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 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 27 * Copyright (c) 2012 by Delphix. All rights reserved. 28 */ 29 30 #pragma ident "%Z%%M% %I% %E% SMI" 31 32 /* 33 * DTrace - Dynamic Tracing for Solaris 34 * 35 * This is the implementation of the Solaris Dynamic Tracing framework 36 * (DTrace). The user-visible interface to DTrace is described at length in 37 * the "Solaris Dynamic Tracing Guide". The interfaces between the libdtrace 38 * library, the in-kernel DTrace framework, and the DTrace providers are 39 * described in the block comments in the <sys/dtrace.h> header file. The 40 * internal architecture of DTrace is described in the block comments in the 41 * <sys/dtrace_impl.h> header file. The comments contained within the DTrace 42 * implementation very much assume mastery of all of these sources; if one has 43 * an unanswered question about the implementation, one should consult them 44 * first. 45 * 46 * The functions here are ordered roughly as follows: 47 * 48 * - Probe context functions 49 * - Probe hashing functions 50 * - Non-probe context utility functions 51 * - Matching functions 52 * - Provider-to-Framework API functions 53 * - Probe management functions 54 * - DIF object functions 55 * - Format functions 56 * - Predicate functions 57 * - ECB functions 58 * - Buffer functions 59 * - Enabling functions 60 * - DOF functions 61 * - Anonymous enabling functions 62 * - Consumer state functions 63 * - Helper functions 64 * - Hook functions 65 * - Driver cookbook functions 66 * 67 * Each group of functions begins with a block comment labelled the "DTrace 68 * [Group] Functions", allowing one to find each block by searching forward 69 * on capital-f functions. 70 */ 71 #include <sys/errno.h> 72 #if !defined(sun) 73 #include <sys/time.h> 74 #endif 75 #include <sys/stat.h> 76 #include <sys/modctl.h> 77 #include <sys/conf.h> 78 #include <sys/systm.h> 79 #if defined(sun) 80 #include <sys/ddi.h> 81 #include <sys/sunddi.h> 82 #endif 83 #include <sys/cpuvar.h> 84 #include <sys/kmem.h> 85 #if defined(sun) 86 #include <sys/strsubr.h> 87 #endif 88 #include <sys/sysmacros.h> 89 #include <sys/dtrace_impl.h> 90 #include <sys/atomic.h> 91 #include <sys/cmn_err.h> 92 #if defined(sun) 93 #include <sys/mutex_impl.h> 94 #include <sys/rwlock_impl.h> 95 #endif 96 #include <sys/ctf_api.h> 97 #if defined(sun) 98 #include <sys/panic.h> 99 #include <sys/priv_impl.h> 100 #endif 101 #include <sys/policy.h> 102 #if defined(sun) 103 #include <sys/cred_impl.h> 104 #include <sys/procfs_isa.h> 105 #endif 106 #include <sys/taskq.h> 107 #if defined(sun) 108 #include <sys/mkdev.h> 109 #include <sys/kdi.h> 110 #endif 111 #include <sys/zone.h> 112 #include <sys/socket.h> 113 #include <netinet/in.h> 114 115 /* FreeBSD includes: */ 116 #if !defined(sun) 117 #include <sys/callout.h> 118 #include <sys/ctype.h> 119 #include <sys/eventhandler.h> 120 #include <sys/limits.h> 121 #include <sys/kdb.h> 122 #include <sys/kernel.h> 123 #include <sys/malloc.h> 124 #include <sys/sysctl.h> 125 #include <sys/lock.h> 126 #include <sys/mutex.h> 127 #include <sys/rwlock.h> 128 #include <sys/sx.h> 129 #include <sys/dtrace_bsd.h> 130 #include <netinet/in.h> 131 #include "dtrace_cddl.h" 132 #include "dtrace_debug.c" 133 #endif 134 135 /* 136 * DTrace Tunable Variables 137 * 138 * The following variables may be tuned by adding a line to /etc/system that 139 * includes both the name of the DTrace module ("dtrace") and the name of the 140 * variable. For example: 141 * 142 * set dtrace:dtrace_destructive_disallow = 1 143 * 144 * In general, the only variables that one should be tuning this way are those 145 * that affect system-wide DTrace behavior, and for which the default behavior 146 * is undesirable. Most of these variables are tunable on a per-consumer 147 * basis using DTrace options, and need not be tuned on a system-wide basis. 148 * When tuning these variables, avoid pathological values; while some attempt 149 * is made to verify the integrity of these variables, they are not considered 150 * part of the supported interface to DTrace, and they are therefore not 151 * checked comprehensively. Further, these variables should not be tuned 152 * dynamically via "mdb -kw" or other means; they should only be tuned via 153 * /etc/system. 154 */ 155 int dtrace_destructive_disallow = 0; 156 dtrace_optval_t dtrace_nonroot_maxsize = (16 * 1024 * 1024); 157 size_t dtrace_difo_maxsize = (256 * 1024); 158 dtrace_optval_t dtrace_dof_maxsize = (8 * 1024 * 1024); 159 size_t dtrace_global_maxsize = (16 * 1024); 160 size_t dtrace_actions_max = (16 * 1024); 161 size_t dtrace_retain_max = 1024; 162 dtrace_optval_t dtrace_helper_actions_max = 128; 163 dtrace_optval_t dtrace_helper_providers_max = 32; 164 dtrace_optval_t dtrace_dstate_defsize = (1 * 1024 * 1024); 165 size_t dtrace_strsize_default = 256; 166 dtrace_optval_t dtrace_cleanrate_default = 9900990; /* 101 hz */ 167 dtrace_optval_t dtrace_cleanrate_min = 200000; /* 5000 hz */ 168 dtrace_optval_t dtrace_cleanrate_max = (uint64_t)60 * NANOSEC; /* 1/minute */ 169 dtrace_optval_t dtrace_aggrate_default = NANOSEC; /* 1 hz */ 170 dtrace_optval_t dtrace_statusrate_default = NANOSEC; /* 1 hz */ 171 dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC; /* 6/minute */ 172 dtrace_optval_t dtrace_switchrate_default = NANOSEC; /* 1 hz */ 173 dtrace_optval_t dtrace_nspec_default = 1; 174 dtrace_optval_t dtrace_specsize_default = 32 * 1024; 175 dtrace_optval_t dtrace_stackframes_default = 20; 176 dtrace_optval_t dtrace_ustackframes_default = 20; 177 dtrace_optval_t dtrace_jstackframes_default = 50; 178 dtrace_optval_t dtrace_jstackstrsize_default = 512; 179 int dtrace_msgdsize_max = 128; 180 hrtime_t dtrace_chill_max = 500 * (NANOSEC / MILLISEC); /* 500 ms */ 181 hrtime_t dtrace_chill_interval = NANOSEC; /* 1000 ms */ 182 int dtrace_devdepth_max = 32; 183 int dtrace_err_verbose; 184 hrtime_t dtrace_deadman_interval = NANOSEC; 185 hrtime_t dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC; 186 hrtime_t dtrace_deadman_user = (hrtime_t)30 * NANOSEC; 187 hrtime_t dtrace_unregister_defunct_reap = (hrtime_t)60 * NANOSEC; 188 #if !defined(sun) 189 int dtrace_memstr_max = 4096; 190 #endif 191 192 /* 193 * DTrace External Variables 194 * 195 * As dtrace(7D) is a kernel module, any DTrace variables are obviously 196 * available to DTrace consumers via the backtick (`) syntax. One of these, 197 * dtrace_zero, is made deliberately so: it is provided as a source of 198 * well-known, zero-filled memory. While this variable is not documented, 199 * it is used by some translators as an implementation detail. 200 */ 201 const char dtrace_zero[256] = { 0 }; /* zero-filled memory */ 202 203 /* 204 * DTrace Internal Variables 205 */ 206 #if defined(sun) 207 static dev_info_t *dtrace_devi; /* device info */ 208 #endif 209 #if defined(sun) 210 static vmem_t *dtrace_arena; /* probe ID arena */ 211 static vmem_t *dtrace_minor; /* minor number arena */ 212 #else 213 static taskq_t *dtrace_taskq; /* task queue */ 214 static struct unrhdr *dtrace_arena; /* Probe ID number. */ 215 #endif 216 static dtrace_probe_t **dtrace_probes; /* array of all probes */ 217 static int dtrace_nprobes; /* number of probes */ 218 static dtrace_provider_t *dtrace_provider; /* provider list */ 219 static dtrace_meta_t *dtrace_meta_pid; /* user-land meta provider */ 220 static int dtrace_opens; /* number of opens */ 221 static int dtrace_helpers; /* number of helpers */ 222 #if defined(sun) 223 static void *dtrace_softstate; /* softstate pointer */ 224 #endif 225 static dtrace_hash_t *dtrace_bymod; /* probes hashed by module */ 226 static dtrace_hash_t *dtrace_byfunc; /* probes hashed by function */ 227 static dtrace_hash_t *dtrace_byname; /* probes hashed by name */ 228 static dtrace_toxrange_t *dtrace_toxrange; /* toxic range array */ 229 static int dtrace_toxranges; /* number of toxic ranges */ 230 static int dtrace_toxranges_max; /* size of toxic range array */ 231 static dtrace_anon_t dtrace_anon; /* anonymous enabling */ 232 static kmem_cache_t *dtrace_state_cache; /* cache for dynamic state */ 233 static uint64_t dtrace_vtime_references; /* number of vtimestamp refs */ 234 static kthread_t *dtrace_panicked; /* panicking thread */ 235 static dtrace_ecb_t *dtrace_ecb_create_cache; /* cached created ECB */ 236 static dtrace_genid_t dtrace_probegen; /* current probe generation */ 237 static dtrace_helpers_t *dtrace_deferred_pid; /* deferred helper list */ 238 static dtrace_enabling_t *dtrace_retained; /* list of retained enablings */ 239 static dtrace_dynvar_t dtrace_dynhash_sink; /* end of dynamic hash chains */ 240 #if !defined(sun) 241 static struct mtx dtrace_unr_mtx; 242 MTX_SYSINIT(dtrace_unr_mtx, &dtrace_unr_mtx, "Unique resource identifier", MTX_DEF); 243 int dtrace_in_probe; /* non-zero if executing a probe */ 244 #if defined(__i386__) || defined(__amd64__) || defined(__mips__) || defined(__powerpc__) 245 uintptr_t dtrace_in_probe_addr; /* Address of invop when already in probe */ 246 #endif 247 static eventhandler_tag dtrace_kld_load_tag; 248 static eventhandler_tag dtrace_kld_unload_try_tag; 249 #endif 250 251 /* 252 * DTrace Locking 253 * DTrace is protected by three (relatively coarse-grained) locks: 254 * 255 * (1) dtrace_lock is required to manipulate essentially any DTrace state, 256 * including enabling state, probes, ECBs, consumer state, helper state, 257 * etc. Importantly, dtrace_lock is _not_ required when in probe context; 258 * probe context is lock-free -- synchronization is handled via the 259 * dtrace_sync() cross call mechanism. 260 * 261 * (2) dtrace_provider_lock is required when manipulating provider state, or 262 * when provider state must be held constant. 263 * 264 * (3) dtrace_meta_lock is required when manipulating meta provider state, or 265 * when meta provider state must be held constant. 266 * 267 * The lock ordering between these three locks is dtrace_meta_lock before 268 * dtrace_provider_lock before dtrace_lock. (In particular, there are 269 * several places where dtrace_provider_lock is held by the framework as it 270 * calls into the providers -- which then call back into the framework, 271 * grabbing dtrace_lock.) 272 * 273 * There are two other locks in the mix: mod_lock and cpu_lock. With respect 274 * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical 275 * role as a coarse-grained lock; it is acquired before both of these locks. 276 * With respect to dtrace_meta_lock, its behavior is stranger: cpu_lock must 277 * be acquired _between_ dtrace_meta_lock and any other DTrace locks. 278 * mod_lock is similar with respect to dtrace_provider_lock in that it must be 279 * acquired _between_ dtrace_provider_lock and dtrace_lock. 280 */ 281 static kmutex_t dtrace_lock; /* probe state lock */ 282 static kmutex_t dtrace_provider_lock; /* provider state lock */ 283 static kmutex_t dtrace_meta_lock; /* meta-provider state lock */ 284 285 #if !defined(sun) 286 /* XXX FreeBSD hacks. */ 287 #define cr_suid cr_svuid 288 #define cr_sgid cr_svgid 289 #define ipaddr_t in_addr_t 290 #define mod_modname pathname 291 #define vuprintf vprintf 292 #define ttoproc(_a) ((_a)->td_proc) 293 #define crgetzoneid(_a) 0 294 #define NCPU MAXCPU 295 #define SNOCD 0 296 #define CPU_ON_INTR(_a) 0 297 298 #define PRIV_EFFECTIVE (1 << 0) 299 #define PRIV_DTRACE_KERNEL (1 << 1) 300 #define PRIV_DTRACE_PROC (1 << 2) 301 #define PRIV_DTRACE_USER (1 << 3) 302 #define PRIV_PROC_OWNER (1 << 4) 303 #define PRIV_PROC_ZONE (1 << 5) 304 #define PRIV_ALL ~0 305 306 SYSCTL_DECL(_debug_dtrace); 307 SYSCTL_DECL(_kern_dtrace); 308 #endif 309 310 #if defined(sun) 311 #define curcpu CPU->cpu_id 312 #endif 313 314 315 /* 316 * DTrace Provider Variables 317 * 318 * These are the variables relating to DTrace as a provider (that is, the 319 * provider of the BEGIN, END, and ERROR probes). 320 */ 321 static dtrace_pattr_t dtrace_provider_attr = { 322 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, 323 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 324 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 325 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, 326 { DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, 327 }; 328 329 static void 330 dtrace_nullop(void) 331 {} 332 333 static dtrace_pops_t dtrace_provider_ops = { 334 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop, 335 (void (*)(void *, modctl_t *))dtrace_nullop, 336 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, 337 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, 338 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, 339 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, 340 NULL, 341 NULL, 342 NULL, 343 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop 344 }; 345 346 static dtrace_id_t dtrace_probeid_begin; /* special BEGIN probe */ 347 static dtrace_id_t dtrace_probeid_end; /* special END probe */ 348 dtrace_id_t dtrace_probeid_error; /* special ERROR probe */ 349 350 /* 351 * DTrace Helper Tracing Variables 352 */ 353 uint32_t dtrace_helptrace_next = 0; 354 uint32_t dtrace_helptrace_nlocals; 355 char *dtrace_helptrace_buffer; 356 int dtrace_helptrace_bufsize = 512 * 1024; 357 358 #ifdef DEBUG 359 int dtrace_helptrace_enabled = 1; 360 #else 361 int dtrace_helptrace_enabled = 0; 362 #endif 363 364 /* 365 * DTrace Error Hashing 366 * 367 * On DEBUG kernels, DTrace will track the errors that has seen in a hash 368 * table. This is very useful for checking coverage of tests that are 369 * expected to induce DIF or DOF processing errors, and may be useful for 370 * debugging problems in the DIF code generator or in DOF generation . The 371 * error hash may be examined with the ::dtrace_errhash MDB dcmd. 372 */ 373 #ifdef DEBUG 374 static dtrace_errhash_t dtrace_errhash[DTRACE_ERRHASHSZ]; 375 static const char *dtrace_errlast; 376 static kthread_t *dtrace_errthread; 377 static kmutex_t dtrace_errlock; 378 #endif 379 380 /* 381 * DTrace Macros and Constants 382 * 383 * These are various macros that are useful in various spots in the 384 * implementation, along with a few random constants that have no meaning 385 * outside of the implementation. There is no real structure to this cpp 386 * mishmash -- but is there ever? 387 */ 388 #define DTRACE_HASHSTR(hash, probe) \ 389 dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs))) 390 391 #define DTRACE_HASHNEXT(hash, probe) \ 392 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs) 393 394 #define DTRACE_HASHPREV(hash, probe) \ 395 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs) 396 397 #define DTRACE_HASHEQ(hash, lhs, rhs) \ 398 (strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \ 399 *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0) 400 401 #define DTRACE_AGGHASHSIZE_SLEW 17 402 403 #define DTRACE_V4MAPPED_OFFSET (sizeof (uint32_t) * 3) 404 405 /* 406 * The key for a thread-local variable consists of the lower 61 bits of the 407 * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL. 408 * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never 409 * equal to a variable identifier. This is necessary (but not sufficient) to 410 * assure that global associative arrays never collide with thread-local 411 * variables. To guarantee that they cannot collide, we must also define the 412 * order for keying dynamic variables. That order is: 413 * 414 * [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ] 415 * 416 * Because the variable-key and the tls-key are in orthogonal spaces, there is 417 * no way for a global variable key signature to match a thread-local key 418 * signature. 419 */ 420 #if defined(sun) 421 #define DTRACE_TLS_THRKEY(where) { \ 422 uint_t intr = 0; \ 423 uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \ 424 for (; actv; actv >>= 1) \ 425 intr++; \ 426 ASSERT(intr < (1 << 3)); \ 427 (where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \ 428 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \ 429 } 430 #else 431 #define DTRACE_TLS_THRKEY(where) { \ 432 solaris_cpu_t *_c = &solaris_cpu[curcpu]; \ 433 uint_t intr = 0; \ 434 uint_t actv = _c->cpu_intr_actv; \ 435 for (; actv; actv >>= 1) \ 436 intr++; \ 437 ASSERT(intr < (1 << 3)); \ 438 (where) = ((curthread->td_tid + DIF_VARIABLE_MAX) & \ 439 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \ 440 } 441 #endif 442 443 #define DT_BSWAP_8(x) ((x) & 0xff) 444 #define DT_BSWAP_16(x) ((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8)) 445 #define DT_BSWAP_32(x) ((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16)) 446 #define DT_BSWAP_64(x) ((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32)) 447 448 #define DT_MASK_LO 0x00000000FFFFFFFFULL 449 450 #define DTRACE_STORE(type, tomax, offset, what) \ 451 *((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what); 452 453 #ifndef __x86 454 #define DTRACE_ALIGNCHECK(addr, size, flags) \ 455 if (addr & (size - 1)) { \ 456 *flags |= CPU_DTRACE_BADALIGN; \ 457 cpu_core[curcpu].cpuc_dtrace_illval = addr; \ 458 return (0); \ 459 } 460 #else 461 #define DTRACE_ALIGNCHECK(addr, size, flags) 462 #endif 463 464 /* 465 * Test whether a range of memory starting at testaddr of size testsz falls 466 * within the range of memory described by addr, sz. We take care to avoid 467 * problems with overflow and underflow of the unsigned quantities, and 468 * disallow all negative sizes. Ranges of size 0 are allowed. 469 */ 470 #define DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \ 471 ((testaddr) - (baseaddr) < (basesz) && \ 472 (testaddr) + (testsz) - (baseaddr) <= (basesz) && \ 473 (testaddr) + (testsz) >= (testaddr)) 474 475 /* 476 * Test whether alloc_sz bytes will fit in the scratch region. We isolate 477 * alloc_sz on the righthand side of the comparison in order to avoid overflow 478 * or underflow in the comparison with it. This is simpler than the INRANGE 479 * check above, because we know that the dtms_scratch_ptr is valid in the 480 * range. Allocations of size zero are allowed. 481 */ 482 #define DTRACE_INSCRATCH(mstate, alloc_sz) \ 483 ((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \ 484 (mstate)->dtms_scratch_ptr >= (alloc_sz)) 485 486 #define DTRACE_LOADFUNC(bits) \ 487 /*CSTYLED*/ \ 488 uint##bits##_t \ 489 dtrace_load##bits(uintptr_t addr) \ 490 { \ 491 size_t size = bits / NBBY; \ 492 /*CSTYLED*/ \ 493 uint##bits##_t rval; \ 494 int i; \ 495 volatile uint16_t *flags = (volatile uint16_t *) \ 496 &cpu_core[curcpu].cpuc_dtrace_flags; \ 497 \ 498 DTRACE_ALIGNCHECK(addr, size, flags); \ 499 \ 500 for (i = 0; i < dtrace_toxranges; i++) { \ 501 if (addr >= dtrace_toxrange[i].dtt_limit) \ 502 continue; \ 503 \ 504 if (addr + size <= dtrace_toxrange[i].dtt_base) \ 505 continue; \ 506 \ 507 /* \ 508 * This address falls within a toxic region; return 0. \ 509 */ \ 510 *flags |= CPU_DTRACE_BADADDR; \ 511 cpu_core[curcpu].cpuc_dtrace_illval = addr; \ 512 return (0); \ 513 } \ 514 \ 515 *flags |= CPU_DTRACE_NOFAULT; \ 516 /*CSTYLED*/ \ 517 rval = *((volatile uint##bits##_t *)addr); \ 518 *flags &= ~CPU_DTRACE_NOFAULT; \ 519 \ 520 return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0); \ 521 } 522 523 #ifdef _LP64 524 #define dtrace_loadptr dtrace_load64 525 #else 526 #define dtrace_loadptr dtrace_load32 527 #endif 528 529 #define DTRACE_DYNHASH_FREE 0 530 #define DTRACE_DYNHASH_SINK 1 531 #define DTRACE_DYNHASH_VALID 2 532 533 #define DTRACE_MATCH_NEXT 0 534 #define DTRACE_MATCH_DONE 1 535 #define DTRACE_ANCHORED(probe) ((probe)->dtpr_func[0] != '\0') 536 #define DTRACE_STATE_ALIGN 64 537 538 #define DTRACE_FLAGS2FLT(flags) \ 539 (((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR : \ 540 ((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP : \ 541 ((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO : \ 542 ((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV : \ 543 ((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV : \ 544 ((flags) & CPU_DTRACE_TUPOFLOW) ? DTRACEFLT_TUPOFLOW : \ 545 ((flags) & CPU_DTRACE_BADALIGN) ? DTRACEFLT_BADALIGN : \ 546 ((flags) & CPU_DTRACE_NOSCRATCH) ? DTRACEFLT_NOSCRATCH : \ 547 ((flags) & CPU_DTRACE_BADSTACK) ? DTRACEFLT_BADSTACK : \ 548 DTRACEFLT_UNKNOWN) 549 550 #define DTRACEACT_ISSTRING(act) \ 551 ((act)->dta_kind == DTRACEACT_DIFEXPR && \ 552 (act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) 553 554 /* Function prototype definitions: */ 555 static size_t dtrace_strlen(const char *, size_t); 556 static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id); 557 static void dtrace_enabling_provide(dtrace_provider_t *); 558 static int dtrace_enabling_match(dtrace_enabling_t *, int *); 559 static void dtrace_enabling_matchall(void); 560 static void dtrace_enabling_reap(void); 561 static dtrace_state_t *dtrace_anon_grab(void); 562 static uint64_t dtrace_helper(int, dtrace_mstate_t *, 563 dtrace_state_t *, uint64_t, uint64_t); 564 static dtrace_helpers_t *dtrace_helpers_create(proc_t *); 565 static void dtrace_buffer_drop(dtrace_buffer_t *); 566 static int dtrace_buffer_consumed(dtrace_buffer_t *, hrtime_t when); 567 static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t, 568 dtrace_state_t *, dtrace_mstate_t *); 569 static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t, 570 dtrace_optval_t); 571 static int dtrace_ecb_create_enable(dtrace_probe_t *, void *); 572 static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *); 573 uint16_t dtrace_load16(uintptr_t); 574 uint32_t dtrace_load32(uintptr_t); 575 uint64_t dtrace_load64(uintptr_t); 576 uint8_t dtrace_load8(uintptr_t); 577 void dtrace_dynvar_clean(dtrace_dstate_t *); 578 dtrace_dynvar_t *dtrace_dynvar(dtrace_dstate_t *, uint_t, dtrace_key_t *, 579 size_t, dtrace_dynvar_op_t, dtrace_mstate_t *, dtrace_vstate_t *); 580 uintptr_t dtrace_dif_varstr(uintptr_t, dtrace_state_t *, dtrace_mstate_t *); 581 582 /* 583 * DTrace Probe Context Functions 584 * 585 * These functions are called from probe context. Because probe context is 586 * any context in which C may be called, arbitrarily locks may be held, 587 * interrupts may be disabled, we may be in arbitrary dispatched state, etc. 588 * As a result, functions called from probe context may only call other DTrace 589 * support functions -- they may not interact at all with the system at large. 590 * (Note that the ASSERT macro is made probe-context safe by redefining it in 591 * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary 592 * loads are to be performed from probe context, they _must_ be in terms of 593 * the safe dtrace_load*() variants. 594 * 595 * Some functions in this block are not actually called from probe context; 596 * for these functions, there will be a comment above the function reading 597 * "Note: not called from probe context." 598 */ 599 void 600 dtrace_panic(const char *format, ...) 601 { 602 va_list alist; 603 604 va_start(alist, format); 605 dtrace_vpanic(format, alist); 606 va_end(alist); 607 } 608 609 int 610 dtrace_assfail(const char *a, const char *f, int l) 611 { 612 dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l); 613 614 /* 615 * We just need something here that even the most clever compiler 616 * cannot optimize away. 617 */ 618 return (a[(uintptr_t)f]); 619 } 620 621 /* 622 * Atomically increment a specified error counter from probe context. 623 */ 624 static void 625 dtrace_error(uint32_t *counter) 626 { 627 /* 628 * Most counters stored to in probe context are per-CPU counters. 629 * However, there are some error conditions that are sufficiently 630 * arcane that they don't merit per-CPU storage. If these counters 631 * are incremented concurrently on different CPUs, scalability will be 632 * adversely affected -- but we don't expect them to be white-hot in a 633 * correctly constructed enabling... 634 */ 635 uint32_t oval, nval; 636 637 do { 638 oval = *counter; 639 640 if ((nval = oval + 1) == 0) { 641 /* 642 * If the counter would wrap, set it to 1 -- assuring 643 * that the counter is never zero when we have seen 644 * errors. (The counter must be 32-bits because we 645 * aren't guaranteed a 64-bit compare&swap operation.) 646 * To save this code both the infamy of being fingered 647 * by a priggish news story and the indignity of being 648 * the target of a neo-puritan witch trial, we're 649 * carefully avoiding any colorful description of the 650 * likelihood of this condition -- but suffice it to 651 * say that it is only slightly more likely than the 652 * overflow of predicate cache IDs, as discussed in 653 * dtrace_predicate_create(). 654 */ 655 nval = 1; 656 } 657 } while (dtrace_cas32(counter, oval, nval) != oval); 658 } 659 660 /* 661 * Use the DTRACE_LOADFUNC macro to define functions for each of loading a 662 * uint8_t, a uint16_t, a uint32_t and a uint64_t. 663 */ 664 DTRACE_LOADFUNC(8) 665 DTRACE_LOADFUNC(16) 666 DTRACE_LOADFUNC(32) 667 DTRACE_LOADFUNC(64) 668 669 static int 670 dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate) 671 { 672 if (dest < mstate->dtms_scratch_base) 673 return (0); 674 675 if (dest + size < dest) 676 return (0); 677 678 if (dest + size > mstate->dtms_scratch_ptr) 679 return (0); 680 681 return (1); 682 } 683 684 static int 685 dtrace_canstore_statvar(uint64_t addr, size_t sz, 686 dtrace_statvar_t **svars, int nsvars) 687 { 688 int i; 689 690 for (i = 0; i < nsvars; i++) { 691 dtrace_statvar_t *svar = svars[i]; 692 693 if (svar == NULL || svar->dtsv_size == 0) 694 continue; 695 696 if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size)) 697 return (1); 698 } 699 700 return (0); 701 } 702 703 /* 704 * Check to see if the address is within a memory region to which a store may 705 * be issued. This includes the DTrace scratch areas, and any DTrace variable 706 * region. The caller of dtrace_canstore() is responsible for performing any 707 * alignment checks that are needed before stores are actually executed. 708 */ 709 static int 710 dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, 711 dtrace_vstate_t *vstate) 712 { 713 /* 714 * First, check to see if the address is in scratch space... 715 */ 716 if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base, 717 mstate->dtms_scratch_size)) 718 return (1); 719 720 /* 721 * Now check to see if it's a dynamic variable. This check will pick 722 * up both thread-local variables and any global dynamically-allocated 723 * variables. 724 */ 725 if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base, 726 vstate->dtvs_dynvars.dtds_size)) { 727 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; 728 uintptr_t base = (uintptr_t)dstate->dtds_base + 729 (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t)); 730 uintptr_t chunkoffs; 731 732 /* 733 * Before we assume that we can store here, we need to make 734 * sure that it isn't in our metadata -- storing to our 735 * dynamic variable metadata would corrupt our state. For 736 * the range to not include any dynamic variable metadata, 737 * it must: 738 * 739 * (1) Start above the hash table that is at the base of 740 * the dynamic variable space 741 * 742 * (2) Have a starting chunk offset that is beyond the 743 * dtrace_dynvar_t that is at the base of every chunk 744 * 745 * (3) Not span a chunk boundary 746 * 747 */ 748 if (addr < base) 749 return (0); 750 751 chunkoffs = (addr - base) % dstate->dtds_chunksize; 752 753 if (chunkoffs < sizeof (dtrace_dynvar_t)) 754 return (0); 755 756 if (chunkoffs + sz > dstate->dtds_chunksize) 757 return (0); 758 759 return (1); 760 } 761 762 /* 763 * Finally, check the static local and global variables. These checks 764 * take the longest, so we perform them last. 765 */ 766 if (dtrace_canstore_statvar(addr, sz, 767 vstate->dtvs_locals, vstate->dtvs_nlocals)) 768 return (1); 769 770 if (dtrace_canstore_statvar(addr, sz, 771 vstate->dtvs_globals, vstate->dtvs_nglobals)) 772 return (1); 773 774 return (0); 775 } 776 777 778 /* 779 * Convenience routine to check to see if the address is within a memory 780 * region in which a load may be issued given the user's privilege level; 781 * if not, it sets the appropriate error flags and loads 'addr' into the 782 * illegal value slot. 783 * 784 * DTrace subroutines (DIF_SUBR_*) should use this helper to implement 785 * appropriate memory access protection. 786 */ 787 static int 788 dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, 789 dtrace_vstate_t *vstate) 790 { 791 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; 792 793 /* 794 * If we hold the privilege to read from kernel memory, then 795 * everything is readable. 796 */ 797 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 798 return (1); 799 800 /* 801 * You can obviously read that which you can store. 802 */ 803 if (dtrace_canstore(addr, sz, mstate, vstate)) 804 return (1); 805 806 /* 807 * We're allowed to read from our own string table. 808 */ 809 if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab, 810 mstate->dtms_difo->dtdo_strlen)) 811 return (1); 812 813 DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV); 814 *illval = addr; 815 return (0); 816 } 817 818 /* 819 * Convenience routine to check to see if a given string is within a memory 820 * region in which a load may be issued given the user's privilege level; 821 * this exists so that we don't need to issue unnecessary dtrace_strlen() 822 * calls in the event that the user has all privileges. 823 */ 824 static int 825 dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, 826 dtrace_vstate_t *vstate) 827 { 828 size_t strsz; 829 830 /* 831 * If we hold the privilege to read from kernel memory, then 832 * everything is readable. 833 */ 834 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 835 return (1); 836 837 strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz); 838 if (dtrace_canload(addr, strsz, mstate, vstate)) 839 return (1); 840 841 return (0); 842 } 843 844 /* 845 * Convenience routine to check to see if a given variable is within a memory 846 * region in which a load may be issued given the user's privilege level. 847 */ 848 static int 849 dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate, 850 dtrace_vstate_t *vstate) 851 { 852 size_t sz; 853 ASSERT(type->dtdt_flags & DIF_TF_BYREF); 854 855 /* 856 * If we hold the privilege to read from kernel memory, then 857 * everything is readable. 858 */ 859 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 860 return (1); 861 862 if (type->dtdt_kind == DIF_TYPE_STRING) 863 sz = dtrace_strlen(src, 864 vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1; 865 else 866 sz = type->dtdt_size; 867 868 return (dtrace_canload((uintptr_t)src, sz, mstate, vstate)); 869 } 870 871 /* 872 * Compare two strings using safe loads. 873 */ 874 static int 875 dtrace_strncmp(char *s1, char *s2, size_t limit) 876 { 877 uint8_t c1, c2; 878 volatile uint16_t *flags; 879 880 if (s1 == s2 || limit == 0) 881 return (0); 882 883 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; 884 885 do { 886 if (s1 == NULL) { 887 c1 = '\0'; 888 } else { 889 c1 = dtrace_load8((uintptr_t)s1++); 890 } 891 892 if (s2 == NULL) { 893 c2 = '\0'; 894 } else { 895 c2 = dtrace_load8((uintptr_t)s2++); 896 } 897 898 if (c1 != c2) 899 return (c1 - c2); 900 } while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT)); 901 902 return (0); 903 } 904 905 /* 906 * Compute strlen(s) for a string using safe memory accesses. The additional 907 * len parameter is used to specify a maximum length to ensure completion. 908 */ 909 static size_t 910 dtrace_strlen(const char *s, size_t lim) 911 { 912 uint_t len; 913 914 for (len = 0; len != lim; len++) { 915 if (dtrace_load8((uintptr_t)s++) == '\0') 916 break; 917 } 918 919 return (len); 920 } 921 922 /* 923 * Check if an address falls within a toxic region. 924 */ 925 static int 926 dtrace_istoxic(uintptr_t kaddr, size_t size) 927 { 928 uintptr_t taddr, tsize; 929 int i; 930 931 for (i = 0; i < dtrace_toxranges; i++) { 932 taddr = dtrace_toxrange[i].dtt_base; 933 tsize = dtrace_toxrange[i].dtt_limit - taddr; 934 935 if (kaddr - taddr < tsize) { 936 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); 937 cpu_core[curcpu].cpuc_dtrace_illval = kaddr; 938 return (1); 939 } 940 941 if (taddr - kaddr < size) { 942 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); 943 cpu_core[curcpu].cpuc_dtrace_illval = taddr; 944 return (1); 945 } 946 } 947 948 return (0); 949 } 950 951 /* 952 * Copy src to dst using safe memory accesses. The src is assumed to be unsafe 953 * memory specified by the DIF program. The dst is assumed to be safe memory 954 * that we can store to directly because it is managed by DTrace. As with 955 * standard bcopy, overlapping copies are handled properly. 956 */ 957 static void 958 dtrace_bcopy(const void *src, void *dst, size_t len) 959 { 960 if (len != 0) { 961 uint8_t *s1 = dst; 962 const uint8_t *s2 = src; 963 964 if (s1 <= s2) { 965 do { 966 *s1++ = dtrace_load8((uintptr_t)s2++); 967 } while (--len != 0); 968 } else { 969 s2 += len; 970 s1 += len; 971 972 do { 973 *--s1 = dtrace_load8((uintptr_t)--s2); 974 } while (--len != 0); 975 } 976 } 977 } 978 979 /* 980 * Copy src to dst using safe memory accesses, up to either the specified 981 * length, or the point that a nul byte is encountered. The src is assumed to 982 * be unsafe memory specified by the DIF program. The dst is assumed to be 983 * safe memory that we can store to directly because it is managed by DTrace. 984 * Unlike dtrace_bcopy(), overlapping regions are not handled. 985 */ 986 static void 987 dtrace_strcpy(const void *src, void *dst, size_t len) 988 { 989 if (len != 0) { 990 uint8_t *s1 = dst, c; 991 const uint8_t *s2 = src; 992 993 do { 994 *s1++ = c = dtrace_load8((uintptr_t)s2++); 995 } while (--len != 0 && c != '\0'); 996 } 997 } 998 999 /* 1000 * Copy src to dst, deriving the size and type from the specified (BYREF) 1001 * variable type. The src is assumed to be unsafe memory specified by the DIF 1002 * program. The dst is assumed to be DTrace variable memory that is of the 1003 * specified type; we assume that we can store to directly. 1004 */ 1005 static void 1006 dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type) 1007 { 1008 ASSERT(type->dtdt_flags & DIF_TF_BYREF); 1009 1010 if (type->dtdt_kind == DIF_TYPE_STRING) { 1011 dtrace_strcpy(src, dst, type->dtdt_size); 1012 } else { 1013 dtrace_bcopy(src, dst, type->dtdt_size); 1014 } 1015 } 1016 1017 /* 1018 * Compare s1 to s2 using safe memory accesses. The s1 data is assumed to be 1019 * unsafe memory specified by the DIF program. The s2 data is assumed to be 1020 * safe memory that we can access directly because it is managed by DTrace. 1021 */ 1022 static int 1023 dtrace_bcmp(const void *s1, const void *s2, size_t len) 1024 { 1025 volatile uint16_t *flags; 1026 1027 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; 1028 1029 if (s1 == s2) 1030 return (0); 1031 1032 if (s1 == NULL || s2 == NULL) 1033 return (1); 1034 1035 if (s1 != s2 && len != 0) { 1036 const uint8_t *ps1 = s1; 1037 const uint8_t *ps2 = s2; 1038 1039 do { 1040 if (dtrace_load8((uintptr_t)ps1++) != *ps2++) 1041 return (1); 1042 } while (--len != 0 && !(*flags & CPU_DTRACE_FAULT)); 1043 } 1044 return (0); 1045 } 1046 1047 /* 1048 * Zero the specified region using a simple byte-by-byte loop. Note that this 1049 * is for safe DTrace-managed memory only. 1050 */ 1051 static void 1052 dtrace_bzero(void *dst, size_t len) 1053 { 1054 uchar_t *cp; 1055 1056 for (cp = dst; len != 0; len--) 1057 *cp++ = 0; 1058 } 1059 1060 static void 1061 dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum) 1062 { 1063 uint64_t result[2]; 1064 1065 result[0] = addend1[0] + addend2[0]; 1066 result[1] = addend1[1] + addend2[1] + 1067 (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0); 1068 1069 sum[0] = result[0]; 1070 sum[1] = result[1]; 1071 } 1072 1073 /* 1074 * Shift the 128-bit value in a by b. If b is positive, shift left. 1075 * If b is negative, shift right. 1076 */ 1077 static void 1078 dtrace_shift_128(uint64_t *a, int b) 1079 { 1080 uint64_t mask; 1081 1082 if (b == 0) 1083 return; 1084 1085 if (b < 0) { 1086 b = -b; 1087 if (b >= 64) { 1088 a[0] = a[1] >> (b - 64); 1089 a[1] = 0; 1090 } else { 1091 a[0] >>= b; 1092 mask = 1LL << (64 - b); 1093 mask -= 1; 1094 a[0] |= ((a[1] & mask) << (64 - b)); 1095 a[1] >>= b; 1096 } 1097 } else { 1098 if (b >= 64) { 1099 a[1] = a[0] << (b - 64); 1100 a[0] = 0; 1101 } else { 1102 a[1] <<= b; 1103 mask = a[0] >> (64 - b); 1104 a[1] |= mask; 1105 a[0] <<= b; 1106 } 1107 } 1108 } 1109 1110 /* 1111 * The basic idea is to break the 2 64-bit values into 4 32-bit values, 1112 * use native multiplication on those, and then re-combine into the 1113 * resulting 128-bit value. 1114 * 1115 * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) = 1116 * hi1 * hi2 << 64 + 1117 * hi1 * lo2 << 32 + 1118 * hi2 * lo1 << 32 + 1119 * lo1 * lo2 1120 */ 1121 static void 1122 dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product) 1123 { 1124 uint64_t hi1, hi2, lo1, lo2; 1125 uint64_t tmp[2]; 1126 1127 hi1 = factor1 >> 32; 1128 hi2 = factor2 >> 32; 1129 1130 lo1 = factor1 & DT_MASK_LO; 1131 lo2 = factor2 & DT_MASK_LO; 1132 1133 product[0] = lo1 * lo2; 1134 product[1] = hi1 * hi2; 1135 1136 tmp[0] = hi1 * lo2; 1137 tmp[1] = 0; 1138 dtrace_shift_128(tmp, 32); 1139 dtrace_add_128(product, tmp, product); 1140 1141 tmp[0] = hi2 * lo1; 1142 tmp[1] = 0; 1143 dtrace_shift_128(tmp, 32); 1144 dtrace_add_128(product, tmp, product); 1145 } 1146 1147 /* 1148 * This privilege check should be used by actions and subroutines to 1149 * verify that the user credentials of the process that enabled the 1150 * invoking ECB match the target credentials 1151 */ 1152 static int 1153 dtrace_priv_proc_common_user(dtrace_state_t *state) 1154 { 1155 cred_t *cr, *s_cr = state->dts_cred.dcr_cred; 1156 1157 /* 1158 * We should always have a non-NULL state cred here, since if cred 1159 * is null (anonymous tracing), we fast-path bypass this routine. 1160 */ 1161 ASSERT(s_cr != NULL); 1162 1163 if ((cr = CRED()) != NULL && 1164 s_cr->cr_uid == cr->cr_uid && 1165 s_cr->cr_uid == cr->cr_ruid && 1166 s_cr->cr_uid == cr->cr_suid && 1167 s_cr->cr_gid == cr->cr_gid && 1168 s_cr->cr_gid == cr->cr_rgid && 1169 s_cr->cr_gid == cr->cr_sgid) 1170 return (1); 1171 1172 return (0); 1173 } 1174 1175 /* 1176 * This privilege check should be used by actions and subroutines to 1177 * verify that the zone of the process that enabled the invoking ECB 1178 * matches the target credentials 1179 */ 1180 static int 1181 dtrace_priv_proc_common_zone(dtrace_state_t *state) 1182 { 1183 #if defined(sun) 1184 cred_t *cr, *s_cr = state->dts_cred.dcr_cred; 1185 1186 /* 1187 * We should always have a non-NULL state cred here, since if cred 1188 * is null (anonymous tracing), we fast-path bypass this routine. 1189 */ 1190 ASSERT(s_cr != NULL); 1191 1192 if ((cr = CRED()) != NULL && 1193 s_cr->cr_zone == cr->cr_zone) 1194 return (1); 1195 1196 return (0); 1197 #else 1198 return (1); 1199 #endif 1200 } 1201 1202 /* 1203 * This privilege check should be used by actions and subroutines to 1204 * verify that the process has not setuid or changed credentials. 1205 */ 1206 static int 1207 dtrace_priv_proc_common_nocd(void) 1208 { 1209 proc_t *proc; 1210 1211 if ((proc = ttoproc(curthread)) != NULL && 1212 !(proc->p_flag & SNOCD)) 1213 return (1); 1214 1215 return (0); 1216 } 1217 1218 static int 1219 dtrace_priv_proc_destructive(dtrace_state_t *state) 1220 { 1221 int action = state->dts_cred.dcr_action; 1222 1223 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) && 1224 dtrace_priv_proc_common_zone(state) == 0) 1225 goto bad; 1226 1227 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) && 1228 dtrace_priv_proc_common_user(state) == 0) 1229 goto bad; 1230 1231 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) && 1232 dtrace_priv_proc_common_nocd() == 0) 1233 goto bad; 1234 1235 return (1); 1236 1237 bad: 1238 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; 1239 1240 return (0); 1241 } 1242 1243 static int 1244 dtrace_priv_proc_control(dtrace_state_t *state) 1245 { 1246 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL) 1247 return (1); 1248 1249 if (dtrace_priv_proc_common_zone(state) && 1250 dtrace_priv_proc_common_user(state) && 1251 dtrace_priv_proc_common_nocd()) 1252 return (1); 1253 1254 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; 1255 1256 return (0); 1257 } 1258 1259 static int 1260 dtrace_priv_proc(dtrace_state_t *state) 1261 { 1262 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC) 1263 return (1); 1264 1265 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; 1266 1267 return (0); 1268 } 1269 1270 static int 1271 dtrace_priv_kernel(dtrace_state_t *state) 1272 { 1273 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL) 1274 return (1); 1275 1276 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; 1277 1278 return (0); 1279 } 1280 1281 static int 1282 dtrace_priv_kernel_destructive(dtrace_state_t *state) 1283 { 1284 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE) 1285 return (1); 1286 1287 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; 1288 1289 return (0); 1290 } 1291 1292 /* 1293 * Note: not called from probe context. This function is called 1294 * asynchronously (and at a regular interval) from outside of probe context to 1295 * clean the dirty dynamic variable lists on all CPUs. Dynamic variable 1296 * cleaning is explained in detail in <sys/dtrace_impl.h>. 1297 */ 1298 void 1299 dtrace_dynvar_clean(dtrace_dstate_t *dstate) 1300 { 1301 dtrace_dynvar_t *dirty; 1302 dtrace_dstate_percpu_t *dcpu; 1303 int i, work = 0; 1304 1305 for (i = 0; i < NCPU; i++) { 1306 dcpu = &dstate->dtds_percpu[i]; 1307 1308 ASSERT(dcpu->dtdsc_rinsing == NULL); 1309 1310 /* 1311 * If the dirty list is NULL, there is no dirty work to do. 1312 */ 1313 if (dcpu->dtdsc_dirty == NULL) 1314 continue; 1315 1316 /* 1317 * If the clean list is non-NULL, then we're not going to do 1318 * any work for this CPU -- it means that there has not been 1319 * a dtrace_dynvar() allocation on this CPU (or from this CPU) 1320 * since the last time we cleaned house. 1321 */ 1322 if (dcpu->dtdsc_clean != NULL) 1323 continue; 1324 1325 work = 1; 1326 1327 /* 1328 * Atomically move the dirty list aside. 