_t), state, NULL)) < 0) {
2349 		dtrace_buffer_drop(dest);
2350 		goto out;
2351 	}
2352 
2353 	/*
2354 	 * We have the space; copy the buffer across.  (Note that this is a
2355 	 * highly subobtimal bcopy(); in the unlikely event that this becomes
2356 	 * a serious performance issue, a high-performance DTrace-specific
2357 	 * bcopy() should obviously be invented.)
2358 	 */
2359 	daddr = (uintptr_t)dest->dtb_tomax + offs;
2360 	dlimit = daddr + src->dtb_offset;
2361 	saddr = (uintptr_t)src->dtb_tomax;
2362 
2363 	/*
2364 	 * First, the aligned portion.
2365 	 */
2366 	while (dlimit - daddr >= sizeof (uint64_t)) {
2367 		*((uint64_t *)daddr) = *((uint64_t *)saddr);
2368 
2369 		daddr += sizeof (uint64_t);
2370 		saddr += sizeof (uint64_t);
2371 	}
2372 
2373 	/*
2374 	 * Now any left-over bit...
2375 	 */
2376 	while (dlimit - daddr)
2377 		*((uint8_t *)daddr++) = *((uint8_t *)saddr++);
2378 
2379 	/*
2380 	 * Finally, commit the reserved space in the destination buffer.
2381 	 */
2382 	dest->dtb_offset = offs + src->dtb_offset;
2383 
2384 out:
2385 	/*
2386 	 * If we're lucky enough to be the only active CPU on this speculation
2387 	 * buffer, we can just set the state back to DTRACESPEC_INACTIVE.
2388 	 */
2389 	if (current == DTRACESPEC_ACTIVE ||
2390 	    (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) {
2391 		uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state,
2392 		    DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE);
2393 
2394 		ASSERT(rval == DTRACESPEC_COMMITTING);
2395 	}
2396 
2397 	src->dtb_offset = 0;
2398 	src->dtb_xamot_drops += src->dtb_drops;
2399 	src->dtb_drops = 0;
2400 }
2401 
2402 /*
2403  * This routine discards an active speculation.  If the specified speculation
2404  * is not in a valid state to perform a discard(), this routine will silently
2405  * do nothing.  The state of the specified speculation is transitioned
2406  * according to the state transition diagram outlined in <sys/dtrace_impl.h>
2407  */
2408 static void
2409 dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu,
2410     dtrace_specid_t which)
2411 {
2412 	dtrace_speculation_t *spec;
2413 	dtrace_speculation_state_t current, new = 0;
2414 	dtrace_buffer_t *buf;
2415 
2416 	if (which == 0)
2417 		return;
2418 
2419 	if (which > state->dts_nspeculations) {
2420 		cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2421 		return;
2422 	}
2423 
2424 	spec = &state->dts_speculations[which - 1];
2425 	buf = &spec->dtsp_buffer[cpu];
2426 
2427 	do {
2428 		current = spec->dtsp_state;
2429 
2430 		switch (current) {
2431 		case DTRACESPEC_INACTIVE:
2432 		case DTRACESPEC_COMMITTINGMANY:
2433 		case DTRACESPEC_COMMITTING:
2434 		case DTRACESPEC_DISCARDING:
2435 			return;
2436 
2437 		case DTRACESPEC_ACTIVE:
2438 		case DTRACESPEC_ACTIVEMANY:
2439 			new = DTRACESPEC_DISCARDING;
2440 			break;
2441 
2442 		case DTRACESPEC_ACTIVEONE:
2443 			if (buf->dtb_offset != 0) {
2444 				new = DTRACESPEC_INACTIVE;
2445 			} else {
2446 				new = DTRACESPEC_DISCARDING;
2447 			}
2448 			break;
2449 
2450 		default:
2451 			ASSERT(0);
2452 		}
2453 	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2454 	    current, new) != current);
2455 
2456 	buf->dtb_offset = 0;
2457 	buf->dtb_drops = 0;
2458 }
2459 
2460 /*
2461  * Note:  not called from probe context.  This function is called
2462  * asynchronously from cross call context to clean any speculations that are
2463  * in the COMMITTINGMANY or DISCARDING states.  These speculations may not be
2464  * transitioned back to the INACTIVE state until all CPUs have cleaned the
2465  * speculation.
2466  */
2467 static void
2468 dtrace_speculation_clean_here(dtrace_state_t *state)
2469 {
2470 	dtrace_icookie_t cookie;
2471 	processorid_t cpu = curcpu;
2472 	dtrace_buffer_t *dest = &state->dts_buffer[cpu];
2473 	dtrace_specid_t i;
2474 
2475 	cookie = dtrace_interrupt_disable();
2476 
2477 	if (dest->dtb_tomax == NULL) {
2478 		dtrace_interrupt_enable(cookie);
2479 		return;
2480 	}
2481 
2482 	for (i = 0; i < state->dts_nspeculations; i++) {
2483 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2484 		dtrace_buffer_t *src = &spec->dtsp_buffer[cpu];
2485 
2486 		if (src->dtb_tomax == NULL)
2487 			continue;
2488 
2489 		if (spec->dtsp_state == DTRACESPEC_DISCARDING) {
2490 			src->dtb_offset = 0;
2491 			continue;
2492 		}
2493 
2494 		if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2495 			continue;
2496 
2497 		if (src->dtb_offset == 0)
2498 			continue;
2499 
2500 		dtrace_speculation_commit(state, cpu, i + 1);
2501 	}
2502 
2503 	dtrace_interrupt_enable(cookie);
2504 }
2505 
2506 /*
2507  * Note:  not called from probe context.  This function is called
2508  * asynchronously (and at a regular interval) to clean any speculations that
2509  * are in the COMMITTINGMANY or DISCARDING states.  If it discovers that there
2510  * is work to be done, it cross calls all CPUs to perform that work;
2511  * COMMITMANY and DISCARDING speculations may not be transitioned back to the
2512  * INACTIVE state until they have been cleaned by all CPUs.
2513  */
2514 static void
2515 dtrace_speculation_clean(dtrace_state_t *state)
2516 {
2517 	int work = 0, rv;
2518 	dtrace_specid_t i;
2519 
2520 	for (i = 0; i < state->dts_nspeculations; i++) {
2521 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2522 
2523 		ASSERT(!spec->dtsp_cleaning);
2524 
2525 		if (spec->dtsp_state != DTRACESPEC_DISCARDING &&
2526 		    spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2527 			continue;
2528 
2529 		work++;
2530 		spec->dtsp_cleaning = 1;
2531 	}
2532 
2533 	if (!work)
2534 		return;
2535 
2536 	dtrace_xcall(DTRACE_CPUALL,
2537 	    (dtrace_xcall_t)dtrace_speculation_clean_here, state);
2538 
2539 	/*
2540 	 * We now know that all CPUs have committed or discarded their
2541 	 * speculation buffers, as appropriate.  We can now set the state
2542 	 * to inactive.
2543 	 */
2544 	for (i = 0; i < state->dts_nspeculations; i++) {
2545 		dtrace_speculation_t *spec = &state->dts_speculations[i];
2546 		dtrace_speculation_state_t current, new;
2547 
2548 		if (!spec->dtsp_cleaning)
2549 			continue;
2550 
2551 		current = spec->dtsp_state;
2552 		ASSERT(current == DTRACESPEC_DISCARDING ||
2553 		    current == DTRACESPEC_COMMITTINGMANY);
2554 
2555 		new = DTRACESPEC_INACTIVE;
2556 
2557 		rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new);
2558 		ASSERT(rv == current);
2559 		spec->dtsp_cleaning = 0;
2560 	}
2561 }
2562 
2563 /*
2564  * Called as part of a speculate() to get the speculative buffer associated
2565  * with a given speculation.  Returns NULL if the specified speculation is not
2566  * in an ACTIVE state.  If the speculation is in the ACTIVEONE state -- and
2567  * the active CPU is not the specified CPU -- the speculation will be
2568  * atomically transitioned into the ACTIVEMANY state.
2569  */
2570 static dtrace_buffer_t *
2571 dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid,
2572     dtrace_specid_t which)
2573 {
2574 	dtrace_speculation_t *spec;
2575 	dtrace_speculation_state_t current, new = 0;
2576 	dtrace_buffer_t *buf;
2577 
2578 	if (which == 0)
2579 		return (NULL);
2580 
2581 	if (which > state->dts_nspeculations) {
2582 		cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2583 		return (NULL);
2584 	}
2585 
2586 	spec = &state->dts_speculations[which - 1];
2587 	buf = &spec->dtsp_buffer[cpuid];
2588 
2589 	do {
2590 		current = spec->dtsp_state;
2591 
2592 		switch (current) {
2593 		case DTRACESPEC_INACTIVE:
2594 		case DTRACESPEC_COMMITTINGMANY:
2595 		case DTRACESPEC_DISCARDING:
2596 			return (NULL);
2597 
2598 		case DTRACESPEC_COMMITTING:
2599 			ASSERT(buf->dtb_offset == 0);
2600 			return (NULL);
2601 
2602 		case DTRACESPEC_ACTIVEONE:
2603 			/*
2604 			 * This speculation is currently active on one CPU.
2605 			 * Check the offset in the buffer; if it's non-zero,
2606 			 * that CPU must be us (and we leave the state alone).
2607 			 * If it's zero, assume that we're starting on a new
2608 			 * CPU -- and change the state to indicate that the
2609 			 * speculation is active on more than one CPU.
2610 			 */
2611 			if (buf->dtb_offset != 0)
2612 				return (buf);
2613 
2614 			new = DTRACESPEC_ACTIVEMANY;
2615 			break;
2616 
2617 		case DTRACESPEC_ACTIVEMANY:
2618 			return (buf);
2619 
2620 		case DTRACESPEC_ACTIVE:
2621 			new = DTRACESPEC_ACTIVEONE;
2622 			break;
2623 
2624 		default:
2625 			ASSERT(0);
2626 		}
2627 	} while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2628 	    current, new) != current);
2629 
2630 	ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY);
2631 	return (buf);
2632 }
2633 
2634 /*
2635  * Return a string.  In the event that the user lacks the privilege to access
2636  * arbitrary kernel memory, we copy the string out to scratch memory so that we
2637  * don't fail access checking.
2638  *
2639  * dtrace_dif_variable() uses this routine as a helper for various
2640  * builtin values such as 'execname' and 'probefunc.'
2641  */
2642 uintptr_t
2643 dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state,
2644     dtrace_mstate_t *mstate)
2645 {
2646 	uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
2647 	uintptr_t ret;
2648 	size_t strsz;
2649 
2650 	/*
2651 	 * The easy case: this probe is allowed to read all of memory, so
2652 	 * we can just return this as a vanilla pointer.
2653 	 */
2654 	if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
2655 		return (addr);
2656 
2657 	/*
2658 	 * This is the tougher case: we copy the string in question from
2659 	 * kernel memory into scratch memory and return it that way: this
2660 	 * ensures that we won't trip up when access checking tests the
2661 	 * BYREF return value.
2662 	 */
2663 	strsz = dtrace_strlen((char *)addr, size) + 1;
2664 
2665 	if (mstate->dtms_scratch_ptr + strsz >
2666 	    mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
2667 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
2668 		return (0);
2669 	}
2670 
2671 	dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
2672 	    strsz);
2673 	ret = mstate->dtms_scratch_ptr;
2674 	mstate->dtms_scratch_ptr += strsz;
2675 	return (ret);
2676 }
2677 
2678 /*
2679  * Return a string from a memoy address which is known to have one or
2680  * more concatenated, individually zero terminated, sub-strings.
2681  * In the event that the user lacks the privilege to access
2682  * arbitrary kernel memory, we copy the string out to scratch memory so that we
2683  * don't fail access checking.
2684  *
2685  * dtrace_dif_variable() uses this routine as a helper for various
2686  * builtin values such as 'execargs'.
2687  */
2688 static uintptr_t
2689 dtrace_dif_varstrz(uintptr_t addr, size_t strsz, dtrace_state_t *state,
2690     dtrace_mstate_t *mstate)
2691 {
2692 	char *p;
2693 	size_t i;
2694 	uintptr_t ret;
2695 
2696 	if (mstate->dtms_scratch_ptr + strsz >
2697 	    mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
2698 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
2699 		return (0);
2700 	}
2701 
2702 	dtrace_bcopy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
2703 	    strsz);
2704 
2705 	/* Replace sub-string termination characters with a space. */
2706 	for (p = (char *) mstate->dtms_scratch_ptr, i = 0; i < strsz - 1;
2707 	    p++, i++)
2708 		if (*p == '\0')
2709 			*p = ' ';
2710 
2711 	ret = mstate->dtms_scratch_ptr;
2712 	mstate->dtms_scratch_ptr += strsz;
2713 	return (ret);
2714 }
2715 
2716 /*
2717  * This function implements the DIF emulator's variable lookups.  The emulator
2718  * passes a reserved variable identifier and optional built-in array index.
2719  */
2720 static uint64_t
2721 dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v,
2722     uint64_t ndx)
2723 {
2724 	/*
2725 	 * If we're accessing one of the uncached arguments, we'll turn this
2726 	 * into a reference in the args array.
2727 	 */
2728 	if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) {
2729 		ndx = v - DIF_VAR_ARG0;
2730 		v = DIF_VAR_ARGS;
2731 	}
2732 
2733 	switch (v) {
2734 	case DIF_VAR_ARGS:
2735 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS);
2736 		if (ndx >= sizeof (mstate->dtms_arg) /
2737 		    sizeof (mstate->dtms_arg[0])) {
2738 			int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2739 			dtrace_provider_t *pv;
2740 			uint64_t val;
2741 
2742 			pv = mstate->dtms_probe->dtpr_provider;
2743 			if (pv->dtpv_pops.dtps_getargval != NULL)
2744 				val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg,
2745 				    mstate->dtms_probe->dtpr_id,
2746 				    mstate->dtms_probe->dtpr_arg, ndx, aframes);
2747 			else
2748 				val = dtrace_getarg(ndx, aframes);
2749 
2750 			/*
2751 			 * This is regrettably required to keep the compiler
2752 			 * from tail-optimizing the call to dtrace_getarg().
2753 			 * The condition always evaluates to true, but the
2754 			 * compiler has no way of figuring that out a priori.
2755 			 * (None of this would be necessary if the compiler
2756 			 * could be relied upon to _always_ tail-optimize
2757 			 * the call to dtrace_getarg() -- but it can't.)
2758 			 */
2759 			if (mstate->dtms_probe != NULL)
2760 				return (val);
2761 
2762 			ASSERT(0);
2763 		}
2764 
2765 		return (mstate->dtms_arg[ndx]);
2766 
2767 #if defined(sun)
2768 	case DIF_VAR_UREGS: {
2769 		klwp_t *lwp;
2770 
2771 		if (!dtrace_priv_proc(state))
2772 			return (0);
2773 
2774 		if ((lwp = curthread->t_lwp) == NULL) {
2775 			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
2776 			cpu_core[curcpu].cpuc_dtrace_illval = NULL;
2777 			return (0);
2778 		}
2779 
2780 		return (dtrace_getreg(lwp->lwp_regs, ndx));
2781 		return (0);
2782 	}
2783 #else
2784 	case DIF_VAR_UREGS: {
2785 		struct trapframe *tframe;
2786 
2787 		if (!dtrace_priv_proc(state))
2788 			return (0);
2789 
2790 		if ((tframe = curthread->td_frame) == NULL) {
2791 			DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
2792 			cpu_core[curcpu].cpuc_dtrace_illval = 0;
2793 			return (0);
2794 		}
2795 
2796 		return (dtrace_getreg(tframe, ndx));
2797 	}
2798 #endif
2799 
2800 	case DIF_VAR_CURTHREAD:
2801 		if (!dtrace_priv_kernel(state))
2802 			return (0);
2803 		return ((uint64_t)(uintptr_t)curthread);
2804 
2805 	case DIF_VAR_TIMESTAMP:
2806 		if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) {
2807 			mstate->dtms_timestamp = dtrace_gethrtime();
2808 			mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP;
2809 		}
2810 		return (mstate->dtms_timestamp);
2811 
2812 	case DIF_VAR_VTIMESTAMP:
2813 		ASSERT(dtrace_vtime_references != 0);
2814 		return (curthread->t_dtrace_vtime);
2815 
2816 	case DIF_VAR_WALLTIMESTAMP:
2817 		if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) {
2818 			mstate->dtms_walltimestamp = dtrace_gethrestime();
2819 			mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP;
2820 		}
2821 		return (mstate->dtms_walltimestamp);
2822 
2823 #if defined(sun)
2824 	case DIF_VAR_IPL:
2825 		if (!dtrace_priv_kernel(state))
2826 			return (0);
2827 		if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) {
2828 			mstate->dtms_ipl = dtrace_getipl();
2829 			mstate->dtms_present |= DTRACE_MSTATE_IPL;
2830 		}
2831 		return (mstate->dtms_ipl);
2832 #endif
2833 
2834 	case DIF_VAR_EPID:
2835 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID);
2836 		return (mstate->dtms_epid);
2837 
2838 	case DIF_VAR_ID:
2839 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2840 		return (mstate->dtms_probe->dtpr_id);
2841 
2842 	case DIF_VAR_STACKDEPTH:
2843 		if (!dtrace_priv_kernel(state))
2844 			return (0);
2845 		if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) {
2846 			int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2847 
2848 			mstate->dtms_stackdepth = dtrace_getstackdepth(aframes);
2849 			mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH;
2850 		}
2851 		return (mstate->dtms_stackdepth);
2852 
2853 	case DIF_VAR_USTACKDEPTH:
2854 		if (!dtrace_priv_proc(state))
2855 			return (0);
2856 		if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) {
2857 			/*
2858 			 * See comment in DIF_VAR_PID.
2859 			 */
2860 			if (DTRACE_ANCHORED(mstate->dtms_probe) &&
2861 			    CPU_ON_INTR(CPU)) {
2862 				mstate->dtms_ustackdepth = 0;
2863 			} else {
2864 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
2865 				mstate->dtms_ustackdepth =
2866 				    dtrace_getustackdepth();
2867 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
2868 			}
2869 			mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH;
2870 		}
2871 		return (mstate->dtms_ustackdepth);
2872 
2873 	case DIF_VAR_CALLER:
2874 		if (!dtrace_priv_kernel(state))
2875 			return (0);
2876 		if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) {
2877 			int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2878 
2879 			if (!DTRACE_ANCHORED(mstate->dtms_probe)) {
2880 				/*
2881 				 * If this is an unanchored probe, we are
2882 				 * required to go through the slow path:
2883 				 * dtrace_caller() only guarantees correct
2884 				 * results for anchored probes.
2885 				 */
2886 				pc_t caller[2] = {0, 0};
2887 
2888 				dtrace_getpcstack(caller, 2, aframes,
2889 				    (uint32_t *)(uintptr_t)mstate->dtms_arg[0]);
2890 				mstate->dtms_caller = caller[1];
2891 			} else if ((mstate->dtms_caller =
2892 			    dtrace_caller(aframes)) == -1) {
2893 				/*
2894 				 * We have failed to do this the quick way;
2895 				 * we must resort to the slower approach of
2896 				 * calling dtrace_getpcstack().
2897 				 */
2898 				pc_t caller = 0;
2899 
2900 				dtrace_getpcstack(&caller, 1, aframes, NULL);
2901 				mstate->dtms_caller = caller;
2902 			}
2903 
2904 			mstate->dtms_present |= DTRACE_MSTATE_CALLER;
2905 		}
2906 		return (mstate->dtms_caller);
2907 
2908 	case DIF_VAR_UCALLER:
2909 		if (!dtrace_priv_proc(state))
2910 			return (0);
2911 
2912 		if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) {
2913 			uint64_t ustack[3];
2914 
2915 			/*
2916 			 * dtrace_getupcstack() fills in the first uint64_t
2917 			 * with the current PID.  The second uint64_t will
2918 			 * be the program counter at user-level.  The third
2919 			 * uint64_t will contain the caller, which is what
2920 			 * we're after.
2921 			 */
2922 			ustack[2] = 0;
2923 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
2924 			dtrace_getupcstack(ustack, 3);
2925 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
2926 			mstate->dtms_ucaller = ustack[2];
2927 			mstate->dtms_present |= DTRACE_MSTATE_UCALLER;
2928 		}
2929 
2930 		return (mstate->dtms_ucaller);
2931 
2932 	case DIF_VAR_PROBEPROV:
2933 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2934 		return (dtrace_dif_varstr(
2935 		    (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name,
2936 		    state, mstate));
2937 
2938 	case DIF_VAR_PROBEMOD:
2939 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2940 		return (dtrace_dif_varstr(
2941 		    (uintptr_t)mstate->dtms_probe->dtpr_mod,
2942 		    state, mstate));
2943 
2944 	case DIF_VAR_PROBEFUNC:
2945 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2946 		return (dtrace_dif_varstr(
2947 		    (uintptr_t)mstate->dtms_probe->dtpr_func,
2948 		    state, mstate));
2949 
2950 	case DIF_VAR_PROBENAME:
2951 		ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2952 		return (dtrace_dif_varstr(
2953 		    (uintptr_t)mstate->dtms_probe->dtpr_name,
2954 		    state, mstate));
2955 
2956 	case DIF_VAR_PID:
2957 		if (!dtrace_priv_proc(state))
2958 			return (0);
2959 
2960 #if defined(sun)
2961 		/*
2962 		 * Note that we are assuming that an unanchored probe is
2963 		 * always due to a high-level interrupt.  (And we're assuming
2964 		 * that there is only a single high level interrupt.)
2965 		 */
2966 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
2967 			return (pid0.pid_id);
2968 
2969 		/*
2970 		 * It is always safe to dereference one's own t_procp pointer:
2971 		 * it always points to a valid, allocated proc structure.
2972 		 * Further, it is always safe to dereference the p_pidp member
2973 		 * of one's own proc structure.  (These are truisms becuase
2974 		 * threads and processes don't clean up their own state --
2975 		 * they leave that task to whomever reaps them.)
2976 		 */
2977 		return ((uint64_t)curthread->t_procp->p_pidp->pid_id);
2978 #else
2979 		return ((uint64_t)curproc->p_pid);
2980 #endif
2981 
2982 	case DIF_VAR_PPID:
2983 		if (!dtrace_priv_proc(state))
2984 			return (0);
2985 
2986 #if defined(sun)
2987 		/*
2988 		 * See comment in DIF_VAR_PID.
2989 		 */
2990 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
2991 			return (pid0.pid_id);
2992 
2993 		/*
2994 		 * It is always safe to dereference one's own t_procp pointer:
2995 		 * it always points to a valid, allocated proc structure.
2996 		 * (This is true because threads don't clean up their own
2997 		 * state -- they leave that task to whomever reaps them.)
2998 		 */
2999 		return ((uint64_t)curthread->t_procp->p_ppid);
3000 #else
3001 		return ((uint64_t)curproc->p_pptr->p_pid);
3002 #endif
3003 
3004 	case DIF_VAR_TID:
3005 #if defined(sun)
3006 		/*
3007 		 * See comment in DIF_VAR_PID.
3008 		 */
3009 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3010 			return (0);
3011 #endif
3012 
3013 		return ((uint64_t)curthread->t_tid);
3014 
3015 	case DIF_VAR_EXECARGS: {
3016 		struct pargs *p_args = curthread->td_proc->p_args;
3017 
3018 		if (p_args == NULL)
3019 			return(0);
3020 
3021 		return (dtrace_dif_varstrz(
3022 		    (uintptr_t) p_args->ar_args, p_args->ar_length, state, mstate));
3023 	}
3024 
3025 	case DIF_VAR_EXECNAME:
3026 #if defined(sun)
3027 		if (!dtrace_priv_proc(state))
3028 			return (0);
3029 
3030 		/*
3031 		 * See comment in DIF_VAR_PID.
3032 		 */
3033 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3034 			return ((uint64_t)(uintptr_t)p0.p_user.u_comm);
3035 
3036 		/*
3037 		 * It is always safe to dereference one's own t_procp pointer:
3038 		 * it always points to a valid, allocated proc structure.
3039 		 * (This is true because threads don't clean up their own
3040 		 * state -- they leave that task to whomever reaps them.)
3041 		 */
3042 		return (dtrace_dif_varstr(
3043 		    (uintptr_t)curthread->t_procp->p_user.u_comm,
3044 		    state, mstate));
3045 #else
3046 		return (dtrace_dif_varstr(
3047 		    (uintptr_t) curthread->td_proc->p_comm, state, mstate));
3048 #endif
3049 
3050 	case DIF_VAR_ZONENAME:
3051 #if defined(sun)
3052 		if (!dtrace_priv_proc(state))
3053 			return (0);
3054 
3055 		/*
3056 		 * See comment in DIF_VAR_PID.
3057 		 */
3058 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3059 			return ((uint64_t)(uintptr_t)p0.p_zone->zone_name);
3060 
3061 		/*
3062 		 * It is always safe to dereference one's own t_procp pointer:
3063 		 * it always points to a valid, allocated proc structure.
3064 		 * (This is true because threads don't clean up their own
3065 		 * state -- they leave that task to whomever reaps them.)
3066 		 */
3067 		return (dtrace_dif_varstr(
3068 		    (uintptr_t)curthread->t_procp->p_zone->zone_name,
3069 		    state, mstate));
3070 #else
3071 		return (0);
3072 #endif
3073 
3074 	case DIF_VAR_UID:
3075 		if (!dtrace_priv_proc(state))
3076 			return (0);
3077 
3078 #if defined(sun)
3079 		/*
3080 		 * See comment in DIF_VAR_PID.
3081 		 */
3082 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3083 			return ((uint64_t)p0.p_cred->cr_uid);
3084 #endif
3085 
3086 		/*
3087 		 * It is always safe to dereference one's own t_procp pointer:
3088 		 * it always points to a valid, allocated proc structure.
3089 		 * (This is true because threads don't clean up their own
3090 		 * state -- they leave that task to whomever reaps them.)
3091 		 *
3092 		 * Additionally, it is safe to dereference one's own process
3093 		 * credential, since this is never NULL after process birth.
3094 		 */
3095 		return ((uint64_t)curthread->t_procp->p_cred->cr_uid);
3096 
3097 	case DIF_VAR_GID:
3098 		if (!dtrace_priv_proc(state))
3099 			return (0);
3100 
3101 #if defined(sun)
3102 		/*
3103 		 * See comment in DIF_VAR_PID.
3104 		 */
3105 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3106 			return ((uint64_t)p0.p_cred->cr_gid);
3107 #endif
3108 
3109 		/*
3110 		 * It is always safe to dereference one's own t_procp pointer:
3111 		 * it always points to a valid, allocated proc structure.
3112 		 * (This is true because threads don't clean up their own
3113 		 * state -- they leave that task to whomever reaps them.)
3114 		 *
3115 		 * Additionally, it is safe to dereference one's own process
3116 		 * credential, since this is never NULL after process birth.
3117 		 */
3118 		return ((uint64_t)curthread->t_procp->p_cred->cr_gid);
3119 
3120 	case DIF_VAR_ERRNO: {
3121 #if defined(sun)
3122 		klwp_t *lwp;
3123 		if (!dtrace_priv_proc(state))
3124 			return (0);
3125 
3126 		/*
3127 		 * See comment in DIF_VAR_PID.
3128 		 */
3129 		if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3130 			return (0);
3131 
3132 		/*
3133 		 * It is always safe to dereference one's own t_lwp pointer in
3134 		 * the event that this pointer is non-NULL.  (This is true
3135 		 * because threads and lwps don't clean up their own state --
3136 		 * they leave that task to whomever reaps them.)
3137 		 */
3138 		if ((lwp = curthread->t_lwp) == NULL)
3139 			return (0);
3140 
3141 		return ((uint64_t)lwp->lwp_errno);
3142 #else
3143 		return (curthread->td_errno);
3144 #endif
3145 	}
3146 	default:
3147 		DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
3148 		return (0);
3149 	}
3150 }
3151 
3152 /*
3153  * Emulate the execution of DTrace ID subroutines invoked by the call opcode.
3154  * Notice that we don't bother validating the proper number of arguments or
3155  * their types in the tuple stack.  This isn't needed because all argument
3156  * interpretation is safe because of our load safety -- the worst that can
3157  * happen is that a bogus program can obtain bogus results.
3158  */
3159 static void
3160 dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs,
3161     dtrace_key_t *tupregs, int nargs,
3162     dtrace_mstate_t *mstate, dtrace_state_t *state)
3163 {
3164 	volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
3165 	volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
3166 	dtrace_vstate_t *vstate = &state->dts_vstate;
3167 
3168 #if defined(sun)
3169 	union {
3170 		mutex_impl_t mi;
3171 		uint64_t mx;
3172 	} m;
3173 
3174 	union {
3175 		krwlock_t ri;
3176 		uintptr_t rw;
3177 	} r;
3178 #else
3179 	struct thread *lowner;
3180 	union {
3181 		struct lock_object *li;
3182 		uintptr_t lx;
3183 	} l;
3184 #endif
3185 
3186 	switch (subr) {
3187 	case DIF_SUBR_RAND:
3188 		regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875;
3189 		break;
3190 
3191 #if defined(sun)
3192 	case DIF_SUBR_MUTEX_OWNED:
3193 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3194 		    mstate, vstate)) {
3195 			regs[rd] = 0;
3196 			break;
3197 		}
3198 
3199 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3200 		if (MUTEX_TYPE_ADAPTIVE(&m.mi))
3201 			regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER;
3202 		else
3203 			regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock);
3204 		break;
3205 
3206 	case DIF_SUBR_MUTEX_OWNER:
3207 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3208 		    mstate, vstate)) {
3209 			regs[rd] = 0;
3210 			break;
3211 		}
3212 
3213 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3214 		if (MUTEX_TYPE_ADAPTIVE(&m.mi) &&
3215 		    MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER)
3216 			regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi);
3217 		else
3218 			regs[rd] = 0;
3219 		break;
3220 
3221 	case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
3222 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3223 		    mstate, vstate)) {
3224 			regs[rd] = 0;
3225 			break;
3226 		}
3227 
3228 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3229 		regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi);
3230 		break;
3231 
3232 	case DIF_SUBR_MUTEX_TYPE_SPIN:
3233 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3234 		    mstate, vstate)) {
3235 			regs[rd] = 0;
3236 			break;
3237 		}
3238 
3239 		m.mx = dtrace_load64(tupregs[0].dttk_value);
3240 		regs[rd] = MUTEX_TYPE_SPIN(&m.mi);
3241 		break;
3242 
3243 	case DIF_SUBR_RW_READ_HELD: {
3244 		uintptr_t tmp;
3245 
3246 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3247 		    mstate, vstate)) {
3248 			regs[rd] = 0;
3249 			break;
3250 		}
3251 
3252 		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3253 		regs[rd] = _RW_READ_HELD(&r.ri, tmp);
3254 		break;
3255 	}
3256 
3257 	case DIF_SUBR_RW_WRITE_HELD:
3258 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3259 		    mstate, vstate)) {
3260 			regs[rd] = 0;
3261 			break;
3262 		}
3263 
3264 		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3265 		regs[rd] = _RW_WRITE_HELD(&r.ri);
3266 		break;
3267 
3268 	case DIF_SUBR_RW_ISWRITER:
3269 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3270 		    mstate, vstate)) {
3271 			regs[rd] = 0;
3272 			break;
3273 		}
3274 
3275 		r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3276 		regs[rd] = _RW_ISWRITER(&r.ri);
3277 		break;
3278 
3279 #else
3280 	case DIF_SUBR_MUTEX_OWNED:
3281 		if (!dtrace_canload(tupregs[0].dttk_value,
3282 			sizeof (struct lock_object), mstate, vstate)) {
3283 			regs[rd] = 0;
3284 			break;
3285 		}
3286 		l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3287 		regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3288 		break;
3289 
3290 	case DIF_SUBR_MUTEX_OWNER:
3291 		if (!dtrace_canload(tupregs[0].dttk_value,
3292 			sizeof (struct lock_object), mstate, vstate)) {
3293 			regs[rd] = 0;
3294 			break;
3295 		}
3296 		l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3297 		LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3298 		regs[rd] = (uintptr_t)lowner;
3299 		break;
3300 
3301 	case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
3302 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx),
3303 		    mstate, vstate)) {
3304 			regs[rd] = 0;
3305 			break;
3306 		}
3307 		l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3308 		/* XXX - should be only LC_SLEEPABLE? */
3309 		regs[rd] = (LOCK_CLASS(l.li)->lc_flags &
3310 		    (LC_SLEEPLOCK | LC_SLEEPABLE)) != 0;
3311 		break;
3312 
3313 	case DIF_SUBR_MUTEX_TYPE_SPIN:
3314 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx),
3315 		    mstate, vstate)) {
3316 			regs[rd] = 0;
3317 			break;
3318 		}
3319 		l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3320 		regs[rd] = (LOCK_CLASS(l.li)->lc_flags & LC_SPINLOCK) != 0;
3321 		break;
3322 
3323 	case DIF_SUBR_RW_READ_HELD:
3324 	case DIF_SUBR_SX_SHARED_HELD:
3325 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3326 		    mstate, vstate)) {
3327 			regs[rd] = 0;
3328 			break;
3329 		}
3330 		l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3331 		regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) &&
3332 		    lowner == NULL;
3333 		break;
3334 
3335 	case DIF_SUBR_RW_WRITE_HELD:
3336 	case DIF_SUBR_SX_EXCLUSIVE_HELD:
3337 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3338 		    mstate, vstate)) {
3339 			regs[rd] = 0;
3340 			break;
3341 		}
3342 		l.lx = dtrace_loadptr(tupregs[0].dttk_value);
3343 		LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3344 		regs[rd] = (lowner == curthread);
3345 		break;
3346 
3347 	case DIF_SUBR_RW_ISWRITER:
3348 	case DIF_SUBR_SX_ISEXCLUSIVE:
3349 		if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3350 		    mstate, vstate)) {
3351 			regs[rd] = 0;
3352 			break;
3353 		}
3354 		l.lx = dtrace_loadptr(tupregs[0].dttk_value);
3355 		regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) &&
3356 		    lowner != NULL;
3357 		break;
3358 #endif /* ! defined(sun) */
3359 
3360 	case DIF_SUBR_BCOPY: {
3361 		/*
3362 		 * We need to be sure that the destination is in the scratch
3363 		 * region -- no other region is allowed.
3364 		 */
3365 		uintptr_t src = tupregs[0].dttk_value;
3366 		uintptr_t dest = tupregs[1].dttk_value;
3367 		size_t size = tupregs[2].dttk_value;
3368 
3369 		if (!dtrace_inscratch(dest, size, mstate)) {
3370 			*flags |= CPU_DTRACE_BADADDR;
3371 			*illval = regs[rd];
3372 			break;
3373 		}
3374 
3375 		if (!dtrace_canload(src, size, mstate, vstate)) {
3376 			regs[rd] = 0;
3377 			break;
3378 		}
3379 
3380 		dtrace_bcopy((void *)src, (void *)dest, size);
3381 		break;
3382 	}
3383 
3384 	case DIF_SUBR_ALLOCA:
3385 	case DIF_SUBR_COPYIN: {
3386 		uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
3387 		uint64_t size =
3388 		    tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value;
3389 		size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size;
3390 
3391 		/*
3392 		 * This action doesn't require any credential checks since
3393 		 * probes will not activate in user contexts to which the
3394 		 * enabling user does not have permissions.
3395 		 */
3396 
3397 		/*
3398 		 * Rounding up the user allocation size could have overflowed
3399 		 * a large, bogus allocation (like -1ULL) to 0.
3400 		 */
3401 		if (scratch_size < size ||
3402 		    !DTRACE_INSCRATCH(mstate, scratch_size)) {
3403 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3404 			regs[rd] = 0;
3405 			break;
3406 		}
3407 
3408 		if (subr == DIF_SUBR_COPYIN) {
3409 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3410 			dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
3411 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3412 		}
3413 
3414 		mstate->dtms_scratch_ptr += scratch_size;
3415 		regs[rd] = dest;
3416 		break;
3417 	}
3418 
3419 	case DIF_SUBR_COPYINTO: {
3420 		uint64_t size = tupregs[1].dttk_value;
3421 		uintptr_t dest = tupregs[2].dttk_value;
3422 
3423 		/*
3424 		 * This action doesn't require any credential checks since
3425 		 * probes will not activate in user contexts to which the
3426 		 * enabling user does not have permissions.
3427 		 */
3428 		if (!dtrace_inscratch(dest, size, mstate)) {
3429 			*flags |= CPU_DTRACE_BADADDR;
3430 			*illval = regs[rd];
3431 			break;
3432 		}
3433 
3434 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3435 		dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
3436 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3437 		break;
3438 	}
3439 
3440 	case DIF_SUBR_COPYINSTR: {
3441 		uintptr_t dest = mstate->dtms_scratch_ptr;
3442 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3443 
3444 		if (nargs > 1 && tupregs[1].dttk_value < size)
3445 			size = tupregs[1].dttk_value + 1;
3446 
3447 		/*
3448 		 * This action doesn't require any credential checks since
3449 		 * probes will not activate in user contexts to which the
3450 		 * enabling user does not have permissions.
3451 		 */
3452 		if (!DTRACE_INSCRATCH(mstate, size)) {
3453 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3454 			regs[rd] = 0;
3455 			break;
3456 		}
3457 
3458 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3459 		dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags);
3460 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3461 
3462 		((char *)dest)[size - 1] = '\0';
3463 		mstate->dtms_scratch_ptr += size;
3464 		regs[rd] = dest;
3465 		break;
3466 	}
3467 
3468 #if defined(sun)
3469 	case DIF_SUBR_MSGSIZE:
3470 	case DIF_SUBR_MSGDSIZE: {
3471 		uintptr_t baddr = tupregs[0].dttk_value, daddr;
3472 		uintptr_t wptr, rptr;
3473 		size_t count = 0;
3474 		int cont = 0;
3475 
3476 		while (baddr != 0 && !(*flags & CPU_DTRACE_FAULT)) {
3477 
3478 			if (!dtrace_canload(baddr, sizeof (mblk_t), mstate,
3479 			    vstate)) {
3480 				regs[rd] = 0;
3481 				break;
3482 			}
3483 
3484 			wptr = dtrace_loadptr(baddr +
3485 			    offsetof(mblk_t, b_wptr));
3486 
3487 			rptr = dtrace_loadptr(baddr +
3488 			    offsetof(mblk_t, b_rptr));
3489 
3490 			if (wptr < rptr) {
3491 				*flags |= CPU_DTRACE_BADADDR;
3492 				*illval = tupregs[0].dttk_value;
3493 				break;
3494 			}
3495 
3496 			daddr = dtrace_loadptr(baddr +
3497 			    offsetof(mblk_t, b_datap));
3498 
3499 			baddr = dtrace_loadptr(baddr +
3500 			    offsetof(mblk_t, b_cont));
3501 
3502 			/*
3503 			 * We want to prevent against denial-of-service here,
3504 			 * so we're only going to search the list for
3505 			 * dtrace_msgdsize_max mblks.
3506 			 */
3507 			if (cont++ > dtrace_msgdsize_max) {
3508 				*flags |= CPU_DTRACE_ILLOP;
3509 				break;
3510 			}
3511 
3512 			if (subr == DIF_SUBR_MSGDSIZE) {
3513 				if (dtrace_load8(daddr +
3514 				    offsetof(dblk_t, db_type)) != M_DATA)
3515 					continue;
3516 			}
3517 
3518 			count += wptr - rptr;
3519 		}
3520 
3521 		if (!(*flags & CPU_DTRACE_FAULT))
3522 			regs[rd] = count;
3523 
3524 		break;
3525 	}
3526 #endif
3527 
3528 	case DIF_SUBR_PROGENYOF: {
3529 		pid_t pid = tupregs[0].dttk_value;
3530 		proc_t *p;
3531 		int rval = 0;
3532 
3533 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3534 
3535 		for (p = curthread->t_procp; p != NULL; p = p->p_parent) {
3536 #if defined(sun)
3537 			if (p->p_pidp->pid_id == pid) {
3538 #else
3539 			if (p->p_pid == pid) {
3540 #endif
3541 				rval = 1;
3542 				break;
3543 			}
3544 		}
3545 
3546 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3547 
3548 		regs[rd] = rval;
3549 		break;
3550 	}
3551 
3552 	case DIF_SUBR_SPECULATION:
3553 		regs[rd] = dtrace_speculation(state);
3554 		break;
3555 
3556 	case DIF_SUBR_COPYOUT: {
3557 		uintptr_t kaddr = tupregs[0].dttk_value;
3558 		uintptr_t uaddr = tupregs[1].dttk_value;
3559 		uint64_t size = tupregs[2].dttk_value;
3560 
3561 		if (!dtrace_destructive_disallow &&
3562 		    dtrace_priv_proc_control(state) &&
3563 		    !dtrace_istoxic(kaddr, size)) {
3564 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3565 			dtrace_copyout(kaddr, uaddr, size, flags);
3566 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3567 		}
3568 		break;
3569 	}
3570 
3571 	case DIF_SUBR_COPYOUTSTR: {
3572 		uintptr_t kaddr = tupregs[0].dttk_value;
3573 		uintptr_t uaddr = tupregs[1].dttk_value;
3574 		uint64_t size = tupregs[2].dttk_value;
3575 
3576 		if (!dtrace_destructive_disallow &&
3577 		    dtrace_priv_proc_control(state) &&
3578 		    !dtrace_istoxic(kaddr, size)) {
3579 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3580 			dtrace_copyoutstr(kaddr, uaddr, size, flags);
3581 			DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3582 		}
3583 		break;
3584 	}
3585 
3586 	case DIF_SUBR_STRLEN: {
3587 		size_t sz;
3588 		uintptr_t addr = (uintptr_t)tupregs[0].dttk_value;
3589 		sz = dtrace_strlen((char *)addr,
3590 		    state->dts_options[DTRACEOPT_STRSIZE]);
3591 
3592 		if (!dtrace_canload(addr, sz + 1, mstate, vstate)) {
3593 			regs[rd] = 0;
3594 			break;
3595 		}
3596 
3597 		regs[rd] = sz;
3598 
3599 		break;
3600 	}
3601 
3602 	case DIF_SUBR_STRCHR:
3603 	case DIF_SUBR_STRRCHR: {
3604 		/*
3605 		 * We're going to iterate over the string looking for the
3606 		 * specified character.  We will iterate until we have reached
3607 		 * the string length or we have found the character.  If this
3608 		 * is DIF_SUBR_STRRCHR, we will look for the last occurrence
3609 		 * of the specified character instead of the first.
3610 		 */
3611 		uintptr_t saddr = tupregs[0].dttk_value;
3612 		uintptr_t addr = tupregs[0].dttk_value;
3613 		uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE];
3614 		char c, target = (char)tupregs[1].dttk_value;
3615 
3616 		for (regs[rd] = 0; addr < limit; addr++) {
3617 			if ((c = dtrace_load8(addr)) == target) {
3618 				regs[rd] = addr;
3619 
3620 				if (subr == DIF_SUBR_STRCHR)
3621 					break;
3622 			}
3623 
3624 			if (c == '\0')
3625 				break;
3626 		}
3627 
3628 		if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) {
3629 			regs[rd] = 0;
3630 			break;
3631 		}
3632 
3633 		break;
3634 	}
3635 
3636 	case DIF_SUBR_STRSTR:
3637 	case DIF_SUBR_INDEX:
3638 	case DIF_SUBR_RINDEX: {
3639 		/*
3640 		 * We're going to iterate over the string looking for the
3641 		 * specified string.  We will iterate until we have reached
3642 		 * the string length or we have found the string.  (Yes, this
3643 		 * is done in the most naive way possible -- but considering
3644 		 * that the string we're searching for is likely to be
3645 		 * relatively short, the complexity of Rabin-Karp or similar
3646 		 * hardly seems merited.)
3647 		 */
3648 		char *addr = (char *)(uintptr_t)tupregs[0].dttk_value;
3649 		char *substr = (char *)(uintptr_t)tupregs[1].dttk_value;
3650 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3651 		size_t len = dtrace_strlen(addr, size);
3652 		size_t sublen = dtrace_strlen(substr, size);
3653 		char *limit = addr + len, *orig = addr;
3654 		int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1;
3655 		int inc = 1;
3656 
3657 		regs[rd] = notfound;
3658 
3659 		if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) {
3660 			regs[rd] = 0;
3661 			break;
3662 		}
3663 
3664 		if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate,
3665 		    vstate)) {
3666 			regs[rd] = 0;
3667 			break;
3668 		}
3669 
3670 		/*
3671 		 * strstr() and index()/rindex() have similar semantics if
3672 		 * both strings are the empty string: strstr() returns a
3673 		 * pointer to the (empty) string, and index() and rindex()
3674 		 * both return index 0 (regardless of any position argument).
3675 		 */
3676 		if (sublen == 0 && len == 0) {
3677 			if (subr == DIF_SUBR_STRSTR)
3678 				regs[rd] = (uintptr_t)addr;
3679 			else
3680 				regs[rd] = 0;
3681 			break;
3682 		}
3683 
3684 		if (subr != DIF_SUBR_STRSTR) {
3685 			if (subr == DIF_SUBR_RINDEX) {
3686 				limit = orig - 1;
3687 				addr += len;
3688 				inc = -1;
3689 			}
3690 
3691 			/*
3692 			 * Both index() and rindex() take an optional position
3693 			 * argument that denotes the starting position.
3694 			 */
3695 			if (nargs == 3) {
3696 				int64_t pos = (int64_t)tupregs[2].dttk_value;
3697 
3698 				/*
3699 				 * If the position argument to index() is
3700 				 * negative, Perl implicitly clamps it at
3701 				 * zero.  This semantic is a little surprising
3702 				 * given the special meaning of negative
3703 				 * positions to similar Perl functions like
3704 				 * substr(), but it appears to reflect a
3705 				 * notion that index() can start from a
3706 				 * negative index and increment its way up to
3707 				 * the string.  Given this notion, Perl's
3708 				 * rindex() is at least self-consistent in
3709 				 * that it implicitly clamps positions greater
3710 				 * than the string length to be the string
3711 				 * length.  Where Perl completely loses
3712 				 * coherence, however, is when the specified
3713 				 * substring is the empty string ("").  In
3714 				 * this case, even if the position is
3715 				 * negative, rindex() returns 0 -- and even if
3716 				 * the position is greater than the length,
3717 				 * index() returns the string length.  These
3718 				 * semantics violate the notion that index()
3719 				 * should never return a value less than the
3720 				 * specified position and that rindex() should
3721 				 * never return a value greater than the
3722 				 * specified position.  (One assumes that
3723 				 * these semantics are artifacts of Perl's
3724 				 * implementation and not the results of
3725 				 * deliberate design -- it beggars belief that
3726 				 * even Larry Wall could desire such oddness.)
3727 				 * While in the abstract one would wish for
3728 				 * consistent position semantics across
3729 				 * substr(), index() and rindex() -- or at the
3730 				 * very least self-consistent position
3731 				 * semantics for index() and rindex() -- we
3732 				 * instead opt to keep with the extant Perl
3733 				 * semantics, in all their broken glory.  (Do
3734 				 * we have more desire to maintain Perl's
3735 				 * semantics than Perl does?  Probably.)
3736 				 */
3737 				if (subr == DIF_SUBR_RINDEX) {
3738 					if (pos < 0) {
3739 						if (sublen == 0)
3740 							regs[rd] = 0;
3741 						break;
3742 					}
3743 
3744 					if (pos > len)
3745 						pos = len;
3746 				} else {
3747 					if (pos < 0)
3748 						pos = 0;
3749 
3750 					if (pos >= len) {
3751 						if (sublen == 0)
3752 							regs[rd] = len;
3753 						break;
3754 					}
3755 				}
3756 
3757 				addr = orig + pos;
3758 			}
3759 		}
3760 
3761 		for (regs[rd] = notfound; addr != limit; addr += inc) {
3762 			if (dtrace_strncmp(addr, substr, sublen) == 0) {
3763 				if (subr != DIF_SUBR_STRSTR) {
3764 					/*
3765 					 * As D index() and rindex() are
3766 					 * modeled on Perl (and not on awk),
3767 					 * we return a zero-based (and not a
3768 					 * one-based) index.  (For you Perl
3769 					 * weenies: no, we're not going to add
3770 					 * $[ -- and shouldn't you be at a con
3771 					 * or something?)
3772 					 */
3773 					regs[rd] = (uintptr_t)(addr - orig);
3774 					break;
3775 				}
3776 
3777 				ASSERT(subr == DIF_SUBR_STRSTR);
3778 				regs[rd] = (uintptr_t)addr;
3779 				break;
3780 			}
3781 		}
3782 
3783 		break;
3784 	}
3785 
3786 	case DIF_SUBR_STRTOK: {
3787 		uintptr_t addr = tupregs[0].dttk_value;
3788 		uintptr_t tokaddr = tupregs[1].dttk_value;
3789 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3790 		uintptr_t limit, toklimit = tokaddr + size;
3791 		uint8_t c = 0, tokmap[32];	 /* 256 / 8 */
3792 		char *dest = (char *)mstate->dtms_scratch_ptr;
3793 		int i;
3794 
3795 		/*
3796 		 * Check both the token buffer and (later) the input buffer,
3797 		 * since both could be non-scratch addresses.
3798 		 */
3799 		if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) {
3800 			regs[rd] = 0;
3801 			break;
3802 		}
3803 
3804 		if (!DTRACE_INSCRATCH(mstate, size)) {
3805 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3806 			regs[rd] = 0;
3807 			break;
3808 		}
3809 
3810 		if (addr == 0) {
3811 			/*
3812 			 * If the address specified is NULL, we use our saved
3813 			 * strtok pointer from the mstate.  Note that this
3814 			 * means that the saved strtok pointer is _only_
3815 			 * valid within multiple enablings of the same probe --
3816 			 * it behaves like an implicit clause-local variable.
3817 			 */
3818 			addr = mstate->dtms_strtok;
3819 		} else {
3820 			/*
3821 			 * If the user-specified address is non-NULL we must
3822 			 * access check it.  This is the only time we have
3823 			 * a chance to do so, since this address may reside
3824 			 * in the string table of this clause-- future calls
3825 			 * (when we fetch addr from mstate->dtms_strtok)
3826 			 * would fail this access check.
3827 			 */
3828 			if (!dtrace_strcanload(addr, size, mstate, vstate)) {
3829 				regs[rd] = 0;
3830 				break;
3831 			}
3832 		}
3833 
3834 		/*
3835 		 * First, zero the token map, and then process the token
3836 		 * string -- setting a bit in the map for every character
3837 		 * found in the token string.
3838 		 */
3839 		for (i = 0; i < sizeof (tokmap); i++)
3840 			tokmap[i] = 0;
3841 
3842 		for (; tokaddr < toklimit; tokaddr++) {
3843 			if ((c = dtrace_load8(tokaddr)) == '\0')
3844 				break;
3845 
3846 			ASSERT((c >> 3) < sizeof (tokmap));
3847 			tokmap[c >> 3] |= (1 << (c & 0x7));
3848 		}
3849 
3850 		for (limit = addr + size; addr < limit; addr++) {
3851 			/*
3852 			 * We're looking for a character that is _not_ contained
3853 			 * in the token string.
3854 			 */
3855 			if ((c = dtrace_load8(addr)) == '\0')
3856 				break;
3857 
3858 			if (!(tokmap[c >> 3] & (1 << (c & 0x7))))
3859 				break;
3860 		}
3861 
3862 		if (c == '\0') {
3863 			/*
3864 			 * We reached the end of the string without finding
3865 			 * any character that was not in the token string.
3866 			 * We return NULL in this case, and we set the saved
3867 			 * address to NULL as well.
3868 			 */
3869 			regs[rd] = 0;
3870 			mstate->dtms_strtok = 0;
3871 			break;
3872 		}
3873 
3874 		/*
3875 		 * From here on, we're copying into the destination string.
3876 		 */
3877 		for (i = 0; addr < limit && i < size - 1; addr++) {
3878 			if ((c = dtrace_load8(addr)) == '\0')
3879 				break;
3880 
3881 			if (tokmap[c >> 3] & (1 << (c & 0x7)))
3882 				break;
3883 
3884 			ASSERT(i < size);
3885 			dest[i++] = c;
3886 		}
3887 
3888 		ASSERT(i < size);
3889 		dest[i] = '\0';
3890 		regs[rd] = (uintptr_t)dest;
3891 		mstate->dtms_scratch_ptr += size;
3892 		mstate->dtms_strtok = addr;
3893 		break;
3894 	}
3895 
3896 	case DIF_SUBR_SUBSTR: {
3897 		uintptr_t s = tupregs[0].dttk_value;
3898 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3899 		char *d = (char *)mstate->dtms_scratch_ptr;
3900 		int64_t index = (int64_t)tupregs[1].dttk_value;
3901 		int64_t remaining = (int64_t)tupregs[2].dttk_value;
3902 		size_t len = dtrace_strlen((char *)s, size);
3903 		int64_t i = 0;
3904 
3905 		if (!dtrace_canload(s, len + 1, mstate, vstate)) {
3906 			regs[rd] = 0;
3907 			break;
3908 		}
3909 
3910 		if (!DTRACE_INSCRATCH(mstate, size)) {
3911 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3912 			regs[rd] = 0;
3913 			break;
3914 		}
3915 
3916 		if (nargs <= 2)
3917 			remaining = (int64_t)size;
3918 
3919 		if (index < 0) {
3920 			index += len;
3921 
3922 			if (index < 0 && index + remaining > 0) {
3923 				remaining += index;
3924 				index = 0;
3925 			}
3926 		}
3927 
3928 		if (index >= len || index < 0) {
3929 			remaining = 0;
3930 		} else if (remaining < 0) {
3931 			remaining += len - index;
3932 		} else if (index + remaining > size) {
3933 			remaining = size - index;
3934 		}
3935 
3936 		for (i = 0; i < remaining; i++) {
3937 			if ((d[i] = dtrace_load8(s + index + i)) == '\0')
3938 				break;
3939 		}
3940 
3941 		d[i] = '\0';
3942 
3943 		mstate->dtms_scratch_ptr += size;
3944 		regs[rd] = (uintptr_t)d;
3945 		break;
3946 	}
3947 
3948 #if defined(sun)
3949 	case DIF_SUBR_GETMAJOR:
3950 #ifdef _LP64
3951 		regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64;
3952 #else
3953 		regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ;
3954 #endif
3955 		break;
3956 
3957 	case DIF_SUBR_GETMINOR:
3958 #ifdef _LP64
3959 		regs[rd] = tupregs[0].dttk_value & MAXMIN64;
3960 #else
3961 		regs[rd] = tupregs[0].dttk_value & MAXMIN;
3962 #endif
3963 		break;
3964 
3965 	case DIF_SUBR_DDI_PATHNAME: {
3966 		/*
3967 		 * This one is a galactic mess.  We are going to roughly
3968 		 * emulate ddi_pathname(), but it's made more complicated
3969 		 * by the fact that we (a) want to include the minor name and
3970 		 * (b) must proceed iteratively instead of recursively.
3971 		 */
3972 		uintptr_t dest = mstate->dtms_scratch_ptr;
3973 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3974 		char *start = (char *)dest, *end = start + size - 1;
3975 		uintptr_t daddr = tupregs[0].dttk_value;
3976 		int64_t minor = (int64_t)tupregs[1].dttk_value;
3977 		char *s;
3978 		int i, len, depth = 0;
3979 
3980 		/*
3981 		 * Due to all the pointer jumping we do and context we must
3982 		 * rely upon, we just mandate that the user must have kernel
3983 		 * read privileges to use this routine.
3984 		 */
3985 		if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) {
3986 			*flags |= CPU_DTRACE_KPRIV;
3987 			*illval = daddr;
3988 			regs[rd] = 0;
3989 		}
3990 
3991 		if (!DTRACE_INSCRATCH(mstate, size)) {
3992 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3993 			regs[rd] = 0;
3994 			break;
3995 		}
3996 
3997 		*end = '\0';
3998 
3999 		/*
4000 		 * We want to have a name for the minor.  In order to do this,
4001 		 * we need to walk the minor list from the devinfo.  We want
4002 		 * to be sure that we don't infinitely walk a circular list,
4003 		 * so we check for circularity by sending a scout pointer
4004 		 * ahead two elements for every element that we iterate over;
4005 		 * if the list is circular, these will ultimately point to the
4006 		 * same element.  You may recognize this little trick as the
4007 		 * answer to a stupid interview question -- one that always
4008 		 * seems to be asked by those who had to have it laboriously
4009 		 * explained to them, and who can't even concisely describe
4010 		 * the conditions under which one would be forced to resort to
4011 		 * this technique.  Needless to say, those conditions are
4012 		 * found here -- and probably only here.  Is this the only use
4013 		 * of this infamous trick in shipping, production code?  If it
4014 		 * isn't, it probably should be...
4015 		 */
4016 		if (minor != -1) {
4017 			uintptr_t maddr = dtrace_loadptr(daddr +
4018 			    offsetof(struct dev_info, devi_minor));
4019 
4020 			uintptr_t next = offsetof(struct ddi_minor_data, next);
4021 			uintptr_t name = offsetof(struct ddi_minor_data,
4022 			    d_minor) + offsetof(struct ddi_minor, name);
4023 			uintptr_t dev = offsetof(struct ddi_minor_data,
4024 			    d_minor) + offsetof(struct ddi_minor, dev);
4025 			uintptr_t scout;
4026 
4027 			if (maddr != NULL)
4028 				scout = dtrace_loadptr(maddr + next);
4029 
4030 			while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4031 				uint64_t m;
4032 #ifdef _LP64
4033 				m = dtrace_load64(maddr + dev) & MAXMIN64;
4034 #else
4035 				m = dtrace_load32(maddr + dev) & MAXMIN;
4036 #endif
4037 				if (m != minor) {
4038 					maddr = dtrace_loadptr(maddr + next);
4039 
4040 					if (scout == NULL)
4041 						continue;
4042 
4043 					scout = dtrace_loadptr(scout + next);
4044 
4045 					if (scout == NULL)
4046 						continue;
4047 
4048 					scout = dtrace_loadptr(scout + next);
4049 
4050 					if (scout == NULL)
4051 						continue;
4052 
4053 					if (scout == maddr) {
4054 						*flags |= CPU_DTRACE_ILLOP;
4055 						break;
4056 					}
4057 
4058 					continue;
4059 				}
4060 
4061 				/*
4062 				 * We have the minor data.  Now we need to
4063 				 * copy the minor's name into the end of the
4064 				 * pathname.
4065 				 */
4066 				s = (char *)dtrace_loadptr(maddr + name);
4067 				len = dtrace_strlen(s, size);
4068 
4069 				if (*flags & CPU_DTRACE_FAULT)
4070 					break;
4071 
4072 				if (len != 0) {
4073 					if ((end -= (len + 1)) < start)
4074 						break;
4075 
4076 					*end = ':';
4077 				}
4078 
4079 				for (i = 1; i <= len; i++)
4080 					end[i] = dtrace_load8((uintptr_t)s++);
4081 				break;
4082 			}
4083 		}
4084 
4085 		while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4086 			ddi_node_state_t devi_state;
4087 
4088 			devi_state = dtrace_load32(daddr +
4089 			    offsetof(struct dev_info, devi_node_state));
4090 
4091 			if (*flags & CPU_DTRACE_FAULT)
4092 				break;
4093 
4094 			if (devi_state >= DS_INITIALIZED) {
4095 				s = (char *)dtrace_loadptr(daddr +
4096 				    offsetof(struct dev_info, devi_addr));
4097 				len = dtrace_strlen(s, size);
4098 
4099 				if (*flags & CPU_DTRACE_FAULT)
4100 					break;
4101 
4102 				if (len != 0) {
4103 					if ((end -= (len + 1)) < start)
4104 						break;
4105 
4106 					*end = '@';
4107 				}
4108 
4109 				for (i = 1; i <= len; i++)
4110 					end[i] = dtrace_load8((uintptr_t)s++);
4111 			}
4112 
4113 			/*
4114 			 * Now for the node name...
4115 			 */
4116 			s = (char *)dtrace_loadptr(daddr +
4117 			    offsetof(struct dev_info, devi_node_name));
4118 
4119 			daddr = dtrace_loadptr(daddr +
4120 			    offsetof(struct dev_info, devi_parent));
4121 
4122 			/*
4123 			 * If our parent is NULL (that is, if we're the root
4124 			 * node), we're going to use the special path
4125 			 * "devices".