1329 */ 1330 do { 1331 dirty = dcpu->dtdsc_dirty; 1332 1333 /* 1334 * Before we zap the dirty list, set the rinsing list. 1335 * (This allows for a potential assertion in 1336 * dtrace_dynvar(): if a free dynamic variable appears 1337 * on a hash chain, either the dirty list or the 1338 * rinsing list for some CPU must be non-NULL.) 1339 */ 1340 dcpu->dtdsc_rinsing = dirty; 1341 dtrace_membar_producer(); 1342 } while (dtrace_casptr(&dcpu->dtdsc_dirty, 1343 dirty, NULL) != dirty); 1344 } 1345 1346 if (!work) { 1347 /* 1348 * We have no work to do; we can simply return. 1349 */ 1350 return; 1351 } 1352 1353 dtrace_sync(); 1354 1355 for (i = 0; i < NCPU; i++) { 1356 dcpu = &dstate->dtds_percpu[i]; 1357 1358 if (dcpu->dtdsc_rinsing == NULL) 1359 continue; 1360 1361 /* 1362 * We are now guaranteed that no hash chain contains a pointer 1363 * into this dirty list; we can make it clean. 1364 */ 1365 ASSERT(dcpu->dtdsc_clean == NULL); 1366 dcpu->dtdsc_clean = dcpu->dtdsc_rinsing; 1367 dcpu->dtdsc_rinsing = NULL; 1368 } 1369 1370 /* 1371 * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make 1372 * sure that all CPUs have seen all of the dtdsc_clean pointers. 1373 * This prevents a race whereby a CPU incorrectly decides that 1374 * the state should be something other than DTRACE_DSTATE_CLEAN 1375 * after dtrace_dynvar_clean() has completed. 1376 */ 1377 dtrace_sync(); 1378 1379 dstate->dtds_state = DTRACE_DSTATE_CLEAN; 1380 } 1381 1382 /* 1383 * Depending on the value of the op parameter, this function looks-up, 1384 * allocates or deallocates an arbitrarily-keyed dynamic variable. If an 1385 * allocation is requested, this function will return a pointer to a 1386 * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no 1387 * variable can be allocated. If NULL is returned, the appropriate counter 1388 * will be incremented. 1389 */ 1390 dtrace_dynvar_t * 1391 dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys, 1392 dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op, 1393 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate) 1394 { 1395 uint64_t hashval = DTRACE_DYNHASH_VALID; 1396 dtrace_dynhash_t *hash = dstate->dtds_hash; 1397 dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL; 1398 processorid_t me = curcpu, cpu = me; 1399 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me]; 1400 size_t bucket, ksize; 1401 size_t chunksize = dstate->dtds_chunksize; 1402 uintptr_t kdata, lock, nstate; 1403 uint_t i; 1404 1405 ASSERT(nkeys != 0); 1406 1407 /* 1408 * Hash the key. As with aggregations, we use Jenkins' "One-at-a-time" 1409 * algorithm. For the by-value portions, we perform the algorithm in 1410 * 16-bit chunks (as opposed to 8-bit chunks). This speeds things up a 1411 * bit, and seems to have only a minute effect on distribution. For 1412 * the by-reference data, we perform "One-at-a-time" iterating (safely) 1413 * over each referenced byte. It's painful to do this, but it's much 1414 * better than pathological hash distribution. The efficacy of the 1415 * hashing algorithm (and a comparison with other algorithms) may be 1416 * found by running the ::dtrace_dynstat MDB dcmd. 1417 */ 1418 for (i = 0; i < nkeys; i++) { 1419 if (key[i].dttk_size == 0) { 1420 uint64_t val = key[i].dttk_value; 1421 1422 hashval += (val >> 48) & 0xffff; 1423 hashval += (hashval << 10); 1424 hashval ^= (hashval >> 6); 1425 1426 hashval += (val >> 32) & 0xffff; 1427 hashval += (hashval << 10); 1428 hashval ^= (hashval >> 6); 1429 1430 hashval += (val >> 16) & 0xffff; 1431 hashval += (hashval << 10); 1432 hashval ^= (hashval >> 6); 1433 1434 hashval += val & 0xffff; 1435 hashval += (hashval << 10); 1436 hashval ^= (hashval >> 6); 1437 } else { 1438 /* 1439 * This is incredibly painful, but it beats the hell 1440 * out of the alternative. 1441 */ 1442 uint64_t j, size = key[i].dttk_size; 1443 uintptr_t base = (uintptr_t)key[i].dttk_value; 1444 1445 if (!dtrace_canload(base, size, mstate, vstate)) 1446 break; 1447 1448 for (j = 0; j < size; j++) { 1449 hashval += dtrace_load8(base + j); 1450 hashval += (hashval << 10); 1451 hashval ^= (hashval >> 6); 1452 } 1453 } 1454 } 1455 1456 if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) 1457 return (NULL); 1458 1459 hashval += (hashval << 3); 1460 hashval ^= (hashval >> 11); 1461 hashval += (hashval << 15); 1462 1463 /* 1464 * There is a remote chance (ideally, 1 in 2^31) that our hashval 1465 * comes out to be one of our two sentinel hash values. If this 1466 * actually happens, we set the hashval to be a value known to be a 1467 * non-sentinel value. 1468 */ 1469 if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK) 1470 hashval = DTRACE_DYNHASH_VALID; 1471 1472 /* 1473 * Yes, it's painful to do a divide here. If the cycle count becomes 1474 * important here, tricks can be pulled to reduce it. (However, it's 1475 * critical that hash collisions be kept to an absolute minimum; 1476 * they're much more painful than a divide.) It's better to have a 1477 * solution that generates few collisions and still keeps things 1478 * relatively simple. 1479 */ 1480 bucket = hashval % dstate->dtds_hashsize; 1481 1482 if (op == DTRACE_DYNVAR_DEALLOC) { 1483 volatile uintptr_t *lockp = &hash[bucket].dtdh_lock; 1484 1485 for (;;) { 1486 while ((lock = *lockp) & 1) 1487 continue; 1488 1489 if (dtrace_casptr((volatile void *)lockp, 1490 (volatile void *)lock, (volatile void *)(lock + 1)) == (void *)lock) 1491 break; 1492 } 1493 1494 dtrace_membar_producer(); 1495 } 1496 1497 top: 1498 prev = NULL; 1499 lock = hash[bucket].dtdh_lock; 1500 1501 dtrace_membar_consumer(); 1502 1503 start = hash[bucket].dtdh_chain; 1504 ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK || 1505 start->dtdv_hashval != DTRACE_DYNHASH_FREE || 1506 op != DTRACE_DYNVAR_DEALLOC)); 1507 1508 for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) { 1509 dtrace_tuple_t *dtuple = &dvar->dtdv_tuple; 1510 dtrace_key_t *dkey = &dtuple->dtt_key[0]; 1511 1512 if (dvar->dtdv_hashval != hashval) { 1513 if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) { 1514 /* 1515 * We've reached the sink, and therefore the 1516 * end of the hash chain; we can kick out of 1517 * the loop knowing that we have seen a valid 1518 * snapshot of state. 1519 */ 1520 ASSERT(dvar->dtdv_next == NULL); 1521 ASSERT(dvar == &dtrace_dynhash_sink); 1522 break; 1523 } 1524 1525 if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) { 1526 /* 1527 * We've gone off the rails: somewhere along 1528 * the line, one of the members of this hash 1529 * chain was deleted. Note that we could also 1530 * detect this by simply letting this loop run 1531 * to completion, as we would eventually hit 1532 * the end of the dirty list. However, we 1533 * want to avoid running the length of the 1534 * dirty list unnecessarily (it might be quite 1535 * long), so we catch this as early as 1536 * possible by detecting the hash marker. In 1537 * this case, we simply set dvar to NULL and 1538 * break; the conditional after the loop will 1539 * send us back to top. 1540 */ 1541 dvar = NULL; 1542 break; 1543 } 1544 1545 goto next; 1546 } 1547 1548 if (dtuple->dtt_nkeys != nkeys) 1549 goto next; 1550 1551 for (i = 0; i < nkeys; i++, dkey++) { 1552 if (dkey->dttk_size != key[i].dttk_size) 1553 goto next; /* size or type mismatch */ 1554 1555 if (dkey->dttk_size != 0) { 1556 if (dtrace_bcmp( 1557 (void *)(uintptr_t)key[i].dttk_value, 1558 (void *)(uintptr_t)dkey->dttk_value, 1559 dkey->dttk_size)) 1560 goto next; 1561 } else { 1562 if (dkey->dttk_value != key[i].dttk_value) 1563 goto next; 1564 } 1565 } 1566 1567 if (op != DTRACE_DYNVAR_DEALLOC) 1568 return (dvar); 1569 1570 ASSERT(dvar->dtdv_next == NULL || 1571 dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE); 1572 1573 if (prev != NULL) { 1574 ASSERT(hash[bucket].dtdh_chain != dvar); 1575 ASSERT(start != dvar); 1576 ASSERT(prev->dtdv_next == dvar); 1577 prev->dtdv_next = dvar->dtdv_next; 1578 } else { 1579 if (dtrace_casptr(&hash[bucket].dtdh_chain, 1580 start, dvar->dtdv_next) != start) { 1581 /* 1582 * We have failed to atomically swing the 1583 * hash table head pointer, presumably because 1584 * of a conflicting allocation on another CPU. 1585 * We need to reread the hash chain and try 1586 * again. 1587 */ 1588 goto top; 1589 } 1590 } 1591 1592 dtrace_membar_producer(); 1593 1594 /* 1595 * Now set the hash value to indicate that it's free. 1596 */ 1597 ASSERT(hash[bucket].dtdh_chain != dvar); 1598 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; 1599 1600 dtrace_membar_producer(); 1601 1602 /* 1603 * Set the next pointer to point at the dirty list, and 1604 * atomically swing the dirty pointer to the newly freed dvar. 1605 */ 1606 do { 1607 next = dcpu->dtdsc_dirty; 1608 dvar->dtdv_next = next; 1609 } while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next); 1610 1611 /* 1612 * Finally, unlock this hash bucket. 1613 */ 1614 ASSERT(hash[bucket].dtdh_lock == lock); 1615 ASSERT(lock & 1); 1616 hash[bucket].dtdh_lock++; 1617 1618 return (NULL); 1619 next: 1620 prev = dvar; 1621 continue; 1622 } 1623 1624 if (dvar == NULL) { 1625 /* 1626 * If dvar is NULL, it is because we went off the rails: 1627 * one of the elements that we traversed in the hash chain 1628 * was deleted while we were traversing it. In this case, 1629 * we assert that we aren't doing a dealloc (deallocs lock 1630 * the hash bucket to prevent themselves from racing with 1631 * one another), and retry the hash chain traversal. 1632 */ 1633 ASSERT(op != DTRACE_DYNVAR_DEALLOC); 1634 goto top; 1635 } 1636 1637 if (op != DTRACE_DYNVAR_ALLOC) { 1638 /* 1639 * If we are not to allocate a new variable, we want to 1640 * return NULL now. Before we return, check that the value 1641 * of the lock word hasn't changed. If it has, we may have 1642 * seen an inconsistent snapshot. 1643 */ 1644 if (op == DTRACE_DYNVAR_NOALLOC) { 1645 if (hash[bucket].dtdh_lock != lock) 1646 goto top; 1647 } else { 1648 ASSERT(op == DTRACE_DYNVAR_DEALLOC); 1649 ASSERT(hash[bucket].dtdh_lock == lock); 1650 ASSERT(lock & 1); 1651 hash[bucket].dtdh_lock++; 1652 } 1653 1654 return (NULL); 1655 } 1656 1657 /* 1658 * We need to allocate a new dynamic variable. The size we need is the 1659 * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the 1660 * size of any auxiliary key data (rounded up to 8-byte alignment) plus 1661 * the size of any referred-to data (dsize). We then round the final 1662 * size up to the chunksize for allocation. 1663 */ 1664 for (ksize = 0, i = 0; i < nkeys; i++) 1665 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); 1666 1667 /* 1668 * This should be pretty much impossible, but could happen if, say, 1669 * strange DIF specified the tuple. Ideally, this should be an 1670 * assertion and not an error condition -- but that requires that the 1671 * chunksize calculation in dtrace_difo_chunksize() be absolutely 1672 * bullet-proof. (That is, it must not be able to be fooled by 1673 * malicious DIF.) Given the lack of backwards branches in DIF, 1674 * solving this would presumably not amount to solving the Halting 1675 * Problem -- but it still seems awfully hard. 1676 */ 1677 if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) + 1678 ksize + dsize > chunksize) { 1679 dcpu->dtdsc_drops++; 1680 return (NULL); 1681 } 1682 1683 nstate = DTRACE_DSTATE_EMPTY; 1684 1685 do { 1686 retry: 1687 free = dcpu->dtdsc_free; 1688 1689 if (free == NULL) { 1690 dtrace_dynvar_t *clean = dcpu->dtdsc_clean; 1691 void *rval; 1692 1693 if (clean == NULL) { 1694 /* 1695 * We're out of dynamic variable space on 1696 * this CPU. Unless we have tried all CPUs, 1697 * we'll try to allocate from a different 1698 * CPU. 1699 */ 1700 switch (dstate->dtds_state) { 1701 case DTRACE_DSTATE_CLEAN: { 1702 void *sp = &dstate->dtds_state; 1703 1704 if (++cpu >= NCPU) 1705 cpu = 0; 1706 1707 if (dcpu->dtdsc_dirty != NULL && 1708 nstate == DTRACE_DSTATE_EMPTY) 1709 nstate = DTRACE_DSTATE_DIRTY; 1710 1711 if (dcpu->dtdsc_rinsing != NULL) 1712 nstate = DTRACE_DSTATE_RINSING; 1713 1714 dcpu = &dstate->dtds_percpu[cpu]; 1715 1716 if (cpu != me) 1717 goto retry; 1718 1719 (void) dtrace_cas32(sp, 1720 DTRACE_DSTATE_CLEAN, nstate); 1721 1722 /* 1723 * To increment the correct bean 1724 * counter, take another lap. 1725 */ 1726 goto retry; 1727 } 1728 1729 case DTRACE_DSTATE_DIRTY: 1730 dcpu->dtdsc_dirty_drops++; 1731 break; 1732 1733 case DTRACE_DSTATE_RINSING: 1734 dcpu->dtdsc_rinsing_drops++; 1735 break; 1736 1737 case DTRACE_DSTATE_EMPTY: 1738 dcpu->dtdsc_drops++; 1739 break; 1740 } 1741 1742 DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP); 1743 return (NULL); 1744 } 1745 1746 /* 1747 * The clean list appears to be non-empty. We want to 1748 * move the clean list to the free list; we start by 1749 * moving the clean pointer aside. 1750 */ 1751 if (dtrace_casptr(&dcpu->dtdsc_clean, 1752 clean, NULL) != clean) { 1753 /* 1754 * We are in one of two situations: 1755 * 1756 * (a) The clean list was switched to the 1757 * free list by another CPU. 1758 * 1759 * (b) The clean list was added to by the 1760 * cleansing cyclic. 1761 * 1762 * In either of these situations, we can 1763 * just reattempt the free list allocation. 1764 */ 1765 goto retry; 1766 } 1767 1768 ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE); 1769 1770 /* 1771 * Now we'll move the clean list to the free list. 1772 * It's impossible for this to fail: the only way 1773 * the free list can be updated is through this 1774 * code path, and only one CPU can own the clean list. 1775 * Thus, it would only be possible for this to fail if 1776 * this code were racing with dtrace_dynvar_clean(). 1777 * (That is, if dtrace_dynvar_clean() updated the clean 1778 * list, and we ended up racing to update the free 1779 * list.) This race is prevented by the dtrace_sync() 1780 * in dtrace_dynvar_clean() -- which flushes the 1781 * owners of the clean lists out before resetting 1782 * the clean lists. 1783 */ 1784 rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean); 1785 ASSERT(rval == NULL); 1786 goto retry; 1787 } 1788 1789 dvar = free; 1790 new_free = dvar->dtdv_next; 1791 } while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free); 1792 1793 /* 1794 * We have now allocated a new chunk. We copy the tuple keys into the 1795 * tuple array and copy any referenced key data into the data space 1796 * following the tuple array. As we do this, we relocate dttk_value 1797 * in the final tuple to point to the key data address in the chunk. 1798 */ 1799 kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys]; 1800 dvar->dtdv_data = (void *)(kdata + ksize); 1801 dvar->dtdv_tuple.dtt_nkeys = nkeys; 1802 1803 for (i = 0; i < nkeys; i++) { 1804 dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i]; 1805 size_t kesize = key[i].dttk_size; 1806 1807 if (kesize != 0) { 1808 dtrace_bcopy( 1809 (const void *)(uintptr_t)key[i].dttk_value, 1810 (void *)kdata, kesize); 1811 dkey->dttk_value = kdata; 1812 kdata += P2ROUNDUP(kesize, sizeof (uint64_t)); 1813 } else { 1814 dkey->dttk_value = key[i].dttk_value; 1815 } 1816 1817 dkey->dttk_size = kesize; 1818 } 1819 1820 ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE); 1821 dvar->dtdv_hashval = hashval; 1822 dvar->dtdv_next = start; 1823 1824 if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start) 1825 return (dvar); 1826 1827 /* 1828 * The cas has failed. Either another CPU is adding an element to 1829 * this hash chain, or another CPU is deleting an element from this 1830 * hash chain. The simplest way to deal with both of these cases 1831 * (though not necessarily the most efficient) is to free our 1832 * allocated block and tail-call ourselves. Note that the free is 1833 * to the dirty list and _not_ to the free list. This is to prevent 1834 * races with allocators, above. 1835 */ 1836 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; 1837 1838 dtrace_membar_producer(); 1839 1840 do { 1841 free = dcpu->dtdsc_dirty; 1842 dvar->dtdv_next = free; 1843 } while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free); 1844 1845 return (dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate)); 1846 } 1847 1848 /*ARGSUSED*/ 1849 static void 1850 dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg) 1851 { 1852 if ((int64_t)nval < (int64_t)*oval) 1853 *oval = nval; 1854 } 1855 1856 /*ARGSUSED*/ 1857 static void 1858 dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg) 1859 { 1860 if ((int64_t)nval > (int64_t)*oval) 1861 *oval = nval; 1862 } 1863 1864 static void 1865 dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr) 1866 { 1867 int i, zero = DTRACE_QUANTIZE_ZEROBUCKET; 1868 int64_t val = (int64_t)nval; 1869 1870 if (val < 0) { 1871 for (i = 0; i < zero; i++) { 1872 if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) { 1873 quanta[i] += incr; 1874 return; 1875 } 1876 } 1877 } else { 1878 for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) { 1879 if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) { 1880 quanta[i - 1] += incr; 1881 return; 1882 } 1883 } 1884 1885 quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr; 1886 return; 1887 } 1888 1889 ASSERT(0); 1890 } 1891 1892 static void 1893 dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr) 1894 { 1895 uint64_t arg = *lquanta++; 1896 int32_t base = DTRACE_LQUANTIZE_BASE(arg); 1897 uint16_t step = DTRACE_LQUANTIZE_STEP(arg); 1898 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg); 1899 int32_t val = (int32_t)nval, level; 1900 1901 ASSERT(step != 0); 1902 ASSERT(levels != 0); 1903 1904 if (val < base) { 1905 /* 1906 * This is an underflow. 1907 */ 1908 lquanta[0] += incr; 1909 return; 1910 } 1911 1912 level = (val - base) / step; 1913 1914 if (level < levels) { 1915 lquanta[level + 1] += incr; 1916 return; 1917 } 1918 1919 /* 1920 * This is an overflow. 1921 */ 1922 lquanta[levels + 1] += incr; 1923 } 1924 1925 static int 1926 dtrace_aggregate_llquantize_bucket(uint16_t factor, uint16_t low, 1927 uint16_t high, uint16_t nsteps, int64_t value) 1928 { 1929 int64_t this = 1, last, next; 1930 int base = 1, order; 1931 1932 ASSERT(factor <= nsteps); 1933 ASSERT(nsteps % factor == 0); 1934 1935 for (order = 0; order < low; order++) 1936 this *= factor; 1937 1938 /* 1939 * If our value is less than our factor taken to the power of the 1940 * low order of magnitude, it goes into the zeroth bucket. 1941 */ 1942 if (value < (last = this)) 1943 return (0); 1944 1945 for (this *= factor; order <= high; order++) { 1946 int nbuckets = this > nsteps ? nsteps : this; 1947 1948 if ((next = this * factor) < this) { 1949 /* 1950 * We should not generally get log/linear quantizations 1951 * with a high magnitude that allows 64-bits to 1952 * overflow, but we nonetheless protect against this 1953 * by explicitly checking for overflow, and clamping 1954 * our value accordingly. 1955 */ 1956 value = this - 1; 1957 } 1958 1959 if (value < this) { 1960 /* 1961 * If our value lies within this order of magnitude, 1962 * determine its position by taking the offset within 1963 * the order of magnitude, dividing by the bucket 1964 * width, and adding to our (accumulated) base. 1965 */ 1966 return (base + (value - last) / (this / nbuckets)); 1967 } 1968 1969 base += nbuckets - (nbuckets / factor); 1970 last = this; 1971 this = next; 1972 } 1973 1974 /* 1975 * Our value is greater than or equal to our factor taken to the 1976 * power of one plus the high magnitude -- return the top bucket. 1977 */ 1978 return (base); 1979 } 1980 1981 static void 1982 dtrace_aggregate_llquantize(uint64_t *llquanta, uint64_t nval, uint64_t incr) 1983 { 1984 uint64_t arg = *llquanta++; 1985 uint16_t factor = DTRACE_LLQUANTIZE_FACTOR(arg); 1986 uint16_t low = DTRACE_LLQUANTIZE_LOW(arg); 1987 uint16_t high = DTRACE_LLQUANTIZE_HIGH(arg); 1988 uint16_t nsteps = DTRACE_LLQUANTIZE_NSTEP(arg); 1989 1990 llquanta[dtrace_aggregate_llquantize_bucket(factor, 1991 low, high, nsteps, nval)] += incr; 1992 } 1993 1994 /*ARGSUSED*/ 1995 static void 1996 dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg) 1997 { 1998 data[0]++; 1999 data[1] += nval; 2000 } 2001 2002 /*ARGSUSED*/ 2003 static void 2004 dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg) 2005 { 2006 int64_t snval = (int64_t)nval; 2007 uint64_t tmp[2]; 2008 2009 data[0]++; 2010 data[1] += nval; 2011 2012 /* 2013 * What we want to say here is: 2014 * 2015 * data[2] += nval * nval; 2016 * 2017 * But given that nval is 64-bit, we could easily overflow, so 2018 * we do this as 128-bit arithmetic. 2019 */ 2020 if (snval < 0) 2021 snval = -snval; 2022 2023 dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp); 2024 dtrace_add_128(data + 2, tmp, data + 2); 2025 } 2026 2027 /*ARGSUSED*/ 2028 static void 2029 dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg) 2030 { 2031 *oval = *oval + 1; 2032 } 2033 2034 /*ARGSUSED*/ 2035 static void 2036 dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg) 2037 { 2038 *oval += nval; 2039 } 2040 2041 /* 2042 * Aggregate given the tuple in the principal data buffer, and the aggregating 2043 * action denoted by the specified dtrace_aggregation_t. The aggregation 2044 * buffer is specified as the buf parameter. This routine does not return 2045 * failure; if there is no space in the aggregation buffer, the data will be 2046 * dropped, and a corresponding counter incremented. 2047 */ 2048 static void 2049 dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf, 2050 intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg) 2051 { 2052 dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec; 2053 uint32_t i, ndx, size, fsize; 2054 uint32_t align = sizeof (uint64_t) - 1; 2055 dtrace_aggbuffer_t *agb; 2056 dtrace_aggkey_t *key; 2057 uint32_t hashval = 0, limit, isstr; 2058 caddr_t tomax, data, kdata; 2059 dtrace_actkind_t action; 2060 dtrace_action_t *act; 2061 uintptr_t offs; 2062 2063 if (buf == NULL) 2064 return; 2065 2066 if (!agg->dtag_hasarg) { 2067 /* 2068 * Currently, only quantize() and lquantize() take additional 2069 * arguments, and they have the same semantics: an increment 2070 * value that defaults to 1 when not present. If additional 2071 * aggregating actions take arguments, the setting of the 2072 * default argument value will presumably have to become more 2073 * sophisticated... 2074 */ 2075 arg = 1; 2076 } 2077 2078 action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION; 2079 size = rec->dtrd_offset - agg->dtag_base; 2080 fsize = size + rec->dtrd_size; 2081 2082 ASSERT(dbuf->dtb_tomax != NULL); 2083 data = dbuf->dtb_tomax + offset + agg->dtag_base; 2084 2085 if ((tomax = buf->dtb_tomax) == NULL) { 2086 dtrace_buffer_drop(buf); 2087 return; 2088 } 2089 2090 /* 2091 * The metastructure is always at the bottom of the buffer. 2092 */ 2093 agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size - 2094 sizeof (dtrace_aggbuffer_t)); 2095 2096 if (buf->dtb_offset == 0) { 2097 /* 2098 * We just kludge up approximately 1/8th of the size to be 2099 * buckets. If this guess ends up being routinely 2100 * off-the-mark, we may need to dynamically readjust this 2101 * based on past performance. 2102 */ 2103 uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t); 2104 2105 if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) < 2106 (uintptr_t)tomax || hashsize == 0) { 2107 /* 2108 * We've been given a ludicrously small buffer; 2109 * increment our drop count and leave. 2110 */ 2111 dtrace_buffer_drop(buf); 2112 return; 2113 } 2114 2115 /* 2116 * And now, a pathetic attempt to try to get a an odd (or 2117 * perchance, a prime) hash size for better hash distribution. 2118 */ 2119 if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3)) 2120 hashsize -= DTRACE_AGGHASHSIZE_SLEW; 2121 2122 agb->dtagb_hashsize = hashsize; 2123 agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb - 2124 agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *)); 2125 agb->dtagb_free = (uintptr_t)agb->dtagb_hash; 2126 2127 for (i = 0; i < agb->dtagb_hashsize; i++) 2128 agb->dtagb_hash[i] = NULL; 2129 } 2130 2131 ASSERT(agg->dtag_first != NULL); 2132 ASSERT(agg->dtag_first->dta_intuple); 2133 2134 /* 2135 * Calculate the hash value based on the key. Note that we _don't_ 2136 * include the aggid in the hashing (but we will store it as part of 2137 * the key). The hashing algorithm is Bob Jenkins' "One-at-a-time" 2138 * algorithm: a simple, quick algorithm that has no known funnels, and 2139 * gets good distribution in practice. The efficacy of the hashing 2140 * algorithm (and a comparison with other algorithms) may be found by 2141 * running the ::dtrace_aggstat MDB dcmd. 2142 */ 2143 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { 2144 i = act->dta_rec.dtrd_offset - agg->dtag_base; 2145 limit = i + act->dta_rec.dtrd_size; 2146 ASSERT(limit <= size); 2147 isstr = DTRACEACT_ISSTRING(act); 2148 2149 for (; i < limit; i++) { 2150 hashval += data[i]; 2151 hashval += (hashval << 10); 2152 hashval ^= (hashval >> 6); 2153 2154 if (isstr && data[i] == '\0') 2155 break; 2156 } 2157 } 2158 2159 hashval += (hashval << 3); 2160 hashval ^= (hashval >> 11); 2161 hashval += (hashval << 15); 2162 2163 /* 2164 * Yes, the divide here is expensive -- but it's generally the least 2165 * of the performance issues given the amount of data that we iterate 2166 * over to compute hash values, compare data, etc. 2167 */ 2168 ndx = hashval % agb->dtagb_hashsize; 2169 2170 for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) { 2171 ASSERT((caddr_t)key >= tomax); 2172 ASSERT((caddr_t)key < tomax + buf->dtb_size); 2173 2174 if (hashval != key->dtak_hashval || key->dtak_size != size) 2175 continue; 2176 2177 kdata = key->dtak_data; 2178 ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size); 2179 2180 for (act = agg->dtag_first; act->dta_intuple; 2181 act = act->dta_next) { 2182 i = act->dta_rec.dtrd_offset - agg->dtag_base; 2183 limit = i + act->dta_rec.dtrd_size; 2184 ASSERT(limit <= size); 2185 isstr = DTRACEACT_ISSTRING(act); 2186 2187 for (; i < limit; i++) { 2188 if (kdata[i] != data[i]) 2189 goto next; 2190 2191 if (isstr && data[i] == '\0') 2192 break; 2193 } 2194 } 2195 2196 if (action != key->dtak_action) { 2197 /* 2198 * We are aggregating on the same value in the same 2199 * aggregation with two different aggregating actions. 2200 * (This should have been picked up in the compiler, 2201 * so we may be dealing with errant or devious DIF.) 2202 * This is an error condition; we indicate as much, 2203 * and return. 2204 */ 2205 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 2206 return; 2207 } 2208 2209 /* 2210 * This is a hit: we need to apply the aggregator to 2211 * the value at this key. 2212 */ 2213 agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg); 2214 return; 2215 next: 2216 continue; 2217 } 2218 2219 /* 2220 * We didn't find it. We need to allocate some zero-filled space, 2221 * link it into the hash table appropriately, and apply the aggregator 2222 * to the (zero-filled) value. 2223 */ 2224 offs = buf->dtb_offset; 2225 while (offs & (align - 1)) 2226 offs += sizeof (uint32_t); 2227 2228 /* 2229 * If we don't have enough room to both allocate a new key _and_ 2230 * its associated data, increment the drop count and return. 2231 */ 2232 if ((uintptr_t)tomax + offs + fsize > 2233 agb->dtagb_free - sizeof (dtrace_aggkey_t)) { 2234 dtrace_buffer_drop(buf); 2235 return; 2236 } 2237 2238 /*CONSTCOND*/ 2239 ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1))); 2240 key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t)); 2241 agb->dtagb_free -= sizeof (dtrace_aggkey_t); 2242 2243 key->dtak_data = kdata = tomax + offs; 2244 buf->dtb_offset = offs + fsize; 2245 2246 /* 2247 * Now copy the data across. 2248 */ 2249 *((dtrace_aggid_t *)kdata) = agg->dtag_id; 2250 2251 for (i = sizeof (dtrace_aggid_t); i < size; i++) 2252 kdata[i] = data[i]; 2253 2254 /* 2255 * Because strings are not zeroed out by default, we need to iterate 2256 * looking for actions that store strings, and we need to explicitly 2257 * pad these strings out with zeroes. 2258 */ 2259 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { 2260 int nul; 2261 2262 if (!DTRACEACT_ISSTRING(act)) 2263 continue; 2264 2265 i = act->dta_rec.dtrd_offset - agg->dtag_base; 2266 limit = i + act->dta_rec.dtrd_size; 2267 ASSERT(limit <= size); 2268 2269 for (nul = 0; i < limit; i++) { 2270 if (nul) { 2271 kdata[i] = '\0'; 2272 continue; 2273 } 2274 2275 if (data[i] != '\0') 2276 continue; 2277 2278 nul = 1; 2279 } 2280 } 2281 2282 for (i = size; i < fsize; i++) 2283 kdata[i] = 0; 2284 2285 key->dtak_hashval = hashval; 2286 key->dtak_size = size; 2287 key->dtak_action = action; 2288 key->dtak_next = agb->dtagb_hash[ndx]; 2289 agb->dtagb_hash[ndx] = key; 2290 2291 /* 2292 * Finally, apply the aggregator. 2293 */ 2294 *((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial; 2295 agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg); 2296 } 2297 2298 /* 2299 * Given consumer state, this routine finds a speculation in the INACTIVE 2300 * state and transitions it into the ACTIVE state. If there is no speculation 2301 * in the INACTIVE state, 0 is returned. In this case, no error counter is 2302 * incremented -- it is up to the caller to take appropriate action. 2303 */ 2304 static int 2305 dtrace_speculation(dtrace_state_t *state) 2306 { 2307 int i = 0; 2308 dtrace_speculation_state_t current; 2309 uint32_t *stat = &state->dts_speculations_unavail, count; 2310 2311 while (i < state->dts_nspeculations) { 2312 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2313 2314 current = spec->dtsp_state; 2315 2316 if (current != DTRACESPEC_INACTIVE) { 2317 if (current == DTRACESPEC_COMMITTINGMANY || 2318 current == DTRACESPEC_COMMITTING || 2319 current == DTRACESPEC_DISCARDING) 2320 stat = &state->dts_speculations_busy; 2321 i++; 2322 continue; 2323 } 2324 2325 if (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2326 current, DTRACESPEC_ACTIVE) == current) 2327 return (i + 1); 2328 } 2329 2330 /* 2331 * We couldn't find a speculation. If we found as much as a single 2332 * busy speculation buffer, we'll attribute this failure as "busy" 2333 * instead of "unavail". 2334 */ 2335 do { 2336 count = *stat; 2337 } while (dtrace_cas32(stat, count, count + 1) != count); 2338 2339 return (0); 2340 } 2341 2342 /* 2343 * This routine commits an active speculation. If the specified speculation 2344 * is not in a valid state to perform a commit(), this routine will silently do 2345 * nothing. The state of the specified speculation is transitioned according 2346 * to the state transition diagram outlined in <sys/dtrace_impl.h> 2347 */ 2348 static void 2349 dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu, 2350 dtrace_specid_t which) 2351 { 2352 dtrace_speculation_t *spec; 2353 dtrace_buffer_t *src, *dest; 2354 uintptr_t daddr, saddr, dlimit, slimit; 2355 dtrace_speculation_state_t current, new = 0; 2356 intptr_t offs; 2357 uint64_t timestamp; 2358 2359 if (which == 0) 2360 return; 2361 2362 if (which > state->dts_nspeculations) { 2363 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 2364 return; 2365 } 2366 2367 spec = &state->dts_speculations[which - 1]; 2368 src = &spec->dtsp_buffer[cpu]; 2369 dest = &state->dts_buffer[cpu]; 2370 2371 do { 2372 current = spec->dtsp_state; 2373 2374 if (current == DTRACESPEC_COMMITTINGMANY) 2375 break; 2376 2377 switch (current) { 2378 case DTRACESPEC_INACTIVE: 2379 case DTRACESPEC_DISCARDING: 2380 return; 2381 2382 case DTRACESPEC_COMMITTING: 2383 /* 2384 * This is only possible if we are (a) commit()'ing 2385 * without having done a prior speculate() on this CPU 2386 * and (b) racing with another commit() on a different 2387 * CPU. There's nothing to do -- we just assert that 2388 * our offset is 0. 2389 */ 2390 ASSERT(src->dtb_offset == 0); 2391 return; 2392 2393 case DTRACESPEC_ACTIVE: 2394 new = DTRACESPEC_COMMITTING; 2395 break; 2396 2397 case DTRACESPEC_ACTIVEONE: 2398 /* 2399 * This speculation is active on one CPU. If our 2400 * buffer offset is non-zero, we know that the one CPU 2401 * must be us. Otherwise, we are committing on a 2402 * different CPU from the speculate(), and we must 2403 * rely on being asynchronously cleaned. 2404 */ 2405 if (src->dtb_offset != 0) { 2406 new = DTRACESPEC_COMMITTING; 2407 break; 2408 } 2409 /*FALLTHROUGH*/ 2410 2411 case DTRACESPEC_ACTIVEMANY: 2412 new = DTRACESPEC_COMMITTINGMANY; 2413 break; 2414 2415 default: 2416 ASSERT(0); 2417 } 2418 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2419 current, new) != current); 2420 2421 /* 2422 * We have set the state to indicate that we are committing this 2423 * speculation. Now reserve the necessary space in the destination 2424 * buffer. 2425 */ 2426 if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset, 2427 sizeof (uint64_t), state, NULL)) < 0) { 2428 dtrace_buffer_drop(dest); 2429 goto out; 2430 } 2431 2432 /* 2433 * We have sufficient space to copy the speculative buffer into the 2434 * primary buffer. First, modify the speculative buffer, filling 2435 * in the timestamp of all entries with the current time. The data 2436 * must have the commit() time rather than the time it was traced, 2437 * so that all entries in the primary buffer are in timestamp order. 2438 */ 2439 timestamp = dtrace_gethrtime(); 2440 saddr = (uintptr_t)src->dtb_tomax; 2441 slimit = saddr + src->dtb_offset; 2442 while (saddr < slimit) { 2443 size_t size; 2444 dtrace_rechdr_t *dtrh = (dtrace_rechdr_t *)saddr; 2445 2446 if (dtrh->dtrh_epid == DTRACE_EPIDNONE) { 2447 saddr += sizeof (dtrace_epid_t); 2448 continue; 2449 } 2450 ASSERT3U(dtrh->dtrh_epid, <=, state->dts_necbs); 2451 size = state->dts_ecbs[dtrh->dtrh_epid - 1]->dte_size; 2452 2453 ASSERT3U(saddr + size, <=, slimit); 2454 ASSERT3U(size, >=, sizeof (dtrace_rechdr_t)); 2455 ASSERT3U(DTRACE_RECORD_LOAD_TIMESTAMP(dtrh), ==, UINT64_MAX); 2456 2457 DTRACE_RECORD_STORE_TIMESTAMP(dtrh, timestamp); 2458 2459 saddr += size; 2460 } 2461 2462 /* 2463 * Copy the buffer across. (Note that this is a 2464 * highly subobtimal bcopy(); in the unlikely event that this becomes 2465 * a serious performance issue, a high-performance DTrace-specific 2466 * bcopy() should obviously be invented.) 2467 */ 2468 daddr = (uintptr_t)dest->dtb_tomax + offs; 2469 dlimit = daddr + src->dtb_offset; 2470 saddr = (uintptr_t)src->dtb_tomax; 2471 2472 /* 2473 * First, the aligned portion. 2474 */ 2475 while (dlimit - daddr >= sizeof (uint64_t)) { 2476 *((uint64_t *)daddr) = *((uint64_t *)saddr); 2477 2478 daddr += sizeof (uint64_t); 2479 saddr += sizeof (uint64_t); 2480 } 2481 2482 /* 2483 * Now any left-over bit... 2484 */ 2485 while (dlimit - daddr) 2486 *((uint8_t *)daddr++) = *((uint8_t *)saddr++); 2487 2488 /* 2489 * Finally, commit the reserved space in the destination buffer. 2490 */ 2491 dest->dtb_offset = offs + src->dtb_offset; 2492 2493 out: 2494 /* 2495 * If we're lucky enough to be the only active CPU on this speculation 2496 * buffer, we can just set the state back to DTRACESPEC_INACTIVE. 2497 */ 2498 if (current == DTRACESPEC_ACTIVE || 2499 (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) { 2500 uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state, 2501 DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE); 2502 2503 ASSERT(rval == DTRACESPEC_COMMITTING); 2504 } 2505 2506 src->dtb_offset = 0; 2507 src->dtb_xamot_drops += src->dtb_drops; 2508 src->dtb_drops = 0; 2509 } 2510 2511 /* 2512 * This routine discards an active speculation. If the specified speculation 2513 * is not in a valid state to perform a discard(), this routine will silently 2514 * do nothing. The state of the specified speculation is transitioned 2515 * according to the state transition diagram outlined in <sys/dtrace_impl.h> 2516 */ 2517 static void 2518 dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu, 2519 dtrace_specid_t which) 2520 { 2521 dtrace_speculation_t *spec; 2522 dtrace_speculation_state_t current, new = 0; 2523 dtrace_buffer_t *buf; 2524 2525 if (which == 0) 2526 return; 2527 2528 if (which > state->dts_nspeculations) { 2529 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 2530 return; 2531 } 2532 2533 spec = &state->dts_speculations[which - 1]; 2534 buf = &spec->dtsp_buffer[cpu]; 2535 2536 do { 2537 current = spec->dtsp_state; 2538 2539 switch (current) { 2540 case DTRACESPEC_INACTIVE: 2541 case DTRACESPEC_COMMITTINGMANY: 2542 case DTRACESPEC_COMMITTING: 2543 case DTRACESPEC_DISCARDING: 2544 return; 2545 2546 case DTRACESPEC_ACTIVE: 2547 case DTRACESPEC_ACTIVEMANY: 2548 new = DTRACESPEC_DISCARDING; 2549 break; 2550 2551 case DTRACESPEC_ACTIVEONE: 2552 if (buf->dtb_offset != 0) { 2553 new = DTRACESPEC_INACTIVE; 2554 } else { 2555 new = DTRACESPEC_DISCARDING; 2556 } 2557 break; 2558 2559 default: 2560 ASSERT(0); 2561 } 2562 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2563 current, new) != current); 2564 2565 buf->dtb_offset = 0; 2566 buf->dtb_drops = 0; 2567 } 2568 2569 /* 2570 * Note: not called from probe context. This function is called 2571 * asynchronously from cross call context to clean any speculations that are 2572 * in the COMMITTINGMANY or DISCARDING states. These speculations may not be 2573 * transitioned back to the INACTIVE state until all CPUs have cleaned the 2574 * speculation. 2575 */ 2576 static void 2577 dtrace_speculation_clean_here(dtrace_state_t *state) 2578 { 2579 dtrace_icookie_t cookie; 2580 processorid_t cpu = curcpu; 2581 dtrace_buffer_t *dest = &state->dts_buffer[cpu]; 2582 dtrace_specid_t i; 2583 2584 cookie = dtrace_interrupt_disable(); 2585 2586 if (dest->dtb_tomax == NULL) { 2587 dtrace_interrupt_enable(cookie); 2588 return; 2589 } 2590 2591 for (i = 0; i < state->dts_nspeculations; i++) { 2592 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2593 dtrace_buffer_t *src = &spec->dtsp_buffer[cpu]; 2594 2595 if (src->dtb_tomax == NULL) 2596 continue; 2597 2598 if (spec->dtsp_state == DTRACESPEC_DISCARDING) { 2599 src->dtb_offset = 0; 2600 continue; 2601 } 2602 2603 if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) 2604 continue; 2605 2606 if (src->dtb_offset == 0) 2607 continue; 2608 2609 dtrace_speculation_commit(state, cpu, i + 1); 2610 } 2611 2612 dtrace_interrupt_enable(cookie); 2613 } 2614 2615 /* 2616 * Note: not called from probe context. This function is called 2617 * asynchronously (and at a regular interval) to clean any speculations that 2618 * are in the COMMITTINGMANY or DISCARDING states. If it discovers that there 2619 * is work to be done, it cross calls all CPUs to perform that work; 2620 * COMMITMANY and DISCARDING speculations may not be transitioned back to the 2621 * INACTIVE state until they have been cleaned by all CPUs. 2622 */ 2623 static void 2624 dtrace_speculation_clean(dtrace_state_t *state) 2625 { 2626 int work = 0, rv; 2627 dtrace_specid_t i; 2628 2629 for (i = 0; i < state->dts_nspeculations; i++) { 2630 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2631 2632 ASSERT(!spec->dtsp_cleaning); 2633 2634 if (spec->dtsp_state != DTRACESPEC_DISCARDING && 2635 spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) 2636 continue; 2637 2638 work++; 2639 spec->dtsp_cleaning = 1; 2640 } 2641 2642 if (!work) 2643 return; 2644 2645 dtrace_xcall(DTRACE_CPUALL, 2646 (dtrace_xcall_t)dtrace_speculation_clean_here, state); 2647 2648 /* 2649 * We now know that all CPUs have committed or discarded their 2650 * speculation buffers, as appropriate. We can now set the state 2651 * to inactive. 2652 */ 2653 for (i = 0; i < state->dts_nspeculations; i++) { 2654 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2655 dtrace_speculation_state_t current, new; 2656 2657 if (!spec->dtsp_cleaning) 2658 continue; 2659 2660 current = spec->dtsp_state; 2661 ASSERT(current == DTRACESPEC_DISCARDING || 2662 current == DTRACESPEC_COMMITTINGMANY); 2663 2664 new = DTRACESPEC_INACTIVE; 2665 2666 rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new); 2667 ASSERT(rv == current); 2668 spec->dtsp_cleaning = 0; 2669 } 2670 } 2671 2672 /* 2673 * Called as part of a speculate() to get the speculative buffer associated 2674 * with a given speculation. Returns NULL if the specified speculation is not 2675 * in an ACTIVE state. If the speculation is in the ACTIVEONE state -- and 2676 * the active CPU is not the specified CPU -- the speculation will be 2677 * atomically transitioned into the ACTIVEMANY state. 2678 */ 2679 static dtrace_buffer_t * 2680 dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid, 2681 dtrace_specid_t which) 2682 { 2683 dtrace_speculation_t *spec; 2684 dtrace_speculation_state_t current, new = 0; 2685 dtrace_buffer_t *buf; 2686 2687 if (which == 0) 2688 return (NULL); 2689 2690 if (which > state->dts_nspeculations) { 2691 cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 2692 return (NULL); 2693 } 2694 2695 spec = &state->dts_speculations[which - 1]; 2696 buf = &spec->dtsp_buffer[cpuid]; 2697 2698 do { 2699 current = spec->dtsp_state; 2700 2701 switch (current) { 2702 case DTRACESPEC_INACTIVE: 2703 case DTRACESPEC_COMMITTINGMANY: 2704 case DTRACESPEC_DISCARDING: 2705 return (NULL); 2706 2707 case DTRACESPEC_COMMITTING: 2708 ASSERT(buf->dtb_offset == 0); 2709 return (NULL); 2710 2711 case DTRACESPEC_ACTIVEONE: 2712 /* 2713 * This speculation is currently active on one CPU. 2714 * Check the offset in the buffer; if it's non-zero, 2715 * that CPU must be us (and we leave the state alone). 2716 * If it's zero, assume that we're starting on a new 2717 * CPU -- and change the state to indicate that the 2718 * speculation is active on more than one CPU. 2719 */ 2720 if (buf->dtb_offset != 0) 2721 return (buf); 2722 2723 new = DTRACESPEC_ACTIVEMANY; 2724 break; 2725 2726 case DTRACESPEC_ACTIVEMANY: 2727 return (buf); 2728 2729 case DTRACESPEC_ACTIVE: 2730 new = DTRACESPEC_ACTIVEONE; 2731 break; 2732 2733 default: 2734 ASSERT(0); 2735 } 2736 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2737 current, new) != current); 2738 2739 ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY); 2740 return (buf); 2741 } 2742 2743 /* 2744 * Return a string. In the event that the user lacks the privilege to access 2745 * arbitrary kernel memory, we copy the string out to scratch memory so that we 2746 * don't fail access checking. 2747 * 2748 * dtrace_dif_variable() uses this routine as a helper for various 2749 * builtin values such as 'execname' and 'probefunc.' 2750 */ 2751 uintptr_t 2752 dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state, 2753 dtrace_mstate_t *mstate) 2754 { 2755 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 2756 uintptr_t ret; 2757 size_t strsz; 2758 2759 /* 2760 * The easy case: this probe is allowed to read all of memory, so 2761 * we can just return this as a vanilla pointer. 2762 */ 2763 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 2764 return (addr); 2765 2766 /* 2767 * This is the tougher case: we copy the string in question from 2768 * kernel memory into scratch memory and return it that way: this 2769 * ensures that we won't trip up when access checking tests the 2770 * BYREF return value. 2771 */ 2772 strsz = dtrace_strlen((char *)addr, size) + 1; 2773 2774 if (mstate->dtms_scratch_ptr + strsz > 2775 mstate->dtms_scratch_base + mstate->dtms_scratch_size) { 2776 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 2777 return (0); 2778 } 2779 2780 dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr, 2781 strsz); 2782 ret = mstate->dtms_scratch_ptr; 2783 mstate->dtms_scratch_ptr += strsz; 2784 return (ret); 2785 } 2786 2787 /* 2788 * Return a string from a memoy address which is known to have one or 2789 * more concatenated, individually zero terminated, sub-strings. 2790 * In the event that the user lacks the privilege to access 2791 * arbitrary kernel memory, we copy the string out to scratch memory so that we 2792 * don't fail access checking. 2793 * 2794 * dtrace_dif_variable() uses this routine as a helper for various 2795 * builtin values such as 'execargs'. 2796 */ 2797 static uintptr_t 2798 dtrace_dif_varstrz(uintptr_t addr, size_t strsz, dtrace_state_t *state, 2799 dtrace_mstate_t *mstate) 2800 { 2801 char *p; 2802 size_t i; 2803 uintptr_t ret; 2804 2805 if (mstate->dtms_scratch_ptr + strsz > 2806 mstate->dtms_scratch_base + mstate->dtms_scratch_size) { 2807 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 2808 return (0); 2809 } 2810 2811 dtrace_bcopy((const void *)addr, (void *)mstate->dtms_scratch_ptr, 2812 strsz); 2813 2814 /* Replace sub-string termination characters with a space. */ 2815 for (p = (char *) mstate->dtms_scratch_ptr, i = 0; i < strsz - 1; 2816 p++, i++) 2817 if (*p == '\0') 2818 *p = ' '; 2819 2820 ret = mstate->dtms_scratch_ptr; 2821 mstate->dtms_scratch_ptr += strsz; 2822 return (ret); 2823 } 2824 2825 /* 2826 * This function implements the DIF emulator's variable lookups. The emulator 2827 * passes a reserved variable identifier and optional built-in array index. 2828 */ 2829 static uint64_t 2830 dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v, 2831 uint64_t ndx) 2832 { 2833 /* 2834 * If we're accessing one of the uncached arguments, we'll turn this 2835 * into a reference in the args array. 2836 */ 2837 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) { 2838 ndx = v - DIF_VAR_ARG0; 2839 v = DIF_VAR_ARGS; 2840 } 2841 2842 switch (v) { 2843 case DIF_VAR_ARGS: 2844 ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS); 2845 if (ndx >= sizeof (mstate->dtms_arg) / 2846 sizeof (mstate->dtms_arg[0])) { 2847 int aframes = mstate->dtms_probe->dtpr_aframes + 2; 2848 dtrace_provider_t *pv; 2849 uint64_t val; 2850 2851 pv = mstate->dtms_probe->dtpr_provider; 2852 if (pv->dtpv_pops.dtps_getargval != NULL) 2853 val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg, 2854 mstate->dtms_probe->dtpr_id, 2855 mstate->dtms_probe->dtpr_arg, ndx, aframes); 2856 else 2857 val = dtrace_getarg(ndx, aframes); 2858 2859 /* 2860 * This is regrettably required to keep the compiler 2861 * from tail-optimizing the call to dtrace_getarg(). 2862 * The condition always evaluates to true, but the 2863 * compiler has no way of figuring that out a priori. 2864 * (None of this would be necessary if the compiler 2865 * could be relied upon to _always_ tail-optimize 2866 * the call to dtrace_getarg() -- but it can't.) 2867 */ 2868 if (mstate->dtms_probe != NULL) 2869 return (val); 2870 2871 ASSERT(0); 2872 } 2873 2874 return (mstate->dtms_arg[ndx]); 2875 2876 #if defined(sun) 2877 case DIF_VAR_UREGS: { 2878 klwp_t *lwp; 2879 2880 if (!dtrace_priv_proc(state)) 2881 return (0); 2882 2883 if ((lwp = curthread->t_lwp) == NULL) { 2884 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); 2885 cpu_core[curcpu].cpuc_dtrace_illval = NULL; 2886 return (0); 2887 } 2888 2889 return (dtrace_getreg(lwp->lwp_regs, ndx)); 2890 return (0); 2891 } 2892 #else 2893 case DIF_VAR_UREGS: { 2894 struct trapframe *tframe; 2895 2896 if (!dtrace_priv_proc(state)) 2897 return (0); 2898 2899 if ((tframe = curthread->td_frame) == NULL) { 2900 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); 2901 cpu_core[curcpu].cpuc_dtrace_illval = 0; 2902 return (0); 2903 } 2904 2905 return (dtrace_getreg(tframe, ndx)); 2906 } 2907 #endif 2908 2909 case DIF_VAR_CURTHREAD: 2910 if (!dtrace_priv_kernel(state)) 2911 return (0); 2912 return ((uint64_t)(uintptr_t)curthread); 2913 2914 case DIF_VAR_TIMESTAMP: 2915 if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) { 2916 mstate->dtms_timestamp = dtrace_gethrtime(); 2917 mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP; 2918 } 2919 return (mstate->dtms_timestamp); 2920 2921 case DIF_VAR_VTIMESTAMP: 2922 ASSERT(dtrace_vtime_references != 0); 2923 return (curthread->t_dtrace_vtime); 2924 2925 case DIF_VAR_WALLTIMESTAMP: 2926 if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) { 2927 mstate->dtms_walltimestamp = dtrace_gethrestime(); 2928 mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP; 2929 } 2930 return (mstate->dtms_walltimestamp); 2931 2932 #if defined(sun) 2933 case DIF_VAR_IPL: 2934 if (!dtrace_priv_kernel(state)) 2935 return (0); 2936 if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) { 2937 mstate->dtms_ipl = dtrace_getipl(); 2938 mstate->dtms_present |= DTRACE_MSTATE_IPL; 2939 } 2940 return (mstate->dtms_ipl); 2941 #endif 2942 2943 case DIF_VAR_EPID: 2944 ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID); 2945 return (mstate->dtms_epid); 2946 2947 case DIF_VAR_ID: 2948 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 2949 return (mstate->dtms_probe->dtpr_id); 2950 2951 case DIF_VAR_STACKDEPTH: 2952 if (!dtrace_priv_kernel(state)) 2953 return (0); 2954 if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) { 2955 int aframes = mstate->dtms_probe->dtpr_aframes + 2; 2956 2957 mstate->dtms_stackdepth = dtrace_getstackdepth(aframes); 2958 mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH; 2959 } 2960 return (mstate->dtms_stackdepth); 2961 2962 case DIF_VAR_USTACKDEPTH: 2963 if (!dtrace_priv_proc(state)) 2964 return (0); 2965 if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) { 2966 /* 2967 * See comment in DIF_VAR_PID. 2968 */ 2969 if (DTRACE_ANCHORED(mstate->dtms_probe) && 2970 CPU_ON_INTR(CPU)) { 2971 mstate->dtms_ustackdepth = 0; 2972 } else { 2973 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 2974 mstate->dtms_ustackdepth = 2975 dtrace_getustackdepth(); 2976 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 2977 } 2978 mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH; 2979 } 2980 return (mstate->dtms_ustackdepth); 2981 2982 case DIF_VAR_CALLER: 2983 if (!dtrace_priv_kernel(state)) 2984 return (0); 2985 if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) { 2986 int aframes = mstate->dtms_probe->dtpr_aframes + 2; 2987 2988 if (!DTRACE_ANCHORED(mstate->dtms_probe)) { 2989 /* 2990 * If this is an unanchored probe, we are 2991 * required to go through the slow path: 2992 * dtrace_caller() only guarantees correct 2993 * results for anchored probes. 2994 */ 2995 pc_t caller[2] = {0, 0}; 2996 2997 dtrace_getpcstack(caller, 2, aframes, 2998 (uint32_t *)(uintptr_t)mstate->dtms_arg[0]); 2999 mstate->dtms_caller = caller[1]; 3000 } else if ((mstate->dtms_caller = 3001 dtrace_caller(aframes)) == -1) { 3002 /* 3003 * We have failed to do this the quick way; 3004 * we must resort to the slower approach of 3005 * calling dtrace_getpcstack(). 3006 */ 3007 pc_t caller = 0; 3008 3009 dtrace_getpcstack(&caller, 1, aframes, NULL); 3010 mstate->dtms_caller = caller; 3011 } 3012 3013 mstate->dtms_present |= DTRACE_MSTATE_CALLER; 3014 } 3015 return (mstate->dtms_caller); 3016 3017 case DIF_VAR_UCALLER: 3018 if (!dtrace_priv_proc(state)) 3019 return (0); 3020 3021 if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) { 3022 uint64_t ustack[3]; 3023 3024 /* 3025 * dtrace_getupcstack() fills in the first uint64_t 3026 * with the current PID. The second uint64_t will 3027 * be the program counter at user-level. The third 3028 * uint64_t will contain the caller, which is what 3029 * we're after. 3030 */ 3031 ustack[2] = 0; 3032 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3033 dtrace_getupcstack(ustack, 3); 3034 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3035 mstate->dtms_ucaller = ustack[2]; 3036 mstate->dtms_present |= DTRACE_MSTATE_UCALLER; 3037 } 3038 3039 return (mstate->dtms_ucaller); 3040 3041 case DIF_VAR_PROBEPROV: 3042 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 3043 return (dtrace_dif_varstr( 3044 (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name, 3045 state, mstate)); 3046 3047 case DIF_VAR_PROBEMOD: 3048 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 3049 return (dtrace_dif_varstr( 3050 (uintptr_t)mstate->dtms_probe->dtpr_mod, 3051 state, mstate)); 3052 3053 case DIF_VAR_PROBEFUNC: 3054 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 3055 return (dtrace_dif_varstr( 3056 (uintptr_t)mstate->dtms_probe->dtpr_func, 3057 state, mstate)); 3058 3059 case DIF_VAR_PROBENAME: 3060 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 3061 return (dtrace_dif_varstr( 3062 (uintptr_t)mstate->dtms_probe->dtpr_name, 3063 state, mstate)); 3064 3065 case DIF_VAR_PID: 3066 if (!dtrace_priv_proc(state)) 3067 return (0); 3068 3069 #if defined(sun) 3070 /* 3071 * Note that we are assuming that an unanchored probe is 3072 * always due to a high-level interrupt. (And we're assuming 3073 * that there is only a single high level interrupt.) 3074 */ 3075 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3076 return (pid0.pid_id); 3077 3078 /* 3079 * It is always safe to dereference one's own t_procp pointer: 3080 * it always points to a valid, allocated proc structure. 3081 * Further, it is always safe to dereference the p_pidp member 3082 * of one's own proc structure. (These are truisms becuase 3083 * threads and processes don't clean up their own state -- 3084 * they leave that task to whomever reaps them.) 3085 */ 3086 return ((uint64_t)curthread->t_procp->p_pidp->pid_id); 3087 #else 3088 return ((uint64_t)curproc->p_pid); 3089 #endif 3090 3091 case DIF_VAR_PPID: 3092 if (!dtrace_priv_proc(state)) 3093 return (0); 3094 3095 #if defined(sun) 3096 /* 3097 * See comment in DIF_VAR_PID. 3098 */ 3099 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3100 return (pid0.pid_id); 3101 3102 /* 3103 * It is always safe to dereference one's own t_procp pointer: 3104 * it always points to a valid, allocated proc structure. 3105 * (This is true because threads don't clean up their own 3106 * state -- they leave that task to whomever reaps them.) 3107 */ 3108 return ((uint64_t)curthread->t_procp->p_ppid); 3109 #else 3110 return ((uint64_t)curproc->p_pptr->p_pid); 3111 #endif 3112 3113 case DIF_VAR_TID: 3114 #if defined(sun) 3115 /* 3116 * See comment in DIF_VAR_PID. 3117 */ 3118 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3119 return (0); 3120 #endif 3121 3122 return ((uint64_t)curthread->t_tid); 3123 3124 case DIF_VAR_EXECARGS: { 3125 struct pargs *p_args = curthread->td_proc->p_args; 3126 3127 if (p_args == NULL) 3128 return(0); 3129 3130 return (dtrace_dif_varstrz( 3131 (uintptr_t) p_args->ar_args, p_args->ar_length, state, mstate)); 3132 } 3133 3134 case DIF_VAR_EXECNAME: 3135 #if defined(sun) 3136 if (!dtrace_priv_proc(state)) 3137 return (0); 3138 3139 /* 3140 * See comment in DIF_VAR_PID. 3141 */ 3142 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3143 return ((uint64_t)(uintptr_t)p0.p_user.u_comm); 3144 3145 /* 3146 * It is always safe to dereference one's own t_procp pointer: 3147 * it always points to a valid, allocated proc structure. 3148 * (This is true because threads don't clean up their own 3149 * state -- they leave that task to whomever reaps them.) 3150 */ 3151 return (dtrace_dif_varstr( 3152 (uintptr_t)curthread->t_procp->p_user.u_comm, 3153 state, mstate)); 3154 #else 3155 return (dtrace_dif_varstr( 3156 (uintptr_t) curthread->td_proc->p_comm, state, mstate)); 3157 #endif 3158 3159 case DIF_VAR_ZONENAME: 3160 #if defined(sun) 3161 if (!dtrace_priv_proc(state)) 3162 return (0); 3163 3164 /* 3165 * See comment in DIF_VAR_PID. 3166 */ 3167 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3168 return ((uint64_t)(uintptr_t)p0.p_zone->zone_name); 3169 3170 /* 3171 * It is always safe to dereference one's own t_procp pointer: 3172 * it always points to a valid, allocated proc structure. 3173 * (This is true because threads don't clean up their own 3174 * state -- they leave that task to whomever reaps them.) 3175 */ 3176 return (dtrace_dif_varstr( 3177 (uintptr_t)curthread->t_procp->p_zone->zone_name, 3178 state, mstate)); 3179 #else 3180 return (0); 3181 #endif 3182 3183 case DIF_VAR_UID: 3184 if (!dtrace_priv_proc(state)) 3185 return (0); 3186 3187 #if defined(sun) 3188 /* 3189 * See comment in DIF_VAR_PID. 3190 */ 3191 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3192 return ((uint64_t)p0.p_cred->cr_uid); 3193 #endif 3194 3195 /* 3196 * It is always safe to dereference one's own t_procp pointer: 3197 * it always points to a valid, allocated proc structure. 3198 * (This is true because threads don't clean up their own 3199 * state -- they leave that task to whomever reaps them.) 3200 * 3201 * Additionally, it is safe to dereference one's own process 3202 * credential, since this is never NULL after process birth. 3203 */ 3204 return ((uint64_t)curthread->t_procp->p_cred->cr_uid); 3205 3206 case DIF_VAR_GID: 3207 if (!dtrace_priv_proc(state)) 3208 return (0); 3209 3210 #if defined(sun) 3211 /* 3212 * See comment in DIF_VAR_PID. 3213 */ 3214 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3215 return ((uint64_t)p0.p_cred->cr_gid); 3216 #endif 3217 3218 /* 3219 * It is always safe to dereference one's own t_procp pointer: 3220 * it always points to a valid, allocated proc structure. 3221 * (This is true because threads don't clean up their own 3222 * state -- they leave that task to whomever reaps them.) 3223 * 3224 * Additionally, it is safe to dereference one's own process 3225 * credential, since this is never NULL after process birth. 3226 */ 3227 return ((uint64_t)curthread->t_procp->p_cred->cr_gid); 3228 3229 case DIF_VAR_ERRNO: { 3230 #if defined(sun) 3231 klwp_t *lwp; 3232 if (!dtrace_priv_proc(state)) 3233 return (0); 3234 3235 /* 3236 * See comment in DIF_VAR_PID. 3237 */ 3238 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3239 return (0); 3240 3241 /* 3242 * It is always safe to dereference one's own t_lwp pointer in 3243 * the event that this pointer is non-NULL. (This is true 3244 * because threads and lwps don't clean up their own state -- 3245 * they leave that task to whomever reaps them.) 3246 */ 3247 if ((lwp = curthread->t_lwp) == NULL) 3248 return (0); 3249 3250 return ((uint64_t)lwp->lwp_errno); 3251 #else 3252 return (curthread->td_errno); 3253 #endif 3254 } 3255 #if !defined(sun) 3256 case DIF_VAR_CPU: { 3257 return curcpu; 3258 } 3259 #endif 3260 default: 3261 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 3262 return (0); 3263 } 3264 } 3265 3266 /* 3267 * Emulate the execution of DTrace ID subroutines invoked by the call opcode. 3268 * Notice that we don't bother validating the proper number of arguments or 3269 * their types in the tuple stack. This isn't needed because all argument 3270 * interpretation is safe because of our load safety -- the worst that can 3271 * happen is that a bogus program can obtain bogus results. 3272 */ 3273 static void 3274 dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs, 3275 dtrace_key_t *tupregs, int nargs, 3276 dtrace_mstate_t *mstate, dtrace_state_t *state) 3277 { 3278 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 3279 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; 3280 dtrace_vstate_t *vstate = &state->dts_vstate; 3281 3282 #if defined(sun) 3283 union { 3284 mutex_impl_t mi; 3285 uint64_t mx; 3286 } m; 3287 3288 union { 3289 krwlock_t ri; 3290 uintptr_t rw; 3291 } r; 3292 #else 3293 struct thread *lowner; 3294 union { 3295 struct lock_object *li; 3296 uintptr_t lx; 3297 } l; 3298 #endif 3299 3300 switch (subr) { 3301 case DIF_SUBR_RAND: 3302 regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875; 3303 break; 3304 3305 #if defined(sun) 3306 case DIF_SUBR_MUTEX_OWNED: 3307 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3308 mstate, vstate)) { 3309 regs[rd] = 0; 3310 break; 3311 } 3312 3313 m.mx = dtrace_load64(tupregs[0].dttk_value); 3314 if (MUTEX_TYPE_ADAPTIVE(&m.mi)) 3315 regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER; 3316 else 3317 regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock); 3318 break; 3319 3320 case DIF_SUBR_MUTEX_OWNER: 3321 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3322 mstate, vstate)) { 3323 regs[rd] = 0; 3324 break; 3325 } 3326 3327 m.mx = dtrace_load64(tupregs[0].dttk_value); 3328 if (MUTEX_TYPE_ADAPTIVE(&m.mi) && 3329 MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER) 3330 regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi); 3331 else 3332 regs[rd] = 0; 3333 break; 3334 3335 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: 3336 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3337 mstate, vstate)) { 3338 regs[rd] = 0; 3339 break; 3340 } 3341 3342 m.mx = dtrace_load64(tupregs[0].dttk_value); 3343 regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi); 3344 break; 3345 3346 case DIF_SUBR_MUTEX_TYPE_SPIN: 3347 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3348 mstate, vstate)) { 3349 regs[rd] = 0; 3350 break; 3351 } 3352 3353 m.mx = dtrace_load64(tupregs[0].dttk_value); 3354 regs[rd] = MUTEX_TYPE_SPIN(&m.mi); 3355 break; 3356 3357 case DIF_SUBR_RW_READ_HELD: { 3358 uintptr_t tmp; 3359 3360 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t), 3361 mstate, vstate)) { 3362 regs[rd] = 0; 3363 break; 3364 } 3365 3366 r.rw = dtrace_loadptr(tupregs[0].dttk_value); 3367 regs[rd] = _RW_READ_HELD(&r.ri, tmp); 3368 break; 3369 } 3370 3371 case DIF_SUBR_RW_WRITE_HELD: 3372 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), 3373 mstate, vstate)) { 3374 regs[rd] = 0; 3375 break; 3376 } 3377 3378 r.rw = dtrace_loadptr(tupregs[0].dttk_value); 3379 regs[rd] = _RW_WRITE_HELD(&r.ri); 3380 break; 3381 3382 case DIF_SUBR_RW_ISWRITER: 3383 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), 3384 mstate, vstate)) { 3385 regs[rd] = 0; 3386 break; 3387 } 3388 3389 r.rw = dtrace_loadptr(tupregs[0].dttk_value); 3390 regs[rd] = _RW_ISWRITER(&r.ri); 3391 break; 3392 3393 #else 3394 case DIF_SUBR_MUTEX_OWNED: 3395 if (!dtrace_canload(tupregs[0].dttk_value, 3396 sizeof (struct lock_object), mstate, vstate)) { 3397 regs[rd] = 0; 3398 break; 3399 } 3400 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value); 3401 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner); 3402 break; 3403 3404 case DIF_SUBR_MUTEX_OWNER: 3405 if (!dtrace_canload(tupregs[0].dttk_value, 3406 sizeof (struct lock_object), mstate, vstate)) { 3407 regs[rd] = 0; 3408 break; 3409 } 3410 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value); 3411 LOCK_CLASS(l.li)->lc_owner(l.li, &lowner); 3412 regs[rd] = (uintptr_t)lowner; 3413 break; 3414 3415 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: 3416 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx), 3417 mstate, vstate)) { 3418 regs[rd] = 0; 3419 break; 3420 } 3421 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value); 3422 /* XXX - should be only LC_SLEEPABLE? */ 3423 regs[rd] = (LOCK_CLASS(l.li)->lc_flags & 3424 (LC_SLEEPLOCK | LC_SLEEPABLE)) != 0; 3425 break; 3426 3427 case DIF_SUBR_MUTEX_TYPE_SPIN: 3428 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx), 3429 mstate, vstate)) { 3430 regs[rd] = 0; 3431 break; 3432 } 3433 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value); 3434 regs[rd] = (LOCK_CLASS(l.li)->lc_flags & LC_SPINLOCK) != 0; 3435 break; 3436 3437 case DIF_SUBR_RW_READ_HELD: 3438 case DIF_SUBR_SX_SHARED_HELD: 3439 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t), 3440 mstate, vstate)) { 3441 regs[rd] = 0; 3442 break; 3443 } 3444 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value); 3445 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) && 3446 lowner == NULL; 3447 break; 3448 3449 case DIF_SUBR_RW_WRITE_HELD: 3450 case DIF_SUBR_SX_EXCLUSIVE_HELD: 3451 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t), 3452 mstate, vstate)) { 3453 regs[rd] = 0; 3454 break; 3455 } 3456 l.lx = dtrace_loadptr(tupregs[0].dttk_value); 3457 LOCK_CLASS(l.li)->lc_owner(l.li, &lowner); 3458 regs[rd] = (lowner == curthread); 3459 break; 3460 3461 case DIF_SUBR_RW_ISWRITER: 3462 case DIF_SUBR_SX_ISEXCLUSIVE: 3463 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t), 3464 mstate, vstate)) { 3465 regs[rd] = 0; 3466 break; 3467 } 3468 l.lx = dtrace_loadptr(tupregs[0].dttk_value); 3469 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) && 3470 lowner != NULL; 3471 break; 3472 #endif /* ! defined(sun) */ 3473 3474 case DIF_SUBR_BCOPY: { 3475 /* 3476 * We need to be sure that the destination is in the scratch 3477 * region -- no other region is allowed. 3478 */ 3479 uintptr_t src = tupregs[0].dttk_value; 3480 uintptr_t dest = tupregs[1].dttk_value; 3481 size_t size = tupregs[2].dttk_value; 3482 3483 if (!dtrace_inscratch(dest, size, mstate)) { 3484 *flags |= CPU_DTRACE_BADADDR; 3485 *illval = regs[rd]; 3486 break; 3487 } 3488 3489 if (!dtrace_canload(src, size, mstate, vstate)) { 3490 regs[rd] = 0; 3491 break; 3492 } 3493 3494 dtrace_bcopy((void *)src, (void *)dest, size); 3495 break; 3496 } 3497 3498 case DIF_SUBR_ALLOCA: 3499 case DIF_SUBR_COPYIN: { 3500 uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); 3501 uint64_t size = 3502 tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value; 3503 size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size; 3504 3505 /* 3506 * This action doesn't require any credential checks since 3507 * probes will not activate in user contexts to which the 3508 * enabling user does not have permissions. 3509 */ 3510 3511 /* 3512 * Rounding up the user allocation size could have overflowed 3513 * a large, bogus allocation (like -1ULL) to 0. 3514 */ 3515 if (scratch_size < size || 3516 !DTRACE_INSCRATCH(mstate, scratch_size)) { 3517 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3518 regs[rd] = 0; 3519 break; 3520 } 3521 3522 if (subr == DIF_SUBR_COPYIN) { 3523 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3524 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); 3525 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3526 } 3527 3528 mstate->dtms_scratch_ptr += scratch_size; 3529 regs[rd] = dest; 3530 break; 3531 } 3532 3533 case DIF_SUBR_COPYINTO: { 3534 uint64_t size = tupregs[1].dttk_value; 3535 uintptr_t dest = tupregs[2].dttk_value; 3536 3537 /* 3538 * This action doesn't require any credential checks since 3539 * probes will not activate in user contexts to which the 3540 * enabling user does not have permissions. 3541 */ 3542 if (!dtrace_inscratch(dest, size, mstate)) { 3543 *flags |= CPU_DTRACE_BADADDR; 3544 *illval = regs[rd]; 3545 break; 3546 } 3547 3548 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3549 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); 3550 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3551 break; 3552 } 3553 3554 case DIF_SUBR_COPYINSTR: { 3555 uintptr_t dest = mstate->dtms_scratch_ptr; 3556 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3557 3558 if (nargs > 1 && tupregs[1].dttk_value < size) 3559 size = tupregs[1].dttk_value + 1; 3560 3561 /* 3562 * This action doesn't require any credential checks since 3563 * probes will not activate in user contexts to which the 3564 * enabling user does not have permissions. 3565 */ 3566 if (!DTRACE_INSCRATCH(mstate, size)) { 3567 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3568 regs[rd] = 0; 3569 break; 3570 } 3571 3572 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3573 dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags); 3574 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3575 3576 ((char *)dest)[size - 1] = '\0'; 3577 mstate->dtms_scratch_ptr += size; 3578 regs[rd] = dest; 3579 break; 3580 } 3581 3582 #if defined(sun) 3583 case DIF_SUBR_MSGSIZE: 3584 case DIF_SUBR_MSGDSIZE: { 3585 uintptr_t baddr = tupregs[0].dttk_value, daddr; 3586 uintptr_t wptr, rptr; 3587 size_t count = 0; 3588 int cont = 0; 3589 3590 while (baddr != 0 && !(*flags & CPU_DTRACE_FAULT)) { 3591 3592 if (!dtrace_canload(baddr, sizeof (mblk_t), mstate, 3593 vstate)) { 3594 regs[rd] = 0; 3595 break; 3596 } 3597 3598 wptr = dtrace_loadptr(baddr + 3599 offsetof(mblk_t, b_wptr)); 3600 3601 rptr = dtrace_loadptr(baddr + 3602 offsetof(mblk_t, b_rptr)); 3603 3604 if (wptr < rptr) { 3605 *flags |= CPU_DTRACE_BADADDR; 3606 *illval = tupregs[0].dttk_value; 3607 break; 3608 } 3609 3610 daddr = dtrace_loadptr(baddr + 3611 offsetof(mblk_t, b_datap)); 3612 3613 baddr = dtrace_loadptr(baddr + 3614 offsetof(mblk_t, b_cont)); 3615 3616 /* 3617 * We want to prevent against denial-of-service here, 3618 * so we're only going to search the list for 3619 * dtrace_msgdsize_max mblks. 3620 */ 3621 if (cont++ > dtrace_msgdsize_max) { 3622 *flags |= CPU_DTRACE_ILLOP; 3623 break; 3624 } 3625 3626 if (subr == DIF_SUBR_MSGDSIZE) { 3627 if (dtrace_load8(daddr + 3628 offsetof(dblk_t, db_type)) != M_DATA) 3629 continue; 3630 } 3631 3632 count += wptr - rptr; 3633 } 3634 3635 if (!(*flags & CPU_DTRACE_FAULT)) 3636 regs[rd] = count; 3637 3638 break; 3639 } 3640 #endif 3641 3642 case DIF_SUBR_PROGENYOF: { 3643 pid_t pid = tupregs[0].dttk_value; 3644 proc_t *p; 3645 int rval = 0; 3646 3647 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3648 3649 for (p = curthread->t_procp; p != NULL; p = p->p_parent) { 3650 #if defined(sun) 3651 if (p->p_pidp->pid_id == pid) { 3652 #else 3653 if (p->p_pid == pid) { 3654 #endif 3655 rval = 1; 3656 break; 3657 } 3658 } 3659 3660 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3661 3662 regs[rd] = rval; 3663 break; 3664 } 3665 3666 case DIF_SUBR_SPECULATION: 3667 regs[rd] = dtrace_speculation(state); 3668 break; 3669 3670 case DIF_SUBR_COPYOUT: { 3671 uintptr_t kaddr = tupregs[0].dttk_value; 3672 uintptr_t uaddr = tupregs[1].dttk_value; 3673 uint64_t size = tupregs[2].dttk_value; 3674 3675 if (!dtrace_destructive_disallow && 3676 dtrace_priv_proc_control(state) && 3677 !dtrace_istoxic(kaddr, size)) { 3678 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3679 dtrace_copyout(kaddr, uaddr, size, flags); 3680 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3681 } 3682 break; 3683 } 3684 3685 case DIF_SUBR_COPYOUTSTR: { 3686 uintptr_t kaddr = tupregs[0].dttk_value; 3687 uintptr_t uaddr = tupregs[1].dttk_value; 3688 uint64_t size = tupregs[2].dttk_value; 3689 3690 if (!dtrace_destructive_disallow && 3691 dtrace_priv_proc_control(state) && 3692 !dtrace_istoxic(kaddr, size)) { 3693 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3694 dtrace_copyoutstr(kaddr, uaddr, size, flags); 3695 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3696 } 3697 break; 3698 } 3699 3700 case DIF_SUBR_STRLEN: { 3701 size_t sz; 3702 uintptr_t addr = (uintptr_t)tupregs[0].dttk_value; 3703 sz = dtrace_strlen((char *)addr, 3704 state->dts_options[DTRACEOPT_STRSIZE]); 3705 3706 if (!dtrace_canload(addr, sz + 1, mstate, vstate)) { 3707 regs[rd] = 0; 3708 break; 3709 } 3710 3711 regs[rd] = sz; 3712 3713 break; 3714 } 3715 3716 case DIF_SUBR_STRCHR: 3717 case DIF_SUBR_STRRCHR: { 3718 /* 3719 * We're going to iterate over the string looking for the 3720 * specified character. We will iterate until we have reached 3721 * the string length or we have found the character. If this 3722 * is DIF_SUBR_STRRCHR, we will look for the last occurrence 3723 * of the specified character instead of the first. 3724 */ 3725 uintptr_t saddr = tupregs[0].dttk_value; 3726 uintptr_t addr = tupregs[0].dttk_value; 3727 uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE]; 3728 char c, target = (char)tupregs[1].dttk_value; 3729 3730 for (regs[rd] = 0; addr < limit; addr++) { 3731 if ((c = dtrace_load8(addr)) == target) { 3732 regs[rd] = addr; 3733 3734 if (subr == DIF_SUBR_STRCHR) 3735 break; 3736 } 3737 3738 if (c == '\0') 3739 break; 3740 } 3741 3742 if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) { 3743 regs[rd] = 0; 3744 break; 3745 } 3746 3747 break; 3748 } 3749 3750 case DIF_SUBR_STRSTR: 3751 case DIF_SUBR_INDEX: 3752 case DIF_SUBR_RINDEX: { 3753 /* 3754 * We're going to iterate over the string looking for the 3755 * specified string. We will iterate until we have reached 3756 * the string length or we have found the string. (Yes, this 3757 * is done in the most naive way possible -- but considering 3758 * that the string we're searching for is likely to be 3759 * relatively short, the complexity of Rabin-Karp or similar 3760 * hardly seems merited.) 3761 */ 3762 char *addr = (char *)(uintptr_t)tupregs[0].dttk_value; 3763 char *substr = (char *)(uintptr_t)tupregs[1].dttk_value; 3764 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3765 size_t len = dtrace_strlen(addr, size); 3766 size_t sublen = dtrace_strlen(substr, size); 3767 char *limit = addr + len, *orig = addr; 3768 int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1; 3769 int inc = 1; 3770 3771 regs[rd] = notfound; 3772 3773 if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) { 3774 regs[rd] = 0; 3775 break; 3776 } 3777 3778 if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate, 3779 vstate)) { 3780 regs[rd] = 0; 3781 break; 3782 } 3783 3784 /* 3785 * strstr() and index()/rindex() have similar semantics if 3786 * both strings are the empty string: strstr() returns a 3787 * pointer to the (empty) string, and index() and rindex() 3788 * both return index 0 (regardless of any position argument). 3789 */ 3790 if (sublen == 0 && len == 0) { 3791 if (subr == DIF_SUBR_STRSTR) 3792 regs[rd] = (uintptr_t)addr; 3793 else 3794 regs[rd] = 0; 3795 break; 3796 } 3797 3798 if (subr != DIF_SUBR_STRSTR) { 3799 if (subr == DIF_SUBR_RINDEX) { 3800 limit = orig - 1; 3801 addr += len; 3802 inc = -1; 3803 } 3804 3805 /* 3806 * Both index() and rindex() take an optional position 3807 * argument that denotes the starting position. 3808 */ 3809 if (nargs == 3) { 3810 int64_t pos = (int64_t)tupregs[2].dttk_value; 3811 3812 /* 3813 * If the position argument to index() is 3814 * negative, Perl implicitly clamps it at 3815 * zero. This semantic is a little surprising 3816 * given the special meaning of negative 3817 * positions to similar Perl functions like 3818 * substr(), but it appears to reflect a 3819 * notion that index() can start from a 3820 * negative index and increment its way up to 3821 * the string. Given this notion, Perl's 3822 * rindex() is at least self-consistent in 3823 * that it implicitly clamps positions greater 3824 * than the string length to be the string 3825 * length. Where Perl completely loses 3826 * coherence, however, is when the specified 3827 * substring is the empty string (""). In 3828 * this case, even if the position is 3829 * negative, rindex() returns 0 -- and even if 3830 * the position is greater than the length, 3831 * index() returns the string length. These 3832 * semantics violate the notion that index() 3833 * should never return a value less than the 3834 * specified position and that rindex() should 3835 * never return a value greater than the 3836 * specified position. (One assumes that 3837 * these semantics are artifacts of Perl's 3838 * implementation and not the results of 3839 * deliberate design -- it beggars belief that 3840 * even Larry Wall could desire such oddness.) 3841 * While in the abstract one would wish for 3842 * consistent position semantics across 3843 * substr(), index() and rindex() -- or at the 3844 * very least self-consistent position 3845 * semantics for index() and rindex() -- we 3846 * instead opt to keep with the extant Perl 3847 * semantics, in all their broken glory. (Do 3848 * we have more desire to maintain Perl's 3849 * semantics than Perl does? Probably.) 3850 */ 3851 if (subr == DIF_SUBR_RINDEX) { 3852 if (pos < 0) { 3853 if (sublen == 0) 3854 regs[rd] = 0; 3855 break; 3856 } 3857 3858 if (pos > len) 3859 pos = len; 3860 } else { 3861 if (pos < 0) 3862 pos = 0; 3863 3864 if (pos >= len) { 3865 if (sublen == 0) 3866 regs[rd] = len; 3867 break; 3868 } 3869 } 3870 3871 addr = orig + pos; 3872 } 3873 } 3874 3875 for (regs[rd] = notfound; addr != limit; addr += inc) { 3876 if (dtrace_strncmp(addr, substr, sublen) == 0) { 3877 if (subr != DIF_SUBR_STRSTR) { 3878 /* 3879 * As D index() and rindex() are 3880 * modeled on Perl (and not on awk), 3881 * we return a zero-based (and not a 3882 * one-based) index. (For you Perl 3883 * weenies: no, we're not going to add 3884 * $[ -- and shouldn't you be at a con 3885 * or something?) 3886 */ 3887 regs[rd] = (uintptr_t)(addr - orig); 3888 break; 3889 } 3890 3891 ASSERT(subr == DIF_SUBR_STRSTR); 3892 regs[rd] = (uintptr_t)addr; 3893 break; 3894 } 3895 } 3896 3897 break; 3898 } 3899 3900 case DIF_SUBR_STRTOK: { 3901 uintptr_t addr = tupregs[0].dttk_value; 3902 uintptr_t tokaddr = tupregs[1].dttk_value; 3903 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3904 uintptr_t limit, toklimit = tokaddr + size; 3905 uint8_t c = 0, tokmap[32]; /* 256 / 8 */ 3906 char *dest = (char *)mstate->dtms_scratch_ptr; 3907 int i; 3908 3909 /* 3910 * Check both the token buffer and (later) the input buffer, 3911 * since both could be non-scratch addresses. 3912 */ 3913 if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) { 3914 regs[rd] = 0; 3915 break; 3916 } 3917 3918 if (!DTRACE_INSCRATCH(mstate, size)) { 3919 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3920 regs[rd] = 0; 3921 break; 3922 } 3923 3924 if (addr == 0) { 3925 /* 3926 * If the address specified is NULL, we use our saved 3927 * strtok pointer from the mstate. Note that this 3928 * means that the saved strtok pointer is _only_ 3929 * valid within multiple enablings of the same probe -- 3930 * it behaves like an implicit clause-local variable. 3931 */ 3932 addr = mstate->dtms_strtok; 3933 } else { 3934 /* 3935 * If the user-specified address is non-NULL we must 3936 * access check it. This is the only time we have 3937 * a chance to do so, since this address may reside 3938 * in the string table of this clause-- future calls 3939 * (when we fetch addr from mstate->dtms_strtok) 3940 * would fail this access check. 3941 */ 3942 if (!dtrace_strcanload(addr, size, mstate, vstate)) { 3943 regs[rd] = 0; 3944 break; 3945 } 3946 } 3947 3948 /* 3949 * First, zero the token map, and then process the token 3950 * string -- setting a bit in the map for every character 3951 * found in the token string. 3952 */ 3953 for (i = 0; i < sizeof (tokmap); i++) 3954 tokmap[i] = 0; 3955 3956 for (; tokaddr < toklimit; tokaddr++) { 3957 if ((c = dtrace_load8(tokaddr)) == '\0') 3958 break; 3959 3960 ASSERT((c >> 3) < sizeof (tokmap)); 3961 tokmap[c >> 3] |= (1 << (c & 0x7)); 3962 } 3963 3964 for (limit = addr + size; addr < limit; addr++) { 3965 /* 3966 * We're looking for a character that is _not_ contained 3967 * in the token string. 3968 */ 3969 if ((c = dtrace_load8(addr)) == '\0') 3970 break; 3971 3972 if (!(tokmap[c >> 3] & (1 << (c & 0x7)))) 3973 break; 3974 } 3975 3976 if (c == '\0') { 3977 /* 3978 * We reached the end of the string without finding 3979 * any character that was not in the token string. 3980 * We return NULL in this case, and we set the saved 3981 * address to NULL as well. 3982 */ 3983 regs[rd] = 0; 3984 mstate->dtms_strtok = 0; 3985 break; 3986 } 3987 3988 /* 3989 * From here on, we're copying into the destination string. 3990 */ 3991 for (i = 0; addr < limit && i < size - 1; addr++) { 3992 if ((c = dtrace_load8(addr)) == '\0') 3993 break; 3994 3995 if (tokmap[c >> 3] & (1 << (c & 0x7))) 3996 break; 3997 3998 ASSERT(i < size); 3999 dest[i++] = c; 4000 } 4001 4002 ASSERT(i < size); 4003 dest[i] = '\0'; 4004 regs[rd] = (uintptr_t)dest; 4005 mstate->dtms_scratch_ptr += size; 4006 mstate->dtms_strtok = addr; 4007 break; 4008 } 4009 4010 case DIF_SUBR_SUBSTR: { 4011 uintptr_t s = tupregs[0].dttk_value; 4012 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4013 char *d = (char *)mstate->dtms_scratch_ptr; 4014 int64_t index = (int64_t)tupregs[1].dttk_value; 4015 int64_t remaining = (int64_t)tupregs[2].dttk_value; 4016 size_t len = dtrace_strlen((char *)s, size); 4017 int64_t i = 0; 4018 4019 if (!dtrace_canload(s, len + 1, mstate, vstate)) { 4020 regs[rd] = 0; 4021 break; 4022 } 4023 4024 if (!DTRACE_INSCRATCH(mstate, size)) { 4025 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4026 regs[rd] = 0; 4027 break; 4028 } 4029 4030 if (nargs <= 2) 4031 remaining = (int64_t)size; 4032 4033 if (index < 0) { 4034 index += len; 4035 4036 if (index < 0 && index + remaining > 0) { 4037 remaining += index; 4038 index = 0; 4039 } 4040 } 4041 4042 if (index >= len || index < 0) { 4043 remaining = 0; 4044 } else if (remaining < 0) { 4045 remaining += len - index; 4046 } else if (index + remaining > size) { 4047 remaining = size - index; 4048 } 4049 4050 for (i = 0; i < remaining; i++) { 4051 if ((d[i] = dtrace_load8(s + index + i)) == '\0') 4052 break; 4053 } 4054 4055 d[i] = '\0'; 4056 4057 mstate->dtms_scratch_ptr += size; 4058 regs[rd] = (uintptr_t)d; 4059 break; 4060 } 4061 4062 case DIF_SUBR_TOUPPER: 4063 case DIF_SUBR_TOLOWER: { 4064 uintptr_t s = tupregs[0].dttk_value; 4065 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4066 char *dest = (char *)mstate->dtms_scratch_ptr, c; 4067 size_t len = dtrace_strlen((char *)s, size); 4068 char lower, upper, convert; 4069 int64_t i; 4070 4071 if (subr == DIF_SUBR_TOUPPER) { 4072 lower = 'a'; 4073 upper = 'z'; 4074 convert = 'A'; 4075 } else { 4076 lower = 'A'; 4077 upper = 'Z'; 4078 convert = 'a'; 4079 } 4080 4081 if (!dtrace_canload(s, len + 1, mstate, vstate)) { 4082 regs[rd] = 0; 4083 break; 4084 } 4085 4086 if (!DTRACE_INSCRATCH(mstate, size)) { 4087 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4088 regs[rd] = 0; 4089 break; 4090 } 4091 4092 for (i = 0; i < size - 1; i++) { 4093 if ((c = dtrace_load8(s + i)) == '\0') 4094 break; 4095 4096 if (c >= lower && c <= upper) 4097 c = convert + (c - lower); 4098 4099 dest[i] = c; 4100 } 4101 4102 ASSERT(i < size); 4103 dest[i] = '\0'; 4104 regs[rd] = (uintptr_t)dest; 4105 mstate->dtms_scratch_ptr += size; 4106 break; 4107 } 4108 4109 #if defined(sun) 4110 case DIF_SUBR_GETMAJOR: 4111 #ifdef _LP64 4112 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64; 4113 #else 4114 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ; 4115 #endif 4116 break; 4117 4118 case DIF_SUBR_GETMINOR: 4119 #ifdef _LP64 4120 regs[rd] = tupregs[0].dttk_value & MAXMIN64; 4121 #else 4122 regs[rd] = tupregs[0].dttk_value & MAXMIN; 4123 #endif 4124 break; 4125 4126 case DIF_SUBR_DDI_PATHNAME: { 4127 /* 4128 * This one is a galactic mess. We are going to roughly 4129 * emulate ddi_pathname(), but it's made more complicated 4130 * by the fact that we (a) want to include the minor name and 4131 * (b) must proceed iteratively instead of recursively. 4132 */ 4133 uintptr_t dest = mstate->dtms_scratch_ptr; 4134 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4135 char *start = (char *)dest, *end = start + size - 1; 4136 uintptr_t daddr = tupregs[0].dttk_value; 4137 int64_t minor = (int64_t)tupregs[1].dttk_value; 4138 char *s; 4139 int i, len, depth = 0; 4140 4141 /* 4142 * Due to all the pointer jumping we do and context we must 4143 * rely upon, we just mandate that the user must have kernel 4144 * read privileges to use this routine. 4145 */ 4146 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) { 4147 *flags |= CPU_DTRACE_KPRIV; 4148 *illval = daddr; 4149 regs[rd] = 0; 4150 } 4151 4152 if (!DTRACE_INSCRATCH(mstate, size)) { 4153 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4154 regs[rd] = 0; 4155 break; 4156 } 4157 4158 *end = '\0'; 4159 4160 /* 4161 * We want to have a name for the minor. In order to do this, 4162 * we need to walk the minor list from the devinfo. We want 4163 * to be sure that we don't infinitely walk a circular list, 4164 * so we check for circularity by sending a scout pointer 4165 * ahead two elements for every element that we iterate over; 4166 * if the list is circular, these will ultimately point to the 4167 * same element. You may recognize this little trick as the 4168 * answer to a stupid interview question -- one that always 4169 * seems to be asked by those who had to have it laboriously 4170 * explained to them, and who can't even concisely describe 4171 * the conditions under which one would be forced to resort to 4172 * this technique. Needless to say, those conditions are 4173 * found here -- and probably only here. Is this the only use 4174 * of this infamous trick in shipping, production code? If it 4175 * isn't, it probably should be... 4176 */ 4177 if (minor != -1) { 4178 uintptr_t maddr = dtrace_loadptr(daddr + 4179 offsetof(struct dev_info, devi_minor)); 4180 4181 uintptr_t next = offsetof(struct ddi_minor_data, next); 4182 uintptr_t name = offsetof(struct ddi_minor_data, 4183 d_minor) + offsetof(struct ddi_minor, name); 4184 uintptr_t dev = offsetof(struct ddi_minor_data, 4185 d_minor) + offsetof(struct ddi_minor, dev); 4186 uintptr_t scout; 4187 4188 if (maddr != NULL) 4189 scout = dtrace_loadptr(maddr + next); 4190 4191 while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { 4192 uint64_t m; 4193 #ifdef _LP64 4194 m = dtrace_load64(maddr + dev) & MAXMIN64; 4195 #else 4196 m = dtrace_load32(maddr + dev) & MAXMIN; 4197 #endif 4198 if (m != minor) { 4199 maddr = dtrace_loadptr(maddr + next); 4200 4201 if (scout == NULL) 4202 continue; 4203 4204 scout = dtrace_loadptr(scout + next); 4205 4206 if (scout == NULL) 4207 continue; 4208 4209 scout = dtrace_loadptr(scout + next); 4210 4211 if (scout == NULL) 4212 continue; 4213 4214 if (scout == maddr) { 4215 *flags |= CPU_DTRACE_ILLOP; 4216 break; 4217 } 4218 4219 continue; 4220 } 4221 4222 /* 4223 * We have the minor data. Now we need to 4224 * copy the minor's name into the end of the 4225 * pathname. 