4126 			 */
4127 			if (daddr == 0)
4128 				s = "devices";
4129 
4130 			len = dtrace_strlen(s, size);
4131 			if (*flags & CPU_DTRACE_FAULT)
4132 				break;
4133 
4134 			if ((end -= (len + 1)) < start)
4135 				break;
4136 
4137 			for (i = 1; i <= len; i++)
4138 				end[i] = dtrace_load8((uintptr_t)s++);
4139 			*end = '/';
4140 
4141 			if (depth++ > dtrace_devdepth_max) {
4142 				*flags |= CPU_DTRACE_ILLOP;
4143 				break;
4144 			}
4145 		}
4146 
4147 		if (end < start)
4148 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4149 
4150 		if (daddr == 0) {
4151 			regs[rd] = (uintptr_t)end;
4152 			mstate->dtms_scratch_ptr += size;
4153 		}
4154 
4155 		break;
4156 	}
4157 #endif
4158 
4159 	case DIF_SUBR_STRJOIN: {
4160 		char *d = (char *)mstate->dtms_scratch_ptr;
4161 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4162 		uintptr_t s1 = tupregs[0].dttk_value;
4163 		uintptr_t s2 = tupregs[1].dttk_value;
4164 		int i = 0;
4165 
4166 		if (!dtrace_strcanload(s1, size, mstate, vstate) ||
4167 		    !dtrace_strcanload(s2, size, mstate, vstate)) {
4168 			regs[rd] = 0;
4169 			break;
4170 		}
4171 
4172 		if (!DTRACE_INSCRATCH(mstate, size)) {
4173 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4174 			regs[rd] = 0;
4175 			break;
4176 		}
4177 
4178 		for (;;) {
4179 			if (i >= size) {
4180 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4181 				regs[rd] = 0;
4182 				break;
4183 			}
4184 
4185 			if ((d[i++] = dtrace_load8(s1++)) == '\0') {
4186 				i--;
4187 				break;
4188 			}
4189 		}
4190 
4191 		for (;;) {
4192 			if (i >= size) {
4193 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4194 				regs[rd] = 0;
4195 				break;
4196 			}
4197 
4198 			if ((d[i++] = dtrace_load8(s2++)) == '\0')
4199 				break;
4200 		}
4201 
4202 		if (i < size) {
4203 			mstate->dtms_scratch_ptr += i;
4204 			regs[rd] = (uintptr_t)d;
4205 		}
4206 
4207 		break;
4208 	}
4209 
4210 	case DIF_SUBR_LLTOSTR: {
4211 		int64_t i = (int64_t)tupregs[0].dttk_value;
4212 		int64_t val = i < 0 ? i * -1 : i;
4213 		uint64_t size = 22;	/* enough room for 2^64 in decimal */
4214 		char *end = (char *)mstate->dtms_scratch_ptr + size - 1;
4215 
4216 		if (!DTRACE_INSCRATCH(mstate, size)) {
4217 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4218 			regs[rd] = 0;
4219 			break;
4220 		}
4221 
4222 		for (*end-- = '\0'; val; val /= 10)
4223 			*end-- = '0' + (val % 10);
4224 
4225 		if (i == 0)
4226 			*end-- = '0';
4227 
4228 		if (i < 0)
4229 			*end-- = '-';
4230 
4231 		regs[rd] = (uintptr_t)end + 1;
4232 		mstate->dtms_scratch_ptr += size;
4233 		break;
4234 	}
4235 
4236 	case DIF_SUBR_HTONS:
4237 	case DIF_SUBR_NTOHS:
4238 #if BYTE_ORDER == BIG_ENDIAN
4239 		regs[rd] = (uint16_t)tupregs[0].dttk_value;
4240 #else
4241 		regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value);
4242 #endif
4243 		break;
4244 
4245 
4246 	case DIF_SUBR_HTONL:
4247 	case DIF_SUBR_NTOHL:
4248 #if BYTE_ORDER == BIG_ENDIAN
4249 		regs[rd] = (uint32_t)tupregs[0].dttk_value;
4250 #else
4251 		regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value);
4252 #endif
4253 		break;
4254 
4255 
4256 	case DIF_SUBR_HTONLL:
4257 	case DIF_SUBR_NTOHLL:
4258 #if BYTE_ORDER == BIG_ENDIAN
4259 		regs[rd] = (uint64_t)tupregs[0].dttk_value;
4260 #else
4261 		regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value);
4262 #endif
4263 		break;
4264 
4265 
4266 	case DIF_SUBR_DIRNAME:
4267 	case DIF_SUBR_BASENAME: {
4268 		char *dest = (char *)mstate->dtms_scratch_ptr;
4269 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4270 		uintptr_t src = tupregs[0].dttk_value;
4271 		int i, j, len = dtrace_strlen((char *)src, size);
4272 		int lastbase = -1, firstbase = -1, lastdir = -1;
4273 		int start, end;
4274 
4275 		if (!dtrace_canload(src, len + 1, mstate, vstate)) {
4276 			regs[rd] = 0;
4277 			break;
4278 		}
4279 
4280 		if (!DTRACE_INSCRATCH(mstate, size)) {
4281 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4282 			regs[rd] = 0;
4283 			break;
4284 		}
4285 
4286 		/*
4287 		 * The basename and dirname for a zero-length string is
4288 		 * defined to be "."
4289 		 */
4290 		if (len == 0) {
4291 			len = 1;
4292 			src = (uintptr_t)".";
4293 		}
4294 
4295 		/*
4296 		 * Start from the back of the string, moving back toward the
4297 		 * front until we see a character that isn't a slash.  That
4298 		 * character is the last character in the basename.
4299 		 */
4300 		for (i = len - 1; i >= 0; i--) {
4301 			if (dtrace_load8(src + i) != '/')
4302 				break;
4303 		}
4304 
4305 		if (i >= 0)
4306 			lastbase = i;
4307 
4308 		/*
4309 		 * Starting from the last character in the basename, move
4310 		 * towards the front until we find a slash.  The character
4311 		 * that we processed immediately before that is the first
4312 		 * character in the basename.
4313 		 */
4314 		for (; i >= 0; i--) {
4315 			if (dtrace_load8(src + i) == '/')
4316 				break;
4317 		}
4318 
4319 		if (i >= 0)
4320 			firstbase = i + 1;
4321 
4322 		/*
4323 		 * Now keep going until we find a non-slash character.  That
4324 		 * character is the last character in the dirname.
4325 		 */
4326 		for (; i >= 0; i--) {
4327 			if (dtrace_load8(src + i) != '/')
4328 				break;
4329 		}
4330 
4331 		if (i >= 0)
4332 			lastdir = i;
4333 
4334 		ASSERT(!(lastbase == -1 && firstbase != -1));
4335 		ASSERT(!(firstbase == -1 && lastdir != -1));
4336 
4337 		if (lastbase == -1) {
4338 			/*
4339 			 * We didn't find a non-slash character.  We know that
4340 			 * the length is non-zero, so the whole string must be
4341 			 * slashes.  In either the dirname or the basename
4342 			 * case, we return '/'.
4343 			 */
4344 			ASSERT(firstbase == -1);
4345 			firstbase = lastbase = lastdir = 0;
4346 		}
4347 
4348 		if (firstbase == -1) {
4349 			/*
4350 			 * The entire string consists only of a basename
4351 			 * component.  If we're looking for dirname, we need
4352 			 * to change our string to be just "."; if we're
4353 			 * looking for a basename, we'll just set the first
4354 			 * character of the basename to be 0.
4355 			 */
4356 			if (subr == DIF_SUBR_DIRNAME) {
4357 				ASSERT(lastdir == -1);
4358 				src = (uintptr_t)".";
4359 				lastdir = 0;
4360 			} else {
4361 				firstbase = 0;
4362 			}
4363 		}
4364 
4365 		if (subr == DIF_SUBR_DIRNAME) {
4366 			if (lastdir == -1) {
4367 				/*
4368 				 * We know that we have a slash in the name --
4369 				 * or lastdir would be set to 0, above.  And
4370 				 * because lastdir is -1, we know that this
4371 				 * slash must be the first character.  (That
4372 				 * is, the full string must be of the form
4373 				 * "/basename".)  In this case, the last
4374 				 * character of the directory name is 0.
4375 				 */
4376 				lastdir = 0;
4377 			}
4378 
4379 			start = 0;
4380 			end = lastdir;
4381 		} else {
4382 			ASSERT(subr == DIF_SUBR_BASENAME);
4383 			ASSERT(firstbase != -1 && lastbase != -1);
4384 			start = firstbase;
4385 			end = lastbase;
4386 		}
4387 
4388 		for (i = start, j = 0; i <= end && j < size - 1; i++, j++)
4389 			dest[j] = dtrace_load8(src + i);
4390 
4391 		dest[j] = '\0';
4392 		regs[rd] = (uintptr_t)dest;
4393 		mstate->dtms_scratch_ptr += size;
4394 		break;
4395 	}
4396 
4397 	case DIF_SUBR_CLEANPATH: {
4398 		char *dest = (char *)mstate->dtms_scratch_ptr, c;
4399 		uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4400 		uintptr_t src = tupregs[0].dttk_value;
4401 		int i = 0, j = 0;
4402 
4403 		if (!dtrace_strcanload(src, size, mstate, vstate)) {
4404 			regs[rd] = 0;
4405 			break;
4406 		}
4407 
4408 		if (!DTRACE_INSCRATCH(mstate, size)) {
4409 			DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4410 			regs[rd] = 0;
4411 			break;
4412 		}
4413 
4414 		/*
4415 		 * Move forward, loading each character.
4416 		 */
4417 		do {
4418 			c = dtrace_load8(src + i++);
4419 next:
4420 			if (j + 5 >= size)	/* 5 = strlen("/..c\0") */
4421 				break;
4422 
4423 			if (c != '/') {
4424 				dest[j++] = c;
4425 				continue;
4426 			}
4427 
4428 			c = dtrace_load8(src + i++);
4429 
4430 			if (c == '/') {
4431 				/*
4432 				 * We have two slashes -- we can just advance
4433 				 * to the next character.
4434 				 */
4435 				goto next;
4436 			}
4437 
4438 			if (c != '.') {
4439 				/*
4440 				 * This is not "." and it's not ".." -- we can
4441 				 * just store the "/" and this character and
4442 				 * drive on.
4443 				 */
4444 				dest[j++] = '/';
4445 				dest[j++] = c;
4446 				continue;
4447 			}
4448 
4449 			c = dtrace_load8(src + i++);
4450 
4451 			if (c == '/') {
4452 				/*
4453 				 * This is a "/./" component.  We're not going
4454 				 * to store anything in the destination buffer;
4455 				 * we're just going to go to the next component.
4456 				 */
4457 				goto next;
4458 			}
4459 
4460 			if (c != '.') {
4461 				/*
4462 				 * This is not ".." -- we can just store the
4463 				 * "/." and this character and continue
4464 				 * processing.
4465 				 */
4466 				dest[j++] = '/';
4467 				dest[j++] = '.';
4468 				dest[j++] = c;
4469 				continue;
4470 			}
4471 
4472 			c = dtrace_load8(src + i++);
4473 
4474 			if (c != '/' && c != '\0') {
4475 				/*
4476 				 * This is not ".." -- it's "..[mumble]".
4477 				 * We'll store the "/.." and this character
4478 				 * and continue processing.
4479 				 */
4480 				dest[j++] = '/';
4481 				dest[j++] = '.';
4482 				dest[j++] = '.';
4483 				dest[j++] = c;
4484 				continue;
4485 			}
4486 
4487 			/*
4488 			 * This is "/../" or "/..\0".  We need to back up
4489 			 * our destination pointer until we find a "/".
4490 			 */
4491 			i--;
4492 			while (j != 0 && dest[--j] != '/')
4493 				continue;
4494 
4495 			if (c == '\0')
4496 				dest[++j] = '/';
4497 		} while (c != '\0');
4498 
4499 		dest[j] = '\0';
4500 		regs[rd] = (uintptr_t)dest;
4501 		mstate->dtms_scratch_ptr += size;
4502 		break;
4503 	}
4504 
4505 	case DIF_SUBR_INET_NTOA:
4506 	case DIF_SUBR_INET_NTOA6:
4507 	case DIF_SUBR_INET_NTOP: {
4508 		size_t size;
4509 		int af, argi, i;
4510 		char *base, *end;
4511 
4512 		if (subr == DIF_SUBR_INET_NTOP) {
4513 			af = (int)tupregs[0].dttk_value;
4514 			argi = 1;
4515 		} else {
4516 			af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6;
4517 			argi = 0;
4518 		}
4519 
4520 		if (af == AF_INET) {
4521 			ipaddr_t ip4;
4522 			uint8_t *ptr8, val;
4523 
4524 			/*
4525 			 * Safely load the IPv4 address.
4526 			 */
4527 			ip4 = dtrace_load32(tupregs[argi].dttk_value);
4528 
4529 			/*
4530 			 * Check an IPv4 string will fit in scratch.
4531 			 */
4532 			size = INET_ADDRSTRLEN;
4533 			if (!DTRACE_INSCRATCH(mstate, size)) {
4534 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4535 				regs[rd] = 0;
4536 				break;
4537 			}
4538 			base = (char *)mstate->dtms_scratch_ptr;
4539 			end = (char *)mstate->dtms_scratch_ptr + size - 1;
4540 
4541 			/*
4542 			 * Stringify as a dotted decimal quad.
4543 			 */
4544 			*end-- = '\0';
4545 			ptr8 = (uint8_t *)&ip4;
4546 			for (i = 3; i >= 0; i--) {
4547 				val = ptr8[i];
4548 
4549 				if (val == 0) {
4550 					*end-- = '0';
4551 				} else {
4552 					for (; val; val /= 10) {
4553 						*end-- = '0' + (val % 10);
4554 					}
4555 				}
4556 
4557 				if (i > 0)
4558 					*end-- = '.';
4559 			}
4560 			ASSERT(end + 1 >= base);
4561 
4562 		} else if (af == AF_INET6) {
4563 			struct in6_addr ip6;
4564 			int firstzero, tryzero, numzero, v6end;
4565 			uint16_t val;
4566 			const char digits[] = "0123456789abcdef";
4567 
4568 			/*
4569 			 * Stringify using RFC 1884 convention 2 - 16 bit
4570 			 * hexadecimal values with a zero-run compression.
4571 			 * Lower case hexadecimal digits are used.
4572 			 * 	eg, fe80::214:4fff:fe0b:76c8.
4573 			 * The IPv4 embedded form is returned for inet_ntop,
4574 			 * just the IPv4 string is returned for inet_ntoa6.
4575 			 */
4576 
4577 			/*
4578 			 * Safely load the IPv6 address.
4579 			 */
4580 			dtrace_bcopy(
4581 			    (void *)(uintptr_t)tupregs[argi].dttk_value,
4582 			    (void *)(uintptr_t)&ip6, sizeof (struct in6_addr));
4583 
4584 			/*
4585 			 * Check an IPv6 string will fit in scratch.
4586 			 */
4587 			size = INET6_ADDRSTRLEN;
4588 			if (!DTRACE_INSCRATCH(mstate, size)) {
4589 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4590 				regs[rd] = 0;
4591 				break;
4592 			}
4593 			base = (char *)mstate->dtms_scratch_ptr;
4594 			end = (char *)mstate->dtms_scratch_ptr + size - 1;
4595 			*end-- = '\0';
4596 
4597 			/*
4598 			 * Find the longest run of 16 bit zero values
4599 			 * for the single allowed zero compression - "::".
4600 			 */
4601 			firstzero = -1;
4602 			tryzero = -1;
4603 			numzero = 1;
4604 			for (i = 0; i < sizeof (struct in6_addr); i++) {
4605 #if defined(sun)
4606 				if (ip6._S6_un._S6_u8[i] == 0 &&
4607 #else
4608 				if (ip6.__u6_addr.__u6_addr8[i] == 0 &&
4609 #endif
4610 				    tryzero == -1 && i % 2 == 0) {
4611 					tryzero = i;
4612 					continue;
4613 				}
4614 
4615 				if (tryzero != -1 &&
4616 #if defined(sun)
4617 				    (ip6._S6_un._S6_u8[i] != 0 ||
4618 #else
4619 				    (ip6.__u6_addr.__u6_addr8[i] != 0 ||
4620 #endif
4621 				    i == sizeof (struct in6_addr) - 1)) {
4622 
4623 					if (i - tryzero <= numzero) {
4624 						tryzero = -1;
4625 						continue;
4626 					}
4627 
4628 					firstzero = tryzero;
4629 					numzero = i - i % 2 - tryzero;
4630 					tryzero = -1;
4631 
4632 #if defined(sun)
4633 					if (ip6._S6_un._S6_u8[i] == 0 &&
4634 #else
4635 					if (ip6.__u6_addr.__u6_addr8[i] == 0 &&
4636 #endif
4637 					    i == sizeof (struct in6_addr) - 1)
4638 						numzero += 2;
4639 				}
4640 			}
4641 			ASSERT(firstzero + numzero <= sizeof (struct in6_addr));
4642 
4643 			/*
4644 			 * Check for an IPv4 embedded address.
4645 			 */
4646 			v6end = sizeof (struct in6_addr) - 2;
4647 			if (IN6_IS_ADDR_V4MAPPED(&ip6) ||
4648 			    IN6_IS_ADDR_V4COMPAT(&ip6)) {
4649 				for (i = sizeof (struct in6_addr) - 1;
4650 				    i >= DTRACE_V4MAPPED_OFFSET; i--) {
4651 					ASSERT(end >= base);
4652 
4653 #if defined(sun)
4654 					val = ip6._S6_un._S6_u8[i];
4655 #else
4656 					val = ip6.__u6_addr.__u6_addr8[i];
4657 #endif
4658 
4659 					if (val == 0) {
4660 						*end-- = '0';
4661 					} else {
4662 						for (; val; val /= 10) {
4663 							*end-- = '0' + val % 10;
4664 						}
4665 					}
4666 
4667 					if (i > DTRACE_V4MAPPED_OFFSET)
4668 						*end-- = '.';
4669 				}
4670 
4671 				if (subr == DIF_SUBR_INET_NTOA6)
4672 					goto inetout;
4673 
4674 				/*
4675 				 * Set v6end to skip the IPv4 address that
4676 				 * we have already stringified.
4677 				 */
4678 				v6end = 10;
4679 			}
4680 
4681 			/*
4682 			 * Build the IPv6 string by working through the
4683 			 * address in reverse.
4684 			 */
4685 			for (i = v6end; i >= 0; i -= 2) {
4686 				ASSERT(end >= base);
4687 
4688 				if (i == firstzero + numzero - 2) {
4689 					*end-- = ':';
4690 					*end-- = ':';
4691 					i -= numzero - 2;
4692 					continue;
4693 				}
4694 
4695 				if (i < 14 && i != firstzero - 2)
4696 					*end-- = ':';
4697 
4698 #if defined(sun)
4699 				val = (ip6._S6_un._S6_u8[i] << 8) +
4700 				    ip6._S6_un._S6_u8[i + 1];
4701 #else
4702 				val = (ip6.__u6_addr.__u6_addr8[i] << 8) +
4703 				    ip6.__u6_addr.__u6_addr8[i + 1];
4704 #endif
4705 
4706 				if (val == 0) {
4707 					*end-- = '0';
4708 				} else {
4709 					for (; val; val /= 16) {
4710 						*end-- = digits[val % 16];
4711 					}
4712 				}
4713 			}
4714 			ASSERT(end + 1 >= base);
4715 
4716 		} else {
4717 			/*
4718 			 * The user didn't use AH_INET or AH_INET6.
4719 			 */
4720 			DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
4721 			regs[rd] = 0;
4722 			break;
4723 		}
4724 
4725 inetout:	regs[rd] = (uintptr_t)end + 1;
4726 		mstate->dtms_scratch_ptr += size;
4727 		break;
4728 	}
4729 
4730 	case DIF_SUBR_MEMREF: {
4731 		uintptr_t size = 2 * sizeof(uintptr_t);
4732 		uintptr_t *memref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t));
4733 		size_t scratch_size = ((uintptr_t) memref - mstate->dtms_scratch_ptr) + size;
4734 
4735 		/* address and length */
4736 		memref[0] = tupregs[0].dttk_value;
4737 		memref[1] = tupregs[1].dttk_value;
4738 
4739 		regs[rd] = (uintptr_t) memref;
4740 		mstate->dtms_scratch_ptr += scratch_size;
4741 		break;
4742 	}
4743 
4744 	case DIF_SUBR_TYPEREF: {
4745 		uintptr_t size = 4 * sizeof(uintptr_t);
4746 		uintptr_t *typeref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t));
4747 		size_t scratch_size = ((uintptr_t) typeref - mstate->dtms_scratch_ptr) + size;
4748 
4749 		/* address, num_elements, type_str, type_len */
4750 		typeref[0] = tupregs[0].dttk_value;
4751 		typeref[1] = tupregs[1].dttk_value;
4752 		typeref[2] = tupregs[2].dttk_value;
4753 		typeref[3] = tupregs[3].dttk_value;
4754 
4755 		regs[rd] = (uintptr_t) typeref;
4756 		mstate->dtms_scratch_ptr += scratch_size;
4757 		break;
4758 	}
4759 	}
4760 }
4761 
4762 /*
4763  * Emulate the execution of DTrace IR instructions specified by the given
4764  * DIF object.  This function is deliberately void of assertions as all of
4765  * the necessary checks are handled by a call to dtrace_difo_validate().
4766  */
4767 static uint64_t
4768 dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate,
4769     dtrace_vstate_t *vstate, dtrace_state_t *state)
4770 {
4771 	const dif_instr_t *text = difo->dtdo_buf;
4772 	const uint_t textlen = difo->dtdo_len;
4773 	const char *strtab = difo->dtdo_strtab;
4774 	const uint64_t *inttab = difo->dtdo_inttab;
4775 
4776 	uint64_t rval = 0;
4777 	dtrace_statvar_t *svar;
4778 	dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
4779 	dtrace_difv_t *v;
4780 	volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
4781 	volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
4782 
4783 	dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
4784 	uint64_t regs[DIF_DIR_NREGS];
4785 	uint64_t *tmp;
4786 
4787 	uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0;
4788 	int64_t cc_r;
4789 	uint_t pc = 0, id, opc = 0;
4790 	uint8_t ttop = 0;
4791 	dif_instr_t instr;
4792 	uint_t r1, r2, rd;
4793 
4794 	/*
4795 	 * We stash the current DIF object into the machine state: we need it
4796 	 * for subsequent access checking.
4797 	 */
4798 	mstate->dtms_difo = difo;
4799 
4800 	regs[DIF_REG_R0] = 0; 		/* %r0 is fixed at zero */
4801 
4802 	while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) {
4803 		opc = pc;
4804 
4805 		instr = text[pc++];
4806 		r1 = DIF_INSTR_R1(instr);
4807 		r2 = DIF_INSTR_R2(instr);
4808 		rd = DIF_INSTR_RD(instr);
4809 
4810 		switch (DIF_INSTR_OP(instr)) {
4811 		case DIF_OP_OR:
4812 			regs[rd] = regs[r1] | regs[r2];
4813 			break;
4814 		case DIF_OP_XOR:
4815 			regs[rd] = regs[r1] ^ regs[r2];
4816 			break;
4817 		case DIF_OP_AND:
4818 			regs[rd] = regs[r1] & regs[r2];
4819 			break;
4820 		case DIF_OP_SLL:
4821 			regs[rd] = regs[r1] << regs[r2];
4822 			break;
4823 		case DIF_OP_SRL:
4824 			regs[rd] = regs[r1] >> regs[r2];
4825 			break;
4826 		case DIF_OP_SUB:
4827 			regs[rd] = regs[r1] - regs[r2];
4828 			break;
4829 		case DIF_OP_ADD:
4830 			regs[rd] = regs[r1] + regs[r2];
4831 			break;
4832 		case DIF_OP_MUL:
4833 			regs[rd] = regs[r1] * regs[r2];
4834 			break;
4835 		case DIF_OP_SDIV:
4836 			if (regs[r2] == 0) {
4837 				regs[rd] = 0;
4838 				*flags |= CPU_DTRACE_DIVZERO;
4839 			} else {
4840 				regs[rd] = (int64_t)regs[r1] /
4841 				    (int64_t)regs[r2];
4842 			}
4843 			break;
4844 
4845 		case DIF_OP_UDIV:
4846 			if (regs[r2] == 0) {
4847 				regs[rd] = 0;
4848 				*flags |= CPU_DTRACE_DIVZERO;
4849 			} else {
4850 				regs[rd] = regs[r1] / regs[r2];
4851 			}
4852 			break;
4853 
4854 		case DIF_OP_SREM:
4855 			if (regs[r2] == 0) {
4856 				regs[rd] = 0;
4857 				*flags |= CPU_DTRACE_DIVZERO;
4858 			} else {
4859 				regs[rd] = (int64_t)regs[r1] %
4860 				    (int64_t)regs[r2];
4861 			}
4862 			break;
4863 
4864 		case DIF_OP_UREM:
4865 			if (regs[r2] == 0) {
4866 				regs[rd] = 0;
4867 				*flags |= CPU_DTRACE_DIVZERO;
4868 			} else {
4869 				regs[rd] = regs[r1] % regs[r2];
4870 			}
4871 			break;
4872 
4873 		case DIF_OP_NOT:
4874 			regs[rd] = ~regs[r1];
4875 			break;
4876 		case DIF_OP_MOV:
4877 			regs[rd] = regs[r1];
4878 			break;
4879 		case DIF_OP_CMP:
4880 			cc_r = regs[r1] - regs[r2];
4881 			cc_n = cc_r < 0;
4882 			cc_z = cc_r == 0;
4883 			cc_v = 0;
4884 			cc_c = regs[r1] < regs[r2];
4885 			break;
4886 		case DIF_OP_TST:
4887 			cc_n = cc_v = cc_c = 0;
4888 			cc_z = regs[r1] == 0;
4889 			break;
4890 		case DIF_OP_BA:
4891 			pc = DIF_INSTR_LABEL(instr);
4892 			break;
4893 		case DIF_OP_BE:
4894 			if (cc_z)
4895 				pc = DIF_INSTR_LABEL(instr);
4896 			break;
4897 		case DIF_OP_BNE:
4898 			if (cc_z == 0)
4899 				pc = DIF_INSTR_LABEL(instr);
4900 			break;
4901 		case DIF_OP_BG:
4902 			if ((cc_z | (cc_n ^ cc_v)) == 0)
4903 				pc = DIF_INSTR_LABEL(instr);
4904 			break;
4905 		case DIF_OP_BGU:
4906 			if ((cc_c | cc_z) == 0)
4907 				pc = DIF_INSTR_LABEL(instr);
4908 			break;
4909 		case DIF_OP_BGE:
4910 			if ((cc_n ^ cc_v) == 0)
4911 				pc = DIF_INSTR_LABEL(instr);
4912 			break;
4913 		case DIF_OP_BGEU:
4914 			if (cc_c == 0)
4915 				pc = DIF_INSTR_LABEL(instr);
4916 			break;
4917 		case DIF_OP_BL:
4918 			if (cc_n ^ cc_v)
4919 				pc = DIF_INSTR_LABEL(instr);
4920 			break;
4921 		case DIF_OP_BLU:
4922 			if (cc_c)
4923 				pc = DIF_INSTR_LABEL(instr);
4924 			break;
4925 		case DIF_OP_BLE:
4926 			if (cc_z | (cc_n ^ cc_v))
4927 				pc = DIF_INSTR_LABEL(instr);
4928 			break;
4929 		case DIF_OP_BLEU:
4930 			if (cc_c | cc_z)
4931 				pc = DIF_INSTR_LABEL(instr);
4932 			break;
4933 		case DIF_OP_RLDSB:
4934 			if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) {
4935 				*flags |= CPU_DTRACE_KPRIV;
4936 				*illval = regs[r1];
4937 				break;
4938 			}
4939 			/*FALLTHROUGH*/
4940 		case DIF_OP_LDSB:
4941 			regs[rd] = (int8_t)dtrace_load8(regs[r1]);
4942 			break;
4943 		case DIF_OP_RLDSH:
4944 			if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) {
4945 				*flags |= CPU_DTRACE_KPRIV;
4946 				*illval = regs[r1];
4947 				break;
4948 			}
4949 			/*FALLTHROUGH*/
4950 		case DIF_OP_LDSH:
4951 			regs[rd] = (int16_t)dtrace_load16(regs[r1]);
4952 			break;
4953 		case DIF_OP_RLDSW:
4954 			if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) {
4955 				*flags |= CPU_DTRACE_KPRIV;
4956 				*illval = regs[r1];
4957 				break;
4958 			}
4959 			/*FALLTHROUGH*/
4960 		case DIF_OP_LDSW:
4961 			regs[rd] = (int32_t)dtrace_load32(regs[r1]);
4962 			break;
4963 		case DIF_OP_RLDUB:
4964 			if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) {
4965 				*flags |= CPU_DTRACE_KPRIV;
4966 				*illval = regs[r1];
4967 				break;
4968 			}
4969 			/*FALLTHROUGH*/
4970 		case DIF_OP_LDUB:
4971 			regs[rd] = dtrace_load8(regs[r1]);
4972 			break;
4973 		case DIF_OP_RLDUH:
4974 			if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) {
4975 				*flags |= CPU_DTRACE_KPRIV;
4976 				*illval = regs[r1];
4977 				break;
4978 			}
4979 			/*FALLTHROUGH*/
4980 		case DIF_OP_LDUH:
4981 			regs[rd] = dtrace_load16(regs[r1]);
4982 			break;
4983 		case DIF_OP_RLDUW:
4984 			if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) {
4985 				*flags |= CPU_DTRACE_KPRIV;
4986 				*illval = regs[r1];
4987 				break;
4988 			}
4989 			/*FALLTHROUGH*/
4990 		case DIF_OP_LDUW:
4991 			regs[rd] = dtrace_load32(regs[r1]);
4992 			break;
4993 		case DIF_OP_RLDX:
4994 			if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) {
4995 				*flags |= CPU_DTRACE_KPRIV;
4996 				*illval = regs[r1];
4997 				break;
4998 			}
4999 			/*FALLTHROUGH*/
5000 		case DIF_OP_LDX:
5001 			regs[rd] = dtrace_load64(regs[r1]);
5002 			break;
5003 		case DIF_OP_ULDSB:
5004 			regs[rd] = (int8_t)
5005 			    dtrace_fuword8((void *)(uintptr_t)regs[r1]);
5006 			break;
5007 		case DIF_OP_ULDSH:
5008 			regs[rd] = (int16_t)
5009 			    dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5010 			break;
5011 		case DIF_OP_ULDSW:
5012 			regs[rd] = (int32_t)
5013 			    dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5014 			break;
5015 		case DIF_OP_ULDUB:
5016 			regs[rd] =
5017 			    dtrace_fuword8((void *)(uintptr_t)regs[r1]);
5018 			break;
5019 		case DIF_OP_ULDUH:
5020 			regs[rd] =
5021 			    dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5022 			break;
5023 		case DIF_OP_ULDUW:
5024 			regs[rd] =
5025 			    dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5026 			break;
5027 		case DIF_OP_ULDX:
5028 			regs[rd] =
5029 			    dtrace_fuword64((void *)(uintptr_t)regs[r1]);
5030 			break;
5031 		case DIF_OP_RET:
5032 			rval = regs[rd];
5033 			pc = textlen;
5034 			break;
5035 		case DIF_OP_NOP:
5036 			break;
5037 		case DIF_OP_SETX:
5038 			regs[rd] = inttab[DIF_INSTR_INTEGER(instr)];
5039 			break;
5040 		case DIF_OP_SETS:
5041 			regs[rd] = (uint64_t)(uintptr_t)
5042 			    (strtab + DIF_INSTR_STRING(instr));
5043 			break;
5044 		case DIF_OP_SCMP: {
5045 			size_t sz = state->dts_options[DTRACEOPT_STRSIZE];
5046 			uintptr_t s1 = regs[r1];
5047 			uintptr_t s2 = regs[r2];
5048 
5049 			if (s1 != 0 &&
5050 			    !dtrace_strcanload(s1, sz, mstate, vstate))
5051 				break;
5052 			if (s2 != 0 &&
5053 			    !dtrace_strcanload(s2, sz, mstate, vstate))
5054 				break;
5055 
5056 			cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz);
5057 
5058 			cc_n = cc_r < 0;
5059 			cc_z = cc_r == 0;
5060 			cc_v = cc_c = 0;
5061 			break;
5062 		}
5063 		case DIF_OP_LDGA:
5064 			regs[rd] = dtrace_dif_variable(mstate, state,
5065 			    r1, regs[r2]);
5066 			break;
5067 		case DIF_OP_LDGS:
5068 			id = DIF_INSTR_VAR(instr);
5069 
5070 			if (id >= DIF_VAR_OTHER_UBASE) {
5071 				uintptr_t a;
5072 
5073 				id -= DIF_VAR_OTHER_UBASE;
5074 				svar = vstate->dtvs_globals[id];
5075 				ASSERT(svar != NULL);
5076 				v = &svar->dtsv_var;
5077 
5078 				if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) {
5079 					regs[rd] = svar->dtsv_data;
5080 					break;
5081 				}
5082 
5083 				a = (uintptr_t)svar->dtsv_data;
5084 
5085 				if (*(uint8_t *)a == UINT8_MAX) {
5086 					/*
5087 					 * If the 0th byte is set to UINT8_MAX
5088 					 * then this is to be treated as a
5089 					 * reference to a NULL variable.
5090 					 */
5091 					regs[rd] = 0;
5092 				} else {
5093 					regs[rd] = a + sizeof (uint64_t);
5094 				}
5095 
5096 				break;
5097 			}
5098 
5099 			regs[rd] = dtrace_dif_variable(mstate, state, id, 0);
5100 			break;
5101 
5102 		case DIF_OP_STGS:
5103 			id = DIF_INSTR_VAR(instr);
5104 
5105 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5106 			id -= DIF_VAR_OTHER_UBASE;
5107 
5108 			svar = vstate->dtvs_globals[id];
5109 			ASSERT(svar != NULL);
5110 			v = &svar->dtsv_var;
5111 
5112 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5113 				uintptr_t a = (uintptr_t)svar->dtsv_data;
5114 
5115 				ASSERT(a != 0);
5116 				ASSERT(svar->dtsv_size != 0);
5117 
5118 				if (regs[rd] == 0) {
5119 					*(uint8_t *)a = UINT8_MAX;
5120 					break;
5121 				} else {
5122 					*(uint8_t *)a = 0;
5123 					a += sizeof (uint64_t);
5124 				}
5125 				if (!dtrace_vcanload(
5126 				    (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5127 				    mstate, vstate))
5128 					break;
5129 
5130 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
5131 				    (void *)a, &v->dtdv_type);
5132 				break;
5133 			}
5134 
5135 			svar->dtsv_data = regs[rd];
5136 			break;
5137 
5138 		case DIF_OP_LDTA:
5139 			/*
5140 			 * There are no DTrace built-in thread-local arrays at
5141 			 * present.  This opcode is saved for future work.
5142 			 */
5143 			*flags |= CPU_DTRACE_ILLOP;
5144 			regs[rd] = 0;
5145 			break;
5146 
5147 		case DIF_OP_LDLS:
5148 			id = DIF_INSTR_VAR(instr);
5149 
5150 			if (id < DIF_VAR_OTHER_UBASE) {
5151 				/*
5152 				 * For now, this has no meaning.
5153 				 */
5154 				regs[rd] = 0;
5155 				break;
5156 			}
5157 
5158 			id -= DIF_VAR_OTHER_UBASE;
5159 
5160 			ASSERT(id < vstate->dtvs_nlocals);
5161 			ASSERT(vstate->dtvs_locals != NULL);
5162 
5163 			svar = vstate->dtvs_locals[id];
5164 			ASSERT(svar != NULL);
5165 			v = &svar->dtsv_var;
5166 
5167 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5168 				uintptr_t a = (uintptr_t)svar->dtsv_data;
5169 				size_t sz = v->dtdv_type.dtdt_size;
5170 
5171 				sz += sizeof (uint64_t);
5172 				ASSERT(svar->dtsv_size == NCPU * sz);
5173 				a += curcpu * sz;
5174 
5175 				if (*(uint8_t *)a == UINT8_MAX) {
5176 					/*
5177 					 * If the 0th byte is set to UINT8_MAX
5178 					 * then this is to be treated as a
5179 					 * reference to a NULL variable.
5180 					 */
5181 					regs[rd] = 0;
5182 				} else {
5183 					regs[rd] = a + sizeof (uint64_t);
5184 				}
5185 
5186 				break;
5187 			}
5188 
5189 			ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5190 			tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5191 			regs[rd] = tmp[curcpu];
5192 			break;
5193 
5194 		case DIF_OP_STLS:
5195 			id = DIF_INSTR_VAR(instr);
5196 
5197 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5198 			id -= DIF_VAR_OTHER_UBASE;
5199 			ASSERT(id < vstate->dtvs_nlocals);
5200 
5201 			ASSERT(vstate->dtvs_locals != NULL);
5202 			svar = vstate->dtvs_locals[id];
5203 			ASSERT(svar != NULL);
5204 			v = &svar->dtsv_var;
5205 
5206 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5207 				uintptr_t a = (uintptr_t)svar->dtsv_data;
5208 				size_t sz = v->dtdv_type.dtdt_size;
5209 
5210 				sz += sizeof (uint64_t);
5211 				ASSERT(svar->dtsv_size == NCPU * sz);
5212 				a += curcpu * sz;
5213 
5214 				if (regs[rd] == 0) {
5215 					*(uint8_t *)a = UINT8_MAX;
5216 					break;
5217 				} else {
5218 					*(uint8_t *)a = 0;
5219 					a += sizeof (uint64_t);
5220 				}
5221 
5222 				if (!dtrace_vcanload(
5223 				    (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5224 				    mstate, vstate))
5225 					break;
5226 
5227 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
5228 				    (void *)a, &v->dtdv_type);
5229 				break;
5230 			}
5231 
5232 			ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5233 			tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5234 			tmp[curcpu] = regs[rd];
5235 			break;
5236 
5237 		case DIF_OP_LDTS: {
5238 			dtrace_dynvar_t *dvar;
5239 			dtrace_key_t *key;
5240 
5241 			id = DIF_INSTR_VAR(instr);
5242 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5243 			id -= DIF_VAR_OTHER_UBASE;
5244 			v = &vstate->dtvs_tlocals[id];
5245 
5246 			key = &tupregs[DIF_DTR_NREGS];
5247 			key[0].dttk_value = (uint64_t)id;
5248 			key[0].dttk_size = 0;
5249 			DTRACE_TLS_THRKEY(key[1].dttk_value);
5250 			key[1].dttk_size = 0;
5251 
5252 			dvar = dtrace_dynvar(dstate, 2, key,
5253 			    sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC,
5254 			    mstate, vstate);
5255 
5256 			if (dvar == NULL) {
5257 				regs[rd] = 0;
5258 				break;
5259 			}
5260 
5261 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5262 				regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
5263 			} else {
5264 				regs[rd] = *((uint64_t *)dvar->dtdv_data);
5265 			}
5266 
5267 			break;
5268 		}
5269 
5270 		case DIF_OP_STTS: {
5271 			dtrace_dynvar_t *dvar;
5272 			dtrace_key_t *key;
5273 
5274 			id = DIF_INSTR_VAR(instr);
5275 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5276 			id -= DIF_VAR_OTHER_UBASE;
5277 
5278 			key = &tupregs[DIF_DTR_NREGS];
5279 			key[0].dttk_value = (uint64_t)id;
5280 			key[0].dttk_size = 0;
5281 			DTRACE_TLS_THRKEY(key[1].dttk_value);
5282 			key[1].dttk_size = 0;
5283 			v = &vstate->dtvs_tlocals[id];
5284 
5285 			dvar = dtrace_dynvar(dstate, 2, key,
5286 			    v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5287 			    v->dtdv_type.dtdt_size : sizeof (uint64_t),
5288 			    regs[rd] ? DTRACE_DYNVAR_ALLOC :
5289 			    DTRACE_DYNVAR_DEALLOC, mstate, vstate);
5290 
5291 			/*
5292 			 * Given that we're storing to thread-local data,
5293 			 * we need to flush our predicate cache.
5294 			 */
5295 			curthread->t_predcache = 0;
5296 
5297 			if (dvar == NULL)
5298 				break;
5299 
5300 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5301 				if (!dtrace_vcanload(
5302 				    (void *)(uintptr_t)regs[rd],
5303 				    &v->dtdv_type, mstate, vstate))
5304 					break;
5305 
5306 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
5307 				    dvar->dtdv_data, &v->dtdv_type);
5308 			} else {
5309 				*((uint64_t *)dvar->dtdv_data) = regs[rd];
5310 			}
5311 
5312 			break;
5313 		}
5314 
5315 		case DIF_OP_SRA:
5316 			regs[rd] = (int64_t)regs[r1] >> regs[r2];
5317 			break;
5318 
5319 		case DIF_OP_CALL:
5320 			dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd,
5321 			    regs, tupregs, ttop, mstate, state);
5322 			break;
5323 
5324 		case DIF_OP_PUSHTR:
5325 			if (ttop == DIF_DTR_NREGS) {
5326 				*flags |= CPU_DTRACE_TUPOFLOW;
5327 				break;
5328 			}
5329 
5330 			if (r1 == DIF_TYPE_STRING) {
5331 				/*
5332 				 * If this is a string type and the size is 0,
5333 				 * we'll use the system-wide default string
5334 				 * size.  Note that we are _not_ looking at
5335 				 * the value of the DTRACEOPT_STRSIZE option;
5336 				 * had this been set, we would expect to have
5337 				 * a non-zero size value in the "pushtr".
5338 				 */
5339 				tupregs[ttop].dttk_size =
5340 				    dtrace_strlen((char *)(uintptr_t)regs[rd],
5341 				    regs[r2] ? regs[r2] :
5342 				    dtrace_strsize_default) + 1;
5343 			} else {
5344 				tupregs[ttop].dttk_size = regs[r2];
5345 			}
5346 
5347 			tupregs[ttop++].dttk_value = regs[rd];
5348 			break;
5349 
5350 		case DIF_OP_PUSHTV:
5351 			if (ttop == DIF_DTR_NREGS) {
5352 				*flags |= CPU_DTRACE_TUPOFLOW;
5353 				break;
5354 			}
5355 
5356 			tupregs[ttop].dttk_value = regs[rd];
5357 			tupregs[ttop++].dttk_size = 0;
5358 			break;
5359 
5360 		case DIF_OP_POPTS:
5361 			if (ttop != 0)
5362 				ttop--;
5363 			break;
5364 
5365 		case DIF_OP_FLUSHTS:
5366 			ttop = 0;
5367 			break;
5368 
5369 		case DIF_OP_LDGAA:
5370 		case DIF_OP_LDTAA: {
5371 			dtrace_dynvar_t *dvar;
5372 			dtrace_key_t *key = tupregs;
5373 			uint_t nkeys = ttop;
5374 
5375 			id = DIF_INSTR_VAR(instr);
5376 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5377 			id -= DIF_VAR_OTHER_UBASE;
5378 
5379 			key[nkeys].dttk_value = (uint64_t)id;
5380 			key[nkeys++].dttk_size = 0;
5381 
5382 			if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) {
5383 				DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
5384 				key[nkeys++].dttk_size = 0;
5385 				v = &vstate->dtvs_tlocals[id];
5386 			} else {
5387 				v = &vstate->dtvs_globals[id]->dtsv_var;
5388 			}
5389 
5390 			dvar = dtrace_dynvar(dstate, nkeys, key,
5391 			    v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5392 			    v->dtdv_type.dtdt_size : sizeof (uint64_t),
5393 			    DTRACE_DYNVAR_NOALLOC, mstate, vstate);
5394 
5395 			if (dvar == NULL) {
5396 				regs[rd] = 0;
5397 				break;
5398 			}
5399 
5400 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5401 				regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
5402 			} else {
5403 				regs[rd] = *((uint64_t *)dvar->dtdv_data);
5404 			}
5405 
5406 			break;
5407 		}
5408 
5409 		case DIF_OP_STGAA:
5410 		case DIF_OP_STTAA: {
5411 			dtrace_dynvar_t *dvar;
5412 			dtrace_key_t *key = tupregs;
5413 			uint_t nkeys = ttop;
5414 
5415 			id = DIF_INSTR_VAR(instr);
5416 			ASSERT(id >= DIF_VAR_OTHER_UBASE);
5417 			id -= DIF_VAR_OTHER_UBASE;
5418 
5419 			key[nkeys].dttk_value = (uint64_t)id;
5420 			key[nkeys++].dttk_size = 0;
5421 
5422 			if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) {
5423 				DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
5424 				key[nkeys++].dttk_size = 0;
5425 				v = &vstate->dtvs_tlocals[id];
5426 			} else {
5427 				v = &vstate->dtvs_globals[id]->dtsv_var;
5428 			}
5429 
5430 			dvar = dtrace_dynvar(dstate, nkeys, key,
5431 			    v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5432 			    v->dtdv_type.dtdt_size : sizeof (uint64_t),
5433 			    regs[rd] ? DTRACE_DYNVAR_ALLOC :
5434 			    DTRACE_DYNVAR_DEALLOC, mstate, vstate);
5435 
5436 			if (dvar == NULL)
5437 				break;
5438 
5439 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5440 				if (!dtrace_vcanload(
5441 				    (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5442 				    mstate, vstate))
5443 					break;
5444 
5445 				dtrace_vcopy((void *)(uintptr_t)regs[rd],
5446 				    dvar->dtdv_data, &v->dtdv_type);
5447 			} else {
5448 				*((uint64_t *)dvar->dtdv_data) = regs[rd];
5449 			}
5450 
5451 			break;
5452 		}
5453 
5454 		case DIF_OP_ALLOCS: {
5455 			uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
5456 			size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1];
5457 
5458 			/*
5459 			 * Rounding up the user allocation size could have
5460 			 * overflowed large, bogus allocations (like -1ULL) to
5461 			 * 0.
5462 			 */
5463 			if (size < regs[r1] ||
5464 			    !DTRACE_INSCRATCH(mstate, size)) {
5465 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5466 				regs[rd] = 0;
5467 				break;
5468 			}
5469 
5470 			dtrace_bzero((void *) mstate->dtms_scratch_ptr, size);
5471 			mstate->dtms_scratch_ptr += size;
5472 			regs[rd] = ptr;
5473 			break;
5474 		}
5475 
5476 		case DIF_OP_COPYS:
5477 			if (!dtrace_canstore(regs[rd], regs[r2],
5478 			    mstate, vstate)) {
5479 				*flags |= CPU_DTRACE_BADADDR;
5480 				*illval = regs[rd];
5481 				break;
5482 			}
5483 
5484 			if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate))
5485 				break;
5486 
5487 			dtrace_bcopy((void *)(uintptr_t)regs[r1],
5488 			    (void *)(uintptr_t)regs[rd], (size_t)regs[r2]);
5489 			break;
5490 
5491 		case DIF_OP_STB:
5492 			if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) {
5493 				*flags |= CPU_DTRACE_BADADDR;
5494 				*illval = regs[rd];
5495 				break;
5496 			}
5497 			*((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1];
5498 			break;
5499 
5500 		case DIF_OP_STH:
5501 			if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) {
5502 				*flags |= CPU_DTRACE_BADADDR;
5503 				*illval = regs[rd];
5504 				break;
5505 			}
5506 			if (regs[rd] & 1) {
5507 				*flags |= CPU_DTRACE_BADALIGN;
5508 				*illval = regs[rd];
5509 				break;
5510 			}
5511 			*((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1];
5512 			break;
5513 
5514 		case DIF_OP_STW:
5515 			if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) {
5516 				*flags |= CPU_DTRACE_BADADDR;
5517 				*illval = regs[rd];
5518 				break;
5519 			}
5520 			if (regs[rd] & 3) {
5521 				*flags |= CPU_DTRACE_BADALIGN;
5522 				*illval = regs[rd];
5523 				break;
5524 			}
5525 			*((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1];
5526 			break;
5527 
5528 		case DIF_OP_STX:
5529 			if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) {
5530 				*flags |= CPU_DTRACE_BADADDR;
5531 				*illval = regs[rd];
5532 				break;
5533 			}
5534 			if (regs[rd] & 7) {
5535 				*flags |= CPU_DTRACE_BADALIGN;
5536 				*illval = regs[rd];
5537 				break;
5538 			}
5539 			*((uint64_t *)(uintptr_t)regs[rd]) = regs[r1];
5540 			break;
5541 		}
5542 	}
5543 
5544 	if (!(*flags & CPU_DTRACE_FAULT))
5545 		return (rval);
5546 
5547 	mstate->dtms_fltoffs = opc * sizeof (dif_instr_t);
5548 	mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS;
5549 
5550 	return (0);
5551 }
5552 
5553 static void
5554 dtrace_action_breakpoint(dtrace_ecb_t *ecb)
5555 {
5556 	dtrace_probe_t *probe = ecb->dte_probe;
5557 	dtrace_provider_t *prov = probe->dtpr_provider;
5558 	char c[DTRACE_FULLNAMELEN + 80], *str;
5559 	char *msg = "dtrace: breakpoint action at probe ";
5560 	char *ecbmsg = " (ecb ";
5561 	uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4));
5562 	uintptr_t val = (uintptr_t)ecb;
5563 	int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0;
5564 
5565 	if (dtrace_destructive_disallow)
5566 		return;
5567 
5568 	/*
5569 	 * It's impossible to be taking action on the NULL probe.
5570 	 */
5571 	ASSERT(probe != NULL);
5572 
5573 	/*
5574 	 * This is a poor man's (destitute man's?) sprintf():  we want to
5575 	 * print the provider name, module name, function name and name of
5576 	 * the probe, along with the hex address of the ECB with the breakpoint
5577 	 * action -- all of which we must place in the character buffer by
5578 	 * hand.
5579 	 */
5580 	while (*msg != '\0')
5581 		c[i++] = *msg++;
5582 
5583 	for (str = prov->dtpv_name; *str != '\0'; str++)
5584 		c[i++] = *str;
5585 	c[i++] = ':';
5586 
5587 	for (str = probe->dtpr_mod; *str != '\0'; str++)
5588 		c[i++] = *str;
5589 	c[i++] = ':';
5590 
5591 	for (str = probe->dtpr_func; *str != '\0'; str++)
5592 		c[i++] = *str;
5593 	c[i++] = ':';
5594 
5595 	for (str = probe->dtpr_name; *str != '\0'; str++)
5596 		c[i++] = *str;
5597 
5598 	while (*ecbmsg != '\0')
5599 		c[i++] = *ecbmsg++;
5600 
5601 	while (shift >= 0) {
5602 		mask = (uintptr_t)0xf << shift;
5603 
5604 		if (val >= ((uintptr_t)1 << shift))
5605 			c[i++] = "0123456789abcdef"[(val & mask) >> shift];
5606 		shift -= 4;
5607 	}
5608 
5609 	c[i++] = ')';
5610 	c[i] = '\0';
5611 
5612 #if defined(sun)
5613 	debug_enter(c);
5614 #else
5615 	kdb_enter(KDB_WHY_DTRACE, "breakpoint action");
5616 #endif
5617 }
5618 
5619 static void
5620 dtrace_action_panic(dtrace_ecb_t *ecb)
5621 {
5622 	dtrace_probe_t *probe = ecb->dte_probe;
5623 
5624 	/*
5625 	 * It's impossible to be taking action on the NULL probe.
5626 	 */
5627 	ASSERT(probe != NULL);
5628 
5629 	if (dtrace_destructive_disallow)
5630 		return;
5631 
5632 	if (dtrace_panicked != NULL)
5633 		return;
5634 
5635 	if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL)
5636 		return;
5637 
5638 	/*
5639 	 * We won the right to panic.  (We want to be sure that only one
5640 	 * thread calls panic() from dtrace_probe(), and that panic() is
5641 	 * called exactly once.)
5642 	 */
5643 	dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)",
5644 	    probe->dtpr_provider->dtpv_name, probe->dtpr_mod,
5645 	    probe->dtpr_func, probe->dtpr_name, (void *)ecb);
5646 }
5647 
5648 static void
5649 dtrace_action_raise(uint64_t sig)
5650 {
5651 	if (dtrace_destructive_disallow)
5652 		return;
5653 
5654 	if (sig >= NSIG) {
5655 		DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
5656 		return;
5657 	}
5658 
5659 #if defined(sun)
5660 	/*
5661 	 * raise() has a queue depth of 1 -- we ignore all subsequent
5662 	 * invocations of the raise() action.
5663 	 */
5664 	if (curthread->t_dtrace_sig == 0)
5665 		curthread->t_dtrace_sig = (uint8_t)sig;
5666 
5667 	curthread->t_sig_check = 1;
5668 	aston(curthread);
5669 #else
5670 	struct proc *p = curproc;
5671 	PROC_LOCK(p);
5672 	kern_psignal(p, sig);
5673 	PROC_UNLOCK(p);
5674 #endif
5675 }
5676 
5677 static void
5678 dtrace_action_stop(void)
5679 {
5680 	if (dtrace_destructive_disallow)
5681 		return;
5682 
5683 #if defined(sun)
5684 	if (!curthread->t_dtrace_stop) {
5685 		curthread->t_dtrace_stop = 1;
5686 		curthread->t_sig_check = 1;
5687 		aston(curthread);
5688 	}
5689 #else
5690 	struct proc *p = curproc;
5691 	PROC_LOCK(p);
5692 	kern_psignal(p, SIGSTOP);
5693 	PROC_UNLOCK(p);
5694 #endif
5695 }
5696 
5697 static void
5698 dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val)
5699 {
5700 	hrtime_t now;
5701 	volatile uint16_t *flags;
5702 #if defined(sun)
5703 	cpu_t *cpu = CPU;
5704 #else
5705 	cpu_t *cpu = &solaris_cpu[curcpu];
5706 #endif
5707 
5708 	if (dtrace_destructive_disallow)
5709 		return;
5710 
5711 	flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags;
5712 
5713 	now = dtrace_gethrtime();
5714 
5715 	if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) {
5716 		/*
5717 		 * We need to advance the mark to the current time.
5718 		 */
5719 		cpu->cpu_dtrace_chillmark = now;
5720 		cpu->cpu_dtrace_chilled = 0;
5721 	}
5722 
5723 	/*
5724 	 * Now check to see if the requested chill time would take us over
5725 	 * the maximum amount of time allowed in the chill interval.  (Or
5726 	 * worse, if the calculation itself induces overflow.)
5727 	 */
5728 	if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max ||
5729 	    cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) {
5730 		*flags |= CPU_DTRACE_ILLOP;
5731 		return;
5732 	}
5733 
5734 	while (dtrace_gethrtime() - now < val)
5735 		continue;
5736 
5737 	/*
5738 	 * Normally, we assure that the value of the variable "timestamp" does
5739 	 * not change within an ECB.  The presence of chill() represents an
5740 	 * exception to this rule, however.
5741 	 */
5742 	mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP;
5743 	cpu->cpu_dtrace_chilled += val;
5744 }
5745 
5746 static void
5747 dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state,
5748     uint64_t *buf, uint64_t arg)
5749 {
5750 	int nframes = DTRACE_USTACK_NFRAMES(arg);
5751 	int strsize = DTRACE_USTACK_STRSIZE(arg);
5752 	uint64_t *pcs = &buf[1], *fps;
5753 	char *str = (char *)&pcs[nframes];
5754 	int size, offs = 0, i, j;
5755 	uintptr_t old = mstate->dtms_scratch_ptr, saved;
5756 	uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
5757 	char *sym;
5758 
5759 	/*
5760 	 * Should be taking a faster path if string space has not been
5761 	 * allocated.
5762 	 */
5763 	ASSERT(strsize != 0);
5764 
5765 	/*
5766 	 * We will first allocate some temporary space for the frame pointers.
5767 	 */
5768 	fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
5769 	size = (uintptr_t)fps - mstate->dtms_scratch_ptr +
5770 	    (nframes * sizeof (uint64_t));
5771 
5772 	if (!DTRACE_INSCRATCH(mstate, size)) {
5773 		/*
5774 		 * Not enough room for our frame pointers -- need to indicate
5775 		 * that we ran out of scratch space.
5776 		 */
5777 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5778 		return;
5779 	}
5780 
5781 	mstate->dtms_scratch_ptr += size;
5782 	saved = mstate->dtms_scratch_ptr;
5783 
5784 	/*
5785 	 * Now get a stack with both program counters and frame pointers.
5786 	 */
5787 	DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5788 	dtrace_getufpstack(buf, fps, nframes + 1);
5789 	DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5790 
5791 	/*
5792 	 * If that faulted, we're cooked.
5793 	 */
5794 	if (*flags & CPU_DTRACE_FAULT)
5795 		goto out;
5796 
5797 	/*
5798 	 * Now we want to walk up the stack, calling the USTACK helper.  For
5799 	 * each iteration, we restore the scratch pointer.
5800 	 */
5801 	for (i = 0; i < nframes; i++) {
5802 		mstate->dtms_scratch_ptr = saved;
5803 
5804 		if (offs >= strsize)
5805 			break;
5806 
5807 		sym = (char *)(uintptr_t)dtrace_helper(
5808 		    DTRACE_HELPER_ACTION_USTACK,
5809 		    mstate, state, pcs[i], fps[i]);
5810 
5811 		/*
5812 		 * If we faulted while running the helper, we're going to
5813 		 * clear the fault and null out the corresponding string.
5814 		 */
5815 		if (*flags & CPU_DTRACE_FAULT) {
5816 			*flags &= ~CPU_DTRACE_FAULT;
5817 			str[offs++] = '\0';
5818 			continue;
5819 		}
5820 
5821 		if (sym == NULL) {
5822 			str[offs++] = '\0';
5823 			continue;
5824 		}
5825 
5826 		DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5827 
5828 		/*
5829 		 * Now copy in the string that the helper returned to us.
5830 		 */
5831 		for (j = 0; offs + j < strsize; j++) {
5832 			if ((str[offs + j] = sym[j]) == '\0')
5833 				break;
5834 		}
5835 
5836 		DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5837 
5838 		offs += j + 1;
5839 	}
5840 
5841 	if (offs >= strsize) {
5842 		/*
5843 		 * If we didn't have room for all of the strings, we don't
5844 		 * abort processing -- this needn't be a fatal error -- but we
5845 		 * still want to increment a counter (dts_stkstroverflows) to
5846 		 * allow this condition to be warned about.  (If this is from
5847 		 * a jstack() action, it is easily tuned via jstackstrsize.)
5848 		 */
5849 		dtrace_error(&state->dts_stkstroverflows);
5850 	}
5851 
5852 	while (offs < strsize)
5853 		str[offs++] = '\0';
5854 
5855 out:
5856 	mstate->dtms_scratch_ptr = old;
5857 }
5858 
5859 /*
5860  * If you're looking for the epicenter of DTrace, you just found it.  This
5861  * is the function called by the provider to fire a probe -- from which all
5862  * subsequent probe-context DTrace activity emanates.
5863  */
5864 void
5865 dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1,
5866     uintptr_t arg2, uintptr_t arg3, uintptr_t arg4)
5867 {
5868 	processorid_t cpuid;
5869 	dtrace_icookie_t cookie;
5870 	dtrace_probe_t *probe;
5871 	dtrace_mstate_t mstate;
5872 	dtrace_ecb_t *ecb;
5873 	dtrace_action_t *act;
5874 	intptr_t offs;
5875 	size_t size;
5876 	int vtime, onintr;
5877 	volatile uint16_t *flags;
5878 	hrtime_t now;
5879 
5880 	if (panicstr != NULL)
5881 		return;
5882 
5883 #if defined(sun)
5884 	/*
5885 	 * Kick out immediately if this CPU is still being born (in which case
5886 	 * curthread will be set to -1) or the current thread can't allow
5887 	 * probes in its current context.
5888 	 */
5889 	if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE))
5890 		return;
5891 #endif
5892 
5893 	cookie = dtrace_interrupt_disable();
5894 	probe = dtrace_probes[id - 1];
5895 	cpuid = curcpu;
5896 	onintr = CPU_ON_INTR(CPU);
5897 
5898 	if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE &&
5899 	    probe->dtpr_predcache == curthread->t_predcache) {
5900 		/*
5901 		 * We have hit in the predicate cache; we know that
5902 		 * this predicate would evaluate to be false.
5903 		 */
5904 		dtrace_interrupt_enable(cookie);
5905 		return;
5906 	}
5907 
5908 #if defined(sun)
5909 	if (panic_quiesce) {
5910 #else
5911 	if (panicstr != NULL) {
5912 #endif
5913 		/*
5914 		 * We don't trace anything if we're panicking.
5915 		 */
5916 		dtrace_interrupt_enable(cookie);
5917 		return;
5918 	}
5919 
5920 	now = dtrace_gethrtime();
5921 	vtime = dtrace_vtime_references != 0;
5922 
5923 	if (vtime && curthread->t_dtrace_start)
5924 		curthread->t_dtrace_vtime += now - curthread->t_dtrace_start;
5925 
5926 	mstate.dtms_difo = NULL;
5927 	mstate.dtms_probe = probe;
5928 	mstate.dtms_strtok = 0;
5929 	mstate.dtms_arg[0] = arg0;
5930 	mstate.dtms_arg[1] = arg1;
5931 	mstate.dtms_arg[2] = arg2;
5932 	mstate.dtms_arg[3] = arg3;
5933 	mstate.dtms_arg[4] = arg4;
5934 
5935 	flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags;
5936 
5937 	for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
5938 		dtrace_predicate_t *pred = ecb->dte_predicate;
5939 		dtrace_state_t *state = ecb->dte_state;
5940 		dtrace_buffer_t *buf = &state->dts_buffer[cpuid];
5941 		dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid];
5942 		dtrace_vstate_t *vstate = &state->dts_vstate;
5943 		dtrace_provider_t *prov = probe->dtpr_provider;
5944 		int committed = 0;
5945 		caddr_t tomax;
5946 
5947 		/*
5948 		 * A little subtlety with the following (seemingly innocuous)
5949 		 * declaration of the automatic 'val':  by looking at the
5950 		 * code, you might think that it could be declared in the
5951 		 * action processing loop, below.  (That is, it's only used in
5952 		 * the action processing loop.)  However, it must be declared
5953 		 * out of that scope because in the case of DIF expression
5954 		 * arguments to aggregating actions, one iteration of the
5955 		 * action loop will use the last iteration's value.
5956 		 */
5957 		uint64_t val = 0;
5958 
5959 		mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE;
5960 		*flags &= ~CPU_DTRACE_ERROR;
5961 
5962 		if (prov == dtrace_provider) {
5963 			/*
5964 			 * If dtrace itself is the provider of this probe,
5965 			 * we're only going to continue processing the ECB if
5966 			 * arg0 (the dtrace_state_t) is equal to the ECB's
5967 			 * creating state.  (This prevents disjoint consumers
5968 			 * from seeing one another's metaprobes.)
5969 			 */
5970 			if (arg0 != (uint64_t)(uintptr_t)state)
5971 				continue;
5972 		}
5973 
5974 		if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) {
5975 			/*
5976 			 * We're not currently active.  If our provider isn't
5977 			 * the dtrace pseudo provider, we're not interested.
5978 			 */
5979 			if (prov != dtrace_provider)
5980 				continue;
5981 
5982 			/*
5983 			 * Now we must further check if we are in the BEGIN
5984 			 * probe.  If we are, we will only continue processing
5985 			 * if we're still in WARMUP -- if one BEGIN enabling
5986 			 * has invoked the exit() action, we don't want to
5987 			 * evaluate subsequent BEGIN enablings.
5988 			 */
5989 			if (probe->dtpr_id == dtrace_probeid_begin &&
5990 			    state->dts_activity != DTRACE_ACTIVITY_WARMUP) {
5991 				ASSERT(state->dts_activity ==
5992 				    DTRACE_ACTIVITY_DRAINING);
5993 				continue;
5994 			}
5995 		}
5996 
5997 		if (ecb->dte_cond) {
5998 			/*
5999 			 * If the dte_cond bits indicate that this
6000 			 * consumer is only allowed to see user-mode firings
6001 			 * of this probe, call the provider's dtps_usermode()
6002 			 * entry point to check that the probe was fired
6003 			 * while in a user context. Skip this ECB if that's
6004 			 * not the case.
6005 			 */
6006 			if ((ecb->dte_cond & DTRACE_COND_USERMODE) &&
6007 			    prov->dtpv_pops.dtps_usermode(prov->dtpv_arg,
6008 			    probe->dtpr_id, probe->dtpr_arg) == 0)
6009 				continue;
6010 
6011 #if defined(sun)
6012 			/*
6013 			 * This is more subtle than it looks. We have to be
6014 			 * absolutely certain that CRED() isn't going to
6015 			 * change out from under us so it's only legit to
6016 			 * examine that structure if we're in constrained
6017 			 * situations. Currently, the only times we'll this
6018 			 * check is if a non-super-user has enabled the
6019 			 * profile or syscall providers -- providers that
6020 			 * allow visibility of all processes. For the
6021 			 * profile case, the check above will ensure that
6022 			 * we're examining a user context.
6023 			 */
6024 			if (ecb->dte_cond & DTRACE_COND_OWNER) {
6025 				cred_t *cr;
6026 				cred_t *s_cr =
6027 				    ecb->dte_state->dts_cred.dcr_cred;
6028 				proc_t *proc;
6029 
6030 				ASSERT(s_cr != NULL);
6031 
6032 				if ((cr = CRED()) == NULL ||
6033 				    s_cr->cr_uid != cr->cr_uid ||
6034 				    s_cr->cr_uid != cr->cr_ruid ||
6035 				    s_cr->cr_uid != cr->cr_suid ||
6036 				    s_cr->cr_gid != cr->cr_gid ||
6037 				    s_cr->cr_gid != cr->cr_rgid ||
6038 				    s_cr->cr_gid != cr->cr_sgid ||
6039 				    (proc = ttoproc(curthread)) == NULL ||
6040 				    (proc->p_flag & SNOCD))
6041 					continue;
6042 			}
6043 
6044 			if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) {
6045 				cred_t *cr;
6046 				cred_t *s_cr =
6047 				    ecb->dte_state->dts_cred.dcr_cred;
6048 
6049 				ASSERT(s_cr != NULL);
6050 
6051 				if ((cr = CRED()) == NULL ||
6052 				    s_cr->cr_zone->zone_id !=
6053 				    cr->cr_zone->zone_id)
6054 					continue;
6055 			}
6056 #endif
6057 		}
6058 
6059 		if (now - state->dts_alive > dtrace_deadman_timeout) {
6060 			/*
6061 			 * We seem to be dead.  Unless we (a) have kernel
6062 			 * destructive permissions (b) have expicitly enabled
6063 			 * destructive actions and (c) destructive actions have
6064 			 * not been disabled, we're going to transition into
6065 			 * the KILLED state, from which no further processing
6066 			 * on this state will be performed.
6067 			 */
6068 			if (!dtrace_priv_kernel_destructive(state) ||
6069 			    !state->dts_cred.dcr_destructive ||
6070 			    dtrace_destructive_disallow) {
6071 				void *activity = &state->dts_activity;
6072 				dtrace_activity_t current;
6073 
6074 				do {
6075 					current = state->dts_activity;
6076 				} while (dtrace_cas32(activity, current,
6077 				    DTRACE_ACTIVITY_KILLED) != current);
6078 
6079 				continue;
6080 			}
6081 		}
6082 
6083 		if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed,
6084 		    ecb->dte_alignment, state, &mstate)) < 0)
6085 			continue;
6086 
6087 		tomax = buf->dtb_tomax;
6088 		ASSERT(tomax != NULL);
6089 
6090 		if (ecb->dte_size != 0)
6091 			DTRACE_STORE(uint32_t, tomax, offs, ecb->dte_epid);
6092 
6093 		mstate.dtms_epid = ecb->dte_epid;
6094 		mstate.dtms_present |= DTRACE_MSTATE_EPID;
6095 
6096 		if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL)
6097 			mstate.dtms_access = DTRACE_ACCESS_KERNEL;
6098 		else
6099 			mstate.dtms_access = 0;
6100 
6101 		if (pred != NULL) {
6102 			dtrace_difo_t *dp = pred->dtp_difo;
6103 			int rval;
6104 
6105 			rval = dtrace_dif_emulate(dp, &mstate, vstate, state);
6106 
6107 			if (!(*flags & CPU_DTRACE_ERROR) && !rval) {
6108 				dtrace_cacheid_t cid = probe->dtpr_predcache;
6109 
6110 				if (cid != DTRACE_CACHEIDNONE && !onintr) {
6111 					/*
6112 					 * Update the predicate cache...
6113 					 */
6114 					ASSERT(cid == pred->dtp_cacheid);
6115 					curthread->t_predcache = cid;
6116 				}
6117 
6118 				continue;
6119 			}
6120 		}
6121 
6122 		for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) &&
6123 		    act != NULL; act = act->dta_next) {
6124 			size_t valoffs;
6125 			dtrace_difo_t *dp;
6126 			dtrace_recdesc_t *rec = &act->dta_rec;
6127 
6128 			size = rec->dtrd_size;
6129 			valoffs = offs + rec->dtrd_offset;
6130 
6131 			if (DTRACEACT_ISAGG(act->dta_kind)) {
6132 				uint64_t v = 0xbad;
6133 				dtrace_aggregation_t *agg;
6134 
6135 				agg = (dtrace_aggregation_t *)act;
6136 
6137 				if ((dp = act->dta_difo) != NULL)
6138 					v = dtrace_dif_emulate(dp,
6139 					    &mstate, vstate, state);
6140 
6141 				if (*flags & CPU_DTRACE_ERROR)
6142 					continue;
6143 
6144 				/*
6145 				 * Note that we always pass the expression
6146 				 * value from the previous iteration of the
6147 				 * action loop.  This value will only be used
6148 				 * if there is an expression argument to the
6149 				 * aggregating action, denoted by the
6150 				 * dtag_hasarg field.
6151 				 */
6152 				dtrace_aggregate(agg, buf,
6153 				    offs, aggbuf, v, val);
6154 				continue;
6155 			}
6156 
6157 			switch (act->dta_kind) {
6158 			case DTRACEACT_STOP:
6159 				if (dtrace_priv_proc_destructive(state))
6160 					dtrace_action_stop();
6161 				continue;
6162 
6163 			case DTRACEACT_BREAKPOINT:
6164 				if (dtrace_priv_kernel_destructive(state))
6165 					dtrace_action_breakpoint(ecb);
6166 				continue;
6167 
6168 			case DTRACEACT_PANIC:
6169 				if (dtrace_priv_kernel_destructive(state))
6170 					dtrace_action_panic(ecb);
6171 				continue;
6172 
6173 			case DTRACEACT_STACK:
6174 				if (!dtrace_priv_kernel(state))
6175 					continue;
6176 
6177 				dtrace_getpcstack((pc_t *)(tomax + valoffs),
6178 				    size / sizeof (pc_t), probe->dtpr_aframes,
6179 				    DTRACE_ANCHORED(probe) ? NULL :
6180 				    (uint32_t *)arg0);
6181 				continue;
6182 
6183 			case DTRACEACT_JSTACK:
6184 			case DTRACEACT_USTACK:
6185 				if (!dtrace_priv_proc(state))
6186 					continue;
6187 
6188 				/*
6189 				 * See comment in DIF_VAR_PID.
6190 				 */
6191 				if (DTRACE_ANCHORED(mstate.dtms_probe) &&
6192 				    CPU_ON_INTR(CPU)) {
6193 					int depth = DTRACE_USTACK_NFRAMES(
6194 					    rec->dtrd_arg) + 1;
6195 
6196 					dtrace_bzero((void *)(tomax + valoffs),
6197 					    DTRACE_USTACK_STRSIZE(rec->dtrd_arg)
6198 					    + depth * sizeof (uint64_t));
6199 
6200 					continue;
6201 				}
6202 
6203 				if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 &&
6204 				    curproc->p_dtrace_helpers != NULL) {
6205 					/*
6206 					 * This is the slow path -- we have
6207 					 * allocated string space, and we're
6208 					 * getting the stack of a process that
6209 					 * has helpers.  Call into a separate
6210 					 * routine to perform this processing.
6211 					 */
6212 					dtrace_action_ustack(&mstate, state,
6213 					    (uint64_t *)(tomax + valoffs),
6214 					    rec->dtrd_arg);
6215 					continue;
6216 				}
6217 
6218 				DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6219 				dtrace_getupcstack((uint64_t *)
6220 				    (tomax + valoffs),
6221 				    DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1);
6222 				DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6223 				continue;
6224 
6225 			default:
6226 				break;
6227 			}
6228 
6229 			dp = act->dta_difo;
6230 			ASSERT(dp != NULL);
6231 
6232 			val = dtrace_dif_emulate(dp, &mstate, vstate, state);
6233 
6234 			if (*flags & CPU_DTRACE_ERROR)
6235 				continue;
6236 
6237 			switch (act->dta_kind) {
6238 			case DTRACEACT_SPECULATE:
6239 				ASSERT(buf == &state->dts_buffer[cpuid]);
6240 				buf = dtrace_speculation_buffer(state,
6241 				    cpuid, val);
6242 
6243 				if (buf == NULL) {
6244 					*flags |= CPU_DTRACE_DROP;
6245 					continue;
6246 				}
6247 
6248 				offs = dtrace_buffer_reserve(buf,
6249 				    ecb->dte_needed, ecb->dte_alignment,
6250 				    state, NULL);
6251 
6252 				if (offs < 0) {
6253 					*flags |= CPU_DTRACE_DROP;
6254 					continue;
6255 				}
6256 
6257 				tomax = buf->dtb_tomax;
6258 				ASSERT(tomax != NULL);
6259 
6260 				if (ecb->dte_size != 0)
6261 					DTRACE_STORE(uint32_t, tomax, offs,
6262 					    ecb->dte_epid);
6263 				continue;
6264 
6265 			case DTRACEACT_PRINTM: {
6266 				/* The DIF returns a 'memref'. */
6267 				uintptr_t *memref = (uintptr_t *)(uintptr_t) val;
6268 
6269 				/* Get the size from the memref. */
6270 				size = memref[1];
6271 
6272 				/*
6273 				 * Check if the size exceeds the allocated
6274 				 * buffer size.
6275 				 */
6276 				if (size + sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) {
6277 					/* Flag a drop! */
6278 					*flags |= CPU_DTRACE_DROP;
6279 					continue;
6280 				}
6281 
6282 				/* Store the size in the buffer first. */
6283 				DTRACE_STORE(uintptr_t, tomax,
6284 				    valoffs, size);
6285 
6286 				/*
6287 				 * Offset the buffer address to the start
6288 				 * of the data.
6289 				 */
6290 				valoffs += sizeof(uintptr_t);
6291 
6292 				/*
6293 				 * Reset to the memory address rather than
6294 				 * the memref array, then let the BYREF
6295 				 * code below do the work to store the
6296 				 * memory data in the buffer.
6297 				 */
6298 				val = memref[0];
6299 				break;
6300 			}
6301 
6302 			case DTRACEACT_PRINTT: {
6303 				/* The DIF returns a 'typeref'. */
6304 				uintptr_t *typeref = (uintptr_t *)(uintptr_t) val;
6305 				char c = '\0' + 1;
6306 				size_t s;
6307 
6308 				/*
6309 				 * Get the type string length and round it
6310 				 * up so that the data that follows is
6311 				 * aligned for easy access.
6312 				 */
6313 				size_t typs = strlen((char *) typeref[2]) + 1;
6314 				typs = roundup(typs,  sizeof(uintptr_t));
6315 
6316 				/*
6317 				 *Get the size from the typeref using the
6318 				 * number of elements and the type size.
6319 				 */
6320 				size = typeref[1] * typeref[3];
6321 
6322 				/*
6323 				 * Check if the size exceeds the allocated
6324 				 * buffer size.
6325 				 */
6326 				if (size + typs + 2 * sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) {
6327 					/* Flag a drop! */
6328 					*flags |= CPU_DTRACE_DROP;
6329 
6330 				}
6331 
6332 				/* Store the size in the buffer first. */
6333 				DTRACE_STORE(uintptr_t, tomax,
6334 				    valoffs, size);
6335 				valoffs += sizeof(uintptr_t);
6336 
6337 				/* Store the type size in the buffer. */
6338 				DTRACE_STORE(uintptr_t, tomax,
6339 				    valoffs, typeref[3]);
6340 				valoffs += sizeof(uintptr_t);
6341 
6342 				val = typeref[2];
6343 
6344 				for (s = 0; s < typs; s++) {
6345 					if (c != '\0')
6346 						c = dtrace_load8(val++);
6347 
6348 					DTRACE_STORE(uint8_t, tomax,
6349 					    valoffs++, c);
6350 				}
6351 
6352 				/*
6353 				 * Reset to the memory address rather than
6354 				 * the typeref array, then let the BYREF
6355 				 * code below do the work to store the
6356 				 * memory data in the buffer.
6357 				 */
6358 				val = typeref[0];
6359 				break;
6360 			}
6361 
6362 			case DTRACEACT_CHILL:
6363 				if (dtrace_priv_kernel_destructive(state))
6364 					dtrace_action_chill(&mstate, val);
6365 				continue;
6366 
6367 			case DTRACEACT_RAISE:
6368 				if (dtrace_priv_proc_destructive(state))
6369 					dtrace_action_raise(val);
6370 				continue;
6371 
6372 			case DTRACEACT_COMMIT:
6373 				ASSERT(!committed);
6374 
6375 				/*
6376 				 * We need to commit our buffer state.
6377 				 */
6378 				if (ecb->dte_size)
6379 					buf->dtb_offset = offs + ecb->dte_size;
6380 				buf = &state->dts_buffer[cpuid];
6381 				dtrace_speculation_commit(state, cpuid, val);
6382 				committed = 1;
6383 				continue;
6384 
6385 			case DTRACEACT_DISCARD:
6386 				dtrace_speculation_discard(state, cpuid, val);
6387 				continue;
6388 
6389 			case DTRACEACT_DIFEXPR:
6390 			case DTRACEACT_LIBACT:
6391 			case DTRACEACT_PRINTF:
6392 			case DTRACEACT_PRINTA:
6393 			case DTRACEACT_SYSTEM:
6394 			case DTRACEACT_FREOPEN:
6395 				break;
6396 
6397 			case DTRACEACT_SYM:
6398 			case DTRACEACT_MOD:
6399 				if (!dtrace_priv_kernel(state))
6400 					continue;
6401 				break;
6402 
6403 			case DTRACEACT_USYM:
6404 			case DTRACEACT_UMOD:
6405 			case DTRACEACT_UADDR: {
6406 #if defined(sun)
6407 				struct pid *pid = curthread->t_procp->p_pidp;
6408 #endif
6409 
6410 				if (!dtrace_priv_proc(state))
6411 					continue;
6412 
6413 				DTRACE_STORE(uint64_t, tomax,
6414 #if defined(sun)
6415 				    valoffs, (uint64_t)pid->pid_id);
6416 #else
6417 				    valoffs, (uint64_t) curproc->p_pid);
6418 #endif
6419 				DTRACE_STORE(uint64_t, tomax,
6420 				    valoffs + sizeof (uint64_t), val);
6421 
6422 				continue;
6423 			}
6424 
6425 			case DTRACEACT_EXIT: {
6426 				/*
6427 				 * For the exit action, we are going to attempt
6428 				 * to atomically set our activity to be
6429 				 * draining.  If this fails (either because
6430 				 * another CPU has beat us to the exit action,
6431 				 * or because our current activity is something
6432 				 * other than ACTIVE or WARMUP), we will
6433 				 * continue.  This assures that the exit action
6434 				 * can be successfully recorded at most once
6435 				 * when we're in the ACTIVE state.  If we're
6436 				 * encountering the exit() action while in
6437 				 * COOLDOWN, however, we want to honor the new
6438 				 * status code.  (We know that we're the only
6439 				 * thread in COOLDOWN, so there is no race.)
6440 				 */
6441 				void *activity = &state->dts_activity;
6442 				dtrace_activity_t current = state->dts_activity;
6443 
6444 				if (current == DTRACE_ACTIVITY_COOLDOWN)
6445 					break;
6446 
6447 				if (current != DTRACE_ACTIVITY_WARMUP)
6448 					current = DTRACE_ACTIVITY_ACTIVE;
6449 
6450 				if (dtrace_cas32(activity, current,
6451 				    DTRACE_ACTIVITY_DRAINING) != current) {
6452 					*flags |= CPU_DTRACE_DROP;
6453 					continue;
6454 				}
6455 
6456 				break;
6457 			}
6458 
6459 			default:
6460 				ASSERT(0);
6461 			}
6462 
6463 			if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) {
6464 				uintptr_t end = valoffs + size;
6465 
6466 				if (!dtrace_vcanload((void *)(uintptr_t)val,
6467 				    &dp->dtdo_rtype, &mstate, vstate))
6468 					continue;
6469 
6470 				/*
6471 				 * If this is a string, we're going to only
6472 				 * load until we find the zero byte -- after
6473 				 * which we'll store zero bytes.
6474 				 */
6475 				if (dp->dtdo_rtype.dtdt_kind ==
6476 				    DIF_TYPE_STRING) {
6477 					char c = '\0' + 1;
6478 					int intuple = act->dta_intuple;
6479 					size_t s;
6480 
6481 					for (s = 0; s < size; s++) {
6482 						if (c != '\0')
6483 							c = dtrace_load8(val++);
6484 
6485 						DTRACE_STORE(uint8_t, tomax,
6486 						    valoffs++, c);
6487 
6488 						if (c == '\0' && intuple)
6489 							break;
6490 					}
6491 
6492 					continue;
6493 				}
6494 
6495 				while (valoffs < end) {
6496 					DTRACE_STORE(uint8_t, tomax, valoffs++,
6497 					    dtrace_load8(val++));
6498 				}
6499 
6500 				continue;
6501 			}
6502 
6503 			switch (size) {
6504 			case 0:
6505 				break;
6506 
6507 			case sizeof (uint8_t):
6508 				DTRACE_STORE(uint8_t, tomax, valoffs, val);
6509 				break;
6510 			case sizeof (uint16_t):
6511 				DTRACE_STORE(uint16_t, tomax, valoffs, val);
6512 				break;
6513 			case sizeof (uint32_t):
6514 				DTRACE_STORE(uint32_t, tomax, valoffs, val);
6515 				break;
6516 			case sizeof (uint64_t):
6517 				DTRACE_STORE(uint64_t, tomax, valoffs, val);
6518 				break;
6519 			default:
6520 				/*
6521 				 * Any other size should have been returned by
6522 				 * reference, not by value.
6523 				 */
6524 				ASSERT(0);
6525 				break;
6526 			}
6527 		}
6528 
6529 		if (*flags & CPU_DTRACE_DROP)
6530 			continue;
6531 
6532 		if (*flags & CPU_DTRACE_FAULT) {
6533 			int ndx;
6534 			dtrace_action_t *err;
6535 
6536 			buf->dtb_errors++;
6537 
6538 			if (probe->dtpr_id == dtrace_probeid_error) {
6539 				/*
6540 				 * There's nothing we can do -- we had an
6541 				 * error on the error probe.  We bump an
6542 				 * error counter to at least indicate that
6543 				 * this condition happened.
6544 				 */
6545 				dtrace_error(&state->dts_dblerrors);
6546 				continue;
6547 			}
6548 
6549 			if (vtime) {
6550 				/*
6551 				 * Before recursing on dtrace_probe(), we
6552 				 * need to explicitly clear out our start
6553 				 * time to prevent it from being accumulated
6554 				 * into t_dtrace_vtime.
6555 				 */
6556 				curthread->t_dtrace_start = 0;
6557 			}
6558 
6559 			/*
6560 			 * Iterate over the actions to figure out which action
6561 			 * we were processing when we experienced the error.
6562 			 * Note that act points _past_ the faulting action; if
6563 			 * act is ecb->dte_action, the fault was in the
6564 			 * predicate, if it's ecb->dte_action->dta_next it's
6565 			 * in action #1, and so on.
6566 			 */
6567 			for (err = ecb->dte_action, ndx = 0;
6568 			    err != act; err = err->dta_next, ndx++)
6569 				continue;
6570 
6571 			dtrace_probe_error(state, ecb->dte_epid, ndx,
6572 			    (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ?
6573 			    mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags),
6574 			    cpu_core[cpuid].cpuc_dtrace_illval);
6575 
6576 			continue;
6577 		}
6578 
6579 		if (!committed)
6580 			buf->dtb_offset = offs + ecb->dte_size;
6581 	}
6582 
6583 	if (vtime)
6584 		curthread->t_dtrace_start = dtrace_gethrtime();
6585 
6586 	dtrace_interrupt_enable(cookie);
6587 }
6588 
6589 /*
6590  * DTrace Probe Hashing Functions
6591  *
6592  * The functions in this section (and indeed, the functions in remaining
6593  * sections) are not _called_ from probe context.  (Any exceptions to this are
6594  * marked with a "Note:".)  Rather, they are called from elsewhere in the
6595  * DTrace framework to look-up probes in, add probes to and remove probes from
6596  * the DTrace probe hashes.  (Each probe is hashed by each element of the
6597  * probe tuple -- allowing for fast lookups, regardless of what was
6598  * specified.)
6599  */
6600 static uint_t
6601 dtrace_hash_str(const char *p)
6602 {
6603 	unsigned int g;
6604 	uint_t hval = 0;
6605 
6606 	while (*p) {
6607 		hval = (hval << 4) + *p++;
6608 		if ((g = (hval & 0xf0000000)) != 0)
6609 			hval ^= g >> 24;
6610 		hval &= ~g;
6611 	}
6612 	return (hval);
6613 }
6614 
6615 static dtrace_hash_t *
6616 dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs)
6617 {
6618 	dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP);
6619 
6620 	hash->dth_stroffs = stroffs;
6621 	hash->dth_nextoffs = nextoffs;
6622 	hash->dth_prevoffs = prevoffs;
6623 
6624 	hash->dth_size = 1;
6625 	hash->dth_mask = hash->dth_size - 1;
6626 
6627 	hash->dth_tab = kmem_zalloc(hash->dth_size *
6628 	    sizeof (dtrace_hashbucket_t *), KM_SLEEP);
6629 
6630 	return (hash);
6631 }
6632 
6633 static void
6634 dtrace_hash_destroy(dtrace_hash_t *hash)
6635 {
6636 #ifdef DEBUG
6637 	int i;
6638 
6639 	for (i = 0; i < hash->dth_size; i++)
6640 		ASSERT(hash->dth_tab[i] == NULL);
6641 #endif
6642 
6643 	kmem_free(hash->dth_tab,
6644 	    hash->dth_size * sizeof (dtrace_hashbucket_t *));
6645 	kmem_free(hash, sizeof (dtrace_hash_t));
6646 }
6647 
6648 static void
6649 dtrace_hash_resize(dtrace_hash_t *hash)
6650 {
6651 	int size = hash->dth_size, i, ndx;
6652 	int new_size = hash->dth_size << 1;
6653 	int new_mask = new_size - 1;
6654 	dtrace_hashbucket_t **new_tab, *bucket, *next;
6655 
6656 	ASSERT((new_size & new_mask) == 0);
6657 
6658 	new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP);
6659 
6660 	for (i = 0; i < size; i++) {
6661 		for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) {
6662 			dtrace_probe_t *probe = bucket->dthb_chain;
6663 
6664 			ASSERT(probe != NULL);
6665 			ndx = DTRACE_HASHSTR(hash, probe) & new_mask;
6666 
6667 			next = bucket->dthb_next;
6668 			bucket->dthb_next = new_tab[ndx];
6669 			new_tab[ndx] = bucket;
6670 		}
6671 	}
6672 
6673 	kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *));
6674 	hash->dth_tab = new_tab;
6675 	hash->dth_size = new_size;
6676 	hash->dth_mask = new_mask;
6677 }
6678 
6679 static void
6680 dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new)
6681 {
6682 	int hashval = DTRACE_HASHSTR(hash, new);
6683 	int ndx = hashval & hash->dth_mask;
6684 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6685 	dtrace_probe_t **nextp, **prevp;
6686 
6687 	for (; bucket != NULL; bucket = bucket->dthb_next) {
6688 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new))
6689 			goto add;
6690 	}
6691 
6692 	if ((hash->dth_nbuckets >> 1) > hash->dth_size) {
6693 		dtrace_hash_resize(hash);
6694 		dtrace_hash_add(hash, new);
6695 		return;
6696 	}
6697 
6698 	bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP);
6699 	bucket->dthb_next = hash->dth_tab[ndx];
6700 	hash->dth_tab[ndx] = bucket;
6701 	hash->dth_nbuckets++;
6702 
6703 add:
6704 	nextp = DTRACE_HASHNEXT(hash, new);
6705 	ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL);
6706 	*nextp = bucket->dthb_chain;
6707 
6708 	if (bucket->dthb_chain != NULL) {
6709 		prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain);
6710 		ASSERT(*prevp == NULL);
6711 		*prevp = new;
6712 	}
6713 
6714 	bucket->dthb_chain = new;
6715 	bucket->dthb_len++;
6716 }
6717 
6718 static dtrace_probe_t *
6719 dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template)
6720 {
6721 	int hashval = DTRACE_HASHSTR(hash, template);
6722 	int ndx = hashval & hash->dth_mask;
6723 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6724 
6725 	for (; bucket != NULL; bucket = bucket->dthb_next) {
6726 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
6727 			return (bucket->dthb_chain);
6728 	}
6729 
6730 	return (NULL);
6731 }
6732 
6733 static int
6734 dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template)
6735 {
6736 	int hashval = DTRACE_HASHSTR(hash, template);
6737 	int ndx = hashval & hash->dth_mask;
6738 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6739 
6740 	for (; bucket != NULL; bucket = bucket->dthb_next) {
6741 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
6742 			return (bucket->dthb_len);
6743 	}
6744 
6745 	return (0);
6746 }
6747 
6748 static void
6749 dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe)
6750 {
6751 	int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask;
6752 	dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6753 
6754 	dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe);
6755 	dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe);
6756 
6757 	/*
6758 	 * Find the bucket that we're removing this probe from.
6759 	 */
6760 	for (; bucket != NULL; bucket = bucket->dthb_next) {
6761 		if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe))
6762 			break;
6763 	}
6764 
6765 	ASSERT(bucket != NULL);
6766 
6767 	if (*prevp == NULL) {
6768 		if (*nextp == NULL) {
6769 			/*
6770 			 * The removed probe was the only probe on this
6771 			 * bucket; we need to remove the bucket.
6772 			 */
6773 			dtrace_hashbucket_t *b = hash->dth_tab[ndx];
6774 
6775 			ASSERT(bucket->dthb_chain == probe);
6776 			ASSERT(b != NULL);
6777 
6778 			if (b == bucket) {
6779 				hash->dth_tab[ndx] = bucket->dthb_next;
6780 			} else {
6781 				while (b->dthb_next != bucket)
6782 					b = b->dthb_next;
6783 				b->dthb_next = bucket->dthb_next;
6784 			}
6785 
6786 			ASSERT(hash->dth_nbuckets > 0);
6787 			hash->dth_nbuckets--;
6788 			kmem_free(bucket, sizeof (dtrace_hashbucket_t));
6789 			return;
6790 		}
6791 
6792 		bucket->dthb_chain = *nextp;
6793 	} else {
6794 		*(DTRACE_HASHNEXT(hash, *prevp)) = *nextp;
6795 	}
6796 
6797 	if (*nextp != NULL)
6798 		*(DTRACE_HASHPREV(hash, *nextp)) = *prevp;
6799 }
6800 
6801 /*
6802  * DTrace Utility Functions
6803  *
6804  * These are random utility functions that are _not_ called from probe context.
6805  */
6806 static int
6807 dtrace_badattr(const dtrace_attribute_t *a)
6808 {
6809 	return (a->dtat_name > DTRACE_STABILITY_MAX ||
6810 	    a->dtat_data > DTRACE_STABILITY_MAX ||
6811 	    a->dtat_class > DTRACE_CLASS_MAX);
6812 }
6813 
6814 /*
6815  * Return a duplicate copy of a string.  If the specified string is NULL,
6816  * this function returns a zero-length string.
6817  */
6818 static char *
6819 dtrace_strdup(const char *str)
6820 {
6821 	char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP);
6822 
6823 	if (str != NULL)
6824 		(void) strcpy(new, str);
6825 
6826 	return (new);
6827 }
6828 
6829 #define	DTRACE_ISALPHA(c)	\
6830 	(((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z'))
6831 
6832 static int
6833 dtrace_badname(const char *s)
6834 {
6835 	char c;
6836 
6837 	if (s == NULL || (c = *s++) == '\0')
6838 		return (0);
6839 
6840 	if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.')
6841 		return (1);
6842 
6843 	while ((c = *s++) != '\0') {
6844 		if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') &&
6845 		    c != '-' && c != '_' && c != '.' && c != '`')
6846 			return (1);
6847 	}
6848 
6849 	return (0);
6850 }
6851 
6852 static void
6853 dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp)
6854 {
6855 	uint32_t priv;
6856 
6857 #if defined(sun)
6858 	if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
6859 		/*
6860 		 * For DTRACE_PRIV_ALL, the uid and zoneid don't matter.
6861 		 */
6862 		priv = DTRACE_PRIV_ALL;
6863 	} else {
6864 		*uidp = crgetuid(cr);
6865 		*zoneidp = crgetzoneid(cr);
6866 
6867 		priv = 0;
6868 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE))
6869 			priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER;
6870 		else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE))
6871 			priv |= DTRACE_PRIV_USER;
6872 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE))
6873 			priv |= DTRACE_PRIV_PROC;
6874 		if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
6875 			priv |= DTRACE_PRIV_OWNER;
6876 		if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
6877 			priv |= DTRACE_PRIV_ZONEOWNER;
6878 	}
6879 #else
6880 	priv = DTRACE_PRIV_ALL;
6881 #endif
6882 
6883 	*privp = priv;
6884 }
6885 
6886 #ifdef DTRACE_ERRDEBUG
6887 static void
6888 dtrace_errdebug(const char *str)
6889 {
6890 	int hval = dtrace_hash_str(str) % DTRACE_ERRHASHSZ;
6891 	int occupied = 0;
6892 
6893 	mutex_enter(&dtrace_errlock);
6894 	dtrace_errlast = str;
6895 	dtrace_errthread = curthread;
6896 
6897 	while (occupied++ < DTRACE_ERRHASHSZ) {
6898 		if (dtrace_errhash[hval].dter_msg == str) {
6899 			dtrace_errhash[hval].dter_count++;
6900 			goto out;
6901 		}
6902 
6903 		if (dtrace_errhash[hval].dter_msg != NULL) {
6904 			hval = (hval + 1) % DTRACE_ERRHASHSZ;
6905 			continue;
6906 		}
6907 
6908 		dtrace_errhash[hval].dter_msg = str;
6909 		dtrace_errhash[hval].dter_count = 1;
6910 		goto out;
6911 	}
6912 
6913 	panic("dtrace: undersized error hash");
6914 out:
6915 	mutex_exit(&dtrace_errlock);
6916 }
6917 #endif
6918 
6919 /*
6920  * DTrace Matching Functions
6921  *
6922  * These functions are used to match groups of probes, given some elements of
6923  * a probe tuple, or some globbed expressions for elements of a probe tuple.
6924  */
6925 static int
6926 dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid,
6927     zoneid_t zoneid)
6928 {
6929 	if (priv != DTRACE_PRIV_ALL) {
6930 		uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags;
6931 		uint32_t match = priv & ppriv;
6932 
6933 		/*
6934 		 * No PRIV_DTRACE_* privileges...
6935 		 */
6936 		if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER |
6937 		    DTRACE_PRIV_KERNEL)) == 0)
6938 			return (0);
6939 
6940 		/*
6941 		 * No matching bits, but there were bits to match...
6942 		 */
6943 		if (match == 0 && ppriv != 0)
6944 			return (0);
6945 
6946 		/*
6947 		 * Need to have permissions to the process, but don't...
6948 		 */
6949 		if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 &&
6950 		    uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) {
6951 			return (0);
6952 		}
6953 
6954 		/*
6955 		 * Need to be in the same zone unless we possess the
6956 		 * privilege to examine all zones.
6957 		 */
6958 		if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 &&
6959 		    zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) {
6960 			return (0);
6961 		}
6962 	}
6963 
6964 	return (1);
6965 }
6966 
6967 /*
6968  * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which
6969  * consists of input pattern strings and an ops-vector to evaluate them.
6970  * This function returns >0 for match, 0 for no match, and <0 for error.
6971  */
6972 static int
6973 dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp,
6974     uint32_t priv, uid_t uid, zoneid_t zoneid)
6975 {
6976 	dtrace_provider_t *pvp = prp->dtpr_provider;
6977 	int rv;
6978 
6979 	if (pvp->dtpv_defunct)
6980 		return (0);
6981 
6982 	if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0)
6983 		return (rv);
6984 
6985 	if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0)
6986 		return (rv);
6987 
6988 	if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0)
6989 		return (rv);
6990 
6991 	if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0)
6992 		return (rv);
6993 
6994 	if (dtrace_match_priv(prp, priv, uid, zoneid) == 0)
6995 		return (0);
6996 
6997 	return (rv);
6998 }
6999 
7000 /*
7001  * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN)
7002  * interface for matching a glob pattern 'p' to an input string 's'.  Unlike
7003  * libc's version, the kernel version only applies to 8-bit ASCII strings.
7004  * In addition, all of the recursion cases except for '*' matching have been
7005  * unwound.  For '*', we still implement recursive evaluation, but a depth
7006  * counter is maintained and matching is aborted if we recurse too deep.
7007  * The function returns 0 if no match, >0 if match, and <0 if recursion error.
7008  */
7009 static int
7010 dtrace_match_glob(const char *s, const char *p, int depth)
7011 {
7012 	const char *olds;
7013 	char s1, c;
7014 	int gs;
7015 
7016 	if (depth > DTRACE_PROBEKEY_MAXDEPTH)
7017 		return (-1);
7018 
7019 	if (s == NULL)
7020 		s = ""; /* treat NULL as empty string */
7021 
7022 top:
7023 	olds = s;
7024 	s1 = *s++;
7025 
7026 	if (p == NULL)
7027 		return (0);
7028 
7029 	if ((c = *p++) == '\0')
7030 		return (s1 == '\0');
7031 
7032 	switch (c) {
7033 	case '[': {
7034 		int ok = 0, notflag = 0;
7035 		char lc = '\0';
7036 
7037 		if (s1 == '\0')
7038 			return (0);
7039 
7040 		if (*p == '!') {
7041 			notflag = 1;
7042 			p++;
7043 		}
7044 
7045 		if ((c = *p++) == '\0')
7046 			return (0);
7047 
7048 		do {
7049 			if (c == '-' && lc != '\0' && *p != ']') {
7050 				if ((c = *p++) == '\0')
7051 					return (0);
7052 				if (c == '\\' && (c = *p++) == '\0')
7053 					return (0);
7054 
7055 				if (notflag) {
7056 					if (s1 < lc || s1 > c)
7057 						ok++;
7058 					else
7059 						return (0);
7060 				} else if (lc <= s1 && s1 <= c)
7061 					ok++;
7062 
7063 			} else if (c == '\\' && (c = *p++) == '\0')
7064 				return (0);
7065 
7066 			lc = c; /* save left-hand 'c' for next iteration */
7067 
7068 			if (notflag) {
7069 				if (s1 != c)
7070 					ok++;
7071 				else
7072 					return (0);
7073 			} else if (s1 == c)
7074 				ok++;
7075 
7076 			if ((c = *p++) == '\0')
7077 				return (0);
7078 
7079 		} while (c != ']');
7080 
7081 		if (ok)
7082 			goto top;
7083 
7084 		return (0);
7085 	}
7086 
7087 	case '\\':
7088 		if ((c = *p++) == '\0')
7089 			return (0);
7090 		/*FALLTHRU*/
7091 
7092 	default:
7093 		if (c != s1)
7094 			return (0);
7095 		/*FALLTHRU*/
7096 
7097 	case '?':
7098 		if (s1 != '\0')
7099 			goto top;
7100 		return (0);
7101 
7102 	case '*':
7103 		while (*p == '*')
7104 			p++; /* consecutive *'s are identical to a single one */
7105 
7106 		if (*p == '\0')
7107 			return (1);
7108 
7109 		for (s = olds; *s != '\0'; s++) {
7110 			if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0)
7111 				return (gs);
7112 		}
7113 
7114 		return (0);
7115 	}
7116 }
7117 
7118 /*ARGSUSED*/
7119 static int
7120 dtrace_match_string(const char *s, const char *p, int depth)
7121 {
7122 	return (s != NULL && strcmp(s, p) == 0);
7123 }
7124 
7125 /*ARGSUSED*/
7126 static int
7127 dtrace_match_nul(const char *s, const char *p, int depth)
7128 {
7129 	return (1); /* always match the empty pattern */
7130 }
7131 
7132 /*ARGSUSED*/
7133 static int
7134 dtrace_match_nonzero(const char *s, const char *p, int depth)
7135 {
7136 	return (s != NULL && s[0] != '\0');
7137 }
7138 
7139 static int
7140 dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid,
7141     zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg)
7142 {
7143 	dtrace_probe_t template, *probe;
7144 	dtrace_hash_t *hash = NULL;
7145 	int len, best = INT_MAX, nmatched = 0;
7146 	dtrace_id_t i;
7147 
7148 	ASSERT(MUTEX_HELD(&dtrace_lock));
7149 
7150 	/*
7151 	 * If the probe ID is specified in the key, just lookup by ID and
7152 	 * invoke the match callback once if a matching probe is found.
7153 	 */
7154 	if (pkp->dtpk_id != DTRACE_IDNONE) {
7155 		if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL &&
7156 		    dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) {
7157 			(void) (*matched)(probe, arg);
7158 			nmatched++;
7159 		}
7160 		return (nmatched);
7161 	}
7162 
7163 	template.dtpr_mod = (char *)pkp->dtpk_mod;
7164 	template.dtpr_func = (char *)pkp->dtpk_func;
7165 	template.dtpr_name = (char *)pkp->dtpk_name;
7166 
7167 	/*
7168 	 * We want to find the most distinct of the module name, function
7169 	 * name, and name.  So for each one that is not a glob pattern or
7170 	 * empty string, we perform a lookup in the corresponding hash and
7171 	 * use the hash table with the fewest collisions to do our search.
7172 	 */
7173 	if (pkp->dtpk_mmatch == &dtrace_match_string &&
7174 	    (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) {
7175 		best = len;
7176 		hash = dtrace_bymod;
7177 	}
7178 
7179 	if (pkp->dtpk_fmatch == &dtrace_match_string &&
7180 	    (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) {
7181 		best = len;
7182 		hash = dtrace_byfunc;
7183 	}
7184 
7185 	if (pkp->dtpk_nmatch == &dtrace_match_string &&
7186 	    (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) {
7187 		best = len;
7188 		hash = dtrace_byname;
7189 	}
7190 
7191 	/*
7192 	 * If we did not select a hash table, iterate over every probe and
7193 	 * invoke our callback for each one that matches our input probe key.
7194 	 */
7195 	if (hash == NULL) {
7196 		for (i = 0; i < dtrace_nprobes; i++) {
7197 			if ((probe = dtrace_probes[i]) == NULL ||
7198 			    dtrace_match_probe(probe, pkp, priv, uid,
7199 			    zoneid) <= 0)
7200 				continue;
7201 
7202 			nmatched++;
7203 
7204 			if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT)
7205 				break;
7206 		}
7207 
7208 		return (nmatched);
7209 	}
7210 
7211 	/*
7212 	 * If we selected a hash table, iterate over each probe of the same key
7213 	 * name and invoke the callback for every probe that matches the other
7214 	 * attributes of our input probe key.
7215 	 */
7216 	for (probe = dtrace_hash_lookup(hash, &template); probe != NULL;
7217 	    probe = *(DTRACE_HASHNEXT(hash, probe))) {
7218 
7219 		if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0)
7220 			continue;
7221 
7222 		nmatched++;
7223 
7224 		if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT)
7225 			break;
7226 	}
7227 
7228 	return (nmatched);
7229 }
7230 
7231 /*
7232  * Return the function pointer dtrace_probecmp() should use to compare the
7233  * specified pattern with a string.  For NULL or empty patterns, we select
7234  * dtrace_match_nul().  For glob pattern strings, we use dtrace_match_glob().
7235  * For non-empty non-glob strings, we use dtrace_match_string().
7236  */
7237 static dtrace_probekey_f *
7238 dtrace_probekey_func(const char *p)
7239 {
7240 	char c;
7241 
7242 	if (p == NULL || *p == '\0')
7243 		return (&dtrace_match_nul);
7244 
7245 	while ((c = *p++) != '\0') {
7246 		if (c == '[' || c == '?' || c == '*' || c == '\\')
7247 			return (&dtrace_match_glob);
7248 	}
7249 
7250 	return (&dtrace_match_string);
7251 }
7252 
7253 /*
7254  * Build a probe comparison key for use with dtrace_match_probe() from the
7255  * given probe description.  By convention, a null key only matches anchored
7256  * probes: if each field is the empty string, reset dtpk_fmatch to
7257  * dtrace_match_nonzero().
7258  */
7259 static void
7260 dtrace_probekey(dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp)
7261 {
7262 	pkp->dtpk_prov = pdp->dtpd_provider;
7263 	pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider);
7264 
7265 	pkp->dtpk_mod = pdp->dtpd_mod;
7266 	pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod);
7267 
7268 	pkp->dtpk_func = pdp->dtpd_func;
7269 	pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func);
7270 
7271 	pkp->dtpk_name = pdp->dtpd_name;
7272 	pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name);
7273 
7274 	pkp->dtpk_id = pdp->dtpd_id;
7275 
7276 	if (pkp->dtpk_id == DTRACE_IDNONE &&
7277 	    pkp->dtpk_pmatch == &dtrace_match_nul &&
7278 	    pkp->dtpk_mmatch == &dtrace_match_nul &&
7279 	    pkp->dtpk_fmatch == &dtrace_match_nul &&
7280 	    pkp->dtpk_nmatch == &dtrace_match_nul)
7281 		pkp->dtpk_fmatch = &dtrace_match_nonzero;
7282 }
7283 
7284 /*
7285  * DTrace Provider-to-Framework API Functions
7286  *
7287  * These functions implement much of the Provider-to-Framework API, as
7288  * described in <sys/dtrace.h>.  The parts of the API not in this section are
7289  * the functions in the API for probe management (found below), and
7290  * dtrace_probe() itself (found above).
7291  */
7292 
7293 /*
7294  * Register the calling provider with the DTrace framework.  This should
7295  * generally be called by DTrace providers in their attach(9E) entry point.
7296  */
7297 int
7298 dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv,
7299     cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp)
7300 {
7301 	dtrace_provider_t *provider;
7302 
7303 	if (name == NULL || pap == NULL || pops == NULL || idp == NULL) {
7304 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7305 		    "arguments", name ? name : "<NULL>");
7306 		return (EINVAL);
7307 	}
7308 
7309 	if (name[0] == '\0' || dtrace_badname(name)) {
7310 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7311 		    "provider name", name);
7312 		return (EINVAL);
7313 	}
7314 
7315 	if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) ||
7316 	    pops->dtps_enable == NULL || pops->dtps_disable == NULL ||
7317 	    pops->dtps_destroy == NULL ||
7318 	    ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) {
7319 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7320 		    "provider ops", name);
7321 		return (EINVAL);
7322 	}
7323 
7324 	if (dtrace_badattr(&pap->dtpa_provider) ||
7325 	    dtrace_badattr(&pap->dtpa_mod) ||
7326 	    dtrace_badattr(&pap->dtpa_func) ||
7327 	    dtrace_badattr(&pap->dtpa_name) ||
7328 	    dtrace_badattr(&pap->dtpa_args)) {
7329 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7330 		    "provider attributes", name);
7331 		return (EINVAL);
7332 	}
7333 
7334 	if (priv & ~DTRACE_PRIV_ALL) {
7335 		cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7336 		    "privilege attributes", name);
7337 		return (EINVAL);
7338 	}
7339 
7340 	if ((priv & DTRACE_PRIV_KERNEL) &&
7341 	    (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) &&
7342 	    pops->dtps_usermode == NULL) {
7343 		cmn_err(CE_WARN, "failed to register provider '%s': need "
7344 		    "dtps_usermode() op for given privilege attributes", name);
7345 		return (EINVAL);
7346 	}
7347 
7348 	provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP);
7349 	provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
7350 	(void) strcpy(provider->dtpv_name, name);
7351 
7352 	provider->dtpv_attr = *pap;
7353 	provider->dtpv_priv.dtpp_flags = priv;
7354 	if (cr != NULL) {
7355 		provider->dtpv_priv.dtpp_uid = crgetuid(cr);
7356 		provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr);
7357 	}
7358 	provider->dtpv_pops = *pops;
7359 
7360 	if (pops->dtps_provide == NULL) {
7361 		ASSERT(pops->dtps_provide_module != NULL);
7362 		provider->dtpv_pops.dtps_provide =
7363 		    (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop;
7364 	}
7365 
7366 	if (pops->dtps_provide_module == NULL) {
7367 		ASSERT(pops->dtps_provide != NULL);
7368 		provider->dtpv_pops.dtps_provide_module =
7369 		    (void (*)(void *, modctl_t *))dtrace_nullop;
7370 	}
7371 
7372 	if (pops->dtps_suspend == NULL) {
7373 		ASSERT(pops->dtps_resume == NULL);
7374 		provider->dtpv_pops.dtps_suspend =
7375 		    (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
7376 		provider->dtpv_pops.dtps_resume =
7377 		    (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
7378 	}
7379 
7380 	provider->dtpv_arg = arg;
7381 	*idp = (dtrace_provider_id_t)provider;
7382 
7383 	if (pops == &dtrace_provider_ops) {
7384 		ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7385 		ASSERT(MUTEX_HELD(&dtrace_lock));
7386 		ASSERT(dtrace_anon.dta_enabling == NULL);
7387 
7388 		/*
7389 		 * We make sure that the DTrace provider is at the head of
7390 		 * the provider chain.
7391 		 */
7392 		provider->dtpv_next = dtrace_provider;
7393 		dtrace_provider = provider;
7394 		return (0);
7395 	}
7396 
7397 	mutex_enter(&dtrace_provider_lock);
7398 	mutex_enter(&dtrace_lock);
7399 
7400 	/*
7401 	 * If there is at least one provider registered, we'll add this
7402 	 * provider after the first provider.
7403 	 */
7404 	if (dtrace_provider != NULL) {
7405 		provider->dtpv_next = dtrace_provider->dtpv_next;
7406 		dtrace_provider->dtpv_next = provider;
7407 	} else {
7408 		dtrace_provider = provider;
7409 	}
7410 
7411 	if (dtrace_retained != NULL) {
7412 		dtrace_enabling_provide(provider);
7413 
7414 		/*
7415 		 * Now we need to call dtrace_enabling_matchall() -- which
7416 		 * will acquire cpu_lock and dtrace_lock.  We therefore need
7417 		 * to drop all of our locks before calling into it...
7418 		 */
7419 		mutex_exit(&dtrace_lock);
7420 		mutex_exit(&dtrace_provider_lock);
7421 		dtrace_enabling_matchall();
7422 
7423 		return (0);
7424 	}
7425 
7426 	mutex_exit(&dtrace_lock);
7427 	mutex_exit(&dtrace_provider_lock);
7428 
7429 	return (0);
7430 }
7431 
7432 /*
7433  * Unregister the specified provider from the DTrace framework.  This should
7434  * generally be called by DTrace providers in their detach(9E) entry point.
7435  */
7436 int
7437 dtrace_unregister(dtrace_provider_id_t id)
7438 {
7439 	dtrace_provider_t *old = (dtrace_provider_t *)id;
7440 	dtrace_provider_t *prev = NULL;
7441 	int i, self = 0;
7442 	dtrace_probe_t *probe, *first = NULL;
7443 
7444 	if (old->dtpv_pops.dtps_enable ==
7445 	    (void (*)(void *, dtrace_id_t, void *))dtrace_nullop) {
7446 		/*
7447 		 * If DTrace itself is the provider, we're called with locks
7448 		 * already held.
7449 		 */
7450 		ASSERT(old == dtrace_provider);
7451 #if defined(sun)
7452 		ASSERT(dtrace_devi != NULL);
7453 #endif
7454 		ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7455 		ASSERT(MUTEX_HELD(&dtrace_lock));
7456 		self = 1;
7457 
7458 		if (dtrace_provider->dtpv_next != NULL) {
7459 			/*
7460 			 * There's another provider here; return failure.
7461 			 */
7462 			return (EBUSY);
7463 		}
7464 	} else {
7465 		mutex_enter(&dtrace_provider_lock);
7466 		mutex_enter(&mod_lock);
7467 		mutex_enter(&dtrace_lock);
7468 	}
7469 
7470 	/*
7471 	 * If anyone has /dev/dtrace open, or if there are anonymous enabled
7472 	 * probes, we refuse to let providers slither away, unless this
7473 	 * provider has already been explicitly invalidated.
7474 	 */
7475 	if (!old->dtpv_defunct &&
7476 	    (dtrace_opens || (dtrace_anon.dta_state != NULL &&
7477 	    dtrace_anon.dta_state->dts_necbs > 0))) {
7478 		if (!self) {
7479 			mutex_exit(&dtrace_lock);
7480 			mutex_exit(&mod_lock);
7481 			mutex_exit(&dtrace_provider_lock);
7482 		}
7483 		return (EBUSY);
7484 	}
7485 
7486 	/*
7487 	 * Attempt to destroy the probes associated with this provider.
7488 	 */
7489 	for (i = 0; i < dtrace_nprobes; i++) {
7490 		if ((probe = dtrace_probes[i]) == NULL)
7491 			continue;
7492 
7493 		if (probe->dtpr_provider != old)
7494 			continue;
7495 
7496 		if (probe->dtpr_ecb == NULL)
7497 			continue;
7498 
7499 		/*
7500 		 * We have at least one ECB; we can't remove this provider.
7501 		 */
7502 		if (!self) {
7503 			mutex_exit(&dtrace_lock);
7504 			mutex_exit(&mod_lock);
7505 			mutex_exit(&dtrace_provider_lock);
7506 		}
7507 		return (EBUSY);
7508 	}
7509 
7510 	/*
7511 	 * All of the probes for this provider are disabled; we can safely
7512 	 * remove all of them from their hash chains and from the probe array.
7513 	 */
7514 	for (i = 0; i < dtrace_nprobes; i++) {
7515 		if ((probe = dtrace_probes[i]) == NULL)
7516 			continue;
7517 
7518 		if (probe->dtpr_provider != old)
7519 			continue;
7520 
7521 		dtrace_probes[i] = NULL;
7522 
7523 		dtrace_hash_remove(dtrace_bymod, probe);
7524 		dtrace_hash_remove(dtrace_byfunc, probe);
7525 		dtrace_hash_remove(dtrace_byname, probe);
7526 
7527 		if (first == NULL) {
7528 			first = probe;
7529 			probe->dtpr_nextmod = NULL;
7530 		} else {
7531 			probe->dtpr_nextmod = first;
7532 			first = probe;
7533 		}
7534 	}
7535 
7536 	/*
7537 	 * The provider's probes have been removed from the hash chains and
7538 	 * from the probe array.  Now issue a dtrace_sync() to be sure that
7539 	 * everyone has cleared out from any probe array processing.
7540 	 */
7541 	dtrace_sync();
7542 
7543 	for (probe = first; probe != NULL; probe = first) {
7544 		first = probe->dtpr_nextmod;
7545 
7546 		old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id,
7547 		    probe->dtpr_arg);
7548 		kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
7549 		kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
7550 		kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
7551 #if defined(sun)
7552 		vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1);
7553 #else
7554 		free_unr(dtrace_arena, probe->dtpr_id);
7555 #endif
7556 		kmem_free(probe, sizeof (dtrace_probe_t));
7557 	}
7558 
7559 	if ((prev = dtrace_provider) == old) {
7560 #if defined(sun)
7561 		ASSERT(self || dtrace_devi == NULL);
7562 		ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL);
7563 #endif
7564 		dtrace_provider = old->dtpv_next;
7565 	} else {
7566 		while (prev != NULL && prev->dtpv_next != old)
7567 			prev = prev->dtpv_next;
7568 
7569 		if (prev == NULL) {
7570 			panic("attempt to unregister non-existent "
7571 			    "dtrace provider %p\n", (void *)id);
7572 		}
7573 
7574 		prev->dtpv_next = old->dtpv_next;
7575 	}
7576 
7577 	if (!self) {
7578 		mutex_exit(&dtrace_lock);
7579 		mutex_exit(&mod_lock);
7580 		mutex_exit(&dtrace_provider_lock);
7581 	}
7582 
7583 	kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1);
7584 	kmem_free(old, sizeof (dtrace_provider_t));
7585 
7586 	return (0);
7587 }
7588 
7589 /*
7590  * Invalidate the specified provider.  All subsequent probe lookups for the
7591  * specified provider will fail, but its probes will not be removed.
7592  */
7593 void
7594 dtrace_invalidate(dtrace_provider_id_t id)
7595 {
7596 	dtrace_provider_t *pvp = (dtrace_provider_t *)id;
7597 
7598 	ASSERT(pvp->dtpv_pops.dtps_enable !=
7599 	    (void (*)(void *, dtrace_id_t, void *))dtrace_nullop);
7600 
7601 	mutex_enter(&dtrace_provider_lock);
7602 	mutex_enter(&dtrace_lock);
7603 
7604 	pvp->dtpv_defunct = 1;
7605 
7606 	mutex_exit(&dtrace_lock);
7607 	mutex_exit(&dtrace_provider_lock);
7608 }
7609 
7610 /*
7611  * Indicate whether or not DTrace has attached.
7612  */
7613 int
7614 dtrace_attached(void)
7615 {
7616 	/*
7617 	 * dtrace_provider will be non-NULL iff the DTrace driver has
7618 	 * attached.  (It's non-NULL because DTrace is always itself a
7619 	 * provider.)
7620 	 */
7621 	return (dtrace_provider != NULL);
7622 }
7623 
7624 /*
7625  * Remove all the unenabled probes for the given provider.  This function is
7626  * not unlike dtrace_unregister(), except that it doesn't remove the provider
7627  * -- just as many of its associated probes as it can.
7628  */
7629 int
7630 dtrace_condense(dtrace_provider_id_t id)
7631 {
7632 	dtrace_provider_t *prov = (dtrace_provider_t *)id;
7633 	int i;
7634 	dtrace_probe_t *probe;
7635 
7636 	/*
7637 	 * Make sure this isn't the dtrace provider itself.
7638 	 */
7639 	ASSERT(prov->dtpv_pops.dtps_enable !=
7640 	    (void (*)(void *, dtrace_id_t, void *))dtrace_nullop);
7641 
7642 	mutex_enter(&dtrace_provider_lock);
7643 	mutex_enter(&dtrace_lock);
7644 
7645 	/*
7646 	 * Attempt to destroy the probes associated with this provider.
7647 	 */
7648 	for (i = 0; i < dtrace_nprobes; i++) {
7649 		if ((probe = dtrace_probes[i]) == NULL)
7650 			continue;
7651 
7652 		if (probe->dtpr_provider != prov)
7653 			continue;
7654 
7655 		if (probe->dtpr_ecb != NULL)
7656 			continue;
7657 
7658 		dtrace_probes[i] = NULL;
7659 
7660 		dtrace_hash_remove(dtrace_bymod, probe);
7661 		dtrace_hash_remove(dtrace_byfunc, probe);
7662 		dtrace_hash_remove(dtrace_byname, probe);
7663 
7664 		prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1,
7665 		    probe->dtpr_arg);
7666 		kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
7667 		kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
7668 		kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
7669 		kmem_free(probe, sizeof (dtrace_probe_t));
7670 #if defined(sun)
7671 		vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1);
7672 #else
7673 		free_unr(dtrace_arena, i + 1);
7674 #endif
7675 	}
7676 
7677 	mutex_exit(&dtrace_lock);
7678 	mutex_exit(&dtrace_provider_lock);
7679 
7680 	return (0);
7681 }
7682 
7683 /*
7684  * DTrace Probe Management Functions
7685  *
7686  * The functions in this section perform the DTrace probe management,
7687  * including functions to create probes, look-up probes, and call into the
7688  * providers to request that probes be provided.  Some of these functions are
7689  * in the Provider-to-Framework API; these functions can be identified by the
7690  * fact that they are not declared "static".
7691  */
7692 
7693 /*
7694  * Create a probe with the specified module name, function name, and name.
7695  */
7696 dtrace_id_t
7697 dtrace_probe_create(dtrace_provider_id_t prov, const char *mod,
7698     const char *func, const char *name, int aframes, void *arg)
7699 {
7700 	dtrace_probe_t *probe, **probes;
7701 	dtrace_provider_t *provider = (dtrace_provider_t *)prov;
7702 	dtrace_id_t id;
7703 
7704 	if (provider == dtrace_provider) {
7705 		ASSERT(MUTEX_HELD(&dtrace_lock));
7706 	} else {
7707 		mutex_enter(&dtrace_lock);
7708 	}
7709 
7710 #if defined(sun)
7711 	id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1,
7712 	    VM_BESTFIT | VM_SLEEP);
7713 #else
7714 	id = alloc_unr(dtrace_arena);
7715 #endif
7716 	probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP);
7717 
7718 	probe->dtpr_id = id;
7719 	probe->dtpr_gen = dtrace_probegen++;
7720 	probe->dtpr_mod = dtrace_strdup(mod);
7721 	probe->dtpr_func = dtrace_strdup(func);
7722 	probe->dtpr_name = dtrace_strdup(name);
7723 	probe->dtpr_arg = arg;
7724 	probe->dtpr_aframes = aframes;
7725 	probe->dtpr_provider = provider;
7726 
7727 	dtrace_hash_add(dtrace_bymod, probe);
7728 	dtrace_hash_add(dtrace_byfunc, probe);
7729 	dtrace_hash_add(dtrace_byname, probe);
7730 
7731 	if (id - 1 >= dtrace_nprobes) {
7732 		size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *);
7733 		size_t nsize = osize << 1;
7734 
7735 		if (nsize == 0) {
7736 			ASSERT(osize == 0);
7737 			ASSERT(dtrace_probes == NULL);
7738 			nsize = sizeof (dtrace_probe_t *);
7739 		}
7740 
7741 		probes = kmem_zalloc(nsize, KM_SLEEP);
7742 
7743 		if (dtrace_probes == NULL) {
7744 			ASSERT(osize == 0);
7745 			dtrace_probes = probes;
7746 			dtrace_nprobes = 1;
7747 		} else {
7748 			dtrace_probe_t **oprobes = dtrace_probes;
7749 
7750 			bcopy(oprobes, probes, osize);
7751 			dtrace_membar_producer();
7752 			dtrace_probes = probes;
7753 
7754 			dtrace_sync();
7755 
7756 			/*
7757 			 * All CPUs are now seeing the new probes array; we can
7758 			 * safely free the old array.