4226 */ 4227 s = (char *)dtrace_loadptr(maddr + name); 4228 len = dtrace_strlen(s, size); 4229 4230 if (*flags & CPU_DTRACE_FAULT) 4231 break; 4232 4233 if (len != 0) { 4234 if ((end -= (len + 1)) < start) 4235 break; 4236 4237 *end = ':'; 4238 } 4239 4240 for (i = 1; i <= len; i++) 4241 end[i] = dtrace_load8((uintptr_t)s++); 4242 break; 4243 } 4244 } 4245 4246 while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { 4247 ddi_node_state_t devi_state; 4248 4249 devi_state = dtrace_load32(daddr + 4250 offsetof(struct dev_info, devi_node_state)); 4251 4252 if (*flags & CPU_DTRACE_FAULT) 4253 break; 4254 4255 if (devi_state >= DS_INITIALIZED) { 4256 s = (char *)dtrace_loadptr(daddr + 4257 offsetof(struct dev_info, devi_addr)); 4258 len = dtrace_strlen(s, size); 4259 4260 if (*flags & CPU_DTRACE_FAULT) 4261 break; 4262 4263 if (len != 0) { 4264 if ((end -= (len + 1)) < start) 4265 break; 4266 4267 *end = '@'; 4268 } 4269 4270 for (i = 1; i <= len; i++) 4271 end[i] = dtrace_load8((uintptr_t)s++); 4272 } 4273 4274 /* 4275 * Now for the node name... 4276 */ 4277 s = (char *)dtrace_loadptr(daddr + 4278 offsetof(struct dev_info, devi_node_name)); 4279 4280 daddr = dtrace_loadptr(daddr + 4281 offsetof(struct dev_info, devi_parent)); 4282 4283 /* 4284 * If our parent is NULL (that is, if we're the root 4285 * node), we're going to use the special path 4286 * "devices". 4287 */ 4288 if (daddr == 0) 4289 s = "devices"; 4290 4291 len = dtrace_strlen(s, size); 4292 if (*flags & CPU_DTRACE_FAULT) 4293 break; 4294 4295 if ((end -= (len + 1)) < start) 4296 break; 4297 4298 for (i = 1; i <= len; i++) 4299 end[i] = dtrace_load8((uintptr_t)s++); 4300 *end = '/'; 4301 4302 if (depth++ > dtrace_devdepth_max) { 4303 *flags |= CPU_DTRACE_ILLOP; 4304 break; 4305 } 4306 } 4307 4308 if (end < start) 4309 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4310 4311 if (daddr == 0) { 4312 regs[rd] = (uintptr_t)end; 4313 mstate->dtms_scratch_ptr += size; 4314 } 4315 4316 break; 4317 } 4318 #endif 4319 4320 case DIF_SUBR_STRJOIN: { 4321 char *d = (char *)mstate->dtms_scratch_ptr; 4322 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4323 uintptr_t s1 = tupregs[0].dttk_value; 4324 uintptr_t s2 = tupregs[1].dttk_value; 4325 int i = 0; 4326 4327 if (!dtrace_strcanload(s1, size, mstate, vstate) || 4328 !dtrace_strcanload(s2, size, mstate, vstate)) { 4329 regs[rd] = 0; 4330 break; 4331 } 4332 4333 if (!DTRACE_INSCRATCH(mstate, size)) { 4334 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4335 regs[rd] = 0; 4336 break; 4337 } 4338 4339 for (;;) { 4340 if (i >= size) { 4341 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4342 regs[rd] = 0; 4343 break; 4344 } 4345 4346 if ((d[i++] = dtrace_load8(s1++)) == '\0') { 4347 i--; 4348 break; 4349 } 4350 } 4351 4352 for (;;) { 4353 if (i >= size) { 4354 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4355 regs[rd] = 0; 4356 break; 4357 } 4358 4359 if ((d[i++] = dtrace_load8(s2++)) == '\0') 4360 break; 4361 } 4362 4363 if (i < size) { 4364 mstate->dtms_scratch_ptr += i; 4365 regs[rd] = (uintptr_t)d; 4366 } 4367 4368 break; 4369 } 4370 4371 case DIF_SUBR_LLTOSTR: { 4372 int64_t i = (int64_t)tupregs[0].dttk_value; 4373 uint64_t val, digit; 4374 uint64_t size = 65; /* enough room for 2^64 in binary */ 4375 char *end = (char *)mstate->dtms_scratch_ptr + size - 1; 4376 int base = 10; 4377 4378 if (nargs > 1) { 4379 if ((base = tupregs[1].dttk_value) <= 1 || 4380 base > ('z' - 'a' + 1) + ('9' - '0' + 1)) { 4381 *flags |= CPU_DTRACE_ILLOP; 4382 break; 4383 } 4384 } 4385 4386 val = (base == 10 && i < 0) ? i * -1 : i; 4387 4388 if (!DTRACE_INSCRATCH(mstate, size)) { 4389 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4390 regs[rd] = 0; 4391 break; 4392 } 4393 4394 for (*end-- = '\0'; val; val /= base) { 4395 if ((digit = val % base) <= '9' - '0') { 4396 *end-- = '0' + digit; 4397 } else { 4398 *end-- = 'a' + (digit - ('9' - '0') - 1); 4399 } 4400 } 4401 4402 if (i == 0 && base == 16) 4403 *end-- = '0'; 4404 4405 if (base == 16) 4406 *end-- = 'x'; 4407 4408 if (i == 0 || base == 8 || base == 16) 4409 *end-- = '0'; 4410 4411 if (i < 0 && base == 10) 4412 *end-- = '-'; 4413 4414 regs[rd] = (uintptr_t)end + 1; 4415 mstate->dtms_scratch_ptr += size; 4416 break; 4417 } 4418 4419 case DIF_SUBR_HTONS: 4420 case DIF_SUBR_NTOHS: 4421 #if BYTE_ORDER == BIG_ENDIAN 4422 regs[rd] = (uint16_t)tupregs[0].dttk_value; 4423 #else 4424 regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value); 4425 #endif 4426 break; 4427 4428 4429 case DIF_SUBR_HTONL: 4430 case DIF_SUBR_NTOHL: 4431 #if BYTE_ORDER == BIG_ENDIAN 4432 regs[rd] = (uint32_t)tupregs[0].dttk_value; 4433 #else 4434 regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value); 4435 #endif 4436 break; 4437 4438 4439 case DIF_SUBR_HTONLL: 4440 case DIF_SUBR_NTOHLL: 4441 #if BYTE_ORDER == BIG_ENDIAN 4442 regs[rd] = (uint64_t)tupregs[0].dttk_value; 4443 #else 4444 regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value); 4445 #endif 4446 break; 4447 4448 4449 case DIF_SUBR_DIRNAME: 4450 case DIF_SUBR_BASENAME: { 4451 char *dest = (char *)mstate->dtms_scratch_ptr; 4452 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4453 uintptr_t src = tupregs[0].dttk_value; 4454 int i, j, len = dtrace_strlen((char *)src, size); 4455 int lastbase = -1, firstbase = -1, lastdir = -1; 4456 int start, end; 4457 4458 if (!dtrace_canload(src, len + 1, mstate, vstate)) { 4459 regs[rd] = 0; 4460 break; 4461 } 4462 4463 if (!DTRACE_INSCRATCH(mstate, size)) { 4464 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4465 regs[rd] = 0; 4466 break; 4467 } 4468 4469 /* 4470 * The basename and dirname for a zero-length string is 4471 * defined to be "." 4472 */ 4473 if (len == 0) { 4474 len = 1; 4475 src = (uintptr_t)"."; 4476 } 4477 4478 /* 4479 * Start from the back of the string, moving back toward the 4480 * front until we see a character that isn't a slash. That 4481 * character is the last character in the basename. 4482 */ 4483 for (i = len - 1; i >= 0; i--) { 4484 if (dtrace_load8(src + i) != '/') 4485 break; 4486 } 4487 4488 if (i >= 0) 4489 lastbase = i; 4490 4491 /* 4492 * Starting from the last character in the basename, move 4493 * towards the front until we find a slash. The character 4494 * that we processed immediately before that is the first 4495 * character in the basename. 4496 */ 4497 for (; i >= 0; i--) { 4498 if (dtrace_load8(src + i) == '/') 4499 break; 4500 } 4501 4502 if (i >= 0) 4503 firstbase = i + 1; 4504 4505 /* 4506 * Now keep going until we find a non-slash character. That 4507 * character is the last character in the dirname. 4508 */ 4509 for (; i >= 0; i--) { 4510 if (dtrace_load8(src + i) != '/') 4511 break; 4512 } 4513 4514 if (i >= 0) 4515 lastdir = i; 4516 4517 ASSERT(!(lastbase == -1 && firstbase != -1)); 4518 ASSERT(!(firstbase == -1 && lastdir != -1)); 4519 4520 if (lastbase == -1) { 4521 /* 4522 * We didn't find a non-slash character. We know that 4523 * the length is non-zero, so the whole string must be 4524 * slashes. In either the dirname or the basename 4525 * case, we return '/'. 4526 */ 4527 ASSERT(firstbase == -1); 4528 firstbase = lastbase = lastdir = 0; 4529 } 4530 4531 if (firstbase == -1) { 4532 /* 4533 * The entire string consists only of a basename 4534 * component. If we're looking for dirname, we need 4535 * to change our string to be just "."; if we're 4536 * looking for a basename, we'll just set the first 4537 * character of the basename to be 0. 4538 */ 4539 if (subr == DIF_SUBR_DIRNAME) { 4540 ASSERT(lastdir == -1); 4541 src = (uintptr_t)"."; 4542 lastdir = 0; 4543 } else { 4544 firstbase = 0; 4545 } 4546 } 4547 4548 if (subr == DIF_SUBR_DIRNAME) { 4549 if (lastdir == -1) { 4550 /* 4551 * We know that we have a slash in the name -- 4552 * or lastdir would be set to 0, above. And 4553 * because lastdir is -1, we know that this 4554 * slash must be the first character. (That 4555 * is, the full string must be of the form 4556 * "/basename".) In this case, the last 4557 * character of the directory name is 0. 4558 */ 4559 lastdir = 0; 4560 } 4561 4562 start = 0; 4563 end = lastdir; 4564 } else { 4565 ASSERT(subr == DIF_SUBR_BASENAME); 4566 ASSERT(firstbase != -1 && lastbase != -1); 4567 start = firstbase; 4568 end = lastbase; 4569 } 4570 4571 for (i = start, j = 0; i <= end && j < size - 1; i++, j++) 4572 dest[j] = dtrace_load8(src + i); 4573 4574 dest[j] = '\0'; 4575 regs[rd] = (uintptr_t)dest; 4576 mstate->dtms_scratch_ptr += size; 4577 break; 4578 } 4579 4580 case DIF_SUBR_CLEANPATH: { 4581 char *dest = (char *)mstate->dtms_scratch_ptr, c; 4582 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4583 uintptr_t src = tupregs[0].dttk_value; 4584 int i = 0, j = 0; 4585 4586 if (!dtrace_strcanload(src, size, mstate, vstate)) { 4587 regs[rd] = 0; 4588 break; 4589 } 4590 4591 if (!DTRACE_INSCRATCH(mstate, size)) { 4592 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4593 regs[rd] = 0; 4594 break; 4595 } 4596 4597 /* 4598 * Move forward, loading each character. 4599 */ 4600 do { 4601 c = dtrace_load8(src + i++); 4602 next: 4603 if (j + 5 >= size) /* 5 = strlen("/..c\0") */ 4604 break; 4605 4606 if (c != '/') { 4607 dest[j++] = c; 4608 continue; 4609 } 4610 4611 c = dtrace_load8(src + i++); 4612 4613 if (c == '/') { 4614 /* 4615 * We have two slashes -- we can just advance 4616 * to the next character. 4617 */ 4618 goto next; 4619 } 4620 4621 if (c != '.') { 4622 /* 4623 * This is not "." and it's not ".." -- we can 4624 * just store the "/" and this character and 4625 * drive on. 4626 */ 4627 dest[j++] = '/'; 4628 dest[j++] = c; 4629 continue; 4630 } 4631 4632 c = dtrace_load8(src + i++); 4633 4634 if (c == '/') { 4635 /* 4636 * This is a "/./" component. We're not going 4637 * to store anything in the destination buffer; 4638 * we're just going to go to the next component. 4639 */ 4640 goto next; 4641 } 4642 4643 if (c != '.') { 4644 /* 4645 * This is not ".." -- we can just store the 4646 * "/." and this character and continue 4647 * processing. 4648 */ 4649 dest[j++] = '/'; 4650 dest[j++] = '.'; 4651 dest[j++] = c; 4652 continue; 4653 } 4654 4655 c = dtrace_load8(src + i++); 4656 4657 if (c != '/' && c != '\0') { 4658 /* 4659 * This is not ".." -- it's "..[mumble]". 4660 * We'll store the "/.." and this character 4661 * and continue processing. 4662 */ 4663 dest[j++] = '/'; 4664 dest[j++] = '.'; 4665 dest[j++] = '.'; 4666 dest[j++] = c; 4667 continue; 4668 } 4669 4670 /* 4671 * This is "/../" or "/..\0". We need to back up 4672 * our destination pointer until we find a "/". 4673 */ 4674 i--; 4675 while (j != 0 && dest[--j] != '/') 4676 continue; 4677 4678 if (c == '\0') 4679 dest[++j] = '/'; 4680 } while (c != '\0'); 4681 4682 dest[j] = '\0'; 4683 regs[rd] = (uintptr_t)dest; 4684 mstate->dtms_scratch_ptr += size; 4685 break; 4686 } 4687 4688 case DIF_SUBR_INET_NTOA: 4689 case DIF_SUBR_INET_NTOA6: 4690 case DIF_SUBR_INET_NTOP: { 4691 size_t size; 4692 int af, argi, i; 4693 char *base, *end; 4694 4695 if (subr == DIF_SUBR_INET_NTOP) { 4696 af = (int)tupregs[0].dttk_value; 4697 argi = 1; 4698 } else { 4699 af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6; 4700 argi = 0; 4701 } 4702 4703 if (af == AF_INET) { 4704 ipaddr_t ip4; 4705 uint8_t *ptr8, val; 4706 4707 /* 4708 * Safely load the IPv4 address. 4709 */ 4710 ip4 = dtrace_load32(tupregs[argi].dttk_value); 4711 4712 /* 4713 * Check an IPv4 string will fit in scratch. 4714 */ 4715 size = INET_ADDRSTRLEN; 4716 if (!DTRACE_INSCRATCH(mstate, size)) { 4717 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4718 regs[rd] = 0; 4719 break; 4720 } 4721 base = (char *)mstate->dtms_scratch_ptr; 4722 end = (char *)mstate->dtms_scratch_ptr + size - 1; 4723 4724 /* 4725 * Stringify as a dotted decimal quad. 4726 */ 4727 *end-- = '\0'; 4728 ptr8 = (uint8_t *)&ip4; 4729 for (i = 3; i >= 0; i--) { 4730 val = ptr8[i]; 4731 4732 if (val == 0) { 4733 *end-- = '0'; 4734 } else { 4735 for (; val; val /= 10) { 4736 *end-- = '0' + (val % 10); 4737 } 4738 } 4739 4740 if (i > 0) 4741 *end-- = '.'; 4742 } 4743 ASSERT(end + 1 >= base); 4744 4745 } else if (af == AF_INET6) { 4746 struct in6_addr ip6; 4747 int firstzero, tryzero, numzero, v6end; 4748 uint16_t val; 4749 const char digits[] = "0123456789abcdef"; 4750 4751 /* 4752 * Stringify using RFC 1884 convention 2 - 16 bit 4753 * hexadecimal values with a zero-run compression. 4754 * Lower case hexadecimal digits are used. 4755 * eg, fe80::214:4fff:fe0b:76c8. 4756 * The IPv4 embedded form is returned for inet_ntop, 4757 * just the IPv4 string is returned for inet_ntoa6. 4758 */ 4759 4760 /* 4761 * Safely load the IPv6 address. 4762 */ 4763 dtrace_bcopy( 4764 (void *)(uintptr_t)tupregs[argi].dttk_value, 4765 (void *)(uintptr_t)&ip6, sizeof (struct in6_addr)); 4766 4767 /* 4768 * Check an IPv6 string will fit in scratch. 4769 */ 4770 size = INET6_ADDRSTRLEN; 4771 if (!DTRACE_INSCRATCH(mstate, size)) { 4772 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4773 regs[rd] = 0; 4774 break; 4775 } 4776 base = (char *)mstate->dtms_scratch_ptr; 4777 end = (char *)mstate->dtms_scratch_ptr + size - 1; 4778 *end-- = '\0'; 4779 4780 /* 4781 * Find the longest run of 16 bit zero values 4782 * for the single allowed zero compression - "::". 4783 */ 4784 firstzero = -1; 4785 tryzero = -1; 4786 numzero = 1; 4787 for (i = 0; i < sizeof (struct in6_addr); i++) { 4788 #if defined(sun) 4789 if (ip6._S6_un._S6_u8[i] == 0 && 4790 #else 4791 if (ip6.__u6_addr.__u6_addr8[i] == 0 && 4792 #endif 4793 tryzero == -1 && i % 2 == 0) { 4794 tryzero = i; 4795 continue; 4796 } 4797 4798 if (tryzero != -1 && 4799 #if defined(sun) 4800 (ip6._S6_un._S6_u8[i] != 0 || 4801 #else 4802 (ip6.__u6_addr.__u6_addr8[i] != 0 || 4803 #endif 4804 i == sizeof (struct in6_addr) - 1)) { 4805 4806 if (i - tryzero <= numzero) { 4807 tryzero = -1; 4808 continue; 4809 } 4810 4811 firstzero = tryzero; 4812 numzero = i - i % 2 - tryzero; 4813 tryzero = -1; 4814 4815 #if defined(sun) 4816 if (ip6._S6_un._S6_u8[i] == 0 && 4817 #else 4818 if (ip6.__u6_addr.__u6_addr8[i] == 0 && 4819 #endif 4820 i == sizeof (struct in6_addr) - 1) 4821 numzero += 2; 4822 } 4823 } 4824 ASSERT(firstzero + numzero <= sizeof (struct in6_addr)); 4825 4826 /* 4827 * Check for an IPv4 embedded address. 4828 */ 4829 v6end = sizeof (struct in6_addr) - 2; 4830 if (IN6_IS_ADDR_V4MAPPED(&ip6) || 4831 IN6_IS_ADDR_V4COMPAT(&ip6)) { 4832 for (i = sizeof (struct in6_addr) - 1; 4833 i >= DTRACE_V4MAPPED_OFFSET; i--) { 4834 ASSERT(end >= base); 4835 4836 #if defined(sun) 4837 val = ip6._S6_un._S6_u8[i]; 4838 #else 4839 val = ip6.__u6_addr.__u6_addr8[i]; 4840 #endif 4841 4842 if (val == 0) { 4843 *end-- = '0'; 4844 } else { 4845 for (; val; val /= 10) { 4846 *end-- = '0' + val % 10; 4847 } 4848 } 4849 4850 if (i > DTRACE_V4MAPPED_OFFSET) 4851 *end-- = '.'; 4852 } 4853 4854 if (subr == DIF_SUBR_INET_NTOA6) 4855 goto inetout; 4856 4857 /* 4858 * Set v6end to skip the IPv4 address that 4859 * we have already stringified. 4860 */ 4861 v6end = 10; 4862 } 4863 4864 /* 4865 * Build the IPv6 string by working through the 4866 * address in reverse. 4867 */ 4868 for (i = v6end; i >= 0; i -= 2) { 4869 ASSERT(end >= base); 4870 4871 if (i == firstzero + numzero - 2) { 4872 *end-- = ':'; 4873 *end-- = ':'; 4874 i -= numzero - 2; 4875 continue; 4876 } 4877 4878 if (i < 14 && i != firstzero - 2) 4879 *end-- = ':'; 4880 4881 #if defined(sun) 4882 val = (ip6._S6_un._S6_u8[i] << 8) + 4883 ip6._S6_un._S6_u8[i + 1]; 4884 #else 4885 val = (ip6.__u6_addr.__u6_addr8[i] << 8) + 4886 ip6.__u6_addr.__u6_addr8[i + 1]; 4887 #endif 4888 4889 if (val == 0) { 4890 *end-- = '0'; 4891 } else { 4892 for (; val; val /= 16) { 4893 *end-- = digits[val % 16]; 4894 } 4895 } 4896 } 4897 ASSERT(end + 1 >= base); 4898 4899 } else { 4900 /* 4901 * The user didn't use AH_INET or AH_INET6. 4902 */ 4903 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 4904 regs[rd] = 0; 4905 break; 4906 } 4907 4908 inetout: regs[rd] = (uintptr_t)end + 1; 4909 mstate->dtms_scratch_ptr += size; 4910 break; 4911 } 4912 4913 case DIF_SUBR_MEMREF: { 4914 uintptr_t size = 2 * sizeof(uintptr_t); 4915 uintptr_t *memref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t)); 4916 size_t scratch_size = ((uintptr_t) memref - mstate->dtms_scratch_ptr) + size; 4917 4918 /* address and length */ 4919 memref[0] = tupregs[0].dttk_value; 4920 memref[1] = tupregs[1].dttk_value; 4921 4922 regs[rd] = (uintptr_t) memref; 4923 mstate->dtms_scratch_ptr += scratch_size; 4924 break; 4925 } 4926 4927 #if !defined(sun) 4928 case DIF_SUBR_MEMSTR: { 4929 char *str = (char *)mstate->dtms_scratch_ptr; 4930 uintptr_t mem = tupregs[0].dttk_value; 4931 char c = tupregs[1].dttk_value; 4932 size_t size = tupregs[2].dttk_value; 4933 uint8_t n; 4934 int i; 4935 4936 regs[rd] = 0; 4937 4938 if (size == 0) 4939 break; 4940 4941 if (!dtrace_canload(mem, size - 1, mstate, vstate)) 4942 break; 4943 4944 if (!DTRACE_INSCRATCH(mstate, size)) { 4945 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4946 break; 4947 } 4948 4949 if (dtrace_memstr_max != 0 && size > dtrace_memstr_max) { 4950 *flags |= CPU_DTRACE_ILLOP; 4951 break; 4952 } 4953 4954 for (i = 0; i < size - 1; i++) { 4955 n = dtrace_load8(mem++); 4956 str[i] = (n == 0) ? c : n; 4957 } 4958 str[size - 1] = 0; 4959 4960 regs[rd] = (uintptr_t)str; 4961 mstate->dtms_scratch_ptr += size; 4962 break; 4963 } 4964 #endif 4965 4966 case DIF_SUBR_TYPEREF: { 4967 uintptr_t size = 4 * sizeof(uintptr_t); 4968 uintptr_t *typeref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t)); 4969 size_t scratch_size = ((uintptr_t) typeref - mstate->dtms_scratch_ptr) + size; 4970 4971 /* address, num_elements, type_str, type_len */ 4972 typeref[0] = tupregs[0].dttk_value; 4973 typeref[1] = tupregs[1].dttk_value; 4974 typeref[2] = tupregs[2].dttk_value; 4975 typeref[3] = tupregs[3].dttk_value; 4976 4977 regs[rd] = (uintptr_t) typeref; 4978 mstate->dtms_scratch_ptr += scratch_size; 4979 break; 4980 } 4981 } 4982 } 4983 4984 /* 4985 * Emulate the execution of DTrace IR instructions specified by the given 4986 * DIF object. This function is deliberately void of assertions as all of 4987 * the necessary checks are handled by a call to dtrace_difo_validate(). 4988 */ 4989 static uint64_t 4990 dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate, 4991 dtrace_vstate_t *vstate, dtrace_state_t *state) 4992 { 4993 const dif_instr_t *text = difo->dtdo_buf; 4994 const uint_t textlen = difo->dtdo_len; 4995 const char *strtab = difo->dtdo_strtab; 4996 const uint64_t *inttab = difo->dtdo_inttab; 4997 4998 uint64_t rval = 0; 4999 dtrace_statvar_t *svar; 5000 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; 5001 dtrace_difv_t *v; 5002 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 5003 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; 5004 5005 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ 5006 uint64_t regs[DIF_DIR_NREGS]; 5007 uint64_t *tmp; 5008 5009 uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0; 5010 int64_t cc_r; 5011 uint_t pc = 0, id, opc = 0; 5012 uint8_t ttop = 0; 5013 dif_instr_t instr; 5014 uint_t r1, r2, rd; 5015 5016 /* 5017 * We stash the current DIF object into the machine state: we need it 5018 * for subsequent access checking. 5019 */ 5020 mstate->dtms_difo = difo; 5021 5022 regs[DIF_REG_R0] = 0; /* %r0 is fixed at zero */ 5023 5024 while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) { 5025 opc = pc; 5026 5027 instr = text[pc++]; 5028 r1 = DIF_INSTR_R1(instr); 5029 r2 = DIF_INSTR_R2(instr); 5030 rd = DIF_INSTR_RD(instr); 5031 5032 switch (DIF_INSTR_OP(instr)) { 5033 case DIF_OP_OR: 5034 regs[rd] = regs[r1] | regs[r2]; 5035 break; 5036 case DIF_OP_XOR: 5037 regs[rd] = regs[r1] ^ regs[r2]; 5038 break; 5039 case DIF_OP_AND: 5040 regs[rd] = regs[r1] & regs[r2]; 5041 break; 5042 case DIF_OP_SLL: 5043 regs[rd] = regs[r1] << regs[r2]; 5044 break; 5045 case DIF_OP_SRL: 5046 regs[rd] = regs[r1] >> regs[r2]; 5047 break; 5048 case DIF_OP_SUB: 5049 regs[rd] = regs[r1] - regs[r2]; 5050 break; 5051 case DIF_OP_ADD: 5052 regs[rd] = regs[r1] + regs[r2]; 5053 break; 5054 case DIF_OP_MUL: 5055 regs[rd] = regs[r1] * regs[r2]; 5056 break; 5057 case DIF_OP_SDIV: 5058 if (regs[r2] == 0) { 5059 regs[rd] = 0; 5060 *flags |= CPU_DTRACE_DIVZERO; 5061 } else { 5062 regs[rd] = (int64_t)regs[r1] / 5063 (int64_t)regs[r2]; 5064 } 5065 break; 5066 5067 case DIF_OP_UDIV: 5068 if (regs[r2] == 0) { 5069 regs[rd] = 0; 5070 *flags |= CPU_DTRACE_DIVZERO; 5071 } else { 5072 regs[rd] = regs[r1] / regs[r2]; 5073 } 5074 break; 5075 5076 case DIF_OP_SREM: 5077 if (regs[r2] == 0) { 5078 regs[rd] = 0; 5079 *flags |= CPU_DTRACE_DIVZERO; 5080 } else { 5081 regs[rd] = (int64_t)regs[r1] % 5082 (int64_t)regs[r2]; 5083 } 5084 break; 5085 5086 case DIF_OP_UREM: 5087 if (regs[r2] == 0) { 5088 regs[rd] = 0; 5089 *flags |= CPU_DTRACE_DIVZERO; 5090 } else { 5091 regs[rd] = regs[r1] % regs[r2]; 5092 } 5093 break; 5094 5095 case DIF_OP_NOT: 5096 regs[rd] = ~regs[r1]; 5097 break; 5098 case DIF_OP_MOV: 5099 regs[rd] = regs[r1]; 5100 break; 5101 case DIF_OP_CMP: 5102 cc_r = regs[r1] - regs[r2]; 5103 cc_n = cc_r < 0; 5104 cc_z = cc_r == 0; 5105 cc_v = 0; 5106 cc_c = regs[r1] < regs[r2]; 5107 break; 5108 case DIF_OP_TST: 5109 cc_n = cc_v = cc_c = 0; 5110 cc_z = regs[r1] == 0; 5111 break; 5112 case DIF_OP_BA: 5113 pc = DIF_INSTR_LABEL(instr); 5114 break; 5115 case DIF_OP_BE: 5116 if (cc_z) 5117 pc = DIF_INSTR_LABEL(instr); 5118 break; 5119 case DIF_OP_BNE: 5120 if (cc_z == 0) 5121 pc = DIF_INSTR_LABEL(instr); 5122 break; 5123 case DIF_OP_BG: 5124 if ((cc_z | (cc_n ^ cc_v)) == 0) 5125 pc = DIF_INSTR_LABEL(instr); 5126 break; 5127 case DIF_OP_BGU: 5128 if ((cc_c | cc_z) == 0) 5129 pc = DIF_INSTR_LABEL(instr); 5130 break; 5131 case DIF_OP_BGE: 5132 if ((cc_n ^ cc_v) == 0) 5133 pc = DIF_INSTR_LABEL(instr); 5134 break; 5135 case DIF_OP_BGEU: 5136 if (cc_c == 0) 5137 pc = DIF_INSTR_LABEL(instr); 5138 break; 5139 case DIF_OP_BL: 5140 if (cc_n ^ cc_v) 5141 pc = DIF_INSTR_LABEL(instr); 5142 break; 5143 case DIF_OP_BLU: 5144 if (cc_c) 5145 pc = DIF_INSTR_LABEL(instr); 5146 break; 5147 case DIF_OP_BLE: 5148 if (cc_z | (cc_n ^ cc_v)) 5149 pc = DIF_INSTR_LABEL(instr); 5150 break; 5151 case DIF_OP_BLEU: 5152 if (cc_c | cc_z) 5153 pc = DIF_INSTR_LABEL(instr); 5154 break; 5155 case DIF_OP_RLDSB: 5156 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { 5157 *flags |= CPU_DTRACE_KPRIV; 5158 *illval = regs[r1]; 5159 break; 5160 } 5161 /*FALLTHROUGH*/ 5162 case DIF_OP_LDSB: 5163 regs[rd] = (int8_t)dtrace_load8(regs[r1]); 5164 break; 5165 case DIF_OP_RLDSH: 5166 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { 5167 *flags |= CPU_DTRACE_KPRIV; 5168 *illval = regs[r1]; 5169 break; 5170 } 5171 /*FALLTHROUGH*/ 5172 case DIF_OP_LDSH: 5173 regs[rd] = (int16_t)dtrace_load16(regs[r1]); 5174 break; 5175 case DIF_OP_RLDSW: 5176 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { 5177 *flags |= CPU_DTRACE_KPRIV; 5178 *illval = regs[r1]; 5179 break; 5180 } 5181 /*FALLTHROUGH*/ 5182 case DIF_OP_LDSW: 5183 regs[rd] = (int32_t)dtrace_load32(regs[r1]); 5184 break; 5185 case DIF_OP_RLDUB: 5186 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { 5187 *flags |= CPU_DTRACE_KPRIV; 5188 *illval = regs[r1]; 5189 break; 5190 } 5191 /*FALLTHROUGH*/ 5192 case DIF_OP_LDUB: 5193 regs[rd] = dtrace_load8(regs[r1]); 5194 break; 5195 case DIF_OP_RLDUH: 5196 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { 5197 *flags |= CPU_DTRACE_KPRIV; 5198 *illval = regs[r1]; 5199 break; 5200 } 5201 /*FALLTHROUGH*/ 5202 case DIF_OP_LDUH: 5203 regs[rd] = dtrace_load16(regs[r1]); 5204 break; 5205 case DIF_OP_RLDUW: 5206 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { 5207 *flags |= CPU_DTRACE_KPRIV; 5208 *illval = regs[r1]; 5209 break; 5210 } 5211 /*FALLTHROUGH*/ 5212 case DIF_OP_LDUW: 5213 regs[rd] = dtrace_load32(regs[r1]); 5214 break; 5215 case DIF_OP_RLDX: 5216 if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) { 5217 *flags |= CPU_DTRACE_KPRIV; 5218 *illval = regs[r1]; 5219 break; 5220 } 5221 /*FALLTHROUGH*/ 5222 case DIF_OP_LDX: 5223 regs[rd] = dtrace_load64(regs[r1]); 5224 break; 5225 case DIF_OP_ULDSB: 5226 regs[rd] = (int8_t) 5227 dtrace_fuword8((void *)(uintptr_t)regs[r1]); 5228 break; 5229 case DIF_OP_ULDSH: 5230 regs[rd] = (int16_t) 5231 dtrace_fuword16((void *)(uintptr_t)regs[r1]); 5232 break; 5233 case DIF_OP_ULDSW: 5234 regs[rd] = (int32_t) 5235 dtrace_fuword32((void *)(uintptr_t)regs[r1]); 5236 break; 5237 case DIF_OP_ULDUB: 5238 regs[rd] = 5239 dtrace_fuword8((void *)(uintptr_t)regs[r1]); 5240 break; 5241 case DIF_OP_ULDUH: 5242 regs[rd] = 5243 dtrace_fuword16((void *)(uintptr_t)regs[r1]); 5244 break; 5245 case DIF_OP_ULDUW: 5246 regs[rd] = 5247 dtrace_fuword32((void *)(uintptr_t)regs[r1]); 5248 break; 5249 case DIF_OP_ULDX: 5250 regs[rd] = 5251 dtrace_fuword64((void *)(uintptr_t)regs[r1]); 5252 break; 5253 case DIF_OP_RET: 5254 rval = regs[rd]; 5255 pc = textlen; 5256 break; 5257 case DIF_OP_NOP: 5258 break; 5259 case DIF_OP_SETX: 5260 regs[rd] = inttab[DIF_INSTR_INTEGER(instr)]; 5261 break; 5262 case DIF_OP_SETS: 5263 regs[rd] = (uint64_t)(uintptr_t) 5264 (strtab + DIF_INSTR_STRING(instr)); 5265 break; 5266 case DIF_OP_SCMP: { 5267 size_t sz = state->dts_options[DTRACEOPT_STRSIZE]; 5268 uintptr_t s1 = regs[r1]; 5269 uintptr_t s2 = regs[r2]; 5270 5271 if (s1 != 0 && 5272 !dtrace_strcanload(s1, sz, mstate, vstate)) 5273 break; 5274 if (s2 != 0 && 5275 !dtrace_strcanload(s2, sz, mstate, vstate)) 5276 break; 5277 5278 cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz); 5279 5280 cc_n = cc_r < 0; 5281 cc_z = cc_r == 0; 5282 cc_v = cc_c = 0; 5283 break; 5284 } 5285 case DIF_OP_LDGA: 5286 regs[rd] = dtrace_dif_variable(mstate, state, 5287 r1, regs[r2]); 5288 break; 5289 case DIF_OP_LDGS: 5290 id = DIF_INSTR_VAR(instr); 5291 5292 if (id >= DIF_VAR_OTHER_UBASE) { 5293 uintptr_t a; 5294 5295 id -= DIF_VAR_OTHER_UBASE; 5296 svar = vstate->dtvs_globals[id]; 5297 ASSERT(svar != NULL); 5298 v = &svar->dtsv_var; 5299 5300 if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) { 5301 regs[rd] = svar->dtsv_data; 5302 break; 5303 } 5304 5305 a = (uintptr_t)svar->dtsv_data; 5306 5307 if (*(uint8_t *)a == UINT8_MAX) { 5308 /* 5309 * If the 0th byte is set to UINT8_MAX 5310 * then this is to be treated as a 5311 * reference to a NULL variable. 5312 */ 5313 regs[rd] = 0; 5314 } else { 5315 regs[rd] = a + sizeof (uint64_t); 5316 } 5317 5318 break; 5319 } 5320 5321 regs[rd] = dtrace_dif_variable(mstate, state, id, 0); 5322 break; 5323 5324 case DIF_OP_STGS: 5325 id = DIF_INSTR_VAR(instr); 5326 5327 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5328 id -= DIF_VAR_OTHER_UBASE; 5329 5330 svar = vstate->dtvs_globals[id]; 5331 ASSERT(svar != NULL); 5332 v = &svar->dtsv_var; 5333 5334 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5335 uintptr_t a = (uintptr_t)svar->dtsv_data; 5336 5337 ASSERT(a != 0); 5338 ASSERT(svar->dtsv_size != 0); 5339 5340 if (regs[rd] == 0) { 5341 *(uint8_t *)a = UINT8_MAX; 5342 break; 5343 } else { 5344 *(uint8_t *)a = 0; 5345 a += sizeof (uint64_t); 5346 } 5347 if (!dtrace_vcanload( 5348 (void *)(uintptr_t)regs[rd], &v->dtdv_type, 5349 mstate, vstate)) 5350 break; 5351 5352 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5353 (void *)a, &v->dtdv_type); 5354 break; 5355 } 5356 5357 svar->dtsv_data = regs[rd]; 5358 break; 5359 5360 case DIF_OP_LDTA: 5361 /* 5362 * There are no DTrace built-in thread-local arrays at 5363 * present. This opcode is saved for future work. 5364 */ 5365 *flags |= CPU_DTRACE_ILLOP; 5366 regs[rd] = 0; 5367 break; 5368 5369 case DIF_OP_LDLS: 5370 id = DIF_INSTR_VAR(instr); 5371 5372 if (id < DIF_VAR_OTHER_UBASE) { 5373 /* 5374 * For now, this has no meaning. 5375 */ 5376 regs[rd] = 0; 5377 break; 5378 } 5379 5380 id -= DIF_VAR_OTHER_UBASE; 5381 5382 ASSERT(id < vstate->dtvs_nlocals); 5383 ASSERT(vstate->dtvs_locals != NULL); 5384 5385 svar = vstate->dtvs_locals[id]; 5386 ASSERT(svar != NULL); 5387 v = &svar->dtsv_var; 5388 5389 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5390 uintptr_t a = (uintptr_t)svar->dtsv_data; 5391 size_t sz = v->dtdv_type.dtdt_size; 5392 5393 sz += sizeof (uint64_t); 5394 ASSERT(svar->dtsv_size == NCPU * sz); 5395 a += curcpu * sz; 5396 5397 if (*(uint8_t *)a == UINT8_MAX) { 5398 /* 5399 * If the 0th byte is set to UINT8_MAX 5400 * then this is to be treated as a 5401 * reference to a NULL variable. 5402 */ 5403 regs[rd] = 0; 5404 } else { 5405 regs[rd] = a + sizeof (uint64_t); 5406 } 5407 5408 break; 5409 } 5410 5411 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); 5412 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; 5413 regs[rd] = tmp[curcpu]; 5414 break; 5415 5416 case DIF_OP_STLS: 5417 id = DIF_INSTR_VAR(instr); 5418 5419 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5420 id -= DIF_VAR_OTHER_UBASE; 5421 ASSERT(id < vstate->dtvs_nlocals); 5422 5423 ASSERT(vstate->dtvs_locals != NULL); 5424 svar = vstate->dtvs_locals[id]; 5425 ASSERT(svar != NULL); 5426 v = &svar->dtsv_var; 5427 5428 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5429 uintptr_t a = (uintptr_t)svar->dtsv_data; 5430 size_t sz = v->dtdv_type.dtdt_size; 5431 5432 sz += sizeof (uint64_t); 5433 ASSERT(svar->dtsv_size == NCPU * sz); 5434 a += curcpu * sz; 5435 5436 if (regs[rd] == 0) { 5437 *(uint8_t *)a = UINT8_MAX; 5438 break; 5439 } else { 5440 *(uint8_t *)a = 0; 5441 a += sizeof (uint64_t); 5442 } 5443 5444 if (!dtrace_vcanload( 5445 (void *)(uintptr_t)regs[rd], &v->dtdv_type, 5446 mstate, vstate)) 5447 break; 5448 5449 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5450 (void *)a, &v->dtdv_type); 5451 break; 5452 } 5453 5454 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); 5455 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; 5456 tmp[curcpu] = regs[rd]; 5457 break; 5458 5459 case DIF_OP_LDTS: { 5460 dtrace_dynvar_t *dvar; 5461 dtrace_key_t *key; 5462 5463 id = DIF_INSTR_VAR(instr); 5464 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5465 id -= DIF_VAR_OTHER_UBASE; 5466 v = &vstate->dtvs_tlocals[id]; 5467 5468 key = &tupregs[DIF_DTR_NREGS]; 5469 key[0].dttk_value = (uint64_t)id; 5470 key[0].dttk_size = 0; 5471 DTRACE_TLS_THRKEY(key[1].dttk_value); 5472 key[1].dttk_size = 0; 5473 5474 dvar = dtrace_dynvar(dstate, 2, key, 5475 sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC, 5476 mstate, vstate); 5477 5478 if (dvar == NULL) { 5479 regs[rd] = 0; 5480 break; 5481 } 5482 5483 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5484 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; 5485 } else { 5486 regs[rd] = *((uint64_t *)dvar->dtdv_data); 5487 } 5488 5489 break; 5490 } 5491 5492 case DIF_OP_STTS: { 5493 dtrace_dynvar_t *dvar; 5494 dtrace_key_t *key; 5495 5496 id = DIF_INSTR_VAR(instr); 5497 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5498 id -= DIF_VAR_OTHER_UBASE; 5499 5500 key = &tupregs[DIF_DTR_NREGS]; 5501 key[0].dttk_value = (uint64_t)id; 5502 key[0].dttk_size = 0; 5503 DTRACE_TLS_THRKEY(key[1].dttk_value); 5504 key[1].dttk_size = 0; 5505 v = &vstate->dtvs_tlocals[id]; 5506 5507 dvar = dtrace_dynvar(dstate, 2, key, 5508 v->dtdv_type.dtdt_size > sizeof (uint64_t) ? 5509 v->dtdv_type.dtdt_size : sizeof (uint64_t), 5510 regs[rd] ? DTRACE_DYNVAR_ALLOC : 5511 DTRACE_DYNVAR_DEALLOC, mstate, vstate); 5512 5513 /* 5514 * Given that we're storing to thread-local data, 5515 * we need to flush our predicate cache. 5516 */ 5517 curthread->t_predcache = 0; 5518 5519 if (dvar == NULL) 5520 break; 5521 5522 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5523 if (!dtrace_vcanload( 5524 (void *)(uintptr_t)regs[rd], 5525 &v->dtdv_type, mstate, vstate)) 5526 break; 5527 5528 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5529 dvar->dtdv_data, &v->dtdv_type); 5530 } else { 5531 *((uint64_t *)dvar->dtdv_data) = regs[rd]; 5532 } 5533 5534 break; 5535 } 5536 5537 case DIF_OP_SRA: 5538 regs[rd] = (int64_t)regs[r1] >> regs[r2]; 5539 break; 5540 5541 case DIF_OP_CALL: 5542 dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd, 5543 regs, tupregs, ttop, mstate, state); 5544 break; 5545 5546 case DIF_OP_PUSHTR: 5547 if (ttop == DIF_DTR_NREGS) { 5548 *flags |= CPU_DTRACE_TUPOFLOW; 5549 break; 5550 } 5551 5552 if (r1 == DIF_TYPE_STRING) { 5553 /* 5554 * If this is a string type and the size is 0, 5555 * we'll use the system-wide default string 5556 * size. Note that we are _not_ looking at 5557 * the value of the DTRACEOPT_STRSIZE option; 5558 * had this been set, we would expect to have 5559 * a non-zero size value in the "pushtr". 5560 */ 5561 tupregs[ttop].dttk_size = 5562 dtrace_strlen((char *)(uintptr_t)regs[rd], 5563 regs[r2] ? regs[r2] : 5564 dtrace_strsize_default) + 1; 5565 } else { 5566 tupregs[ttop].dttk_size = regs[r2]; 5567 } 5568 5569 tupregs[ttop++].dttk_value = regs[rd]; 5570 break; 5571 5572 case DIF_OP_PUSHTV: 5573 if (ttop == DIF_DTR_NREGS) { 5574 *flags |= CPU_DTRACE_TUPOFLOW; 5575 break; 5576 } 5577 5578 tupregs[ttop].dttk_value = regs[rd]; 5579 tupregs[ttop++].dttk_size = 0; 5580 break; 5581 5582 case DIF_OP_POPTS: 5583 if (ttop != 0) 5584 ttop--; 5585 break; 5586 5587 case DIF_OP_FLUSHTS: 5588 ttop = 0; 5589 break; 5590 5591 case DIF_OP_LDGAA: 5592 case DIF_OP_LDTAA: { 5593 dtrace_dynvar_t *dvar; 5594 dtrace_key_t *key = tupregs; 5595 uint_t nkeys = ttop; 5596 5597 id = DIF_INSTR_VAR(instr); 5598 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5599 id -= DIF_VAR_OTHER_UBASE; 5600 5601 key[nkeys].dttk_value = (uint64_t)id; 5602 key[nkeys++].dttk_size = 0; 5603 5604 if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) { 5605 DTRACE_TLS_THRKEY(key[nkeys].dttk_value); 5606 key[nkeys++].dttk_size = 0; 5607 v = &vstate->dtvs_tlocals[id]; 5608 } else { 5609 v = &vstate->dtvs_globals[id]->dtsv_var; 5610 } 5611 5612 dvar = dtrace_dynvar(dstate, nkeys, key, 5613 v->dtdv_type.dtdt_size > sizeof (uint64_t) ? 5614 v->dtdv_type.dtdt_size : sizeof (uint64_t), 5615 DTRACE_DYNVAR_NOALLOC, mstate, vstate); 5616 5617 if (dvar == NULL) { 5618 regs[rd] = 0; 5619 break; 5620 } 5621 5622 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5623 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; 5624 } else { 5625 regs[rd] = *((uint64_t *)dvar->dtdv_data); 5626 } 5627 5628 break; 5629 } 5630 5631 case DIF_OP_STGAA: 5632 case DIF_OP_STTAA: { 5633 dtrace_dynvar_t *dvar; 5634 dtrace_key_t *key = tupregs; 5635 uint_t nkeys = ttop; 5636 5637 id = DIF_INSTR_VAR(instr); 5638 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5639 id -= DIF_VAR_OTHER_UBASE; 5640 5641 key[nkeys].dttk_value = (uint64_t)id; 5642 key[nkeys++].dttk_size = 0; 5643 5644 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) { 5645 DTRACE_TLS_THRKEY(key[nkeys].dttk_value); 5646 key[nkeys++].dttk_size = 0; 5647 v = &vstate->dtvs_tlocals[id]; 5648 } else { 5649 v = &vstate->dtvs_globals[id]->dtsv_var; 5650 } 5651 5652 dvar = dtrace_dynvar(dstate, nkeys, key, 5653 v->dtdv_type.dtdt_size > sizeof (uint64_t) ? 5654 v->dtdv_type.dtdt_size : sizeof (uint64_t), 5655 regs[rd] ? DTRACE_DYNVAR_ALLOC : 5656 DTRACE_DYNVAR_DEALLOC, mstate, vstate); 5657 5658 if (dvar == NULL) 5659 break; 5660 5661 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5662 if (!dtrace_vcanload( 5663 (void *)(uintptr_t)regs[rd], &v->dtdv_type, 5664 mstate, vstate)) 5665 break; 5666 5667 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5668 dvar->dtdv_data, &v->dtdv_type); 5669 } else { 5670 *((uint64_t *)dvar->dtdv_data) = regs[rd]; 5671 } 5672 5673 break; 5674 } 5675 5676 case DIF_OP_ALLOCS: { 5677 uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); 5678 size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1]; 5679 5680 /* 5681 * Rounding up the user allocation size could have 5682 * overflowed large, bogus allocations (like -1ULL) to 5683 * 0. 5684 */ 5685 if (size < regs[r1] || 5686 !DTRACE_INSCRATCH(mstate, size)) { 5687 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 5688 regs[rd] = 0; 5689 break; 5690 } 5691 5692 dtrace_bzero((void *) mstate->dtms_scratch_ptr, size); 5693 mstate->dtms_scratch_ptr += size; 5694 regs[rd] = ptr; 5695 break; 5696 } 5697 5698 case DIF_OP_COPYS: 5699 if (!dtrace_canstore(regs[rd], regs[r2], 5700 mstate, vstate)) { 5701 *flags |= CPU_DTRACE_BADADDR; 5702 *illval = regs[rd]; 5703 break; 5704 } 5705 5706 if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate)) 5707 break; 5708 5709 dtrace_bcopy((void *)(uintptr_t)regs[r1], 5710 (void *)(uintptr_t)regs[rd], (size_t)regs[r2]); 5711 break; 5712 5713 case DIF_OP_STB: 5714 if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) { 5715 *flags |= CPU_DTRACE_BADADDR; 5716 *illval = regs[rd]; 5717 break; 5718 } 5719 *((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1]; 5720 break; 5721 5722 case DIF_OP_STH: 5723 if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) { 5724 *flags |= CPU_DTRACE_BADADDR; 5725 *illval = regs[rd]; 5726 break; 5727 } 5728 if (regs[rd] & 1) { 5729 *flags |= CPU_DTRACE_BADALIGN; 5730 *illval = regs[rd]; 5731 break; 5732 } 5733 *((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1]; 5734 break; 5735 5736 case DIF_OP_STW: 5737 if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) { 5738 *flags |= CPU_DTRACE_BADADDR; 5739 *illval = regs[rd]; 5740 break; 5741 } 5742 if (regs[rd] & 3) { 5743 *flags |= CPU_DTRACE_BADALIGN; 5744 *illval = regs[rd]; 5745 break; 5746 } 5747 *((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1]; 5748 break; 5749 5750 case DIF_OP_STX: 5751 if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) { 5752 *flags |= CPU_DTRACE_BADADDR; 5753 *illval = regs[rd]; 5754 break; 5755 } 5756 if (regs[rd] & 7) { 5757 *flags |= CPU_DTRACE_BADALIGN; 5758 *illval = regs[rd]; 5759 break; 5760 } 5761 *((uint64_t *)(uintptr_t)regs[rd]) = regs[r1]; 5762 break; 5763 } 5764 } 5765 5766 if (!(*flags & CPU_DTRACE_FAULT)) 5767 return (rval); 5768 5769 mstate->dtms_fltoffs = opc * sizeof (dif_instr_t); 5770 mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS; 5771 5772 return (0); 5773 } 5774 5775 static void 5776 dtrace_action_breakpoint(dtrace_ecb_t *ecb) 5777 { 5778 dtrace_probe_t *probe = ecb->dte_probe; 5779 dtrace_provider_t *prov = probe->dtpr_provider; 5780 char c[DTRACE_FULLNAMELEN + 80], *str; 5781 char *msg = "dtrace: breakpoint action at probe "; 5782 char *ecbmsg = " (ecb "; 5783 uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4)); 5784 uintptr_t val = (uintptr_t)ecb; 5785 int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0; 5786 5787 if (dtrace_destructive_disallow) 5788 return; 5789 5790 /* 5791 * It's impossible to be taking action on the NULL probe. 5792 */ 5793 ASSERT(probe != NULL); 5794 5795 /* 5796 * This is a poor man's (destitute man's?) sprintf(): we want to 5797 * print the provider name, module name, function name and name of 5798 * the probe, along with the hex address of the ECB with the breakpoint 5799 * action -- all of which we must place in the character buffer by 5800 * hand. 5801 */ 5802 while (*msg != '\0') 5803 c[i++] = *msg++; 5804 5805 for (str = prov->dtpv_name; *str != '\0'; str++) 5806 c[i++] = *str; 5807 c[i++] = ':'; 5808 5809 for (str = probe->dtpr_mod; *str != '\0'; str++) 5810 c[i++] = *str; 5811 c[i++] = ':'; 5812 5813 for (str = probe->dtpr_func; *str != '\0'; str++) 5814 c[i++] = *str; 5815 c[i++] = ':'; 5816 5817 for (str = probe->dtpr_name; *str != '\0'; str++) 5818 c[i++] = *str; 5819 5820 while (*ecbmsg != '\0') 5821 c[i++] = *ecbmsg++; 5822 5823 while (shift >= 0) { 5824 mask = (uintptr_t)0xf << shift; 5825 5826 if (val >= ((uintptr_t)1 << shift)) 5827 c[i++] = "0123456789abcdef"[(val & mask) >> shift]; 5828 shift -= 4; 5829 } 5830 5831 c[i++] = ')'; 5832 c[i] = '\0'; 5833 5834 #if defined(sun) 5835 debug_enter(c); 5836 #else 5837 kdb_enter(KDB_WHY_DTRACE, "breakpoint action"); 5838 #endif 5839 } 5840 5841 static void 5842 dtrace_action_panic(dtrace_ecb_t *ecb) 5843 { 5844 dtrace_probe_t *probe = ecb->dte_probe; 5845 5846 /* 5847 * It's impossible to be taking action on the NULL probe. 5848 */ 5849 ASSERT(probe != NULL); 5850 5851 if (dtrace_destructive_disallow) 5852 return; 5853 5854 if (dtrace_panicked != NULL) 5855 return; 5856 5857 if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL) 5858 return; 5859 5860 /* 5861 * We won the right to panic. (We want to be sure that only one 5862 * thread calls panic() from dtrace_probe(), and that panic() is 5863 * called exactly once.) 5864 */ 5865 dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)", 5866 probe->dtpr_provider->dtpv_name, probe->dtpr_mod, 5867 probe->dtpr_func, probe->dtpr_name, (void *)ecb); 5868 } 5869 5870 static void 5871 dtrace_action_raise(uint64_t sig) 5872 { 5873 if (dtrace_destructive_disallow) 5874 return; 5875 5876 if (sig >= NSIG) { 5877 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 5878 return; 5879 } 5880 5881 #if defined(sun) 5882 /* 5883 * raise() has a queue depth of 1 -- we ignore all subsequent 5884 * invocations of the raise() action. 