7759 			 */
7760 			kmem_free(oprobes, osize);
7761 			dtrace_nprobes <<= 1;
7762 		}
7763 
7764 		ASSERT(id - 1 < dtrace_nprobes);
7765 	}
7766 
7767 	ASSERT(dtrace_probes[id - 1] == NULL);
7768 	dtrace_probes[id - 1] = probe;
7769 
7770 	if (provider != dtrace_provider)
7771 		mutex_exit(&dtrace_lock);
7772 
7773 	return (id);
7774 }
7775 
7776 static dtrace_probe_t *
7777 dtrace_probe_lookup_id(dtrace_id_t id)
7778 {
7779 	ASSERT(MUTEX_HELD(&dtrace_lock));
7780 
7781 	if (id == 0 || id > dtrace_nprobes)
7782 		return (NULL);
7783 
7784 	return (dtrace_probes[id - 1]);
7785 }
7786 
7787 static int
7788 dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg)
7789 {
7790 	*((dtrace_id_t *)arg) = probe->dtpr_id;
7791 
7792 	return (DTRACE_MATCH_DONE);
7793 }
7794 
7795 /*
7796  * Look up a probe based on provider and one or more of module name, function
7797  * name and probe name.
7798  */
7799 dtrace_id_t
7800 dtrace_probe_lookup(dtrace_provider_id_t prid, char *mod,
7801     char *func, char *name)
7802 {
7803 	dtrace_probekey_t pkey;
7804 	dtrace_id_t id;
7805 	int match;
7806 
7807 	pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name;
7808 	pkey.dtpk_pmatch = &dtrace_match_string;
7809 	pkey.dtpk_mod = mod;
7810 	pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul;
7811 	pkey.dtpk_func = func;
7812 	pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul;
7813 	pkey.dtpk_name = name;
7814 	pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul;
7815 	pkey.dtpk_id = DTRACE_IDNONE;
7816 
7817 	mutex_enter(&dtrace_lock);
7818 	match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0,
7819 	    dtrace_probe_lookup_match, &id);
7820 	mutex_exit(&dtrace_lock);
7821 
7822 	ASSERT(match == 1 || match == 0);
7823 	return (match ? id : 0);
7824 }
7825 
7826 /*
7827  * Returns the probe argument associated with the specified probe.
7828  */
7829 void *
7830 dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid)
7831 {
7832 	dtrace_probe_t *probe;
7833 	void *rval = NULL;
7834 
7835 	mutex_enter(&dtrace_lock);
7836 
7837 	if ((probe = dtrace_probe_lookup_id(pid)) != NULL &&
7838 	    probe->dtpr_provider == (dtrace_provider_t *)id)
7839 		rval = probe->dtpr_arg;
7840 
7841 	mutex_exit(&dtrace_lock);
7842 
7843 	return (rval);
7844 }
7845 
7846 /*
7847  * Copy a probe into a probe description.
7848  */
7849 static void
7850 dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp)
7851 {
7852 	bzero(pdp, sizeof (dtrace_probedesc_t));
7853 	pdp->dtpd_id = prp->dtpr_id;
7854 
7855 	(void) strncpy(pdp->dtpd_provider,
7856 	    prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1);
7857 
7858 	(void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1);
7859 	(void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1);
7860 	(void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1);
7861 }
7862 
7863 #if !defined(sun)
7864 static int
7865 dtrace_probe_provide_cb(linker_file_t lf, void *arg)
7866 {
7867 	dtrace_provider_t *prv = (dtrace_provider_t *) arg;
7868 
7869 	prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, lf);
7870 
7871 	return(0);
7872 }
7873 #endif
7874 
7875 
7876 /*
7877  * Called to indicate that a probe -- or probes -- should be provided by a
7878  * specfied provider.  If the specified description is NULL, the provider will
7879  * be told to provide all of its probes.  (This is done whenever a new
7880  * consumer comes along, or whenever a retained enabling is to be matched.) If
7881  * the specified description is non-NULL, the provider is given the
7882  * opportunity to dynamically provide the specified probe, allowing providers
7883  * to support the creation of probes on-the-fly.  (So-called _autocreated_
7884  * probes.)  If the provider is NULL, the operations will be applied to all
7885  * providers; if the provider is non-NULL the operations will only be applied
7886  * to the specified provider.  The dtrace_provider_lock must be held, and the
7887  * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation
7888  * will need to grab the dtrace_lock when it reenters the framework through
7889  * dtrace_probe_lookup(), dtrace_probe_create(), etc.
7890  */
7891 static void
7892 dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv)
7893 {
7894 #if defined(sun)
7895 	modctl_t *ctl;
7896 #endif
7897 	int all = 0;
7898 
7899 	ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7900 
7901 	if (prv == NULL) {
7902 		all = 1;
7903 		prv = dtrace_provider;
7904 	}
7905 
7906 	do {
7907 		/*
7908 		 * First, call the blanket provide operation.
7909 		 */
7910 		prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc);
7911 
7912 		/*
7913 		 * Now call the per-module provide operation.  We will grab
7914 		 * mod_lock to prevent the list from being modified.  Note
7915 		 * that this also prevents the mod_busy bits from changing.
7916 		 * (mod_busy can only be changed with mod_lock held.)
7917 		 */
7918 		mutex_enter(&mod_lock);
7919 
7920 #if defined(sun)
7921 		ctl = &modules;
7922 		do {
7923 			if (ctl->mod_busy || ctl->mod_mp == NULL)
7924 				continue;
7925 
7926 			prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
7927 
7928 		} while ((ctl = ctl->mod_next) != &modules);
7929 #else
7930 		(void) linker_file_foreach(dtrace_probe_provide_cb, prv);
7931 #endif
7932 
7933 		mutex_exit(&mod_lock);
7934 	} while (all && (prv = prv->dtpv_next) != NULL);
7935 }
7936 
7937 #if defined(sun)
7938 /*
7939  * Iterate over each probe, and call the Framework-to-Provider API function
7940  * denoted by offs.
7941  */
7942 static void
7943 dtrace_probe_foreach(uintptr_t offs)
7944 {
7945 	dtrace_provider_t *prov;
7946 	void (*func)(void *, dtrace_id_t, void *);
7947 	dtrace_probe_t *probe;
7948 	dtrace_icookie_t cookie;
7949 	int i;
7950 
7951 	/*
7952 	 * We disable interrupts to walk through the probe array.  This is
7953 	 * safe -- the dtrace_sync() in dtrace_unregister() assures that we
7954 	 * won't see stale data.
7955 	 */
7956 	cookie = dtrace_interrupt_disable();
7957 
7958 	for (i = 0; i < dtrace_nprobes; i++) {
7959 		if ((probe = dtrace_probes[i]) == NULL)
7960 			continue;
7961 
7962 		if (probe->dtpr_ecb == NULL) {
7963 			/*
7964 			 * This probe isn't enabled -- don't call the function.
7965 			 */
7966 			continue;
7967 		}
7968 
7969 		prov = probe->dtpr_provider;
7970 		func = *((void(**)(void *, dtrace_id_t, void *))
7971 		    ((uintptr_t)&prov->dtpv_pops + offs));
7972 
7973 		func(prov->dtpv_arg, i + 1, probe->dtpr_arg);
7974 	}
7975 
7976 	dtrace_interrupt_enable(cookie);
7977 }
7978 #endif
7979 
7980 static int
7981 dtrace_probe_enable(dtrace_probedesc_t *desc, dtrace_enabling_t *enab)
7982 {
7983 	dtrace_probekey_t pkey;
7984 	uint32_t priv;
7985 	uid_t uid;
7986 	zoneid_t zoneid;
7987 
7988 	ASSERT(MUTEX_HELD(&dtrace_lock));
7989 	dtrace_ecb_create_cache = NULL;
7990 
7991 	if (desc == NULL) {
7992 		/*
7993 		 * If we're passed a NULL description, we're being asked to
7994 		 * create an ECB with a NULL probe.
7995 		 */
7996 		(void) dtrace_ecb_create_enable(NULL, enab);
7997 		return (0);
7998 	}
7999 
8000 	dtrace_probekey(desc, &pkey);
8001 	dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred,
8002 	    &priv, &uid, &zoneid);
8003 
8004 	return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable,
8005 	    enab));
8006 }
8007 
8008 /*
8009  * DTrace Helper Provider Functions
8010  */
8011 static void
8012 dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr)
8013 {
8014 	attr->dtat_name = DOF_ATTR_NAME(dofattr);
8015 	attr->dtat_data = DOF_ATTR_DATA(dofattr);
8016 	attr->dtat_class = DOF_ATTR_CLASS(dofattr);
8017 }
8018 
8019 static void
8020 dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov,
8021     const dof_provider_t *dofprov, char *strtab)
8022 {
8023 	hprov->dthpv_provname = strtab + dofprov->dofpv_name;
8024 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider,
8025 	    dofprov->dofpv_provattr);
8026 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod,
8027 	    dofprov->dofpv_modattr);
8028 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func,
8029 	    dofprov->dofpv_funcattr);
8030 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name,
8031 	    dofprov->dofpv_nameattr);
8032 	dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args,
8033 	    dofprov->dofpv_argsattr);
8034 }
8035 
8036 static void
8037 dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8038 {
8039 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8040 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8041 	dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
8042 	dof_provider_t *provider;
8043 	dof_probe_t *probe;
8044 	uint32_t *off, *enoff;
8045 	uint8_t *arg;
8046 	char *strtab;
8047 	uint_t i, nprobes;
8048 	dtrace_helper_provdesc_t dhpv;
8049 	dtrace_helper_probedesc_t dhpb;
8050 	dtrace_meta_t *meta = dtrace_meta_pid;
8051 	dtrace_mops_t *mops = &meta->dtm_mops;
8052 	void *parg;
8053 
8054 	provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8055 	str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8056 	    provider->dofpv_strtab * dof->dofh_secsize);
8057 	prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8058 	    provider->dofpv_probes * dof->dofh_secsize);
8059 	arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8060 	    provider->dofpv_prargs * dof->dofh_secsize);
8061 	off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8062 	    provider->dofpv_proffs * dof->dofh_secsize);
8063 
8064 	strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8065 	off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset);
8066 	arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
8067 	enoff = NULL;
8068 
8069 	/*
8070 	 * See dtrace_helper_provider_validate().
8071 	 */
8072 	if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
8073 	    provider->dofpv_prenoffs != DOF_SECT_NONE) {
8074 		enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8075 		    provider->dofpv_prenoffs * dof->dofh_secsize);
8076 		enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset);
8077 	}
8078 
8079 	nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
8080 
8081 	/*
8082 	 * Create the provider.
8083 	 */
8084 	dtrace_dofprov2hprov(&dhpv, provider, strtab);
8085 
8086 	if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL)
8087 		return;
8088 
8089 	meta->dtm_count++;
8090 
8091 	/*
8092 	 * Create the probes.
8093 	 */
8094 	for (i = 0; i < nprobes; i++) {
8095 		probe = (dof_probe_t *)(uintptr_t)(daddr +
8096 		    prb_sec->dofs_offset + i * prb_sec->dofs_entsize);
8097 
8098 		dhpb.dthpb_mod = dhp->dofhp_mod;
8099 		dhpb.dthpb_func = strtab + probe->dofpr_func;
8100 		dhpb.dthpb_name = strtab + probe->dofpr_name;
8101 		dhpb.dthpb_base = probe->dofpr_addr;
8102 		dhpb.dthpb_offs = off + probe->dofpr_offidx;
8103 		dhpb.dthpb_noffs = probe->dofpr_noffs;
8104 		if (enoff != NULL) {
8105 			dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx;
8106 			dhpb.dthpb_nenoffs = probe->dofpr_nenoffs;
8107 		} else {
8108 			dhpb.dthpb_enoffs = NULL;
8109 			dhpb.dthpb_nenoffs = 0;
8110 		}
8111 		dhpb.dthpb_args = arg + probe->dofpr_argidx;
8112 		dhpb.dthpb_nargc = probe->dofpr_nargc;
8113 		dhpb.dthpb_xargc = probe->dofpr_xargc;
8114 		dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv;
8115 		dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv;
8116 
8117 		mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb);
8118 	}
8119 }
8120 
8121 static void
8122 dtrace_helper_provide(dof_helper_t *dhp, pid_t pid)
8123 {
8124 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8125 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8126 	int i;
8127 
8128 	ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8129 
8130 	for (i = 0; i < dof->dofh_secnum; i++) {
8131 		dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8132 		    dof->dofh_secoff + i * dof->dofh_secsize);
8133 
8134 		if (sec->dofs_type != DOF_SECT_PROVIDER)
8135 			continue;
8136 
8137 		dtrace_helper_provide_one(dhp, sec, pid);
8138 	}
8139 
8140 	/*
8141 	 * We may have just created probes, so we must now rematch against
8142 	 * any retained enablings.  Note that this call will acquire both
8143 	 * cpu_lock and dtrace_lock; the fact that we are holding
8144 	 * dtrace_meta_lock now is what defines the ordering with respect to
8145 	 * these three locks.
8146 	 */
8147 	dtrace_enabling_matchall();
8148 }
8149 
8150 static void
8151 dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8152 {
8153 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8154 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8155 	dof_sec_t *str_sec;
8156 	dof_provider_t *provider;
8157 	char *strtab;
8158 	dtrace_helper_provdesc_t dhpv;
8159 	dtrace_meta_t *meta = dtrace_meta_pid;
8160 	dtrace_mops_t *mops = &meta->dtm_mops;
8161 
8162 	provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8163 	str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8164 	    provider->dofpv_strtab * dof->dofh_secsize);
8165 
8166 	strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8167 
8168 	/*
8169 	 * Create the provider.
8170 	 */
8171 	dtrace_dofprov2hprov(&dhpv, provider, strtab);
8172 
8173 	mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid);
8174 
8175 	meta->dtm_count--;
8176 }
8177 
8178 static void
8179 dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid)
8180 {
8181 	uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8182 	dof_hdr_t *dof = (dof_hdr_t *)daddr;
8183 	int i;
8184 
8185 	ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8186 
8187 	for (i = 0; i < dof->dofh_secnum; i++) {
8188 		dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8189 		    dof->dofh_secoff + i * dof->dofh_secsize);
8190 
8191 		if (sec->dofs_type != DOF_SECT_PROVIDER)
8192 			continue;
8193 
8194 		dtrace_helper_provider_remove_one(dhp, sec, pid);
8195 	}
8196 }
8197 
8198 /*
8199  * DTrace Meta Provider-to-Framework API Functions
8200  *
8201  * These functions implement the Meta Provider-to-Framework API, as described
8202  * in <sys/dtrace.h>.
8203  */
8204 int
8205 dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg,
8206     dtrace_meta_provider_id_t *idp)
8207 {
8208 	dtrace_meta_t *meta;
8209 	dtrace_helpers_t *help, *next;
8210 	int i;
8211 
8212 	*idp = DTRACE_METAPROVNONE;
8213 
8214 	/*
8215 	 * We strictly don't need the name, but we hold onto it for
8216 	 * debuggability. All hail error queues!
8217 	 */
8218 	if (name == NULL) {
8219 		cmn_err(CE_WARN, "failed to register meta-provider: "
8220 		    "invalid name");
8221 		return (EINVAL);
8222 	}
8223 
8224 	if (mops == NULL ||
8225 	    mops->dtms_create_probe == NULL ||
8226 	    mops->dtms_provide_pid == NULL ||
8227 	    mops->dtms_remove_pid == NULL) {
8228 		cmn_err(CE_WARN, "failed to register meta-register %s: "
8229 		    "invalid ops", name);
8230 		return (EINVAL);
8231 	}
8232 
8233 	meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP);
8234 	meta->dtm_mops = *mops;
8235 	meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
8236 	(void) strcpy(meta->dtm_name, name);
8237 	meta->dtm_arg = arg;
8238 
8239 	mutex_enter(&dtrace_meta_lock);
8240 	mutex_enter(&dtrace_lock);
8241 
8242 	if (dtrace_meta_pid != NULL) {
8243 		mutex_exit(&dtrace_lock);
8244 		mutex_exit(&dtrace_meta_lock);
8245 		cmn_err(CE_WARN, "failed to register meta-register %s: "
8246 		    "user-land meta-provider exists", name);
8247 		kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1);
8248 		kmem_free(meta, sizeof (dtrace_meta_t));
8249 		return (EINVAL);
8250 	}
8251 
8252 	dtrace_meta_pid = meta;
8253 	*idp = (dtrace_meta_provider_id_t)meta;
8254 
8255 	/*
8256 	 * If there are providers and probes ready to go, pass them
8257 	 * off to the new meta provider now.
8258 	 */
8259 
8260 	help = dtrace_deferred_pid;
8261 	dtrace_deferred_pid = NULL;
8262 
8263 	mutex_exit(&dtrace_lock);
8264 
8265 	while (help != NULL) {
8266 		for (i = 0; i < help->dthps_nprovs; i++) {
8267 			dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
8268 			    help->dthps_pid);
8269 		}
8270 
8271 		next = help->dthps_next;
8272 		help->dthps_next = NULL;
8273 		help->dthps_prev = NULL;
8274 		help->dthps_deferred = 0;
8275 		help = next;
8276 	}
8277 
8278 	mutex_exit(&dtrace_meta_lock);
8279 
8280 	return (0);
8281 }
8282 
8283 int
8284 dtrace_meta_unregister(dtrace_meta_provider_id_t id)
8285 {
8286 	dtrace_meta_t **pp, *old = (dtrace_meta_t *)id;
8287 
8288 	mutex_enter(&dtrace_meta_lock);
8289 	mutex_enter(&dtrace_lock);
8290 
8291 	if (old == dtrace_meta_pid) {
8292 		pp = &dtrace_meta_pid;
8293 	} else {
8294 		panic("attempt to unregister non-existent "
8295 		    "dtrace meta-provider %p\n", (void *)old);
8296 	}
8297 
8298 	if (old->dtm_count != 0) {
8299 		mutex_exit(&dtrace_lock);
8300 		mutex_exit(&dtrace_meta_lock);
8301 		return (EBUSY);
8302 	}
8303 
8304 	*pp = NULL;
8305 
8306 	mutex_exit(&dtrace_lock);
8307 	mutex_exit(&dtrace_meta_lock);
8308 
8309 	kmem_free(old->dtm_name, strlen(old->dtm_name) + 1);
8310 	kmem_free(old, sizeof (dtrace_meta_t));
8311 
8312 	return (0);
8313 }
8314 
8315 
8316 /*
8317  * DTrace DIF Object Functions
8318  */
8319 static int
8320 dtrace_difo_err(uint_t pc, const char *format, ...)
8321 {
8322 	if (dtrace_err_verbose) {
8323 		va_list alist;
8324 
8325 		(void) uprintf("dtrace DIF object error: [%u]: ", pc);
8326 		va_start(alist, format);
8327 		(void) vuprintf(format, alist);
8328 		va_end(alist);
8329 	}
8330 
8331 #ifdef DTRACE_ERRDEBUG
8332 	dtrace_errdebug(format);
8333 #endif
8334 	return (1);
8335 }
8336 
8337 /*
8338  * Validate a DTrace DIF object by checking the IR instructions.  The following
8339  * rules are currently enforced by dtrace_difo_validate():
8340  *
8341  * 1. Each instruction must have a valid opcode
8342  * 2. Each register, string, variable, or subroutine reference must be valid
8343  * 3. No instruction can modify register %r0 (must be zero)
8344  * 4. All instruction reserved bits must be set to zero
8345  * 5. The last instruction must be a "ret" instruction
8346  * 6. All branch targets must reference a valid instruction _after_ the branch
8347  */
8348 static int
8349 dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs,
8350     cred_t *cr)
8351 {
8352 	int err = 0, i;
8353 	int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
8354 	int kcheckload;
8355 	uint_t pc;
8356 
8357 	kcheckload = cr == NULL ||
8358 	    (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0;
8359 
8360 	dp->dtdo_destructive = 0;
8361 
8362 	for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) {
8363 		dif_instr_t instr = dp->dtdo_buf[pc];
8364 
8365 		uint_t r1 = DIF_INSTR_R1(instr);
8366 		uint_t r2 = DIF_INSTR_R2(instr);
8367 		uint_t rd = DIF_INSTR_RD(instr);
8368 		uint_t rs = DIF_INSTR_RS(instr);
8369 		uint_t label = DIF_INSTR_LABEL(instr);
8370 		uint_t v = DIF_INSTR_VAR(instr);
8371 		uint_t subr = DIF_INSTR_SUBR(instr);
8372 		uint_t type = DIF_INSTR_TYPE(instr);
8373 		uint_t op = DIF_INSTR_OP(instr);
8374 
8375 		switch (op) {
8376 		case DIF_OP_OR:
8377 		case DIF_OP_XOR:
8378 		case DIF_OP_AND:
8379 		case DIF_OP_SLL:
8380 		case DIF_OP_SRL:
8381 		case DIF_OP_SRA:
8382 		case DIF_OP_SUB:
8383 		case DIF_OP_ADD:
8384 		case DIF_OP_MUL:
8385 		case DIF_OP_SDIV:
8386 		case DIF_OP_UDIV:
8387 		case DIF_OP_SREM:
8388 		case DIF_OP_UREM:
8389 		case DIF_OP_COPYS:
8390 			if (r1 >= nregs)
8391 				err += efunc(pc, "invalid register %u\n", r1);
8392 			if (r2 >= nregs)
8393 				err += efunc(pc, "invalid register %u\n", r2);
8394 			if (rd >= nregs)
8395 				err += efunc(pc, "invalid register %u\n", rd);
8396 			if (rd == 0)
8397 				err += efunc(pc, "cannot write to %r0\n");
8398 			break;
8399 		case DIF_OP_NOT:
8400 		case DIF_OP_MOV:
8401 		case DIF_OP_ALLOCS:
8402 			if (r1 >= nregs)
8403 				err += efunc(pc, "invalid register %u\n", r1);
8404 			if (r2 != 0)
8405 				err += efunc(pc, "non-zero reserved bits\n");
8406 			if (rd >= nregs)
8407 				err += efunc(pc, "invalid register %u\n", rd);
8408 			if (rd == 0)
8409 				err += efunc(pc, "cannot write to %r0\n");
8410 			break;
8411 		case DIF_OP_LDSB:
8412 		case DIF_OP_LDSH:
8413 		case DIF_OP_LDSW:
8414 		case DIF_OP_LDUB:
8415 		case DIF_OP_LDUH:
8416 		case DIF_OP_LDUW:
8417 		case DIF_OP_LDX:
8418 			if (r1 >= nregs)
8419 				err += efunc(pc, "invalid register %u\n", r1);
8420 			if (r2 != 0)
8421 				err += efunc(pc, "non-zero reserved bits\n");
8422 			if (rd >= nregs)
8423 				err += efunc(pc, "invalid register %u\n", rd);
8424 			if (rd == 0)
8425 				err += efunc(pc, "cannot write to %r0\n");
8426 			if (kcheckload)
8427 				dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op +
8428 				    DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd);
8429 			break;
8430 		case DIF_OP_RLDSB:
8431 		case DIF_OP_RLDSH:
8432 		case DIF_OP_RLDSW:
8433 		case DIF_OP_RLDUB:
8434 		case DIF_OP_RLDUH:
8435 		case DIF_OP_RLDUW:
8436 		case DIF_OP_RLDX:
8437 			if (r1 >= nregs)
8438 				err += efunc(pc, "invalid register %u\n", r1);
8439 			if (r2 != 0)
8440 				err += efunc(pc, "non-zero reserved bits\n");
8441 			if (rd >= nregs)
8442 				err += efunc(pc, "invalid register %u\n", rd);
8443 			if (rd == 0)
8444 				err += efunc(pc, "cannot write to %r0\n");
8445 			break;
8446 		case DIF_OP_ULDSB:
8447 		case DIF_OP_ULDSH:
8448 		case DIF_OP_ULDSW:
8449 		case DIF_OP_ULDUB:
8450 		case DIF_OP_ULDUH:
8451 		case DIF_OP_ULDUW:
8452 		case DIF_OP_ULDX:
8453 			if (r1 >= nregs)
8454 				err += efunc(pc, "invalid register %u\n", r1);
8455 			if (r2 != 0)
8456 				err += efunc(pc, "non-zero reserved bits\n");
8457 			if (rd >= nregs)
8458 				err += efunc(pc, "invalid register %u\n", rd);
8459 			if (rd == 0)
8460 				err += efunc(pc, "cannot write to %r0\n");
8461 			break;
8462 		case DIF_OP_STB:
8463 		case DIF_OP_STH:
8464 		case DIF_OP_STW:
8465 		case DIF_OP_STX:
8466 			if (r1 >= nregs)
8467 				err += efunc(pc, "invalid register %u\n", r1);
8468 			if (r2 != 0)
8469 				err += efunc(pc, "non-zero reserved bits\n");
8470 			if (rd >= nregs)
8471 				err += efunc(pc, "invalid register %u\n", rd);
8472 			if (rd == 0)
8473 				err += efunc(pc, "cannot write to 0 address\n");
8474 			break;
8475 		case DIF_OP_CMP:
8476 		case DIF_OP_SCMP:
8477 			if (r1 >= nregs)
8478 				err += efunc(pc, "invalid register %u\n", r1);
8479 			if (r2 >= nregs)
8480 				err += efunc(pc, "invalid register %u\n", r2);
8481 			if (rd != 0)
8482 				err += efunc(pc, "non-zero reserved bits\n");
8483 			break;
8484 		case DIF_OP_TST:
8485 			if (r1 >= nregs)
8486 				err += efunc(pc, "invalid register %u\n", r1);
8487 			if (r2 != 0 || rd != 0)
8488 				err += efunc(pc, "non-zero reserved bits\n");
8489 			break;
8490 		case DIF_OP_BA:
8491 		case DIF_OP_BE:
8492 		case DIF_OP_BNE:
8493 		case DIF_OP_BG:
8494 		case DIF_OP_BGU:
8495 		case DIF_OP_BGE:
8496 		case DIF_OP_BGEU:
8497 		case DIF_OP_BL:
8498 		case DIF_OP_BLU:
8499 		case DIF_OP_BLE:
8500 		case DIF_OP_BLEU:
8501 			if (label >= dp->dtdo_len) {
8502 				err += efunc(pc, "invalid branch target %u\n",
8503 				    label);
8504 			}
8505 			if (label <= pc) {
8506 				err += efunc(pc, "backward branch to %u\n",
8507 				    label);
8508 			}
8509 			break;
8510 		case DIF_OP_RET:
8511 			if (r1 != 0 || r2 != 0)
8512 				err += efunc(pc, "non-zero reserved bits\n");
8513 			if (rd >= nregs)
8514 				err += efunc(pc, "invalid register %u\n", rd);
8515 			break;
8516 		case DIF_OP_NOP:
8517 		case DIF_OP_POPTS:
8518 		case DIF_OP_FLUSHTS:
8519 			if (r1 != 0 || r2 != 0 || rd != 0)
8520 				err += efunc(pc, "non-zero reserved bits\n");
8521 			break;
8522 		case DIF_OP_SETX:
8523 			if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) {
8524 				err += efunc(pc, "invalid integer ref %u\n",
8525 				    DIF_INSTR_INTEGER(instr));
8526 			}
8527 			if (rd >= nregs)
8528 				err += efunc(pc, "invalid register %u\n", rd);
8529 			if (rd == 0)
8530 				err += efunc(pc, "cannot write to %r0\n");
8531 			break;
8532 		case DIF_OP_SETS:
8533 			if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) {
8534 				err += efunc(pc, "invalid string ref %u\n",
8535 				    DIF_INSTR_STRING(instr));
8536 			}
8537 			if (rd >= nregs)
8538 				err += efunc(pc, "invalid register %u\n", rd);
8539 			if (rd == 0)
8540 				err += efunc(pc, "cannot write to %r0\n");
8541 			break;
8542 		case DIF_OP_LDGA:
8543 		case DIF_OP_LDTA:
8544 			if (r1 > DIF_VAR_ARRAY_MAX)
8545 				err += efunc(pc, "invalid array %u\n", r1);
8546 			if (r2 >= nregs)
8547 				err += efunc(pc, "invalid register %u\n", r2);
8548 			if (rd >= nregs)
8549 				err += efunc(pc, "invalid register %u\n", rd);
8550 			if (rd == 0)
8551 				err += efunc(pc, "cannot write to %r0\n");
8552 			break;
8553 		case DIF_OP_LDGS:
8554 		case DIF_OP_LDTS:
8555 		case DIF_OP_LDLS:
8556 		case DIF_OP_LDGAA:
8557 		case DIF_OP_LDTAA:
8558 			if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX)
8559 				err += efunc(pc, "invalid variable %u\n", v);
8560 			if (rd >= nregs)
8561 				err += efunc(pc, "invalid register %u\n", rd);
8562 			if (rd == 0)
8563 				err += efunc(pc, "cannot write to %r0\n");
8564 			break;
8565 		case DIF_OP_STGS:
8566 		case DIF_OP_STTS:
8567 		case DIF_OP_STLS:
8568 		case DIF_OP_STGAA:
8569 		case DIF_OP_STTAA:
8570 			if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX)
8571 				err += efunc(pc, "invalid variable %u\n", v);
8572 			if (rs >= nregs)
8573 				err += efunc(pc, "invalid register %u\n", rd);
8574 			break;
8575 		case DIF_OP_CALL:
8576 			if (subr > DIF_SUBR_MAX)
8577 				err += efunc(pc, "invalid subr %u\n", subr);
8578 			if (rd >= nregs)
8579 				err += efunc(pc, "invalid register %u\n", rd);
8580 			if (rd == 0)
8581 				err += efunc(pc, "cannot write to %r0\n");
8582 
8583 			if (subr == DIF_SUBR_COPYOUT ||
8584 			    subr == DIF_SUBR_COPYOUTSTR) {
8585 				dp->dtdo_destructive = 1;
8586 			}
8587 			break;
8588 		case DIF_OP_PUSHTR:
8589 			if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF)
8590 				err += efunc(pc, "invalid ref type %u\n", type);
8591 			if (r2 >= nregs)
8592 				err += efunc(pc, "invalid register %u\n", r2);
8593 			if (rs >= nregs)
8594 				err += efunc(pc, "invalid register %u\n", rs);
8595 			break;
8596 		case DIF_OP_PUSHTV:
8597 			if (type != DIF_TYPE_CTF)
8598 				err += efunc(pc, "invalid val type %u\n", type);
8599 			if (r2 >= nregs)
8600 				err += efunc(pc, "invalid register %u\n", r2);
8601 			if (rs >= nregs)
8602 				err += efunc(pc, "invalid register %u\n", rs);
8603 			break;
8604 		default:
8605 			err += efunc(pc, "invalid opcode %u\n",
8606 			    DIF_INSTR_OP(instr));
8607 		}
8608 	}
8609 
8610 	if (dp->dtdo_len != 0 &&
8611 	    DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) {
8612 		err += efunc(dp->dtdo_len - 1,
8613 		    "expected 'ret' as last DIF instruction\n");
8614 	}
8615 
8616 	if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) {
8617 		/*
8618 		 * If we're not returning by reference, the size must be either
8619 		 * 0 or the size of one of the base types.
8620 		 */
8621 		switch (dp->dtdo_rtype.dtdt_size) {
8622 		case 0:
8623 		case sizeof (uint8_t):
8624 		case sizeof (uint16_t):
8625 		case sizeof (uint32_t):
8626 		case sizeof (uint64_t):
8627 			break;
8628 
8629 		default:
8630 			err += efunc(dp->dtdo_len - 1, "bad return size");
8631 		}
8632 	}
8633 
8634 	for (i = 0; i < dp->dtdo_varlen && err == 0; i++) {
8635 		dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL;
8636 		dtrace_diftype_t *vt, *et;
8637 		uint_t id, ndx;
8638 
8639 		if (v->dtdv_scope != DIFV_SCOPE_GLOBAL &&
8640 		    v->dtdv_scope != DIFV_SCOPE_THREAD &&
8641 		    v->dtdv_scope != DIFV_SCOPE_LOCAL) {
8642 			err += efunc(i, "unrecognized variable scope %d\n",
8643 			    v->dtdv_scope);
8644 			break;
8645 		}
8646 
8647 		if (v->dtdv_kind != DIFV_KIND_ARRAY &&
8648 		    v->dtdv_kind != DIFV_KIND_SCALAR) {
8649 			err += efunc(i, "unrecognized variable type %d\n",
8650 			    v->dtdv_kind);
8651 			break;
8652 		}
8653 
8654 		if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) {
8655 			err += efunc(i, "%d exceeds variable id limit\n", id);
8656 			break;
8657 		}
8658 
8659 		if (id < DIF_VAR_OTHER_UBASE)
8660 			continue;
8661 
8662 		/*
8663 		 * For user-defined variables, we need to check that this
8664 		 * definition is identical to any previous definition that we
8665 		 * encountered.
8666 		 */
8667 		ndx = id - DIF_VAR_OTHER_UBASE;
8668 
8669 		switch (v->dtdv_scope) {
8670 		case DIFV_SCOPE_GLOBAL:
8671 			if (ndx < vstate->dtvs_nglobals) {
8672 				dtrace_statvar_t *svar;
8673 
8674 				if ((svar = vstate->dtvs_globals[ndx]) != NULL)
8675 					existing = &svar->dtsv_var;
8676 			}
8677 
8678 			break;
8679 
8680 		case DIFV_SCOPE_THREAD:
8681 			if (ndx < vstate->dtvs_ntlocals)
8682 				existing = &vstate->dtvs_tlocals[ndx];
8683 			break;
8684 
8685 		case DIFV_SCOPE_LOCAL:
8686 			if (ndx < vstate->dtvs_nlocals) {
8687 				dtrace_statvar_t *svar;
8688 
8689 				if ((svar = vstate->dtvs_locals[ndx]) != NULL)
8690 					existing = &svar->dtsv_var;
8691 			}
8692 
8693 			break;
8694 		}
8695 
8696 		vt = &v->dtdv_type;
8697 
8698 		if (vt->dtdt_flags & DIF_TF_BYREF) {
8699 			if (vt->dtdt_size == 0) {
8700 				err += efunc(i, "zero-sized variable\n");
8701 				break;
8702 			}
8703 
8704 			if (v->dtdv_scope == DIFV_SCOPE_GLOBAL &&
8705 			    vt->dtdt_size > dtrace_global_maxsize) {
8706 				err += efunc(i, "oversized by-ref global\n");
8707 				break;
8708 			}
8709 		}
8710 
8711 		if (existing == NULL || existing->dtdv_id == 0)
8712 			continue;
8713 
8714 		ASSERT(existing->dtdv_id == v->dtdv_id);
8715 		ASSERT(existing->dtdv_scope == v->dtdv_scope);
8716 
8717 		if (existing->dtdv_kind != v->dtdv_kind)
8718 			err += efunc(i, "%d changed variable kind\n", id);
8719 
8720 		et = &existing->dtdv_type;
8721 
8722 		if (vt->dtdt_flags != et->dtdt_flags) {
8723 			err += efunc(i, "%d changed variable type flags\n", id);
8724 			break;
8725 		}
8726 
8727 		if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) {
8728 			err += efunc(i, "%d changed variable type size\n", id);
8729 			break;
8730 		}
8731 	}
8732 
8733 	return (err);
8734 }
8735 
8736 /*
8737  * Validate a DTrace DIF object that it is to be used as a helper.  Helpers
8738  * are much more constrained than normal DIFOs.  Specifically, they may
8739  * not:
8740  *
8741  * 1. Make calls to subroutines other than copyin(), copyinstr() or
8742  *    miscellaneous string routines
8743  * 2. Access DTrace variables other than the args[] array, and the
8744  *    curthread, pid, ppid, tid, execname, zonename, uid and gid variables.
8745  * 3. Have thread-local variables.
8746  * 4. Have dynamic variables.
8747  */
8748 static int
8749 dtrace_difo_validate_helper(dtrace_difo_t *dp)
8750 {
8751 	int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
8752 	int err = 0;
8753 	uint_t pc;
8754 
8755 	for (pc = 0; pc < dp->dtdo_len; pc++) {
8756 		dif_instr_t instr = dp->dtdo_buf[pc];
8757 
8758 		uint_t v = DIF_INSTR_VAR(instr);
8759 		uint_t subr = DIF_INSTR_SUBR(instr);
8760 		uint_t op = DIF_INSTR_OP(instr);
8761 
8762 		switch (op) {
8763 		case DIF_OP_OR:
8764 		case DIF_OP_XOR:
8765 		case DIF_OP_AND:
8766 		case DIF_OP_SLL:
8767 		case DIF_OP_SRL:
8768 		case DIF_OP_SRA:
8769 		case DIF_OP_SUB:
8770 		case DIF_OP_ADD:
8771 		case DIF_OP_MUL:
8772 		case DIF_OP_SDIV:
8773 		case DIF_OP_UDIV:
8774 		case DIF_OP_SREM:
8775 		case DIF_OP_UREM:
8776 		case DIF_OP_COPYS:
8777 		case DIF_OP_NOT:
8778 		case DIF_OP_MOV:
8779 		case DIF_OP_RLDSB:
8780 		case DIF_OP_RLDSH:
8781 		case DIF_OP_RLDSW:
8782 		case DIF_OP_RLDUB:
8783 		case DIF_OP_RLDUH:
8784 		case DIF_OP_RLDUW:
8785 		case DIF_OP_RLDX:
8786 		case DIF_OP_ULDSB:
8787 		case DIF_OP_ULDSH:
8788 		case DIF_OP_ULDSW:
8789 		case DIF_OP_ULDUB:
8790 		case DIF_OP_ULDUH:
8791 		case DIF_OP_ULDUW:
8792 		case DIF_OP_ULDX:
8793 		case DIF_OP_STB:
8794 		case DIF_OP_STH:
8795 		case DIF_OP_STW:
8796 		case DIF_OP_STX:
8797 		case DIF_OP_ALLOCS:
8798 		case DIF_OP_CMP:
8799 		case DIF_OP_SCMP:
8800 		case DIF_OP_TST:
8801 		case DIF_OP_BA:
8802 		case DIF_OP_BE:
8803 		case DIF_OP_BNE:
8804 		case DIF_OP_BG:
8805 		case DIF_OP_BGU:
8806 		case DIF_OP_BGE:
8807 		case DIF_OP_BGEU:
8808 		case DIF_OP_BL:
8809 		case DIF_OP_BLU:
8810 		case DIF_OP_BLE:
8811 		case DIF_OP_BLEU:
8812 		case DIF_OP_RET:
8813 		case DIF_OP_NOP:
8814 		case DIF_OP_POPTS:
8815 		case DIF_OP_FLUSHTS:
8816 		case DIF_OP_SETX:
8817 		case DIF_OP_SETS:
8818 		case DIF_OP_LDGA:
8819 		case DIF_OP_LDLS:
8820 		case DIF_OP_STGS:
8821 		case DIF_OP_STLS:
8822 		case DIF_OP_PUSHTR:
8823 		case DIF_OP_PUSHTV:
8824 			break;
8825 
8826 		case DIF_OP_LDGS:
8827 			if (v >= DIF_VAR_OTHER_UBASE)
8828 				break;
8829 
8830 			if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9)
8831 				break;
8832 
8833 			if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID ||
8834 			    v == DIF_VAR_PPID || v == DIF_VAR_TID ||
8835 			    v == DIF_VAR_EXECARGS ||
8836 			    v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME ||
8837 			    v == DIF_VAR_UID || v == DIF_VAR_GID)
8838 				break;
8839 
8840 			err += efunc(pc, "illegal variable %u\n", v);
8841 			break;
8842 
8843 		case DIF_OP_LDTA:
8844 		case DIF_OP_LDTS:
8845 		case DIF_OP_LDGAA:
8846 		case DIF_OP_LDTAA:
8847 			err += efunc(pc, "illegal dynamic variable load\n");
8848 			break;
8849 
8850 		case DIF_OP_STTS:
8851 		case DIF_OP_STGAA:
8852 		case DIF_OP_STTAA:
8853 			err += efunc(pc, "illegal dynamic variable store\n");
8854 			break;
8855 
8856 		case DIF_OP_CALL:
8857 			if (subr == DIF_SUBR_ALLOCA ||
8858 			    subr == DIF_SUBR_BCOPY ||
8859 			    subr == DIF_SUBR_COPYIN ||
8860 			    subr == DIF_SUBR_COPYINTO ||
8861 			    subr == DIF_SUBR_COPYINSTR ||
8862 			    subr == DIF_SUBR_INDEX ||
8863 			    subr == DIF_SUBR_INET_NTOA ||
8864 			    subr == DIF_SUBR_INET_NTOA6 ||
8865 			    subr == DIF_SUBR_INET_NTOP ||
8866 			    subr == DIF_SUBR_LLTOSTR ||
8867 			    subr == DIF_SUBR_RINDEX ||
8868 			    subr == DIF_SUBR_STRCHR ||
8869 			    subr == DIF_SUBR_STRJOIN ||
8870 			    subr == DIF_SUBR_STRRCHR ||
8871 			    subr == DIF_SUBR_STRSTR ||
8872 			    subr == DIF_SUBR_HTONS ||
8873 			    subr == DIF_SUBR_HTONL ||
8874 			    subr == DIF_SUBR_HTONLL ||
8875 			    subr == DIF_SUBR_NTOHS ||
8876 			    subr == DIF_SUBR_NTOHL ||
8877 			    subr == DIF_SUBR_NTOHLL ||
8878 			    subr == DIF_SUBR_MEMREF ||
8879 			    subr == DIF_SUBR_TYPEREF)
8880 				break;
8881 
8882 			err += efunc(pc, "invalid subr %u\n", subr);
8883 			break;
8884 
8885 		default:
8886 			err += efunc(pc, "invalid opcode %u\n",
8887 			    DIF_INSTR_OP(instr));
8888 		}
8889 	}
8890 
8891 	return (err);
8892 }
8893 
8894 /*
8895  * Returns 1 if the expression in the DIF object can be cached on a per-thread
8896  * basis; 0 if not.
8897  */
8898 static int
8899 dtrace_difo_cacheable(dtrace_difo_t *dp)
8900 {
8901 	int i;
8902 
8903 	if (dp == NULL)
8904 		return (0);
8905 
8906 	for (i = 0; i < dp->dtdo_varlen; i++) {
8907 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
8908 
8909 		if (v->dtdv_scope != DIFV_SCOPE_GLOBAL)
8910 			continue;
8911 
8912 		switch (v->dtdv_id) {
8913 		case DIF_VAR_CURTHREAD:
8914 		case DIF_VAR_PID:
8915 		case DIF_VAR_TID:
8916 		case DIF_VAR_EXECARGS:
8917 		case DIF_VAR_EXECNAME:
8918 		case DIF_VAR_ZONENAME:
8919 			break;
8920 
8921 		default:
8922 			return (0);
8923 		}
8924 	}
8925 
8926 	/*
8927 	 * This DIF object may be cacheable.  Now we need to look for any
8928 	 * array loading instructions, any memory loading instructions, or
8929 	 * any stores to thread-local variables.
8930 	 */
8931 	for (i = 0; i < dp->dtdo_len; i++) {
8932 		uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]);
8933 
8934 		if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) ||
8935 		    (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) ||
8936 		    (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) ||
8937 		    op == DIF_OP_LDGA || op == DIF_OP_STTS)
8938 			return (0);
8939 	}
8940 
8941 	return (1);
8942 }
8943 
8944 static void
8945 dtrace_difo_hold(dtrace_difo_t *dp)
8946 {
8947 	int i;
8948 
8949 	ASSERT(MUTEX_HELD(&dtrace_lock));
8950 
8951 	dp->dtdo_refcnt++;
8952 	ASSERT(dp->dtdo_refcnt != 0);
8953 
8954 	/*
8955 	 * We need to check this DIF object for references to the variable
8956 	 * DIF_VAR_VTIMESTAMP.
8957 	 */
8958 	for (i = 0; i < dp->dtdo_varlen; i++) {
8959 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
8960 
8961 		if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
8962 			continue;
8963 
8964 		if (dtrace_vtime_references++ == 0)
8965 			dtrace_vtime_enable();
8966 	}
8967 }
8968 
8969 /*
8970  * This routine calculates the dynamic variable chunksize for a given DIF
8971  * object.  The calculation is not fool-proof, and can probably be tricked by
8972  * malicious DIF -- but it works for all compiler-generated DIF.  Because this
8973  * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail
8974  * if a dynamic variable size exceeds the chunksize.
8975  */
8976 static void
8977 dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
8978 {
8979 	uint64_t sval = 0;
8980 	dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
8981 	const dif_instr_t *text = dp->dtdo_buf;
8982 	uint_t pc, srd = 0;
8983 	uint_t ttop = 0;
8984 	size_t size, ksize;
8985 	uint_t id, i;
8986 
8987 	for (pc = 0; pc < dp->dtdo_len; pc++) {
8988 		dif_instr_t instr = text[pc];
8989 		uint_t op = DIF_INSTR_OP(instr);
8990 		uint_t rd = DIF_INSTR_RD(instr);
8991 		uint_t r1 = DIF_INSTR_R1(instr);
8992 		uint_t nkeys = 0;
8993 		uchar_t scope = 0;
8994 
8995 		dtrace_key_t *key = tupregs;
8996 
8997 		switch (op) {
8998 		case DIF_OP_SETX:
8999 			sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)];
9000 			srd = rd;
9001 			continue;
9002 
9003 		case DIF_OP_STTS:
9004 			key = &tupregs[DIF_DTR_NREGS];
9005 			key[0].dttk_size = 0;
9006 			key[1].dttk_size = 0;
9007 			nkeys = 2;
9008 			scope = DIFV_SCOPE_THREAD;
9009 			break;
9010 
9011 		case DIF_OP_STGAA:
9012 		case DIF_OP_STTAA:
9013 			nkeys = ttop;
9014 
9015 			if (DIF_INSTR_OP(instr) == DIF_OP_STTAA)
9016 				key[nkeys++].dttk_size = 0;
9017 
9018 			key[nkeys++].dttk_size = 0;
9019 
9020 			if (op == DIF_OP_STTAA) {
9021 				scope = DIFV_SCOPE_THREAD;
9022 			} else {
9023 				scope = DIFV_SCOPE_GLOBAL;
9024 			}
9025 
9026 			break;
9027 
9028 		case DIF_OP_PUSHTR:
9029 			if (ttop == DIF_DTR_NREGS)
9030 				return;
9031 
9032 			if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) {
9033 				/*
9034 				 * If the register for the size of the "pushtr"
9035 				 * is %r0 (or the value is 0) and the type is
9036 				 * a string, we'll use the system-wide default
9037 				 * string size.
9038 				 */
9039 				tupregs[ttop++].dttk_size =
9040 				    dtrace_strsize_default;
9041 			} else {
9042 				if (srd == 0)
9043 					return;
9044 
9045 				tupregs[ttop++].dttk_size = sval;
9046 			}
9047 
9048 			break;
9049 
9050 		case DIF_OP_PUSHTV:
9051 			if (ttop == DIF_DTR_NREGS)
9052 				return;
9053 
9054 			tupregs[ttop++].dttk_size = 0;
9055 			break;
9056 
9057 		case DIF_OP_FLUSHTS:
9058 			ttop = 0;
9059 			break;
9060 
9061 		case DIF_OP_POPTS:
9062 			if (ttop != 0)
9063 				ttop--;
9064 			break;
9065 		}
9066 
9067 		sval = 0;
9068 		srd = 0;
9069 
9070 		if (nkeys == 0)
9071 			continue;
9072 
9073 		/*
9074 		 * We have a dynamic variable allocation; calculate its size.
9075 		 */
9076 		for (ksize = 0, i = 0; i < nkeys; i++)
9077 			ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
9078 
9079 		size = sizeof (dtrace_dynvar_t);
9080 		size += sizeof (dtrace_key_t) * (nkeys - 1);
9081 		size += ksize;
9082 
9083 		/*
9084 		 * Now we need to determine the size of the stored data.
9085 		 */
9086 		id = DIF_INSTR_VAR(instr);
9087 
9088 		for (i = 0; i < dp->dtdo_varlen; i++) {
9089 			dtrace_difv_t *v = &dp->dtdo_vartab[i];
9090 
9091 			if (v->dtdv_id == id && v->dtdv_scope == scope) {
9092 				size += v->dtdv_type.dtdt_size;
9093 				break;
9094 			}
9095 		}
9096 
9097 		if (i == dp->dtdo_varlen)
9098 			return;
9099 
9100 		/*
9101 		 * We have the size.  If this is larger than the chunk size
9102 		 * for our dynamic variable state, reset the chunk size.
9103 		 */
9104 		size = P2ROUNDUP(size, sizeof (uint64_t));
9105 
9106 		if (size > vstate->dtvs_dynvars.dtds_chunksize)
9107 			vstate->dtvs_dynvars.dtds_chunksize = size;
9108 	}
9109 }
9110 
9111 static void
9112 dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9113 {
9114 	int i, oldsvars, osz, nsz, otlocals, ntlocals;
9115 	uint_t id;
9116 
9117 	ASSERT(MUTEX_HELD(&dtrace_lock));
9118 	ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0);
9119 
9120 	for (i = 0; i < dp->dtdo_varlen; i++) {
9121 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9122 		dtrace_statvar_t *svar, ***svarp = NULL;
9123 		size_t dsize = 0;
9124 		uint8_t scope = v->dtdv_scope;
9125 		int *np = NULL;
9126 
9127 		if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9128 			continue;
9129 
9130 		id -= DIF_VAR_OTHER_UBASE;
9131 
9132 		switch (scope) {
9133 		case DIFV_SCOPE_THREAD:
9134 			while (id >= (otlocals = vstate->dtvs_ntlocals)) {
9135 				dtrace_difv_t *tlocals;
9136 
9137 				if ((ntlocals = (otlocals << 1)) == 0)
9138 					ntlocals = 1;
9139 
9140 				osz = otlocals * sizeof (dtrace_difv_t);
9141 				nsz = ntlocals * sizeof (dtrace_difv_t);
9142 
9143 				tlocals = kmem_zalloc(nsz, KM_SLEEP);
9144 
9145 				if (osz != 0) {
9146 					bcopy(vstate->dtvs_tlocals,
9147 					    tlocals, osz);
9148 					kmem_free(vstate->dtvs_tlocals, osz);
9149 				}
9150 
9151 				vstate->dtvs_tlocals = tlocals;
9152 				vstate->dtvs_ntlocals = ntlocals;
9153 			}
9154 
9155 			vstate->dtvs_tlocals[id] = *v;
9156 			continue;
9157 
9158 		case DIFV_SCOPE_LOCAL:
9159 			np = &vstate->dtvs_nlocals;
9160 			svarp = &vstate->dtvs_locals;
9161 
9162 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9163 				dsize = NCPU * (v->dtdv_type.dtdt_size +
9164 				    sizeof (uint64_t));
9165 			else
9166 				dsize = NCPU * sizeof (uint64_t);
9167 
9168 			break;
9169 
9170 		case DIFV_SCOPE_GLOBAL:
9171 			np = &vstate->dtvs_nglobals;
9172 			svarp = &vstate->dtvs_globals;
9173 
9174 			if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9175 				dsize = v->dtdv_type.dtdt_size +
9176 				    sizeof (uint64_t);
9177 
9178 			break;
9179 
9180 		default:
9181 			ASSERT(0);
9182 		}
9183 
9184 		while (id >= (oldsvars = *np)) {
9185 			dtrace_statvar_t **statics;
9186 			int newsvars, oldsize, newsize;
9187 
9188 			if ((newsvars = (oldsvars << 1)) == 0)
9189 				newsvars = 1;
9190 
9191 			oldsize = oldsvars * sizeof (dtrace_statvar_t *);
9192 			newsize = newsvars * sizeof (dtrace_statvar_t *);
9193 
9194 			statics = kmem_zalloc(newsize, KM_SLEEP);
9195 
9196 			if (oldsize != 0) {
9197 				bcopy(*svarp, statics, oldsize);
9198 				kmem_free(*svarp, oldsize);
9199 			}
9200 
9201 			*svarp = statics;
9202 			*np = newsvars;
9203 		}
9204 
9205 		if ((svar = (*svarp)[id]) == NULL) {
9206 			svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP);
9207 			svar->dtsv_var = *v;
9208 
9209 			if ((svar->dtsv_size = dsize) != 0) {
9210 				svar->dtsv_data = (uint64_t)(uintptr_t)
9211 				    kmem_zalloc(dsize, KM_SLEEP);
9212 			}
9213 
9214 			(*svarp)[id] = svar;
9215 		}
9216 
9217 		svar->dtsv_refcnt++;
9218 	}
9219 
9220 	dtrace_difo_chunksize(dp, vstate);
9221 	dtrace_difo_hold(dp);
9222 }
9223 
9224 static dtrace_difo_t *
9225 dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9226 {
9227 	dtrace_difo_t *new;
9228 	size_t sz;
9229 
9230 	ASSERT(dp->dtdo_buf != NULL);
9231 	ASSERT(dp->dtdo_refcnt != 0);
9232 
9233 	new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
9234 
9235 	ASSERT(dp->dtdo_buf != NULL);
9236 	sz = dp->dtdo_len * sizeof (dif_instr_t);
9237 	new->dtdo_buf = kmem_alloc(sz, KM_SLEEP);
9238 	bcopy(dp->dtdo_buf, new->dtdo_buf, sz);
9239 	new->dtdo_len = dp->dtdo_len;
9240 
9241 	if (dp->dtdo_strtab != NULL) {
9242 		ASSERT(dp->dtdo_strlen != 0);
9243 		new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP);
9244 		bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen);
9245 		new->dtdo_strlen = dp->dtdo_strlen;
9246 	}
9247 
9248 	if (dp->dtdo_inttab != NULL) {
9249 		ASSERT(dp->dtdo_intlen != 0);
9250 		sz = dp->dtdo_intlen * sizeof (uint64_t);
9251 		new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP);
9252 		bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz);
9253 		new->dtdo_intlen = dp->dtdo_intlen;
9254 	}
9255 
9256 	if (dp->dtdo_vartab != NULL) {
9257 		ASSERT(dp->dtdo_varlen != 0);
9258 		sz = dp->dtdo_varlen * sizeof (dtrace_difv_t);
9259 		new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP);
9260 		bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz);
9261 		new->dtdo_varlen = dp->dtdo_varlen;
9262 	}
9263 
9264 	dtrace_difo_init(new, vstate);
9265 	return (new);
9266 }
9267 
9268 static void
9269 dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9270 {
9271 	int i;
9272 
9273 	ASSERT(dp->dtdo_refcnt == 0);
9274 
9275 	for (i = 0; i < dp->dtdo_varlen; i++) {
9276 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9277 		dtrace_statvar_t *svar, **svarp = NULL;
9278 		uint_t id;
9279 		uint8_t scope = v->dtdv_scope;
9280 		int *np = NULL;
9281 
9282 		switch (scope) {
9283 		case DIFV_SCOPE_THREAD:
9284 			continue;
9285 
9286 		case DIFV_SCOPE_LOCAL:
9287 			np = &vstate->dtvs_nlocals;
9288 			svarp = vstate->dtvs_locals;
9289 			break;
9290 
9291 		case DIFV_SCOPE_GLOBAL:
9292 			np = &vstate->dtvs_nglobals;
9293 			svarp = vstate->dtvs_globals;
9294 			break;
9295 
9296 		default:
9297 			ASSERT(0);
9298 		}
9299 
9300 		if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9301 			continue;
9302 
9303 		id -= DIF_VAR_OTHER_UBASE;
9304 		ASSERT(id < *np);
9305 
9306 		svar = svarp[id];
9307 		ASSERT(svar != NULL);
9308 		ASSERT(svar->dtsv_refcnt > 0);
9309 
9310 		if (--svar->dtsv_refcnt > 0)
9311 			continue;
9312 
9313 		if (svar->dtsv_size != 0) {
9314 			ASSERT(svar->dtsv_data != 0);
9315 			kmem_free((void *)(uintptr_t)svar->dtsv_data,
9316 			    svar->dtsv_size);
9317 		}
9318 
9319 		kmem_free(svar, sizeof (dtrace_statvar_t));
9320 		svarp[id] = NULL;
9321 	}
9322 
9323 	if (dp->dtdo_buf != NULL)
9324 		kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
9325 	if (dp->dtdo_inttab != NULL)
9326 		kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
9327 	if (dp->dtdo_strtab != NULL)
9328 		kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
9329 	if (dp->dtdo_vartab != NULL)
9330 		kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
9331 
9332 	kmem_free(dp, sizeof (dtrace_difo_t));
9333 }
9334 
9335 static void
9336 dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9337 {
9338 	int i;
9339 
9340 	ASSERT(MUTEX_HELD(&dtrace_lock));
9341 	ASSERT(dp->dtdo_refcnt != 0);
9342 
9343 	for (i = 0; i < dp->dtdo_varlen; i++) {
9344 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
9345 
9346 		if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
9347 			continue;
9348 
9349 		ASSERT(dtrace_vtime_references > 0);
9350 		if (--dtrace_vtime_references == 0)
9351 			dtrace_vtime_disable();
9352 	}
9353 
9354 	if (--dp->dtdo_refcnt == 0)
9355 		dtrace_difo_destroy(dp, vstate);
9356 }
9357 
9358 /*
9359  * DTrace Format Functions
9360  */
9361 static uint16_t
9362 dtrace_format_add(dtrace_state_t *state, char *str)
9363 {
9364 	char *fmt, **new;
9365 	uint16_t ndx, len = strlen(str) + 1;
9366 
9367 	fmt = kmem_zalloc(len, KM_SLEEP);
9368 	bcopy(str, fmt, len);
9369 
9370 	for (ndx = 0; ndx < state->dts_nformats; ndx++) {
9371 		if (state->dts_formats[ndx] == NULL) {
9372 			state->dts_formats[ndx] = fmt;
9373 			return (ndx + 1);
9374 		}
9375 	}
9376 
9377 	if (state->dts_nformats == USHRT_MAX) {
9378 		/*
9379 		 * This is only likely if a denial-of-service attack is being
9380 		 * attempted.  As such, it's okay to fail silently here.
9381 		 */
9382 		kmem_free(fmt, len);
9383 		return (0);
9384 	}
9385 
9386 	/*
9387 	 * For simplicity, we always resize the formats array to be exactly the
9388 	 * number of formats.
9389 	 */
9390 	ndx = state->dts_nformats++;
9391 	new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP);
9392 
9393 	if (state->dts_formats != NULL) {
9394 		ASSERT(ndx != 0);
9395 		bcopy(state->dts_formats, new, ndx * sizeof (char *));
9396 		kmem_free(state->dts_formats, ndx * sizeof (char *));
9397 	}
9398 
9399 	state->dts_formats = new;
9400 	state->dts_formats[ndx] = fmt;
9401 
9402 	return (ndx + 1);
9403 }
9404 
9405 static void
9406 dtrace_format_remove(dtrace_state_t *state, uint16_t format)
9407 {
9408 	char *fmt;
9409 
9410 	ASSERT(state->dts_formats != NULL);
9411 	ASSERT(format <= state->dts_nformats);
9412 	ASSERT(state->dts_formats[format - 1] != NULL);
9413 
9414 	fmt = state->dts_formats[format - 1];
9415 	kmem_free(fmt, strlen(fmt) + 1);
9416 	state->dts_formats[format - 1] = NULL;
9417 }
9418 
9419 static void
9420 dtrace_format_destroy(dtrace_state_t *state)
9421 {
9422 	int i;
9423 
9424 	if (state->dts_nformats == 0) {
9425 		ASSERT(state->dts_formats == NULL);
9426 		return;
9427 	}
9428 
9429 	ASSERT(state->dts_formats != NULL);
9430 
9431 	for (i = 0; i < state->dts_nformats; i++) {
9432 		char *fmt = state->dts_formats[i];
9433 
9434 		if (fmt == NULL)
9435 			continue;
9436 
9437 		kmem_free(fmt, strlen(fmt) + 1);
9438 	}
9439 
9440 	kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *));
9441 	state->dts_nformats = 0;
9442 	state->dts_formats = NULL;
9443 }
9444 
9445 /*
9446  * DTrace Predicate Functions
9447  */
9448 static dtrace_predicate_t *
9449 dtrace_predicate_create(dtrace_difo_t *dp)
9450 {
9451 	dtrace_predicate_t *pred;
9452 
9453 	ASSERT(MUTEX_HELD(&dtrace_lock));
9454 	ASSERT(dp->dtdo_refcnt != 0);
9455 
9456 	pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP);
9457 	pred->dtp_difo = dp;
9458 	pred->dtp_refcnt = 1;
9459 
9460 	if (!dtrace_difo_cacheable(dp))
9461 		return (pred);
9462 
9463 	if (dtrace_predcache_id == DTRACE_CACHEIDNONE) {
9464 		/*
9465 		 * This is only theoretically possible -- we have had 2^32
9466 		 * cacheable predicates on this machine.  We cannot allow any
9467 		 * more predicates to become cacheable:  as unlikely as it is,
9468 		 * there may be a thread caching a (now stale) predicate cache
9469 		 * ID. (N.B.: the temptation is being successfully resisted to
9470 		 * have this cmn_err() "Holy shit -- we executed this code!")
9471 		 */
9472 		return (pred);
9473 	}
9474 
9475 	pred->dtp_cacheid = dtrace_predcache_id++;
9476 
9477 	return (pred);
9478 }
9479 
9480 static void
9481 dtrace_predicate_hold(dtrace_predicate_t *pred)
9482 {
9483 	ASSERT(MUTEX_HELD(&dtrace_lock));
9484 	ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0);
9485 	ASSERT(pred->dtp_refcnt > 0);
9486 
9487 	pred->dtp_refcnt++;
9488 }
9489 
9490 static void
9491 dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate)
9492 {
9493 	dtrace_difo_t *dp = pred->dtp_difo;
9494 
9495 	ASSERT(MUTEX_HELD(&dtrace_lock));
9496 	ASSERT(dp != NULL && dp->dtdo_refcnt != 0);
9497 	ASSERT(pred->dtp_refcnt > 0);
9498 
9499 	if (--pred->dtp_refcnt == 0) {
9500 		dtrace_difo_release(pred->dtp_difo, vstate);
9501 		kmem_free(pred, sizeof (dtrace_predicate_t));
9502 	}
9503 }
9504 
9505 /*
9506  * DTrace Action Description Functions
9507  */
9508 static dtrace_actdesc_t *
9509 dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple,
9510     uint64_t uarg, uint64_t arg)
9511 {
9512 	dtrace_actdesc_t *act;
9513 
9514 #if defined(sun)
9515 	ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL &&
9516 	    arg >= KERNELBASE) || (arg == NULL && kind == DTRACEACT_PRINTA));
9517 #endif
9518 
9519 	act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP);
9520 	act->dtad_kind = kind;
9521 	act->dtad_ntuple = ntuple;
9522 	act->dtad_uarg = uarg;
9523 	act->dtad_arg = arg;
9524 	act->dtad_refcnt = 1;
9525 
9526 	return (act);
9527 }
9528 
9529 static void
9530 dtrace_actdesc_hold(dtrace_actdesc_t *act)
9531 {
9532 	ASSERT(act->dtad_refcnt >= 1);
9533 	act->dtad_refcnt++;
9534 }
9535 
9536 static void
9537 dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate)
9538 {
9539 	dtrace_actkind_t kind = act->dtad_kind;
9540 	dtrace_difo_t *dp;
9541 
9542 	ASSERT(act->dtad_refcnt >= 1);
9543 
9544 	if (--act->dtad_refcnt != 0)
9545 		return;
9546 
9547 	if ((dp = act->dtad_difo) != NULL)
9548 		dtrace_difo_release(dp, vstate);
9549 
9550 	if (DTRACEACT_ISPRINTFLIKE(kind)) {
9551 		char *str = (char *)(uintptr_t)act->dtad_arg;
9552 
9553 #if defined(sun)
9554 		ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) ||
9555 		    (str == NULL && act->dtad_kind == DTRACEACT_PRINTA));
9556 #endif
9557 
9558 		if (str != NULL)
9559 			kmem_free(str, strlen(str) + 1);
9560 	}
9561 
9562 	kmem_free(act, sizeof (dtrace_actdesc_t));
9563 }
9564 
9565 /*
9566  * DTrace ECB Functions
9567  */
9568 static dtrace_ecb_t *
9569 dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe)
9570 {
9571 	dtrace_ecb_t *ecb;
9572 	dtrace_epid_t epid;
9573 
9574 	ASSERT(MUTEX_HELD(&dtrace_lock));
9575 
9576 	ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP);
9577 	ecb->dte_predicate = NULL;
9578 	ecb->dte_probe = probe;
9579 
9580 	/*
9581 	 * The default size is the size of the default action: recording
9582 	 * the epid.
9583 	 */
9584 	ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t);
9585 	ecb->dte_alignment = sizeof (dtrace_epid_t);
9586 
9587 	epid = state->dts_epid++;
9588 
9589 	if (epid - 1 >= state->dts_necbs) {
9590 		dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs;
9591 		int necbs = state->dts_necbs << 1;
9592 
9593 		ASSERT(epid == state->dts_necbs + 1);
9594 
9595 		if (necbs == 0) {
9596 			ASSERT(oecbs == NULL);
9597 			necbs = 1;
9598 		}
9599 
9600 		ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP);
9601 
9602 		if (oecbs != NULL)
9603 			bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs));
9604 
9605 		dtrace_membar_producer();
9606 		state->dts_ecbs = ecbs;
9607 
9608 		if (oecbs != NULL) {
9609 			/*
9610 			 * If this state is active, we must dtrace_sync()
9611 			 * before we can free the old dts_ecbs array:  we're
9612 			 * coming in hot, and there may be active ring
9613 			 * buffer processing (which indexes into the dts_ecbs
9614 			 * array) on another CPU.
9615 			 */
9616 			if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
9617 				dtrace_sync();
9618 
9619 			kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs));
9620 		}
9621 
9622 		dtrace_membar_producer();
9623 		state->dts_necbs = necbs;
9624 	}
9625 
9626 	ecb->dte_state = state;
9627 
9628 	ASSERT(state->dts_ecbs[epid - 1] == NULL);
9629 	dtrace_membar_producer();
9630 	state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb;
9631 
9632 	return (ecb);
9633 }
9634 
9635 static void
9636 dtrace_ecb_enable(dtrace_ecb_t *ecb)
9637 {
9638 	dtrace_probe_t *probe = ecb->dte_probe;
9639 
9640 	ASSERT(MUTEX_HELD(&cpu_lock));
9641 	ASSERT(MUTEX_HELD(&dtrace_lock));
9642 	ASSERT(ecb->dte_next == NULL);
9643 
9644 	if (probe == NULL) {
9645 		/*
9646 		 * This is the NULL probe -- there's nothing to do.
9647 		 */
9648 		return;
9649 	}
9650 
9651 	if (probe->dtpr_ecb == NULL) {
9652 		dtrace_provider_t *prov = probe->dtpr_provider;
9653 
9654 		/*
9655 		 * We're the first ECB on this probe.
9656 		 */
9657 		probe->dtpr_ecb = probe->dtpr_ecb_last = ecb;
9658 
9659 		if (ecb->dte_predicate != NULL)
9660 			probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid;
9661 
9662 		prov->dtpv_pops.dtps_enable(prov->dtpv_arg,
9663 		    probe->dtpr_id, probe->dtpr_arg);
9664 	} else {
9665 		/*
9666 		 * This probe is already active.  Swing the last pointer to
9667 		 * point to the new ECB, and issue a dtrace_sync() to assure
9668 		 * that all CPUs have seen the change.
9669 		 */
9670 		ASSERT(probe->dtpr_ecb_last != NULL);
9671 		probe->dtpr_ecb_last->dte_next = ecb;
9672 		probe->dtpr_ecb_last = ecb;
9673 		probe->dtpr_predcache = 0;
9674 
9675 		dtrace_sync();
9676 	}
9677 }
9678 
9679 static void
9680 dtrace_ecb_resize(dtrace_ecb_t *ecb)
9681 {
9682 	uint32_t maxalign = sizeof (dtrace_epid_t);
9683 	uint32_t align = sizeof (uint8_t), offs, diff;
9684 	dtrace_action_t *act;
9685 	int wastuple = 0;
9686 	uint32_t aggbase = UINT32_MAX;
9687 	dtrace_state_t *state = ecb->dte_state;
9688 
9689 	/*
9690 	 * If we record anything, we always record the epid.  (And we always
9691 	 * record it first.)
9692 	 */
9693 	offs = sizeof (dtrace_epid_t);
9694 	ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t);
9695 
9696 	for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
9697 		dtrace_recdesc_t *rec = &act->dta_rec;
9698 
9699 		if ((align = rec->dtrd_alignment) > maxalign)
9700 			maxalign = align;
9701 
9702 		if (!wastuple && act->dta_intuple) {
9703 			/*
9704 			 * This is the first record in a tuple.  Align the
9705 			 * offset to be at offset 4 in an 8-byte aligned
9706 			 * block.
9707 			 */
9708 			diff = offs + sizeof (dtrace_aggid_t);
9709 
9710 			if ((diff = (diff & (sizeof (uint64_t) - 1))))
9711 				offs += sizeof (uint64_t) - diff;
9712 
9713 			aggbase = offs - sizeof (dtrace_aggid_t);
9714 			ASSERT(!(aggbase & (sizeof (uint64_t) - 1)));
9715 		}
9716 
9717 		/*LINTED*/
9718 		if (rec->dtrd_size != 0 && (diff = (offs & (align - 1)))) {
9719 			/*
9720 			 * The current offset is not properly aligned; align it.
9721 			 */
9722 			offs += align - diff;
9723 		}
9724 
9725 		rec->dtrd_offset = offs;
9726 
9727 		if (offs + rec->dtrd_size > ecb->dte_needed) {
9728 			ecb->dte_needed = offs + rec->dtrd_size;
9729 
9730 			if (ecb->dte_needed > state->dts_needed)
9731 				state->dts_needed = ecb->dte_needed;
9732 		}
9733 
9734 		if (DTRACEACT_ISAGG(act->dta_kind)) {
9735 			dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
9736 			dtrace_action_t *first = agg->dtag_first, *prev;
9737 
9738 			ASSERT(rec->dtrd_size != 0 && first != NULL);
9739 			ASSERT(wastuple);
9740 			ASSERT(aggbase != UINT32_MAX);
9741 
9742 			agg->dtag_base = aggbase;
9743 
9744 			while ((prev = first->dta_prev) != NULL &&
9745 			    DTRACEACT_ISAGG(prev->dta_kind)) {
9746 				agg = (dtrace_aggregation_t *)prev;
9747 				first = agg->dtag_first;
9748 			}
9749 
9750 			if (prev != NULL) {
9751 				offs = prev->dta_rec.dtrd_offset +
9752 				    prev->dta_rec.dtrd_size;
9753 			} else {
9754 				offs = sizeof (dtrace_epid_t);
9755 			}
9756 			wastuple = 0;
9757 		} else {
9758 			if (!act->dta_intuple)
9759 				ecb->dte_size = offs + rec->dtrd_size;
9760 
9761 			offs += rec->dtrd_size;
9762 		}
9763 
9764 		wastuple = act->dta_intuple;
9765 	}
9766 
9767 	if ((act = ecb->dte_action) != NULL &&
9768 	    !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) &&
9769 	    ecb->dte_size == sizeof (dtrace_epid_t)) {
9770 		/*
9771 		 * If the size is still sizeof (dtrace_epid_t), then all
9772 		 * actions store no data; set the size to 0.
9773 		 */
9774 		ecb->dte_alignment = maxalign;
9775 		ecb->dte_size = 0;
9776 
9777 		/*
9778 		 * If the needed space is still sizeof (dtrace_epid_t), then
9779 		 * all actions need no additional space; set the needed
9780 		 * size to 0.
9781 		 */
9782 		if (ecb->dte_needed == sizeof (dtrace_epid_t))
9783 			ecb->dte_needed = 0;
9784 
9785 		return;
9786 	}
9787 
9788 	/*
9789 	 * Set our alignment, and make sure that the dte_size and dte_needed
9790 	 * are aligned to the size of an EPID.
9791 	 */
9792 	ecb->dte_alignment = maxalign;
9793 	ecb->dte_size = (ecb->dte_size + (sizeof (dtrace_epid_t) - 1)) &
9794 	    ~(sizeof (dtrace_epid_t) - 1);
9795 	ecb->dte_needed = (ecb->dte_needed + (sizeof (dtrace_epid_t) - 1)) &
9796 	    ~(sizeof (dtrace_epid_t) - 1);
9797 	ASSERT(ecb->dte_size <= ecb->dte_needed);
9798 }
9799 
9800 static dtrace_action_t *
9801 dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
9802 {
9803 	dtrace_aggregation_t *agg;
9804 	size_t size = sizeof (uint64_t);
9805 	int ntuple = desc->dtad_ntuple;
9806 	dtrace_action_t *act;
9807 	dtrace_recdesc_t *frec;
9808 	dtrace_aggid_t aggid;
9809 	dtrace_state_t *state = ecb->dte_state;
9810 
9811 	agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP);
9812 	agg->dtag_ecb = ecb;
9813 
9814 	ASSERT(DTRACEACT_ISAGG(desc->dtad_kind));
9815 
9816 	switch (desc->dtad_kind) {
9817 	case DTRACEAGG_MIN:
9818 		agg->dtag_initial = INT64_MAX;
9819 		agg->dtag_aggregate = dtrace_aggregate_min;
9820 		break;
9821 
9822 	case DTRACEAGG_MAX:
9823 		agg->dtag_initial = INT64_MIN;
9824 		agg->dtag_aggregate = dtrace_aggregate_max;
9825 		break;
9826 
9827 	case DTRACEAGG_COUNT:
9828 		agg->dtag_aggregate = dtrace_aggregate_count;
9829 		break;
9830 
9831 	case DTRACEAGG_QUANTIZE:
9832 		agg->dtag_aggregate = dtrace_aggregate_quantize;
9833 		size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) *
9834 		    sizeof (uint64_t);
9835 		break;
9836 
9837 	case DTRACEAGG_LQUANTIZE: {
9838 		uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg);
9839 		uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg);
9840 
9841 		agg->dtag_initial = desc->dtad_arg;
9842 		agg->dtag_aggregate = dtrace_aggregate_lquantize;
9843 
9844 		if (step == 0 || levels == 0)
9845 			goto err;
9846 
9847 		size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t);
9848 		break;
9849 	}
9850 
9851 	case DTRACEAGG_AVG:
9852 		agg->dtag_aggregate = dtrace_aggregate_avg;
9853 		size = sizeof (uint64_t) * 2;
9854 		break;
9855 
9856 	case DTRACEAGG_STDDEV:
9857 		agg->dtag_aggregate = dtrace_aggregate_stddev;
9858 		size = sizeof (uint64_t) * 4;
9859 		break;
9860 
9861 	case DTRACEAGG_SUM:
9862 		agg->dtag_aggregate = dtrace_aggregate_sum;
9863 		break;
9864 
9865 	default:
9866 		goto err;
9867 	}
9868 
9869 	agg->dtag_action.dta_rec.dtrd_size = size;
9870 
9871 	if (ntuple == 0)
9872 		goto err;
9873 
9874 	/*
9875 	 * We must make sure that we have enough actions for the n-tuple.
9876 	 */
9877 	for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) {
9878 		if (DTRACEACT_ISAGG(act->dta_kind))
9879 			break;
9880 
9881 		if (--ntuple == 0) {
9882 			/*
9883 			 * This is the action with which our n-tuple begins.
9884 			 */
9885 			agg->dtag_first = act;
9886 			goto success;
9887 		}
9888 	}
9889 
9890 	/*
9891 	 * This n-tuple is short by ntuple elements.  Return failure.
9892 	 */
9893 	ASSERT(ntuple != 0);
9894 err:
9895 	kmem_free(agg, sizeof (dtrace_aggregation_t));
9896 	return (NULL);
9897 
9898 success:
9899 	/*
9900 	 * If the last action in the tuple has a size of zero, it's actually
9901 	 * an expression argument for the aggregating action.
9902 	 */
9903 	ASSERT(ecb->dte_action_last != NULL);
9904 	act = ecb->dte_action_last;
9905 
9906 	if (act->dta_kind == DTRACEACT_DIFEXPR) {
9907 		ASSERT(act->dta_difo != NULL);
9908 
9909 		if (act->dta_difo->dtdo_rtype.dtdt_size == 0)
9910 			agg->dtag_hasarg = 1;
9911 	}
9912 
9913 	/*
9914 	 * We need to allocate an id for this aggregation.
9915 	 */
9916 #if defined(sun)
9917 	aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1,
9918 	    VM_BESTFIT | VM_SLEEP);
9919 #else
9920 	aggid = alloc_unr(state->dts_aggid_arena);
9921 #endif
9922 
9923 	if (aggid - 1 >= state->dts_naggregations) {
9924 		dtrace_aggregation_t **oaggs = state->dts_aggregations;
9925 		dtrace_aggregation_t **aggs;
9926 		int naggs = state->dts_naggregations << 1;
9927 		int onaggs = state->dts_naggregations;
9928 
9929 		ASSERT(aggid == state->dts_naggregations + 1);
9930 
9931 		if (naggs == 0) {
9932 			ASSERT(oaggs == NULL);
9933 			naggs = 1;
9934 		}
9935 
9936 		aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP);
9937 
9938 		if (oaggs != NULL) {
9939 			bcopy(oaggs, aggs, onaggs * sizeof (*aggs));
9940 			kmem_free(oaggs, onaggs * sizeof (*aggs));
9941 		}
9942 
9943 		state->dts_aggregations = aggs;
9944 		state->dts_naggregations = naggs;
9945 	}
9946 
9947 	ASSERT(state->dts_aggregations[aggid - 1] == NULL);
9948 	state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg;
9949 
9950 	frec = &agg->dtag_first->dta_rec;
9951 	if (frec->dtrd_alignment < sizeof (dtrace_aggid_t))
9952 		frec->dtrd_alignment = sizeof (dtrace_aggid_t);
9953 
9954 	for (act = agg->dtag_first; act != NULL; act = act->dta_next) {
9955 		ASSERT(!act->dta_intuple);
9956 		act->dta_intuple = 1;
9957 	}
9958 
9959 	return (&agg->dtag_action);
9960 }
9961 
9962 static void
9963 dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act)
9964 {
9965 	dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
9966 	dtrace_state_t *state = ecb->dte_state;
9967 	dtrace_aggid_t aggid = agg->dtag_id;
9968 
9969 	ASSERT(DTRACEACT_ISAGG(act->dta_kind));
9970 #if defined(sun)
9971 	vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1);
9972 #else
9973 	free_unr(state->dts_aggid_arena, aggid);
9974 #endif
9975 
9976 	ASSERT(state->dts_aggregations[aggid - 1] == agg);
9977 	state->dts_aggregations[aggid - 1] = NULL;
9978 
9979 	kmem_free(agg, sizeof (dtrace_aggregation_t));
9980 }
9981 
9982 static int
9983 dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
9984 {
9985 	dtrace_action_t *action, *last;
9986 	dtrace_difo_t *dp = desc->dtad_difo;
9987 	uint32_t size = 0, align = sizeof (uint8_t), mask;
9988 	uint16_t format = 0;
9989 	dtrace_recdesc_t *rec;
9990 	dtrace_state_t *state = ecb->dte_state;
9991 	dtrace_optval_t *opt = state->dts_options, nframes = 0, strsize;
9992 	uint64_t arg = desc->dtad_arg;
9993 
9994 	ASSERT(MUTEX_HELD(&dtrace_lock));
9995 	ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1);
9996 
9997 	if (DTRACEACT_ISAGG(desc->dtad_kind)) {
9998 		/*
9999 		 * If this is an aggregating action, there must be neither
10000 		 * a speculate nor a commit on the action chain.
10001 		 */
10002 		dtrace_action_t *act;
10003 
10004 		for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
10005 			if (act->dta_kind == DTRACEACT_COMMIT)
10006 				return (EINVAL);
10007 
10008 			if (act->dta_kind == DTRACEACT_SPECULATE)
10009 				return (EINVAL);
10010 		}
10011 
10012 		action = dtrace_ecb_aggregation_create(ecb, desc);
10013 
10014 		if (action == NULL)
10015 			return (EINVAL);
10016 	} else {
10017 		if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) ||
10018 		    (desc->dtad_kind == DTRACEACT_DIFEXPR &&
10019 		    dp != NULL && dp->dtdo_destructive)) {
10020 			state->dts_destructive = 1;
10021 		}
10022 
10023 		switch (desc->dtad_kind) {
10024 		case DTRACEACT_PRINTF:
10025 		case DTRACEACT_PRINTA:
10026 		case DTRACEACT_SYSTEM:
10027 		case DTRACEACT_FREOPEN:
10028 			/*
10029 			 * We know that our arg is a string -- turn it into a
10030 			 * format.
10031 			 */
10032 			if (arg == 0) {
10033 				ASSERT(desc->dtad_kind == DTRACEACT_PRINTA);
10034 				format = 0;
10035 			} else {
10036 				ASSERT(arg != 0);
10037 #if defined(sun)
10038 				ASSERT(arg > KERNELBASE);
10039 #endif
10040 				format = dtrace_format_add(state,
10041 				    (char *)(uintptr_t)arg);
10042 			}
10043 
10044 			/*FALLTHROUGH*/
10045 		case DTRACEACT_LIBACT:
10046 		case DTRACEACT_DIFEXPR:
10047 			if (dp == NULL)
10048 				return (EINVAL);
10049 
10050 			if ((size = dp->dtdo_rtype.dtdt_size) != 0)
10051 				break;
10052 
10053 			if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) {
10054 				if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10055 					return (EINVAL);
10056 
10057 				size = opt[DTRACEOPT_STRSIZE];
10058 			}
10059 
10060 			break;
10061 
10062 		case DTRACEACT_STACK:
10063 			if ((nframes = arg) == 0) {
10064 				nframes = opt[DTRACEOPT_STACKFRAMES];
10065 				ASSERT(nframes > 0);
10066 				arg = nframes;
10067 			}
10068 
10069 			size = nframes * sizeof (pc_t);
10070 			break;
10071 
10072 		case DTRACEACT_JSTACK:
10073 			if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0)
10074 				strsize = opt[DTRACEOPT_JSTACKSTRSIZE];
10075 
10076 			if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0)
10077 				nframes = opt[DTRACEOPT_JSTACKFRAMES];
10078 
10079 			arg = DTRACE_USTACK_ARG(nframes, strsize);
10080 
10081 			/*FALLTHROUGH*/
10082 		case DTRACEACT_USTACK:
10083 			if (desc->dtad_kind != DTRACEACT_JSTACK &&
10084 			    (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) {
10085 				strsize = DTRACE_USTACK_STRSIZE(arg);
10086 				nframes = opt[DTRACEOPT_USTACKFRAMES];
10087 				ASSERT(nframes > 0);
10088 				arg = DTRACE_USTACK_ARG(nframes, strsize);
10089 			}
10090 
10091 			/*
10092 			 * Save a slot for the pid.
10093 			 */
10094 			size = (nframes + 1) * sizeof (uint64_t);
10095 			size += DTRACE_USTACK_STRSIZE(arg);
10096 			size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t)));
10097 
10098 			break;
10099 
10100 		case DTRACEACT_SYM:
10101 		case DTRACEACT_MOD:
10102 			if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) !=
10103 			    sizeof (uint64_t)) ||
10104 			    (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10105 				return (EINVAL);
10106 			break;
10107 
10108 		case DTRACEACT_USYM:
10109 		case DTRACEACT_UMOD:
10110 		case DTRACEACT_UADDR:
10111 			if (dp == NULL ||
10112 			    (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) ||
10113 			    (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10114 				return (EINVAL);
10115 
10116 			/*
10117 			 * We have a slot for the pid, plus a slot for the
10118 			 * argument.  To keep things simple (aligned with
10119 			 * bitness-neutral sizing), we store each as a 64-bit
10120 			 * quantity.
10121 			 */
10122 			size = 2 * sizeof (uint64_t);
10123 			break;
10124 
10125 		case DTRACEACT_STOP:
10126 		case DTRACEACT_BREAKPOINT:
10127 		case DTRACEACT_PANIC:
10128 			break;
10129 
10130 		case DTRACEACT_CHILL:
10131 		case DTRACEACT_DISCARD:
10132 		case DTRACEACT_RAISE:
10133 			if (dp == NULL)
10134 				return (EINVAL);
10135 			break;
10136 
10137 		case DTRACEACT_EXIT:
10138 			if (dp == NULL ||
10139 			    (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) ||
10140 			    (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10141 				return (EINVAL);
10142 			break;
10143 
10144 		case DTRACEACT_SPECULATE:
10145 			if (ecb->dte_size > sizeof (dtrace_epid_t))
10146 				return (EINVAL);
10147 
10148 			if (dp == NULL)
10149 				return (EINVAL);
10150 
10151 			state->dts_speculates = 1;
10152 			break;
10153 
10154 		case DTRACEACT_PRINTM:
10155 		    	size = dp->dtdo_rtype.dtdt_size;
10156 			break;
10157 
10158 		case DTRACEACT_PRINTT:
10159 		    	size = dp->dtdo_rtype.dtdt_size;
10160 			break;
10161 
10162 		case DTRACEACT_COMMIT: {
10163 			dtrace_action_t *act = ecb->dte_action;
10164 
10165 			for (; act != NULL; act = act->dta_next) {
10166 				if (act->dta_kind == DTRACEACT_COMMIT)
10167 					return (EINVAL);
10168 			}
10169 
10170 			if (dp == NULL)
10171 				return (EINVAL);
10172 			break;
10173 		}
10174 
10175 		default:
10176 			return (EINVAL);
10177 		}
10178 
10179 		if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) {
10180 			/*
10181 			 * If this is a data-storing action or a speculate,
10182 			 * we must be sure that there isn't a commit on the
10183 			 * action chain.
10184 			 */
10185 			dtrace_action_t *act = ecb->dte_action;
10186 
10187 			for (; act != NULL; act = act->dta_next) {
10188 				if (act->dta_kind == DTRACEACT_COMMIT)
10189 					return (EINVAL);
10190 			}
10191 		}
10192 
10193 		action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP);
10194 		action->dta_rec.dtrd_size = size;
10195 	}
10196 
10197 	action->dta_refcnt = 1;
10198 	rec = &action->dta_rec;
10199 	size = rec->dtrd_size;
10200 
10201 	for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) {
10202 		if (!(size & mask)) {
10203 			align = mask + 1;
10204 			break;
10205 		}
10206 	}
10207 
10208 	action->dta_kind = desc->dtad_kind;
10209 
10210 	if ((action->dta_difo = dp) != NULL)
10211 		dtrace_difo_hold(dp);
10212 
10213 	rec->dtrd_action = action->dta_kind;
10214 	rec->dtrd_arg = arg;
10215 	rec->dtrd_uarg = desc->dtad_uarg;
10216 	rec->dtrd_alignment = (uint16_t)align;
10217 	rec->dtrd_format = format;
10218 
10219 	if ((last = ecb->dte_action_last) != NULL) {
10220 		ASSERT(ecb->dte_action != NULL);
10221 		action->dta_prev = last;
10222 		last->dta_next = action;
10223 	} else {
10224 		ASSERT(ecb->dte_action == NULL);
10225 		ecb->dte_action = action;
10226 	}
10227 
10228 	ecb->dte_action_last = action;
10229 
10230 	return (0);
10231 }
10232 
10233 static void
10234 dtrace_ecb_action_remove(dtrace_ecb_t *ecb)
10235 {
10236 	dtrace_action_t *act = ecb->dte_action, *next;
10237 	dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate;
10238 	dtrace_difo_t *dp;
10239 	uint16_t format;
10240 
10241 	if (act != NULL && act->dta_refcnt > 1) {
10242 		ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1);
10243 		act->dta_refcnt--;
10244 	} else {
10245 		for (; act != NULL; act = next) {
10246 			next = act->dta_next;
10247 			ASSERT(next != NULL || act == ecb->dte_action_last);
10248 			ASSERT(act->dta_refcnt == 1);
10249 
10250 			if ((format = act->dta_rec.dtrd_format) != 0)
10251 				dtrace_format_remove(ecb->dte_state, format);
10252 
10253 			if ((dp = act->dta_difo) != NULL)
10254 				dtrace_difo_release(dp, vstate);
10255 
10256 			if (DTRACEACT_ISAGG(act->dta_kind)) {
10257 				dtrace_ecb_aggregation_destroy(ecb, act);
10258 			} else {
10259 				kmem_free(act, sizeof (dtrace_action_t));
10260 			}
10261 		}
10262 	}
10263 
10264 	ecb->dte_action = NULL;
10265 	ecb->dte_action_last = NULL;
10266 	ecb->dte_size = sizeof (dtrace_epid_t);
10267 }
10268 
10269 static void
10270 dtrace_ecb_disable(dtrace_ecb_t *ecb)
10271 {
10272 	/*
10273 	 * We disable the ECB by removing it from its probe.
10274 	 */
10275 	dtrace_ecb_t *pecb, *prev = NULL;
10276 	dtrace_probe_t *probe = ecb->dte_probe;
10277 
10278 	ASSERT(MUTEX_HELD(&dtrace_lock));
10279 
10280 	if (probe == NULL) {
10281 		/*
10282 		 * This is the NULL probe; there is nothing to disable.
10283 		 */
10284 		return;
10285 	}
10286 
10287 	for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) {
10288 		if (pecb == ecb)
10289 			break;
10290 		prev = pecb;
10291 	}
10292 
10293 	ASSERT(pecb != NULL);
10294 
10295 	if (prev == NULL) {
10296 		probe->dtpr_ecb = ecb->dte_next;
10297 	} else {
10298 		prev->dte_next = ecb->dte_next;
10299 	}
10300 
10301 	if (ecb == probe->dtpr_ecb_last) {
10302 		ASSERT(ecb->dte_next == NULL);
10303 		probe->dtpr_ecb_last = prev;
10304 	}
10305 
10306 	/*
10307 	 * The ECB has been disconnected from the probe; now sync to assure
10308 	 * that all CPUs have seen the change before returning.
10309 	 */
10310 	dtrace_sync();
10311 
10312 	if (probe->dtpr_ecb == NULL) {
10313 		/*
10314 		 * That was the last ECB on the probe; clear the predicate
10315 		 * cache ID for the probe, disable it and sync one more time
10316 		 * to assure that we'll never hit it again.
10317 		 */
10318 		dtrace_provider_t *prov = probe->dtpr_provider;
10319 
10320 		ASSERT(ecb->dte_next == NULL);
10321 		ASSERT(probe->dtpr_ecb_last == NULL);
10322 		probe->dtpr_predcache = DTRACE_CACHEIDNONE;
10323 		prov->dtpv_pops.dtps_disable(prov->dtpv_arg,
10324 		    probe->dtpr_id, probe->dtpr_arg);
10325 		dtrace_sync();
10326 	} else {
10327 		/*
10328 		 * There is at least one ECB remaining on the probe.  If there
10329 		 * is _exactly_ one, set the probe's predicate cache ID to be
10330 		 * the predicate cache ID of the remaining ECB.
10331 		 */
10332 		ASSERT(probe->dtpr_ecb_last != NULL);
10333 		ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE);
10334 
10335 		if (probe->dtpr_ecb == probe->dtpr_ecb_last) {
10336 			dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate;
10337 
10338 			ASSERT(probe->dtpr_ecb->dte_next == NULL);
10339 
10340 			if (p != NULL)
10341 				probe->dtpr_predcache = p->dtp_cacheid;
10342 		}
10343 
10344 		ecb->dte_next = NULL;
10345 	}
10346 }
10347 
10348 static void
10349 dtrace_ecb_destroy(dtrace_ecb_t *ecb)
10350 {
10351 	dtrace_state_t *state = ecb->dte_state;
10352 	dtrace_vstate_t *vstate = &state->dts_vstate;
10353 	dtrace_predicate_t *pred;
10354 	dtrace_epid_t epid = ecb->dte_epid;
10355 
10356 	ASSERT(MUTEX_HELD(&dtrace_lock));
10357 	ASSERT(ecb->dte_next == NULL);
10358 	ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb);
10359 
10360 	if ((pred = ecb->dte_predicate) != NULL)
10361 		dtrace_predicate_release(pred, vstate);
10362 
10363 	dtrace_ecb_action_remove(ecb);
10364 
10365 	ASSERT(state->dts_ecbs[epid - 1] == ecb);
10366 	state->dts_ecbs[epid - 1] = NULL;
10367 
10368 	kmem_free(ecb, sizeof (dtrace_ecb_t));
10369 }
10370 
10371 static dtrace_ecb_t *
10372 dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe,
10373     dtrace_enabling_t *enab)
10374 {
10375 	dtrace_ecb_t *ecb;
10376 	dtrace_predicate_t *pred;
10377 	dtrace_actdesc_t *act;
10378 	dtrace_provider_t *prov;
10379 	dtrace_ecbdesc_t *desc = enab->dten_current;
10380 
10381 	ASSERT(MUTEX_HELD(&dtrace_lock));
10382 	ASSERT(state != NULL);
10383 
10384 	ecb = dtrace_ecb_add(state, probe);
10385 	ecb->dte_uarg = desc->dted_uarg;
10386 
10387 	if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) {
10388 		dtrace_predicate_hold(pred);
10389 		ecb->dte_predicate = pred;
10390 	}
10391 
10392 	if (probe != NULL) {
10393 		/*
10394 		 * If the provider shows more leg than the consumer is old
10395 		 * enough to see, we need to enable the appropriate implicit
10396 		 * predicate bits to prevent the ecb from activating at
10397 		 * revealing times.
10398 		 *
10399 		 * Providers specifying DTRACE_PRIV_USER at register time
10400 		 * are stating that they need the /proc-style privilege
10401 		 * model to be enforced, and this is what DTRACE_COND_OWNER
10402 		 * and DTRACE_COND_ZONEOWNER will then do at probe time.
10403 		 */
10404 		prov = probe->dtpr_provider;
10405 		if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) &&
10406 		    (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
10407 			ecb->dte_cond |= DTRACE_COND_OWNER;
10408 
10409 		if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) &&
10410 		    (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
10411 			ecb->dte_cond |= DTRACE_COND_ZONEOWNER;
10412 
10413 		/*
10414 		 * If the provider shows us kernel innards and the user
10415 		 * is lacking sufficient privilege, enable the
10416 		 * DTRACE_COND_USERMODE implicit predicate.
10417 		 */
10418 		if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) &&
10419 		    (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL))
10420 			ecb->dte_cond |= DTRACE_COND_USERMODE;
10421 	}
10422 
10423 	if (dtrace_ecb_create_cache != NULL) {
10424 		/*
10425 		 * If we have a cached ecb, we'll use its action list instead
10426 		 * of creating our own (saving both time and space).
10427 		 */
10428 		dtrace_ecb_t *cached = dtrace_ecb_create_cache;
10429 		dtrace_action_t *act = cached->dte_action;
10430 
10431 		if (act != NULL) {
10432 			ASSERT(act->dta_refcnt > 0);
10433 			act->dta_refcnt++;
10434 			ecb->dte_action = act;
10435 			ecb->dte_action_last = cached->dte_action_last;
10436 			ecb->dte_needed = cached->dte_needed;
10437 			ecb->dte_size = cached->dte_size;
10438 			ecb->dte_alignment = cached->dte_alignment;
10439 		}
10440 
10441 		return (ecb);
10442 	}
10443 
10444 	for (act = desc->dted_action; act != NULL; act = act->dtad_next) {
10445 		if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) {
10446 			dtrace_ecb_destroy(ecb);
10447 			return (NULL);
10448 		}
10449 	}
10450 
10451 	dtrace_ecb_resize(ecb);
10452 
10453 	return (dtrace_ecb_create_cache = ecb);
10454 }
10455 
10456 static int
10457 dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg)
10458 {
10459 	dtrace_ecb_t *ecb;
10460 	dtrace_enabling_t *enab = arg;
10461 	dtrace_state_t *state = enab->dten_vstate->dtvs_state;
10462 
10463 	ASSERT(state != NULL);
10464 
10465 	if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) {
10466 		/*
10467 		 * This probe was created in a generation for which this
10468 		 * enabling has previously created ECBs; we don't want to
10469 		 * enable it again, so just kick out.
10470 		 */
10471 		return (DTRACE_MATCH_NEXT);
10472 	}
10473 
10474 	if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL)
10475 		return (DTRACE_MATCH_DONE);
10476 
10477 	dtrace_ecb_enable(ecb);
10478 	return (DTRACE_MATCH_NEXT);
10479 }
10480 
10481 static dtrace_ecb_t *
10482 dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id)
10483 {
10484 	dtrace_ecb_t *ecb;
10485 
10486 	ASSERT(MUTEX_HELD(&dtrace_lock));
10487 
10488 	if (id == 0 || id > state->dts_necbs)
10489 		return (NULL);
10490 
10491 	ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL);
10492 	ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id);
10493 
10494 	return (state->dts_ecbs[id - 1]);
10495 }
10496 
10497 static dtrace_aggregation_t *
10498 dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id)
10499 {
10500 	dtrace_aggregation_t *agg;
10501 
10502 	ASSERT(MUTEX_HELD(&dtrace_lock));
10503 
10504 	if (id == 0 || id > state->dts_naggregations)
10505 		return (NULL);
10506 
10507 	ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL);
10508 	ASSERT((agg = state->dts_aggregations[id - 1]) == NULL ||
10509 	    agg->dtag_id == id);
10510 
10511 	return (state->dts_aggregations[id - 1]);
10512 }
10513 
10514 /*
10515  * DTrace Buffer Functions
10516  *
10517  * The following functions manipulate DTrace buffers.  Most of these functions
10518  * are called in the context of establishing or processing consumer state;
10519  * exceptions are explicitly noted.
10520  */
10521 
10522 /*
10523  * Note:  called from cross call context.  This function switches the two
10524  * buffers on a given CPU.  The atomicity of this operation is assured by
10525  * disabling interrupts while the actual switch takes place; the disabling of
10526  * interrupts serializes the execution with any execution of dtrace_probe() on
10527  * the same CPU.
10528  */
10529 static void
10530 dtrace_buffer_switch(dtrace_buffer_t *buf)
10531 {
10532 	caddr_t tomax = buf->dtb_tomax;
10533 	caddr_t xamot = buf->dtb_xamot;
10534 	dtrace_icookie_t cookie;
10535 
10536 	ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
10537 	ASSERT(!(buf->dtb_flags & DTRACEBUF_RING));
10538 
10539 	cookie = dtrace_interrupt_disable();
10540 	buf->dtb_tomax = xamot;
10541 	buf->dtb_xamot = tomax;
10542 	buf->dtb_xamot_drops = buf->dtb_drops;
10543 	buf->dtb_xamot_offset = buf->dtb_offset;
10544 	buf->dtb_xamot_errors = buf->dtb_errors;
10545 	buf->dtb_xamot_flags = buf->dtb_flags;
10546 	buf->dtb_offset = 0;
10547 	buf->dtb_drops = 0;
10548 	buf->dtb_errors = 0;
10549 	buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED);
10550 	dtrace_interrupt_enable(cookie);
10551 }
10552 
10553 /*
10554  * Note:  called from cross call context.  This function activates a buffer
10555  * on a CPU.  As with dtrace_buffer_switch(), the atomicity of the operation
10556  * is guaranteed by the disabling of interrupts.
10557  */
10558 static void
10559 dtrace_buffer_activate(dtrace_state_t *state)
10560 {
10561 	dtrace_buffer_t *buf;
10562 	dtrace_icookie_t cookie = dtrace_interrupt_disable();
10563 
10564 	buf = &state->dts_buffer[curcpu];
10565 
10566 	if (buf->dtb_tomax != NULL) {
10567 		/*
10568 		 * We might like to assert that the buffer is marked inactive,
10569 		 * but this isn't necessarily true:  the buffer for the CPU
10570 		 * that processes the BEGIN probe has its buffer activated
10571 		 * manually.  In this case, we take the (harmless) action
10572 		 * re-clearing the bit INACTIVE bit.
10573 		 */
10574 		buf->dtb_flags &= ~DTRACEBUF_INACTIVE;
10575 	}
10576 
10577 	dtrace_interrupt_enable(cookie);
10578 }
10579 
10580 static int
10581 dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags,
10582     processorid_t cpu)
10583 {
10584 #if defined(sun)
10585 	cpu_t *cp;
10586 #endif
10587 	dtrace_buffer_t *buf;
10588 
10589 #if defined(sun)
10590 	ASSERT(MUTEX_HELD(&cpu_lock));
10591 	ASSERT(MUTEX_HELD(&dtrace_lock));
10592 
10593 	if (size > dtrace_nonroot_maxsize &&
10594 	    !PRIV_POLICY_CHOICE(CRED(), PRIV_ALL, B_FALSE))
10595 		return (EFBIG);
10596 
10597 	cp = cpu_list;
10598 
10599 	do {
10600 		if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id)
10601 			continue;
10602 
10603 		buf = &bufs[cp->cpu_id];
10604 
10605 		/*
10606 		 * If there is already a buffer allocated for this CPU, it
10607 		 * is only possible that this is a DR event.  In this case,
10608 		 */
10609 		if (buf->dtb_tomax != NULL) {
10610 			ASSERT(buf->dtb_size == size);
10611 			continue;
10612 		}
10613 
10614 		ASSERT(buf->dtb_xamot == NULL);
10615 
10616 		if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10617 			goto err;
10618 
10619 		buf->dtb_size = size;
10620 		buf->dtb_flags = flags;
10621 		buf->dtb_offset = 0;
10622 		buf->dtb_drops = 0;
10623 
10624 		if (flags & DTRACEBUF_NOSWITCH)
10625 			continue;
10626 
10627 		if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10628 			goto err;
10629 	} while ((cp = cp->cpu_next) != cpu_list);
10630 
10631 	return (0);
10632 
10633 err:
10634 	cp = cpu_list;
10635 
10636 	do {
10637 		if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id)
10638 			continue;
10639 
10640 		buf = &bufs[cp->cpu_id];
10641 
10642 		if (buf->dtb_xamot != NULL) {
10643 			ASSERT(buf->dtb_tomax != NULL);
10644 			ASSERT(buf->dtb_size == size);
10645 			kmem_free(buf->dtb_xamot, size);
10646 		}
10647 
10648 		if (buf->dtb_tomax != NULL) {
10649 			ASSERT(buf->dtb_size == size);
10650 			kmem_free(buf->dtb_tomax, size);
10651 		}
10652 
10653 		buf->dtb_tomax = NULL;
10654 		buf->dtb_xamot = NULL;
10655 		buf->dtb_size = 0;
10656 	} while ((cp = cp->cpu_next) != cpu_list);
10657 
10658 	return (ENOMEM);
10659 #else
10660 	int i;
10661 
10662 #if defined(__amd64__)
10663 	/*
10664 	 * FreeBSD isn't good at limiting the amount of memory we
10665 	 * ask to malloc, so let's place a limit here before trying
10666 	 * to do something that might well end in tears at bedtime.
10667 	 */
10668 	if (size > physmem * PAGE_SIZE / (128 * (mp_maxid + 1)))
10669 		return(ENOMEM);
10670 #endif
10671 
10672 	ASSERT(MUTEX_HELD(&dtrace_lock));
10673 	CPU_FOREACH(i) {
10674 		if (cpu != DTRACE_CPUALL && cpu != i)
10675 			continue;
10676 
10677 		buf = &bufs[i];
10678 
10679 		/*
10680 		 * If there is already a buffer allocated for this CPU, it
10681 		 * is only possible that this is a DR event.  In this case,
10682 		 * the buffer size must match our specified size.
10683 		 */
10684 		if (buf->dtb_tomax != NULL) {
10685 			ASSERT(buf->dtb_size == size);
10686 			continue;
10687 		}
10688 
10689 		ASSERT(buf->dtb_xamot == NULL);
10690 
10691 		if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10692 			goto err;
10693 
10694 		buf->dtb_size = size;
10695 		buf->dtb_flags = flags;
10696 		buf->dtb_offset = 0;
10697 		buf->dtb_drops = 0;
10698 
10699 		if (flags & DTRACEBUF_NOSWITCH)
10700 			continue;
10701 
10702 		if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10703 			goto err;
10704 	}
10705 
10706 	return (0);
10707 
10708 err:
10709 	/*
10710 	 * Error allocating memory, so free the buffers that were
10711 	 * allocated before the failed allocation.
10712 	 */
10713 	CPU_FOREACH(i) {
10714 		if (cpu != DTRACE_CPUALL && cpu != i)
10715 			continue;
10716 
10717 		buf = &bufs[i];
10718 
10719 		if (buf->dtb_xamot != NULL) {
10720 			ASSERT(buf->dtb_tomax != NULL);
10721 			ASSERT(buf->dtb_size == size);
10722 			kmem_free(buf->dtb_xamot, size);
10723 		}
10724 
10725 		if (buf->dtb_tomax != NULL) {
10726 			ASSERT(buf->dtb_size == size);
10727 			kmem_free(buf->dtb_tomax, size);
10728 		}
10729 
10730 		buf->dtb_tomax = NULL;
10731 		buf->dtb_xamot = NULL;
10732 		buf->dtb_size = 0;
10733 
10734 	}
10735 
10736 	return (ENOMEM);
10737 #endif
10738 }
10739 
10740 /*
10741  * Note:  called from probe context.  This function just increments the drop
10742  * count on a buffer.  It has been made a function to allow for the
10743  * possibility of understanding the source of mysterious drop counts.  (A
10744  * problem for which one may be particularly disappointed that DTrace cannot
10745  * be used to understand DTrace.)
10746  */
10747 static void
10748 dtrace_buffer_drop(dtrace_buffer_t *buf)
10749 {
10750 	buf->dtb_drops++;
10751 }
10752 
10753 /*
10754  * Note:  called from probe context.  This function is called to reserve space
10755  * in a buffer.  If mstate is non-NULL, sets the scratch base and size in the
10756  * mstate.  Returns the new offset in the buffer, or a negative value if an
10757  * error has occurred.
10758  */
10759 static intptr_t
10760 dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align,
10761     dtrace_state_t *state, dtrace_mstate_t *mstate)
10762 {
10763 	intptr_t offs = buf->dtb_offset, soffs;
10764 	intptr_t woffs;
10765 	caddr_t tomax;
10766 	size_t total;
10767 
10768 	if (buf->dtb_flags & DTRACEBUF_INACTIVE)
10769 		return (-1);
10770 
10771 	if ((tomax = buf->dtb_tomax) == NULL) {
10772 		dtrace_buffer_drop(buf);
10773 		return (-1);
10774 	}
10775 
10776 	if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) {
10777 		while (offs & (align - 1)) {
10778 			/*
10779 			 * Assert that our alignment is off by a number which
10780 			 * is itself sizeof (uint32_t) aligned.
10781 			 */
10782 			ASSERT(!((align - (offs & (align - 1))) &
10783 			    (sizeof (uint32_t) - 1)));
10784 			DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
10785 			offs += sizeof (uint32_t);
10786 		}
10787 
10788 		if ((soffs = offs + needed) > buf->dtb_size) {
10789 			dtrace_buffer_drop(buf);
10790 			return (-1);
10791 		}
10792 
10793 		if (mstate == NULL)
10794 			return (offs);
10795 
10796 		mstate->dtms_scratch_base = (uintptr_t)tomax + soffs;
10797 		mstate->dtms_scratch_size = buf->dtb_size - soffs;
10798 		mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
10799 
10800 		return (offs);
10801 	}
10802 
10803 	if (buf->dtb_flags & DTRACEBUF_FILL) {
10804 		if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN &&
10805 		    (buf->dtb_flags & DTRACEBUF_FULL))
10806 			return (-1);
10807 		goto out;
10808 	}
10809 
10810 	total = needed + (offs & (align - 1));
10811 
10812 	/*
10813 	 * For a ring buffer, life is quite a bit more complicated.  Before
10814 	 * we can store any padding, we need to adjust our wrapping offset.
10815 	 * (If we've never before wrapped or we're not about to, no adjustment
10816 	 * is required.)
10817 	 */
10818 	if ((buf->dtb_flags & DTRACEBUF_WRAPPED) ||
10819 	    offs + total > buf->dtb_size) {
10820 		woffs = buf->dtb_xamot_offset;
10821 
10822 		if (offs + total > buf->dtb_size) {
10823 			/*
10824 			 * We can't fit in the end of the buffer.  First, a
10825 			 * sanity check that we can fit in the buffer at all.
10826 			 */
10827 			if (total > buf->dtb_size) {
10828 				dtrace_buffer_drop(buf);
10829 				return (-1);
10830 			}
10831 
10832 			/*
10833 			 * We're going to be storing at the top of the buffer,
10834 			 * so now we need to deal with the wrapped offset.  We
10835 			 * only reset our wrapped offset to 0 if it is
10836 			 * currently greater than the current offset.  If it
10837 			 * is less than the current offset, it is because a
10838 			 * previous allocation induced a wrap -- but the
10839 			 * allocation didn't subsequently take the space due
10840 			 * to an error or false predicate evaluation.  In this
10841 			 * case, we'll just leave the wrapped offset alone: if
10842 			 * the wrapped offset hasn't been advanced far enough
10843 			 * for this allocation, it will be adjusted in the
10844 			 * lower loop.
10845 			 */
10846 			if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
10847 				if (woffs >= offs)
10848 					woffs = 0;
10849 			} else {
10850 				woffs = 0;
10851 			}
10852 
10853 			/*
10854 			 * Now we know that we're going to be storing to the
10855 			 * top of the buffer and that there is room for us
10856 			 * there.  We need to clear the buffer from the current
10857 			 * offset to the end (there may be old gunk there).
10858 			 */
10859 			while (offs < buf->dtb_size)
10860 				tomax[offs++] = 0;
10861 
10862 			/*
10863 			 * We need to set our offset to zero.  And because we
10864 			 * are wrapping, we need to set the bit indicating as
10865 			 * much.  We can also adjust our needed space back
10866 			 * down to the space required by the ECB -- we know
10867 			 * that the top of the buffer is aligned.
10868 			 */
10869 			offs = 0;
10870 			total = needed;
10871 			buf->dtb_flags |= DTRACEBUF_WRAPPED;
10872 		} else {
10873 			/*
10874 			 * There is room for us in the buffer, so we simply
10875 			 * need to check the wrapped offset.
10876 			 */
10877 			if (woffs < offs) {
10878 				/*
10879 				 * The wrapped offset is less than the offset.
10880 				 * This can happen if we allocated buffer space
10881 				 * that induced a wrap, but then we didn't
10882 				 * subsequently take the space due to an error
10883 				 * or false predicate evaluation.  This is
10884 				 * okay; we know that _this_ allocation isn't
10885 				 * going to induce a wrap.  We still can't
10886 				 * reset the wrapped offset to be zero,
10887 				 * however: the space may have been trashed in
10888 				 * the previous failed probe attempt.  But at
10889 				 * least the wrapped offset doesn't need to
10890 				 * be adjusted at all...
10891 				 */
10892 				goto out;
10893 			}
10894 		}
10895 
10896 		while (offs + total > woffs) {
10897 			dtrace_epid_t epid = *(uint32_t *)(tomax + woffs);
10898 			size_t size;
10899 
10900 			if (epid == DTRACE_EPIDNONE) {
10901 				size = sizeof (uint32_t);
10902 			} else {
10903 				ASSERT(epid <= state->dts_necbs);
10904 				ASSERT(state->dts_ecbs[epid - 1] != NULL);
10905 
10906 				size = state->dts_ecbs[epid - 1]->dte_size;
10907 			}
10908 
10909 			ASSERT(woffs + size <= buf->dtb_size);
10910 			ASSERT(size != 0);
10911 
10912 			if (woffs + size == buf->dtb_size) {
10913 				/*
10914 				 * We've reached the end of the buffer; we want
10915 				 * to set the wrapped offset to 0 and break
10916 				 * out.  However, if the offs is 0, then we're
10917 				 * in a strange edge-condition:  the amount of
10918 				 * space that we want to reserve plus the size
10919 				 * of the record that we're overwriting is
10920 				 * greater than the size of the buffer.  This
10921 				 * is problematic because if we reserve the
10922 				 * space but subsequently don't consume it (due
10923 				 * to a failed predicate or error) the wrapped
10924 				 * offset will be 0 -- yet the EPID at offset 0
10925 				 * will not be committed.  This situation is
10926 				 * relatively easy to deal with:  if we're in
10927 				 * this case, the buffer is indistinguishable
10928 				 * from one that hasn't wrapped; we need only
10929 				 * finish the job by clearing the wrapped bit,
10930 				 * explicitly setting the offset to be 0, and
10931 				 * zero'ing out the old data in the buffer.
10932 				 */
10933 				if (offs == 0) {
10934 					buf->dtb_flags &= ~DTRACEBUF_WRAPPED;
10935 					buf->dtb_offset = 0;
10936 					woffs = total;
10937 
10938 					while (woffs < buf->dtb_size)
10939 						tomax[woffs++] = 0;
10940 				}
10941 
10942 				woffs = 0;
10943 				break;
10944 			}
10945 
10946 			woffs += size;
10947 		}
10948 
10949 		/*
10950 		 * We have a wrapped offset.  It may be that the wrapped offset
10951 		 * has become zero -- that's okay.
10952 		 */
10953 		buf->dtb_xamot_offset = woffs;
10954 	}
10955 
10956 out:
10957 	/*
10958 	 * Now we can plow the buffer with any necessary padding.
10959 	 */
10960 	while (offs & (align - 1)) {
10961 		/*
10962 		 * Assert that our alignment is off by a number which
10963 		 * is itself sizeof (uint32_t) aligned.
10964 		 */
10965 		ASSERT(!((align - (offs & (align - 1))) &
10966 		    (sizeof (uint32_t) - 1)));
10967 		DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
10968 		offs += sizeof (uint32_t);
10969 	}
10970 
10971 	if (buf->dtb_flags & DTRACEBUF_FILL) {
10972 		if (offs + needed > buf->dtb_size - state->dts_reserve) {
10973 			buf->dtb_flags |= DTRACEBUF_FULL;
10974 			return (-1);
10975 		}
10976 	}
10977 
10978 	if (mstate == NULL)
10979 		return (offs);
10980 
10981 	/*
10982 	 * For ring buffers and fill buffers, the scratch space is always
10983 	 * the inactive buffer.
10984 	 */
10985 	mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot;
10986 	mstate->dtms_scratch_size = buf->dtb_size;
10987 	mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
10988 
10989 	return (offs);
10990 }
10991 
10992 static void
10993 dtrace_buffer_polish(dtrace_buffer_t *buf)
10994 {
10995 	ASSERT(buf->dtb_flags & DTRACEBUF_RING);
10996 	ASSERT(MUTEX_HELD(&dtrace_lock));
10997 
10998 	if (!(buf->dtb_flags & DTRACEBUF_WRAPPED))
10999 		return;
11000 
11001 	/*
11002 	 * We need to polish the ring buffer.  There are three cases:
11003 	 *
11004 	 * - The first (and presumably most common) is that there is no gap
11005 	 *   between the buffer offset and the wrapped offset.  In this case,
11006 	 *   there is nothing in the buffer that isn't valid data; we can
11007 	 *   mark the buffer as polished and return.
11008 	 *
11009 	 * - The second (less common than the first but still more common
11010 	 *   than the third) is that there is a gap between the buffer offset
11011 	 *   and the wrapped offset, and the wrapped offset is larger than the
11012 	 *   buffer offset.  This can happen because of an alignment issue, or
11013 	 *   can happen because of a call to dtrace_buffer_reserve() that
11014 	 *   didn't subsequently consume the buffer space.  In this case,
11015 	 *   we need to zero the data from the buffer offset to the wrapped
11016 	 *   offset.
11017 	 *
11018 	 * - The third (and least common) is that there is a gap between the
11019 	 *   buffer offset and the wrapped offset, but the wrapped offset is
11020 	 *   _less_ than the buffer offset.  This can only happen because a
11021 	 *   call to dtrace_buffer_reserve() induced a wrap, but the space
11022 	 *   was not subsequently consumed.  In this case, we need to zero the
11023 	 *   space from the offset to the end of the buffer _and_ from the
11024 	 *   top of the buffer to the wrapped offset.
11025 	 */
11026 	if (buf->dtb_offset < buf->dtb_xamot_offset) {
11027 		bzero(buf->dtb_tomax + buf->dtb_offset,
11028 		    buf->dtb_xamot_offset - buf->dtb_offset);
11029 	}
11030 
11031 	if (buf->dtb_offset > buf->dtb_xamot_offset) {
11032 		bzero(buf->dtb_tomax + buf->dtb_offset,
11033 		    buf->dtb_size - buf->dtb_offset);
11034 		bzero(buf->dtb_tomax, buf->dtb_xamot_offset);
11035 	}
11036 }
11037 
11038 static void
11039 dtrace_buffer_free(dtrace_buffer_t *bufs)
11040 {
11041 	int i;
11042 
11043 	for (i = 0; i < NCPU; i++) {
11044 		dtrace_buffer_t *buf = &bufs[i];
11045 
11046 		if (buf->dtb_tomax == NULL) {
11047 			ASSERT(buf->dtb_xamot == NULL);
11048 			ASSERT(buf->dtb_size == 0);
11049 			continue;
11050 		}
11051 
11052 		if (buf->dtb_xamot != NULL) {
11053 			ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
11054 			kmem_free(buf->dtb_xamot, buf->dtb_size);
11055 		}
11056 
11057 		kmem_free(buf->dtb_tomax, buf->dtb_size);
11058 		buf->dtb_size = 0;
11059 		buf->dtb_tomax = NULL;
11060 		buf->dtb_xamot = NULL;
11061 	}
11062 }
11063 
11064 /*
11065  * DTrace Enabling Functions
11066  */
11067 static dtrace_enabling_t *
11068 dtrace_enabling_create(dtrace_vstate_t *vstate)
11069 {
11070 	dtrace_enabling_t *enab;
11071 
11072 	enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP);
11073 	enab->dten_vstate = vstate;
11074 
11075 	return (enab);
11076 }
11077 
11078 static void
11079 dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb)
11080 {
11081 	dtrace_ecbdesc_t **ndesc;
11082 	size_t osize, nsize;
11083 
11084 	/*
11085 	 * We can't add to enablings after we've enabled them, or after we've
11086 	 * retained them.
11087 	 */
11088 	ASSERT(enab->dten_probegen == 0);
11089 	ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11090 
11091 	if (enab->dten_ndesc < enab->dten_maxdesc) {
11092 		enab->dten_desc[enab->dten_ndesc++] = ecb;
11093 		return;
11094 	}
11095 
11096 	osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11097 
11098 	if (enab->dten_maxdesc == 0) {
11099 		enab->dten_maxdesc = 1;
11100 	} else {
11101 		enab->dten_maxdesc <<= 1;
11102 	}
11103 
11104 	ASSERT(enab->dten_ndesc < enab->dten_maxdesc);
11105 
11106 	nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11107 	ndesc = kmem_zalloc(nsize, KM_SLEEP);
11108 	bcopy(enab->dten_desc, ndesc, osize);
11109 	if (enab->dten_desc != NULL)
11110 		kmem_free(enab->dten_desc, osize);
11111 
11112 	enab->dten_desc = ndesc;
11113 	enab->dten_desc[enab->dten_ndesc++] = ecb;
11114 }
11115 
11116 static void
11117 dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb,
11118     dtrace_probedesc_t *pd)
11119 {
11120 	dtrace_ecbdesc_t *new;
11121 	dtrace_predicate_t *pred;
11122 	dtrace_actdesc_t *act;
11123 
11124 	/*
11125 	 * We're going to create a new ECB description that matches the
11126 	 * specified ECB in every way, but has the specified probe description.
11127 	 */
11128 	new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
11129 
11130 	if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL)
11131 		dtrace_predicate_hold(pred);
11132 
11133 	for (act = ecb->dted_action; act != NULL; act = act->dtad_next)
11134 		dtrace_actdesc_hold(act);
11135 
11136 	new->dted_action = ecb->dted_action;
11137 	new->dted_pred = ecb->dted_pred;
11138 	new->dted_probe = *pd;
11139 	new->dted_uarg = ecb->dted_uarg;
11140 
11141 	dtrace_enabling_add(enab, new);
11142 }
11143 
11144 static void
11145 dtrace_enabling_dump(dtrace_enabling_t *enab)
11146 {
11147 	int i;
11148 
11149 	for (i = 0; i < enab->dten_ndesc; i++) {
11150 		dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe;
11151 
11152 		cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i,
11153 		    desc->dtpd_provider, desc->dtpd_mod,
11154 		    desc->dtpd_func, desc->dtpd_name);
11155 	}
11156 }
11157 
11158 static void
11159 dtrace_enabling_destroy(dtrace_enabling_t *enab)
11160 {
11161 	int i;
11162 	dtrace_ecbdesc_t *ep;
11163 	dtrace_vstate_t *vstate = enab->dten_vstate;
11164 
11165 	ASSERT(MUTEX_HELD(&dtrace_lock));
11166 
11167 	for (i = 0; i < enab->dten_ndesc; i++) {
11168 		dtrace_actdesc_t *act, *next;
11169 		dtrace_predicate_t *pred;
11170 
11171 		ep = enab->dten_desc[i];
11172 
11173 		if ((pred = ep->dted_pred.dtpdd_predicate) != NULL)
11174 			dtrace_predicate_release(pred, vstate);
11175 
11176 		for (act = ep->dted_action; act != NULL; act = next) {
11177 			next = act->dtad_next;
11178 			dtrace_actdesc_release(act, vstate);
11179 		}
11180 
11181 		kmem_free(ep, sizeof (dtrace_ecbdesc_t));
11182 	}
11183 
11184 	if (enab->dten_desc != NULL)
11185 		kmem_free(enab->dten_desc,
11186 		    enab->dten_maxdesc * sizeof (dtrace_enabling_t *));
11187 
11188 	/*
11189 	 * If this was a retained enabling, decrement the dts_nretained count
11190 	 * and take it off of the dtrace_retained list.