5885 */ 5886 if (curthread->t_dtrace_sig == 0) 5887 curthread->t_dtrace_sig = (uint8_t)sig; 5888 5889 curthread->t_sig_check = 1; 5890 aston(curthread); 5891 #else 5892 struct proc *p = curproc; 5893 PROC_LOCK(p); 5894 kern_psignal(p, sig); 5895 PROC_UNLOCK(p); 5896 #endif 5897 } 5898 5899 static void 5900 dtrace_action_stop(void) 5901 { 5902 if (dtrace_destructive_disallow) 5903 return; 5904 5905 #if defined(sun) 5906 if (!curthread->t_dtrace_stop) { 5907 curthread->t_dtrace_stop = 1; 5908 curthread->t_sig_check = 1; 5909 aston(curthread); 5910 } 5911 #else 5912 struct proc *p = curproc; 5913 PROC_LOCK(p); 5914 kern_psignal(p, SIGSTOP); 5915 PROC_UNLOCK(p); 5916 #endif 5917 } 5918 5919 static void 5920 dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val) 5921 { 5922 hrtime_t now; 5923 volatile uint16_t *flags; 5924 #if defined(sun) 5925 cpu_t *cpu = CPU; 5926 #else 5927 cpu_t *cpu = &solaris_cpu[curcpu]; 5928 #endif 5929 5930 if (dtrace_destructive_disallow) 5931 return; 5932 5933 flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags; 5934 5935 now = dtrace_gethrtime(); 5936 5937 if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) { 5938 /* 5939 * We need to advance the mark to the current time. 5940 */ 5941 cpu->cpu_dtrace_chillmark = now; 5942 cpu->cpu_dtrace_chilled = 0; 5943 } 5944 5945 /* 5946 * Now check to see if the requested chill time would take us over 5947 * the maximum amount of time allowed in the chill interval. (Or 5948 * worse, if the calculation itself induces overflow.) 5949 */ 5950 if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max || 5951 cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) { 5952 *flags |= CPU_DTRACE_ILLOP; 5953 return; 5954 } 5955 5956 while (dtrace_gethrtime() - now < val) 5957 continue; 5958 5959 /* 5960 * Normally, we assure that the value of the variable "timestamp" does 5961 * not change within an ECB. The presence of chill() represents an 5962 * exception to this rule, however. 5963 */ 5964 mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP; 5965 cpu->cpu_dtrace_chilled += val; 5966 } 5967 5968 static void 5969 dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state, 5970 uint64_t *buf, uint64_t arg) 5971 { 5972 int nframes = DTRACE_USTACK_NFRAMES(arg); 5973 int strsize = DTRACE_USTACK_STRSIZE(arg); 5974 uint64_t *pcs = &buf[1], *fps; 5975 char *str = (char *)&pcs[nframes]; 5976 int size, offs = 0, i, j; 5977 uintptr_t old = mstate->dtms_scratch_ptr, saved; 5978 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 5979 char *sym; 5980 5981 /* 5982 * Should be taking a faster path if string space has not been 5983 * allocated. 5984 */ 5985 ASSERT(strsize != 0); 5986 5987 /* 5988 * We will first allocate some temporary space for the frame pointers. 5989 */ 5990 fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8); 5991 size = (uintptr_t)fps - mstate->dtms_scratch_ptr + 5992 (nframes * sizeof (uint64_t)); 5993 5994 if (!DTRACE_INSCRATCH(mstate, size)) { 5995 /* 5996 * Not enough room for our frame pointers -- need to indicate 5997 * that we ran out of scratch space. 5998 */ 5999 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 6000 return; 6001 } 6002 6003 mstate->dtms_scratch_ptr += size; 6004 saved = mstate->dtms_scratch_ptr; 6005 6006 /* 6007 * Now get a stack with both program counters and frame pointers. 6008 */ 6009 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 6010 dtrace_getufpstack(buf, fps, nframes + 1); 6011 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 6012 6013 /* 6014 * If that faulted, we're cooked. 6015 */ 6016 if (*flags & CPU_DTRACE_FAULT) 6017 goto out; 6018 6019 /* 6020 * Now we want to walk up the stack, calling the USTACK helper. For 6021 * each iteration, we restore the scratch pointer. 6022 */ 6023 for (i = 0; i < nframes; i++) { 6024 mstate->dtms_scratch_ptr = saved; 6025 6026 if (offs >= strsize) 6027 break; 6028 6029 sym = (char *)(uintptr_t)dtrace_helper( 6030 DTRACE_HELPER_ACTION_USTACK, 6031 mstate, state, pcs[i], fps[i]); 6032 6033 /* 6034 * If we faulted while running the helper, we're going to 6035 * clear the fault and null out the corresponding string. 6036 */ 6037 if (*flags & CPU_DTRACE_FAULT) { 6038 *flags &= ~CPU_DTRACE_FAULT; 6039 str[offs++] = '\0'; 6040 continue; 6041 } 6042 6043 if (sym == NULL) { 6044 str[offs++] = '\0'; 6045 continue; 6046 } 6047 6048 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 6049 6050 /* 6051 * Now copy in the string that the helper returned to us. 6052 */ 6053 for (j = 0; offs + j < strsize; j++) { 6054 if ((str[offs + j] = sym[j]) == '\0') 6055 break; 6056 } 6057 6058 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 6059 6060 offs += j + 1; 6061 } 6062 6063 if (offs >= strsize) { 6064 /* 6065 * If we didn't have room for all of the strings, we don't 6066 * abort processing -- this needn't be a fatal error -- but we 6067 * still want to increment a counter (dts_stkstroverflows) to 6068 * allow this condition to be warned about. (If this is from 6069 * a jstack() action, it is easily tuned via jstackstrsize.) 6070 */ 6071 dtrace_error(&state->dts_stkstroverflows); 6072 } 6073 6074 while (offs < strsize) 6075 str[offs++] = '\0'; 6076 6077 out: 6078 mstate->dtms_scratch_ptr = old; 6079 } 6080 6081 /* 6082 * If you're looking for the epicenter of DTrace, you just found it. This 6083 * is the function called by the provider to fire a probe -- from which all 6084 * subsequent probe-context DTrace activity emanates. 6085 */ 6086 void 6087 dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1, 6088 uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) 6089 { 6090 processorid_t cpuid; 6091 dtrace_icookie_t cookie; 6092 dtrace_probe_t *probe; 6093 dtrace_mstate_t mstate; 6094 dtrace_ecb_t *ecb; 6095 dtrace_action_t *act; 6096 intptr_t offs; 6097 size_t size; 6098 int vtime, onintr; 6099 volatile uint16_t *flags; 6100 hrtime_t now; 6101 6102 if (panicstr != NULL) 6103 return; 6104 6105 #if defined(sun) 6106 /* 6107 * Kick out immediately if this CPU is still being born (in which case 6108 * curthread will be set to -1) or the current thread can't allow 6109 * probes in its current context. 6110 */ 6111 if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE)) 6112 return; 6113 #endif 6114 6115 cookie = dtrace_interrupt_disable(); 6116 probe = dtrace_probes[id - 1]; 6117 cpuid = curcpu; 6118 onintr = CPU_ON_INTR(CPU); 6119 6120 if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE && 6121 probe->dtpr_predcache == curthread->t_predcache) { 6122 /* 6123 * We have hit in the predicate cache; we know that 6124 * this predicate would evaluate to be false. 6125 */ 6126 dtrace_interrupt_enable(cookie); 6127 return; 6128 } 6129 6130 #if defined(sun) 6131 if (panic_quiesce) { 6132 #else 6133 if (panicstr != NULL) { 6134 #endif 6135 /* 6136 * We don't trace anything if we're panicking. 6137 */ 6138 dtrace_interrupt_enable(cookie); 6139 return; 6140 } 6141 6142 now = dtrace_gethrtime(); 6143 vtime = dtrace_vtime_references != 0; 6144 6145 if (vtime && curthread->t_dtrace_start) 6146 curthread->t_dtrace_vtime += now - curthread->t_dtrace_start; 6147 6148 mstate.dtms_difo = NULL; 6149 mstate.dtms_probe = probe; 6150 mstate.dtms_strtok = 0; 6151 mstate.dtms_arg[0] = arg0; 6152 mstate.dtms_arg[1] = arg1; 6153 mstate.dtms_arg[2] = arg2; 6154 mstate.dtms_arg[3] = arg3; 6155 mstate.dtms_arg[4] = arg4; 6156 6157 flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags; 6158 6159 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { 6160 dtrace_predicate_t *pred = ecb->dte_predicate; 6161 dtrace_state_t *state = ecb->dte_state; 6162 dtrace_buffer_t *buf = &state->dts_buffer[cpuid]; 6163 dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid]; 6164 dtrace_vstate_t *vstate = &state->dts_vstate; 6165 dtrace_provider_t *prov = probe->dtpr_provider; 6166 uint64_t tracememsize = 0; 6167 int committed = 0; 6168 caddr_t tomax; 6169 6170 /* 6171 * A little subtlety with the following (seemingly innocuous) 6172 * declaration of the automatic 'val': by looking at the 6173 * code, you might think that it could be declared in the 6174 * action processing loop, below. (That is, it's only used in 6175 * the action processing loop.) However, it must be declared 6176 * out of that scope because in the case of DIF expression 6177 * arguments to aggregating actions, one iteration of the 6178 * action loop will use the last iteration's value. 6179 */ 6180 uint64_t val = 0; 6181 6182 mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE; 6183 *flags &= ~CPU_DTRACE_ERROR; 6184 6185 if (prov == dtrace_provider) { 6186 /* 6187 * If dtrace itself is the provider of this probe, 6188 * we're only going to continue processing the ECB if 6189 * arg0 (the dtrace_state_t) is equal to the ECB's 6190 * creating state. (This prevents disjoint consumers 6191 * from seeing one another's metaprobes.) 6192 */ 6193 if (arg0 != (uint64_t)(uintptr_t)state) 6194 continue; 6195 } 6196 6197 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) { 6198 /* 6199 * We're not currently active. If our provider isn't 6200 * the dtrace pseudo provider, we're not interested. 6201 */ 6202 if (prov != dtrace_provider) 6203 continue; 6204 6205 /* 6206 * Now we must further check if we are in the BEGIN 6207 * probe. If we are, we will only continue processing 6208 * if we're still in WARMUP -- if one BEGIN enabling 6209 * has invoked the exit() action, we don't want to 6210 * evaluate subsequent BEGIN enablings. 6211 */ 6212 if (probe->dtpr_id == dtrace_probeid_begin && 6213 state->dts_activity != DTRACE_ACTIVITY_WARMUP) { 6214 ASSERT(state->dts_activity == 6215 DTRACE_ACTIVITY_DRAINING); 6216 continue; 6217 } 6218 } 6219 6220 if (ecb->dte_cond) { 6221 /* 6222 * If the dte_cond bits indicate that this 6223 * consumer is only allowed to see user-mode firings 6224 * of this probe, call the provider's dtps_usermode() 6225 * entry point to check that the probe was fired 6226 * while in a user context. Skip this ECB if that's 6227 * not the case. 6228 */ 6229 if ((ecb->dte_cond & DTRACE_COND_USERMODE) && 6230 prov->dtpv_pops.dtps_usermode(prov->dtpv_arg, 6231 probe->dtpr_id, probe->dtpr_arg) == 0) 6232 continue; 6233 6234 #if defined(sun) 6235 /* 6236 * This is more subtle than it looks. We have to be 6237 * absolutely certain that CRED() isn't going to 6238 * change out from under us so it's only legit to 6239 * examine that structure if we're in constrained 6240 * situations. Currently, the only times we'll this 6241 * check is if a non-super-user has enabled the 6242 * profile or syscall providers -- providers that 6243 * allow visibility of all processes. For the 6244 * profile case, the check above will ensure that 6245 * we're examining a user context. 6246 */ 6247 if (ecb->dte_cond & DTRACE_COND_OWNER) { 6248 cred_t *cr; 6249 cred_t *s_cr = 6250 ecb->dte_state->dts_cred.dcr_cred; 6251 proc_t *proc; 6252 6253 ASSERT(s_cr != NULL); 6254 6255 if ((cr = CRED()) == NULL || 6256 s_cr->cr_uid != cr->cr_uid || 6257 s_cr->cr_uid != cr->cr_ruid || 6258 s_cr->cr_uid != cr->cr_suid || 6259 s_cr->cr_gid != cr->cr_gid || 6260 s_cr->cr_gid != cr->cr_rgid || 6261 s_cr->cr_gid != cr->cr_sgid || 6262 (proc = ttoproc(curthread)) == NULL || 6263 (proc->p_flag & SNOCD)) 6264 continue; 6265 } 6266 6267 if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) { 6268 cred_t *cr; 6269 cred_t *s_cr = 6270 ecb->dte_state->dts_cred.dcr_cred; 6271 6272 ASSERT(s_cr != NULL); 6273 6274 if ((cr = CRED()) == NULL || 6275 s_cr->cr_zone->zone_id != 6276 cr->cr_zone->zone_id) 6277 continue; 6278 } 6279 #endif 6280 } 6281 6282 if (now - state->dts_alive > dtrace_deadman_timeout) { 6283 /* 6284 * We seem to be dead. Unless we (a) have kernel 6285 * destructive permissions (b) have explicitly enabled 6286 * destructive actions and (c) destructive actions have 6287 * not been disabled, we're going to transition into 6288 * the KILLED state, from which no further processing 6289 * on this state will be performed. 6290 */ 6291 if (!dtrace_priv_kernel_destructive(state) || 6292 !state->dts_cred.dcr_destructive || 6293 dtrace_destructive_disallow) { 6294 void *activity = &state->dts_activity; 6295 dtrace_activity_t current; 6296 6297 do { 6298 current = state->dts_activity; 6299 } while (dtrace_cas32(activity, current, 6300 DTRACE_ACTIVITY_KILLED) != current); 6301 6302 continue; 6303 } 6304 } 6305 6306 if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed, 6307 ecb->dte_alignment, state, &mstate)) < 0) 6308 continue; 6309 6310 tomax = buf->dtb_tomax; 6311 ASSERT(tomax != NULL); 6312 6313 if (ecb->dte_size != 0) { 6314 dtrace_rechdr_t dtrh; 6315 if (!(mstate.dtms_present & DTRACE_MSTATE_TIMESTAMP)) { 6316 mstate.dtms_timestamp = dtrace_gethrtime(); 6317 mstate.dtms_present |= DTRACE_MSTATE_TIMESTAMP; 6318 } 6319 ASSERT3U(ecb->dte_size, >=, sizeof (dtrace_rechdr_t)); 6320 dtrh.dtrh_epid = ecb->dte_epid; 6321 DTRACE_RECORD_STORE_TIMESTAMP(&dtrh, 6322 mstate.dtms_timestamp); 6323 *((dtrace_rechdr_t *)(tomax + offs)) = dtrh; 6324 } 6325 6326 mstate.dtms_epid = ecb->dte_epid; 6327 mstate.dtms_present |= DTRACE_MSTATE_EPID; 6328 6329 if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) 6330 mstate.dtms_access = DTRACE_ACCESS_KERNEL; 6331 else 6332 mstate.dtms_access = 0; 6333 6334 if (pred != NULL) { 6335 dtrace_difo_t *dp = pred->dtp_difo; 6336 int rval; 6337 6338 rval = dtrace_dif_emulate(dp, &mstate, vstate, state); 6339 6340 if (!(*flags & CPU_DTRACE_ERROR) && !rval) { 6341 dtrace_cacheid_t cid = probe->dtpr_predcache; 6342 6343 if (cid != DTRACE_CACHEIDNONE && !onintr) { 6344 /* 6345 * Update the predicate cache... 6346 */ 6347 ASSERT(cid == pred->dtp_cacheid); 6348 curthread->t_predcache = cid; 6349 } 6350 6351 continue; 6352 } 6353 } 6354 6355 for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) && 6356 act != NULL; act = act->dta_next) { 6357 size_t valoffs; 6358 dtrace_difo_t *dp; 6359 dtrace_recdesc_t *rec = &act->dta_rec; 6360 6361 size = rec->dtrd_size; 6362 valoffs = offs + rec->dtrd_offset; 6363 6364 if (DTRACEACT_ISAGG(act->dta_kind)) { 6365 uint64_t v = 0xbad; 6366 dtrace_aggregation_t *agg; 6367 6368 agg = (dtrace_aggregation_t *)act; 6369 6370 if ((dp = act->dta_difo) != NULL) 6371 v = dtrace_dif_emulate(dp, 6372 &mstate, vstate, state); 6373 6374 if (*flags & CPU_DTRACE_ERROR) 6375 continue; 6376 6377 /* 6378 * Note that we always pass the expression 6379 * value from the previous iteration of the 6380 * action loop. This value will only be used 6381 * if there is an expression argument to the 6382 * aggregating action, denoted by the 6383 * dtag_hasarg field. 6384 */ 6385 dtrace_aggregate(agg, buf, 6386 offs, aggbuf, v, val); 6387 continue; 6388 } 6389 6390 switch (act->dta_kind) { 6391 case DTRACEACT_STOP: 6392 if (dtrace_priv_proc_destructive(state)) 6393 dtrace_action_stop(); 6394 continue; 6395 6396 case DTRACEACT_BREAKPOINT: 6397 if (dtrace_priv_kernel_destructive(state)) 6398 dtrace_action_breakpoint(ecb); 6399 continue; 6400 6401 case DTRACEACT_PANIC: 6402 if (dtrace_priv_kernel_destructive(state)) 6403 dtrace_action_panic(ecb); 6404 continue; 6405 6406 case DTRACEACT_STACK: 6407 if (!dtrace_priv_kernel(state)) 6408 continue; 6409 6410 dtrace_getpcstack((pc_t *)(tomax + valoffs), 6411 size / sizeof (pc_t), probe->dtpr_aframes, 6412 DTRACE_ANCHORED(probe) ? NULL : 6413 (uint32_t *)arg0); 6414 continue; 6415 6416 case DTRACEACT_JSTACK: 6417 case DTRACEACT_USTACK: 6418 if (!dtrace_priv_proc(state)) 6419 continue; 6420 6421 /* 6422 * See comment in DIF_VAR_PID. 6423 */ 6424 if (DTRACE_ANCHORED(mstate.dtms_probe) && 6425 CPU_ON_INTR(CPU)) { 6426 int depth = DTRACE_USTACK_NFRAMES( 6427 rec->dtrd_arg) + 1; 6428 6429 dtrace_bzero((void *)(tomax + valoffs), 6430 DTRACE_USTACK_STRSIZE(rec->dtrd_arg) 6431 + depth * sizeof (uint64_t)); 6432 6433 continue; 6434 } 6435 6436 if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 && 6437 curproc->p_dtrace_helpers != NULL) { 6438 /* 6439 * This is the slow path -- we have 6440 * allocated string space, and we're 6441 * getting the stack of a process that 6442 * has helpers. Call into a separate 6443 * routine to perform this processing. 6444 */ 6445 dtrace_action_ustack(&mstate, state, 6446 (uint64_t *)(tomax + valoffs), 6447 rec->dtrd_arg); 6448 continue; 6449 } 6450 6451 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 6452 dtrace_getupcstack((uint64_t *) 6453 (tomax + valoffs), 6454 DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1); 6455 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 6456 continue; 6457 6458 default: 6459 break; 6460 } 6461 6462 dp = act->dta_difo; 6463 ASSERT(dp != NULL); 6464 6465 val = dtrace_dif_emulate(dp, &mstate, vstate, state); 6466 6467 if (*flags & CPU_DTRACE_ERROR) 6468 continue; 6469 6470 switch (act->dta_kind) { 6471 case DTRACEACT_SPECULATE: { 6472 dtrace_rechdr_t *dtrh; 6473 6474 ASSERT(buf == &state->dts_buffer[cpuid]); 6475 buf = dtrace_speculation_buffer(state, 6476 cpuid, val); 6477 6478 if (buf == NULL) { 6479 *flags |= CPU_DTRACE_DROP; 6480 continue; 6481 } 6482 6483 offs = dtrace_buffer_reserve(buf, 6484 ecb->dte_needed, ecb->dte_alignment, 6485 state, NULL); 6486 6487 if (offs < 0) { 6488 *flags |= CPU_DTRACE_DROP; 6489 continue; 6490 } 6491 6492 tomax = buf->dtb_tomax; 6493 ASSERT(tomax != NULL); 6494 6495 if (ecb->dte_size == 0) 6496 continue; 6497 6498 ASSERT3U(ecb->dte_size, >=, 6499 sizeof (dtrace_rechdr_t)); 6500 dtrh = ((void *)(tomax + offs)); 6501 dtrh->dtrh_epid = ecb->dte_epid; 6502 /* 6503 * When the speculation is committed, all of 6504 * the records in the speculative buffer will 6505 * have their timestamps set to the commit 6506 * time. Until then, it is set to a sentinel 6507 * value, for debugability. 6508 */ 6509 DTRACE_RECORD_STORE_TIMESTAMP(dtrh, UINT64_MAX); 6510 continue; 6511 } 6512 6513 case DTRACEACT_PRINTM: { 6514 /* The DIF returns a 'memref'. */ 6515 uintptr_t *memref = (uintptr_t *)(uintptr_t) val; 6516 6517 /* Get the size from the memref. */ 6518 size = memref[1]; 6519 6520 /* 6521 * Check if the size exceeds the allocated 6522 * buffer size. 6523 */ 6524 if (size + sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) { 6525 /* Flag a drop! */ 6526 *flags |= CPU_DTRACE_DROP; 6527 continue; 6528 } 6529 6530 /* Store the size in the buffer first. */ 6531 DTRACE_STORE(uintptr_t, tomax, 6532 valoffs, size); 6533 6534 /* 6535 * Offset the buffer address to the start 6536 * of the data. 6537 */ 6538 valoffs += sizeof(uintptr_t); 6539 6540 /* 6541 * Reset to the memory address rather than 6542 * the memref array, then let the BYREF 6543 * code below do the work to store the 6544 * memory data in the buffer. 6545 */ 6546 val = memref[0]; 6547 break; 6548 } 6549 6550 case DTRACEACT_PRINTT: { 6551 /* The DIF returns a 'typeref'. */ 6552 uintptr_t *typeref = (uintptr_t *)(uintptr_t) val; 6553 char c = '\0' + 1; 6554 size_t s; 6555 6556 /* 6557 * Get the type string length and round it 6558 * up so that the data that follows is 6559 * aligned for easy access. 6560 */ 6561 size_t typs = strlen((char *) typeref[2]) + 1; 6562 typs = roundup(typs, sizeof(uintptr_t)); 6563 6564 /* 6565 *Get the size from the typeref using the 6566 * number of elements and the type size. 6567 */ 6568 size = typeref[1] * typeref[3]; 6569 6570 /* 6571 * Check if the size exceeds the allocated 6572 * buffer size. 6573 */ 6574 if (size + typs + 2 * sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) { 6575 /* Flag a drop! */ 6576 *flags |= CPU_DTRACE_DROP; 6577 6578 } 6579 6580 /* Store the size in the buffer first. */ 6581 DTRACE_STORE(uintptr_t, tomax, 6582 valoffs, size); 6583 valoffs += sizeof(uintptr_t); 6584 6585 /* Store the type size in the buffer. */ 6586 DTRACE_STORE(uintptr_t, tomax, 6587 valoffs, typeref[3]); 6588 valoffs += sizeof(uintptr_t); 6589 6590 val = typeref[2]; 6591 6592 for (s = 0; s < typs; s++) { 6593 if (c != '\0') 6594 c = dtrace_load8(val++); 6595 6596 DTRACE_STORE(uint8_t, tomax, 6597 valoffs++, c); 6598 } 6599 6600 /* 6601 * Reset to the memory address rather than 6602 * the typeref array, then let the BYREF 6603 * code below do the work to store the 6604 * memory data in the buffer. 6605 */ 6606 val = typeref[0]; 6607 break; 6608 } 6609 6610 case DTRACEACT_CHILL: 6611 if (dtrace_priv_kernel_destructive(state)) 6612 dtrace_action_chill(&mstate, val); 6613 continue; 6614 6615 case DTRACEACT_RAISE: 6616 if (dtrace_priv_proc_destructive(state)) 6617 dtrace_action_raise(val); 6618 continue; 6619 6620 case DTRACEACT_COMMIT: 6621 ASSERT(!committed); 6622 6623 /* 6624 * We need to commit our buffer state. 6625 */ 6626 if (ecb->dte_size) 6627 buf->dtb_offset = offs + ecb->dte_size; 6628 buf = &state->dts_buffer[cpuid]; 6629 dtrace_speculation_commit(state, cpuid, val); 6630 committed = 1; 6631 continue; 6632 6633 case DTRACEACT_DISCARD: 6634 dtrace_speculation_discard(state, cpuid, val); 6635 continue; 6636 6637 case DTRACEACT_DIFEXPR: 6638 case DTRACEACT_LIBACT: 6639 case DTRACEACT_PRINTF: 6640 case DTRACEACT_PRINTA: 6641 case DTRACEACT_SYSTEM: 6642 case DTRACEACT_FREOPEN: 6643 case DTRACEACT_TRACEMEM: 6644 break; 6645 6646 case DTRACEACT_TRACEMEM_DYNSIZE: 6647 tracememsize = val; 6648 break; 6649 6650 case DTRACEACT_SYM: 6651 case DTRACEACT_MOD: 6652 if (!dtrace_priv_kernel(state)) 6653 continue; 6654 break; 6655 6656 case DTRACEACT_USYM: 6657 case DTRACEACT_UMOD: 6658 case DTRACEACT_UADDR: { 6659 #if defined(sun) 6660 struct pid *pid = curthread->t_procp->p_pidp; 6661 #endif 6662 6663 if (!dtrace_priv_proc(state)) 6664 continue; 6665 6666 DTRACE_STORE(uint64_t, tomax, 6667 #if defined(sun) 6668 valoffs, (uint64_t)pid->pid_id); 6669 #else 6670 valoffs, (uint64_t) curproc->p_pid); 6671 #endif 6672 DTRACE_STORE(uint64_t, tomax, 6673 valoffs + sizeof (uint64_t), val); 6674 6675 continue; 6676 } 6677 6678 case DTRACEACT_EXIT: { 6679 /* 6680 * For the exit action, we are going to attempt 6681 * to atomically set our activity to be 6682 * draining. If this fails (either because 6683 * another CPU has beat us to the exit action, 6684 * or because our current activity is something 6685 * other than ACTIVE or WARMUP), we will 6686 * continue. This assures that the exit action 6687 * can be successfully recorded at most once 6688 * when we're in the ACTIVE state. If we're 6689 * encountering the exit() action while in 6690 * COOLDOWN, however, we want to honor the new 6691 * status code. (We know that we're the only 6692 * thread in COOLDOWN, so there is no race.) 6693 */ 6694 void *activity = &state->dts_activity; 6695 dtrace_activity_t current = state->dts_activity; 6696 6697 if (current == DTRACE_ACTIVITY_COOLDOWN) 6698 break; 6699 6700 if (current != DTRACE_ACTIVITY_WARMUP) 6701 current = DTRACE_ACTIVITY_ACTIVE; 6702 6703 if (dtrace_cas32(activity, current, 6704 DTRACE_ACTIVITY_DRAINING) != current) { 6705 *flags |= CPU_DTRACE_DROP; 6706 continue; 6707 } 6708 6709 break; 6710 } 6711 6712 default: 6713 ASSERT(0); 6714 } 6715 6716 if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) { 6717 uintptr_t end = valoffs + size; 6718 6719 if (tracememsize != 0 && 6720 valoffs + tracememsize < end) { 6721 end = valoffs + tracememsize; 6722 tracememsize = 0; 6723 } 6724 6725 if (!dtrace_vcanload((void *)(uintptr_t)val, 6726 &dp->dtdo_rtype, &mstate, vstate)) 6727 continue; 6728 6729 /* 6730 * If this is a string, we're going to only 6731 * load until we find the zero byte -- after 6732 * which we'll store zero bytes. 6733 */ 6734 if (dp->dtdo_rtype.dtdt_kind == 6735 DIF_TYPE_STRING) { 6736 char c = '\0' + 1; 6737 int intuple = act->dta_intuple; 6738 size_t s; 6739 6740 for (s = 0; s < size; s++) { 6741 if (c != '\0') 6742 c = dtrace_load8(val++); 6743 6744 DTRACE_STORE(uint8_t, tomax, 6745 valoffs++, c); 6746 6747 if (c == '\0' && intuple) 6748 break; 6749 } 6750 6751 continue; 6752 } 6753 6754 while (valoffs < end) { 6755 DTRACE_STORE(uint8_t, tomax, valoffs++, 6756 dtrace_load8(val++)); 6757 } 6758 6759 continue; 6760 } 6761 6762 switch (size) { 6763 case 0: 6764 break; 6765 6766 case sizeof (uint8_t): 6767 DTRACE_STORE(uint8_t, tomax, valoffs, val); 6768 break; 6769 case sizeof (uint16_t): 6770 DTRACE_STORE(uint16_t, tomax, valoffs, val); 6771 break; 6772 case sizeof (uint32_t): 6773 DTRACE_STORE(uint32_t, tomax, valoffs, val); 6774 break; 6775 case sizeof (uint64_t): 6776 DTRACE_STORE(uint64_t, tomax, valoffs, val); 6777 break; 6778 default: 6779 /* 6780 * Any other size should have been returned by 6781 * reference, not by value. 6782 */ 6783 ASSERT(0); 6784 break; 6785 } 6786 } 6787 6788 if (*flags & CPU_DTRACE_DROP) 6789 continue; 6790 6791 if (*flags & CPU_DTRACE_FAULT) { 6792 int ndx; 6793 dtrace_action_t *err; 6794 6795 buf->dtb_errors++; 6796 6797 if (probe->dtpr_id == dtrace_probeid_error) { 6798 /* 6799 * There's nothing we can do -- we had an 6800 * error on the error probe. We bump an 6801 * error counter to at least indicate that 6802 * this condition happened. 6803 */ 6804 dtrace_error(&state->dts_dblerrors); 6805 continue; 6806 } 6807 6808 if (vtime) { 6809 /* 6810 * Before recursing on dtrace_probe(), we 6811 * need to explicitly clear out our start 6812 * time to prevent it from being accumulated 6813 * into t_dtrace_vtime. 6814 */ 6815 curthread->t_dtrace_start = 0; 6816 } 6817 6818 /* 6819 * Iterate over the actions to figure out which action 6820 * we were processing when we experienced the error. 6821 * Note that act points _past_ the faulting action; if 6822 * act is ecb->dte_action, the fault was in the 6823 * predicate, if it's ecb->dte_action->dta_next it's 6824 * in action #1, and so on. 6825 */ 6826 for (err = ecb->dte_action, ndx = 0; 6827 err != act; err = err->dta_next, ndx++) 6828 continue; 6829 6830 dtrace_probe_error(state, ecb->dte_epid, ndx, 6831 (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ? 6832 mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags), 6833 cpu_core[cpuid].cpuc_dtrace_illval); 6834 6835 continue; 6836 } 6837 6838 if (!committed) 6839 buf->dtb_offset = offs + ecb->dte_size; 6840 } 6841 6842 if (vtime) 6843 curthread->t_dtrace_start = dtrace_gethrtime(); 6844 6845 dtrace_interrupt_enable(cookie); 6846 } 6847 6848 /* 6849 * DTrace Probe Hashing Functions 6850 * 6851 * The functions in this section (and indeed, the functions in remaining 6852 * sections) are not _called_ from probe context. (Any exceptions to this are 6853 * marked with a "Note:".) Rather, they are called from elsewhere in the 6854 * DTrace framework to look-up probes in, add probes to and remove probes from 6855 * the DTrace probe hashes. (Each probe is hashed by each element of the 6856 * probe tuple -- allowing for fast lookups, regardless of what was 6857 * specified.) 6858 */ 6859 static uint_t 6860 dtrace_hash_str(const char *p) 6861 { 6862 unsigned int g; 6863 uint_t hval = 0; 6864 6865 while (*p) { 6866 hval = (hval << 4) + *p++; 6867 if ((g = (hval & 0xf0000000)) != 0) 6868 hval ^= g >> 24; 6869 hval &= ~g; 6870 } 6871 return (hval); 6872 } 6873 6874 static dtrace_hash_t * 6875 dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs) 6876 { 6877 dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP); 6878 6879 hash->dth_stroffs = stroffs; 6880 hash->dth_nextoffs = nextoffs; 6881 hash->dth_prevoffs = prevoffs; 6882 6883 hash->dth_size = 1; 6884 hash->dth_mask = hash->dth_size - 1; 6885 6886 hash->dth_tab = kmem_zalloc(hash->dth_size * 6887 sizeof (dtrace_hashbucket_t *), KM_SLEEP); 6888 6889 return (hash); 6890 } 6891 6892 static void 6893 dtrace_hash_destroy(dtrace_hash_t *hash) 6894 { 6895 #ifdef DEBUG 6896 int i; 6897 6898 for (i = 0; i < hash->dth_size; i++) 6899 ASSERT(hash->dth_tab[i] == NULL); 6900 #endif 6901 6902 kmem_free(hash->dth_tab, 6903 hash->dth_size * sizeof (dtrace_hashbucket_t *)); 6904 kmem_free(hash, sizeof (dtrace_hash_t)); 6905 } 6906 6907 static void 6908 dtrace_hash_resize(dtrace_hash_t *hash) 6909 { 6910 int size = hash->dth_size, i, ndx; 6911 int new_size = hash->dth_size << 1; 6912 int new_mask = new_size - 1; 6913 dtrace_hashbucket_t **new_tab, *bucket, *next; 6914 6915 ASSERT((new_size & new_mask) == 0); 6916 6917 new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP); 6918 6919 for (i = 0; i < size; i++) { 6920 for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) { 6921 dtrace_probe_t *probe = bucket->dthb_chain; 6922 6923 ASSERT(probe != NULL); 6924 ndx = DTRACE_HASHSTR(hash, probe) & new_mask; 6925 6926 next = bucket->dthb_next; 6927 bucket->dthb_next = new_tab[ndx]; 6928 new_tab[ndx] = bucket; 6929 } 6930 } 6931 6932 kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *)); 6933 hash->dth_tab = new_tab; 6934 hash->dth_size = new_size; 6935 hash->dth_mask = new_mask; 6936 } 6937 6938 static void 6939 dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new) 6940 { 6941 int hashval = DTRACE_HASHSTR(hash, new); 6942 int ndx = hashval & hash->dth_mask; 6943 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 6944 dtrace_probe_t **nextp, **prevp; 6945 6946 for (; bucket != NULL; bucket = bucket->dthb_next) { 6947 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new)) 6948 goto add; 6949 } 6950 6951 if ((hash->dth_nbuckets >> 1) > hash->dth_size) { 6952 dtrace_hash_resize(hash); 6953 dtrace_hash_add(hash, new); 6954 return; 6955 } 6956 6957 bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP); 6958 bucket->dthb_next = hash->dth_tab[ndx]; 6959 hash->dth_tab[ndx] = bucket; 6960 hash->dth_nbuckets++; 6961 6962 add: 6963 nextp = DTRACE_HASHNEXT(hash, new); 6964 ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL); 6965 *nextp = bucket->dthb_chain; 6966 6967 if (bucket->dthb_chain != NULL) { 6968 prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain); 6969 ASSERT(*prevp == NULL); 6970 *prevp = new; 6971 } 6972 6973 bucket->dthb_chain = new; 6974 bucket->dthb_len++; 6975 } 6976 6977 static dtrace_probe_t * 6978 dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template) 6979 { 6980 int hashval = DTRACE_HASHSTR(hash, template); 6981 int ndx = hashval & hash->dth_mask; 6982 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 6983 6984 for (; bucket != NULL; bucket = bucket->dthb_next) { 6985 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) 6986 return (bucket->dthb_chain); 6987 } 6988 6989 return (NULL); 6990 } 6991 6992 static int 6993 dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template) 6994 { 6995 int hashval = DTRACE_HASHSTR(hash, template); 6996 int ndx = hashval & hash->dth_mask; 6997 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 6998 6999 for (; bucket != NULL; bucket = bucket->dthb_next) { 7000 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) 7001 return (bucket->dthb_len); 7002 } 7003 7004 return (0); 7005 } 7006 7007 static void 7008 dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe) 7009 { 7010 int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask; 7011 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 7012 7013 dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe); 7014 dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe); 7015 7016 /* 7017 * Find the bucket that we're removing this probe from. 7018 */ 7019 for (; bucket != NULL; bucket = bucket->dthb_next) { 7020 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe)) 7021 break; 7022 } 7023 7024 ASSERT(bucket != NULL); 7025 7026 if (*prevp == NULL) { 7027 if (*nextp == NULL) { 7028 /* 7029 * The removed probe was the only probe on this 7030 * bucket; we need to remove the bucket. 7031 */ 7032 dtrace_hashbucket_t *b = hash->dth_tab[ndx]; 7033 7034 ASSERT(bucket->dthb_chain == probe); 7035 ASSERT(b != NULL); 7036 7037 if (b == bucket) { 7038 hash->dth_tab[ndx] = bucket->dthb_next; 7039 } else { 7040 while (b->dthb_next != bucket) 7041 b = b->dthb_next; 7042 b->dthb_next = bucket->dthb_next; 7043 } 7044 7045 ASSERT(hash->dth_nbuckets > 0); 7046 hash->dth_nbuckets--; 7047 kmem_free(bucket, sizeof (dtrace_hashbucket_t)); 7048 return; 7049 } 7050 7051 bucket->dthb_chain = *nextp; 7052 } else { 7053 *(DTRACE_HASHNEXT(hash, *prevp)) = *nextp; 7054 } 7055 7056 if (*nextp != NULL) 7057 *(DTRACE_HASHPREV(hash, *nextp)) = *prevp; 7058 } 7059 7060 /* 7061 * DTrace Utility Functions 7062 * 7063 * These are random utility functions that are _not_ called from probe context. 7064 */ 7065 static int 7066 dtrace_badattr(const dtrace_attribute_t *a) 7067 { 7068 return (a->dtat_name > DTRACE_STABILITY_MAX || 7069 a->dtat_data > DTRACE_STABILITY_MAX || 7070 a->dtat_class > DTRACE_CLASS_MAX); 7071 } 7072 7073 /* 7074 * Return a duplicate copy of a string. If the specified string is NULL, 7075 * this function returns a zero-length string. 7076 */ 7077 static char * 7078 dtrace_strdup(const char *str) 7079 { 7080 char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP); 7081 7082 if (str != NULL) 7083 (void) strcpy(new, str); 7084 7085 return (new); 7086 } 7087 7088 #define DTRACE_ISALPHA(c) \ 7089 (((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z')) 7090 7091 static int 7092 dtrace_badname(const char *s) 7093 { 7094 char c; 7095 7096 if (s == NULL || (c = *s++) == '\0') 7097 return (0); 7098 7099 if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.') 7100 return (1); 7101 7102 while ((c = *s++) != '\0') { 7103 if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') && 7104 c != '-' && c != '_' && c != '.' && c != '`') 7105 return (1); 7106 } 7107 7108 return (0); 7109 } 7110 7111 static void 7112 dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp) 7113 { 7114 uint32_t priv; 7115 7116 #if defined(sun) 7117 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { 7118 /* 7119 * For DTRACE_PRIV_ALL, the uid and zoneid don't matter. 7120 */ 7121 priv = DTRACE_PRIV_ALL; 7122 } else { 7123 *uidp = crgetuid(cr); 7124 *zoneidp = crgetzoneid(cr); 7125 7126 priv = 0; 7127 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) 7128 priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER; 7129 else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) 7130 priv |= DTRACE_PRIV_USER; 7131 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) 7132 priv |= DTRACE_PRIV_PROC; 7133 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) 7134 priv |= DTRACE_PRIV_OWNER; 7135 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) 7136 priv |= DTRACE_PRIV_ZONEOWNER; 7137 } 7138 #else 7139 priv = DTRACE_PRIV_ALL; 7140 #endif 7141 7142 *privp = priv; 7143 } 7144 7145 #ifdef DTRACE_ERRDEBUG 7146 static void 7147 dtrace_errdebug(const char *str) 7148 { 7149 int hval = dtrace_hash_str(str) % DTRACE_ERRHASHSZ; 7150 int occupied = 0; 7151 7152 mutex_enter(&dtrace_errlock); 7153 dtrace_errlast = str; 7154 dtrace_errthread = curthread; 7155 7156 while (occupied++ < DTRACE_ERRHASHSZ) { 7157 if (dtrace_errhash[hval].dter_msg == str) { 7158 dtrace_errhash[hval].dter_count++; 7159 goto out; 7160 } 7161 7162 if (dtrace_errhash[hval].dter_msg != NULL) { 7163 hval = (hval + 1) % DTRACE_ERRHASHSZ; 7164 continue; 7165 } 7166 7167 dtrace_errhash[hval].dter_msg = str; 7168 dtrace_errhash[hval].dter_count = 1; 7169 goto out; 7170 } 7171 7172 panic("dtrace: undersized error hash"); 7173 out: 7174 mutex_exit(&dtrace_errlock); 7175 } 7176 #endif 7177 7178 /* 7179 * DTrace Matching Functions 7180 * 7181 * These functions are used to match groups of probes, given some elements of 7182 * a probe tuple, or some globbed expressions for elements of a probe tuple. 7183 */ 7184 static int 7185 dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid, 7186 zoneid_t zoneid) 7187 { 7188 if (priv != DTRACE_PRIV_ALL) { 7189 uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags; 7190 uint32_t match = priv & ppriv; 7191 7192 /* 7193 * No PRIV_DTRACE_* privileges... 7194 */ 7195 if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER | 7196 DTRACE_PRIV_KERNEL)) == 0) 7197 return (0); 7198 7199 /* 7200 * No matching bits, but there were bits to match... 7201 */ 7202 if (match == 0 && ppriv != 0) 7203 return (0); 7204 7205 /* 7206 * Need to have permissions to the process, but don't... 7207 */ 7208 if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 && 7209 uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) { 7210 return (0); 7211 } 7212 7213 /* 7214 * Need to be in the same zone unless we possess the 7215 * privilege to examine all zones. 7216 */ 7217 if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 && 7218 zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) { 7219 return (0); 7220 } 7221 } 7222 7223 return (1); 7224 } 7225 7226 /* 7227 * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which 7228 * consists of input pattern strings and an ops-vector to evaluate them. 7229 * This function returns >0 for match, 0 for no match, and <0 for error. 7230 */ 7231 static int 7232 dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp, 7233 uint32_t priv, uid_t uid, zoneid_t zoneid) 7234 { 7235 dtrace_provider_t *pvp = prp->dtpr_provider; 7236 int rv; 7237 7238 if (pvp->dtpv_defunct) 7239 return (0); 7240 7241 if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0) 7242 return (rv); 7243 7244 if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0) 7245 return (rv); 7246 7247 if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0) 7248 return (rv); 7249 7250 if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0) 7251 return (rv); 7252 7253 if (dtrace_match_priv(prp, priv, uid, zoneid) == 0) 7254 return (0); 7255 7256 return (rv); 7257 } 7258 7259 /* 7260 * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN) 7261 * interface for matching a glob pattern 'p' to an input string 's'. Unlike 7262 * libc's version, the kernel version only applies to 8-bit ASCII strings. 7263 * In addition, all of the recursion cases except for '*' matching have been 7264 * unwound. For '*', we still implement recursive evaluation, but a depth 7265 * counter is maintained and matching is aborted if we recurse too deep. 7266 * The function returns 0 if no match, >0 if match, and <0 if recursion error. 7267 */ 7268 static int 7269 dtrace_match_glob(const char *s, const char *p, int depth) 7270 { 7271 const char *olds; 7272 char s1, c; 7273 int gs; 7274 7275 if (depth > DTRACE_PROBEKEY_MAXDEPTH) 7276 return (-1); 7277 7278 if (s == NULL) 7279 s = ""; /* treat NULL as empty string */ 7280 7281 top: 7282 olds = s; 7283 s1 = *s++; 7284 7285 if (p == NULL) 7286 return (0); 7287 7288 if ((c = *p++) == '\0') 7289 return (s1 == '\0'); 7290 7291 switch (c) { 7292 case '[': { 7293 int ok = 0, notflag = 0; 7294 char lc = '\0'; 7295 7296 if (s1 == '\0') 7297 return (0); 7298 7299 if (*p == '!') { 7300 notflag = 1; 7301 p++; 7302 } 7303 7304 if ((c = *p++) == '\0') 7305 return (0); 7306 7307 do { 7308 if (c == '-' && lc != '\0' && *p != ']') { 7309 if ((c = *p++) == '\0') 7310 return (0); 7311 if (c == '\\' && (c = *p++) == '\0') 7312 return (0); 7313 7314 if (notflag) { 7315 if (s1 < lc || s1 > c) 7316 ok++; 7317 else 7318 return (0); 7319 } else if (lc <= s1 && s1 <= c) 7320 ok++; 7321 7322 } else if (c == '\\' && (c = *p++) == '\0') 7323 return (0); 7324 7325 lc = c; /* save left-hand 'c' for next iteration */ 7326 7327 if (notflag) { 7328 if (s1 != c) 7329 ok++; 7330 else 7331 return (0); 7332 } else if (s1 == c) 7333 ok++; 7334 7335 if ((c = *p++) == '\0') 7336 return (0); 7337 7338 } while (c != ']'); 7339 7340 if (ok) 7341 goto top; 7342 7343 return (0); 7344 } 7345 7346 case '\\': 7347 if ((c = *p++) == '\0') 7348 return (0); 7349 /*FALLTHRU*/ 7350 7351 default: 7352 if (c != s1) 7353 return (0); 7354 /*FALLTHRU*/ 7355 7356 case '?': 7357 if (s1 != '\0') 7358 goto top; 7359 return (0); 7360 7361 case '*': 7362 while (*p == '*') 7363 p++; /* consecutive *'s are identical to a single one */ 7364 7365 if (*p == '\0') 7366 return (1); 7367 7368 for (s = olds; *s != '\0'; s++) { 7369 if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0) 7370 return (gs); 7371 } 7372 7373 return (0); 7374 } 7375 } 7376 7377 /*ARGSUSED*/ 7378 static int 7379 dtrace_match_string(const char *s, const char *p, int depth) 7380 { 7381 return (s != NULL && strcmp(s, p) == 0); 7382 } 7383 7384 /*ARGSUSED*/ 7385 static int 7386 dtrace_match_nul(const char *s, const char *p, int depth) 7387 { 7388 return (1); /* always match the empty pattern */ 7389 } 7390 7391 /*ARGSUSED*/ 7392 static int 7393 dtrace_match_nonzero(const char *s, const char *p, int depth) 7394 { 7395 return (s != NULL && s[0] != '\0'); 7396 } 7397 7398 static int 7399 dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid, 7400 zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg) 7401 { 7402 dtrace_probe_t template, *probe; 7403 dtrace_hash_t *hash = NULL; 7404 int len, best = INT_MAX, nmatched = 0; 7405 dtrace_id_t i; 7406 7407 ASSERT(MUTEX_HELD(&dtrace_lock)); 7408 7409 /* 7410 * If the probe ID is specified in the key, just lookup by ID and 7411 * invoke the match callback once if a matching probe is found. 7412 */ 7413 if (pkp->dtpk_id != DTRACE_IDNONE) { 7414 if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL && 7415 dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) { 7416 (void) (*matched)(probe, arg); 7417 nmatched++; 7418 } 7419 return (nmatched); 7420 } 7421 7422 template.dtpr_mod = (char *)pkp->dtpk_mod; 7423 template.dtpr_func = (char *)pkp->dtpk_func; 7424 template.dtpr_name = (char *)pkp->dtpk_name; 7425 7426 /* 7427 * We want to find the most distinct of the module name, function 7428 * name, and name. So for each one that is not a glob pattern or 7429 * empty string, we perform a lookup in the corresponding hash and 7430 * use the hash table with the fewest collisions to do our search. 7431 */ 7432 if (pkp->dtpk_mmatch == &dtrace_match_string && 7433 (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) { 7434 best = len; 7435 hash = dtrace_bymod; 7436 } 7437 7438 if (pkp->dtpk_fmatch == &dtrace_match_string && 7439 (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) { 7440 best = len; 7441 hash = dtrace_byfunc; 7442 } 7443 7444 if (pkp->dtpk_nmatch == &dtrace_match_string && 7445 (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) { 7446 best = len; 7447 hash = dtrace_byname; 7448 } 7449 7450 /* 7451 * If we did not select a hash table, iterate over every probe and 7452 * invoke our callback for each one that matches our input probe key. 7453 */ 7454 if (hash == NULL) { 7455 for (i = 0; i < dtrace_nprobes; i++) { 7456 if ((probe = dtrace_probes[i]) == NULL || 7457 dtrace_match_probe(probe, pkp, priv, uid, 7458 zoneid) <= 0) 7459 continue; 7460 7461 nmatched++; 7462 7463 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT) 7464 break; 7465 } 7466 7467 return (nmatched); 7468 } 7469 7470 /* 7471 * If we selected a hash table, iterate over each probe of the same key 7472 * name and invoke the callback for every probe that matches the other 7473 * attributes of our input probe key. 7474 */ 7475 for (probe = dtrace_hash_lookup(hash, &template); probe != NULL; 7476 probe = *(DTRACE_HASHNEXT(hash, probe))) { 7477 7478 if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0) 7479 continue; 7480 7481 nmatched++; 7482 7483 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT) 7484 break; 7485 } 7486 7487 return (nmatched); 7488 } 7489 7490 /* 7491 * Return the function pointer dtrace_probecmp() should use to compare the 7492 * specified pattern with a string. For NULL or empty patterns, we select 7493 * dtrace_match_nul(). For glob pattern strings, we use dtrace_match_glob(). 7494 * For non-empty non-glob strings, we use dtrace_match_string(). 7495 */ 7496 static dtrace_probekey_f * 7497 dtrace_probekey_func(const char *p) 7498 { 7499 char c; 7500 7501 if (p == NULL || *p == '\0') 7502 return (&dtrace_match_nul); 7503 7504 while ((c = *p++) != '\0') { 7505 if (c == '[' || c == '?' || c == '*' || c == '\\') 7506 return (&dtrace_match_glob); 7507 } 7508 7509 return (&dtrace_match_string); 7510 } 7511 7512 /* 7513 * Build a probe comparison key for use with dtrace_match_probe() from the 7514 * given probe description. By convention, a null key only matches anchored 7515 * probes: if each field is the empty string, reset dtpk_fmatch to 7516 * dtrace_match_nonzero(). 7517 */ 7518 static void 7519 dtrace_probekey(dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp) 7520 { 7521 pkp->dtpk_prov = pdp->dtpd_provider; 7522 pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider); 7523 7524 pkp->dtpk_mod = pdp->dtpd_mod; 7525 pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod); 7526 7527 pkp->dtpk_func = pdp->dtpd_func; 7528 pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func); 7529 7530 pkp->dtpk_name = pdp->dtpd_name; 7531 pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name); 7532 7533 pkp->dtpk_id = pdp->dtpd_id; 7534 7535 if (pkp->dtpk_id == DTRACE_IDNONE && 7536 pkp->dtpk_pmatch == &dtrace_match_nul && 7537 pkp->dtpk_mmatch == &dtrace_match_nul && 7538 pkp->dtpk_fmatch == &dtrace_match_nul && 7539 pkp->dtpk_nmatch == &dtrace_match_nul) 7540 pkp->dtpk_fmatch = &dtrace_match_nonzero; 7541 } 7542 7543 /* 7544 * DTrace Provider-to-Framework API Functions 7545 * 7546 * These functions implement much of the Provider-to-Framework API, as 7547 * described in <sys/dtrace.h>. The parts of the API not in this section are 7548 * the functions in the API for probe management (found below), and 7549 * dtrace_probe() itself (found above). 7550 */ 7551 7552 /* 7553 * Register the calling provider with the DTrace framework. This should 7554 * generally be called by DTrace providers in their attach(9E) entry point. 7555 */ 7556 int 7557 dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv, 7558 cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp) 7559 { 7560 dtrace_provider_t *provider; 7561 7562 if (name == NULL || pap == NULL || pops == NULL || idp == NULL) { 7563 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7564 "arguments", name ? name : "<NULL>"); 7565 return (EINVAL); 7566 } 7567 7568 if (name[0] == '\0' || dtrace_badname(name)) { 7569 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7570 "provider name", name); 7571 return (EINVAL); 7572 } 7573 7574 if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) || 7575 pops->dtps_enable == NULL || pops->dtps_disable == NULL || 7576 pops->dtps_destroy == NULL || 7577 ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) { 7578 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7579 "provider ops", name); 7580 return (EINVAL); 7581 } 7582 7583 if (dtrace_badattr(&pap->dtpa_provider) || 7584 dtrace_badattr(&pap->dtpa_mod) || 7585 dtrace_badattr(&pap->dtpa_func) || 7586 dtrace_badattr(&pap->dtpa_name) || 7587 dtrace_badattr(&pap->dtpa_args)) { 7588 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7589 "provider attributes", name); 7590 return (EINVAL); 7591 } 7592 7593 if (priv & ~DTRACE_PRIV_ALL) { 7594 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7595 "privilege attributes", name); 7596 return (EINVAL); 7597 } 7598 7599 if ((priv & DTRACE_PRIV_KERNEL) && 7600 (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) && 7601 pops->dtps_usermode == NULL) { 7602 cmn_err(CE_WARN, "failed to register provider '%s': need " 7603 "dtps_usermode() op for given privilege attributes", name); 7604 return (EINVAL); 7605 } 7606 7607 provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP); 7608 provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); 7609 (void) strcpy(provider->dtpv_name, name); 7610 7611 provider->dtpv_attr = *pap; 7612 provider->dtpv_priv.dtpp_flags = priv; 7613 if (cr != NULL) { 7614 provider->dtpv_priv.dtpp_uid = crgetuid(cr); 7615 provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr); 7616 } 7617 provider->dtpv_pops = *pops; 7618 7619 if (pops->dtps_provide == NULL) { 7620 ASSERT(pops->dtps_provide_module != NULL); 7621 provider->dtpv_pops.dtps_provide = 7622 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop; 7623 } 7624 7625 if (pops->dtps_provide_module == NULL) { 7626 ASSERT(pops->dtps_provide != NULL); 7627 provider->dtpv_pops.dtps_provide_module = 7628 (void (*)(void *, modctl_t *))dtrace_nullop; 7629 } 7630 7631 if (pops->dtps_suspend == NULL) { 7632 ASSERT(pops->dtps_resume == NULL); 7633 provider->dtpv_pops.dtps_suspend = 7634 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; 7635 provider->dtpv_pops.dtps_resume = 7636 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; 7637 } 7638 7639 provider->dtpv_arg = arg; 7640 *idp = (dtrace_provider_id_t)provider; 7641 7642 if (pops == &dtrace_provider_ops) { 7643 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 7644 ASSERT(MUTEX_HELD(&dtrace_lock)); 7645 ASSERT(dtrace_anon.dta_enabling == NULL); 7646 7647 /* 7648 * We make sure that the DTrace provider is at the head of 7649 * the provider chain. 7650 */ 7651 provider->dtpv_next = dtrace_provider; 7652 dtrace_provider = provider; 7653 return (0); 7654 } 7655 7656 mutex_enter(&dtrace_provider_lock); 7657 mutex_enter(&dtrace_lock); 7658 7659 /* 7660 * If there is at least one provider registered, we'll add this 7661 * provider after the first provider. 7662 */ 7663 if (dtrace_provider != NULL) { 7664 provider->dtpv_next = dtrace_provider->dtpv_next; 7665 dtrace_provider->dtpv_next = provider; 7666 } else { 7667 dtrace_provider = provider; 7668 } 7669 7670 if (dtrace_retained != NULL) { 7671 dtrace_enabling_provide(provider); 7672 7673 /* 7674 * Now we need to call dtrace_enabling_matchall() -- which 7675 * will acquire cpu_lock and dtrace_lock. We therefore need 7676 * to drop all of our locks before calling into it... 7677 */ 7678 mutex_exit(&dtrace_lock); 7679 mutex_exit(&dtrace_provider_lock); 7680 dtrace_enabling_matchall(); 7681 7682 return (0); 7683 } 7684 7685 mutex_exit(&dtrace_lock); 7686 mutex_exit(&dtrace_provider_lock); 7687 7688 return (0); 7689 } 7690 7691 /* 7692 * Unregister the specified provider from the DTrace framework. This should 7693 * generally be called by DTrace providers in their detach(9E) entry point. 7694 */ 7695 int 7696 dtrace_unregister(dtrace_provider_id_t id) 7697 { 7698 dtrace_provider_t *old = (dtrace_provider_t *)id; 7699 dtrace_provider_t *prev = NULL; 7700 int i, self = 0, noreap = 0; 7701 dtrace_probe_t *probe, *first = NULL; 7702 7703 if (old->dtpv_pops.dtps_enable == 7704 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop) { 7705 /* 7706 * If DTrace itself is the provider, we're called with locks 7707 * already held. 7708 */ 7709 ASSERT(old == dtrace_provider); 7710 #if defined(sun) 7711 ASSERT(dtrace_devi != NULL); 7712 #endif 7713 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 7714 ASSERT(MUTEX_HELD(&dtrace_lock)); 7715 self = 1; 7716 7717 if (dtrace_provider->dtpv_next != NULL) { 7718 /* 7719 * There's another provider here; return failure. 7720 */ 7721 return (EBUSY); 7722 } 7723 } else { 7724 mutex_enter(&dtrace_provider_lock); 7725 #if defined(sun) 7726 mutex_enter(&mod_lock); 7727 #endif 7728 mutex_enter(&dtrace_lock); 7729 } 7730 7731 /* 7732 * If anyone has /dev/dtrace open, or if there are anonymous enabled 7733 * probes, we refuse to let providers slither away, unless this 7734 * provider has already been explicitly invalidated. 7735 */ 7736 if (!old->dtpv_defunct && 7737 (dtrace_opens || (dtrace_anon.dta_state != NULL && 7738 dtrace_anon.dta_state->dts_necbs > 0))) { 7739 if (!self) { 7740 mutex_exit(&dtrace_lock); 7741 #if defined(sun) 7742 mutex_exit(&mod_lock); 7743 #endif 7744 mutex_exit(&dtrace_provider_lock); 7745 } 7746 return (EBUSY); 7747 } 7748 7749 /* 7750 * Attempt to destroy the probes associated with this provider. 7751 */ 7752 for (i = 0; i < dtrace_nprobes; i++) { 7753 if ((probe = dtrace_probes[i]) == NULL) 7754 continue; 7755 7756 if (probe->dtpr_provider != old) 7757 continue; 7758 7759 if (probe->dtpr_ecb == NULL) 7760 continue; 7761 7762 /* 7763 * If we are trying to unregister a defunct provider, and the 7764 * provider was made defunct within the interval dictated by 7765 * dtrace_unregister_defunct_reap, we'll (asynchronously) 7766 * attempt to reap our enablings. To denote that the provider 7767 * should reattempt to unregister itself at some point in the 7768 * future, we will return a differentiable error code (EAGAIN 7769 * instead of EBUSY) in this case. 7770 */ 7771 if (dtrace_gethrtime() - old->dtpv_defunct > 7772 dtrace_unregister_defunct_reap) 7773 noreap = 1; 7774 7775 if (!self) { 7776 mutex_exit(&dtrace_lock); 7777 #if defined(sun) 7778 mutex_exit(&mod_lock); 7779 #endif 7780 mutex_exit(&dtrace_provider_lock); 7781 } 7782 7783 if (noreap) 7784 return (EBUSY); 7785 7786 (void) taskq_dispatch(dtrace_taskq, 7787 (task_func_t *)dtrace_enabling_reap, NULL, TQ_SLEEP); 7788 7789 return (EAGAIN); 7790 } 7791 7792 /* 7793 * All of the probes for this provider are disabled; we can safely 7794 * remove all of them from their hash chains and from the probe array. 7795 */ 7796 for (i = 0; i < dtrace_nprobes; i++) { 7797 if ((probe = dtrace_probes[i]) == NULL) 7798 continue; 7799 7800 if (probe->dtpr_provider != old) 7801 continue; 7802 7803 dtrace_probes[i] = NULL; 7804 7805 dtrace_hash_remove(dtrace_bymod, probe); 7806 dtrace_hash_remove(dtrace_byfunc, probe); 7807 dtrace_hash_remove(dtrace_byname, probe); 7808 7809 if (first == NULL) { 7810 first = probe; 7811 probe->dtpr_nextmod = NULL; 7812 } else { 7813 probe->dtpr_nextmod = first; 7814 first = probe; 7815 } 7816 } 7817 7818 /* 7819 * The provider's probes have been removed from the hash chains and 7820 * from the probe array. Now issue a dtrace_sync() to be sure that 7821 * everyone has cleared out from any probe array processing. 7822 */ 7823 dtrace_sync(); 7824 7825 for (probe = first; probe != NULL; probe = first) { 7826 first = probe->dtpr_nextmod; 7827 7828 old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id, 7829 probe->dtpr_arg); 7830 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); 7831 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); 7832 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); 7833 #if defined(sun) 7834 vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1); 7835 #else 7836 free_unr(dtrace_arena, probe->dtpr_id); 7837 #endif 7838 kmem_free(probe, sizeof (dtrace_probe_t)); 7839 } 7840 7841 if ((prev = dtrace_provider) == old) { 7842 #if defined(sun) 7843 ASSERT(self || dtrace_devi == NULL); 7844 ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL); 7845 #endif 7846 dtrace_provider = old->dtpv_next; 7847 } else { 7848 while (prev != NULL && prev->dtpv_next != old) 7849 prev = prev->dtpv_next; 7850 7851 if (prev == NULL) { 7852 panic("attempt to unregister non-existent " 7853 "dtrace provider %p\n", (void *)id); 7854 } 7855 7856 prev->dtpv_next = old->dtpv_next; 7857 } 7858 7859 if (!self) { 7860 mutex_exit(&dtrace_lock); 7861 #if defined(sun) 7862 mutex_exit(&mod_lock); 7863 #endif 7864 mutex_exit(&dtrace_provider_lock); 7865 } 7866 7867 kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1); 7868 kmem_free(old, sizeof (dtrace_provider_t)); 7869 7870 return (0); 7871 } 7872 7873 /* 7874 * Invalidate the specified provider. All subsequent probe lookups for the 7875 * specified provider will fail, but its probes will not be removed. 7876 */ 7877 void 7878 dtrace_invalidate(dtrace_provider_id_t id) 7879 { 7880 dtrace_provider_t *pvp = (dtrace_provider_t *)id; 7881 7882 ASSERT(pvp->dtpv_pops.dtps_enable != 7883 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop); 7884 7885 mutex_enter(&dtrace_provider_lock); 7886 mutex_enter(&dtrace_lock); 7887 7888 pvp->dtpv_defunct = dtrace_gethrtime(); 7889 7890 mutex_exit(&dtrace_lock); 7891 mutex_exit(&dtrace_provider_lock); 7892 } 7893 7894 /* 7895 * Indicate whether or not DTrace has attached. 7896 */ 7897 int 7898 dtrace_attached(void) 7899 { 7900 /* 7901 * dtrace_provider will be non-NULL iff the DTrace driver has 7902 * attached. (It's non-NULL because DTrace is always itself a 7903 * provider.) 7904 */ 7905 return (dtrace_provider != NULL); 7906 } 7907 7908 /* 7909 * Remove all the unenabled probes for the given provider. This function is 7910 * not unlike dtrace_unregister(), except that it doesn't remove the provider 7911 * -- just as many of its associated probes as it can. 7912 */ 7913 int 7914 dtrace_condense(dtrace_provider_id_t id) 7915 { 7916 dtrace_provider_t *prov = (dtrace_provider_t *)id; 7917 int i; 7918 dtrace_probe_t *probe; 7919 7920 /* 7921 * Make sure this isn't the dtrace provider itself. 7922 */ 7923 ASSERT(prov->dtpv_pops.dtps_enable != 7924 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop); 7925 7926 mutex_enter(&dtrace_provider_lock); 7927 mutex_enter(&dtrace_lock); 7928 7929 /* 7930 * Attempt to destroy the probes associated with this provider. 7931 */ 7932 for (i = 0; i < dtrace_nprobes; i++) { 7933 if ((probe = dtrace_probes[i]) == NULL) 7934 continue; 7935 7936 if (probe->dtpr_provider != prov) 7937 continue; 7938 7939 if (probe->dtpr_ecb != NULL) 7940 continue; 7941 7942 dtrace_probes[i] = NULL; 7943 7944 dtrace_hash_remove(dtrace_bymod, probe); 7945 dtrace_hash_remove(dtrace_byfunc, probe); 7946 dtrace_hash_remove(dtrace_byname, probe); 7947 7948 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1, 7949 probe->dtpr_arg); 7950 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); 7951 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); 7952 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); 7953 kmem_free(probe, sizeof (dtrace_probe_t)); 7954 #if defined(sun) 7955 vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1); 7956 #else 7957 free_unr(dtrace_arena, i + 1); 7958 #endif 7959 } 7960 7961 mutex_exit(&dtrace_lock); 7962 mutex_exit(&dtrace_provider_lock); 7963 7964 return (0); 7965 } 7966 7967 /* 7968 * DTrace Probe Management Functions 7969 * 7970 * The functions in this section perform the DTrace probe management, 7971 * including functions to create probes, look-up probes, and call into the 7972 * providers to request that probes be provided. Some of these functions are 7973 * in the Provider-to-Framework API; these functions can be identified by the 7974 * fact that they are not declared "static". 7975 */ 7976 7977 /* 7978 * Create a probe with the specified module name, function name, and name. 7979 */ 7980 dtrace_id_t 7981 dtrace_probe_create(dtrace_provider_id_t prov, const char *mod, 7982 const char *func, const char *name, int aframes, void *arg) 7983 { 7984 dtrace_probe_t *probe, **probes; 7985 dtrace_provider_t *provider = (dtrace_provider_t *)prov; 7986 dtrace_id_t id; 7987 7988 if (provider == dtrace_provider) { 7989 ASSERT(MUTEX_HELD(&dtrace_lock)); 7990 } else { 7991 mutex_enter(&dtrace_lock); 7992 } 7993 7994 #if defined(sun) 7995 id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1, 7996 VM_BESTFIT | VM_SLEEP); 7997 #else 7998 id = alloc_unr(dtrace_arena); 7999 #endif 8000 probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP); 8001 8002 probe->dtpr_id = id; 8003 probe->dtpr_gen = dtrace_probegen++; 8004 probe->dtpr_mod = dtrace_strdup(mod); 8005 probe->dtpr_func = dtrace_strdup(func); 8006 probe->dtpr_name = dtrace_strdup(name); 8007 probe->dtpr_arg = arg; 8008 probe->dtpr_aframes = aframes; 8009 probe->dtpr_provider = provider; 8010 8011 dtrace_hash_add(dtrace_bymod, probe); 8012 dtrace_hash_add(dtrace_byfunc, probe); 8013 dtrace_hash_add(dtrace_byname, probe); 8014 8015 if (id - 1 >= dtrace_nprobes) { 8016 size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *); 8017 size_t nsize = osize << 1; 8018 8019 if (nsize == 0) { 8020 ASSERT(osize == 0); 8021 ASSERT(dtrace_probes == NULL); 8022 nsize = sizeof (dtrace_probe_t *); 8023 } 8024 8025 probes = kmem_zalloc(nsize, KM_SLEEP); 8026 8027 if (dtrace_probes == NULL) { 8028 ASSERT(osize == 0); 8029 dtrace_probes = probes; 8030 dtrace_nprobes = 1; 8031 } else { 8032 dtrace_probe_t **oprobes = dtrace_probes; 8033 8034 bcopy(oprobes, probes, osize); 8035 dtrace_membar_producer(); 8036 dtrace_probes = probes; 8037 8038 dtrace_sync(); 8039 8040 /* 8041 * All CPUs are now seeing the new probes array; we can 8042 * safely free the old array. 8043 */ 8044 kmem_free(oprobes, osize); 8045 dtrace_nprobes <<= 1; 8046 } 8047 8048 ASSERT(id - 1 < dtrace_nprobes); 8049 } 8050 8051 ASSERT(dtrace_probes[id - 1] == NULL); 8052 dtrace_probes[id - 1] = probe; 8053 8054 if (provider != dtrace_provider) 8055 mutex_exit(&dtrace_lock); 8056 8057 return (id); 8058 } 8059 8060 static dtrace_probe_t * 8061 dtrace_probe_lookup_id(dtrace_id_t id) 8062 { 8063 ASSERT(MUTEX_HELD(&dtrace_lock)); 8064 8065 if (id == 0 || id > dtrace_nprobes) 8066 return (NULL); 8067 8068 return (dtrace_probes[id - 1]); 8069 } 8070 8071 static int 8072 dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg) 8073 { 8074 *((dtrace_id_t *)arg) = probe->dtpr_id; 8075 8076 return (DTRACE_MATCH_DONE); 8077 } 8078 8079 /* 8080 * Look up a probe based on provider and one or more of module name, function 8081 * name and probe name. 8082 */ 8083 dtrace_id_t 8084 dtrace_probe_lookup(dtrace_provider_id_t prid, char *mod, 8085 char *func, char *name) 8086 { 8087 dtrace_probekey_t pkey; 8088 dtrace_id_t id; 8089 int match; 8090 8091 pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name; 8092 pkey.dtpk_pmatch = &dtrace_match_string; 8093 pkey.dtpk_mod = mod; 8094 pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul; 8095 pkey.dtpk_func = func; 8096 pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul; 8097 pkey.dtpk_name = name; 8098 pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul; 8099 pkey.dtpk_id = DTRACE_IDNONE; 8100 8101 mutex_enter(&dtrace_lock); 8102 match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0, 8103 dtrace_probe_lookup_match, &id); 8104 mutex_exit(&dtrace_lock); 8105 8106 ASSERT(match == 1 || match == 0); 8107 return (match ? id : 0); 8108 } 8109 8110 /* 8111 * Returns the probe argument associated with the specified probe. 8112 */ 8113 void * 8114 dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid) 8115 { 8116 dtrace_probe_t *probe; 8117 void *rval = NULL; 8118 8119 mutex_enter(&dtrace_lock); 8120 8121 if ((probe = dtrace_probe_lookup_id(pid)) != NULL && 8122 probe->dtpr_provider == (dtrace_provider_t *)id) 8123 rval = probe->dtpr_arg; 8124 8125 mutex_exit(&dtrace_lock); 8126 8127 return (rval); 8128 } 8129 8130 /* 8131 * Copy a probe into a probe description. 8132 */ 8133 static void 8134 dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp) 8135 { 8136 bzero(pdp, sizeof (dtrace_probedesc_t)); 8137 pdp->dtpd_id = prp->dtpr_id; 8138 8139 (void) strncpy(pdp->dtpd_provider, 8140 prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1); 8141 8142 (void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1); 8143 (void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1); 8144 (void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1); 8145 } 8146 8147 /* 8148 * Called to indicate that a probe -- or probes -- should be provided by a 8149 * specfied provider. If the specified description is NULL, the provider will 8150 * be told to provide all of its probes. (This is done whenever a new 8151 * consumer comes along, or whenever a retained enabling is to be matched.) If 8152 * the specified description is non-NULL, the provider is given the 8153 * opportunity to dynamically provide the specified probe, allowing providers 8154 * to support the creation of probes on-the-fly. (So-called _autocreated_ 8155 * probes.) If the provider is NULL, the operations will be applied to all 8156 * providers; if the provider is non-NULL the operations will only be applied 8157 * to the specified provider. The dtrace_provider_lock must be held, and the 8158 * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation 8159 * will need to grab the dtrace_lock when it reenters the framework through 8160 * dtrace_probe_lookup(), dtrace_probe_create(), etc. 8161 */ 8162 static void 8163 dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv) 8164 { 8165 #if defined(sun) 8166 modctl_t *ctl; 8167 #endif 8168 int all = 0; 8169 8170 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 8171 8172 if (prv == NULL) { 8173 all = 1; 8174 prv = dtrace_provider; 8175 } 8176 8177 do { 8178 /* 8179 * First, call the blanket provide operation. 8180 */ 8181 prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc); 8182 8183 #if defined(sun) 8184 /* 8185 * Now call the per-module provide operation. We will grab 8186 * mod_lock to prevent the list from being modified. Note 8187 * that this also prevents the mod_busy bits from changing. 8188 * (mod_busy can only be changed with mod_lock held.) 8189 */ 8190 mutex_enter(&mod_lock); 8191 8192 ctl = &modules; 8193 do { 8194 if (ctl->mod_busy || ctl->mod_mp == NULL) 8195 continue; 8196 8197 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); 8198 8199 } while ((ctl = ctl->mod_next) != &modules); 8200 8201 mutex_exit(&mod_lock); 8202 #endif 8203 } while (all && (prv = prv->dtpv_next) != NULL); 8204 } 8205 8206 #if defined(sun) 8207 /* 8208 * Iterate over each probe, and call the Framework-to-Provider API function 8209 * denoted by offs. 8210 */ 8211 static void 8212 dtrace_probe_foreach(uintptr_t offs) 8213 { 8214 dtrace_provider_t *prov; 8215 void (*func)(void *, dtrace_id_t, void *); 8216 dtrace_probe_t *probe; 8217 dtrace_icookie_t cookie; 8218 int i; 8219 8220 /* 8221 * We disable interrupts to walk through the probe array. This is 8222 * safe -- the dtrace_sync() in dtrace_unregister() assures that we 8223 * won't see stale data. 8224 */ 8225 cookie = dtrace_interrupt_disable(); 8226 8227 for (i = 0; i < dtrace_nprobes; i++) { 8228 if ((probe = dtrace_probes[i]) == NULL) 8229 continue; 8230 8231 if (probe->dtpr_ecb == NULL) { 8232 /* 8233 * This probe isn't enabled -- don't call the function. 8234 */ 8235 continue; 8236 } 8237 8238 prov = probe->dtpr_provider; 8239 func = *((void(**)(void *, dtrace_id_t, void *)) 8240 ((uintptr_t)&prov->dtpv_pops + offs)); 8241 8242 func(prov->dtpv_arg, i + 1, probe->dtpr_arg); 8243 } 8244 8245 dtrace_interrupt_enable(cookie); 8246 } 8247 #endif 8248 8249 static int 8250 dtrace_probe_enable(dtrace_probedesc_t *desc, dtrace_enabling_t *enab) 8251 { 8252 dtrace_probekey_t pkey; 8253 uint32_t priv; 8254 uid_t uid; 8255 zoneid_t zoneid; 8256 8257 ASSERT(MUTEX_HELD(&dtrace_lock)); 8258 dtrace_ecb_create_cache = NULL; 8259 8260 if (desc == NULL) { 8261 /* 8262 * If we're passed a NULL description, we're being asked to 8263 * create an ECB with a NULL probe. 8264 */ 8265 (void) dtrace_ecb_create_enable(NULL, enab); 8266 return (0); 8267 } 8268 8269 dtrace_probekey(desc, &pkey); 8270 dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred, 8271 &priv, &uid, &zoneid); 8272 8273 return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable, 8274 enab)); 8275 } 8276 8277 /* 8278 * DTrace Helper Provider Functions 8279 */ 8280 static void 8281 dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr) 8282 { 8283 attr->dtat_name = DOF_ATTR_NAME(dofattr); 8284 attr->dtat_data = DOF_ATTR_DATA(dofattr); 8285 attr->dtat_class = DOF_ATTR_CLASS(dofattr); 8286 } 8287 8288 static void 8289 dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov, 8290 const dof_provider_t *dofprov, char *strtab) 8291 { 8292 hprov->dthpv_provname = strtab + dofprov->dofpv_name; 8293 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider, 8294 dofprov->dofpv_provattr); 8295 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod, 8296 dofprov->dofpv_modattr); 8297 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func, 8298 dofprov->dofpv_funcattr); 8299 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name, 8300 dofprov->dofpv_nameattr); 8301 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args, 8302 dofprov->dofpv_argsattr); 8303 } 8304 8305 static void 8306 dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) 8307 { 8308 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8309 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8310 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; 8311 dof_provider_t *provider; 8312 dof_probe_t *probe; 8313 uint32_t *off, *enoff; 8314 uint8_t *arg; 8315 char *strtab; 8316 uint_t i, nprobes; 8317 dtrace_helper_provdesc_t dhpv; 8318 dtrace_helper_probedesc_t dhpb; 8319 dtrace_meta_t *meta = dtrace_meta_pid; 8320 dtrace_mops_t *mops = &meta->dtm_mops; 8321 void *parg; 8322 8323 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); 8324 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8325 provider->dofpv_strtab * dof->dofh_secsize); 8326 prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8327 provider->dofpv_probes * dof->dofh_secsize); 8328 arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8329 provider->dofpv_prargs * dof->dofh_secsize); 8330 off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8331 provider->dofpv_proffs * dof->dofh_secsize); 8332 8333 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); 8334 off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset); 8335 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); 8336 enoff = NULL; 8337 8338 /* 8339 * See dtrace_helper_provider_validate(). 8340 */ 8341 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && 8342 provider->dofpv_prenoffs != DOF_SECT_NONE) { 8343 enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8344 provider->dofpv_prenoffs * dof->dofh_secsize); 8345 enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset); 8346 } 8347 8348 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; 8349 8350 /* 8351 * Create the provider. 8352 */ 8353 dtrace_dofprov2hprov(&dhpv, provider, strtab); 8354 8355 if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL) 8356 return; 8357 8358 meta->dtm_count++; 8359 8360 /* 8361 * Create the probes. 8362 */ 8363 for (i = 0; i < nprobes; i++) { 8364 probe = (dof_probe_t *)(uintptr_t)(daddr + 8365 prb_sec->dofs_offset + i * prb_sec->dofs_entsize); 8366 8367 dhpb.dthpb_mod = dhp->dofhp_mod; 8368 dhpb.dthpb_func = strtab + probe->dofpr_func; 8369 dhpb.dthpb_name = strtab + probe->dofpr_name; 8370 dhpb.dthpb_base = probe->dofpr_addr; 8371 dhpb.dthpb_offs = off + probe->dofpr_offidx; 8372 dhpb.dthpb_noffs = probe->dofpr_noffs; 8373 if (enoff != NULL) { 8374 dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx; 8375 dhpb.dthpb_nenoffs = probe->dofpr_nenoffs; 8376 } else { 8377 dhpb.dthpb_enoffs = NULL; 8378 dhpb.dthpb_nenoffs = 0; 8379 } 8380 dhpb.dthpb_args = arg + probe->dofpr_argidx; 8381 dhpb.dthpb_nargc = probe->dofpr_nargc; 8382 dhpb.dthpb_xargc = probe->dofpr_xargc; 8383 dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv; 8384 dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv; 8385 8386 mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb); 8387 } 8388 } 8389 8390 static void 8391 dtrace_helper_provide(dof_helper_t *dhp, pid_t pid) 8392 { 8393 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8394 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8395 int i; 8396 8397 ASSERT(MUTEX_HELD(&dtrace_meta_lock)); 8398 8399 for (i = 0; i < dof->dofh_secnum; i++) { 8400 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + 8401 dof->dofh_secoff + i * dof->dofh_secsize); 8402 8403 if (sec->dofs_type != DOF_SECT_PROVIDER) 8404 continue; 8405 8406 dtrace_helper_provide_one(dhp, sec, pid); 8407 } 8408 8409 /* 8410 * We may have just created probes, so we must now rematch against 8411 * any retained enablings. Note that this call will acquire both 8412 * cpu_lock and dtrace_lock; the fact that we are holding 8413 * dtrace_meta_lock now is what defines the ordering with respect to 8414 * these three locks. 8415 */ 8416 dtrace_enabling_matchall(); 8417 } 8418 8419 static void 8420 dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) 8421 { 8422 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8423 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8424 dof_sec_t *str_sec; 8425 dof_provider_t *provider; 8426 char *strtab; 8427 dtrace_helper_provdesc_t dhpv; 8428 dtrace_meta_t *meta = dtrace_meta_pid; 8429 dtrace_mops_t *mops = &meta->dtm_mops; 8430 8431 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); 8432 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8433 provider->dofpv_strtab * dof->dofh_secsize); 8434 8435 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); 8436 8437 /* 8438 * Create the provider. 8439 */ 8440 dtrace_dofprov2hprov(&dhpv, provider, strtab); 8441 8442 mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid); 8443 8444 meta->dtm_count--; 8445 } 8446 8447 static void 8448 dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid) 8449 { 8450 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8451 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8452 int i; 8453 8454 ASSERT(MUTEX_HELD(&dtrace_meta_lock)); 8455 8456 for (i = 0; i < dof->dofh_secnum; i++) { 8457 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + 8458 dof->dofh_secoff + i * dof->dofh_secsize); 8459 8460 if (sec->dofs_type != DOF_SECT_PROVIDER) 8461 continue; 8462 8463 dtrace_helper_provider_remove_one(dhp, sec, pid); 8464 } 8465 } 8466 8467 /* 8468 * DTrace Meta Provider-to-Framework API Functions 8469 * 8470 * These functions implement the Meta Provider-to-Framework API, as described 8471 * in <sys/dtrace.h>. 8472 */ 8473 int 8474 dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg, 8475 dtrace_meta_provider_id_t *idp) 8476 { 8477 dtrace_meta_t *meta; 8478 dtrace_helpers_t *help, *next; 8479 int i; 8480 8481 *idp = DTRACE_METAPROVNONE; 8482 8483 /* 8484 * We strictly don't need the name, but we hold onto it for 8485 * debuggability. All hail error queues! 8486 */ 8487 if (name == NULL) { 8488 cmn_err(CE_WARN, "failed to register meta-provider: " 8489 "invalid name"); 8490 return (EINVAL); 8491 } 8492 8493 if (mops == NULL || 8494 mops->dtms_create_probe == NULL || 8495 mops->dtms_provide_pid == NULL || 8496 mops->dtms_remove_pid == NULL) { 8497 cmn_err(CE_WARN, "failed to register meta-register %s: " 8498 "invalid ops", name); 8499 return (EINVAL); 8500 } 8501 8502 meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP); 8503 meta->dtm_mops = *mops; 8504 meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); 8505 (void) strcpy(meta->dtm_name, name); 8506 meta->dtm_arg = arg; 8507 8508 mutex_enter(&dtrace_meta_lock); 8509 mutex_enter(&dtrace_lock); 8510 8511 if (dtrace_meta_pid != NULL) { 8512 mutex_exit(&dtrace_lock); 8513 mutex_exit(&dtrace_meta_lock); 8514 cmn_err(CE_WARN, "failed to register meta-register %s: " 8515 "user-land meta-provider exists", name); 8516 kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1); 8517 kmem_free(meta, sizeof (dtrace_meta_t)); 8518 return (EINVAL); 8519 } 8520 8521 dtrace_meta_pid = meta; 8522 *idp = (dtrace_meta_provider_id_t)meta; 8523 8524 /* 8525 * If there are providers and probes ready to go, pass them 8526 * off to the new meta provider now. 8527 */ 8528 8529 help = dtrace_deferred_pid; 8530 dtrace_deferred_pid = NULL; 8531 8532 mutex_exit(&dtrace_lock); 8533 8534 while (help != NULL) { 8535 for (i = 0; i < help->dthps_nprovs; i++) { 8536 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, 8537 help->dthps_pid); 8538 } 8539 8540 next = help->dthps_next; 8541 help->dthps_next = NULL; 8542 help->dthps_prev = NULL; 8543 help->dthps_deferred = 0; 8544 help = next; 8545 } 8546 8547 mutex_exit(&dtrace_meta_lock); 8548 8549 return (0); 8550 } 8551 8552 int 8553 dtrace_meta_unregister(dtrace_meta_provider_id_t id) 8554 { 8555 dtrace_meta_t **pp, *old = (dtrace_meta_t *)id; 8556 8557 mutex_enter(&dtrace_meta_lock); 8558 mutex_enter(&dtrace_lock); 8559 8560 if (old == dtrace_meta_pid) { 8561 pp = &dtrace_meta_pid; 8562 } else { 8563 panic("attempt to unregister non-existent " 8564 "dtrace meta-provider %p\n", (void *)old); 8565 } 8566 8567 if (old->dtm_count != 0) { 8568 mutex_exit(&dtrace_lock); 8569 mutex_exit(&dtrace_meta_lock); 8570 return (EBUSY); 8571 } 8572 8573 *pp = NULL; 8574 8575 mutex_exit(&dtrace_lock); 8576 mutex_exit(&dtrace_meta_lock); 8577 8578 kmem_free(old->dtm_name, strlen(old->dtm_name) + 1); 8579 kmem_free(old, sizeof (dtrace_meta_t)); 8580 8581 return (0); 8582 } 8583 8584 8585 /* 8586 * DTrace DIF Object Functions 8587 */ 8588 static int 8589 dtrace_difo_err(uint_t pc, const char *format, ...) 8590 { 8591 if (dtrace_err_verbose) { 8592 va_list alist; 8593 8594 (void) uprintf("dtrace DIF object error: [%u]: ", pc); 8595 va_start(alist, format); 8596 (void) vuprintf(format, alist); 8597 va_end(alist); 8598 } 8599 8600 #ifdef DTRACE_ERRDEBUG 8601 dtrace_errdebug(format); 8602 #endif 8603 return (1); 8604 } 8605 8606 /* 8607 * Validate a DTrace DIF object by checking the IR instructions. The following 8608 * rules are currently enforced by dtrace_difo_validate(): 8609 * 8610 * 1. Each instruction must have a valid opcode 8611 * 2. Each register, string, variable, or subroutine reference must be valid 8612 * 3. No instruction can modify register %r0 (must be zero) 8613 * 4. All instruction reserved bits must be set to zero 8614 * 5. The last instruction must be a "ret" instruction 8615 * 6. All branch targets must reference a valid instruction _after_ the branch 8616 */ 8617 static int 8618 dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs, 8619 cred_t *cr) 8620 { 8621 int err = 0, i; 8622 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; 8623 int kcheckload; 8624 uint_t pc; 8625 8626 kcheckload = cr == NULL || 8627 (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0; 8628 8629 dp->dtdo_destructive = 0; 8630 8631 for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) { 8632 dif_instr_t instr = dp->dtdo_buf[pc]; 8633 8634 uint_t r1 = DIF_INSTR_R1(instr); 8635 uint_t r2 = DIF_INSTR_R2(instr); 8636 uint_t rd = DIF_INSTR_RD(instr); 8637 uint_t rs = DIF_INSTR_RS(instr); 8638 uint_t label = DIF_INSTR_LABEL(instr); 8639 uint_t v = DIF_INSTR_VAR(instr); 8640 uint_t subr = DIF_INSTR_SUBR(instr); 8641 uint_t type = DIF_INSTR_TYPE(instr); 8642 uint_t op = DIF_INSTR_OP(instr); 8643 8644 switch (op) { 8645 case DIF_OP_OR: 8646 case DIF_OP_XOR: 8647 case DIF_OP_AND: 8648 case DIF_OP_SLL: 8649 case DIF_OP_SRL: 8650 case DIF_OP_SRA: 8651 case DIF_OP_SUB: 8652 case DIF_OP_ADD: 8653 case DIF_OP_MUL: 8654 case DIF_OP_SDIV: 8655 case DIF_OP_UDIV: 8656 case DIF_OP_SREM: 8657 case DIF_OP_UREM: 8658 case DIF_OP_COPYS: 8659 if (r1 >= nregs) 8660 err += efunc(pc, "invalid register %u\n", r1); 8661 if (r2 >= nregs) 8662 err += efunc(pc, "invalid register %u\n", r2); 8663 if (rd >= nregs) 8664 err += efunc(pc, "invalid register %u\n", rd); 8665 if (rd == 0) 8666 err += efunc(pc, "cannot write to %r0\n"); 8667 break; 8668 case DIF_OP_NOT: 8669 case DIF_OP_MOV: 8670 case DIF_OP_ALLOCS: 8671 if (r1 >= nregs) 8672 err += efunc(pc, "invalid register %u\n", r1); 8673 if (r2 != 0) 8674 err += efunc(pc, "non-zero reserved bits\n"); 8675 if (rd >= nregs) 8676 err += efunc(pc, "invalid register %u\n", rd); 8677 if (rd == 0) 8678 err += efunc(pc, "cannot write to %r0\n"); 8679 break; 8680 case DIF_OP_LDSB: 8681 case DIF_OP_LDSH: 8682 case DIF_OP_LDSW: 8683 case DIF_OP_LDUB: 8684 case DIF_OP_LDUH: 8685 case DIF_OP_LDUW: 8686 case DIF_OP_LDX: 8687 if (r1 >= nregs) 8688 err += efunc(pc, "invalid register %u\n", r1); 8689 if (r2 != 0) 8690 err += efunc(pc, "non-zero reserved bits\n"); 8691 if (rd >= nregs) 8692 err += efunc(pc, "invalid register %u\n", rd); 8693 if (rd == 0) 8694 err += efunc(pc, "cannot write to %r0\n"); 8695 if (kcheckload) 8696 dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op + 8697 DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd); 8698 break; 8699 case DIF_OP_RLDSB: 8700 case DIF_OP_RLDSH: 8701 case DIF_OP_RLDSW: 8702 case DIF_OP_RLDUB: 8703 case DIF_OP_RLDUH: 8704 case DIF_OP_RLDUW: 8705 case DIF_OP_RLDX: 8706 if (r1 >= nregs) 8707 err += efunc(pc, "invalid register %u\n", r1); 8708 if (r2 != 0) 8709 err += efunc(pc, "non-zero reserved bits\n"); 8710 if (rd >= nregs) 8711 err += efunc(pc, "invalid register %u\n", rd); 8712 if (rd == 0) 8713 err += efunc(pc, "cannot write to %r0\n"); 8714 break; 8715 case DIF_OP_ULDSB: 8716 case DIF_OP_ULDSH: 8717 case DIF_OP_ULDSW: 8718 case DIF_OP_ULDUB: 8719 case DIF_OP_ULDUH: 8720 case DIF_OP_ULDUW: 8721 case DIF_OP_ULDX: 8722 if (r1 >= nregs) 8723 err += efunc(pc, "invalid register %u\n", r1); 8724 if (r2 != 0) 8725 err += efunc(pc, "non-zero reserved bits\n"); 8726 if (rd >= nregs) 8727 err += efunc(pc, "invalid register %u\n", rd); 8728 if (rd == 0) 8729 err += efunc(pc, "cannot write to %r0\n"); 8730 break; 8731 case DIF_OP_STB: 8732 case DIF_OP_STH: 8733 case DIF_OP_STW: 8734 case DIF_OP_STX: 8735 if (r1 >= nregs) 8736 err += efunc(pc, "invalid register %u\n", r1); 8737 if (r2 != 0) 8738 err += efunc(pc, "non-zero reserved bits\n"); 8739 if (rd >= nregs) 8740 err += efunc(pc, "invalid register %u\n", rd); 8741 if (rd == 0) 8742 err += efunc(pc, "cannot write to 0 address\n"); 8743 break; 8744 case DIF_OP_CMP: 8745 case DIF_OP_SCMP: 8746 if (r1 >= nregs) 8747 err += efunc(pc, "invalid register %u\n", r1); 8748 if (r2 >= nregs) 8749 err += efunc(pc, "invalid register %u\n", r2); 8750 if (rd != 0) 8751 err += efunc(pc, "non-zero reserved bits\n"); 8752 break; 8753 case DIF_OP_TST: 8754 if (r1 >= nregs) 8755 err += efunc(pc, "invalid register %u\n", r1); 8756 if (r2 != 0 || rd != 0) 8757 err += efunc(pc, "non-zero reserved bits\n"); 8758 break; 8759 case DIF_OP_BA: 8760 case DIF_OP_BE: 8761 case DIF_OP_BNE: 8762 case DIF_OP_BG: 8763 case DIF_OP_BGU: 8764 case DIF_OP_BGE: 8765 case DIF_OP_BGEU: 8766 case DIF_OP_BL: 8767 case DIF_OP_BLU: 8768 case DIF_OP_BLE: 8769 case DIF_OP_BLEU: 8770 if (label >= dp->dtdo_len) { 8771 err += efunc(pc, "invalid branch target %u\n", 8772 label); 8773 } 8774 if (label <= pc) { 8775 err += efunc(pc, "backward branch to %u\n", 8776 label); 8777 } 8778 break; 8779 case DIF_OP_RET: 8780 if (r1 != 0 || r2 != 0) 8781 err += efunc(pc, "non-zero reserved bits\n"); 8782 if (rd >= nregs) 8783 err += efunc(pc, "invalid register %u\n", rd); 8784 break; 8785 case DIF_OP_NOP: 8786 case DIF_OP_POPTS: 8787 case DIF_OP_FLUSHTS: 8788 if (r1 != 0 || r2 != 0 || rd != 0) 8789 err += efunc(pc, "non-zero reserved bits\n"); 8790 break; 8791 case DIF_OP_SETX: 8792 if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) { 8793 err += efunc(pc, "invalid integer ref %u\n", 8794 DIF_INSTR_INTEGER(instr)); 8795 } 8796 if (rd >= nregs) 8797 err += efunc(pc, "invalid register %u\n", rd); 8798 if (rd == 0) 8799 err += efunc(pc, "cannot write to %r0\n"); 8800 break; 8801 case DIF_OP_SETS: 8802 if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) { 8803 err += efunc(pc, "invalid string ref %u\n", 8804 DIF_INSTR_STRING(instr)); 8805 } 8806 if (rd >= nregs) 8807 err += efunc(pc, "invalid register %u\n", rd); 8808 if (rd == 0) 8809 err += efunc(pc, "cannot write to %r0\n"); 8810 break; 8811 case DIF_OP_LDGA: 8812 case DIF_OP_LDTA: 8813 if (r1 > DIF_VAR_ARRAY_MAX) 8814 err += efunc(pc, "invalid array %u\n", r1); 8815 if (r2 >= nregs) 8816 err += efunc(pc, "invalid register %u\n", r2); 8817 if (rd >= nregs) 8818 err += efunc(pc, "invalid register %u\n", rd); 8819 if (rd == 0) 8820 err += efunc(pc, "cannot write to %r0\n"); 8821 break; 8822 case DIF_OP_LDGS: 8823 case DIF_OP_LDTS: 8824 case DIF_OP_LDLS: 8825 case DIF_OP_LDGAA: 8826 case DIF_OP_LDTAA: 8827 if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX) 8828 err += efunc(pc, "invalid variable %u\n", v); 8829 if (rd >= nregs) 8830 err += efunc(pc, "invalid register %u\n", rd); 8831 if (rd == 0) 8832 err += efunc(pc, "cannot write to %r0\n"); 8833 break; 8834 case DIF_OP_STGS: 8835 case DIF_OP_STTS: 8836 case DIF_OP_STLS: 8837 case DIF_OP_STGAA: 8838 case DIF_OP_STTAA: 8839 if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX) 8840 err += efunc(pc, "invalid variable %u\n", v); 8841 if (rs >= nregs) 8842 err += efunc(pc, "invalid register %u\n", rd); 8843 break; 8844 case DIF_OP_CALL: 8845 if (subr > DIF_SUBR_MAX) 8846 err += efunc(pc, "invalid subr %u\n", subr); 8847 if (rd >= nregs) 8848 err += efunc(pc, "invalid register %u\n", rd); 8849 if (rd == 0) 8850 err += efunc(pc, "cannot write to %r0\n"); 8851 8852 if (subr == DIF_SUBR_COPYOUT || 8853 subr == DIF_SUBR_COPYOUTSTR) { 8854 dp->dtdo_destructive = 1; 8855 } 8856 break; 8857 case DIF_OP_PUSHTR: 8858 if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF) 8859 err += efunc(pc, "invalid ref type %u\n", type); 8860 if (r2 >= nregs) 8861 err += efunc(pc, "invalid register %u\n", r2); 8862 if (rs >= nregs) 8863 err += efunc(pc, "invalid register %u\n", rs); 8864 break; 8865 case DIF_OP_PUSHTV: 8866 if (type != DIF_TYPE_CTF) 8867 err += efunc(pc, "invalid val type %u\n", type); 8868 if (r2 >= nregs) 8869 err += efunc(pc, "invalid register %u\n", r2); 8870 if (rs >= nregs) 8871 err += efunc(pc, "invalid register %u\n", rs); 8872 break; 8873 default: 8874 err += efunc(pc, "invalid opcode %u\n", 8875 DIF_INSTR_OP(instr)); 8876 } 8877 } 8878 8879 if (dp->dtdo_len != 0 && 8880 DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) { 8881 err += efunc(dp->dtdo_len - 1, 8882 "expected 'ret' as last DIF instruction\n"); 8883 } 8884 8885 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) { 8886 /* 8887 * If we're not returning by reference, the size must be either 8888 * 0 or the size of one of the base types. 