11191 	 */
11192 	if (enab->dten_prev != NULL || enab->dten_next != NULL ||
11193 	    dtrace_retained == enab) {
11194 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
11195 		ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0);
11196 		enab->dten_vstate->dtvs_state->dts_nretained--;
11197 	}
11198 
11199 	if (enab->dten_prev == NULL) {
11200 		if (dtrace_retained == enab) {
11201 			dtrace_retained = enab->dten_next;
11202 
11203 			if (dtrace_retained != NULL)
11204 				dtrace_retained->dten_prev = NULL;
11205 		}
11206 	} else {
11207 		ASSERT(enab != dtrace_retained);
11208 		ASSERT(dtrace_retained != NULL);
11209 		enab->dten_prev->dten_next = enab->dten_next;
11210 	}
11211 
11212 	if (enab->dten_next != NULL) {
11213 		ASSERT(dtrace_retained != NULL);
11214 		enab->dten_next->dten_prev = enab->dten_prev;
11215 	}
11216 
11217 	kmem_free(enab, sizeof (dtrace_enabling_t));
11218 }
11219 
11220 static int
11221 dtrace_enabling_retain(dtrace_enabling_t *enab)
11222 {
11223 	dtrace_state_t *state;
11224 
11225 	ASSERT(MUTEX_HELD(&dtrace_lock));
11226 	ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11227 	ASSERT(enab->dten_vstate != NULL);
11228 
11229 	state = enab->dten_vstate->dtvs_state;
11230 	ASSERT(state != NULL);
11231 
11232 	/*
11233 	 * We only allow each state to retain dtrace_retain_max enablings.
11234 	 */
11235 	if (state->dts_nretained >= dtrace_retain_max)
11236 		return (ENOSPC);
11237 
11238 	state->dts_nretained++;
11239 
11240 	if (dtrace_retained == NULL) {
11241 		dtrace_retained = enab;
11242 		return (0);
11243 	}
11244 
11245 	enab->dten_next = dtrace_retained;
11246 	dtrace_retained->dten_prev = enab;
11247 	dtrace_retained = enab;
11248 
11249 	return (0);
11250 }
11251 
11252 static int
11253 dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match,
11254     dtrace_probedesc_t *create)
11255 {
11256 	dtrace_enabling_t *new, *enab;
11257 	int found = 0, err = ENOENT;
11258 
11259 	ASSERT(MUTEX_HELD(&dtrace_lock));
11260 	ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN);
11261 	ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN);
11262 	ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN);
11263 	ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN);
11264 
11265 	new = dtrace_enabling_create(&state->dts_vstate);
11266 
11267 	/*
11268 	 * Iterate over all retained enablings, looking for enablings that
11269 	 * match the specified state.
11270 	 */
11271 	for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11272 		int i;
11273 
11274 		/*
11275 		 * dtvs_state can only be NULL for helper enablings -- and
11276 		 * helper enablings can't be retained.
11277 		 */
11278 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
11279 
11280 		if (enab->dten_vstate->dtvs_state != state)
11281 			continue;
11282 
11283 		/*
11284 		 * Now iterate over each probe description; we're looking for
11285 		 * an exact match to the specified probe description.
11286 		 */
11287 		for (i = 0; i < enab->dten_ndesc; i++) {
11288 			dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11289 			dtrace_probedesc_t *pd = &ep->dted_probe;
11290 
11291 			if (strcmp(pd->dtpd_provider, match->dtpd_provider))
11292 				continue;
11293 
11294 			if (strcmp(pd->dtpd_mod, match->dtpd_mod))
11295 				continue;
11296 
11297 			if (strcmp(pd->dtpd_func, match->dtpd_func))
11298 				continue;
11299 
11300 			if (strcmp(pd->dtpd_name, match->dtpd_name))
11301 				continue;
11302 
11303 			/*
11304 			 * We have a winning probe!  Add it to our growing
11305 			 * enabling.
11306 			 */
11307 			found = 1;
11308 			dtrace_enabling_addlike(new, ep, create);
11309 		}
11310 	}
11311 
11312 	if (!found || (err = dtrace_enabling_retain(new)) != 0) {
11313 		dtrace_enabling_destroy(new);
11314 		return (err);
11315 	}
11316 
11317 	return (0);
11318 }
11319 
11320 static void
11321 dtrace_enabling_retract(dtrace_state_t *state)
11322 {
11323 	dtrace_enabling_t *enab, *next;
11324 
11325 	ASSERT(MUTEX_HELD(&dtrace_lock));
11326 
11327 	/*
11328 	 * Iterate over all retained enablings, destroy the enablings retained
11329 	 * for the specified state.
11330 	 */
11331 	for (enab = dtrace_retained; enab != NULL; enab = next) {
11332 		next = enab->dten_next;
11333 
11334 		/*
11335 		 * dtvs_state can only be NULL for helper enablings -- and
11336 		 * helper enablings can't be retained.
11337 		 */
11338 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
11339 
11340 		if (enab->dten_vstate->dtvs_state == state) {
11341 			ASSERT(state->dts_nretained > 0);
11342 			dtrace_enabling_destroy(enab);
11343 		}
11344 	}
11345 
11346 	ASSERT(state->dts_nretained == 0);
11347 }
11348 
11349 static int
11350 dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched)
11351 {
11352 	int i = 0;
11353 	int matched = 0;
11354 
11355 	ASSERT(MUTEX_HELD(&cpu_lock));
11356 	ASSERT(MUTEX_HELD(&dtrace_lock));
11357 
11358 	for (i = 0; i < enab->dten_ndesc; i++) {
11359 		dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11360 
11361 		enab->dten_current = ep;
11362 		enab->dten_error = 0;
11363 
11364 		matched += dtrace_probe_enable(&ep->dted_probe, enab);
11365 
11366 		if (enab->dten_error != 0) {
11367 			/*
11368 			 * If we get an error half-way through enabling the
11369 			 * probes, we kick out -- perhaps with some number of
11370 			 * them enabled.  Leaving enabled probes enabled may
11371 			 * be slightly confusing for user-level, but we expect
11372 			 * that no one will attempt to actually drive on in
11373 			 * the face of such errors.  If this is an anonymous
11374 			 * enabling (indicated with a NULL nmatched pointer),
11375 			 * we cmn_err() a message.  We aren't expecting to
11376 			 * get such an error -- such as it can exist at all,
11377 			 * it would be a result of corrupted DOF in the driver
11378 			 * properties.
11379 			 */
11380 			if (nmatched == NULL) {
11381 				cmn_err(CE_WARN, "dtrace_enabling_match() "
11382 				    "error on %p: %d", (void *)ep,
11383 				    enab->dten_error);
11384 			}
11385 
11386 			return (enab->dten_error);
11387 		}
11388 	}
11389 
11390 	enab->dten_probegen = dtrace_probegen;
11391 	if (nmatched != NULL)
11392 		*nmatched = matched;
11393 
11394 	return (0);
11395 }
11396 
11397 static void
11398 dtrace_enabling_matchall(void)
11399 {
11400 	dtrace_enabling_t *enab;
11401 
11402 	mutex_enter(&cpu_lock);
11403 	mutex_enter(&dtrace_lock);
11404 
11405 	/*
11406 	 * Iterate over all retained enablings to see if any probes match
11407 	 * against them.  We only perform this operation on enablings for which
11408 	 * we have sufficient permissions by virtue of being in the global zone
11409 	 * or in the same zone as the DTrace client.  Because we can be called
11410 	 * after dtrace_detach() has been called, we cannot assert that there
11411 	 * are retained enablings.  We can safely load from dtrace_retained,
11412 	 * however:  the taskq_destroy() at the end of dtrace_detach() will
11413 	 * block pending our completion.
11414 	 */
11415 	for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11416 #if defined(sun)
11417 		cred_t *cr = enab->dten_vstate->dtvs_state->dts_cred.dcr_cred;
11418 
11419 		if (INGLOBALZONE(curproc) || getzoneid() == crgetzoneid(cr))
11420 #endif
11421 			(void) dtrace_enabling_match(enab, NULL);
11422 	}
11423 
11424 	mutex_exit(&dtrace_lock);
11425 	mutex_exit(&cpu_lock);
11426 }
11427 
11428 /*
11429  * If an enabling is to be enabled without having matched probes (that is, if
11430  * dtrace_state_go() is to be called on the underlying dtrace_state_t), the
11431  * enabling must be _primed_ by creating an ECB for every ECB description.
11432  * This must be done to assure that we know the number of speculations, the
11433  * number of aggregations, the minimum buffer size needed, etc. before we
11434  * transition out of DTRACE_ACTIVITY_INACTIVE.  To do this without actually
11435  * enabling any probes, we create ECBs for every ECB decription, but with a
11436  * NULL probe -- which is exactly what this function does.
11437  */
11438 static void
11439 dtrace_enabling_prime(dtrace_state_t *state)
11440 {
11441 	dtrace_enabling_t *enab;
11442 	int i;
11443 
11444 	for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11445 		ASSERT(enab->dten_vstate->dtvs_state != NULL);
11446 
11447 		if (enab->dten_vstate->dtvs_state != state)
11448 			continue;
11449 
11450 		/*
11451 		 * We don't want to prime an enabling more than once, lest
11452 		 * we allow a malicious user to induce resource exhaustion.
11453 		 * (The ECBs that result from priming an enabling aren't
11454 		 * leaked -- but they also aren't deallocated until the
11455 		 * consumer state is destroyed.)
11456 		 */
11457 		if (enab->dten_primed)
11458 			continue;
11459 
11460 		for (i = 0; i < enab->dten_ndesc; i++) {
11461 			enab->dten_current = enab->dten_desc[i];
11462 			(void) dtrace_probe_enable(NULL, enab);
11463 		}
11464 
11465 		enab->dten_primed = 1;
11466 	}
11467 }
11468 
11469 /*
11470  * Called to indicate that probes should be provided due to retained
11471  * enablings.  This is implemented in terms of dtrace_probe_provide(), but it
11472  * must take an initial lap through the enabling calling the dtps_provide()
11473  * entry point explicitly to allow for autocreated probes.
11474  */
11475 static void
11476 dtrace_enabling_provide(dtrace_provider_t *prv)
11477 {
11478 	int i, all = 0;
11479 	dtrace_probedesc_t desc;
11480 
11481 	ASSERT(MUTEX_HELD(&dtrace_lock));
11482 	ASSERT(MUTEX_HELD(&dtrace_provider_lock));
11483 
11484 	if (prv == NULL) {
11485 		all = 1;
11486 		prv = dtrace_provider;
11487 	}
11488 
11489 	do {
11490 		dtrace_enabling_t *enab = dtrace_retained;
11491 		void *parg = prv->dtpv_arg;
11492 
11493 		for (; enab != NULL; enab = enab->dten_next) {
11494 			for (i = 0; i < enab->dten_ndesc; i++) {
11495 				desc = enab->dten_desc[i]->dted_probe;
11496 				mutex_exit(&dtrace_lock);
11497 				prv->dtpv_pops.dtps_provide(parg, &desc);
11498 				mutex_enter(&dtrace_lock);
11499 			}
11500 		}
11501 	} while (all && (prv = prv->dtpv_next) != NULL);
11502 
11503 	mutex_exit(&dtrace_lock);
11504 	dtrace_probe_provide(NULL, all ? NULL : prv);
11505 	mutex_enter(&dtrace_lock);
11506 }
11507 
11508 /*
11509  * DTrace DOF Functions
11510  */
11511 /*ARGSUSED*/
11512 static void
11513 dtrace_dof_error(dof_hdr_t *dof, const char *str)
11514 {
11515 	if (dtrace_err_verbose)
11516 		cmn_err(CE_WARN, "failed to process DOF: %s", str);
11517 
11518 #ifdef DTRACE_ERRDEBUG
11519 	dtrace_errdebug(str);
11520 #endif
11521 }
11522 
11523 /*
11524  * Create DOF out of a currently enabled state.  Right now, we only create
11525  * DOF containing the run-time options -- but this could be expanded to create
11526  * complete DOF representing the enabled state.
11527  */
11528 static dof_hdr_t *
11529 dtrace_dof_create(dtrace_state_t *state)
11530 {
11531 	dof_hdr_t *dof;
11532 	dof_sec_t *sec;
11533 	dof_optdesc_t *opt;
11534 	int i, len = sizeof (dof_hdr_t) +
11535 	    roundup(sizeof (dof_sec_t), sizeof (uint64_t)) +
11536 	    sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
11537 
11538 	ASSERT(MUTEX_HELD(&dtrace_lock));
11539 
11540 	dof = kmem_zalloc(len, KM_SLEEP);
11541 	dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0;
11542 	dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1;
11543 	dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2;
11544 	dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3;
11545 
11546 	dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE;
11547 	dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE;
11548 	dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION;
11549 	dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION;
11550 	dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS;
11551 	dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS;
11552 
11553 	dof->dofh_flags = 0;
11554 	dof->dofh_hdrsize = sizeof (dof_hdr_t);
11555 	dof->dofh_secsize = sizeof (dof_sec_t);
11556 	dof->dofh_secnum = 1;	/* only DOF_SECT_OPTDESC */
11557 	dof->dofh_secoff = sizeof (dof_hdr_t);
11558 	dof->dofh_loadsz = len;
11559 	dof->dofh_filesz = len;
11560 	dof->dofh_pad = 0;
11561 
11562 	/*
11563 	 * Fill in the option section header...
11564 	 */
11565 	sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t));
11566 	sec->dofs_type = DOF_SECT_OPTDESC;
11567 	sec->dofs_align = sizeof (uint64_t);
11568 	sec->dofs_flags = DOF_SECF_LOAD;
11569 	sec->dofs_entsize = sizeof (dof_optdesc_t);
11570 
11571 	opt = (dof_optdesc_t *)((uintptr_t)sec +
11572 	    roundup(sizeof (dof_sec_t), sizeof (uint64_t)));
11573 
11574 	sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof;
11575 	sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
11576 
11577 	for (i = 0; i < DTRACEOPT_MAX; i++) {
11578 		opt[i].dofo_option = i;
11579 		opt[i].dofo_strtab = DOF_SECIDX_NONE;
11580 		opt[i].dofo_value = state->dts_options[i];
11581 	}
11582 
11583 	return (dof);
11584 }
11585 
11586 static dof_hdr_t *
11587 dtrace_dof_copyin(uintptr_t uarg, int *errp)
11588 {
11589 	dof_hdr_t hdr, *dof;
11590 
11591 	ASSERT(!MUTEX_HELD(&dtrace_lock));
11592 
11593 	/*
11594 	 * First, we're going to copyin() the sizeof (dof_hdr_t).
11595 	 */
11596 	if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) {
11597 		dtrace_dof_error(NULL, "failed to copyin DOF header");
11598 		*errp = EFAULT;
11599 		return (NULL);
11600 	}
11601 
11602 	/*
11603 	 * Now we'll allocate the entire DOF and copy it in -- provided
11604 	 * that the length isn't outrageous.
11605 	 */
11606 	if (hdr.dofh_loadsz >= dtrace_dof_maxsize) {
11607 		dtrace_dof_error(&hdr, "load size exceeds maximum");
11608 		*errp = E2BIG;
11609 		return (NULL);
11610 	}
11611 
11612 	if (hdr.dofh_loadsz < sizeof (hdr)) {
11613 		dtrace_dof_error(&hdr, "invalid load size");
11614 		*errp = EINVAL;
11615 		return (NULL);
11616 	}
11617 
11618 	dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP);
11619 
11620 	if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0) {
11621 		kmem_free(dof, hdr.dofh_loadsz);
11622 		*errp = EFAULT;
11623 		return (NULL);
11624 	}
11625 
11626 	return (dof);
11627 }
11628 
11629 #if !defined(sun)
11630 static __inline uchar_t
11631 dtrace_dof_char(char c) {
11632 	switch (c) {
11633 	case '0':
11634 	case '1':
11635 	case '2':
11636 	case '3':
11637 	case '4':
11638 	case '5':
11639 	case '6':
11640 	case '7':
11641 	case '8':
11642 	case '9':
11643 		return (c - '0');
11644 	case 'A':
11645 	case 'B':
11646 	case 'C':
11647 	case 'D':
11648 	case 'E':
11649 	case 'F':
11650 		return (c - 'A' + 10);
11651 	case 'a':
11652 	case 'b':
11653 	case 'c':
11654 	case 'd':
11655 	case 'e':
11656 	case 'f':
11657 		return (c - 'a' + 10);
11658 	}
11659 	/* Should not reach here. */
11660 	return (0);
11661 }
11662 #endif
11663 
11664 static dof_hdr_t *
11665 dtrace_dof_property(const char *name)
11666 {
11667 	uchar_t *buf;
11668 	uint64_t loadsz;
11669 	unsigned int len, i;
11670 	dof_hdr_t *dof;
11671 
11672 #if defined(sun)
11673 	/*
11674 	 * Unfortunately, array of values in .conf files are always (and
11675 	 * only) interpreted to be integer arrays.  We must read our DOF
11676 	 * as an integer array, and then squeeze it into a byte array.
11677 	 */
11678 	if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0,
11679 	    (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS)
11680 		return (NULL);
11681 
11682 	for (i = 0; i < len; i++)
11683 		buf[i] = (uchar_t)(((int *)buf)[i]);
11684 
11685 	if (len < sizeof (dof_hdr_t)) {
11686 		ddi_prop_free(buf);
11687 		dtrace_dof_error(NULL, "truncated header");
11688 		return (NULL);
11689 	}
11690 
11691 	if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) {
11692 		ddi_prop_free(buf);
11693 		dtrace_dof_error(NULL, "truncated DOF");
11694 		return (NULL);
11695 	}
11696 
11697 	if (loadsz >= dtrace_dof_maxsize) {
11698 		ddi_prop_free(buf);
11699 		dtrace_dof_error(NULL, "oversized DOF");
11700 		return (NULL);
11701 	}
11702 
11703 	dof = kmem_alloc(loadsz, KM_SLEEP);
11704 	bcopy(buf, dof, loadsz);
11705 	ddi_prop_free(buf);
11706 #else
11707 	char *p;
11708 	char *p_env;
11709 
11710 	if ((p_env = getenv(name)) == NULL)
11711 		return (NULL);
11712 
11713 	len = strlen(p_env) / 2;
11714 
11715 	buf = kmem_alloc(len, KM_SLEEP);
11716 
11717 	dof = (dof_hdr_t *) buf;
11718 
11719 	p = p_env;
11720 
11721 	for (i = 0; i < len; i++) {
11722 		buf[i] = (dtrace_dof_char(p[0]) << 4) |
11723 		     dtrace_dof_char(p[1]);
11724 		p += 2;
11725 	}
11726 
11727 	freeenv(p_env);
11728 
11729 	if (len < sizeof (dof_hdr_t)) {
11730 		kmem_free(buf, 0);
11731 		dtrace_dof_error(NULL, "truncated header");
11732 		return (NULL);
11733 	}
11734 
11735 	if (len < (loadsz = dof->dofh_loadsz)) {
11736 		kmem_free(buf, 0);
11737 		dtrace_dof_error(NULL, "truncated DOF");
11738 		return (NULL);
11739 	}
11740 
11741 	if (loadsz >= dtrace_dof_maxsize) {
11742 		kmem_free(buf, 0);
11743 		dtrace_dof_error(NULL, "oversized DOF");
11744 		return (NULL);
11745 	}
11746 #endif
11747 
11748 	return (dof);
11749 }
11750 
11751 static void
11752 dtrace_dof_destroy(dof_hdr_t *dof)
11753 {
11754 	kmem_free(dof, dof->dofh_loadsz);
11755 }
11756 
11757 /*
11758  * Return the dof_sec_t pointer corresponding to a given section index.  If the
11759  * index is not valid, dtrace_dof_error() is called and NULL is returned.  If
11760  * a type other than DOF_SECT_NONE is specified, the header is checked against
11761  * this type and NULL is returned if the types do not match.
11762  */
11763 static dof_sec_t *
11764 dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i)
11765 {
11766 	dof_sec_t *sec = (dof_sec_t *)(uintptr_t)
11767 	    ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize);
11768 
11769 	if (i >= dof->dofh_secnum) {
11770 		dtrace_dof_error(dof, "referenced section index is invalid");
11771 		return (NULL);
11772 	}
11773 
11774 	if (!(sec->dofs_flags & DOF_SECF_LOAD)) {
11775 		dtrace_dof_error(dof, "referenced section is not loadable");
11776 		return (NULL);
11777 	}
11778 
11779 	if (type != DOF_SECT_NONE && type != sec->dofs_type) {
11780 		dtrace_dof_error(dof, "referenced section is the wrong type");
11781 		return (NULL);
11782 	}
11783 
11784 	return (sec);
11785 }
11786 
11787 static dtrace_probedesc_t *
11788 dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc)
11789 {
11790 	dof_probedesc_t *probe;
11791 	dof_sec_t *strtab;
11792 	uintptr_t daddr = (uintptr_t)dof;
11793 	uintptr_t str;
11794 	size_t size;
11795 
11796 	if (sec->dofs_type != DOF_SECT_PROBEDESC) {
11797 		dtrace_dof_error(dof, "invalid probe section");
11798 		return (NULL);
11799 	}
11800 
11801 	if (sec->dofs_align != sizeof (dof_secidx_t)) {
11802 		dtrace_dof_error(dof, "bad alignment in probe description");
11803 		return (NULL);
11804 	}
11805 
11806 	if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) {
11807 		dtrace_dof_error(dof, "truncated probe description");
11808 		return (NULL);
11809 	}
11810 
11811 	probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset);
11812 	strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab);
11813 
11814 	if (strtab == NULL)
11815 		return (NULL);
11816 
11817 	str = daddr + strtab->dofs_offset;
11818 	size = strtab->dofs_size;
11819 
11820 	if (probe->dofp_provider >= strtab->dofs_size) {
11821 		dtrace_dof_error(dof, "corrupt probe provider");
11822 		return (NULL);
11823 	}
11824 
11825 	(void) strncpy(desc->dtpd_provider,
11826 	    (char *)(str + probe->dofp_provider),
11827 	    MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider));
11828 
11829 	if (probe->dofp_mod >= strtab->dofs_size) {
11830 		dtrace_dof_error(dof, "corrupt probe module");
11831 		return (NULL);
11832 	}
11833 
11834 	(void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod),
11835 	    MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod));
11836 
11837 	if (probe->dofp_func >= strtab->dofs_size) {
11838 		dtrace_dof_error(dof, "corrupt probe function");
11839 		return (NULL);
11840 	}
11841 
11842 	(void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func),
11843 	    MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func));
11844 
11845 	if (probe->dofp_name >= strtab->dofs_size) {
11846 		dtrace_dof_error(dof, "corrupt probe name");
11847 		return (NULL);
11848 	}
11849 
11850 	(void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name),
11851 	    MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name));
11852 
11853 	return (desc);
11854 }
11855 
11856 static dtrace_difo_t *
11857 dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
11858     cred_t *cr)
11859 {
11860 	dtrace_difo_t *dp;
11861 	size_t ttl = 0;
11862 	dof_difohdr_t *dofd;
11863 	uintptr_t daddr = (uintptr_t)dof;
11864 	size_t max = dtrace_difo_maxsize;
11865 	int i, l, n;
11866 
11867 	static const struct {
11868 		int section;
11869 		int bufoffs;
11870 		int lenoffs;
11871 		int entsize;
11872 		int align;
11873 		const char *msg;
11874 	} difo[] = {
11875 		{ DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf),
11876 		offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t),
11877 		sizeof (dif_instr_t), "multiple DIF sections" },
11878 
11879 		{ DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab),
11880 		offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t),
11881 		sizeof (uint64_t), "multiple integer tables" },
11882 
11883 		{ DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab),
11884 		offsetof(dtrace_difo_t, dtdo_strlen), 0,
11885 		sizeof (char), "multiple string tables" },
11886 
11887 		{ DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab),
11888 		offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t),
11889 		sizeof (uint_t), "multiple variable tables" },
11890 
11891 		{ DOF_SECT_NONE, 0, 0, 0, 0, NULL }
11892 	};
11893 
11894 	if (sec->dofs_type != DOF_SECT_DIFOHDR) {
11895 		dtrace_dof_error(dof, "invalid DIFO header section");
11896 		return (NULL);
11897 	}
11898 
11899 	if (sec->dofs_align != sizeof (dof_secidx_t)) {
11900 		dtrace_dof_error(dof, "bad alignment in DIFO header");
11901 		return (NULL);
11902 	}
11903 
11904 	if (sec->dofs_size < sizeof (dof_difohdr_t) ||
11905 	    sec->dofs_size % sizeof (dof_secidx_t)) {
11906 		dtrace_dof_error(dof, "bad size in DIFO header");
11907 		return (NULL);
11908 	}
11909 
11910 	dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
11911 	n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1;
11912 
11913 	dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
11914 	dp->dtdo_rtype = dofd->dofd_rtype;
11915 
11916 	for (l = 0; l < n; l++) {
11917 		dof_sec_t *subsec;
11918 		void **bufp;
11919 		uint32_t *lenp;
11920 
11921 		if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE,
11922 		    dofd->dofd_links[l])) == NULL)
11923 			goto err; /* invalid section link */
11924 
11925 		if (ttl + subsec->dofs_size > max) {
11926 			dtrace_dof_error(dof, "exceeds maximum size");
11927 			goto err;
11928 		}
11929 
11930 		ttl += subsec->dofs_size;
11931 
11932 		for (i = 0; difo[i].section != DOF_SECT_NONE; i++) {
11933 			if (subsec->dofs_type != difo[i].section)
11934 				continue;
11935 
11936 			if (!(subsec->dofs_flags & DOF_SECF_LOAD)) {
11937 				dtrace_dof_error(dof, "section not loaded");
11938 				goto err;
11939 			}
11940 
11941 			if (subsec->dofs_align != difo[i].align) {
11942 				dtrace_dof_error(dof, "bad alignment");
11943 				goto err;
11944 			}
11945 
11946 			bufp = (void **)((uintptr_t)dp + difo[i].bufoffs);
11947 			lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs);
11948 
11949 			if (*bufp != NULL) {
11950 				dtrace_dof_error(dof, difo[i].msg);
11951 				goto err;
11952 			}
11953 
11954 			if (difo[i].entsize != subsec->dofs_entsize) {
11955 				dtrace_dof_error(dof, "entry size mismatch");
11956 				goto err;
11957 			}
11958 
11959 			if (subsec->dofs_entsize != 0 &&
11960 			    (subsec->dofs_size % subsec->dofs_entsize) != 0) {
11961 				dtrace_dof_error(dof, "corrupt entry size");
11962 				goto err;
11963 			}
11964 
11965 			*lenp = subsec->dofs_size;
11966 			*bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP);
11967 			bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset),
11968 			    *bufp, subsec->dofs_size);
11969 
11970 			if (subsec->dofs_entsize != 0)
11971 				*lenp /= subsec->dofs_entsize;
11972 
11973 			break;
11974 		}
11975 
11976 		/*
11977 		 * If we encounter a loadable DIFO sub-section that is not
11978 		 * known to us, assume this is a broken program and fail.
11979 		 */
11980 		if (difo[i].section == DOF_SECT_NONE &&
11981 		    (subsec->dofs_flags & DOF_SECF_LOAD)) {
11982 			dtrace_dof_error(dof, "unrecognized DIFO subsection");
11983 			goto err;
11984 		}
11985 	}
11986 
11987 	if (dp->dtdo_buf == NULL) {
11988 		/*
11989 		 * We can't have a DIF object without DIF text.
11990 		 */
11991 		dtrace_dof_error(dof, "missing DIF text");
11992 		goto err;
11993 	}
11994 
11995 	/*
11996 	 * Before we validate the DIF object, run through the variable table
11997 	 * looking for the strings -- if any of their size are under, we'll set
11998 	 * their size to be the system-wide default string size.  Note that
11999 	 * this should _not_ happen if the "strsize" option has been set --
12000 	 * in this case, the compiler should have set the size to reflect the
12001 	 * setting of the option.
12002 	 */
12003 	for (i = 0; i < dp->dtdo_varlen; i++) {
12004 		dtrace_difv_t *v = &dp->dtdo_vartab[i];
12005 		dtrace_diftype_t *t = &v->dtdv_type;
12006 
12007 		if (v->dtdv_id < DIF_VAR_OTHER_UBASE)
12008 			continue;
12009 
12010 		if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0)
12011 			t->dtdt_size = dtrace_strsize_default;
12012 	}
12013 
12014 	if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0)
12015 		goto err;
12016 
12017 	dtrace_difo_init(dp, vstate);
12018 	return (dp);
12019 
12020 err:
12021 	kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
12022 	kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
12023 	kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
12024 	kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
12025 
12026 	kmem_free(dp, sizeof (dtrace_difo_t));
12027 	return (NULL);
12028 }
12029 
12030 static dtrace_predicate_t *
12031 dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12032     cred_t *cr)
12033 {
12034 	dtrace_difo_t *dp;
12035 
12036 	if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL)
12037 		return (NULL);
12038 
12039 	return (dtrace_predicate_create(dp));
12040 }
12041 
12042 static dtrace_actdesc_t *
12043 dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12044     cred_t *cr)
12045 {
12046 	dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next;
12047 	dof_actdesc_t *desc;
12048 	dof_sec_t *difosec;
12049 	size_t offs;
12050 	uintptr_t daddr = (uintptr_t)dof;
12051 	uint64_t arg;
12052 	dtrace_actkind_t kind;
12053 
12054 	if (sec->dofs_type != DOF_SECT_ACTDESC) {
12055 		dtrace_dof_error(dof, "invalid action section");
12056 		return (NULL);
12057 	}
12058 
12059 	if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) {
12060 		dtrace_dof_error(dof, "truncated action description");
12061 		return (NULL);
12062 	}
12063 
12064 	if (sec->dofs_align != sizeof (uint64_t)) {
12065 		dtrace_dof_error(dof, "bad alignment in action description");
12066 		return (NULL);
12067 	}
12068 
12069 	if (sec->dofs_size < sec->dofs_entsize) {
12070 		dtrace_dof_error(dof, "section entry size exceeds total size");
12071 		return (NULL);
12072 	}
12073 
12074 	if (sec->dofs_entsize != sizeof (dof_actdesc_t)) {
12075 		dtrace_dof_error(dof, "bad entry size in action description");
12076 		return (NULL);
12077 	}
12078 
12079 	if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) {
12080 		dtrace_dof_error(dof, "actions exceed dtrace_actions_max");
12081 		return (NULL);
12082 	}
12083 
12084 	for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) {
12085 		desc = (dof_actdesc_t *)(daddr +
12086 		    (uintptr_t)sec->dofs_offset + offs);
12087 		kind = (dtrace_actkind_t)desc->dofa_kind;
12088 
12089 		if (DTRACEACT_ISPRINTFLIKE(kind) &&
12090 		    (kind != DTRACEACT_PRINTA ||
12091 		    desc->dofa_strtab != DOF_SECIDX_NONE)) {
12092 			dof_sec_t *strtab;
12093 			char *str, *fmt;
12094 			uint64_t i;
12095 
12096 			/*
12097 			 * printf()-like actions must have a format string.
12098 			 */
12099 			if ((strtab = dtrace_dof_sect(dof,
12100 			    DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL)
12101 				goto err;
12102 
12103 			str = (char *)((uintptr_t)dof +
12104 			    (uintptr_t)strtab->dofs_offset);
12105 
12106 			for (i = desc->dofa_arg; i < strtab->dofs_size; i++) {
12107 				if (str[i] == '\0')
12108 					break;
12109 			}
12110 
12111 			if (i >= strtab->dofs_size) {
12112 				dtrace_dof_error(dof, "bogus format string");
12113 				goto err;
12114 			}
12115 
12116 			if (i == desc->dofa_arg) {
12117 				dtrace_dof_error(dof, "empty format string");
12118 				goto err;
12119 			}
12120 
12121 			i -= desc->dofa_arg;
12122 			fmt = kmem_alloc(i + 1, KM_SLEEP);
12123 			bcopy(&str[desc->dofa_arg], fmt, i + 1);
12124 			arg = (uint64_t)(uintptr_t)fmt;
12125 		} else {
12126 			if (kind == DTRACEACT_PRINTA) {
12127 				ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE);
12128 				arg = 0;
12129 			} else {
12130 				arg = desc->dofa_arg;
12131 			}
12132 		}
12133 
12134 		act = dtrace_actdesc_create(kind, desc->dofa_ntuple,
12135 		    desc->dofa_uarg, arg);
12136 
12137 		if (last != NULL) {
12138 			last->dtad_next = act;
12139 		} else {
12140 			first = act;
12141 		}
12142 
12143 		last = act;
12144 
12145 		if (desc->dofa_difo == DOF_SECIDX_NONE)
12146 			continue;
12147 
12148 		if ((difosec = dtrace_dof_sect(dof,
12149 		    DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL)
12150 			goto err;
12151 
12152 		act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr);
12153 
12154 		if (act->dtad_difo == NULL)
12155 			goto err;
12156 	}
12157 
12158 	ASSERT(first != NULL);
12159 	return (first);
12160 
12161 err:
12162 	for (act = first; act != NULL; act = next) {
12163 		next = act->dtad_next;
12164 		dtrace_actdesc_release(act, vstate);
12165 	}
12166 
12167 	return (NULL);
12168 }
12169 
12170 static dtrace_ecbdesc_t *
12171 dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12172     cred_t *cr)
12173 {
12174 	dtrace_ecbdesc_t *ep;
12175 	dof_ecbdesc_t *ecb;
12176 	dtrace_probedesc_t *desc;
12177 	dtrace_predicate_t *pred = NULL;
12178 
12179 	if (sec->dofs_size < sizeof (dof_ecbdesc_t)) {
12180 		dtrace_dof_error(dof, "truncated ECB description");
12181 		return (NULL);
12182 	}
12183 
12184 	if (sec->dofs_align != sizeof (uint64_t)) {
12185 		dtrace_dof_error(dof, "bad alignment in ECB description");
12186 		return (NULL);
12187 	}
12188 
12189 	ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset);
12190 	sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes);
12191 
12192 	if (sec == NULL)
12193 		return (NULL);
12194 
12195 	ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
12196 	ep->dted_uarg = ecb->dofe_uarg;
12197 	desc = &ep->dted_probe;
12198 
12199 	if (dtrace_dof_probedesc(dof, sec, desc) == NULL)
12200 		goto err;
12201 
12202 	if (ecb->dofe_pred != DOF_SECIDX_NONE) {
12203 		if ((sec = dtrace_dof_sect(dof,
12204 		    DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL)
12205 			goto err;
12206 
12207 		if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL)
12208 			goto err;
12209 
12210 		ep->dted_pred.dtpdd_predicate = pred;
12211 	}
12212 
12213 	if (ecb->dofe_actions != DOF_SECIDX_NONE) {
12214 		if ((sec = dtrace_dof_sect(dof,
12215 		    DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL)
12216 			goto err;
12217 
12218 		ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr);
12219 
12220 		if (ep->dted_action == NULL)
12221 			goto err;
12222 	}
12223 
12224 	return (ep);
12225 
12226 err:
12227 	if (pred != NULL)
12228 		dtrace_predicate_release(pred, vstate);
12229 	kmem_free(ep, sizeof (dtrace_ecbdesc_t));
12230 	return (NULL);
12231 }
12232 
12233 /*
12234  * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the
12235  * specified DOF.  At present, this amounts to simply adding 'ubase' to the
12236  * site of any user SETX relocations to account for load object base address.
12237  * In the future, if we need other relocations, this function can be extended.
12238  */
12239 static int
12240 dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
12241 {
12242 	uintptr_t daddr = (uintptr_t)dof;
12243 	dof_relohdr_t *dofr =
12244 	    (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
12245 	dof_sec_t *ss, *rs, *ts;
12246 	dof_relodesc_t *r;
12247 	uint_t i, n;
12248 
12249 	if (sec->dofs_size < sizeof (dof_relohdr_t) ||
12250 	    sec->dofs_align != sizeof (dof_secidx_t)) {
12251 		dtrace_dof_error(dof, "invalid relocation header");
12252 		return (-1);
12253 	}
12254 
12255 	ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab);
12256 	rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec);
12257 	ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec);
12258 
12259 	if (ss == NULL || rs == NULL || ts == NULL)
12260 		return (-1); /* dtrace_dof_error() has been called already */
12261 
12262 	if (rs->dofs_entsize < sizeof (dof_relodesc_t) ||
12263 	    rs->dofs_align != sizeof (uint64_t)) {
12264 		dtrace_dof_error(dof, "invalid relocation section");
12265 		return (-1);
12266 	}
12267 
12268 	r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset);
12269 	n = rs->dofs_size / rs->dofs_entsize;
12270 
12271 	for (i = 0; i < n; i++) {
12272 		uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset;
12273 
12274 		switch (r->dofr_type) {
12275 		case DOF_RELO_NONE:
12276 			break;
12277 		case DOF_RELO_SETX:
12278 			if (r->dofr_offset >= ts->dofs_size || r->dofr_offset +
12279 			    sizeof (uint64_t) > ts->dofs_size) {
12280 				dtrace_dof_error(dof, "bad relocation offset");
12281 				return (-1);
12282 			}
12283 
12284 			if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) {
12285 				dtrace_dof_error(dof, "misaligned setx relo");
12286 				return (-1);
12287 			}
12288 
12289 			*(uint64_t *)taddr += ubase;
12290 			break;
12291 		default:
12292 			dtrace_dof_error(dof, "invalid relocation type");
12293 			return (-1);
12294 		}
12295 
12296 		r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize);
12297 	}
12298 
12299 	return (0);
12300 }
12301 
12302 /*
12303  * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated
12304  * header:  it should be at the front of a memory region that is at least
12305  * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in
12306  * size.  It need not be validated in any other way.
12307  */
12308 static int
12309 dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr,
12310     dtrace_enabling_t **enabp, uint64_t ubase, int noprobes)
12311 {
12312 	uint64_t len = dof->dofh_loadsz, seclen;
12313 	uintptr_t daddr = (uintptr_t)dof;
12314 	dtrace_ecbdesc_t *ep;
12315 	dtrace_enabling_t *enab;
12316 	uint_t i;
12317 
12318 	ASSERT(MUTEX_HELD(&dtrace_lock));
12319 	ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t));
12320 
12321 	/*
12322 	 * Check the DOF header identification bytes.  In addition to checking
12323 	 * valid settings, we also verify that unused bits/bytes are zeroed so
12324 	 * we can use them later without fear of regressing existing binaries.
12325 	 */
12326 	if (bcmp(&dof->dofh_ident[DOF_ID_MAG0],
12327 	    DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) {
12328 		dtrace_dof_error(dof, "DOF magic string mismatch");
12329 		return (-1);
12330 	}
12331 
12332 	if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 &&
12333 	    dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) {
12334 		dtrace_dof_error(dof, "DOF has invalid data model");
12335 		return (-1);
12336 	}
12337 
12338 	if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) {
12339 		dtrace_dof_error(dof, "DOF encoding mismatch");
12340 		return (-1);
12341 	}
12342 
12343 	if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
12344 	    dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) {
12345 		dtrace_dof_error(dof, "DOF version mismatch");
12346 		return (-1);
12347 	}
12348 
12349 	if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) {
12350 		dtrace_dof_error(dof, "DOF uses unsupported instruction set");
12351 		return (-1);
12352 	}
12353 
12354 	if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) {
12355 		dtrace_dof_error(dof, "DOF uses too many integer registers");
12356 		return (-1);
12357 	}
12358 
12359 	if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) {
12360 		dtrace_dof_error(dof, "DOF uses too many tuple registers");
12361 		return (-1);
12362 	}
12363 
12364 	for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) {
12365 		if (dof->dofh_ident[i] != 0) {
12366 			dtrace_dof_error(dof, "DOF has invalid ident byte set");
12367 			return (-1);
12368 		}
12369 	}
12370 
12371 	if (dof->dofh_flags & ~DOF_FL_VALID) {
12372 		dtrace_dof_error(dof, "DOF has invalid flag bits set");
12373 		return (-1);
12374 	}
12375 
12376 	if (dof->dofh_secsize == 0) {
12377 		dtrace_dof_error(dof, "zero section header size");
12378 		return (-1);
12379 	}
12380 
12381 	/*
12382 	 * Check that the section headers don't exceed the amount of DOF
12383 	 * data.  Note that we cast the section size and number of sections
12384 	 * to uint64_t's to prevent possible overflow in the multiplication.
12385 	 */
12386 	seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize;
12387 
12388 	if (dof->dofh_secoff > len || seclen > len ||
12389 	    dof->dofh_secoff + seclen > len) {
12390 		dtrace_dof_error(dof, "truncated section headers");
12391 		return (-1);
12392 	}
12393 
12394 	if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) {
12395 		dtrace_dof_error(dof, "misaligned section headers");
12396 		return (-1);
12397 	}
12398 
12399 	if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) {
12400 		dtrace_dof_error(dof, "misaligned section size");
12401 		return (-1);
12402 	}
12403 
12404 	/*
12405 	 * Take an initial pass through the section headers to be sure that
12406 	 * the headers don't have stray offsets.  If the 'noprobes' flag is
12407 	 * set, do not permit sections relating to providers, probes, or args.
12408 	 */
12409 	for (i = 0; i < dof->dofh_secnum; i++) {
12410 		dof_sec_t *sec = (dof_sec_t *)(daddr +
12411 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12412 
12413 		if (noprobes) {
12414 			switch (sec->dofs_type) {
12415 			case DOF_SECT_PROVIDER:
12416 			case DOF_SECT_PROBES:
12417 			case DOF_SECT_PRARGS:
12418 			case DOF_SECT_PROFFS:
12419 				dtrace_dof_error(dof, "illegal sections "
12420 				    "for enabling");
12421 				return (-1);
12422 			}
12423 		}
12424 
12425 		if (!(sec->dofs_flags & DOF_SECF_LOAD))
12426 			continue; /* just ignore non-loadable sections */
12427 
12428 		if (sec->dofs_align & (sec->dofs_align - 1)) {
12429 			dtrace_dof_error(dof, "bad section alignment");
12430 			return (-1);
12431 		}
12432 
12433 		if (sec->dofs_offset & (sec->dofs_align - 1)) {
12434 			dtrace_dof_error(dof, "misaligned section");
12435 			return (-1);
12436 		}
12437 
12438 		if (sec->dofs_offset > len || sec->dofs_size > len ||
12439 		    sec->dofs_offset + sec->dofs_size > len) {
12440 			dtrace_dof_error(dof, "corrupt section header");
12441 			return (-1);
12442 		}
12443 
12444 		if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr +
12445 		    sec->dofs_offset + sec->dofs_size - 1) != '\0') {
12446 			dtrace_dof_error(dof, "non-terminating string table");
12447 			return (-1);
12448 		}
12449 	}
12450 
12451 	/*
12452 	 * Take a second pass through the sections and locate and perform any
12453 	 * relocations that are present.  We do this after the first pass to
12454 	 * be sure that all sections have had their headers validated.
12455 	 */
12456 	for (i = 0; i < dof->dofh_secnum; i++) {
12457 		dof_sec_t *sec = (dof_sec_t *)(daddr +
12458 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12459 
12460 		if (!(sec->dofs_flags & DOF_SECF_LOAD))
12461 			continue; /* skip sections that are not loadable */
12462 
12463 		switch (sec->dofs_type) {
12464 		case DOF_SECT_URELHDR:
12465 			if (dtrace_dof_relocate(dof, sec, ubase) != 0)
12466 				return (-1);
12467 			break;
12468 		}
12469 	}
12470 
12471 	if ((enab = *enabp) == NULL)
12472 		enab = *enabp = dtrace_enabling_create(vstate);
12473 
12474 	for (i = 0; i < dof->dofh_secnum; i++) {
12475 		dof_sec_t *sec = (dof_sec_t *)(daddr +
12476 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12477 
12478 		if (sec->dofs_type != DOF_SECT_ECBDESC)
12479 			continue;
12480 
12481 		if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) {
12482 			dtrace_enabling_destroy(enab);
12483 			*enabp = NULL;
12484 			return (-1);
12485 		}
12486 
12487 		dtrace_enabling_add(enab, ep);
12488 	}
12489 
12490 	return (0);
12491 }
12492 
12493 /*
12494  * Process DOF for any options.  This routine assumes that the DOF has been
12495  * at least processed by dtrace_dof_slurp().
12496  */
12497 static int
12498 dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state)
12499 {
12500 	int i, rval;
12501 	uint32_t entsize;
12502 	size_t offs;
12503 	dof_optdesc_t *desc;
12504 
12505 	for (i = 0; i < dof->dofh_secnum; i++) {
12506 		dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof +
12507 		    (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12508 
12509 		if (sec->dofs_type != DOF_SECT_OPTDESC)
12510 			continue;
12511 
12512 		if (sec->dofs_align != sizeof (uint64_t)) {
12513 			dtrace_dof_error(dof, "bad alignment in "
12514 			    "option description");
12515 			return (EINVAL);
12516 		}
12517 
12518 		if ((entsize = sec->dofs_entsize) == 0) {
12519 			dtrace_dof_error(dof, "zeroed option entry size");
12520 			return (EINVAL);
12521 		}
12522 
12523 		if (entsize < sizeof (dof_optdesc_t)) {
12524 			dtrace_dof_error(dof, "bad option entry size");
12525 			return (EINVAL);
12526 		}
12527 
12528 		for (offs = 0; offs < sec->dofs_size; offs += entsize) {
12529 			desc = (dof_optdesc_t *)((uintptr_t)dof +
12530 			    (uintptr_t)sec->dofs_offset + offs);
12531 
12532 			if (desc->dofo_strtab != DOF_SECIDX_NONE) {
12533 				dtrace_dof_error(dof, "non-zero option string");
12534 				return (EINVAL);
12535 			}
12536 
12537 			if (desc->dofo_value == DTRACEOPT_UNSET) {
12538 				dtrace_dof_error(dof, "unset option");
12539 				return (EINVAL);
12540 			}
12541 
12542 			if ((rval = dtrace_state_option(state,
12543 			    desc->dofo_option, desc->dofo_value)) != 0) {
12544 				dtrace_dof_error(dof, "rejected option");
12545 				return (rval);
12546 			}
12547 		}
12548 	}
12549 
12550 	return (0);
12551 }
12552 
12553 /*
12554  * DTrace Consumer State Functions
12555  */
12556 static int
12557 dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size)
12558 {
12559 	size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize;
12560 	void *base;
12561 	uintptr_t limit;
12562 	dtrace_dynvar_t *dvar, *next, *start;
12563 	int i;
12564 
12565 	ASSERT(MUTEX_HELD(&dtrace_lock));
12566 	ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL);
12567 
12568 	bzero(dstate, sizeof (dtrace_dstate_t));
12569 
12570 	if ((dstate->dtds_chunksize = chunksize) == 0)
12571 		dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE;
12572 
12573 	if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)))
12574 		size = min;
12575 
12576 	if ((base = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
12577 		return (ENOMEM);
12578 
12579 	dstate->dtds_size = size;
12580 	dstate->dtds_base = base;
12581 	dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP);
12582 	bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t));
12583 
12584 	hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t));
12585 
12586 	if (hashsize != 1 && (hashsize & 1))
12587 		hashsize--;
12588 
12589 	dstate->dtds_hashsize = hashsize;
12590 	dstate->dtds_hash = dstate->dtds_base;
12591 
12592 	/*
12593 	 * Set all of our hash buckets to point to the single sink, and (if
12594 	 * it hasn't already been set), set the sink's hash value to be the
12595 	 * sink sentinel value.  The sink is needed for dynamic variable
12596 	 * lookups to know that they have iterated over an entire, valid hash
12597 	 * chain.
12598 	 */
12599 	for (i = 0; i < hashsize; i++)
12600 		dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink;
12601 
12602 	if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK)
12603 		dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK;
12604 
12605 	/*
12606 	 * Determine number of active CPUs.  Divide free list evenly among
12607 	 * active CPUs.
12608 	 */
12609 	start = (dtrace_dynvar_t *)
12610 	    ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t));
12611 	limit = (uintptr_t)base + size;
12612 
12613 	maxper = (limit - (uintptr_t)start) / NCPU;
12614 	maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize;
12615 
12616 #if !defined(sun)
12617 	CPU_FOREACH(i) {
12618 #else
12619 	for (i = 0; i < NCPU; i++) {
12620 #endif
12621 		dstate->dtds_percpu[i].dtdsc_free = dvar = start;
12622 
12623 		/*
12624 		 * If we don't even have enough chunks to make it once through
12625 		 * NCPUs, we're just going to allocate everything to the first
12626 		 * CPU.  And if we're on the last CPU, we're going to allocate
12627 		 * whatever is left over.  In either case, we set the limit to
12628 		 * be the limit of the dynamic variable space.
12629 		 */
12630 		if (maxper == 0 || i == NCPU - 1) {
12631 			limit = (uintptr_t)base + size;
12632 			start = NULL;
12633 		} else {
12634 			limit = (uintptr_t)start + maxper;
12635 			start = (dtrace_dynvar_t *)limit;
12636 		}
12637 
12638 		ASSERT(limit <= (uintptr_t)base + size);
12639 
12640 		for (;;) {
12641 			next = (dtrace_dynvar_t *)((uintptr_t)dvar +
12642 			    dstate->dtds_chunksize);
12643 
12644 			if ((uintptr_t)next + dstate->dtds_chunksize >= limit)
12645 				break;
12646 
12647 			dvar->dtdv_next = next;
12648 			dvar = next;
12649 		}
12650 
12651 		if (maxper == 0)
12652 			break;
12653 	}
12654 
12655 	return (0);
12656 }
12657 
12658 static void
12659 dtrace_dstate_fini(dtrace_dstate_t *dstate)
12660 {
12661 	ASSERT(MUTEX_HELD(&cpu_lock));
12662 
12663 	if (dstate->dtds_base == NULL)
12664 		return;
12665 
12666 	kmem_free(dstate->dtds_base, dstate->dtds_size);
12667 	kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu);
12668 }
12669 
12670 static void
12671 dtrace_vstate_fini(dtrace_vstate_t *vstate)
12672 {
12673 	/*
12674 	 * Logical XOR, where are you?
12675 	 */
12676 	ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL));
12677 
12678 	if (vstate->dtvs_nglobals > 0) {
12679 		kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals *
12680 		    sizeof (dtrace_statvar_t *));
12681 	}
12682 
12683 	if (vstate->dtvs_ntlocals > 0) {
12684 		kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals *
12685 		    sizeof (dtrace_difv_t));
12686 	}
12687 
12688 	ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL));
12689 
12690 	if (vstate->dtvs_nlocals > 0) {
12691 		kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals *
12692 		    sizeof (dtrace_statvar_t *));
12693 	}
12694 }
12695 
12696 #if defined(sun)
12697 static void
12698 dtrace_state_clean(dtrace_state_t *state)
12699 {
12700 	if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE)
12701 		return;
12702 
12703 	dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars);
12704 	dtrace_speculation_clean(state);
12705 }
12706 
12707 static void
12708 dtrace_state_deadman(dtrace_state_t *state)
12709 {
12710 	hrtime_t now;
12711 
12712 	dtrace_sync();
12713 
12714 	now = dtrace_gethrtime();
12715 
12716 	if (state != dtrace_anon.dta_state &&
12717 	    now - state->dts_laststatus >= dtrace_deadman_user)
12718 		return;
12719 
12720 	/*
12721 	 * We must be sure that dts_alive never appears to be less than the
12722 	 * value upon entry to dtrace_state_deadman(), and because we lack a
12723 	 * dtrace_cas64(), we cannot store to it atomically.  We thus instead
12724 	 * store INT64_MAX to it, followed by a memory barrier, followed by
12725 	 * the new value.  This assures that dts_alive never appears to be
12726 	 * less than its true value, regardless of the order in which the
12727 	 * stores to the underlying storage are issued.
12728 	 */
12729 	state->dts_alive = INT64_MAX;
12730 	dtrace_membar_producer();
12731 	state->dts_alive = now;
12732 }
12733 #else
12734 static void
12735 dtrace_state_clean(void *arg)
12736 {
12737 	dtrace_state_t *state = arg;
12738 	dtrace_optval_t *opt = state->dts_options;
12739 
12740 	if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE)
12741 		return;
12742 
12743 	dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars);
12744 	dtrace_speculation_clean(state);
12745 
12746 	callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC,
12747 	    dtrace_state_clean, state);
12748 }
12749 
12750 static void
12751 dtrace_state_deadman(void *arg)
12752 {
12753 	dtrace_state_t *state = arg;
12754 	hrtime_t now;
12755 
12756 	dtrace_sync();
12757 
12758 	dtrace_debug_output();
12759 
12760 	now = dtrace_gethrtime();
12761 
12762 	if (state != dtrace_anon.dta_state &&
12763 	    now - state->dts_laststatus >= dtrace_deadman_user)
12764 		return;
12765 
12766 	/*
12767 	 * We must be sure that dts_alive never appears to be less than the
12768 	 * value upon entry to dtrace_state_deadman(), and because we lack a
12769 	 * dtrace_cas64(), we cannot store to it atomically.  We thus instead
12770 	 * store INT64_MAX to it, followed by a memory barrier, followed by
12771 	 * the new value.  This assures that dts_alive never appears to be
12772 	 * less than its true value, regardless of the order in which the
12773 	 * stores to the underlying storage are issued.
12774 	 */
12775 	state->dts_alive = INT64_MAX;
12776 	dtrace_membar_producer();
12777 	state->dts_alive = now;
12778 
12779 	callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC,
12780 	    dtrace_state_deadman, state);
12781 }
12782 #endif
12783 
12784 static dtrace_state_t *
12785 #if defined(sun)
12786 dtrace_state_create(dev_t *devp, cred_t *cr)
12787 #else
12788 dtrace_state_create(struct cdev *dev)
12789 #endif
12790 {
12791 #if defined(sun)
12792 	minor_t minor;
12793 	major_t major;
12794 #else
12795 	cred_t *cr = NULL;
12796 	int m = 0;
12797 #endif
12798 	char c[30];
12799 	dtrace_state_t *state;
12800 	dtrace_optval_t *opt;
12801 	int bufsize = NCPU * sizeof (dtrace_buffer_t), i;
12802 
12803 	ASSERT(MUTEX_HELD(&dtrace_lock));
12804 	ASSERT(MUTEX_HELD(&cpu_lock));
12805 
12806 #if defined(sun)
12807 	minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1,
12808 	    VM_BESTFIT | VM_SLEEP);
12809 
12810 	if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) {
12811 		vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
12812 		return (NULL);
12813 	}
12814 
12815 	state = ddi_get_soft_state(dtrace_softstate, minor);
12816 #else
12817 	if (dev != NULL) {
12818 		cr = dev->si_cred;
12819 		m = dev2unit(dev);
12820 		}
12821 
12822 	/* Allocate memory for the state. */
12823 	state = kmem_zalloc(sizeof(dtrace_state_t), KM_SLEEP);
12824 #endif
12825 
12826 	state->dts_epid = DTRACE_EPIDNONE + 1;
12827 
12828 	(void) snprintf(c, sizeof (c), "dtrace_aggid_%d", m);
12829 #if defined(sun)
12830 	state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1,
12831 	    NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
12832 
12833 	if (devp != NULL) {
12834 		major = getemajor(*devp);
12835 	} else {
12836 		major = ddi_driver_major(dtrace_devi);
12837 	}
12838 
12839 	state->dts_dev = makedevice(major, minor);
12840 
12841 	if (devp != NULL)
12842 		*devp = state->dts_dev;
12843 #else
12844 	state->dts_aggid_arena = new_unrhdr(1, INT_MAX, &dtrace_unr_mtx);
12845 	state->dts_dev = dev;
12846 #endif
12847 
12848 	/*
12849 	 * We allocate NCPU buffers.  On the one hand, this can be quite
12850 	 * a bit of memory per instance (nearly 36K on a Starcat).  On the
12851 	 * other hand, it saves an additional memory reference in the probe
12852 	 * path.
12853 	 */
12854 	state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP);
12855 	state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP);
12856 
12857 #if defined(sun)
12858 	state->dts_cleaner = CYCLIC_NONE;
12859 	state->dts_deadman = CYCLIC_NONE;
12860 #else
12861 	callout_init(&state->dts_cleaner, CALLOUT_MPSAFE);
12862 	callout_init(&state->dts_deadman, CALLOUT_MPSAFE);
12863 #endif
12864 	state->dts_vstate.dtvs_state = state;
12865 
12866 	for (i = 0; i < DTRACEOPT_MAX; i++)
12867 		state->dts_options[i] = DTRACEOPT_UNSET;
12868 
12869 	/*
12870 	 * Set the default options.
12871 	 */
12872 	opt = state->dts_options;
12873 	opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH;
12874 	opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO;
12875 	opt[DTRACEOPT_NSPEC] = dtrace_nspec_default;
12876 	opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default;
12877 	opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL;
12878 	opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default;
12879 	opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default;
12880 	opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default;
12881 	opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default;
12882 	opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default;
12883 	opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default;
12884 	opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default;
12885 	opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default;
12886 	opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default;
12887 
12888 	state->dts_activity = DTRACE_ACTIVITY_INACTIVE;
12889 
12890 	/*
12891 	 * Depending on the user credentials, we set flag bits which alter probe
12892 	 * visibility or the amount of destructiveness allowed.  In the case of
12893 	 * actual anonymous tracing, or the possession of all privileges, all of
12894 	 * the normal checks are bypassed.
12895 	 */
12896 	if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
12897 		state->dts_cred.dcr_visible = DTRACE_CRV_ALL;
12898 		state->dts_cred.dcr_action = DTRACE_CRA_ALL;
12899 	} else {
12900 		/*
12901 		 * Set up the credentials for this instantiation.  We take a
12902 		 * hold on the credential to prevent it from disappearing on
12903 		 * us; this in turn prevents the zone_t referenced by this
12904 		 * credential from disappearing.  This means that we can
12905 		 * examine the credential and the zone from probe context.
12906 		 */
12907 		crhold(cr);
12908 		state->dts_cred.dcr_cred = cr;
12909 
12910 		/*
12911 		 * CRA_PROC means "we have *some* privilege for dtrace" and
12912 		 * unlocks the use of variables like pid, zonename, etc.
12913 		 */
12914 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) ||
12915 		    PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
12916 			state->dts_cred.dcr_action |= DTRACE_CRA_PROC;
12917 		}
12918 
12919 		/*
12920 		 * dtrace_user allows use of syscall and profile providers.
12921 		 * If the user also has proc_owner and/or proc_zone, we
12922 		 * extend the scope to include additional visibility and
12923 		 * destructive power.
12924 		 */
12925 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) {
12926 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) {
12927 				state->dts_cred.dcr_visible |=
12928 				    DTRACE_CRV_ALLPROC;
12929 
12930 				state->dts_cred.dcr_action |=
12931 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
12932 			}
12933 
12934 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) {
12935 				state->dts_cred.dcr_visible |=
12936 				    DTRACE_CRV_ALLZONE;
12937 
12938 				state->dts_cred.dcr_action |=
12939 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
12940 			}
12941 
12942 			/*
12943 			 * If we have all privs in whatever zone this is,
12944 			 * we can do destructive things to processes which
12945 			 * have altered credentials.
12946 			 */
12947 #if defined(sun)
12948 			if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
12949 			    cr->cr_zone->zone_privset)) {
12950 				state->dts_cred.dcr_action |=
12951 				    DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
12952 			}
12953 #endif
12954 		}
12955 
12956 		/*
12957 		 * Holding the dtrace_kernel privilege also implies that
12958 		 * the user has the dtrace_user privilege from a visibility
12959 		 * perspective.  But without further privileges, some
12960 		 * destructive actions are not available.
12961 		 */
12962 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) {
12963 			/*
12964 			 * Make all probes in all zones visible.  However,
12965 			 * this doesn't mean that all actions become available
12966 			 * to all zones.
12967 			 */
12968 			state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL |
12969 			    DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE;
12970 
12971 			state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL |
12972 			    DTRACE_CRA_PROC;
12973 			/*
12974 			 * Holding proc_owner means that destructive actions
12975 			 * for *this* zone are allowed.
12976 			 */
12977 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
12978 				state->dts_cred.dcr_action |=
12979 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
12980 
12981 			/*
12982 			 * Holding proc_zone means that destructive actions
12983 			 * for this user/group ID in all zones is allowed.
12984 			 */
12985 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
12986 				state->dts_cred.dcr_action |=
12987 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
12988 
12989 #if defined(sun)
12990 			/*
12991 			 * If we have all privs in whatever zone this is,
12992 			 * we can do destructive things to processes which
12993 			 * have altered credentials.
12994 			 */
12995 			if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
12996 			    cr->cr_zone->zone_privset)) {
12997 				state->dts_cred.dcr_action |=
12998 				    DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
12999 			}
13000 #endif
13001 		}
13002 
13003 		/*
13004 		 * Holding the dtrace_proc privilege gives control over fasttrap
13005 		 * and pid providers.  We need to grant wider destructive
13006 		 * privileges in the event that the user has proc_owner and/or
13007 		 * proc_zone.
13008 		 */
13009 		if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
13010 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
13011 				state->dts_cred.dcr_action |=
13012 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13013 
13014 			if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
13015 				state->dts_cred.dcr_action |=
13016 				    DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13017 		}
13018 	}
13019 
13020 	return (state);
13021 }
13022 
13023 static int
13024 dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which)
13025 {
13026 	dtrace_optval_t *opt = state->dts_options, size;
13027 	processorid_t cpu = 0;;
13028 	int flags = 0, rval;
13029 
13030 	ASSERT(MUTEX_HELD(&dtrace_lock));
13031 	ASSERT(MUTEX_HELD(&cpu_lock));
13032 	ASSERT(which < DTRACEOPT_MAX);
13033 	ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE ||
13034 	    (state == dtrace_anon.dta_state &&
13035 	    state->dts_activity == DTRACE_ACTIVITY_ACTIVE));
13036 
13037 	if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0)
13038 		return (0);
13039 
13040 	if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET)
13041 		cpu = opt[DTRACEOPT_CPU];
13042 
13043 	if (which == DTRACEOPT_SPECSIZE)
13044 		flags |= DTRACEBUF_NOSWITCH;
13045 
13046 	if (which == DTRACEOPT_BUFSIZE) {
13047 		if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING)
13048 			flags |= DTRACEBUF_RING;
13049 
13050 		if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL)
13051 			flags |= DTRACEBUF_FILL;
13052 
13053 		if (state != dtrace_anon.dta_state ||
13054 		    state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
13055 			flags |= DTRACEBUF_INACTIVE;
13056 	}
13057 
13058 	for (size = opt[which]; size >= sizeof (uint64_t); size >>= 1) {
13059 		/*
13060 		 * The size must be 8-byte aligned.  If the size is not 8-byte
13061 		 * aligned, drop it down by the difference.
13062 		 */
13063 		if (size & (sizeof (uint64_t) - 1))
13064 			size -= size & (sizeof (uint64_t) - 1);
13065 
13066 		if (size < state->dts_reserve) {
13067 			/*
13068 			 * Buffers always must be large enough to accommodate
13069 			 * their prereserved space.  We return E2BIG instead
13070 			 * of ENOMEM in this case to allow for user-level
13071 			 * software to differentiate the cases.
13072 			 */
13073 			return (E2BIG);
13074 		}
13075 
13076 		rval = dtrace_buffer_alloc(buf, size, flags, cpu);
13077 
13078 		if (rval != ENOMEM) {
13079 			opt[which] = size;
13080 			return (rval);
13081 		}
13082 
13083 		if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13084 			return (rval);
13085 	}
13086 
13087 	return (ENOMEM);
13088 }
13089 
13090 static int
13091 dtrace_state_buffers(dtrace_state_t *state)
13092 {
13093 	dtrace_speculation_t *spec = state->dts_speculations;
13094 	int rval, i;
13095 
13096 	if ((rval = dtrace_state_buffer(state, state->dts_buffer,
13097 	    DTRACEOPT_BUFSIZE)) != 0)
13098 		return (rval);
13099 
13100 	if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer,
13101 	    DTRACEOPT_AGGSIZE)) != 0)
13102 		return (rval);
13103 
13104 	for (i = 0; i < state->dts_nspeculations; i++) {
13105 		if ((rval = dtrace_state_buffer(state,
13106 		    spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0)
13107 			return (rval);
13108 	}
13109 
13110 	return (0);
13111 }
13112 
13113 static void
13114 dtrace_state_prereserve(dtrace_state_t *state)
13115 {
13116 	dtrace_ecb_t *ecb;
13117 	dtrace_probe_t *probe;
13118 
13119 	state->dts_reserve = 0;
13120 
13121 	if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL)
13122 		return;
13123 
13124 	/*
13125 	 * If our buffer policy is a "fill" buffer policy, we need to set the
13126 	 * prereserved space to be the space required by the END probes.
13127 	 */
13128 	probe = dtrace_probes[dtrace_probeid_end - 1];
13129 	ASSERT(probe != NULL);
13130 
13131 	for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
13132 		if (ecb->dte_state != state)
13133 			continue;
13134 
13135 		state->dts_reserve += ecb->dte_needed + ecb->dte_alignment;
13136 	}
13137 }
13138 
13139 static int
13140 dtrace_state_go(dtrace_state_t *state, processorid_t *cpu)
13141 {
13142 	dtrace_optval_t *opt = state->dts_options, sz, nspec;
13143 	dtrace_speculation_t *spec;
13144 	dtrace_buffer_t *buf;
13145 #if defined(sun)
13146 	cyc_handler_t hdlr;
13147 	cyc_time_t when;
13148 #endif
13149 	int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t);
13150 	dtrace_icookie_t cookie;
13151 
13152 	mutex_enter(&cpu_lock);
13153 	mutex_enter(&dtrace_lock);
13154 
13155 	if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
13156 		rval = EBUSY;
13157 		goto out;
13158 	}
13159 
13160 	/*
13161 	 * Before we can perform any checks, we must prime all of the
13162 	 * retained enablings that correspond to this state.
13163 	 */
13164 	dtrace_enabling_prime(state);
13165 
13166 	if (state->dts_destructive && !state->dts_cred.dcr_destructive) {
13167 		rval = EACCES;
13168 		goto out;
13169 	}
13170 
13171 	dtrace_state_prereserve(state);
13172 
13173 	/*
13174 	 * Now we want to do is try to allocate our speculations.
13175 	 * We do not automatically resize the number of speculations; if
13176 	 * this fails, we will fail the operation.
13177 	 */
13178 	nspec = opt[DTRACEOPT_NSPEC];
13179 	ASSERT(nspec != DTRACEOPT_UNSET);
13180 
13181 	if (nspec > INT_MAX) {
13182 		rval = ENOMEM;
13183 		goto out;
13184 	}
13185 
13186 	spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t), KM_NOSLEEP);
13187 
13188 	if (spec == NULL) {
13189 		rval = ENOMEM;
13190 		goto out;
13191 	}
13192 
13193 	state->dts_speculations = spec;
13194 	state->dts_nspeculations = (int)nspec;
13195 
13196 	for (i = 0; i < nspec; i++) {
13197 		if ((buf = kmem_zalloc(bufsize, KM_NOSLEEP)) == NULL) {
13198 			rval = ENOMEM;
13199 			goto err;
13200 		}
13201 
13202 		spec[i].dtsp_buffer = buf;
13203 	}
13204 
13205 	if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) {
13206 		if (dtrace_anon.dta_state == NULL) {
13207 			rval = ENOENT;
13208 			goto out;
13209 		}
13210 
13211 		if (state->dts_necbs != 0) {
13212 			rval = EALREADY;
13213 			goto out;
13214 		}
13215 
13216 		state->dts_anon = dtrace_anon_grab();
13217 		ASSERT(state->dts_anon != NULL);
13218 		state = state->dts_anon;
13219 
13220 		/*
13221 		 * We want "grabanon" to be set in the grabbed state, so we'll
13222 		 * copy that option value from the grabbing state into the
13223 		 * grabbed state.
13224 		 */
13225 		state->dts_options[DTRACEOPT_GRABANON] =
13226 		    opt[DTRACEOPT_GRABANON];
13227 
13228 		*cpu = dtrace_anon.dta_beganon;
13229 
13230 		/*
13231 		 * If the anonymous state is active (as it almost certainly
13232 		 * is if the anonymous enabling ultimately matched anything),
13233 		 * we don't allow any further option processing -- but we
13234 		 * don't return failure.
13235 		 */
13236 		if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
13237 			goto out;
13238 	}
13239 
13240 	if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET &&
13241 	    opt[DTRACEOPT_AGGSIZE] != 0) {
13242 		if (state->dts_aggregations == NULL) {
13243 			/*
13244 			 * We're not going to create an aggregation buffer
13245 			 * because we don't have any ECBs that contain
13246 			 * aggregations -- set this option to 0.
13247 			 */
13248 			opt[DTRACEOPT_AGGSIZE] = 0;
13249 		} else {
13250 			/*
13251 			 * If we have an aggregation buffer, we must also have
13252 			 * a buffer to use as scratch.
13253 			 */
13254 			if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET ||
13255 			    opt[DTRACEOPT_BUFSIZE] < state->dts_needed) {
13256 				opt[DTRACEOPT_BUFSIZE] = state->dts_needed;
13257 			}
13258 		}
13259 	}
13260 
13261 	if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET &&
13262 	    opt[DTRACEOPT_SPECSIZE] != 0) {
13263 		if (!state->dts_speculates) {
13264 			/*
13265 			 * We're not going to create speculation buffers
13266 			 * because we don't have any ECBs that actually
13267 			 * speculate -- set the speculation size to 0.
13268 			 */
13269 			opt[DTRACEOPT_SPECSIZE] = 0;
13270 		}
13271 	}
13272 
13273 	/*
13274 	 * The bare minimum size for any buffer that we're actually going to
13275 	 * do anything to is sizeof (uint64_t).
13276 	 */
13277 	sz = sizeof (uint64_t);
13278 
13279 	if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) ||
13280 	    (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) ||
13281 	    (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) {
13282 		/*
13283 		 * A buffer size has been explicitly set to 0 (or to a size
13284 		 * that will be adjusted to 0) and we need the space -- we
13285 		 * need to return failure.  We return ENOSPC to differentiate
13286 		 * it from failing to allocate a buffer due to failure to meet
13287 		 * the reserve (for which we return E2BIG).
13288 		 */
13289 		rval = ENOSPC;
13290 		goto out;
13291 	}
13292 
13293 	if ((rval = dtrace_state_buffers(state)) != 0)
13294 		goto err;
13295 
13296 	if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET)
13297 		sz = dtrace_dstate_defsize;
13298 
13299 	do {
13300 		rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz);
13301 
13302 		if (rval == 0)
13303 			break;
13304 
13305 		if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13306 			goto err;
13307 	} while (sz >>= 1);
13308 
13309 	opt[DTRACEOPT_DYNVARSIZE] = sz;
13310 
13311 	if (rval != 0)
13312 		goto err;
13313 
13314 	if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max)
13315 		opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max;
13316 
13317 	if (opt[DTRACEOPT_CLEANRATE] == 0)
13318 		opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13319 
13320 	if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min)
13321 		opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min;
13322 
13323 	if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max)
13324 		opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13325 
13326 	state->dts_alive = state->dts_laststatus = dtrace_gethrtime();
13327 #if defined(sun)
13328 	hdlr.cyh_func = (cyc_func_t)dtrace_state_clean;
13329 	hdlr.cyh_arg = state;
13330 	hdlr.cyh_level = CY_LOW_LEVEL;
13331 
13332 	when.cyt_when = 0;
13333 	when.cyt_interval = opt[DTRACEOPT_CLEANRATE];
13334 
13335 	state->dts_cleaner = cyclic_add(&hdlr, &when);
13336 
13337 	hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman;
13338 	hdlr.cyh_arg = state;
13339 	hdlr.cyh_level = CY_LOW_LEVEL;
13340 
13341 	when.cyt_when = 0;
13342 	when.cyt_interval = dtrace_deadman_interval;
13343 
13344 	state->dts_deadman = cyclic_add(&hdlr, &when);
13345 #else
13346 	callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC,
13347 	    dtrace_state_clean, state);
13348 	callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC,
13349 	    dtrace_state_deadman, state);
13350 #endif
13351 
13352 	state->dts_activity = DTRACE_ACTIVITY_WARMUP;
13353 
13354 	/*
13355 	 * Now it's time to actually fire the BEGIN probe.  We need to disable
13356 	 * interrupts here both to record the CPU on which we fired the BEGIN
13357 	 * probe (the data from this CPU will be processed first at user
13358 	 * level) and to manually activate the buffer for this CPU.
13359 	 */
13360 	cookie = dtrace_interrupt_disable();
13361 	*cpu = curcpu;
13362 	ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE);
13363 	state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE;
13364 
13365 	dtrace_probe(dtrace_probeid_begin,
13366 	    (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
13367 	dtrace_interrupt_enable(cookie);
13368 	/*
13369 	 * We may have had an exit action from a BEGIN probe; only change our
13370 	 * state to ACTIVE if we're still in WARMUP.
13371 	 */
13372 	ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP ||
13373 	    state->dts_activity == DTRACE_ACTIVITY_DRAINING);
13374 
13375 	if (state->dts_activity == DTRACE_ACTIVITY_WARMUP)
13376 		state->dts_activity = DTRACE_ACTIVITY_ACTIVE;
13377 
13378 	/*
13379 	 * Regardless of whether or not now we're in ACTIVE or DRAINING, we
13380 	 * want each CPU to transition its principal buffer out of the
13381 	 * INACTIVE state.  Doing this assures that no CPU will suddenly begin
13382 	 * processing an ECB halfway down a probe's ECB chain; all CPUs will
13383 	 * atomically transition from processing none of a state's ECBs to
13384 	 * processing all of them.
13385 	 */
13386 	dtrace_xcall(DTRACE_CPUALL,
13387 	    (dtrace_xcall_t)dtrace_buffer_activate, state);
13388 	goto out;
13389 
13390 err:
13391 	dtrace_buffer_free(state->dts_buffer);
13392 	dtrace_buffer_free(state->dts_aggbuffer);
13393 
13394 	if ((nspec = state->dts_nspeculations) == 0) {
13395 		ASSERT(state->dts_speculations == NULL);
13396 		goto out;
13397 	}
13398 
13399 	spec = state->dts_speculations;
13400 	ASSERT(spec != NULL);
13401 
13402 	for (i = 0; i < state->dts_nspeculations; i++) {
13403 		if ((buf = spec[i].dtsp_buffer) == NULL)
13404 			break;
13405 
13406 		dtrace_buffer_free(buf);
13407 		kmem_free(buf, bufsize);
13408 	}
13409 
13410 	kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
13411 	state->dts_nspeculations = 0;
13412 	state->dts_speculations = NULL;
13413 
13414 out:
13415 	mutex_exit(&dtrace_lock);
13416 	mutex_exit(&cpu_lock);
13417 
13418 	return (rval);
13419 }
13420 
13421 static int
13422 dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu)
13423 {
13424 	dtrace_icookie_t cookie;
13425 
13426 	ASSERT(MUTEX_HELD(&dtrace_lock));
13427 
13428 	if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE &&
13429 	    state->dts_activity != DTRACE_ACTIVITY_DRAINING)
13430 		return (EINVAL);
13431 
13432 	/*
13433 	 * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync
13434 	 * to be sure that every CPU has seen it.  See below for the details
13435 	 * on why this is done.
13436 	 */
13437 	state->dts_activity = DTRACE_ACTIVITY_DRAINING;
13438 	dtrace_sync();
13439 
13440 	/*
13441 	 * By this point, it is impossible for any CPU to be still processing
13442 	 * with DTRACE_ACTIVITY_ACTIVE.  We can thus set our activity to
13443 	 * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any
13444 	 * other CPU in dtrace_buffer_reserve().  This allows dtrace_probe()
13445 	 * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN
13446 	 * iff we're in the END probe.
13447 	 */
13448 	state->dts_activity = DTRACE_ACTIVITY_COOLDOWN;
13449 	dtrace_sync();
13450 	ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN);
13451 
13452 	/*
13453 	 * Finally, we can release the reserve and call the END probe.  We
13454 	 * disable interrupts across calling the END probe to allow us to
13455 	 * return the CPU on which we actually called the END probe.  This
13456 	 * allows user-land to be sure that this CPU's principal buffer is
13457 	 * processed last.
13458 	 */
13459 	state->dts_reserve = 0;
13460 
13461 	cookie = dtrace_interrupt_disable();
13462 	*cpu = curcpu;
13463 	dtrace_probe(dtrace_probeid_end,
13464 	    (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
13465 	dtrace_interrupt_enable(cookie);
13466 
13467 	state->dts_activity = DTRACE_ACTIVITY_STOPPED;
13468 	dtrace_sync();
13469 
13470 	return (0);
13471 }
13472 
13473 static int
13474 dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option,
13475     dtrace_optval_t val)
13476 {
13477 	ASSERT(MUTEX_HELD(&dtrace_lock));
13478 
13479 	if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
13480 		return (EBUSY);
13481 
13482 	if (option >= DTRACEOPT_MAX)
13483 		return (EINVAL);
13484 
13485 	if (option != DTRACEOPT_CPU && val < 0)
13486 		return (EINVAL);
13487 
13488 	switch (option) {
13489 	case DTRACEOPT_DESTRUCTIVE:
13490 		if (dtrace_destructive_disallow)
13491 			return (EACCES);
13492 
13493 		state->dts_cred.dcr_destructive = 1;
13494 		break;
13495 
13496 	case DTRACEOPT_BUFSIZE:
13497 	case DTRACEOPT_DYNVARSIZE:
13498 	case DTRACEOPT_AGGSIZE:
13499 	case DTRACEOPT_SPECSIZE:
13500 	case DTRACEOPT_STRSIZE:
13501 		if (val < 0)
13502 			return (EINVAL);
13503 
13504 		if (val >= LONG_MAX) {
13505 			/*
13506 			 * If this is an otherwise negative value, set it to
13507 			 * the highest multiple of 128m less than LONG_MAX.
13508 			 * Technically, we're adjusting the size without
13509 			 * regard to the buffer resizing policy, but in fact,
13510 			 * this has no effect -- if we set the buffer size to
13511 			 * ~LONG_MAX and the buffer policy is ultimately set to
13512 			 * be "manual", the buffer allocation is guaranteed to
13513 			 * fail, if only because the allocation requires two
13514 			 * buffers.  (We set the the size to the highest
13515 			 * multiple of 128m because it ensures that the size
13516 			 * will remain a multiple of a megabyte when
13517 			 * repeatedly halved -- all the way down to 15m.)
13518 			 */
13519 			val = LONG_MAX - (1 << 27) + 1;
13520 		}
13521 	}
13522 
13523 	state->dts_options[option] = val;
13524 
13525 	return (0);
13526 }
13527 
13528 static void
13529 dtrace_state_destroy(dtrace_state_t *state)
13530 {
13531 	dtrace_ecb_t *ecb;
13532 	dtrace_vstate_t *vstate = &state->dts_vstate;
13533 #if defined(sun)
13534 	minor_t minor = getminor(state->dts_dev);
13535 #endif
13536 	int i, bufsize = NCPU * sizeof (dtrace_buffer_t);
13537 	dtrace_speculation_t *spec = state->dts_speculations;
13538 	int nspec = state->dts_nspeculations;
13539 	uint32_t match;
13540 
13541 	ASSERT(MUTEX_HELD(&dtrace_lock));
13542 	ASSERT(MUTEX_HELD(&cpu_lock));
13543 
13544 	/*
13545 	 * First, retract any retained enablings for this state.
13546 	 */
13547 	dtrace_enabling_retract(state);
13548 	ASSERT(state->dts_nretained == 0);
13549 
13550 	if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE ||
13551 	    state->dts_activity == DTRACE_ACTIVITY_DRAINING) {
13552 		/*
13553 		 * We have managed to come into dtrace_state_destroy() on a
13554 		 * hot enabling -- almost certainly because of a disorderly
13555 		 * shutdown of a consumer.  (That is, a consumer that is
13556 		 * exiting without having called dtrace_stop().) In this case,
13557 		 * we're going to set our activity to be KILLED, and then
13558 		 * issue a sync to be sure that everyone is out of probe
13559 		 * context before we start blowing away ECBs.
13560 		 */
13561 		state->dts_activity = DTRACE_ACTIVITY_KILLED;
13562 		dtrace_sync();
13563 	}
13564 
13565 	/*
13566 	 * Release the credential hold we took in dtrace_state_create().
13567 	 */
13568 	if (state->dts_cred.dcr_cred != NULL)
13569 		crfree(state->dts_cred.dcr_cred);
13570 
13571 	/*
13572 	 * Now we can safely disable and destroy any enabled probes.  Because
13573 	 * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress
13574 	 * (especially if they're all enabled), we take two passes through the
13575 	 * ECBs:  in the first, we disable just DTRACE_PRIV_KERNEL probes, and
13576 	 * in the second we disable whatever is left over.
13577 	 */
13578 	for (match = DTRACE_PRIV_KERNEL; ; match = 0) {
13579 		for (i = 0; i < state->dts_necbs; i++) {
13580 			if ((ecb = state->dts_ecbs[i]) == NULL)
13581 				continue;
13582 
13583 			if (match && ecb->dte_probe != NULL) {
13584 				dtrace_probe_t *probe = ecb->dte_probe;
13585 				dtrace_provider_t *prov = probe->dtpr_provider;
13586 
13587 				if (!(prov->dtpv_priv.dtpp_flags & match))
13588 					continue;
13589 			}
13590 
13591 			dtrace_ecb_disable(ecb);
13592 			dtrace_ecb_destroy(ecb);
13593 		}
13594 
13595 		if (!match)
13596 			break;
13597 	}
13598 
13599 	/*
13600 	 * Before we free the buffers, perform one more sync to assure that
13601 	 * every CPU is out of probe context.
13602 	 */
13603 	dtrace_sync();
13604 
13605 	dtrace_buffer_free(state->dts_buffer);
13606 	dtrace_buffer_free(state->dts_aggbuffer);
13607 
13608 	for (i = 0; i < nspec; i++)
13609 		dtrace_buffer_free(spec[i].dtsp_buffer);
13610 
13611 #if defined(sun)
13612 	if (state->dts_cleaner != CYCLIC_NONE)
13613 		cyclic_remove(state->dts_cleaner);
13614 
13615 	if (state->dts_deadman != CYCLIC_NONE)
13616 		cyclic_remove(state->dts_deadman);
13617 #else
13618 	callout_stop(&state->dts_cleaner);
13619 	callout_drain(&state->dts_cleaner);
13620 	callout_stop(&state->dts_deadman);
13621 	callout_drain(&state->dts_deadman);
13622 #endif
13623 
13624 	dtrace_dstate_fini(&vstate->dtvs_dynvars);
13625 	dtrace_vstate_fini(vstate);
13626 	if (state->dts_ecbs != NULL)
13627 		kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *));
13628 
13629 	if (state->dts_aggregations != NULL) {
13630 #ifdef DEBUG
13631 		for (i = 0; i < state->dts_naggregations; i++)
13632 			ASSERT(state->dts_aggregations[i] == NULL);
13633 #endif
13634 		ASSERT(state->dts_naggregations > 0);
13635 		kmem_free(state->dts_aggregations,
13636 		    state->dts_naggregations * sizeof (dtrace_aggregation_t *));
13637 	}
13638 
13639 	kmem_free(state->dts_buffer, bufsize);
13640 	kmem_free(state->dts_aggbuffer, bufsize);
13641 
13642 	for (i = 0; i < nspec; i++)
13643 		kmem_free(spec[i].dtsp_buffer, bufsize);
13644 
13645 	if (spec != NULL)
13646 		kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
13647 
13648 	dtrace_format_destroy(state);
13649 
13650 	if (state->dts_aggid_arena != NULL) {
13651 #if defined(sun)
13652 		vmem_destroy(state->dts_aggid_arena);
13653 #else
13654 		delete_unrhdr(state->dts_aggid_arena);
13655 #endif
13656 		state->dts_aggid_arena = NULL;
13657 	}
13658 #if defined(sun)
13659 	ddi_soft_state_free(dtrace_softstate, minor);
13660 	vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
13661 #endif
13662 }
13663 
13664 /*
13665  * DTrace Anonymous Enabling Functions
13666  */
13667 static dtrace_state_t *
13668 dtrace_anon_grab(void)
13669 {
13670 	dtrace_state_t *state;
13671 
13672 	ASSERT(MUTEX_HELD(&dtrace_lock));
13673 
13674 	if ((state = dtrace_anon.dta_state) == NULL) {
13675 		ASSERT(dtrace_anon.dta_enabling == NULL);
13676 		return (NULL);
13677 	}
13678 
13679 	ASSERT(dtrace_anon.dta_enabling != NULL);
13680 	ASSERT(dtrace_retained != NULL);
13681 
13682 	dtrace_enabling_destroy(dtrace_anon.dta_enabling);
13683 	dtrace_anon.dta_enabling = NULL;
13684 	dtrace_anon.dta_state = NULL;
13685 
13686 	return (state);
13687 }
13688 
13689 static void
13690 dtrace_anon_property(void)
13691 {
13692 	int i, rv;
13693 	dtrace_state_t *state;
13694 	dof_hdr_t *dof;
13695 	char c[32];		/* enough for "dof-data-" + digits */
13696 
13697 	ASSERT(MUTEX_HELD(&dtrace_lock));
13698 	ASSERT(MUTEX_HELD(&cpu_lock));
13699 
13700 	for (i = 0; ; i++) {
13701 		(void) snprintf(c, sizeof (c), "dof-data-%d", i);
13702 
13703 		dtrace_err_verbose = 1;
13704 
13705 		if ((dof = dtrace_dof_property(c)) == NULL) {
13706 			dtrace_err_verbose = 0;
13707 			break;
13708 		}
13709 
13710 #if defined(sun)
13711 		/*
13712 		 * We want to create anonymous state, so we need to transition
13713 		 * the kernel debugger to indicate that DTrace is active.  If
13714 		 * this fails (e.g. because the debugger has modified text in
13715 		 * some way), we won't continue with the processing.
13716 		 */
13717 		if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
13718 			cmn_err(CE_NOTE, "kernel debugger active; anonymous "
13719 			    "enabling ignored.");
13720 			dtrace_dof_destroy(dof);
13721 			break;
13722 		}
13723 #endif
13724 
13725 		/*
13726 		 * If we haven't allocated an anonymous state, we'll do so now.
13727 		 */
13728 		if ((state = dtrace_anon.dta_state) == NULL) {
13729 #if defined(sun)
13730 			state = dtrace_state_create(NULL, NULL);
13731 #else
13732 			state = dtrace_state_create(NULL);
13733 #endif
13734 			dtrace_anon.dta_state = state;
13735 
13736 			if (state == NULL) {
13737 				/*
13738 				 * This basically shouldn't happen:  the only
13739 				 * failure mode from dtrace_state_create() is a
13740 				 * failure of ddi_soft_state_zalloc() that
13741 				 * itself should never happen.  Still, the
13742 				 * interface allows for a failure mode, and
13743 				 * we want to fail as gracefully as possible:
13744 				 * we'll emit an error message and cease
13745 				 * processing anonymous state in this case.
13746 				 */
13747 				cmn_err(CE_WARN, "failed to create "
13748 				    "anonymous state");
13749 				dtrace_dof_destroy(dof);
13750 				break;
13751 			}
13752 		}
13753 
13754 		rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(),
13755 		    &dtrace_anon.dta_enabling, 0, B_TRUE);
13756 
13757 		if (rv == 0)
13758 			rv = dtrace_dof_options(dof, state);
13759 
13760 		dtrace_err_verbose = 0;
13761 		dtrace_dof_destroy(dof);
13762 
13763 		if (rv != 0) {
13764 			/*
13765 			 * This is malformed DOF; chuck any anonymous state
13766 			 * that we created.
13767 			 */
13768 			ASSERT(dtrace_anon.dta_enabling == NULL);
13769 			dtrace_state_destroy(state);
13770 			dtrace_anon.dta_state = NULL;
13771 			break;
13772 		}
13773 
13774 		ASSERT(dtrace_anon.dta_enabling != NULL);
13775 	}
13776 
13777 	if (dtrace_anon.dta_enabling != NULL) {
13778 		int rval;
13779 
13780 		/*
13781 		 * dtrace_enabling_retain() can only fail because we are
13782 		 * trying to retain more enablings than are allowed -- but
13783 		 * we only have one anonymous enabling, and we are guaranteed
13784 		 * to be allowed at least one retained enabling; we assert
13785 		 * that dtrace_enabling_retain() returns success.
13786 		 */
13787 		rval = dtrace_enabling_retain(dtrace_anon.dta_enabling);
13788 		ASSERT(rval == 0);
13789 
13790 		dtrace_enabling_dump(dtrace_anon.dta_enabling);
13791 	}
13792 }
13793 
13794 /*
13795  * DTrace Helper Functions
13796  */
13797 static void
13798 dtrace_helper_trace(dtrace_helper_action_t *helper,
13799     dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where)
13800 {
13801 	uint32_t size, next, nnext, i;
13802 	dtrace_helptrace_t *ent;
13803 	uint16_t flags = cpu_core[curcpu].cpuc_dtrace_flags;
13804 
13805 	if (!dtrace_helptrace_enabled)
13806 		return;
13807 
13808 	ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals);
13809 
13810 	/*
13811 	 * What would a tracing framework be without its own tracing
13812 	 * framework?  (Well, a hell of a lot simpler, for starters...)
13813 	 */
13814 	size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals *
13815 	    sizeof (uint64_t) - sizeof (uint64_t);
13816 
13817 	/*
13818 	 * Iterate until we can allocate a slot in the trace buffer.
13819 	 */
13820 	do {
13821 		next = dtrace_helptrace_next;
13822 
13823 		if (next + size < dtrace_helptrace_bufsize) {
13824 			nnext = next + size;
13825 		} else {
13826 			nnext = size;
13827 		}
13828 	} while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next);
13829 
13830 	/*
13831 	 * We have our slot; fill it in.
13832 	 */
13833 	if (nnext == size)
13834 		next = 0;
13835 
13836 	ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next];
13837 	ent->dtht_helper = helper;
13838 	ent->dtht_where = where;
13839 	ent->dtht_nlocals = vstate->dtvs_nlocals;
13840 
13841 	ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ?
13842 	    mstate->dtms_fltoffs : -1;
13843 	ent->dtht_fault = DTRACE_FLAGS2FLT(flags);
13844 	ent->dtht_illval = cpu_core[curcpu].cpuc_dtrace_illval;
13845 
13846 	for (i = 0; i < vstate->dtvs_nlocals; i++) {
13847 		dtrace_statvar_t *svar;
13848 
13849 		if ((svar = vstate->dtvs_locals[i]) == NULL)
13850 			continue;
13851 
13852 		ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t));
13853 		ent->dtht_locals[i] =
13854 		    ((uint64_t *)(uintptr_t)svar->dtsv_data)[curcpu];
13855 	}
13856 }
13857 
13858 static uint64_t
13859 dtrace_helper(int which, dtrace_mstate_t *mstate,
13860     dtrace_state_t *state, uint64_t arg0, uint64_t arg1)
13861 {
13862 	uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
13863 	uint64_t sarg0 = mstate->dtms_arg[0];
13864 	uint64_t sarg1 = mstate->dtms_arg[1];
13865 	uint64_t rval = 0;
13866 	dtrace_helpers_t *helpers = curproc->p_dtrace_helpers;
13867 	dtrace_helper_action_t *helper;
13868 	dtrace_vstate_t *vstate;
13869 	dtrace_difo_t *pred;
13870 	int i, trace = dtrace_helptrace_enabled;
13871 
13872 	ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS);
13873 
13874 	if (helpers == NULL)
13875 		return (0);
13876 
13877 	if ((helper = helpers->dthps_actions[which]) == NULL)
13878 		return (0);
13879 
13880 	vstate = &helpers->dthps_vstate;
13881 	mstate->dtms_arg[0] = arg0;
13882 	mstate->dtms_arg[1] = arg1;
13883 
13884 	/*
13885 	 * Now iterate over each helper.  If its predicate evaluates to 'true',
13886 	 * we'll call the corresponding actions.  Note that the below calls
13887 	 * to dtrace_dif_emulate() may set faults in machine state.  This is
13888 	 * okay:  our caller (the outer dtrace_dif_emulate()) will simply plow
13889 	 * the stored DIF offset with its own (which is the desired behavior).
13890 	 * Also, note the calls to dtrace_dif_emulate() may allocate scratch
13891 	 * from machine state; this is okay, too.
13892 	 */
13893 	for (; helper != NULL; helper = helper->dtha_next) {
13894 		if ((pred = helper->dtha_predicate) != NULL) {
13895 			if (trace)
13896 				dtrace_helper_trace(helper, mstate, vstate, 0);
13897 
13898 			if (!dtrace_dif_emulate(pred, mstate, vstate, state))
13899 				goto next;
13900 
13901 			if (*flags & CPU_DTRACE_FAULT)
13902 				goto err;
13903 		}
13904 
13905 		for (i = 0; i < helper->dtha_nactions; i++) {
13906 			if (trace)
13907 				dtrace_helper_trace(helper,
13908 				    mstate, vstate, i + 1);
13909 
13910 			rval = dtrace_dif_emulate(helper->dtha_actions[i],
13911 			    mstate, vstate, state);
13912 
13913 			if (*flags & CPU_DTRACE_FAULT)
13914 				goto err;
13915 		}
13916 
13917 next:
13918 		if (trace)
13919 			dtrace_helper_trace(helper, mstate, vstate,
13920 			    DTRACE_HELPTRACE_NEXT);
13921 	}
13922 
13923 	if (trace)
13924 		dtrace_helper_trace(helper, mstate, vstate,
13925 		    DTRACE_HELPTRACE_DONE);
13926 
13927 	/*
13928 	 * Restore the arg0 that we saved upon entry.
13929 	 */
13930 	mstate->dtms_arg[0] = sarg0;
13931 	mstate->dtms_arg[1] = sarg1;
13932 
13933 	return (rval);
13934 
13935 err:
13936 	if (trace)
13937 		dtrace_helper_trace(helper, mstate, vstate,
13938 		    DTRACE_HELPTRACE_ERR);
13939 
13940 	/*
13941 	 * Restore the arg0 that we saved upon entry.
13942 	 */
13943 	mstate->dtms_arg[0] = sarg0;
13944 	mstate->dtms_arg[1] = sarg1;
13945 
13946 	return (0);
13947 }
13948 
13949 static void
13950 dtrace_helper_action_destroy(dtrace_helper_action_t *helper,
13951     dtrace_vstate_t *vstate)
13952 {
13953 	int i;
13954 
13955 	if (helper->dtha_predicate != NULL)
13956 		dtrace_difo_release(helper->dtha_predicate, vstate);
13957 
13958 	for (i = 0; i < helper->dtha_nactions; i++) {
13959 		ASSERT(helper->dtha_actions[i] != NULL);
13960 		dtrace_difo_release(helper->dtha_actions[i], vstate);
13961 	}
13962 
13963 	kmem_free(helper->dtha_actions,
13964 	    helper->dtha_nactions * sizeof (dtrace_difo_t *));
13965 	kmem_free(helper, sizeof (dtrace_helper_action_t));
13966 }
13967 
13968 static int
13969 dtrace_helper_destroygen(int gen)
13970 {
13971 	proc_t *p = curproc;
13972 	dtrace_helpers_t *help = p->p_dtrace_helpers;
13973 	dtrace_vstate_t *vstate;
13974 	int i;
13975 
13976 	ASSERT(MUTEX_HELD(&dtrace_lock));
13977 
13978 	if (help == NULL || gen > help->dthps_generation)
13979 		return (EINVAL);
13980 
13981 	vstate = &help->dthps_vstate;
13982 
13983 	for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
13984 		dtrace_helper_action_t *last = NULL, *h, *next;
13985 
13986 		for (h = help->dthps_actions[i]; h != NULL; h = next) {
13987 			next = h->dtha_next;
13988 
13989 			if (h->dtha_generation == gen) {
13990 				if (last != NULL) {
13991 					last->dtha_next = next;
13992 				} else {
13993 					help->dthps_actions[i] = next;
13994 				}
13995 
13996 				dtrace_helper_action_destroy(h, vstate);
13997 			} else {
13998 				last = h;
13999 			}
14000 		}
14001 	}
14002 
14003 	/*
14004 	 * Interate until we've cleared out all helper providers with the
14005 	 * given generation number.
14006 	 */
14007 	for (;;) {
14008 		dtrace_helper_provider_t *prov;
14009 
14010 		/*
14011 		 * Look for a helper provider with the right generation. We
14012 		 * have to start back at the beginning of the list each time
14013 		 * because we drop dtrace_lock. It's unlikely that we'll make
14014 		 * more than two passes.
14015 		 */
14016 		for (i = 0; i < help->dthps_nprovs; i++) {
14017 			prov = help->dthps_provs[i];
14018 
14019 			if (prov->dthp_generation == gen)
14020 				break;
14021 		}
14022 
14023 		/*
14024 		 * If there were no matches, we're done.
14025 		 */
14026 		if (i == help->dthps_nprovs)
14027 			break;
14028 
14029 		/*
14030 		 * Move the last helper provider into this slot.
14031 		 */
14032 		help->dthps_nprovs--;
14033 		help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs];
14034 		help->dthps_provs[help->dthps_nprovs] = NULL;
14035 
14036 		mutex_exit(&dtrace_lock);
14037 
14038 		/*
14039 		 * If we have a meta provider, remove this helper provider.
14040 		 */
14041 		mutex_enter(&dtrace_meta_lock);
14042 		if (dtrace_meta_pid != NULL) {
14043 			ASSERT(dtrace_deferred_pid == NULL);
14044 			dtrace_helper_provider_remove(&prov->dthp_prov,
14045 			    p->p_pid);
14046 		}
14047 		mutex_exit(&dtrace_meta_lock);
14048 
14049 		dtrace_helper_provider_destroy(prov);
14050 
14051 		mutex_enter(&dtrace_lock);
14052 	}
14053 
14054 	return (0);
14055 }
14056 
14057 static int
14058 dtrace_helper_validate(dtrace_helper_action_t *helper)
14059 {
14060 	int err = 0, i;
14061 	dtrace_difo_t *dp;
14062 
14063 	if ((dp = helper->dtha_predicate) != NULL)
14064 		err += dtrace_difo_validate_helper(dp);
14065 
14066 	for (i = 0; i < helper->dtha_nactions; i++)
14067 		err += dtrace_difo_validate_helper(helper->dtha_actions[i]);
14068 
14069 	return (err == 0);
14070 }
14071 
14072 static int
14073 dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep)
14074 {
14075 	dtrace_helpers_t *help;
14076 	dtrace_helper_action_t *helper, *last;
14077 	dtrace_actdesc_t *act;
14078 	dtrace_vstate_t *vstate;
14079 	dtrace_predicate_t *pred;
14080 	int count = 0, nactions = 0, i;
14081 
14082 	if (which < 0 || which >= DTRACE_NHELPER_ACTIONS)
14083 		return (EINVAL);
14084 
14085 	help = curproc->p_dtrace_helpers;
14086 	last = help->dthps_actions[which];
14087 	vstate = &help->dthps_vstate;
14088 
14089 	for (count = 0; last != NULL; last = last->dtha_next) {
14090 		count++;
14091 		if (last->dtha_next == NULL)
14092 			break;
14093 	}
14094 
14095 	/*
14096 	 * If we already have dtrace_helper_actions_max helper actions for this
14097 	 * helper action type, we'll refuse to add a new one.
14098 	 */
14099 	if (count >= dtrace_helper_actions_max)
14100 		return (ENOSPC);
14101 
14102 	helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP);
14103 	helper->dtha_generation = help->dthps_generation;
14104 
14105 	if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) {
14106 		ASSERT(pred->dtp_difo != NULL);
14107 		dtrace_difo_hold(pred->dtp_difo);
14108 		helper->dtha_predicate = pred->dtp_difo;
14109 	}
14110 
14111 	for (act = ep->dted_action; act != NULL; act = act->dtad_next) {
14112 		if (act->dtad_kind != DTRACEACT_DIFEXPR)
14113 			goto err;
14114 
14115 		if (act->dtad_difo == NULL)
14116 			goto err;
14117 
14118 		nactions++;
14119 	}
14120 
14121 	helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) *
14122 	    (helper->dtha_nactions = nactions), KM_SLEEP);
14123 
14124 	for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) {
14125 		dtrace_difo_hold(act->dtad_difo);
14126 		helper->dtha_actions[i++] = act->dtad_difo;
14127 	}
14128 
14129 	if (!dtrace_helper_validate(helper))
14130 		goto err;
14131 
14132 	if (last == NULL) {
14133 		help->dthps_actions[which] = helper;
14134 	} else {
14135 		last->dtha_next = helper;
14136 	}
14137 
14138 	if (vstate->dtvs_nlocals > dtrace_helptrace_nlocals) {
14139 		dtrace_helptrace_nlocals = vstate->dtvs_nlocals;
14140 		dtrace_helptrace_next = 0;
14141 	}
14142 
14143 	return (0);
14144 err:
14145 	dtrace_helper_action_destroy(helper, vstate);
14146 	return (EINVAL);
14147 }
14148 
14149 static void
14150 dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help,
14151     dof_helper_t *dofhp)
14152 {
14153 	ASSERT(MUTEX_NOT_HELD(&dtrace_lock));
14154 
14155 	mutex_enter(&dtrace_meta_lock);
14156 	mutex_enter(&dtrace_lock);
14157 
14158 	if (!dtrace_attached() || dtrace_meta_pid == NULL) {
14159 		/*
14160 		 * If the dtrace module is loaded but not attached, or if
14161 		 * there aren't isn't a meta provider registered to deal with
14162 		 * these provider descriptions, we need to postpone creating
14163 		 * the actual providers until later.
14164 		 */
14165 
14166 		if (help->dthps_next == NULL && help->dthps_prev == NULL &&
14167 		    dtrace_deferred_pid != help) {
14168 			help->dthps_deferred = 1;
14169 			help->dthps_pid = p->p_pid;
14170 			help->dthps_next = dtrace_deferred_pid;
14171 			help->dthps_prev = NULL;
14172 			if (dtrace_deferred_pid != NULL)
14173 				dtrace_deferred_pid->dthps_prev = help;
14174 			dtrace_deferred_pid = help;
14175 		}
14176 
14177 		mutex_exit(&dtrace_lock);
14178 
14179 	} else if (dofhp != NULL) {
14180 		/*
14181 		 * If the dtrace module is loaded and we have a particular
14182 		 * helper provider description, pass that off to the
14183 		 * meta provider.
14184 		 */
14185 
14186 		mutex_exit(&dtrace_lock);
14187 
14188 		dtrace_helper_provide(dofhp, p->p_pid);
14189 
14190 	} else {
14191 		/*
14192 		 * Otherwise, just pass all the helper provider descriptions
14193 		 * off to the meta provider.
14194 		 */
14195 
14196 		int i;
14197 		mutex_exit(&dtrace_lock);
14198 
14199 		for (i = 0; i < help->dthps_nprovs; i++) {
14200 			dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
14201 			    p->p_pid);
14202 		}
14203 	}
14204 
14205 	mutex_exit(&dtrace_meta_lock);
14206 }
14207 
14208 static int
14209 dtrace_helper_provider_add(dof_helper_t *dofhp, int gen)
14210 {
14211 	dtrace_helpers_t *help;
14212 	dtrace_helper_provider_t *hprov, **tmp_provs;
14213 	uint_t tmp_maxprovs, i;
14214 
14215 	ASSERT(MUTEX_HELD(&dtrace_lock));
14216 
14217 	help = curproc->p_dtrace_helpers;
14218 	ASSERT(help != NULL);
14219 
14220 	/*
14221 	 * If we already have dtrace_helper_providers_max helper providers,
14222 	 * we're refuse to add a new one.
14223 	 */
14224 	if (help->dthps_nprovs >= dtrace_helper_providers_max)
14225 		return (ENOSPC);
14226 
14227 	/*
14228 	 * Check to make sure this isn't a duplicate.
14229 	 */
14230 	for (i = 0; i < help->dthps_nprovs; i++) {
14231 		if (dofhp->dofhp_addr ==
14232 		    help->dthps_provs[i]->dthp_prov.dofhp_addr)
14233 			return (EALREADY);
14234 	}
14235 
14236 	hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP);
14237 	hprov->dthp_prov = *dofhp;
14238 	hprov->dthp_ref = 1;
14239 	hprov->dthp_generation = gen;
14240 
14241 	/*
14242 	 * Allocate a bigger table for helper providers if it's already full.
14243 	 */
14244 	if (help->dthps_maxprovs == help->dthps_nprovs) {
14245 		tmp_maxprovs = help->dthps_maxprovs;
14246 		tmp_provs = help->dthps_provs;
14247 
14248 		if (help->dthps_maxprovs == 0)
14249 			help->dthps_maxprovs = 2;
14250 		else
14251 			help->dthps_maxprovs *= 2;
14252 		if (help->dthps_maxprovs > dtrace_helper_providers_max)
14253 			help->dthps_maxprovs = dtrace_helper_providers_max;
14254 
14255 		ASSERT(tmp_maxprovs < help->dthps_maxprovs);
14256 
14257 		help->dthps_provs = kmem_zalloc(help->dthps_maxprovs *
14258 		    sizeof (dtrace_helper_provider_t *), KM_SLEEP);
14259 
14260 		if (tmp_provs != NULL) {
14261 			bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs *
14262 			    sizeof (dtrace_helper_provider_t *));
14263 			kmem_free(tmp_provs, tmp_maxprovs *
14264 			    sizeof (dtrace_helper_provider_t *));
14265 		}
14266 	}
14267 
14268 	help->dthps_provs[help->dthps_nprovs] = hprov;
14269 	help->dthps_nprovs++;
14270 
14271 	return (0);
14272 }
14273 
14274 static void
14275 dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov)
14276 {
14277 	mutex_enter(&dtrace_lock);
14278 
14279 	if (--hprov->dthp_ref == 0) {
14280 		dof_hdr_t *dof;
14281 		mutex_exit(&dtrace_lock);
14282 		dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof;
14283 		dtrace_dof_destroy(dof);
14284 		kmem_free(hprov, sizeof (dtrace_helper_provider_t));
14285 	} else {
14286 		mutex_exit(&dtrace_lock);
14287 	}
14288 }
14289 
14290 static int
14291 dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec)
14292 {
14293 	uintptr_t daddr = (uintptr_t)dof;
14294 	dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
14295 	dof_provider_t *provider;
14296 	dof_probe_t *probe;
14297 	uint8_t *arg;
14298 	char *strtab, *typestr;
14299 	dof_stridx_t typeidx;
14300 	size_t typesz;
14301 	uint_t nprobes, j, k;
14302 
14303 	ASSERT(sec->dofs_type == DOF_SECT_PROVIDER);
14304 
14305 	if (sec->dofs_offset & (sizeof (uint_t) - 1)) {
14306 		dtrace_dof_error(dof, "misaligned section offset");
14307 		return (-1);
14308 	}
14309 
14310 	/*
14311 	 * The section needs to be large enough to contain the DOF provider
14312 	 * structure appropriate for the given version.
14313 	 */
14314 	if (sec->dofs_size <
14315 	    ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ?
14316 	    offsetof(dof_provider_t, dofpv_prenoffs) :
14317 	    sizeof (dof_provider_t))) {
14318 		dtrace_dof_error(dof, "provider section too small");
14319 		return (-1);
14320 	}
14321 
14322 	provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
14323 	str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab);
14324 	prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes);
14325 	arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs);
14326 	off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs);
14327 
14328 	if (str_sec == NULL || prb_sec == NULL ||
14329 	    arg_sec == NULL || off_sec == NULL)
14330 		return (-1);
14331 
14332 	enoff_sec = NULL;
14333 
14334 	if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
14335 	    provider->dofpv_prenoffs != DOF_SECT_NONE &&
14336 	    (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS,
14337 	    provider->dofpv_prenoffs)) == NULL)
14338 		return (-1);
14339 
14340 	strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
14341 
14342 	if (provider->dofpv_name >= str_sec->dofs_size ||
14343 	    strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) {
14344 		dtrace_dof_error(dof, "invalid provider name");
14345 		return (-1);
14346 	}
14347 
14348 	if (prb_sec->dofs_entsize == 0 ||
14349 	    prb_sec->dofs_entsize > prb_sec->dofs_size) {
14350 		dtrace_dof_error(dof, "invalid entry size");
14351 		return (-1);
14352 	}
14353 
14354 	if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) {
14355 		dtrace_dof_error(dof, "misaligned entry size");
14356 		return (-1);
14357 	}
14358 
14359 	if (off_sec->dofs_entsize != sizeof (uint32_t)) {
14360 		dtrace_dof_error(dof, "invalid entry size");
14361 		return (-1);
14362 	}
14363 
14364 	if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) {
14365 		dtrace_dof_error(dof, "misaligned section offset");
14366 		return (-1);
14367 	}
14368 
14369 	if (arg_sec->dofs_entsize != sizeof (uint8_t)) {
14370 		dtrace_dof_error(dof, "invalid entry size");
14371 		return (-1);
14372 	}
14373 
14374 	arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
14375 
14376 	nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
14377 
14378 	/*
14379 	 * Take a pass through the probes to check for errors.
14380 	 */
14381 	for (j = 0; j < nprobes; j++) {
14382 		probe = (dof_probe_t *)(uintptr_t)(daddr +
14383 		    prb_sec->dofs_offset + j * prb_sec->dofs_entsize);
14384 
14385 		if (probe->dofpr_func >= str_sec->dofs_size) {
14386 			dtrace_dof_error(dof, "invalid function name");
14387 			return (-1);
14388 		}
14389 
14390 		if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) {
14391 			dtrace_dof_error(dof, "function name too long");
14392 			return (-1);
14393 		}
14394 
14395 		if (probe->dofpr_name >= str_sec->dofs_size ||
14396 		    strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) {
14397 			dtrace_dof_error(dof, "invalid probe name");
14398 			return (-1);
14399 		}
14400 
14401 		/*
14402 		 * The offset count must not wrap the index, and the offsets
14403 		 * must also not overflow the section's data.
14404 		 */
14405 		if (probe->dofpr_offidx + probe->dofpr_noffs <
14406 		    probe->dofpr_offidx ||
14407 		    (probe->dofpr_offidx + probe->dofpr_noffs) *
14408 		    off_sec->dofs_entsize > off_sec->dofs_size) {
14409 			dtrace_dof_error(dof, "invalid probe offset");
14410 			return (-1);
14411 		}
14412 
14413 		if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) {
14414 			/*
14415 			 * If there's no is-enabled offset section, make sure
14416 			 * there aren't any is-enabled offsets. Otherwise
14417 			 * perform the same checks as for probe offsets
14418 			 * (immediately above).
14419 			 */
14420 			if (enoff_sec == NULL) {
14421 				if (probe->dofpr_enoffidx != 0 ||
14422 				    probe->dofpr_nenoffs != 0) {
14423 					dtrace_dof_error(dof, "is-enabled "
14424 					    "offsets with null section");
14425 					return (-1);
14426 				}
14427 			} else if (probe->dofpr_enoffidx +
14428 			    probe->dofpr_nenoffs < probe->dofpr_enoffidx ||
14429 			    (probe->dofpr_enoffidx + probe->dofpr_nenoffs) *
14430 			    enoff_sec->dofs_entsize > enoff_sec->dofs_size) {
14431 				dtrace_dof_error(dof, "invalid is-enabled "
14432 				    "offset");
14433 				return (-1);
14434 			}
14435 
14436 			if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) {
14437 				dtrace_dof_error(dof, "zero probe and "
14438 				    "is-enabled offsets");
14439 				return (-1);
14440 			}
14441 		} else if (probe->dofpr_noffs == 0) {
14442 			dtrace_dof_error(dof, "zero probe offsets");
14443 			return (-1);
14444 		}
14445 
14446 		if (probe->dofpr_argidx + probe->dofpr_xargc <
14447 		    probe->dofpr_argidx ||
14448 		    (probe->dofpr_argidx + probe->dofpr_xargc) *
14449 		    arg_sec->dofs_entsize > arg_sec->dofs_size) {
14450 			dtrace_dof_error(dof, "invalid args");
14451 			return (-1);
14452 		}
14453 
14454 		typeidx = probe->dofpr_nargv;
14455 		typestr = strtab + probe->dofpr_nargv;
14456 		for (k = 0; k < probe->dofpr_nargc; k++) {
14457 			if (typeidx >= str_sec->dofs_size) {
14458 				dtrace_dof_error(dof, "bad "
14459 				    "native argument type");
14460 				return (-1);
14461 			}
14462 
14463 			typesz = strlen(typestr) + 1;
14464 			if (typesz > DTRACE_ARGTYPELEN) {
14465 				dtrace_dof_error(dof, "native "
14466 				    "argument type too long");
14467 				return (-1);
14468 			}
14469 			typeidx += typesz;
14470 			typestr += typesz;
14471 		}
14472 
14473 		typeidx = probe->dofpr_xargv;
14474 		typestr = strtab + probe->dofpr_xargv;
14475 		for (k = 0; k < probe->dofpr_xargc; k++) {
14476 			if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) {
14477 				dtrace_dof_error(dof, "bad "
14478 				    "native argument index");
14479 				return (-1);
14480 			}
14481 
14482 			if (typeidx >= str_sec->dofs_size) {
14483 				dtrace_dof_error(dof, "bad "
14484 				    "translated argument type");
14485 				return (-1);
14486 			}
14487 
14488 			typesz = strlen(typestr) + 1;
14489 			if (typesz > DTRACE_ARGTYPELEN) {
14490 				dtrace_dof_error(dof, "translated argument "
14491 				    "type too long");
14492 				return (-1);
14493 			}
14494 
14495 			typeidx += typesz;
14496 			typestr += typesz;
14497 		}
14498 	}
14499 
14500 	return (0);
14501 }
14502 
14503 static int
14504 dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp)
14505 {
14506 	dtrace_helpers_t *help;
14507 	dtrace_vstate_t *vstate;
14508 	dtrace_enabling_t *enab = NULL;
14509 	int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1;
14510 	uintptr_t daddr = (uintptr_t)dof;
14511 
14512 	ASSERT(MUTEX_HELD(&dtrace_lock));
14513 
14514 	if ((help = curproc->p_dtrace_helpers) == NULL)
14515 		help = dtrace_helpers_create(curproc);
14516 
14517 	vstate = &help->dthps_vstate;
14518 
14519 	if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab,
14520 	    dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) {
14521 		dtrace_dof_destroy(dof);
14522 		return (rv);
14523 	}
14524 
14525 	/*
14526 	 * Look for helper providers and validate their descriptions.
14527 	 */
14528 	if (dhp != NULL) {
14529 		for (i = 0; i < dof->dofh_secnum; i++) {
14530 			dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
14531 			    dof->dofh_secoff + i * dof->dofh_secsize);
14532 
14533 			if (sec->dofs_type != DOF_SECT_PROVIDER)
14534 				continue;
14535 
14536 			if (dtrace_helper_provider_validate(dof, sec) != 0) {
14537 				dtrace_enabling_destroy(enab);
14538 				dtrace_dof_destroy(dof);
14539 				return (-1);
14540 			}
14541 
14542 			nprovs++;
14543 		}
14544 	}
14545 
14546 	/*
14547 	 * Now we need to walk through the ECB descriptions in the enabling.
14548 	 */
14549 	for (i = 0; i < enab->dten_ndesc; i++) {
14550 		dtrace_ecbdesc_t *ep = enab->dten_desc[i];
14551 		dtrace_probedesc_t *desc = &ep->dted_probe;
14552 
14553 		if (strcmp(desc->dtpd_provider, "dtrace") != 0)
14554 			continue;
14555 
14556 		if (strcmp(desc->dtpd_mod, "helper") != 0)
14557 			continue;
14558 
14559 		if (strcmp(desc->dtpd_func, "ustack") != 0)
14560 			continue;
14561 
14562 		if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK,
14563 		    ep)) != 0) {
14564 			/*
14565 			 * Adding this helper action failed -- we are now going
14566 			 * to rip out the entire generation and return failure.
14567 			 */
14568 			(void) dtrace_helper_destroygen(help->dthps_generation);
14569 			dtrace_enabling_destroy(enab);
14570 			dtrace_dof_destroy(dof);
14571 			return (-1);
14572 		}
14573 
14574 		nhelpers++;
14575 	}
14576 
14577 	if (nhelpers < enab->dten_ndesc)
14578 		dtrace_dof_error(dof, "unmatched helpers");
14579 
14580 	gen = help->dthps_generation++;
14581 	dtrace_enabling_destroy(enab);
14582 
14583 	if (dhp != NULL && nprovs > 0) {
14584 		dhp->dofhp_dof = (uint64_t)(uintptr_t)dof;
14585 		if (dtrace_helper_provider_add(dhp, gen) == 0) {
14586 			mutex_exit(&dtrace_lock);
14587 			dtrace_helper_provider_register(curproc, help, dhp);
14588 			mutex_enter(&dtrace_lock);
14589 
14590 			destroy = 0;
14591 		}
14592 	}
14593 
14594 	if (destroy)
14595 		dtrace_dof_destroy(dof);
14596 
14597 	return (gen);
14598 }
14599 
14600 static dtrace_helpers_t *
14601 dtrace_helpers_create(proc_t *p)
14602 {
14603 	dtrace_helpers_t *help;
14604 
14605 	ASSERT(MUTEX_HELD(&dtrace_lock));
14606 	ASSERT(p->p_dtrace_helpers == NULL);
14607 
14608 	help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP);
14609 	help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) *
14610 	    DTRACE_NHELPER_ACTIONS, KM_SLEEP);
14611 
14612 	p->p_dtrace_helpers = help;
14613 	dtrace_helpers++;
14614 
14615 	return (help);
14616 }
14617 
14618 #if defined(sun)
14619 static
14620 #endif
14621 void
14622 dtrace_helpers_destroy(proc_t *p)
14623 {
14624 	dtrace_helpers_t *help;
14625 	dtrace_vstate_t *vstate;
14626 #if defined(sun)
14627 	proc_t *p = curproc;
14628 #endif
14629 	int i;
14630 
14631 	mutex_enter(&dtrace_lock);
14632 
14633 	ASSERT(p->p_dtrace_helpers != NULL);
14634 	ASSERT(dtrace_helpers > 0);
14635 
14636 	help = p->p_dtrace_helpers;
14637 	vstate = &help->dthps_vstate;
14638 
14639 	/*
14640 	 * We're now going to lose the help from this process.
14641 	 */
14642 	p->p_dtrace_helpers = NULL;
14643 	dtrace_sync();
14644 
14645 	/*
14646 	 * Destory the helper actions.
14647 	 */
14648 	for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
14649 		dtrace_helper_action_t *h, *next;
14650 
14651 		for (h = help->dthps_actions[i]; h != NULL; h = next) {
14652 			next = h->dtha_next;
14653 			dtrace_helper_action_destroy(h, vstate);
14654 			h = next;
14655 		}
14656 	}
14657 
14658 	mutex_exit(&dtrace_lock);
14659 
14660 	/*
14661 	 * Destroy the helper providers.
14662 	 */
14663 	if (help->dthps_maxprovs > 0) {
14664 		mutex_enter(&dtrace_meta_lock);
14665 		if (dtrace_meta_pid != NULL) {
14666 			ASSERT(dtrace_deferred_pid == NULL);
14667 
14668 			for (i = 0; i < help->dthps_nprovs; i++) {
14669 				dtrace_helper_provider_remove(
14670 				    &help->dthps_provs[i]->dthp_prov, p->p_pid);
14671 			}
14672 		} else {
14673 			mutex_enter(&dtrace_lock);
14674 			ASSERT(help->dthps_deferred == 0 ||
14675 			    help->dthps_next != NULL ||
14676 			    help->dthps_prev != NULL ||
14677 			    help == dtrace_deferred_pid);
14678 
14679 			/*
14680 			 * Remove the helper from the deferred list.
14681 			 */
14682 			if (help->dthps_next != NULL)
14683 				help->dthps_next->dthps_prev = help->dthps_prev;
14684 			if (help->dthps_prev != NULL)
14685 				help->dthps_prev->dthps_next = help->dthps_next;
14686 			if (dtrace_deferred_pid == help) {
14687 				dtrace_deferred_pid = help->dthps_next;
14688 				ASSERT(help->dthps_prev == NULL);
14689 			}
14690 
14691 			mutex_exit(&dtrace_lock);
14692 		}
14693 
14694 		mutex_exit(&dtrace_meta_lock);
14695 
14696 		for (i = 0; i < help->dthps_nprovs; i++) {
14697 			dtrace_helper_provider_destroy(help->dthps_provs[i]);
14698 		}
14699 
14700 		kmem_free(help->dthps_provs, help->dthps_maxprovs *
14701 		    sizeof (dtrace_helper_provider_t *));
14702 	}
14703 
14704 	mutex_enter(&dtrace_lock);
14705 
14706 	dtrace_vstate_fini(&help->dthps_vstate);
14707 	kmem_free(help->dthps_actions,
14708 	    sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS);
14709 	kmem_free(help, sizeof (dtrace_helpers_t));
14710 
14711 	--dtrace_helpers;
14712 	mutex_exit(&dtrace_lock);
14713 }
14714 
14715 #if defined(sun)
14716 static
14717 #endif
14718 void
14719 dtrace_helpers_duplicate(proc_t *from, proc_t *to)
14720 {
14721 	dtrace_helpers_t *help, *newhelp;
14722 	dtrace_helper_action_t *helper, *new, *last;
14723 	dtrace_difo_t *dp;
14724 	dtrace_vstate_t *vstate;
14725 	int i, j, sz, hasprovs = 0;
14726 
14727 	mutex_enter(&dtrace_lock);
14728 	ASSERT(from->p_dtrace_helpers != NULL);
14729 	ASSERT(dtrace_helpers > 0);
14730 
14731 	help = from->p_dtrace_helpers;
14732 	newhelp = dtrace_helpers_create(to);
14733 	ASSERT(to->p_dtrace_helpers != NULL);
14734 
14735 	newhelp->dthps_generation = help->dthps_generation;
14736 	vstate = &newhelp->dthps_vstate;
14737 
14738 	/*
14739 	 * Duplicate the helper actions.