8889 */ 8890 switch (dp->dtdo_rtype.dtdt_size) { 8891 case 0: 8892 case sizeof (uint8_t): 8893 case sizeof (uint16_t): 8894 case sizeof (uint32_t): 8895 case sizeof (uint64_t): 8896 break; 8897 8898 default: 8899 err += efunc(dp->dtdo_len - 1, "bad return size"); 8900 } 8901 } 8902 8903 for (i = 0; i < dp->dtdo_varlen && err == 0; i++) { 8904 dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL; 8905 dtrace_diftype_t *vt, *et; 8906 uint_t id, ndx; 8907 8908 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL && 8909 v->dtdv_scope != DIFV_SCOPE_THREAD && 8910 v->dtdv_scope != DIFV_SCOPE_LOCAL) { 8911 err += efunc(i, "unrecognized variable scope %d\n", 8912 v->dtdv_scope); 8913 break; 8914 } 8915 8916 if (v->dtdv_kind != DIFV_KIND_ARRAY && 8917 v->dtdv_kind != DIFV_KIND_SCALAR) { 8918 err += efunc(i, "unrecognized variable type %d\n", 8919 v->dtdv_kind); 8920 break; 8921 } 8922 8923 if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) { 8924 err += efunc(i, "%d exceeds variable id limit\n", id); 8925 break; 8926 } 8927 8928 if (id < DIF_VAR_OTHER_UBASE) 8929 continue; 8930 8931 /* 8932 * For user-defined variables, we need to check that this 8933 * definition is identical to any previous definition that we 8934 * encountered. 8935 */ 8936 ndx = id - DIF_VAR_OTHER_UBASE; 8937 8938 switch (v->dtdv_scope) { 8939 case DIFV_SCOPE_GLOBAL: 8940 if (ndx < vstate->dtvs_nglobals) { 8941 dtrace_statvar_t *svar; 8942 8943 if ((svar = vstate->dtvs_globals[ndx]) != NULL) 8944 existing = &svar->dtsv_var; 8945 } 8946 8947 break; 8948 8949 case DIFV_SCOPE_THREAD: 8950 if (ndx < vstate->dtvs_ntlocals) 8951 existing = &vstate->dtvs_tlocals[ndx]; 8952 break; 8953 8954 case DIFV_SCOPE_LOCAL: 8955 if (ndx < vstate->dtvs_nlocals) { 8956 dtrace_statvar_t *svar; 8957 8958 if ((svar = vstate->dtvs_locals[ndx]) != NULL) 8959 existing = &svar->dtsv_var; 8960 } 8961 8962 break; 8963 } 8964 8965 vt = &v->dtdv_type; 8966 8967 if (vt->dtdt_flags & DIF_TF_BYREF) { 8968 if (vt->dtdt_size == 0) { 8969 err += efunc(i, "zero-sized variable\n"); 8970 break; 8971 } 8972 8973 if (v->dtdv_scope == DIFV_SCOPE_GLOBAL && 8974 vt->dtdt_size > dtrace_global_maxsize) { 8975 err += efunc(i, "oversized by-ref global\n"); 8976 break; 8977 } 8978 } 8979 8980 if (existing == NULL || existing->dtdv_id == 0) 8981 continue; 8982 8983 ASSERT(existing->dtdv_id == v->dtdv_id); 8984 ASSERT(existing->dtdv_scope == v->dtdv_scope); 8985 8986 if (existing->dtdv_kind != v->dtdv_kind) 8987 err += efunc(i, "%d changed variable kind\n", id); 8988 8989 et = &existing->dtdv_type; 8990 8991 if (vt->dtdt_flags != et->dtdt_flags) { 8992 err += efunc(i, "%d changed variable type flags\n", id); 8993 break; 8994 } 8995 8996 if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) { 8997 err += efunc(i, "%d changed variable type size\n", id); 8998 break; 8999 } 9000 } 9001 9002 return (err); 9003 } 9004 9005 /* 9006 * Validate a DTrace DIF object that it is to be used as a helper. Helpers 9007 * are much more constrained than normal DIFOs. Specifically, they may 9008 * not: 9009 * 9010 * 1. Make calls to subroutines other than copyin(), copyinstr() or 9011 * miscellaneous string routines 9012 * 2. Access DTrace variables other than the args[] array, and the 9013 * curthread, pid, ppid, tid, execname, zonename, uid and gid variables. 9014 * 3. Have thread-local variables. 9015 * 4. Have dynamic variables. 9016 */ 9017 static int 9018 dtrace_difo_validate_helper(dtrace_difo_t *dp) 9019 { 9020 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; 9021 int err = 0; 9022 uint_t pc; 9023 9024 for (pc = 0; pc < dp->dtdo_len; pc++) { 9025 dif_instr_t instr = dp->dtdo_buf[pc]; 9026 9027 uint_t v = DIF_INSTR_VAR(instr); 9028 uint_t subr = DIF_INSTR_SUBR(instr); 9029 uint_t op = DIF_INSTR_OP(instr); 9030 9031 switch (op) { 9032 case DIF_OP_OR: 9033 case DIF_OP_XOR: 9034 case DIF_OP_AND: 9035 case DIF_OP_SLL: 9036 case DIF_OP_SRL: 9037 case DIF_OP_SRA: 9038 case DIF_OP_SUB: 9039 case DIF_OP_ADD: 9040 case DIF_OP_MUL: 9041 case DIF_OP_SDIV: 9042 case DIF_OP_UDIV: 9043 case DIF_OP_SREM: 9044 case DIF_OP_UREM: 9045 case DIF_OP_COPYS: 9046 case DIF_OP_NOT: 9047 case DIF_OP_MOV: 9048 case DIF_OP_RLDSB: 9049 case DIF_OP_RLDSH: 9050 case DIF_OP_RLDSW: 9051 case DIF_OP_RLDUB: 9052 case DIF_OP_RLDUH: 9053 case DIF_OP_RLDUW: 9054 case DIF_OP_RLDX: 9055 case DIF_OP_ULDSB: 9056 case DIF_OP_ULDSH: 9057 case DIF_OP_ULDSW: 9058 case DIF_OP_ULDUB: 9059 case DIF_OP_ULDUH: 9060 case DIF_OP_ULDUW: 9061 case DIF_OP_ULDX: 9062 case DIF_OP_STB: 9063 case DIF_OP_STH: 9064 case DIF_OP_STW: 9065 case DIF_OP_STX: 9066 case DIF_OP_ALLOCS: 9067 case DIF_OP_CMP: 9068 case DIF_OP_SCMP: 9069 case DIF_OP_TST: 9070 case DIF_OP_BA: 9071 case DIF_OP_BE: 9072 case DIF_OP_BNE: 9073 case DIF_OP_BG: 9074 case DIF_OP_BGU: 9075 case DIF_OP_BGE: 9076 case DIF_OP_BGEU: 9077 case DIF_OP_BL: 9078 case DIF_OP_BLU: 9079 case DIF_OP_BLE: 9080 case DIF_OP_BLEU: 9081 case DIF_OP_RET: 9082 case DIF_OP_NOP: 9083 case DIF_OP_POPTS: 9084 case DIF_OP_FLUSHTS: 9085 case DIF_OP_SETX: 9086 case DIF_OP_SETS: 9087 case DIF_OP_LDGA: 9088 case DIF_OP_LDLS: 9089 case DIF_OP_STGS: 9090 case DIF_OP_STLS: 9091 case DIF_OP_PUSHTR: 9092 case DIF_OP_PUSHTV: 9093 break; 9094 9095 case DIF_OP_LDGS: 9096 if (v >= DIF_VAR_OTHER_UBASE) 9097 break; 9098 9099 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) 9100 break; 9101 9102 if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID || 9103 v == DIF_VAR_PPID || v == DIF_VAR_TID || 9104 v == DIF_VAR_EXECARGS || 9105 v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME || 9106 v == DIF_VAR_UID || v == DIF_VAR_GID) 9107 break; 9108 9109 err += efunc(pc, "illegal variable %u\n", v); 9110 break; 9111 9112 case DIF_OP_LDTA: 9113 case DIF_OP_LDTS: 9114 case DIF_OP_LDGAA: 9115 case DIF_OP_LDTAA: 9116 err += efunc(pc, "illegal dynamic variable load\n"); 9117 break; 9118 9119 case DIF_OP_STTS: 9120 case DIF_OP_STGAA: 9121 case DIF_OP_STTAA: 9122 err += efunc(pc, "illegal dynamic variable store\n"); 9123 break; 9124 9125 case DIF_OP_CALL: 9126 if (subr == DIF_SUBR_ALLOCA || 9127 subr == DIF_SUBR_BCOPY || 9128 subr == DIF_SUBR_COPYIN || 9129 subr == DIF_SUBR_COPYINTO || 9130 subr == DIF_SUBR_COPYINSTR || 9131 subr == DIF_SUBR_INDEX || 9132 subr == DIF_SUBR_INET_NTOA || 9133 subr == DIF_SUBR_INET_NTOA6 || 9134 subr == DIF_SUBR_INET_NTOP || 9135 subr == DIF_SUBR_LLTOSTR || 9136 subr == DIF_SUBR_RINDEX || 9137 subr == DIF_SUBR_STRCHR || 9138 subr == DIF_SUBR_STRJOIN || 9139 subr == DIF_SUBR_STRRCHR || 9140 subr == DIF_SUBR_STRSTR || 9141 subr == DIF_SUBR_HTONS || 9142 subr == DIF_SUBR_HTONL || 9143 subr == DIF_SUBR_HTONLL || 9144 subr == DIF_SUBR_NTOHS || 9145 subr == DIF_SUBR_NTOHL || 9146 subr == DIF_SUBR_NTOHLL || 9147 subr == DIF_SUBR_MEMREF || 9148 #if !defined(sun) 9149 subr == DIF_SUBR_MEMSTR || 9150 #endif 9151 subr == DIF_SUBR_TYPEREF) 9152 break; 9153 9154 err += efunc(pc, "invalid subr %u\n", subr); 9155 break; 9156 9157 default: 9158 err += efunc(pc, "invalid opcode %u\n", 9159 DIF_INSTR_OP(instr)); 9160 } 9161 } 9162 9163 return (err); 9164 } 9165 9166 /* 9167 * Returns 1 if the expression in the DIF object can be cached on a per-thread 9168 * basis; 0 if not. 9169 */ 9170 static int 9171 dtrace_difo_cacheable(dtrace_difo_t *dp) 9172 { 9173 int i; 9174 9175 if (dp == NULL) 9176 return (0); 9177 9178 for (i = 0; i < dp->dtdo_varlen; i++) { 9179 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9180 9181 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL) 9182 continue; 9183 9184 switch (v->dtdv_id) { 9185 case DIF_VAR_CURTHREAD: 9186 case DIF_VAR_PID: 9187 case DIF_VAR_TID: 9188 case DIF_VAR_EXECARGS: 9189 case DIF_VAR_EXECNAME: 9190 case DIF_VAR_ZONENAME: 9191 break; 9192 9193 default: 9194 return (0); 9195 } 9196 } 9197 9198 /* 9199 * This DIF object may be cacheable. Now we need to look for any 9200 * array loading instructions, any memory loading instructions, or 9201 * any stores to thread-local variables. 9202 */ 9203 for (i = 0; i < dp->dtdo_len; i++) { 9204 uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]); 9205 9206 if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) || 9207 (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) || 9208 (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) || 9209 op == DIF_OP_LDGA || op == DIF_OP_STTS) 9210 return (0); 9211 } 9212 9213 return (1); 9214 } 9215 9216 static void 9217 dtrace_difo_hold(dtrace_difo_t *dp) 9218 { 9219 int i; 9220 9221 ASSERT(MUTEX_HELD(&dtrace_lock)); 9222 9223 dp->dtdo_refcnt++; 9224 ASSERT(dp->dtdo_refcnt != 0); 9225 9226 /* 9227 * We need to check this DIF object for references to the variable 9228 * DIF_VAR_VTIMESTAMP. 9229 */ 9230 for (i = 0; i < dp->dtdo_varlen; i++) { 9231 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9232 9233 if (v->dtdv_id != DIF_VAR_VTIMESTAMP) 9234 continue; 9235 9236 if (dtrace_vtime_references++ == 0) 9237 dtrace_vtime_enable(); 9238 } 9239 } 9240 9241 /* 9242 * This routine calculates the dynamic variable chunksize for a given DIF 9243 * object. The calculation is not fool-proof, and can probably be tricked by 9244 * malicious DIF -- but it works for all compiler-generated DIF. Because this 9245 * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail 9246 * if a dynamic variable size exceeds the chunksize. 9247 */ 9248 static void 9249 dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9250 { 9251 uint64_t sval = 0; 9252 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ 9253 const dif_instr_t *text = dp->dtdo_buf; 9254 uint_t pc, srd = 0; 9255 uint_t ttop = 0; 9256 size_t size, ksize; 9257 uint_t id, i; 9258 9259 for (pc = 0; pc < dp->dtdo_len; pc++) { 9260 dif_instr_t instr = text[pc]; 9261 uint_t op = DIF_INSTR_OP(instr); 9262 uint_t rd = DIF_INSTR_RD(instr); 9263 uint_t r1 = DIF_INSTR_R1(instr); 9264 uint_t nkeys = 0; 9265 uchar_t scope = 0; 9266 9267 dtrace_key_t *key = tupregs; 9268 9269 switch (op) { 9270 case DIF_OP_SETX: 9271 sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)]; 9272 srd = rd; 9273 continue; 9274 9275 case DIF_OP_STTS: 9276 key = &tupregs[DIF_DTR_NREGS]; 9277 key[0].dttk_size = 0; 9278 key[1].dttk_size = 0; 9279 nkeys = 2; 9280 scope = DIFV_SCOPE_THREAD; 9281 break; 9282 9283 case DIF_OP_STGAA: 9284 case DIF_OP_STTAA: 9285 nkeys = ttop; 9286 9287 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) 9288 key[nkeys++].dttk_size = 0; 9289 9290 key[nkeys++].dttk_size = 0; 9291 9292 if (op == DIF_OP_STTAA) { 9293 scope = DIFV_SCOPE_THREAD; 9294 } else { 9295 scope = DIFV_SCOPE_GLOBAL; 9296 } 9297 9298 break; 9299 9300 case DIF_OP_PUSHTR: 9301 if (ttop == DIF_DTR_NREGS) 9302 return; 9303 9304 if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) { 9305 /* 9306 * If the register for the size of the "pushtr" 9307 * is %r0 (or the value is 0) and the type is 9308 * a string, we'll use the system-wide default 9309 * string size. 9310 */ 9311 tupregs[ttop++].dttk_size = 9312 dtrace_strsize_default; 9313 } else { 9314 if (srd == 0) 9315 return; 9316 9317 tupregs[ttop++].dttk_size = sval; 9318 } 9319 9320 break; 9321 9322 case DIF_OP_PUSHTV: 9323 if (ttop == DIF_DTR_NREGS) 9324 return; 9325 9326 tupregs[ttop++].dttk_size = 0; 9327 break; 9328 9329 case DIF_OP_FLUSHTS: 9330 ttop = 0; 9331 break; 9332 9333 case DIF_OP_POPTS: 9334 if (ttop != 0) 9335 ttop--; 9336 break; 9337 } 9338 9339 sval = 0; 9340 srd = 0; 9341 9342 if (nkeys == 0) 9343 continue; 9344 9345 /* 9346 * We have a dynamic variable allocation; calculate its size. 9347 */ 9348 for (ksize = 0, i = 0; i < nkeys; i++) 9349 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); 9350 9351 size = sizeof (dtrace_dynvar_t); 9352 size += sizeof (dtrace_key_t) * (nkeys - 1); 9353 size += ksize; 9354 9355 /* 9356 * Now we need to determine the size of the stored data. 9357 */ 9358 id = DIF_INSTR_VAR(instr); 9359 9360 for (i = 0; i < dp->dtdo_varlen; i++) { 9361 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9362 9363 if (v->dtdv_id == id && v->dtdv_scope == scope) { 9364 size += v->dtdv_type.dtdt_size; 9365 break; 9366 } 9367 } 9368 9369 if (i == dp->dtdo_varlen) 9370 return; 9371 9372 /* 9373 * We have the size. If this is larger than the chunk size 9374 * for our dynamic variable state, reset the chunk size. 9375 */ 9376 size = P2ROUNDUP(size, sizeof (uint64_t)); 9377 9378 if (size > vstate->dtvs_dynvars.dtds_chunksize) 9379 vstate->dtvs_dynvars.dtds_chunksize = size; 9380 } 9381 } 9382 9383 static void 9384 dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9385 { 9386 int i, oldsvars, osz, nsz, otlocals, ntlocals; 9387 uint_t id; 9388 9389 ASSERT(MUTEX_HELD(&dtrace_lock)); 9390 ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0); 9391 9392 for (i = 0; i < dp->dtdo_varlen; i++) { 9393 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9394 dtrace_statvar_t *svar, ***svarp = NULL; 9395 size_t dsize = 0; 9396 uint8_t scope = v->dtdv_scope; 9397 int *np = NULL; 9398 9399 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) 9400 continue; 9401 9402 id -= DIF_VAR_OTHER_UBASE; 9403 9404 switch (scope) { 9405 case DIFV_SCOPE_THREAD: 9406 while (id >= (otlocals = vstate->dtvs_ntlocals)) { 9407 dtrace_difv_t *tlocals; 9408 9409 if ((ntlocals = (otlocals << 1)) == 0) 9410 ntlocals = 1; 9411 9412 osz = otlocals * sizeof (dtrace_difv_t); 9413 nsz = ntlocals * sizeof (dtrace_difv_t); 9414 9415 tlocals = kmem_zalloc(nsz, KM_SLEEP); 9416 9417 if (osz != 0) { 9418 bcopy(vstate->dtvs_tlocals, 9419 tlocals, osz); 9420 kmem_free(vstate->dtvs_tlocals, osz); 9421 } 9422 9423 vstate->dtvs_tlocals = tlocals; 9424 vstate->dtvs_ntlocals = ntlocals; 9425 } 9426 9427 vstate->dtvs_tlocals[id] = *v; 9428 continue; 9429 9430 case DIFV_SCOPE_LOCAL: 9431 np = &vstate->dtvs_nlocals; 9432 svarp = &vstate->dtvs_locals; 9433 9434 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) 9435 dsize = NCPU * (v->dtdv_type.dtdt_size + 9436 sizeof (uint64_t)); 9437 else 9438 dsize = NCPU * sizeof (uint64_t); 9439 9440 break; 9441 9442 case DIFV_SCOPE_GLOBAL: 9443 np = &vstate->dtvs_nglobals; 9444 svarp = &vstate->dtvs_globals; 9445 9446 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) 9447 dsize = v->dtdv_type.dtdt_size + 9448 sizeof (uint64_t); 9449 9450 break; 9451 9452 default: 9453 ASSERT(0); 9454 } 9455 9456 while (id >= (oldsvars = *np)) { 9457 dtrace_statvar_t **statics; 9458 int newsvars, oldsize, newsize; 9459 9460 if ((newsvars = (oldsvars << 1)) == 0) 9461 newsvars = 1; 9462 9463 oldsize = oldsvars * sizeof (dtrace_statvar_t *); 9464 newsize = newsvars * sizeof (dtrace_statvar_t *); 9465 9466 statics = kmem_zalloc(newsize, KM_SLEEP); 9467 9468 if (oldsize != 0) { 9469 bcopy(*svarp, statics, oldsize); 9470 kmem_free(*svarp, oldsize); 9471 } 9472 9473 *svarp = statics; 9474 *np = newsvars; 9475 } 9476 9477 if ((svar = (*svarp)[id]) == NULL) { 9478 svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP); 9479 svar->dtsv_var = *v; 9480 9481 if ((svar->dtsv_size = dsize) != 0) { 9482 svar->dtsv_data = (uint64_t)(uintptr_t) 9483 kmem_zalloc(dsize, KM_SLEEP); 9484 } 9485 9486 (*svarp)[id] = svar; 9487 } 9488 9489 svar->dtsv_refcnt++; 9490 } 9491 9492 dtrace_difo_chunksize(dp, vstate); 9493 dtrace_difo_hold(dp); 9494 } 9495 9496 static dtrace_difo_t * 9497 dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9498 { 9499 dtrace_difo_t *new; 9500 size_t sz; 9501 9502 ASSERT(dp->dtdo_buf != NULL); 9503 ASSERT(dp->dtdo_refcnt != 0); 9504 9505 new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); 9506 9507 ASSERT(dp->dtdo_buf != NULL); 9508 sz = dp->dtdo_len * sizeof (dif_instr_t); 9509 new->dtdo_buf = kmem_alloc(sz, KM_SLEEP); 9510 bcopy(dp->dtdo_buf, new->dtdo_buf, sz); 9511 new->dtdo_len = dp->dtdo_len; 9512 9513 if (dp->dtdo_strtab != NULL) { 9514 ASSERT(dp->dtdo_strlen != 0); 9515 new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP); 9516 bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen); 9517 new->dtdo_strlen = dp->dtdo_strlen; 9518 } 9519 9520 if (dp->dtdo_inttab != NULL) { 9521 ASSERT(dp->dtdo_intlen != 0); 9522 sz = dp->dtdo_intlen * sizeof (uint64_t); 9523 new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP); 9524 bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz); 9525 new->dtdo_intlen = dp->dtdo_intlen; 9526 } 9527 9528 if (dp->dtdo_vartab != NULL) { 9529 ASSERT(dp->dtdo_varlen != 0); 9530 sz = dp->dtdo_varlen * sizeof (dtrace_difv_t); 9531 new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP); 9532 bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz); 9533 new->dtdo_varlen = dp->dtdo_varlen; 9534 } 9535 9536 dtrace_difo_init(new, vstate); 9537 return (new); 9538 } 9539 9540 static void 9541 dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9542 { 9543 int i; 9544 9545 ASSERT(dp->dtdo_refcnt == 0); 9546 9547 for (i = 0; i < dp->dtdo_varlen; i++) { 9548 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9549 dtrace_statvar_t *svar, **svarp = NULL; 9550 uint_t id; 9551 uint8_t scope = v->dtdv_scope; 9552 int *np = NULL; 9553 9554 switch (scope) { 9555 case DIFV_SCOPE_THREAD: 9556 continue; 9557 9558 case DIFV_SCOPE_LOCAL: 9559 np = &vstate->dtvs_nlocals; 9560 svarp = vstate->dtvs_locals; 9561 break; 9562 9563 case DIFV_SCOPE_GLOBAL: 9564 np = &vstate->dtvs_nglobals; 9565 svarp = vstate->dtvs_globals; 9566 break; 9567 9568 default: 9569 ASSERT(0); 9570 } 9571 9572 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) 9573 continue; 9574 9575 id -= DIF_VAR_OTHER_UBASE; 9576 ASSERT(id < *np); 9577 9578 svar = svarp[id]; 9579 ASSERT(svar != NULL); 9580 ASSERT(svar->dtsv_refcnt > 0); 9581 9582 if (--svar->dtsv_refcnt > 0) 9583 continue; 9584 9585 if (svar->dtsv_size != 0) { 9586 ASSERT(svar->dtsv_data != 0); 9587 kmem_free((void *)(uintptr_t)svar->dtsv_data, 9588 svar->dtsv_size); 9589 } 9590 9591 kmem_free(svar, sizeof (dtrace_statvar_t)); 9592 svarp[id] = NULL; 9593 } 9594 9595 if (dp->dtdo_buf != NULL) 9596 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); 9597 if (dp->dtdo_inttab != NULL) 9598 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); 9599 if (dp->dtdo_strtab != NULL) 9600 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); 9601 if (dp->dtdo_vartab != NULL) 9602 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); 9603 9604 kmem_free(dp, sizeof (dtrace_difo_t)); 9605 } 9606 9607 static void 9608 dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9609 { 9610 int i; 9611 9612 ASSERT(MUTEX_HELD(&dtrace_lock)); 9613 ASSERT(dp->dtdo_refcnt != 0); 9614 9615 for (i = 0; i < dp->dtdo_varlen; i++) { 9616 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9617 9618 if (v->dtdv_id != DIF_VAR_VTIMESTAMP) 9619 continue; 9620 9621 ASSERT(dtrace_vtime_references > 0); 9622 if (--dtrace_vtime_references == 0) 9623 dtrace_vtime_disable(); 9624 } 9625 9626 if (--dp->dtdo_refcnt == 0) 9627 dtrace_difo_destroy(dp, vstate); 9628 } 9629 9630 /* 9631 * DTrace Format Functions 9632 */ 9633 static uint16_t 9634 dtrace_format_add(dtrace_state_t *state, char *str) 9635 { 9636 char *fmt, **new; 9637 uint16_t ndx, len = strlen(str) + 1; 9638 9639 fmt = kmem_zalloc(len, KM_SLEEP); 9640 bcopy(str, fmt, len); 9641 9642 for (ndx = 0; ndx < state->dts_nformats; ndx++) { 9643 if (state->dts_formats[ndx] == NULL) { 9644 state->dts_formats[ndx] = fmt; 9645 return (ndx + 1); 9646 } 9647 } 9648 9649 if (state->dts_nformats == USHRT_MAX) { 9650 /* 9651 * This is only likely if a denial-of-service attack is being 9652 * attempted. As such, it's okay to fail silently here. 9653 */ 9654 kmem_free(fmt, len); 9655 return (0); 9656 } 9657 9658 /* 9659 * For simplicity, we always resize the formats array to be exactly the 9660 * number of formats. 9661 */ 9662 ndx = state->dts_nformats++; 9663 new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP); 9664 9665 if (state->dts_formats != NULL) { 9666 ASSERT(ndx != 0); 9667 bcopy(state->dts_formats, new, ndx * sizeof (char *)); 9668 kmem_free(state->dts_formats, ndx * sizeof (char *)); 9669 } 9670 9671 state->dts_formats = new; 9672 state->dts_formats[ndx] = fmt; 9673 9674 return (ndx + 1); 9675 } 9676 9677 static void 9678 dtrace_format_remove(dtrace_state_t *state, uint16_t format) 9679 { 9680 char *fmt; 9681 9682 ASSERT(state->dts_formats != NULL); 9683 ASSERT(format <= state->dts_nformats); 9684 ASSERT(state->dts_formats[format - 1] != NULL); 9685 9686 fmt = state->dts_formats[format - 1]; 9687 kmem_free(fmt, strlen(fmt) + 1); 9688 state->dts_formats[format - 1] = NULL; 9689 } 9690 9691 static void 9692 dtrace_format_destroy(dtrace_state_t *state) 9693 { 9694 int i; 9695 9696 if (state->dts_nformats == 0) { 9697 ASSERT(state->dts_formats == NULL); 9698 return; 9699 } 9700 9701 ASSERT(state->dts_formats != NULL); 9702 9703 for (i = 0; i < state->dts_nformats; i++) { 9704 char *fmt = state->dts_formats[i]; 9705 9706 if (fmt == NULL) 9707 continue; 9708 9709 kmem_free(fmt, strlen(fmt) + 1); 9710 } 9711 9712 kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *)); 9713 state->dts_nformats = 0; 9714 state->dts_formats = NULL; 9715 } 9716 9717 /* 9718 * DTrace Predicate Functions 9719 */ 9720 static dtrace_predicate_t * 9721 dtrace_predicate_create(dtrace_difo_t *dp) 9722 { 9723 dtrace_predicate_t *pred; 9724 9725 ASSERT(MUTEX_HELD(&dtrace_lock)); 9726 ASSERT(dp->dtdo_refcnt != 0); 9727 9728 pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP); 9729 pred->dtp_difo = dp; 9730 pred->dtp_refcnt = 1; 9731 9732 if (!dtrace_difo_cacheable(dp)) 9733 return (pred); 9734 9735 if (dtrace_predcache_id == DTRACE_CACHEIDNONE) { 9736 /* 9737 * This is only theoretically possible -- we have had 2^32 9738 * cacheable predicates on this machine. We cannot allow any 9739 * more predicates to become cacheable: as unlikely as it is, 9740 * there may be a thread caching a (now stale) predicate cache 9741 * ID. (N.B.: the temptation is being successfully resisted to 9742 * have this cmn_err() "Holy shit -- we executed this code!") 9743 */ 9744 return (pred); 9745 } 9746 9747 pred->dtp_cacheid = dtrace_predcache_id++; 9748 9749 return (pred); 9750 } 9751 9752 static void 9753 dtrace_predicate_hold(dtrace_predicate_t *pred) 9754 { 9755 ASSERT(MUTEX_HELD(&dtrace_lock)); 9756 ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0); 9757 ASSERT(pred->dtp_refcnt > 0); 9758 9759 pred->dtp_refcnt++; 9760 } 9761 9762 static void 9763 dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate) 9764 { 9765 dtrace_difo_t *dp = pred->dtp_difo; 9766 9767 ASSERT(MUTEX_HELD(&dtrace_lock)); 9768 ASSERT(dp != NULL && dp->dtdo_refcnt != 0); 9769 ASSERT(pred->dtp_refcnt > 0); 9770 9771 if (--pred->dtp_refcnt == 0) { 9772 dtrace_difo_release(pred->dtp_difo, vstate); 9773 kmem_free(pred, sizeof (dtrace_predicate_t)); 9774 } 9775 } 9776 9777 /* 9778 * DTrace Action Description Functions 9779 */ 9780 static dtrace_actdesc_t * 9781 dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple, 9782 uint64_t uarg, uint64_t arg) 9783 { 9784 dtrace_actdesc_t *act; 9785 9786 #if defined(sun) 9787 ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL && 9788 arg >= KERNELBASE) || (arg == NULL && kind == DTRACEACT_PRINTA)); 9789 #endif 9790 9791 act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP); 9792 act->dtad_kind = kind; 9793 act->dtad_ntuple = ntuple; 9794 act->dtad_uarg = uarg; 9795 act->dtad_arg = arg; 9796 act->dtad_refcnt = 1; 9797 9798 return (act); 9799 } 9800 9801 static void 9802 dtrace_actdesc_hold(dtrace_actdesc_t *act) 9803 { 9804 ASSERT(act->dtad_refcnt >= 1); 9805 act->dtad_refcnt++; 9806 } 9807 9808 static void 9809 dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate) 9810 { 9811 dtrace_actkind_t kind = act->dtad_kind; 9812 dtrace_difo_t *dp; 9813 9814 ASSERT(act->dtad_refcnt >= 1); 9815 9816 if (--act->dtad_refcnt != 0) 9817 return; 9818 9819 if ((dp = act->dtad_difo) != NULL) 9820 dtrace_difo_release(dp, vstate); 9821 9822 if (DTRACEACT_ISPRINTFLIKE(kind)) { 9823 char *str = (char *)(uintptr_t)act->dtad_arg; 9824 9825 #if defined(sun) 9826 ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) || 9827 (str == NULL && act->dtad_kind == DTRACEACT_PRINTA)); 9828 #endif 9829 9830 if (str != NULL) 9831 kmem_free(str, strlen(str) + 1); 9832 } 9833 9834 kmem_free(act, sizeof (dtrace_actdesc_t)); 9835 } 9836 9837 /* 9838 * DTrace ECB Functions 9839 */ 9840 static dtrace_ecb_t * 9841 dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe) 9842 { 9843 dtrace_ecb_t *ecb; 9844 dtrace_epid_t epid; 9845 9846 ASSERT(MUTEX_HELD(&dtrace_lock)); 9847 9848 ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP); 9849 ecb->dte_predicate = NULL; 9850 ecb->dte_probe = probe; 9851 9852 /* 9853 * The default size is the size of the default action: recording 9854 * the header. 9855 */ 9856 ecb->dte_size = ecb->dte_needed = sizeof (dtrace_rechdr_t); 9857 ecb->dte_alignment = sizeof (dtrace_epid_t); 9858 9859 epid = state->dts_epid++; 9860 9861 if (epid - 1 >= state->dts_necbs) { 9862 dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs; 9863 int necbs = state->dts_necbs << 1; 9864 9865 ASSERT(epid == state->dts_necbs + 1); 9866 9867 if (necbs == 0) { 9868 ASSERT(oecbs == NULL); 9869 necbs = 1; 9870 } 9871 9872 ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP); 9873 9874 if (oecbs != NULL) 9875 bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs)); 9876 9877 dtrace_membar_producer(); 9878 state->dts_ecbs = ecbs; 9879 9880 if (oecbs != NULL) { 9881 /* 9882 * If this state is active, we must dtrace_sync() 9883 * before we can free the old dts_ecbs array: we're 9884 * coming in hot, and there may be active ring 9885 * buffer processing (which indexes into the dts_ecbs 9886 * array) on another CPU. 9887 */ 9888 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) 9889 dtrace_sync(); 9890 9891 kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs)); 9892 } 9893 9894 dtrace_membar_producer(); 9895 state->dts_necbs = necbs; 9896 } 9897 9898 ecb->dte_state = state; 9899 9900 ASSERT(state->dts_ecbs[epid - 1] == NULL); 9901 dtrace_membar_producer(); 9902 state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb; 9903 9904 return (ecb); 9905 } 9906 9907 static void 9908 dtrace_ecb_enable(dtrace_ecb_t *ecb) 9909 { 9910 dtrace_probe_t *probe = ecb->dte_probe; 9911 9912 ASSERT(MUTEX_HELD(&cpu_lock)); 9913 ASSERT(MUTEX_HELD(&dtrace_lock)); 9914 ASSERT(ecb->dte_next == NULL); 9915 9916 if (probe == NULL) { 9917 /* 9918 * This is the NULL probe -- there's nothing to do. 9919 */ 9920 return; 9921 } 9922 9923 if (probe->dtpr_ecb == NULL) { 9924 dtrace_provider_t *prov = probe->dtpr_provider; 9925 9926 /* 9927 * We're the first ECB on this probe. 9928 */ 9929 probe->dtpr_ecb = probe->dtpr_ecb_last = ecb; 9930 9931 if (ecb->dte_predicate != NULL) 9932 probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid; 9933 9934 prov->dtpv_pops.dtps_enable(prov->dtpv_arg, 9935 probe->dtpr_id, probe->dtpr_arg); 9936 } else { 9937 /* 9938 * This probe is already active. Swing the last pointer to 9939 * point to the new ECB, and issue a dtrace_sync() to assure 9940 * that all CPUs have seen the change. 9941 */ 9942 ASSERT(probe->dtpr_ecb_last != NULL); 9943 probe->dtpr_ecb_last->dte_next = ecb; 9944 probe->dtpr_ecb_last = ecb; 9945 probe->dtpr_predcache = 0; 9946 9947 dtrace_sync(); 9948 } 9949 } 9950 9951 static void 9952 dtrace_ecb_resize(dtrace_ecb_t *ecb) 9953 { 9954 dtrace_action_t *act; 9955 uint32_t curneeded = UINT32_MAX; 9956 uint32_t aggbase = UINT32_MAX; 9957 9958 /* 9959 * If we record anything, we always record the dtrace_rechdr_t. (And 9960 * we always record it first.) 9961 */ 9962 ecb->dte_size = sizeof (dtrace_rechdr_t); 9963 ecb->dte_alignment = sizeof (dtrace_epid_t); 9964 9965 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 9966 dtrace_recdesc_t *rec = &act->dta_rec; 9967 ASSERT(rec->dtrd_size > 0 || rec->dtrd_alignment == 1); 9968 9969 ecb->dte_alignment = MAX(ecb->dte_alignment, 9970 rec->dtrd_alignment); 9971 9972 if (DTRACEACT_ISAGG(act->dta_kind)) { 9973 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; 9974 9975 ASSERT(rec->dtrd_size != 0); 9976 ASSERT(agg->dtag_first != NULL); 9977 ASSERT(act->dta_prev->dta_intuple); 9978 ASSERT(aggbase != UINT32_MAX); 9979 ASSERT(curneeded != UINT32_MAX); 9980 9981 agg->dtag_base = aggbase; 9982 9983 curneeded = P2ROUNDUP(curneeded, rec->dtrd_alignment); 9984 rec->dtrd_offset = curneeded; 9985 curneeded += rec->dtrd_size; 9986 ecb->dte_needed = MAX(ecb->dte_needed, curneeded); 9987 9988 aggbase = UINT32_MAX; 9989 curneeded = UINT32_MAX; 9990 } else if (act->dta_intuple) { 9991 if (curneeded == UINT32_MAX) { 9992 /* 9993 * This is the first record in a tuple. Align 9994 * curneeded to be at offset 4 in an 8-byte 9995 * aligned block. 9996 */ 9997 ASSERT(act->dta_prev == NULL || 9998 !act->dta_prev->dta_intuple); 9999 ASSERT3U(aggbase, ==, UINT32_MAX); 10000 curneeded = P2PHASEUP(ecb->dte_size, 10001 sizeof (uint64_t), sizeof (dtrace_aggid_t)); 10002 10003 aggbase = curneeded - sizeof (dtrace_aggid_t); 10004 ASSERT(IS_P2ALIGNED(aggbase, 10005 sizeof (uint64_t))); 10006 } 10007 curneeded = P2ROUNDUP(curneeded, rec->dtrd_alignment); 10008 rec->dtrd_offset = curneeded; 10009 curneeded += rec->dtrd_size; 10010 } else { 10011 /* tuples must be followed by an aggregation */ 10012 ASSERT(act->dta_prev == NULL || 10013 !act->dta_prev->dta_intuple); 10014 10015 ecb->dte_size = P2ROUNDUP(ecb->dte_size, 10016 rec->dtrd_alignment); 10017 rec->dtrd_offset = ecb->dte_size; 10018 ecb->dte_size += rec->dtrd_size; 10019 ecb->dte_needed = MAX(ecb->dte_needed, ecb->dte_size); 10020 } 10021 } 10022 10023 if ((act = ecb->dte_action) != NULL && 10024 !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) && 10025 ecb->dte_size == sizeof (dtrace_rechdr_t)) { 10026 /* 10027 * If the size is still sizeof (dtrace_rechdr_t), then all 10028 * actions store no data; set the size to 0. 10029 */ 10030 ecb->dte_size = 0; 10031 } 10032 10033 ecb->dte_size = P2ROUNDUP(ecb->dte_size, sizeof (dtrace_epid_t)); 10034 ecb->dte_needed = P2ROUNDUP(ecb->dte_needed, (sizeof (dtrace_epid_t))); 10035 ecb->dte_state->dts_needed = MAX(ecb->dte_state->dts_needed, 10036 ecb->dte_needed); 10037 } 10038 10039 static dtrace_action_t * 10040 dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) 10041 { 10042 dtrace_aggregation_t *agg; 10043 size_t size = sizeof (uint64_t); 10044 int ntuple = desc->dtad_ntuple; 10045 dtrace_action_t *act; 10046 dtrace_recdesc_t *frec; 10047 dtrace_aggid_t aggid; 10048 dtrace_state_t *state = ecb->dte_state; 10049 10050 agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP); 10051 agg->dtag_ecb = ecb; 10052 10053 ASSERT(DTRACEACT_ISAGG(desc->dtad_kind)); 10054 10055 switch (desc->dtad_kind) { 10056 case DTRACEAGG_MIN: 10057 agg->dtag_initial = INT64_MAX; 10058 agg->dtag_aggregate = dtrace_aggregate_min; 10059 break; 10060 10061 case DTRACEAGG_MAX: 10062 agg->dtag_initial = INT64_MIN; 10063 agg->dtag_aggregate = dtrace_aggregate_max; 10064 break; 10065 10066 case DTRACEAGG_COUNT: 10067 agg->dtag_aggregate = dtrace_aggregate_count; 10068 break; 10069 10070 case DTRACEAGG_QUANTIZE: 10071 agg->dtag_aggregate = dtrace_aggregate_quantize; 10072 size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) * 10073 sizeof (uint64_t); 10074 break; 10075 10076 case DTRACEAGG_LQUANTIZE: { 10077 uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg); 10078 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg); 10079 10080 agg->dtag_initial = desc->dtad_arg; 10081 agg->dtag_aggregate = dtrace_aggregate_lquantize; 10082 10083 if (step == 0 || levels == 0) 10084 goto err; 10085 10086 size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t); 10087 break; 10088 } 10089 10090 case DTRACEAGG_LLQUANTIZE: { 10091 uint16_t factor = DTRACE_LLQUANTIZE_FACTOR(desc->dtad_arg); 10092 uint16_t low = DTRACE_LLQUANTIZE_LOW(desc->dtad_arg); 10093 uint16_t high = DTRACE_LLQUANTIZE_HIGH(desc->dtad_arg); 10094 uint16_t nsteps = DTRACE_LLQUANTIZE_NSTEP(desc->dtad_arg); 10095 int64_t v; 10096 10097 agg->dtag_initial = desc->dtad_arg; 10098 agg->dtag_aggregate = dtrace_aggregate_llquantize; 10099 10100 if (factor < 2 || low >= high || nsteps < factor) 10101 goto err; 10102 10103 /* 10104 * Now check that the number of steps evenly divides a power 10105 * of the factor. (This assures both integer bucket size and 10106 * linearity within each magnitude.) 10107 */ 10108 for (v = factor; v < nsteps; v *= factor) 10109 continue; 10110 10111 if ((v % nsteps) || (nsteps % factor)) 10112 goto err; 10113 10114 size = (dtrace_aggregate_llquantize_bucket(factor, 10115 low, high, nsteps, INT64_MAX) + 2) * sizeof (uint64_t); 10116 break; 10117 } 10118 10119 case DTRACEAGG_AVG: 10120 agg->dtag_aggregate = dtrace_aggregate_avg; 10121 size = sizeof (uint64_t) * 2; 10122 break; 10123 10124 case DTRACEAGG_STDDEV: 10125 agg->dtag_aggregate = dtrace_aggregate_stddev; 10126 size = sizeof (uint64_t) * 4; 10127 break; 10128 10129 case DTRACEAGG_SUM: 10130 agg->dtag_aggregate = dtrace_aggregate_sum; 10131 break; 10132 10133 default: 10134 goto err; 10135 } 10136 10137 agg->dtag_action.dta_rec.dtrd_size = size; 10138 10139 if (ntuple == 0) 10140 goto err; 10141 10142 /* 10143 * We must make sure that we have enough actions for the n-tuple. 10144 */ 10145 for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) { 10146 if (DTRACEACT_ISAGG(act->dta_kind)) 10147 break; 10148 10149 if (--ntuple == 0) { 10150 /* 10151 * This is the action with which our n-tuple begins. 10152 */ 10153 agg->dtag_first = act; 10154 goto success; 10155 } 10156 } 10157 10158 /* 10159 * This n-tuple is short by ntuple elements. Return failure. 10160 */ 10161 ASSERT(ntuple != 0); 10162 err: 10163 kmem_free(agg, sizeof (dtrace_aggregation_t)); 10164 return (NULL); 10165 10166 success: 10167 /* 10168 * If the last action in the tuple has a size of zero, it's actually 10169 * an expression argument for the aggregating action. 10170 */ 10171 ASSERT(ecb->dte_action_last != NULL); 10172 act = ecb->dte_action_last; 10173 10174 if (act->dta_kind == DTRACEACT_DIFEXPR) { 10175 ASSERT(act->dta_difo != NULL); 10176 10177 if (act->dta_difo->dtdo_rtype.dtdt_size == 0) 10178 agg->dtag_hasarg = 1; 10179 } 10180 10181 /* 10182 * We need to allocate an id for this aggregation. 10183 */ 10184 #if defined(sun) 10185 aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1, 10186 VM_BESTFIT | VM_SLEEP); 10187 #else 10188 aggid = alloc_unr(state->dts_aggid_arena); 10189 #endif 10190 10191 if (aggid - 1 >= state->dts_naggregations) { 10192 dtrace_aggregation_t **oaggs = state->dts_aggregations; 10193 dtrace_aggregation_t **aggs; 10194 int naggs = state->dts_naggregations << 1; 10195 int onaggs = state->dts_naggregations; 10196 10197 ASSERT(aggid == state->dts_naggregations + 1); 10198 10199 if (naggs == 0) { 10200 ASSERT(oaggs == NULL); 10201 naggs = 1; 10202 } 10203 10204 aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP); 10205 10206 if (oaggs != NULL) { 10207 bcopy(oaggs, aggs, onaggs * sizeof (*aggs)); 10208 kmem_free(oaggs, onaggs * sizeof (*aggs)); 10209 } 10210 10211 state->dts_aggregations = aggs; 10212 state->dts_naggregations = naggs; 10213 } 10214 10215 ASSERT(state->dts_aggregations[aggid - 1] == NULL); 10216 state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg; 10217 10218 frec = &agg->dtag_first->dta_rec; 10219 if (frec->dtrd_alignment < sizeof (dtrace_aggid_t)) 10220 frec->dtrd_alignment = sizeof (dtrace_aggid_t); 10221 10222 for (act = agg->dtag_first; act != NULL; act = act->dta_next) { 10223 ASSERT(!act->dta_intuple); 10224 act->dta_intuple = 1; 10225 } 10226 10227 return (&agg->dtag_action); 10228 } 10229 10230 static void 10231 dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act) 10232 { 10233 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; 10234 dtrace_state_t *state = ecb->dte_state; 10235 dtrace_aggid_t aggid = agg->dtag_id; 10236 10237 ASSERT(DTRACEACT_ISAGG(act->dta_kind)); 10238 #if defined(sun) 10239 vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1); 10240 #else 10241 free_unr(state->dts_aggid_arena, aggid); 10242 #endif 10243 10244 ASSERT(state->dts_aggregations[aggid - 1] == agg); 10245 state->dts_aggregations[aggid - 1] = NULL; 10246 10247 kmem_free(agg, sizeof (dtrace_aggregation_t)); 10248 } 10249 10250 static int 10251 dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) 10252 { 10253 dtrace_action_t *action, *last; 10254 dtrace_difo_t *dp = desc->dtad_difo; 10255 uint32_t size = 0, align = sizeof (uint8_t), mask; 10256 uint16_t format = 0; 10257 dtrace_recdesc_t *rec; 10258 dtrace_state_t *state = ecb->dte_state; 10259 dtrace_optval_t *opt = state->dts_options, nframes = 0, strsize; 10260 uint64_t arg = desc->dtad_arg; 10261 10262 ASSERT(MUTEX_HELD(&dtrace_lock)); 10263 ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1); 10264 10265 if (DTRACEACT_ISAGG(desc->dtad_kind)) { 10266 /* 10267 * If this is an aggregating action, there must be neither 10268 * a speculate nor a commit on the action chain. 10269 */ 10270 dtrace_action_t *act; 10271 10272 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 10273 if (act->dta_kind == DTRACEACT_COMMIT) 10274 return (EINVAL); 10275 10276 if (act->dta_kind == DTRACEACT_SPECULATE) 10277 return (EINVAL); 10278 } 10279 10280 action = dtrace_ecb_aggregation_create(ecb, desc); 10281 10282 if (action == NULL) 10283 return (EINVAL); 10284 } else { 10285 if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) || 10286 (desc->dtad_kind == DTRACEACT_DIFEXPR && 10287 dp != NULL && dp->dtdo_destructive)) { 10288 state->dts_destructive = 1; 10289 } 10290 10291 switch (desc->dtad_kind) { 10292 case DTRACEACT_PRINTF: 10293 case DTRACEACT_PRINTA: 10294 case DTRACEACT_SYSTEM: 10295 case DTRACEACT_FREOPEN: 10296 case DTRACEACT_DIFEXPR: 10297 /* 10298 * We know that our arg is a string -- turn it into a 10299 * format. 10300 */ 10301 if (arg == 0) { 10302 ASSERT(desc->dtad_kind == DTRACEACT_PRINTA || 10303 desc->dtad_kind == DTRACEACT_DIFEXPR); 10304 format = 0; 10305 } else { 10306 ASSERT(arg != 0); 10307 #if defined(sun) 10308 ASSERT(arg > KERNELBASE); 10309 #endif 10310 format = dtrace_format_add(state, 10311 (char *)(uintptr_t)arg); 10312 } 10313 10314 /*FALLTHROUGH*/ 10315 case DTRACEACT_LIBACT: 10316 case DTRACEACT_TRACEMEM: 10317 case DTRACEACT_TRACEMEM_DYNSIZE: 10318 if (dp == NULL) 10319 return (EINVAL); 10320 10321 if ((size = dp->dtdo_rtype.dtdt_size) != 0) 10322 break; 10323 10324 if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) { 10325 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10326 return (EINVAL); 10327 10328 size = opt[DTRACEOPT_STRSIZE]; 10329 } 10330 10331 break; 10332 10333 case DTRACEACT_STACK: 10334 if ((nframes = arg) == 0) { 10335 nframes = opt[DTRACEOPT_STACKFRAMES]; 10336 ASSERT(nframes > 0); 10337 arg = nframes; 10338 } 10339 10340 size = nframes * sizeof (pc_t); 10341 break; 10342 10343 case DTRACEACT_JSTACK: 10344 if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0) 10345 strsize = opt[DTRACEOPT_JSTACKSTRSIZE]; 10346 10347 if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) 10348 nframes = opt[DTRACEOPT_JSTACKFRAMES]; 10349 10350 arg = DTRACE_USTACK_ARG(nframes, strsize); 10351 10352 /*FALLTHROUGH*/ 10353 case DTRACEACT_USTACK: 10354 if (desc->dtad_kind != DTRACEACT_JSTACK && 10355 (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) { 10356 strsize = DTRACE_USTACK_STRSIZE(arg); 10357 nframes = opt[DTRACEOPT_USTACKFRAMES]; 10358 ASSERT(nframes > 0); 10359 arg = DTRACE_USTACK_ARG(nframes, strsize); 10360 } 10361 10362 /* 10363 * Save a slot for the pid. 10364 */ 10365 size = (nframes + 1) * sizeof (uint64_t); 10366 size += DTRACE_USTACK_STRSIZE(arg); 10367 size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t))); 10368 10369 break; 10370 10371 case DTRACEACT_SYM: 10372 case DTRACEACT_MOD: 10373 if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) != 10374 sizeof (uint64_t)) || 10375 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10376 return (EINVAL); 10377 break; 10378 10379 case DTRACEACT_USYM: 10380 case DTRACEACT_UMOD: 10381 case DTRACEACT_UADDR: 10382 if (dp == NULL || 10383 (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) || 10384 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10385 return (EINVAL); 10386 10387 /* 10388 * We have a slot for the pid, plus a slot for the 10389 * argument. To keep things simple (aligned with 10390 * bitness-neutral sizing), we store each as a 64-bit 10391 * quantity. 10392 */ 10393 size = 2 * sizeof (uint64_t); 10394 break; 10395 10396 case DTRACEACT_STOP: 10397 case DTRACEACT_BREAKPOINT: 10398 case DTRACEACT_PANIC: 10399 break; 10400 10401 case DTRACEACT_CHILL: 10402 case DTRACEACT_DISCARD: 10403 case DTRACEACT_RAISE: 10404 if (dp == NULL) 10405 return (EINVAL); 10406 break; 10407 10408 case DTRACEACT_EXIT: 10409 if (dp == NULL || 10410 (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) || 10411 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10412 return (EINVAL); 10413 break; 10414 10415 case DTRACEACT_SPECULATE: 10416 if (ecb->dte_size > sizeof (dtrace_rechdr_t)) 10417 return (EINVAL); 10418 10419 if (dp == NULL) 10420 return (EINVAL); 10421 10422 state->dts_speculates = 1; 10423 break; 10424 10425 case DTRACEACT_PRINTM: 10426 size = dp->dtdo_rtype.dtdt_size; 10427 break; 10428 10429 case DTRACEACT_PRINTT: 10430 size = dp->dtdo_rtype.dtdt_size; 10431 break; 10432 10433 case DTRACEACT_COMMIT: { 10434 dtrace_action_t *act = ecb->dte_action; 10435 10436 for (; act != NULL; act = act->dta_next) { 10437 if (act->dta_kind == DTRACEACT_COMMIT) 10438 return (EINVAL); 10439 } 10440 10441 if (dp == NULL) 10442 return (EINVAL); 10443 break; 10444 } 10445 10446 default: 10447 return (EINVAL); 10448 } 10449 10450 if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) { 10451 /* 10452 * If this is a data-storing action or a speculate, 10453 * we must be sure that there isn't a commit on the 10454 * action chain. 10455 */ 10456 dtrace_action_t *act = ecb->dte_action; 10457 10458 for (; act != NULL; act = act->dta_next) { 10459 if (act->dta_kind == DTRACEACT_COMMIT) 10460 return (EINVAL); 10461 } 10462 } 10463 10464 action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP); 10465 action->dta_rec.dtrd_size = size; 10466 } 10467 10468 action->dta_refcnt = 1; 10469 rec = &action->dta_rec; 10470 size = rec->dtrd_size; 10471 10472 for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) { 10473 if (!