14740 	 */
14741 	for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
14742 		if ((helper = help->dthps_actions[i]) == NULL)
14743 			continue;
14744 
14745 		for (last = NULL; helper != NULL; helper = helper->dtha_next) {
14746 			new = kmem_zalloc(sizeof (dtrace_helper_action_t),
14747 			    KM_SLEEP);
14748 			new->dtha_generation = helper->dtha_generation;
14749 
14750 			if ((dp = helper->dtha_predicate) != NULL) {
14751 				dp = dtrace_difo_duplicate(dp, vstate);
14752 				new->dtha_predicate = dp;
14753 			}
14754 
14755 			new->dtha_nactions = helper->dtha_nactions;
14756 			sz = sizeof (dtrace_difo_t *) * new->dtha_nactions;
14757 			new->dtha_actions = kmem_alloc(sz, KM_SLEEP);
14758 
14759 			for (j = 0; j < new->dtha_nactions; j++) {
14760 				dtrace_difo_t *dp = helper->dtha_actions[j];
14761 
14762 				ASSERT(dp != NULL);
14763 				dp = dtrace_difo_duplicate(dp, vstate);
14764 				new->dtha_actions[j] = dp;
14765 			}
14766 
14767 			if (last != NULL) {
14768 				last->dtha_next = new;
14769 			} else {
14770 				newhelp->dthps_actions[i] = new;
14771 			}
14772 
14773 			last = new;
14774 		}
14775 	}
14776 
14777 	/*
14778 	 * Duplicate the helper providers and register them with the
14779 	 * DTrace framework.
14780 	 */
14781 	if (help->dthps_nprovs > 0) {
14782 		newhelp->dthps_nprovs = help->dthps_nprovs;
14783 		newhelp->dthps_maxprovs = help->dthps_nprovs;
14784 		newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs *
14785 		    sizeof (dtrace_helper_provider_t *), KM_SLEEP);
14786 		for (i = 0; i < newhelp->dthps_nprovs; i++) {
14787 			newhelp->dthps_provs[i] = help->dthps_provs[i];
14788 			newhelp->dthps_provs[i]->dthp_ref++;
14789 		}
14790 
14791 		hasprovs = 1;
14792 	}
14793 
14794 	mutex_exit(&dtrace_lock);
14795 
14796 	if (hasprovs)
14797 		dtrace_helper_provider_register(to, newhelp, NULL);
14798 }
14799 
14800 #if defined(sun)
14801 /*
14802  * DTrace Hook Functions
14803  */
14804 static void
14805 dtrace_module_loaded(modctl_t *ctl)
14806 {
14807 	dtrace_provider_t *prv;
14808 
14809 	mutex_enter(&dtrace_provider_lock);
14810 	mutex_enter(&mod_lock);
14811 
14812 	ASSERT(ctl->mod_busy);
14813 
14814 	/*
14815 	 * We're going to call each providers per-module provide operation
14816 	 * specifying only this module.
14817 	 */
14818 	for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next)
14819 		prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
14820 
14821 	mutex_exit(&mod_lock);
14822 	mutex_exit(&dtrace_provider_lock);
14823 
14824 	/*
14825 	 * If we have any retained enablings, we need to match against them.
14826 	 * Enabling probes requires that cpu_lock be held, and we cannot hold
14827 	 * cpu_lock here -- it is legal for cpu_lock to be held when loading a
14828 	 * module.  (In particular, this happens when loading scheduling
14829 	 * classes.)  So if we have any retained enablings, we need to dispatch
14830 	 * our task queue to do the match for us.
14831 	 */
14832 	mutex_enter(&dtrace_lock);
14833 
14834 	if (dtrace_retained == NULL) {
14835 		mutex_exit(&dtrace_lock);
14836 		return;
14837 	}
14838 
14839 	(void) taskq_dispatch(dtrace_taskq,
14840 	    (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP);
14841 
14842 	mutex_exit(&dtrace_lock);
14843 
14844 	/*
14845 	 * And now, for a little heuristic sleaze:  in general, we want to
14846 	 * match modules as soon as they load.  However, we cannot guarantee
14847 	 * this, because it would lead us to the lock ordering violation
14848 	 * outlined above.  The common case, of course, is that cpu_lock is
14849 	 * _not_ held -- so we delay here for a clock tick, hoping that that's
14850 	 * long enough for the task queue to do its work.  If it's not, it's
14851 	 * not a serious problem -- it just means that the module that we
14852 	 * just loaded may not be immediately instrumentable.
14853 	 */
14854 	delay(1);
14855 }
14856 
14857 static void
14858 dtrace_module_unloaded(modctl_t *ctl)
14859 {
14860 	dtrace_probe_t template, *probe, *first, *next;
14861 	dtrace_provider_t *prov;
14862 
14863 	template.dtpr_mod = ctl->mod_modname;
14864 
14865 	mutex_enter(&dtrace_provider_lock);
14866 	mutex_enter(&mod_lock);
14867 	mutex_enter(&dtrace_lock);
14868 
14869 	if (dtrace_bymod == NULL) {
14870 		/*
14871 		 * The DTrace module is loaded (obviously) but not attached;
14872 		 * we don't have any work to do.
14873 		 */
14874 		mutex_exit(&dtrace_provider_lock);
14875 		mutex_exit(&mod_lock);
14876 		mutex_exit(&dtrace_lock);
14877 		return;
14878 	}
14879 
14880 	for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template);
14881 	    probe != NULL; probe = probe->dtpr_nextmod) {
14882 		if (probe->dtpr_ecb != NULL) {
14883 			mutex_exit(&dtrace_provider_lock);
14884 			mutex_exit(&mod_lock);
14885 			mutex_exit(&dtrace_lock);
14886 
14887 			/*
14888 			 * This shouldn't _actually_ be possible -- we're
14889 			 * unloading a module that has an enabled probe in it.
14890 			 * (It's normally up to the provider to make sure that
14891 			 * this can't happen.)  However, because dtps_enable()
14892 			 * doesn't have a failure mode, there can be an
14893 			 * enable/unload race.  Upshot:  we don't want to
14894 			 * assert, but we're not going to disable the
14895 			 * probe, either.
14896 			 */
14897 			if (dtrace_err_verbose) {
14898 				cmn_err(CE_WARN, "unloaded module '%s' had "
14899 				    "enabled probes", ctl->mod_modname);
14900 			}
14901 
14902 			return;
14903 		}
14904 	}
14905 
14906 	probe = first;
14907 
14908 	for (first = NULL; probe != NULL; probe = next) {
14909 		ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe);
14910 
14911 		dtrace_probes[probe->dtpr_id - 1] = NULL;
14912 
14913 		next = probe->dtpr_nextmod;
14914 		dtrace_hash_remove(dtrace_bymod, probe);
14915 		dtrace_hash_remove(dtrace_byfunc, probe);
14916 		dtrace_hash_remove(dtrace_byname, probe);
14917 
14918 		if (first == NULL) {
14919 			first = probe;
14920 			probe->dtpr_nextmod = NULL;
14921 		} else {
14922 			probe->dtpr_nextmod = first;
14923 			first = probe;
14924 		}
14925 	}
14926 
14927 	/*
14928 	 * We've removed all of the module's probes from the hash chains and
14929 	 * from the probe array.  Now issue a dtrace_sync() to be sure that
14930 	 * everyone has cleared out from any probe array processing.
14931 	 */
14932 	dtrace_sync();
14933 
14934 	for (probe = first; probe != NULL; probe = first) {
14935 		first = probe->dtpr_nextmod;
14936 		prov = probe->dtpr_provider;
14937 		prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id,
14938 		    probe->dtpr_arg);
14939 		kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
14940 		kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
14941 		kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
14942 		vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1);
14943 		kmem_free(probe, sizeof (dtrace_probe_t));
14944 	}
14945 
14946 	mutex_exit(&dtrace_lock);
14947 	mutex_exit(&mod_lock);
14948 	mutex_exit(&dtrace_provider_lock);
14949 }
14950 
14951 static void
14952 dtrace_suspend(void)
14953 {
14954 	dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend));
14955 }
14956 
14957 static void
14958 dtrace_resume(void)
14959 {
14960 	dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume));
14961 }
14962 #endif
14963 
14964 static int
14965 dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu)
14966 {
14967 	ASSERT(MUTEX_HELD(&cpu_lock));
14968 	mutex_enter(&dtrace_lock);
14969 
14970 	switch (what) {
14971 	case CPU_CONFIG: {
14972 		dtrace_state_t *state;
14973 		dtrace_optval_t *opt, rs, c;
14974 
14975 		/*
14976 		 * For now, we only allocate a new buffer for anonymous state.
14977 		 */
14978 		if ((state = dtrace_anon.dta_state) == NULL)
14979 			break;
14980 
14981 		if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
14982 			break;
14983 
14984 		opt = state->dts_options;
14985 		c = opt[DTRACEOPT_CPU];
14986 
14987 		if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu)
14988 			break;
14989 
14990 		/*
14991 		 * Regardless of what the actual policy is, we're going to
14992 		 * temporarily set our resize policy to be manual.  We're
14993 		 * also going to temporarily set our CPU option to denote
14994 		 * the newly configured CPU.
14995 		 */
14996 		rs = opt[DTRACEOPT_BUFRESIZE];
14997 		opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL;
14998 		opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu;
14999 
15000 		(void) dtrace_state_buffers(state);
15001 
15002 		opt[DTRACEOPT_BUFRESIZE] = rs;
15003 		opt[DTRACEOPT_CPU] = c;
15004 
15005 		break;
15006 	}
15007 
15008 	case CPU_UNCONFIG:
15009 		/*
15010 		 * We don't free the buffer in the CPU_UNCONFIG case.  (The
15011 		 * buffer will be freed when the consumer exits.)
15012 		 */
15013 		break;
15014 
15015 	default:
15016 		break;
15017 	}
15018 
15019 	mutex_exit(&dtrace_lock);
15020 	return (0);
15021 }
15022 
15023 #if defined(sun)
15024 static void
15025 dtrace_cpu_setup_initial(processorid_t cpu)
15026 {
15027 	(void) dtrace_cpu_setup(CPU_CONFIG, cpu);
15028 }
15029 #endif
15030 
15031 static void
15032 dtrace_toxrange_add(uintptr_t base, uintptr_t limit)
15033 {
15034 	if (dtrace_toxranges >= dtrace_toxranges_max) {
15035 		int osize, nsize;
15036 		dtrace_toxrange_t *range;
15037 
15038 		osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15039 
15040 		if (osize == 0) {
15041 			ASSERT(dtrace_toxrange == NULL);
15042 			ASSERT(dtrace_toxranges_max == 0);
15043 			dtrace_toxranges_max = 1;
15044 		} else {
15045 			dtrace_toxranges_max <<= 1;
15046 		}
15047 
15048 		nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15049 		range = kmem_zalloc(nsize, KM_SLEEP);
15050 
15051 		if (dtrace_toxrange != NULL) {
15052 			ASSERT(osize != 0);
15053 			bcopy(dtrace_toxrange, range, osize);
15054 			kmem_free(dtrace_toxrange, osize);
15055 		}
15056 
15057 		dtrace_toxrange = range;
15058 	}
15059 
15060 	ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == 0);
15061 	ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == 0);
15062 
15063 	dtrace_toxrange[dtrace_toxranges].dtt_base = base;
15064 	dtrace_toxrange[dtrace_toxranges].dtt_limit = limit;
15065 	dtrace_toxranges++;
15066 }
15067 
15068 /*
15069  * DTrace Driver Cookbook Functions
15070  */
15071 #if defined(sun)
15072 /*ARGSUSED*/
15073 static int
15074 dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
15075 {
15076 	dtrace_provider_id_t id;
15077 	dtrace_state_t *state = NULL;
15078 	dtrace_enabling_t *enab;
15079 
15080 	mutex_enter(&cpu_lock);
15081 	mutex_enter(&dtrace_provider_lock);
15082 	mutex_enter(&dtrace_lock);
15083 
15084 	if (ddi_soft_state_init(&dtrace_softstate,
15085 	    sizeof (dtrace_state_t), 0) != 0) {
15086 		cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state");
15087 		mutex_exit(&cpu_lock);
15088 		mutex_exit(&dtrace_provider_lock);
15089 		mutex_exit(&dtrace_lock);
15090 		return (DDI_FAILURE);
15091 	}
15092 
15093 	if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR,
15094 	    DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE ||
15095 	    ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR,
15096 	    DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) {
15097 		cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes");
15098 		ddi_remove_minor_node(devi, NULL);
15099 		ddi_soft_state_fini(&dtrace_softstate);
15100 		mutex_exit(&cpu_lock);
15101 		mutex_exit(&dtrace_provider_lock);
15102 		mutex_exit(&dtrace_lock);
15103 		return (DDI_FAILURE);
15104 	}
15105 
15106 	ddi_report_dev(devi);
15107 	dtrace_devi = devi;
15108 
15109 	dtrace_modload = dtrace_module_loaded;
15110 	dtrace_modunload = dtrace_module_unloaded;
15111 	dtrace_cpu_init = dtrace_cpu_setup_initial;
15112 	dtrace_helpers_cleanup = dtrace_helpers_destroy;
15113 	dtrace_helpers_fork = dtrace_helpers_duplicate;
15114 	dtrace_cpustart_init = dtrace_suspend;
15115 	dtrace_cpustart_fini = dtrace_resume;
15116 	dtrace_debugger_init = dtrace_suspend;
15117 	dtrace_debugger_fini = dtrace_resume;
15118 
15119 	register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
15120 
15121 	ASSERT(MUTEX_HELD(&cpu_lock));
15122 
15123 	dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1,
15124 	    NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
15125 	dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE,
15126 	    UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0,
15127 	    VM_SLEEP | VMC_IDENTIFIER);
15128 	dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri,
15129 	    1, INT_MAX, 0);
15130 
15131 	dtrace_state_cache = kmem_cache_create("dtrace_state_cache",
15132 	    sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN,
15133 	    NULL, NULL, NULL, NULL, NULL, 0);
15134 
15135 	ASSERT(MUTEX_HELD(&cpu_lock));
15136 	dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod),
15137 	    offsetof(dtrace_probe_t, dtpr_nextmod),
15138 	    offsetof(dtrace_probe_t, dtpr_prevmod));
15139 
15140 	dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func),
15141 	    offsetof(dtrace_probe_t, dtpr_nextfunc),
15142 	    offsetof(dtrace_probe_t, dtpr_prevfunc));
15143 
15144 	dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name),
15145 	    offsetof(dtrace_probe_t, dtpr_nextname),
15146 	    offsetof(dtrace_probe_t, dtpr_prevname));
15147 
15148 	if (dtrace_retain_max < 1) {
15149 		cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; "
15150 		    "setting to 1", dtrace_retain_max);
15151 		dtrace_retain_max = 1;
15152 	}
15153 
15154 	/*
15155 	 * Now discover our toxic ranges.
15156 	 */
15157 	dtrace_toxic_ranges(dtrace_toxrange_add);
15158 
15159 	/*
15160 	 * Before we register ourselves as a provider to our own framework,
15161 	 * we would like to assert that dtrace_provider is NULL -- but that's
15162 	 * not true if we were loaded as a dependency of a DTrace provider.
15163 	 * Once we've registered, we can assert that dtrace_provider is our
15164 	 * pseudo provider.
15165 	 */
15166 	(void) dtrace_register("dtrace", &dtrace_provider_attr,
15167 	    DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id);
15168 
15169 	ASSERT(dtrace_provider != NULL);
15170 	ASSERT((dtrace_provider_id_t)dtrace_provider == id);
15171 
15172 	dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t)
15173 	    dtrace_provider, NULL, NULL, "BEGIN", 0, NULL);
15174 	dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t)
15175 	    dtrace_provider, NULL, NULL, "END", 0, NULL);
15176 	dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t)
15177 	    dtrace_provider, NULL, NULL, "ERROR", 1, NULL);
15178 
15179 	dtrace_anon_property();
15180 	mutex_exit(&cpu_lock);
15181 
15182 	/*
15183 	 * If DTrace helper tracing is enabled, we need to allocate the
15184 	 * trace buffer and initialize the values.
15185 	 */
15186 	if (dtrace_helptrace_enabled) {
15187 		ASSERT(dtrace_helptrace_buffer == NULL);
15188 		dtrace_helptrace_buffer =
15189 		    kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP);
15190 		dtrace_helptrace_next = 0;
15191 	}
15192 
15193 	/*
15194 	 * If there are already providers, we must ask them to provide their
15195 	 * probes, and then match any anonymous enabling against them.  Note
15196 	 * that there should be no other retained enablings at this time:
15197 	 * the only retained enablings at this time should be the anonymous
15198 	 * enabling.
15199 	 */
15200 	if (dtrace_anon.dta_enabling != NULL) {
15201 		ASSERT(dtrace_retained == dtrace_anon.dta_enabling);
15202 
15203 		dtrace_enabling_provide(NULL);
15204 		state = dtrace_anon.dta_state;
15205 
15206 		/*
15207 		 * We couldn't hold cpu_lock across the above call to
15208 		 * dtrace_enabling_provide(), but we must hold it to actually
15209 		 * enable the probes.  We have to drop all of our locks, pick
15210 		 * up cpu_lock, and regain our locks before matching the
15211 		 * retained anonymous enabling.
15212 		 */
15213 		mutex_exit(&dtrace_lock);
15214 		mutex_exit(&dtrace_provider_lock);
15215 
15216 		mutex_enter(&cpu_lock);
15217 		mutex_enter(&dtrace_provider_lock);
15218 		mutex_enter(&dtrace_lock);
15219 
15220 		if ((enab = dtrace_anon.dta_enabling) != NULL)
15221 			(void) dtrace_enabling_match(enab, NULL);
15222 
15223 		mutex_exit(&cpu_lock);
15224 	}
15225 
15226 	mutex_exit(&dtrace_lock);
15227 	mutex_exit(&dtrace_provider_lock);
15228 
15229 	if (state != NULL) {
15230 		/*
15231 		 * If we created any anonymous state, set it going now.
15232 		 */
15233 		(void) dtrace_state_go(state, &dtrace_anon.dta_beganon);
15234 	}
15235 
15236 	return (DDI_SUCCESS);
15237 }
15238 #endif
15239 
15240 #if !defined(sun)
15241 #if __FreeBSD_version >= 800039
15242 static void
15243 dtrace_dtr(void *data __unused)
15244 {
15245 }
15246 #endif
15247 #endif
15248 
15249 /*ARGSUSED*/
15250 static int
15251 #if defined(sun)
15252 dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
15253 #else
15254 dtrace_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
15255 #endif
15256 {
15257 	dtrace_state_t *state;
15258 	uint32_t priv;
15259 	uid_t uid;
15260 	zoneid_t zoneid;
15261 
15262 #if defined(sun)
15263 	if (getminor(*devp) == DTRACEMNRN_HELPER)
15264 		return (0);
15265 
15266 	/*
15267 	 * If this wasn't an open with the "helper" minor, then it must be
15268 	 * the "dtrace" minor.
15269 	 */
15270 	ASSERT(getminor(*devp) == DTRACEMNRN_DTRACE);
15271 #else
15272 	cred_t *cred_p = NULL;
15273 
15274 #if __FreeBSD_version < 800039
15275 	/*
15276 	 * The first minor device is the one that is cloned so there is
15277 	 * nothing more to do here.
15278 	 */
15279 	if (dev2unit(dev) == 0)
15280 		return 0;
15281 
15282 	/*
15283 	 * Devices are cloned, so if the DTrace state has already
15284 	 * been allocated, that means this device belongs to a
15285 	 * different client. Each client should open '/dev/dtrace'
15286 	 * to get a cloned device.
15287 	 */
15288 	if (dev->si_drv1 != NULL)
15289 		return (EBUSY);
15290 #endif
15291 
15292 	cred_p = dev->si_cred;
15293 #endif
15294 
15295 	/*
15296 	 * If no DTRACE_PRIV_* bits are set in the credential, then the
15297 	 * caller lacks sufficient permission to do anything with DTrace.
15298 	 */
15299 	dtrace_cred2priv(cred_p, &priv, &uid, &zoneid);
15300 	if (priv == DTRACE_PRIV_NONE) {
15301 #if !defined(sun)
15302 #if __FreeBSD_version < 800039
15303 		/* Destroy the cloned device. */
15304                 destroy_dev(dev);
15305 #endif
15306 #endif
15307 
15308 		return (EACCES);
15309 	}
15310 
15311 	/*
15312 	 * Ask all providers to provide all their probes.
15313 	 */
15314 	mutex_enter(&dtrace_provider_lock);
15315 	dtrace_probe_provide(NULL, NULL);
15316 	mutex_exit(&dtrace_provider_lock);
15317 
15318 	mutex_enter(&cpu_lock);
15319 	mutex_enter(&dtrace_lock);
15320 	dtrace_opens++;
15321 	dtrace_membar_producer();
15322 
15323 #if defined(sun)
15324 	/*
15325 	 * If the kernel debugger is active (that is, if the kernel debugger
15326 	 * modified text in some way), we won't allow the open.
15327 	 */
15328 	if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
15329 		dtrace_opens--;
15330 		mutex_exit(&cpu_lock);
15331 		mutex_exit(&dtrace_lock);
15332 		return (EBUSY);
15333 	}
15334 
15335 	state = dtrace_state_create(devp, cred_p);
15336 #else
15337 	state = dtrace_state_create(dev);
15338 #if __FreeBSD_version < 800039
15339 	dev->si_drv1 = state;
15340 #else
15341 	devfs_set_cdevpriv(state, dtrace_dtr);
15342 #endif
15343 #endif
15344 
15345 	mutex_exit(&cpu_lock);
15346 
15347 	if (state == NULL) {
15348 #if defined(sun)
15349 		if (--dtrace_opens == 0)
15350 			(void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15351 #else
15352 		--dtrace_opens;
15353 #endif
15354 		mutex_exit(&dtrace_lock);
15355 #if !defined(sun)
15356 #if __FreeBSD_version < 800039
15357 		/* Destroy the cloned device. */
15358                 destroy_dev(dev);
15359 #endif
15360 #endif
15361 		return (EAGAIN);
15362 	}
15363 
15364 	mutex_exit(&dtrace_lock);
15365 
15366 	return (0);
15367 }
15368 
15369 /*ARGSUSED*/
15370 static int
15371 #if defined(sun)
15372 dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
15373 #else
15374 dtrace_close(struct cdev *dev, int flags, int fmt __unused, struct thread *td)
15375 #endif
15376 {
15377 #if defined(sun)
15378 	minor_t minor = getminor(dev);
15379 	dtrace_state_t *state;
15380 
15381 	if (minor == DTRACEMNRN_HELPER)
15382 		return (0);
15383 
15384 	state = ddi_get_soft_state(dtrace_softstate, minor);
15385 #else
15386 #if __FreeBSD_version < 800039
15387 	dtrace_state_t *state = dev->si_drv1;
15388 
15389 	/* Check if this is not a cloned device. */
15390 	if (dev2unit(dev) == 0)
15391 		return (0);
15392 #else
15393 	dtrace_state_t *state;
15394 	devfs_get_cdevpriv((void **) &state);
15395 #endif
15396 
15397 #endif
15398 
15399 	mutex_enter(&cpu_lock);
15400 	mutex_enter(&dtrace_lock);
15401 
15402 	if (state != NULL) {
15403 		if (state->dts_anon) {
15404 			/*
15405 			 * There is anonymous state. Destroy that first.
15406 			 */
15407 			ASSERT(dtrace_anon.dta_state == NULL);
15408 			dtrace_state_destroy(state->dts_anon);
15409 		}
15410 
15411 		dtrace_state_destroy(state);
15412 
15413 #if !defined(sun)
15414 		kmem_free(state, 0);
15415 #if __FreeBSD_version < 800039
15416 		dev->si_drv1 = NULL;
15417 #else
15418 		devfs_clear_cdevpriv();
15419 #endif
15420 #endif
15421 	}
15422 
15423 	ASSERT(dtrace_opens > 0);
15424 #if defined(sun)
15425 	if (--dtrace_opens == 0)
15426 		(void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15427 #else
15428 	--dtrace_opens;
15429 #endif
15430 
15431 	mutex_exit(&dtrace_lock);
15432 	mutex_exit(&cpu_lock);
15433 
15434 #if __FreeBSD_version < 800039
15435 	/* Schedule this cloned device to be destroyed. */
15436 	destroy_dev_sched(dev);
15437 #endif
15438 
15439 	return (0);
15440 }
15441 
15442 #if defined(sun)
15443 /*ARGSUSED*/
15444 static int
15445 dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv)
15446 {
15447 	int rval;
15448 	dof_helper_t help, *dhp = NULL;
15449 
15450 	switch (cmd) {
15451 	case DTRACEHIOC_ADDDOF:
15452 		if (copyin((void *)arg, &help, sizeof (help)) != 0) {
15453 			dtrace_dof_error(NULL, "failed to copyin DOF helper");
15454 			return (EFAULT);
15455 		}
15456 
15457 		dhp = &help;
15458 		arg = (intptr_t)help.dofhp_dof;
15459 		/*FALLTHROUGH*/
15460 
15461 	case DTRACEHIOC_ADD: {
15462 		dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval);
15463 
15464 		if (dof == NULL)
15465 			return (rval);
15466 
15467 		mutex_enter(&dtrace_lock);
15468 
15469 		/*
15470 		 * dtrace_helper_slurp() takes responsibility for the dof --
15471 		 * it may free it now or it may save it and free it later.
15472 		 */
15473 		if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) {
15474 			*rv = rval;
15475 			rval = 0;
15476 		} else {
15477 			rval = EINVAL;
15478 		}
15479 
15480 		mutex_exit(&dtrace_lock);
15481 		return (rval);
15482 	}
15483 
15484 	case DTRACEHIOC_REMOVE: {
15485 		mutex_enter(&dtrace_lock);
15486 		rval = dtrace_helper_destroygen(arg);
15487 		mutex_exit(&dtrace_lock);
15488 
15489 		return (rval);
15490 	}
15491 
15492 	default:
15493 		break;
15494 	}
15495 
15496 	return (ENOTTY);
15497 }
15498 
15499 /*ARGSUSED*/
15500 static int
15501 dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
15502 {
15503 	minor_t minor = getminor(dev);
15504 	dtrace_state_t *state;
15505 	int rval;
15506 
15507 	if (minor == DTRACEMNRN_HELPER)
15508 		return (dtrace_ioctl_helper(cmd, arg, rv));
15509 
15510 	state = ddi_get_soft_state(dtrace_softstate, minor);
15511 
15512 	if (state->dts_anon) {
15513 		ASSERT(dtrace_anon.dta_state == NULL);
15514 		state = state->dts_anon;
15515 	}
15516 
15517 	switch (cmd) {
15518 	case DTRACEIOC_PROVIDER: {
15519 		dtrace_providerdesc_t pvd;
15520 		dtrace_provider_t *pvp;
15521 
15522 		if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0)
15523 			return (EFAULT);
15524 
15525 		pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0';
15526 		mutex_enter(&dtrace_provider_lock);
15527 
15528 		for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) {
15529 			if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0)
15530 				break;
15531 		}
15532 
15533 		mutex_exit(&dtrace_provider_lock);
15534 
15535 		if (pvp == NULL)
15536 			return (ESRCH);
15537 
15538 		bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t));
15539 		bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t));
15540 
15541 		if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0)
15542 			return (EFAULT);
15543 
15544 		return (0);
15545 	}
15546 
15547 	case DTRACEIOC_EPROBE: {
15548 		dtrace_eprobedesc_t epdesc;
15549 		dtrace_ecb_t *ecb;
15550 		dtrace_action_t *act;
15551 		void *buf;
15552 		size_t size;
15553 		uintptr_t dest;
15554 		int nrecs;
15555 
15556 		if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0)
15557 			return (EFAULT);
15558 
15559 		mutex_enter(&dtrace_lock);
15560 
15561 		if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) {
15562 			mutex_exit(&dtrace_lock);
15563 			return (EINVAL);
15564 		}
15565 
15566 		if (ecb->dte_probe == NULL) {
15567 			mutex_exit(&dtrace_lock);
15568 			return (EINVAL);
15569 		}
15570 
15571 		epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id;
15572 		epdesc.dtepd_uarg = ecb->dte_uarg;
15573 		epdesc.dtepd_size = ecb->dte_size;
15574 
15575 		nrecs = epdesc.dtepd_nrecs;
15576 		epdesc.dtepd_nrecs = 0;
15577 		for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
15578 			if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
15579 				continue;
15580 
15581 			epdesc.dtepd_nrecs++;
15582 		}
15583 
15584 		/*
15585 		 * Now that we have the size, we need to allocate a temporary
15586 		 * buffer in which to store the complete description.  We need
15587 		 * the temporary buffer to be able to drop dtrace_lock()
15588 		 * across the copyout(), below.
15589 		 */
15590 		size = sizeof (dtrace_eprobedesc_t) +
15591 		    (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t));
15592 
15593 		buf = kmem_alloc(size, KM_SLEEP);
15594 		dest = (uintptr_t)buf;
15595 
15596 		bcopy(&epdesc, (void *)dest, sizeof (epdesc));
15597 		dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]);
15598 
15599 		for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
15600 			if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
15601 				continue;
15602 
15603 			if (nrecs-- == 0)
15604 				break;
15605 
15606 			bcopy(&act->dta_rec, (void *)dest,
15607 			    sizeof (dtrace_recdesc_t));
15608 			dest += sizeof (dtrace_recdesc_t);
15609 		}
15610 
15611 		mutex_exit(&dtrace_lock);
15612 
15613 		if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
15614 			kmem_free(buf, size);
15615 			return (EFAULT);
15616 		}
15617 
15618 		kmem_free(buf, size);
15619 		return (0);
15620 	}
15621 
15622 	case DTRACEIOC_AGGDESC: {
15623 		dtrace_aggdesc_t aggdesc;
15624 		dtrace_action_t *act;
15625 		dtrace_aggregation_t *agg;
15626 		int nrecs;
15627 		uint32_t offs;
15628 		dtrace_recdesc_t *lrec;
15629 		void *buf;
15630 		size_t size;
15631 		uintptr_t dest;
15632 
15633 		if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0)
15634 			return (EFAULT);
15635 
15636 		mutex_enter(&dtrace_lock);
15637 
15638 		if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) {
15639 			mutex_exit(&dtrace_lock);
15640 			return (EINVAL);
15641 		}
15642 
15643 		aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid;
15644 
15645 		nrecs = aggdesc.dtagd_nrecs;
15646 		aggdesc.dtagd_nrecs = 0;
15647 
15648 		offs = agg->dtag_base;
15649 		lrec = &agg->dtag_action.dta_rec;
15650 		aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs;
15651 
15652 		for (act = agg->dtag_first; ; act = act->dta_next) {
15653 			ASSERT(act->dta_intuple ||
15654 			    DTRACEACT_ISAGG(act->dta_kind));
15655 
15656 			/*
15657 			 * If this action has a record size of zero, it
15658 			 * denotes an argument to the aggregating action.
15659 			 * Because the presence of this record doesn't (or
15660 			 * shouldn't) affect the way the data is interpreted,
15661 			 * we don't copy it out to save user-level the
15662 			 * confusion of dealing with a zero-length record.
15663 			 */
15664 			if (act->dta_rec.dtrd_size == 0) {
15665 				ASSERT(agg->dtag_hasarg);
15666 				continue;
15667 			}
15668 
15669 			aggdesc.dtagd_nrecs++;
15670 
15671 			if (act == &agg->dtag_action)
15672 				break;
15673 		}
15674 
15675 		/*
15676 		 * Now that we have the size, we need to allocate a temporary
15677 		 * buffer in which to store the complete description.  We need
15678 		 * the temporary buffer to be able to drop dtrace_lock()
15679 		 * across the copyout(), below.
15680 		 */
15681 		size = sizeof (dtrace_aggdesc_t) +
15682 		    (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t));
15683 
15684 		buf = kmem_alloc(size, KM_SLEEP);
15685 		dest = (uintptr_t)buf;
15686 
15687 		bcopy(&aggdesc, (void *)dest, sizeof (aggdesc));
15688 		dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]);
15689 
15690 		for (act = agg->dtag_first; ; act = act->dta_next) {
15691 			dtrace_recdesc_t rec = act->dta_rec;
15692 
15693 			/*
15694 			 * See the comment in the above loop for why we pass
15695 			 * over zero-length records.
15696 			 */
15697 			if (rec.dtrd_size == 0) {
15698 				ASSERT(agg->dtag_hasarg);
15699 				continue;
15700 			}
15701 
15702 			if (nrecs-- == 0)
15703 				break;
15704 
15705 			rec.dtrd_offset -= offs;
15706 			bcopy(&rec, (void *)dest, sizeof (rec));
15707 			dest += sizeof (dtrace_recdesc_t);
15708 
15709 			if (act == &agg->dtag_action)
15710 				break;
15711 		}
15712 
15713 		mutex_exit(&dtrace_lock);
15714 
15715 		if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
15716 			kmem_free(buf, size);
15717 			return (EFAULT);
15718 		}
15719 
15720 		kmem_free(buf, size);
15721 		return (0);
15722 	}
15723 
15724 	case DTRACEIOC_ENABLE: {
15725 		dof_hdr_t *dof;
15726 		dtrace_enabling_t *enab = NULL;
15727 		dtrace_vstate_t *vstate;
15728 		int err = 0;
15729 
15730 		*rv = 0;
15731 
15732 		/*
15733 		 * If a NULL argument has been passed, we take this as our
15734 		 * cue to reevaluate our enablings.
15735 		 */
15736 		if (arg == NULL) {
15737 			dtrace_enabling_matchall();
15738 
15739 			return (0);
15740 		}
15741 
15742 		if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL)
15743 			return (rval);
15744 
15745 		mutex_enter(&cpu_lock);
15746 		mutex_enter(&dtrace_lock);
15747 		vstate = &state->dts_vstate;
15748 
15749 		if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
15750 			mutex_exit(&dtrace_lock);
15751 			mutex_exit(&cpu_lock);
15752 			dtrace_dof_destroy(dof);
15753 			return (EBUSY);
15754 		}
15755 
15756 		if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) {
15757 			mutex_exit(&dtrace_lock);
15758 			mutex_exit(&cpu_lock);
15759 			dtrace_dof_destroy(dof);
15760 			return (EINVAL);
15761 		}
15762 
15763 		if ((rval = dtrace_dof_options(dof, state)) != 0) {
15764 			dtrace_enabling_destroy(enab);
15765 			mutex_exit(&dtrace_lock);
15766 			mutex_exit(&cpu_lock);
15767 			dtrace_dof_destroy(dof);
15768 			return (rval);
15769 		}
15770 
15771 		if ((err = dtrace_enabling_match(enab, rv)) == 0) {
15772 			err = dtrace_enabling_retain(enab);
15773 		} else {
15774 			dtrace_enabling_destroy(enab);
15775 		}
15776 
15777 		mutex_exit(&cpu_lock);
15778 		mutex_exit(&dtrace_lock);
15779 		dtrace_dof_destroy(dof);
15780 
15781 		return (err);
15782 	}
15783 
15784 	case DTRACEIOC_REPLICATE: {
15785 		dtrace_repldesc_t desc;
15786 		dtrace_probedesc_t *match = &desc.dtrpd_match;
15787 		dtrace_probedesc_t *create = &desc.dtrpd_create;
15788 		int err;
15789 
15790 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15791 			return (EFAULT);
15792 
15793 		match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
15794 		match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
15795 		match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
15796 		match->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
15797 
15798 		create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
15799 		create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
15800 		create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
15801 		create->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
15802 
15803 		mutex_enter(&dtrace_lock);
15804 		err = dtrace_enabling_replicate(state, match, create);
15805 		mutex_exit(&dtrace_lock);
15806 
15807 		return (err);
15808 	}
15809 
15810 	case DTRACEIOC_PROBEMATCH:
15811 	case DTRACEIOC_PROBES: {
15812 		dtrace_probe_t *probe = NULL;
15813 		dtrace_probedesc_t desc;
15814 		dtrace_probekey_t pkey;
15815 		dtrace_id_t i;
15816 		int m = 0;
15817 		uint32_t priv;
15818 		uid_t uid;
15819 		zoneid_t zoneid;
15820 
15821 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15822 			return (EFAULT);
15823 
15824 		desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
15825 		desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
15826 		desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
15827 		desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0';
15828 
15829 		/*
15830 		 * Before we attempt to match this probe, we want to give
15831 		 * all providers the opportunity to provide it.
15832 		 */
15833 		if (desc.dtpd_id == DTRACE_IDNONE) {
15834 			mutex_enter(&dtrace_provider_lock);
15835 			dtrace_probe_provide(&desc, NULL);
15836 			mutex_exit(&dtrace_provider_lock);
15837 			desc.dtpd_id++;
15838 		}
15839 
15840 		if (cmd == DTRACEIOC_PROBEMATCH)  {
15841 			dtrace_probekey(&desc, &pkey);
15842 			pkey.dtpk_id = DTRACE_IDNONE;
15843 		}
15844 
15845 		dtrace_cred2priv(cr, &priv, &uid, &zoneid);
15846 
15847 		mutex_enter(&dtrace_lock);
15848 
15849 		if (cmd == DTRACEIOC_PROBEMATCH) {
15850 			for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
15851 				if ((probe = dtrace_probes[i - 1]) != NULL &&
15852 				    (m = dtrace_match_probe(probe, &pkey,
15853 				    priv, uid, zoneid)) != 0)
15854 					break;
15855 			}
15856 
15857 			if (m < 0) {
15858 				mutex_exit(&dtrace_lock);
15859 				return (EINVAL);
15860 			}
15861 
15862 		} else {
15863 			for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
15864 				if ((probe = dtrace_probes[i - 1]) != NULL &&
15865 				    dtrace_match_priv(probe, priv, uid, zoneid))
15866 					break;
15867 			}
15868 		}
15869 
15870 		if (probe == NULL) {
15871 			mutex_exit(&dtrace_lock);
15872 			return (ESRCH);
15873 		}
15874 
15875 		dtrace_probe_description(probe, &desc);
15876 		mutex_exit(&dtrace_lock);
15877 
15878 		if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
15879 			return (EFAULT);
15880 
15881 		return (0);
15882 	}
15883 
15884 	case DTRACEIOC_PROBEARG: {
15885 		dtrace_argdesc_t desc;
15886 		dtrace_probe_t *probe;
15887 		dtrace_provider_t *prov;
15888 
15889 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15890 			return (EFAULT);
15891 
15892 		if (desc.dtargd_id == DTRACE_IDNONE)
15893 			return (EINVAL);
15894 
15895 		if (desc.dtargd_ndx == DTRACE_ARGNONE)
15896 			return (EINVAL);
15897 
15898 		mutex_enter(&dtrace_provider_lock);
15899 		mutex_enter(&mod_lock);
15900 		mutex_enter(&dtrace_lock);
15901 
15902 		if (desc.dtargd_id > dtrace_nprobes) {
15903 			mutex_exit(&dtrace_lock);
15904 			mutex_exit(&mod_lock);
15905 			mutex_exit(&dtrace_provider_lock);
15906 			return (EINVAL);
15907 		}
15908 
15909 		if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) {
15910 			mutex_exit(&dtrace_lock);
15911 			mutex_exit(&mod_lock);
15912 			mutex_exit(&dtrace_provider_lock);
15913 			return (EINVAL);
15914 		}
15915 
15916 		mutex_exit(&dtrace_lock);
15917 
15918 		prov = probe->dtpr_provider;
15919 
15920 		if (prov->dtpv_pops.dtps_getargdesc == NULL) {
15921 			/*
15922 			 * There isn't any typed information for this probe.
15923 			 * Set the argument number to DTRACE_ARGNONE.
15924 			 */
15925 			desc.dtargd_ndx = DTRACE_ARGNONE;
15926 		} else {
15927 			desc.dtargd_native[0] = '\0';
15928 			desc.dtargd_xlate[0] = '\0';
15929 			desc.dtargd_mapping = desc.dtargd_ndx;
15930 
15931 			prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg,
15932 			    probe->dtpr_id, probe->dtpr_arg, &desc);
15933 		}
15934 
15935 		mutex_exit(&mod_lock);
15936 		mutex_exit(&dtrace_provider_lock);
15937 
15938 		if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
15939 			return (EFAULT);
15940 
15941 		return (0);
15942 	}
15943 
15944 	case DTRACEIOC_GO: {
15945 		processorid_t cpuid;
15946 		rval = dtrace_state_go(state, &cpuid);
15947 
15948 		if (rval != 0)
15949 			return (rval);
15950 
15951 		if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
15952 			return (EFAULT);
15953 
15954 		return (0);
15955 	}
15956 
15957 	case DTRACEIOC_STOP: {
15958 		processorid_t cpuid;
15959 
15960 		mutex_enter(&dtrace_lock);
15961 		rval = dtrace_state_stop(state, &cpuid);
15962 		mutex_exit(&dtrace_lock);
15963 
15964 		if (rval != 0)
15965 			return (rval);
15966 
15967 		if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
15968 			return (EFAULT);
15969 
15970 		return (0);
15971 	}
15972 
15973 	case DTRACEIOC_DOFGET: {
15974 		dof_hdr_t hdr, *dof;
15975 		uint64_t len;
15976 
15977 		if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0)
15978 			return (EFAULT);
15979 
15980 		mutex_enter(&dtrace_lock);
15981 		dof = dtrace_dof_create(state);
15982 		mutex_exit(&dtrace_lock);
15983 
15984 		len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz);
15985 		rval = copyout(dof, (void *)arg, len);
15986 		dtrace_dof_destroy(dof);
15987 
15988 		return (rval == 0 ? 0 : EFAULT);
15989 	}
15990 
15991 	case DTRACEIOC_AGGSNAP:
15992 	case DTRACEIOC_BUFSNAP: {
15993 		dtrace_bufdesc_t desc;
15994 		caddr_t cached;
15995 		dtrace_buffer_t *buf;
15996 
15997 		if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15998 			return (EFAULT);
15999 
16000 		if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU)
16001 			return (EINVAL);
16002 
16003 		mutex_enter(&dtrace_lock);
16004 
16005 		if (cmd == DTRACEIOC_BUFSNAP) {
16006 			buf = &state->dts_buffer[desc.dtbd_cpu];
16007 		} else {
16008 			buf = &state->dts_aggbuffer[desc.dtbd_cpu];
16009 		}
16010 
16011 		if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) {
16012 			size_t sz = buf->dtb_offset;
16013 
16014 			if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) {
16015 				mutex_exit(&dtrace_lock);
16016 				return (EBUSY);
16017 			}
16018 
16019 			/*
16020 			 * If this buffer has already been consumed, we're
16021 			 * going to indicate that there's nothing left here
16022 			 * to consume.
16023 			 */
16024 			if (buf->dtb_flags & DTRACEBUF_CONSUMED) {
16025 				mutex_exit(&dtrace_lock);
16026 
16027 				desc.dtbd_size = 0;
16028 				desc.dtbd_drops = 0;
16029 				desc.dtbd_errors = 0;
16030 				desc.dtbd_oldest = 0;
16031 				sz = sizeof (desc);
16032 
16033 				if (copyout(&desc, (void *)arg, sz) != 0)
16034 					return (EFAULT);
16035 
16036 				return (0);
16037 			}
16038 
16039 			/*
16040 			 * If this is a ring buffer that has wrapped, we want
16041 			 * to copy the whole thing out.
16042 			 */
16043 			if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
16044 				dtrace_buffer_polish(buf);
16045 				sz = buf->dtb_size;
16046 			}
16047 
16048 			if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) {
16049 				mutex_exit(&dtrace_lock);
16050 				return (EFAULT);
16051 			}
16052 
16053 			desc.dtbd_size = sz;
16054 			desc.dtbd_drops = buf->dtb_drops;
16055 			desc.dtbd_errors = buf->dtb_errors;
16056 			desc.dtbd_oldest = buf->dtb_xamot_offset;
16057 
16058 			mutex_exit(&dtrace_lock);
16059 
16060 			if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16061 				return (EFAULT);
16062 
16063 			buf->dtb_flags |= DTRACEBUF_CONSUMED;
16064 
16065 			return (0);
16066 		}
16067 
16068 		if (buf->dtb_tomax == NULL) {
16069 			ASSERT(buf->dtb_xamot == NULL);
16070 			mutex_exit(&dtrace_lock);
16071 			return (ENOENT);
16072 		}
16073 
16074 		cached = buf->dtb_tomax;
16075 		ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
16076 
16077 		dtrace_xcall(desc.dtbd_cpu,
16078 		    (dtrace_xcall_t)dtrace_buffer_switch, buf);
16079 
16080 		state->dts_errors += buf->dtb_xamot_errors;
16081 
16082 		/*
16083 		 * If the buffers did not actually switch, then the cross call
16084 		 * did not take place -- presumably because the given CPU is
16085 		 * not in the ready set.  If this is the case, we'll return
16086 		 * ENOENT.
16087 		 */
16088 		if (buf->dtb_tomax == cached) {
16089 			ASSERT(buf->dtb_xamot != cached);
16090 			mutex_exit(&dtrace_lock);
16091 			return (ENOENT);
16092 		}
16093 
16094 		ASSERT(cached == buf->dtb_xamot);
16095 
16096 		/*
16097 		 * We have our snapshot; now copy it out.
16098 		 */
16099 		if (copyout(buf->dtb_xamot, desc.dtbd_data,
16100 		    buf->dtb_xamot_offset) != 0) {
16101 			mutex_exit(&dtrace_lock);
16102 			return (EFAULT);
16103 		}
16104 
16105 		desc.dtbd_size = buf->dtb_xamot_offset;
16106 		desc.dtbd_drops = buf->dtb_xamot_drops;
16107 		desc.dtbd_errors = buf->dtb_xamot_errors;
16108 		desc.dtbd_oldest = 0;
16109 
16110 		mutex_exit(&dtrace_lock);
16111 
16112 		/*
16113 		 * Finally, copy out the buffer description.
16114 		 */
16115 		if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16116 			return (EFAULT);
16117 
16118 		return (0);
16119 	}
16120 
16121 	case DTRACEIOC_CONF: {
16122 		dtrace_conf_t conf;
16123 
16124 		bzero(&conf, sizeof (conf));
16125 		conf.dtc_difversion = DIF_VERSION;
16126 		conf.dtc_difintregs = DIF_DIR_NREGS;
16127 		conf.dtc_diftupregs = DIF_DTR_NREGS;
16128 		conf.dtc_ctfmodel = CTF_MODEL_NATIVE;
16129 
16130 		if (copyout(&conf, (void *)arg, sizeof (conf)) != 0)
16131 			return (EFAULT);
16132 
16133 		return (0);
16134 	}
16135 
16136 	case DTRACEIOC_STATUS: {
16137 		dtrace_status_t stat;
16138 		dtrace_dstate_t *dstate;
16139 		int i, j;
16140 		uint64_t nerrs;
16141 
16142 		/*
16143 		 * See the comment in dtrace_state_deadman() for the reason
16144 		 * for setting dts_laststatus to INT64_MAX before setting
16145 		 * it to the correct value.
16146 		 */
16147 		state->dts_laststatus = INT64_MAX;
16148 		dtrace_membar_producer();
16149 		state->dts_laststatus = dtrace_gethrtime();
16150 
16151 		bzero(&stat, sizeof (stat));
16152 
16153 		mutex_enter(&dtrace_lock);
16154 
16155 		if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) {
16156 			mutex_exit(&dtrace_lock);
16157 			return (ENOENT);
16158 		}
16159 
16160 		if (state->dts_activity == DTRACE_ACTIVITY_DRAINING)
16161 			stat.dtst_exiting = 1;
16162 
16163 		nerrs = state->dts_errors;
16164 		dstate = &state->dts_vstate.dtvs_dynvars;
16165 
16166 		for (i = 0; i < NCPU; i++) {
16167 			dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i];
16168 
16169 			stat.dtst_dyndrops += dcpu->dtdsc_drops;
16170 			stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops;
16171 			stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops;
16172 
16173 			if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL)
16174 				stat.dtst_filled++;
16175 
16176 			nerrs += state->dts_buffer[i].dtb_errors;
16177 
16178 			for (j = 0; j < state->dts_nspeculations; j++) {
16179 				dtrace_speculation_t *spec;
16180 				dtrace_buffer_t *buf;
16181 
16182 				spec = &state->dts_speculations[j];
16183 				buf = &spec->dtsp_buffer[i];
16184 				stat.dtst_specdrops += buf->dtb_xamot_drops;
16185 			}
16186 		}
16187 
16188 		stat.dtst_specdrops_busy = state->dts_speculations_busy;
16189 		stat.dtst_specdrops_unavail = state->dts_speculations_unavail;
16190 		stat.dtst_stkstroverflows = state->dts_stkstroverflows;
16191 		stat.dtst_dblerrors = state->dts_dblerrors;
16192 		stat.dtst_killed =
16193 		    (state->dts_activity == DTRACE_ACTIVITY_KILLED);
16194 		stat.dtst_errors = nerrs;
16195 
16196 		mutex_exit(&dtrace_lock);
16197 
16198 		if (copyout(&stat, (void *)arg, sizeof (stat)) != 0)
16199 			return (EFAULT);
16200 
16201 		return (0);
16202 	}
16203 
16204 	case DTRACEIOC_FORMAT: {
16205 		dtrace_fmtdesc_t fmt;
16206 		char *str;
16207 		int len;
16208 
16209 		if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0)
16210 			return (EFAULT);
16211 
16212 		mutex_enter(&dtrace_lock);
16213 
16214 		if (fmt.dtfd_format == 0 ||
16215 		    fmt.dtfd_format > state->dts_nformats) {
16216 			mutex_exit(&dtrace_lock);
16217 			return (EINVAL);
16218 		}
16219 
16220 		/*
16221 		 * Format strings are allocated contiguously and they are
16222 		 * never freed; if a format index is less than the number
16223 		 * of formats, we can assert that the format map is non-NULL
16224 		 * and that the format for the specified index is non-NULL.
16225 		 */
16226 		ASSERT(state->dts_formats != NULL);
16227 		str = state->dts_formats[fmt.dtfd_format - 1];
16228 		ASSERT(str != NULL);
16229 
16230 		len = strlen(str) + 1;
16231 
16232 		if (len > fmt.dtfd_length) {
16233 			fmt.dtfd_length = len;
16234 
16235 			if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) {
16236 				mutex_exit(&dtrace_lock);
16237 				return (EINVAL);
16238 			}
16239 		} else {
16240 			if (copyout(str, fmt.dtfd_string, len) != 0) {
16241 				mutex_exit(&dtrace_lock);
16242 				return (EINVAL);
16243 			}
16244 		}
16245 
16246 		mutex_exit(&dtrace_lock);
16247 		return (0);
16248 	}
16249 
16250 	default:
16251 		break;
16252 	}
16253 
16254 	return (ENOTTY);
16255 }
16256 
16257 /*ARGSUSED*/
16258 static int
16259 dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
16260 {
16261 	dtrace_state_t *state;
16262 
16263 	switch (cmd) {
16264 	case DDI_DETACH:
16265 		break;
16266 
16267 	case DDI_SUSPEND:
16268 		return (DDI_SUCCESS);
16269 
16270 	default:
16271 		return (DDI_FAILURE);
16272 	}
16273 
16274 	mutex_enter(&cpu_lock);
16275 	mutex_enter(&dtrace_provider_lock);
16276 	mutex_enter(&dtrace_lock);
16277 
16278 	ASSERT(dtrace_opens == 0);
16279 
16280 	if (dtrace_helpers > 0) {
16281 		mutex_exit(&dtrace_provider_lock);
16282 		mutex_exit(&dtrace_lock);
16283 		mutex_exit(&cpu_lock);
16284 		return (DDI_FAILURE);
16285 	}
16286 
16287 	if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) {
16288 		mutex_exit(&dtrace_provider_lock);
16289 		mutex_exit(&dtrace_lock);
16290 		mutex_exit(&cpu_lock);
16291 		return (DDI_FAILURE);
16292 	}
16293 
16294 	dtrace_provider = NULL;
16295 
16296 	if ((state = dtrace_anon_grab()) != NULL) {
16297 		/*
16298 		 * If there were ECBs on this state, the provider should
16299 		 * have not been allowed to detach; assert that there is
16300 		 * none.
16301 		 */
16302 		ASSERT(state->dts_necbs == 0);
16303 		dtrace_state_destroy(state);
16304 
16305 		/*
16306 		 * If we're being detached with anonymous state, we need to
16307 		 * indicate to the kernel debugger that DTrace is now inactive.
16308 		 */
16309 		(void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
16310 	}
16311 
16312 	bzero(&dtrace_anon, sizeof (dtrace_anon_t));
16313 	unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
16314 	dtrace_cpu_init = NULL;
16315 	dtrace_helpers_cleanup = NULL;
16316 	dtrace_helpers_fork = NULL;
16317 	dtrace_cpustart_init = NULL;
16318 	dtrace_cpustart_fini = NULL;
16319 	dtrace_debugger_init = NULL;
16320 	dtrace_debugger_fini = NULL;
16321 	dtrace_modload = NULL;
16322 	dtrace_modunload = NULL;
16323 
16324 	mutex_exit(&cpu_lock);
16325 
16326 	if (dtrace_helptrace_enabled) {
16327 		kmem_free(dtrace_helptrace_buffer, dtrace_helptrace_bufsize);
16328 		dtrace_helptrace_buffer = NULL;
16329 	}
16330 
16331 	kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *));
16332 	dtrace_probes = NULL;
16333 	dtrace_nprobes = 0;
16334 
16335 	dtrace_hash_destroy(dtrace_bymod);
16336 	dtrace_hash_destroy(dtrace_byfunc);
16337 	dtrace_hash_destroy(dtrace_byname);
16338 	dtrace_bymod = NULL;
16339 	dtrace_byfunc = NULL;
16340 	dtrace_byname = NULL;
16341 
16342 	kmem_cache_destroy(dtrace_state_cache);
16343 	vmem_destroy(dtrace_minor);
16344 	vmem_destroy(dtrace_arena);
16345 
16346 	if (dtrace_toxrange != NULL) {
16347 		kmem_free(dtrace_toxrange,
16348 		    dtrace_toxranges_max * sizeof (dtrace_toxrange_t));
16349 		dtrace_toxrange = NULL;
16350 		dtrace_toxranges = 0;
16351 		dtrace_toxranges_max = 0;
16352 	}
16353 
16354 	ddi_remove_minor_node(dtrace_devi, NULL);
16355 	dtrace_devi = NULL;
16356 
16357 	ddi_soft_state_fini(&dtrace_softstate);
16358 
16359 	ASSERT(dtrace_vtime_references == 0);
16360 	ASSERT(dtrace_opens == 0);
16361 	ASSERT(dtrace_retained == NULL);
16362 
16363 	mutex_exit(&dtrace_lock);
16364 	mutex_exit(&dtrace_provider_lock);
16365 
16366 	/*
16367 	 * We don't destroy the task queue until after we have dropped our
16368 	 * locks (taskq_destroy() may block on running tasks).  To prevent
16369 	 * attempting to do work after we have effectively detached but before
16370 	 * the task queue has been destroyed, all tasks dispatched via the
16371 	 * task queue must check that DTrace is still attached before
16372 	 * performing any operation.
16373 	 */
16374 	taskq_destroy(dtrace_taskq);
16375 	dtrace_taskq = NULL;
16376 
16377 	return (DDI_SUCCESS);
16378 }
16379 #endif
16380 
16381 #if defined(sun)
16382 /*ARGSUSED*/
16383 static int
16384 dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
16385 {
16386 	int error;
16387 
16388 	switch (infocmd) {
16389 	case DDI_INFO_DEVT2DEVINFO:
16390 		*result = (void *)dtrace_devi;
16391 		error = DDI_SUCCESS;
16392 		break;
16393 	case DDI_INFO_DEVT2INSTANCE:
16394 		*result = (void *)0;
16395 		error = DDI_SUCCESS;
16396 		break;
16397 	default:
16398 		error = DDI_FAILURE;
16399 	}
16400 	return (error);
16401 }
16402 #endif
16403 
16404 #if defined(sun)
16405 static struct cb_ops dtrace_cb_ops = {
16406 	dtrace_open,		/* open */
16407 	dtrace_close,		/* close */
16408 	nulldev,		/* strategy */
16409 	nulldev,		/* print */
16410 	nodev,			/* dump */
16411 	nodev,			/* read */
16412 	nodev,			/* write */
16413 	dtrace_ioctl,		/* ioctl */
16414 	nodev,			/* devmap */
16415 	nodev,			/* mmap */
16416 	nodev,			/* segmap */
16417 	nochpoll,		/* poll */
16418 	ddi_prop_op,		/* cb_prop_op */
16419 	0,			/* streamtab  */
16420 	D_NEW | D_MP		/* Driver compatibility flag */
16421 };
16422 
16423 static struct dev_ops dtrace_ops = {
16424 	DEVO_REV,		/* devo_rev */
16425 	0,			/* refcnt */
16426 	dtrace_info,		/* get_dev_info */
16427 	nulldev,		/* identify */
16428 	nulldev,		/* probe */
16429 	dtrace_attach,		/* attach */
16430 	dtrace_detach,		/* detach */
16431 	nodev,			/* reset */
16432 	&dtrace_cb_ops,		/* driver operations */
16433 	NULL,			/* bus operations */
16434 	nodev			/* dev power */
16435 };
16436 
16437 static struct modldrv modldrv = {
16438 	&mod_driverops,		/* module type (this is a pseudo driver) */
16439 	"Dynamic Tracing",	/* name of module */
16440 	&dtrace_ops,		/* driver ops */
16441 };
16442 
16443 static struct modlinkage modlinkage = {
16444 	MODREV_1,
16445 	(void *)&modldrv,
16446 	NULL
16447 };
16448 
16449 int
16450 _init(void)
16451 {
16452 	return (mod_install(&modlinkage));
16453 }
16454 
16455 int
16456 _info(struct modinfo *modinfop)
16457 {
16458 	return (mod_info(&modlinkage, modinfop));
16459 }
16460 
16461 int
16462 _fini(void)
16463 {
16464 	return (mod_remove(&modlinkage));
16465 }
16466 #else
16467 
16468 static d_ioctl_t	dtrace_ioctl;
16469 static d_ioctl_t	dtrace_ioctl_helper;
16470 static void		dtrace_load(void *);
16471 static int		dtrace_unload(void);
16472 #if __FreeBSD_version < 800039
16473 static void		dtrace_clone(void *, struct ucred *, char *, int , struct cdev **);
16474 static struct clonedevs	*dtrace_clones;		/* Ptr to the array of cloned devices. */
16475 static eventhandler_tag	eh_tag;			/* Event handler tag. */
16476 #else
16477 static struct cdev	*dtrace_dev;
16478 static struct cdev	*helper_dev;
16479 #endif
16480 
16481 void dtrace_invop_init(void);
16482 void dtrace_invop_uninit(void);
16483 
16484 static struct cdevsw dtrace_cdevsw = {
16485 	.d_version	= D_VERSION,
16486 	.d_flags	= D_TRACKCLOSE | D_NEEDMINOR,
16487 	.d_close	= dtrace_close,
16488 	.d_ioctl	= dtrace_ioctl,
16489 	.d_open		= dtrace_open,
16490 	.d_name		= "dtrace",
16491 };
16492 
16493 static struct cdevsw helper_cdevsw = {
16494 	.d_version	= D_VERSION,
16495 	.d_flags	= D_TRACKCLOSE | D_NEEDMINOR,
16496 	.d_ioctl	= dtrace_ioctl_helper,
16497 	.d_name		= "helper",
16498 };
16499 
16500 #include <dtrace_anon.c>
16501 #if __FreeBSD_version < 800039
16502 #include <dtrace_clone.c>
16503 #endif
16504 #include <dtrace_ioctl.c>
16505 #include <dtrace_load.c>
16506 #include <dtrace_modevent.c>
16507 #include <dtrace_sysctl.c>
16508 #include <dtrace_unload.c>
16509 #include <dtrace_vtime.c>
16510 #include <dtrace_hacks.c>
16511 #include <dtrace_isa.c>
16512 
16513 SYSINIT(dtrace_load, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_load, NULL);
16514 SYSUNINIT(dtrace_unload, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_unload, NULL);
16515 SYSINIT(dtrace_anon_init, SI_SUB_DTRACE_ANON, SI_ORDER_FIRST, dtrace_anon_init, NULL);
16516 
16517 DEV_MODULE(dtrace, dtrace_modevent, NULL);
16518 MODULE_VERSION(dtrace, 1);
16519 MODULE_DEPEND(dtrace, cyclic, 1, 1, 1);
16520 MODULE_DEPEND(dtrace, opensolaris, 1, 1, 1);
16521 #endif
16522