(size & mask)) { 10474 align = mask + 1; 10475 break; 10476 } 10477 } 10478 10479 action->dta_kind = desc->dtad_kind; 10480 10481 if ((action->dta_difo = dp) != NULL) 10482 dtrace_difo_hold(dp); 10483 10484 rec->dtrd_action = action->dta_kind; 10485 rec->dtrd_arg = arg; 10486 rec->dtrd_uarg = desc->dtad_uarg; 10487 rec->dtrd_alignment = (uint16_t)align; 10488 rec->dtrd_format = format; 10489 10490 if ((last = ecb->dte_action_last) != NULL) { 10491 ASSERT(ecb->dte_action != NULL); 10492 action->dta_prev = last; 10493 last->dta_next = action; 10494 } else { 10495 ASSERT(ecb->dte_action == NULL); 10496 ecb->dte_action = action; 10497 } 10498 10499 ecb->dte_action_last = action; 10500 10501 return (0); 10502 } 10503 10504 static void 10505 dtrace_ecb_action_remove(dtrace_ecb_t *ecb) 10506 { 10507 dtrace_action_t *act = ecb->dte_action, *next; 10508 dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate; 10509 dtrace_difo_t *dp; 10510 uint16_t format; 10511 10512 if (act != NULL && act->dta_refcnt > 1) { 10513 ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1); 10514 act->dta_refcnt--; 10515 } else { 10516 for (; act != NULL; act = next) { 10517 next = act->dta_next; 10518 ASSERT(next != NULL || act == ecb->dte_action_last); 10519 ASSERT(act->dta_refcnt == 1); 10520 10521 if ((format = act->dta_rec.dtrd_format) != 0) 10522 dtrace_format_remove(ecb->dte_state, format); 10523 10524 if ((dp = act->dta_difo) != NULL) 10525 dtrace_difo_release(dp, vstate); 10526 10527 if (DTRACEACT_ISAGG(act->dta_kind)) { 10528 dtrace_ecb_aggregation_destroy(ecb, act); 10529 } else { 10530 kmem_free(act, sizeof (dtrace_action_t)); 10531 } 10532 } 10533 } 10534 10535 ecb->dte_action = NULL; 10536 ecb->dte_action_last = NULL; 10537 ecb->dte_size = 0; 10538 } 10539 10540 static void 10541 dtrace_ecb_disable(dtrace_ecb_t *ecb) 10542 { 10543 /* 10544 * We disable the ECB by removing it from its probe. 10545 */ 10546 dtrace_ecb_t *pecb, *prev = NULL; 10547 dtrace_probe_t *probe = ecb->dte_probe; 10548 10549 ASSERT(MUTEX_HELD(&dtrace_lock)); 10550 10551 if (probe == NULL) { 10552 /* 10553 * This is the NULL probe; there is nothing to disable. 10554 */ 10555 return; 10556 } 10557 10558 for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) { 10559 if (pecb == ecb) 10560 break; 10561 prev = pecb; 10562 } 10563 10564 ASSERT(pecb != NULL); 10565 10566 if (prev == NULL) { 10567 probe->dtpr_ecb = ecb->dte_next; 10568 } else { 10569 prev->dte_next = ecb->dte_next; 10570 } 10571 10572 if (ecb == probe->dtpr_ecb_last) { 10573 ASSERT(ecb->dte_next == NULL); 10574 probe->dtpr_ecb_last = prev; 10575 } 10576 10577 /* 10578 * The ECB has been disconnected from the probe; now sync to assure 10579 * that all CPUs have seen the change before returning. 10580 */ 10581 dtrace_sync(); 10582 10583 if (probe->dtpr_ecb == NULL) { 10584 /* 10585 * That was the last ECB on the probe; clear the predicate 10586 * cache ID for the probe, disable it and sync one more time 10587 * to assure that we'll never hit it again. 10588 */ 10589 dtrace_provider_t *prov = probe->dtpr_provider; 10590 10591 ASSERT(ecb->dte_next == NULL); 10592 ASSERT(probe->dtpr_ecb_last == NULL); 10593 probe->dtpr_predcache = DTRACE_CACHEIDNONE; 10594 prov->dtpv_pops.dtps_disable(prov->dtpv_arg, 10595 probe->dtpr_id, probe->dtpr_arg); 10596 dtrace_sync(); 10597 } else { 10598 /* 10599 * There is at least one ECB remaining on the probe. If there 10600 * is _exactly_ one, set the probe's predicate cache ID to be 10601 * the predicate cache ID of the remaining ECB. 10602 */ 10603 ASSERT(probe->dtpr_ecb_last != NULL); 10604 ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE); 10605 10606 if (probe->dtpr_ecb == probe->dtpr_ecb_last) { 10607 dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate; 10608 10609 ASSERT(probe->dtpr_ecb->dte_next == NULL); 10610 10611 if (p != NULL) 10612 probe->dtpr_predcache = p->dtp_cacheid; 10613 } 10614 10615 ecb->dte_next = NULL; 10616 } 10617 } 10618 10619 static void 10620 dtrace_ecb_destroy(dtrace_ecb_t *ecb) 10621 { 10622 dtrace_state_t *state = ecb->dte_state; 10623 dtrace_vstate_t *vstate = &state->dts_vstate; 10624 dtrace_predicate_t *pred; 10625 dtrace_epid_t epid = ecb->dte_epid; 10626 10627 ASSERT(MUTEX_HELD(&dtrace_lock)); 10628 ASSERT(ecb->dte_next == NULL); 10629 ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb); 10630 10631 if ((pred = ecb->dte_predicate) != NULL) 10632 dtrace_predicate_release(pred, vstate); 10633 10634 dtrace_ecb_action_remove(ecb); 10635 10636 ASSERT(state->dts_ecbs[epid - 1] == ecb); 10637 state->dts_ecbs[epid - 1] = NULL; 10638 10639 kmem_free(ecb, sizeof (dtrace_ecb_t)); 10640 } 10641 10642 static dtrace_ecb_t * 10643 dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe, 10644 dtrace_enabling_t *enab) 10645 { 10646 dtrace_ecb_t *ecb; 10647 dtrace_predicate_t *pred; 10648 dtrace_actdesc_t *act; 10649 dtrace_provider_t *prov; 10650 dtrace_ecbdesc_t *desc = enab->dten_current; 10651 10652 ASSERT(MUTEX_HELD(&dtrace_lock)); 10653 ASSERT(state != NULL); 10654 10655 ecb = dtrace_ecb_add(state, probe); 10656 ecb->dte_uarg = desc->dted_uarg; 10657 10658 if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) { 10659 dtrace_predicate_hold(pred); 10660 ecb->dte_predicate = pred; 10661 } 10662 10663 if (probe != NULL) { 10664 /* 10665 * If the provider shows more leg than the consumer is old 10666 * enough to see, we need to enable the appropriate implicit 10667 * predicate bits to prevent the ecb from activating at 10668 * revealing times. 10669 * 10670 * Providers specifying DTRACE_PRIV_USER at register time 10671 * are stating that they need the /proc-style privilege 10672 * model to be enforced, and this is what DTRACE_COND_OWNER 10673 * and DTRACE_COND_ZONEOWNER will then do at probe time. 10674 */ 10675 prov = probe->dtpr_provider; 10676 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) && 10677 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) 10678 ecb->dte_cond |= DTRACE_COND_OWNER; 10679 10680 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) && 10681 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) 10682 ecb->dte_cond |= DTRACE_COND_ZONEOWNER; 10683 10684 /* 10685 * If the provider shows us kernel innards and the user 10686 * is lacking sufficient privilege, enable the 10687 * DTRACE_COND_USERMODE implicit predicate. 10688 */ 10689 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) && 10690 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL)) 10691 ecb->dte_cond |= DTRACE_COND_USERMODE; 10692 } 10693 10694 if (dtrace_ecb_create_cache != NULL) { 10695 /* 10696 * If we have a cached ecb, we'll use its action list instead 10697 * of creating our own (saving both time and space). 10698 */ 10699 dtrace_ecb_t *cached = dtrace_ecb_create_cache; 10700 dtrace_action_t *act = cached->dte_action; 10701 10702 if (act != NULL) { 10703 ASSERT(act->dta_refcnt > 0); 10704 act->dta_refcnt++; 10705 ecb->dte_action = act; 10706 ecb->dte_action_last = cached->dte_action_last; 10707 ecb->dte_needed = cached->dte_needed; 10708 ecb->dte_size = cached->dte_size; 10709 ecb->dte_alignment = cached->dte_alignment; 10710 } 10711 10712 return (ecb); 10713 } 10714 10715 for (act = desc->dted_action; act != NULL; act = act->dtad_next) { 10716 if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) { 10717 dtrace_ecb_destroy(ecb); 10718 return (NULL); 10719 } 10720 } 10721 10722 dtrace_ecb_resize(ecb); 10723 10724 return (dtrace_ecb_create_cache = ecb); 10725 } 10726 10727 static int 10728 dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg) 10729 { 10730 dtrace_ecb_t *ecb; 10731 dtrace_enabling_t *enab = arg; 10732 dtrace_state_t *state = enab->dten_vstate->dtvs_state; 10733 10734 ASSERT(state != NULL); 10735 10736 if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) { 10737 /* 10738 * This probe was created in a generation for which this 10739 * enabling has previously created ECBs; we don't want to 10740 * enable it again, so just kick out. 10741 */ 10742 return (DTRACE_MATCH_NEXT); 10743 } 10744 10745 if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL) 10746 return (DTRACE_MATCH_DONE); 10747 10748 dtrace_ecb_enable(ecb); 10749 return (DTRACE_MATCH_NEXT); 10750 } 10751 10752 static dtrace_ecb_t * 10753 dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id) 10754 { 10755 dtrace_ecb_t *ecb; 10756 10757 ASSERT(MUTEX_HELD(&dtrace_lock)); 10758 10759 if (id == 0 || id > state->dts_necbs) 10760 return (NULL); 10761 10762 ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL); 10763 ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id); 10764 10765 return (state->dts_ecbs[id - 1]); 10766 } 10767 10768 static dtrace_aggregation_t * 10769 dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id) 10770 { 10771 dtrace_aggregation_t *agg; 10772 10773 ASSERT(MUTEX_HELD(&dtrace_lock)); 10774 10775 if (id == 0 || id > state->dts_naggregations) 10776 return (NULL); 10777 10778 ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL); 10779 ASSERT((agg = state->dts_aggregations[id - 1]) == NULL || 10780 agg->dtag_id == id); 10781 10782 return (state->dts_aggregations[id - 1]); 10783 } 10784 10785 /* 10786 * DTrace Buffer Functions 10787 * 10788 * The following functions manipulate DTrace buffers. Most of these functions 10789 * are called in the context of establishing or processing consumer state; 10790 * exceptions are explicitly noted. 10791 */ 10792 10793 /* 10794 * Note: called from cross call context. This function switches the two 10795 * buffers on a given CPU. The atomicity of this operation is assured by 10796 * disabling interrupts while the actual switch takes place; the disabling of 10797 * interrupts serializes the execution with any execution of dtrace_probe() on 10798 * the same CPU. 10799 */ 10800 static void 10801 dtrace_buffer_switch(dtrace_buffer_t *buf) 10802 { 10803 caddr_t tomax = buf->dtb_tomax; 10804 caddr_t xamot = buf->dtb_xamot; 10805 dtrace_icookie_t cookie; 10806 hrtime_t now; 10807 10808 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); 10809 ASSERT(!(buf->dtb_flags & DTRACEBUF_RING)); 10810 10811 cookie = dtrace_interrupt_disable(); 10812 now = dtrace_gethrtime(); 10813 buf->dtb_tomax = xamot; 10814 buf->dtb_xamot = tomax; 10815 buf->dtb_xamot_drops = buf->dtb_drops; 10816 buf->dtb_xamot_offset = buf->dtb_offset; 10817 buf->dtb_xamot_errors = buf->dtb_errors; 10818 buf->dtb_xamot_flags = buf->dtb_flags; 10819 buf->dtb_offset = 0; 10820 buf->dtb_drops = 0; 10821 buf->dtb_errors = 0; 10822 buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED); 10823 buf->dtb_interval = now - buf->dtb_switched; 10824 buf->dtb_switched = now; 10825 dtrace_interrupt_enable(cookie); 10826 } 10827 10828 /* 10829 * Note: called from cross call context. This function activates a buffer 10830 * on a CPU. As with dtrace_buffer_switch(), the atomicity of the operation 10831 * is guaranteed by the disabling of interrupts. 10832 */ 10833 static void 10834 dtrace_buffer_activate(dtrace_state_t *state) 10835 { 10836 dtrace_buffer_t *buf; 10837 dtrace_icookie_t cookie = dtrace_interrupt_disable(); 10838 10839 buf = &state->dts_buffer[curcpu]; 10840 10841 if (buf->dtb_tomax != NULL) { 10842 /* 10843 * We might like to assert that the buffer is marked inactive, 10844 * but this isn't necessarily true: the buffer for the CPU 10845 * that processes the BEGIN probe has its buffer activated 10846 * manually. In this case, we take the (harmless) action 10847 * re-clearing the bit INACTIVE bit. 10848 */ 10849 buf->dtb_flags &= ~DTRACEBUF_INACTIVE; 10850 } 10851 10852 dtrace_interrupt_enable(cookie); 10853 } 10854 10855 static int 10856 dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags, 10857 processorid_t cpu, int *factor) 10858 { 10859 #if defined(sun) 10860 cpu_t *cp; 10861 #endif 10862 dtrace_buffer_t *buf; 10863 int allocated = 0, desired = 0; 10864 10865 #if defined(sun) 10866 ASSERT(MUTEX_HELD(&cpu_lock)); 10867 ASSERT(MUTEX_HELD(&dtrace_lock)); 10868 10869 *factor = 1; 10870 10871 if (size > dtrace_nonroot_maxsize && 10872 !PRIV_POLICY_CHOICE(CRED(), PRIV_ALL, B_FALSE)) 10873 return (EFBIG); 10874 10875 cp = cpu_list; 10876 10877 do { 10878 if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) 10879 continue; 10880 10881 buf = &bufs[cp->cpu_id]; 10882 10883 /* 10884 * If there is already a buffer allocated for this CPU, it 10885 * is only possible that this is a DR event. In this case, 10886 */ 10887 if (buf->dtb_tomax != NULL) { 10888 ASSERT(buf->dtb_size == size); 10889 continue; 10890 } 10891 10892 ASSERT(buf->dtb_xamot == NULL); 10893 10894 if ((buf->dtb_tomax = kmem_zalloc(size, 10895 KM_NOSLEEP | KM_NORMALPRI)) == NULL) 10896 goto err; 10897 10898 buf->dtb_size = size; 10899 buf->dtb_flags = flags; 10900 buf->dtb_offset = 0; 10901 buf->dtb_drops = 0; 10902 10903 if (flags & DTRACEBUF_NOSWITCH) 10904 continue; 10905 10906 if ((buf->dtb_xamot = kmem_zalloc(size, 10907 KM_NOSLEEP | KM_NORMALPRI)) == NULL) 10908 goto err; 10909 } while ((cp = cp->cpu_next) != cpu_list); 10910 10911 return (0); 10912 10913 err: 10914 cp = cpu_list; 10915 10916 do { 10917 if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) 10918 continue; 10919 10920 buf = &bufs[cp->cpu_id]; 10921 desired += 2; 10922 10923 if (buf->dtb_xamot != NULL) { 10924 ASSERT(buf->dtb_tomax != NULL); 10925 ASSERT(buf->dtb_size == size); 10926 kmem_free(buf->dtb_xamot, size); 10927 allocated++; 10928 } 10929 10930 if (buf->dtb_tomax != NULL) { 10931 ASSERT(buf->dtb_size == size); 10932 kmem_free(buf->dtb_tomax, size); 10933 allocated++; 10934 } 10935 10936 buf->dtb_tomax = NULL; 10937 buf->dtb_xamot = NULL; 10938 buf->dtb_size = 0; 10939 } while ((cp = cp->cpu_next) != cpu_list); 10940 #else 10941 int i; 10942 10943 *factor = 1; 10944 #if defined(__amd64__) || defined(__mips__) || defined(__powerpc__) 10945 /* 10946 * FreeBSD isn't good at limiting the amount of memory we 10947 * ask to malloc, so let's place a limit here before trying 10948 * to do something that might well end in tears at bedtime. 10949 */ 10950 if (size > physmem * PAGE_SIZE / (128 * (mp_maxid + 1))) 10951 return (ENOMEM); 10952 #endif 10953 10954 ASSERT(MUTEX_HELD(&dtrace_lock)); 10955 CPU_FOREACH(i) { 10956 if (cpu != DTRACE_CPUALL && cpu != i) 10957 continue; 10958 10959 buf = &bufs[i]; 10960 10961 /* 10962 * If there is already a buffer allocated for this CPU, it 10963 * is only possible that this is a DR event. In this case, 10964 * the buffer size must match our specified size. 10965 */ 10966 if (buf->dtb_tomax != NULL) { 10967 ASSERT(buf->dtb_size == size); 10968 continue; 10969 } 10970 10971 ASSERT(buf->dtb_xamot == NULL); 10972 10973 if ((buf->dtb_tomax = kmem_zalloc(size, 10974 KM_NOSLEEP | KM_NORMALPRI)) == NULL) 10975 goto err; 10976 10977 buf->dtb_size = size; 10978 buf->dtb_flags = flags; 10979 buf->dtb_offset = 0; 10980 buf->dtb_drops = 0; 10981 10982 if (flags & DTRACEBUF_NOSWITCH) 10983 continue; 10984 10985 if ((buf->dtb_xamot = kmem_zalloc(size, 10986 KM_NOSLEEP | KM_NORMALPRI)) == NULL) 10987 goto err; 10988 } 10989 10990 return (0); 10991 10992 err: 10993 /* 10994 * Error allocating memory, so free the buffers that were 10995 * allocated before the failed allocation. 10996 */ 10997 CPU_FOREACH(i) { 10998 if (cpu != DTRACE_CPUALL && cpu != i) 10999 continue; 11000 11001 buf = &bufs[i]; 11002 desired += 2; 11003 11004 if (buf->dtb_xamot != NULL) { 11005 ASSERT(buf->dtb_tomax != NULL); 11006 ASSERT(buf->dtb_size == size); 11007 kmem_free(buf->dtb_xamot, size); 11008 allocated++; 11009 } 11010 11011 if (buf->dtb_tomax != NULL) { 11012 ASSERT(buf->dtb_size == size); 11013 kmem_free(buf->dtb_tomax, size); 11014 allocated++; 11015 } 11016 11017 buf->dtb_tomax = NULL; 11018 buf->dtb_xamot = NULL; 11019 buf->dtb_size = 0; 11020 11021 } 11022 #endif 11023 *factor = desired / (allocated > 0 ? allocated : 1); 11024 11025 return (ENOMEM); 11026 } 11027 11028 /* 11029 * Note: called from probe context. This function just increments the drop 11030 * count on a buffer. It has been made a function to allow for the 11031 * possibility of understanding the source of mysterious drop counts. (A 11032 * problem for which one may be particularly disappointed that DTrace cannot 11033 * be used to understand DTrace.) 11034 */ 11035 static void 11036 dtrace_buffer_drop(dtrace_buffer_t *buf) 11037 { 11038 buf->dtb_drops++; 11039 } 11040 11041 /* 11042 * Note: called from probe context. This function is called to reserve space 11043 * in a buffer. If mstate is non-NULL, sets the scratch base and size in the 11044 * mstate. Returns the new offset in the buffer, or a negative value if an 11045 * error has occurred. 11046 */ 11047 static intptr_t 11048 dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align, 11049 dtrace_state_t *state, dtrace_mstate_t *mstate) 11050 { 11051 intptr_t offs = buf->dtb_offset, soffs; 11052 intptr_t woffs; 11053 caddr_t tomax; 11054 size_t total; 11055 11056 if (buf->dtb_flags & DTRACEBUF_INACTIVE) 11057 return (-1); 11058 11059 if ((tomax = buf->dtb_tomax) == NULL) { 11060 dtrace_buffer_drop(buf); 11061 return (-1); 11062 } 11063 11064 if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) { 11065 while (offs & (align - 1)) { 11066 /* 11067 * Assert that our alignment is off by a number which 11068 * is itself sizeof (uint32_t) aligned. 11069 */ 11070 ASSERT(!((align - (offs & (align - 1))) & 11071 (sizeof (uint32_t) - 1))); 11072 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); 11073 offs += sizeof (uint32_t); 11074 } 11075 11076 if ((soffs = offs + needed) > buf->dtb_size) { 11077 dtrace_buffer_drop(buf); 11078 return (-1); 11079 } 11080 11081 if (mstate == NULL) 11082 return (offs); 11083 11084 mstate->dtms_scratch_base = (uintptr_t)tomax + soffs; 11085 mstate->dtms_scratch_size = buf->dtb_size - soffs; 11086 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; 11087 11088 return (offs); 11089 } 11090 11091 if (buf->dtb_flags & DTRACEBUF_FILL) { 11092 if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN && 11093 (buf->dtb_flags & DTRACEBUF_FULL)) 11094 return (-1); 11095 goto out; 11096 } 11097 11098 total = needed + (offs & (align - 1)); 11099 11100 /* 11101 * For a ring buffer, life is quite a bit more complicated. Before 11102 * we can store any padding, we need to adjust our wrapping offset. 11103 * (If we've never before wrapped or we're not about to, no adjustment 11104 * is required.) 11105 */ 11106 if ((buf->dtb_flags & DTRACEBUF_WRAPPED) || 11107 offs + total > buf->dtb_size) { 11108 woffs = buf->dtb_xamot_offset; 11109 11110 if (offs + total > buf->dtb_size) { 11111 /* 11112 * We can't fit in the end of the buffer. First, a 11113 * sanity check that we can fit in the buffer at all. 11114 */ 11115 if (total > buf->dtb_size) { 11116 dtrace_buffer_drop(buf); 11117 return (-1); 11118 } 11119 11120 /* 11121 * We're going to be storing at the top of the buffer, 11122 * so now we need to deal with the wrapped offset. We 11123 * only reset our wrapped offset to 0 if it is 11124 * currently greater than the current offset. If it 11125 * is less than the current offset, it is because a 11126 * previous allocation induced a wrap -- but the 11127 * allocation didn't subsequently take the space due 11128 * to an error or false predicate evaluation. In this 11129 * case, we'll just leave the wrapped offset alone: if 11130 * the wrapped offset hasn't been advanced far enough 11131 * for this allocation, it will be adjusted in the 11132 * lower loop. 11133 */ 11134 if (buf->dtb_flags & DTRACEBUF_WRAPPED) { 11135 if (woffs >= offs) 11136 woffs = 0; 11137 } else { 11138 woffs = 0; 11139 } 11140 11141 /* 11142 * Now we know that we're going to be storing to the 11143 * top of the buffer and that there is room for us 11144 * there. We need to clear the buffer from the current 11145 * offset to the end (there may be old gunk there). 11146 */ 11147 while (offs < buf->dtb_size) 11148 tomax[offs++] = 0; 11149 11150 /* 11151 * We need to set our offset to zero. And because we 11152 * are wrapping, we need to set the bit indicating as 11153 * much. We can also adjust our needed space back 11154 * down to the space required by the ECB -- we know 11155 * that the top of the buffer is aligned. 11156 */ 11157 offs = 0; 11158 total = needed; 11159 buf->dtb_flags |= DTRACEBUF_WRAPPED; 11160 } else { 11161 /* 11162 * There is room for us in the buffer, so we simply 11163 * need to check the wrapped offset. 11164 */ 11165 if (woffs < offs) { 11166 /* 11167 * The wrapped offset is less than the offset. 11168 * This can happen if we allocated buffer space 11169 * that induced a wrap, but then we didn't 11170 * subsequently take the space due to an error 11171 * or false predicate evaluation. This is 11172 * okay; we know that _this_ allocation isn't 11173 * going to induce a wrap. We still can't 11174 * reset the wrapped offset to be zero, 11175 * however: the space may have been trashed in 11176 * the previous failed probe attempt. But at 11177 * least the wrapped offset doesn't need to 11178 * be adjusted at all... 11179 */ 11180 goto out; 11181 } 11182 } 11183 11184 while (offs + total > woffs) { 11185 dtrace_epid_t epid = *(uint32_t *)(tomax + woffs); 11186 size_t size; 11187 11188 if (epid == DTRACE_EPIDNONE) { 11189 size = sizeof (uint32_t); 11190 } else { 11191 ASSERT3U(epid, <=, state->dts_necbs); 11192 ASSERT(state->dts_ecbs[epid - 1] != NULL); 11193 11194 size = state->dts_ecbs[epid - 1]->dte_size; 11195 } 11196 11197 ASSERT(woffs + size <= buf->dtb_size); 11198 ASSERT(size != 0); 11199 11200 if (woffs + size == buf->dtb_size) { 11201 /* 11202 * We've reached the end of the buffer; we want 11203 * to set the wrapped offset to 0 and break 11204 * out. However, if the offs is 0, then we're 11205 * in a strange edge-condition: the amount of 11206 * space that we want to reserve plus the size 11207 * of the record that we're overwriting is 11208 * greater than the size of the buffer. This 11209 * is problematic because if we reserve the 11210 * space but subsequently don't consume it (due 11211 * to a failed predicate or error) the wrapped 11212 * offset will be 0 -- yet the EPID at offset 0 11213 * will not be committed. This situation is 11214 * relatively easy to deal with: if we're in 11215 * this case, the buffer is indistinguishable 11216 * from one that hasn't wrapped; we need only 11217 * finish the job by clearing the wrapped bit, 11218 * explicitly setting the offset to be 0, and 11219 * zero'ing out the old data in the buffer. 11220 */ 11221 if (offs == 0) { 11222 buf->dtb_flags &= ~DTRACEBUF_WRAPPED; 11223 buf->dtb_offset = 0; 11224 woffs = total; 11225 11226 while (woffs < buf->dtb_size) 11227 tomax[woffs++] = 0; 11228 } 11229 11230 woffs = 0; 11231 break; 11232 } 11233 11234 woffs += size; 11235 } 11236 11237 /* 11238 * We have a wrapped offset. It may be that the wrapped offset 11239 * has become zero -- that's okay. 11240 */ 11241 buf->dtb_xamot_offset = woffs; 11242 } 11243 11244 out: 11245 /* 11246 * Now we can plow the buffer with any necessary padding. 11247 */ 11248 while (offs & (align - 1)) { 11249 /* 11250 * Assert that our alignment is off by a number which 11251 * is itself sizeof (uint32_t) aligned. 11252 */ 11253 ASSERT(!((align - (offs & (align - 1))) & 11254 (sizeof (uint32_t) - 1))); 11255 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); 11256 offs += sizeof (uint32_t); 11257 } 11258 11259 if (buf->dtb_flags & DTRACEBUF_FILL) { 11260 if (offs + needed > buf->dtb_size - state->dts_reserve) { 11261 buf->dtb_flags |= DTRACEBUF_FULL; 11262 return (-1); 11263 } 11264 } 11265 11266 if (mstate == NULL) 11267 return (offs); 11268 11269 /* 11270 * For ring buffers and fill buffers, the scratch space is always 11271 * the inactive buffer. 11272 */ 11273 mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot; 11274 mstate->dtms_scratch_size = buf->dtb_size; 11275 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; 11276 11277 return (offs); 11278 } 11279 11280 static void 11281 dtrace_buffer_polish(dtrace_buffer_t *buf) 11282 { 11283 ASSERT(buf->dtb_flags & DTRACEBUF_RING); 11284 ASSERT(MUTEX_HELD(&dtrace_lock)); 11285 11286 if (!(buf->dtb_flags & DTRACEBUF_WRAPPED)) 11287 return; 11288 11289 /* 11290 * We need to polish the ring buffer. There are three cases: 11291 * 11292 * - The first (and presumably most common) is that there is no gap 11293 * between the buffer offset and the wrapped offset. In this case, 11294 * there is nothing in the buffer that isn't valid data; we can 11295 * mark the buffer as polished and return. 11296 * 11297 * - The second (less common than the first but still more common 11298 * than the third) is that there is a gap between the buffer offset 11299 * and the wrapped offset, and the wrapped offset is larger than the 11300 * buffer offset. This can happen because of an alignment issue, or 11301 * can happen because of a call to dtrace_buffer_reserve() that 11302 * didn't subsequently consume the buffer space. In this case, 11303 * we need to zero the data from the buffer offset to the wrapped 11304 * offset. 11305 * 11306 * - The third (and least common) is that there is a gap between the 11307 * buffer offset and the wrapped offset, but the wrapped offset is 11308 * _less_ than the buffer offset. This can only happen because a 11309 * call to dtrace_buffer_reserve() induced a wrap, but the space 11310 * was not subsequently consumed. In this case, we need to zero the 11311 * space from the offset to the end of the buffer _and_ from the 11312 * top of the buffer to the wrapped offset. 11313 */ 11314 if (buf->dtb_offset < buf->dtb_xamot_offset) { 11315 bzero(buf->dtb_tomax + buf->dtb_offset, 11316 buf->dtb_xamot_offset - buf->dtb_offset); 11317 } 11318 11319 if (buf->dtb_offset > buf->dtb_xamot_offset) { 11320 bzero(buf->dtb_tomax + buf->dtb_offset, 11321 buf->dtb_size - buf->dtb_offset); 11322 bzero(buf->dtb_tomax, buf->dtb_xamot_offset); 11323 } 11324 } 11325 11326 /* 11327 * This routine determines if data generated at the specified time has likely 11328 * been entirely consumed at user-level. This routine is called to determine 11329 * if an ECB on a defunct probe (but for an active enabling) can be safely 11330 * disabled and destroyed. 11331 */ 11332 static int 11333 dtrace_buffer_consumed(dtrace_buffer_t *bufs, hrtime_t when) 11334 { 11335 int i; 11336 11337 for (i = 0; i < NCPU; i++) { 11338 dtrace_buffer_t *buf = &bufs[i]; 11339 11340 if (buf->dtb_size == 0) 11341 continue; 11342 11343 if (buf->dtb_flags & DTRACEBUF_RING) 11344 return (0); 11345 11346 if (!buf->dtb_switched && buf->dtb_offset != 0) 11347 return (0); 11348 11349 if (buf->dtb_switched - buf->dtb_interval < when) 11350 return (0); 11351 } 11352 11353 return (1); 11354 } 11355 11356 static void 11357 dtrace_buffer_free(dtrace_buffer_t *bufs) 11358 { 11359 int i; 11360 11361 for (i = 0; i < NCPU; i++) { 11362 dtrace_buffer_t *buf = &bufs[i]; 11363 11364 if (buf->dtb_tomax == NULL) { 11365 ASSERT(buf->dtb_xamot == NULL); 11366 ASSERT(buf->dtb_size == 0); 11367 continue; 11368 } 11369 11370 if (buf->dtb_xamot != NULL) { 11371 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); 11372 kmem_free(buf->dtb_xamot, buf->dtb_size); 11373 } 11374 11375 kmem_free(buf->dtb_tomax, buf->dtb_size); 11376 buf->dtb_size = 0; 11377 buf->dtb_tomax = NULL; 11378 buf->dtb_xamot = NULL; 11379 } 11380 } 11381 11382 /* 11383 * DTrace Enabling Functions 11384 */ 11385 static dtrace_enabling_t * 11386 dtrace_enabling_create(dtrace_vstate_t *vstate) 11387 { 11388 dtrace_enabling_t *enab; 11389 11390 enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP); 11391 enab->dten_vstate = vstate; 11392 11393 return (enab); 11394 } 11395 11396 static void 11397 dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb) 11398 { 11399 dtrace_ecbdesc_t **ndesc; 11400 size_t osize, nsize; 11401 11402 /* 11403 * We can't add to enablings after we've enabled them, or after we've 11404 * retained them. 11405 */ 11406 ASSERT(enab->dten_probegen == 0); 11407 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); 11408 11409 if (enab->dten_ndesc < enab->dten_maxdesc) { 11410 enab->dten_desc[enab->dten_ndesc++] = ecb; 11411 return; 11412 } 11413 11414 osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); 11415 11416 if (enab->dten_maxdesc == 0) { 11417 enab->dten_maxdesc = 1; 11418 } else { 11419 enab->dten_maxdesc <<= 1; 11420 } 11421 11422 ASSERT(enab->dten_ndesc < enab->dten_maxdesc); 11423 11424 nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); 11425 ndesc = kmem_zalloc(nsize, KM_SLEEP); 11426 bcopy(enab->dten_desc, ndesc, osize); 11427 if (enab->dten_desc != NULL) 11428 kmem_free(enab->dten_desc, osize); 11429 11430 enab->dten_desc = ndesc; 11431 enab->dten_desc[enab->dten_ndesc++] = ecb; 11432 } 11433 11434 static void 11435 dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb, 11436 dtrace_probedesc_t *pd) 11437 { 11438 dtrace_ecbdesc_t *new; 11439 dtrace_predicate_t *pred; 11440 dtrace_actdesc_t *act; 11441 11442 /* 11443 * We're going to create a new ECB description that matches the 11444 * specified ECB in every way, but has the specified probe description. 11445 */ 11446 new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); 11447 11448 if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL) 11449 dtrace_predicate_hold(pred); 11450 11451 for (act = ecb->dted_action; act != NULL; act = act->dtad_next) 11452 dtrace_actdesc_hold(act); 11453 11454 new->dted_action = ecb->dted_action; 11455 new->dted_pred = ecb->dted_pred; 11456 new->dted_probe = *pd; 11457 new->dted_uarg = ecb->dted_uarg; 11458 11459 dtrace_enabling_add(enab, new); 11460 } 11461 11462 static void 11463 dtrace_enabling_dump(dtrace_enabling_t *enab) 11464 { 11465 int i; 11466 11467 for (i = 0; i < enab->dten_ndesc; i++) { 11468 dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe; 11469 11470 cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i, 11471 desc->dtpd_provider, desc->dtpd_mod, 11472 desc->dtpd_func, desc->dtpd_name); 11473 } 11474 } 11475 11476 static void 11477 dtrace_enabling_destroy(dtrace_enabling_t *enab) 11478 { 11479 int i; 11480 dtrace_ecbdesc_t *ep; 11481 dtrace_vstate_t *vstate = enab->dten_vstate; 11482 11483 ASSERT(MUTEX_HELD(&dtrace_lock)); 11484 11485 for (i = 0; i < enab->dten_ndesc; i++) { 11486 dtrace_actdesc_t *act, *next; 11487 dtrace_predicate_t *pred; 11488 11489 ep = enab->dten_desc[i]; 11490 11491 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) 11492 dtrace_predicate_release(pred, vstate); 11493 11494 for (act = ep->dted_action; act != NULL; act = next) { 11495 next = act->dtad_next; 11496 dtrace_actdesc_release(act, vstate); 11497 } 11498 11499 kmem_free(ep, sizeof (dtrace_ecbdesc_t)); 11500 } 11501 11502 if (enab->dten_desc != NULL) 11503 kmem_free(enab->dten_desc, 11504 enab->dten_maxdesc * sizeof (dtrace_enabling_t *)); 11505 11506 /* 11507 * If this was a retained enabling, decrement the dts_nretained count 11508 * and take it off of the dtrace_retained list. 11509 */ 11510 if (enab->dten_prev != NULL || enab->dten_next != NULL || 11511 dtrace_retained == enab) { 11512 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11513 ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0); 11514 enab->dten_vstate->dtvs_state->dts_nretained--; 11515 } 11516 11517 if (enab->dten_prev == NULL) { 11518 if (dtrace_retained == enab) { 11519 dtrace_retained = enab->dten_next; 11520 11521 if (dtrace_retained != NULL) 11522 dtrace_retained->dten_prev = NULL; 11523 } 11524 } else { 11525 ASSERT(enab != dtrace_retained); 11526 ASSERT(dtrace_retained != NULL); 11527 enab->dten_prev->dten_next = enab->dten_next; 11528 } 11529 11530 if (enab->dten_next != NULL) { 11531 ASSERT(dtrace_retained != NULL); 11532 enab->dten_next->dten_prev = enab->dten_prev; 11533 } 11534 11535 kmem_free(enab, sizeof (dtrace_enabling_t)); 11536 } 11537 11538 static int 11539 dtrace_enabling_retain(dtrace_enabling_t *enab) 11540 { 11541 dtrace_state_t *state; 11542 11543 ASSERT(MUTEX_HELD(&dtrace_lock)); 11544 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); 11545 ASSERT(enab->dten_vstate != NULL); 11546 11547 state = enab->dten_vstate->dtvs_state; 11548 ASSERT(state != NULL); 11549 11550 /* 11551 * We only allow each state to retain dtrace_retain_max enablings. 11552 */ 11553 if (state->dts_nretained >= dtrace_retain_max) 11554 return (ENOSPC); 11555 11556 state->dts_nretained++; 11557 11558 if (dtrace_retained == NULL) { 11559 dtrace_retained = enab; 11560 return (0); 11561 } 11562 11563 enab->dten_next = dtrace_retained; 11564 dtrace_retained->dten_prev = enab; 11565 dtrace_retained = enab; 11566 11567 return (0); 11568 } 11569 11570 static int 11571 dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match, 11572 dtrace_probedesc_t *create) 11573 { 11574 dtrace_enabling_t *new, *enab; 11575 int found = 0, err = ENOENT; 11576 11577 ASSERT(MUTEX_HELD(&dtrace_lock)); 11578 ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN); 11579 ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN); 11580 ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN); 11581 ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN); 11582 11583 new = dtrace_enabling_create(&state->dts_vstate); 11584 11585 /* 11586 * Iterate over all retained enablings, looking for enablings that 11587 * match the specified state. 11588 */ 11589 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { 11590 int i; 11591 11592 /* 11593 * dtvs_state can only be NULL for helper enablings -- and 11594 * helper enablings can't be retained. 11595 */ 11596 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11597 11598 if (enab->dten_vstate->dtvs_state != state) 11599 continue; 11600 11601 /* 11602 * Now iterate over each probe description; we're looking for 11603 * an exact match to the specified probe description. 11604 */ 11605 for (i = 0; i < enab->dten_ndesc; i++) { 11606 dtrace_ecbdesc_t *ep = enab->dten_desc[i]; 11607 dtrace_probedesc_t *pd = &ep->dted_probe; 11608 11609 if (strcmp(pd->dtpd_provider, match->dtpd_provider)) 11610 continue; 11611 11612 if (strcmp(pd->dtpd_mod, match->dtpd_mod)) 11613 continue; 11614 11615 if (strcmp(pd->dtpd_func, match->dtpd_func)) 11616 continue; 11617 11618 if (strcmp(pd->dtpd_name, match->dtpd_name)) 11619 continue; 11620 11621 /* 11622 * We have a winning probe! Add it to our growing 11623 * enabling. 11624 */ 11625 found = 1; 11626 dtrace_enabling_addlike(new, ep, create); 11627 } 11628 } 11629 11630 if (!found || (err = dtrace_enabling_retain(new)) != 0) { 11631 dtrace_enabling_destroy(new); 11632 return (err); 11633 } 11634 11635 return (0); 11636 } 11637 11638 static void 11639 dtrace_enabling_retract(dtrace_state_t *state) 11640 { 11641 dtrace_enabling_t *enab, *next; 11642 11643 ASSERT(MUTEX_HELD(&dtrace_lock)); 11644 11645 /* 11646 * Iterate over all retained enablings, destroy the enablings retained 11647 * for the specified state. 11648 */ 11649 for (enab = dtrace_retained; enab != NULL; enab = next) { 11650 next = enab->dten_next; 11651 11652 /* 11653 * dtvs_state can only be NULL for helper enablings -- and 11654 * helper enablings can't be retained. 11655 */ 11656 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11657 11658 if (enab->dten_vstate->dtvs_state == state) { 11659 ASSERT(state->dts_nretained > 0); 11660 dtrace_enabling_destroy(enab); 11661 } 11662 } 11663 11664 ASSERT(state->dts_nretained == 0); 11665 } 11666 11667 static int 11668 dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched) 11669 { 11670 int i = 0; 11671 int matched = 0; 11672 11673 ASSERT(MUTEX_HELD(&cpu_lock)); 11674 ASSERT(MUTEX_HELD(&dtrace_lock)); 11675 11676 for (i = 0; i < enab->dten_ndesc; i++) { 11677 dtrace_ecbdesc_t *ep = enab->dten_desc[i]; 11678 11679 enab->dten_current = ep; 11680 enab->dten_error = 0; 11681 11682 matched += dtrace_probe_enable(&ep->dted_probe, enab); 11683 11684 if (enab->dten_error != 0) { 11685 /* 11686 * If we get an error half-way through enabling the 11687 * probes, we kick out -- perhaps with some number of 11688 * them enabled. Leaving enabled probes enabled may 11689 * be slightly confusing for user-level, but we expect 11690 * that no one will attempt to actually drive on in 11691 * the face of such errors. If this is an anonymous 11692 * enabling (indicated with a NULL nmatched pointer), 11693 * we cmn_err() a message. We aren't expecting to 11694 * get such an error -- such as it can exist at all, 11695 * it would be a result of corrupted DOF in the driver 11696 * properties. 11697 */ 11698 if (nmatched == NULL) { 11699 cmn_err(CE_WARN, "dtrace_enabling_match() " 11700 "error on %p: %d", (void *)ep, 11701 enab->dten_error); 11702 } 11703 11704 return (enab->dten_error); 11705 } 11706 } 11707 11708 enab->dten_probegen = dtrace_probegen; 11709 if (nmatched != NULL) 11710 *nmatched = matched; 11711 11712 return (0); 11713 } 11714 11715 static void 11716 dtrace_enabling_matchall(void) 11717 { 11718 dtrace_enabling_t *enab; 11719 11720 mutex_enter(&cpu_lock); 11721 mutex_enter(&dtrace_lock); 11722 11723 /* 11724 * Iterate over all retained enablings to see if any probes match 11725 * against them. We only perform this operation on enablings for which 11726 * we have sufficient permissions by virtue of being in the global zone 11727 * or in the same zone as the DTrace client. Because we can be called 11728 * after dtrace_detach() has been called, we cannot assert that there 11729 * are retained enablings. We can safely load from dtrace_retained, 11730 * however: the taskq_destroy() at the end of dtrace_detach() will 11731 * block pending our completion. 11732 */ 11733 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { 11734 #if defined(sun) 11735 cred_t *cr = enab->dten_vstate->dtvs_state->dts_cred.dcr_cred; 11736 11737 if (INGLOBALZONE(curproc) || getzoneid() == crgetzoneid(cr)) 11738 #endif 11739 (void) dtrace_enabling_match(enab, NULL); 11740 } 11741 11742 mutex_exit(&dtrace_lock); 11743 mutex_exit(&cpu_lock); 11744 } 11745 11746 /* 11747 * If an enabling is to be enabled without having matched probes (that is, if 11748 * dtrace_state_go() is to be called on the underlying dtrace_state_t), the 11749 * enabling must be _primed_ by creating an ECB for every ECB description. 11750 * This must be done to assure that we know the number of speculations, the 11751 * number of aggregations, the minimum buffer size needed, etc. before we 11752 * transition out of DTRACE_ACTIVITY_INACTIVE. To do this without actually 11753 * enabling any probes, we create ECBs for every ECB decription, but with a 11754 * NULL probe -- which is exactly what this function does. 11755 */ 11756 static void 11757 dtrace_enabling_prime(dtrace_state_t *state) 11758 { 11759 dtrace_enabling_t *enab; 11760 int i; 11761 11762 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { 11763 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11764 11765 if (enab->dten_vstate->dtvs_state != state) 11766 continue; 11767 11768 /* 11769 * We don't want to prime an enabling more than once, lest 11770 * we allow a malicious user to induce resource exhaustion. 11771 * (The ECBs that result from priming an enabling aren't 11772 * leaked -- but they also aren't deallocated until the 11773 * consumer state is destroyed.) 11774 */ 11775 if (enab->dten_primed) 11776 continue; 11777 11778 for (i = 0; i < enab->dten_ndesc; i++) { 11779 enab->dten_current = enab->dten_desc[i]; 11780 (void) dtrace_probe_enable(NULL, enab); 11781 } 11782 11783 enab->dten_primed = 1; 11784 } 11785 } 11786 11787 /* 11788 * Called to indicate that probes should be provided due to retained 11789 * enablings. This is implemented in terms of dtrace_probe_provide(), but it 11790 * must take an initial lap through the enabling calling the dtps_provide() 11791 * entry point explicitly to allow for autocreated probes. 11792 */ 11793 static void 11794 dtrace_enabling_provide(dtrace_provider_t *prv) 11795 { 11796 int i, all = 0; 11797 dtrace_probedesc_t desc; 11798 11799 ASSERT(MUTEX_HELD(&dtrace_lock)); 11800 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 11801 11802 if (prv == NULL) { 11803 all = 1; 11804 prv = dtrace_provider; 11805 } 11806 11807 do { 11808 dtrace_enabling_t *enab = dtrace_retained; 11809 void *parg = prv->dtpv_arg; 11810 11811 for (; enab != NULL; enab = enab->dten_next) { 11812 for (i = 0; i < enab->dten_ndesc; i++) { 11813 desc = enab->dten_desc[i]->dted_probe; 11814 mutex_exit(&dtrace_lock); 11815 prv->dtpv_pops.dtps_provide(parg, &desc); 11816 mutex_enter(&dtrace_lock); 11817 } 11818 } 11819 } while (all && (prv = prv->dtpv_next) != NULL); 11820 11821 mutex_exit(&dtrace_lock); 11822 dtrace_probe_provide(NULL, all ? NULL : prv); 11823 mutex_enter(&dtrace_lock); 11824 } 11825 11826 /* 11827 * Called to reap ECBs that are attached to probes from defunct providers. 11828 */ 11829 static void 11830 dtrace_enabling_reap(void) 11831 { 11832 dtrace_provider_t *prov; 11833 dtrace_probe_t *probe; 11834 dtrace_ecb_t *ecb; 11835 hrtime_t when; 11836 int i; 11837 11838 mutex_enter(&cpu_lock); 11839 mutex_enter(&dtrace_lock); 11840 11841 for (i = 0; i < dtrace_nprobes; i++) { 11842 if ((probe = dtrace_probes[i]) == NULL) 11843 continue; 11844 11845 if (probe->dtpr_ecb == NULL) 11846 continue; 11847 11848 prov = probe->dtpr_provider; 11849 11850 if ((when = prov->dtpv_defunct) == 0) 11851 continue; 11852 11853 /* 11854 * We have ECBs on a defunct provider: we want to reap these 11855 * ECBs to allow the provider to unregister. The destruction 11856 * of these ECBs must be done carefully: if we destroy the ECB 11857 * and the consumer later wishes to consume an EPID that 11858 * corresponds to the destroyed ECB (and if the EPID metadata 11859 * has not been previously consumed), the consumer will abort 11860 * processing on the unknown EPID. To reduce (but not, sadly, 11861 * eliminate) the possibility of this, we will only destroy an 11862 * ECB for a defunct provider if, for the state that 11863 * corresponds to the ECB: 11864 * 11865 * (a) There is no speculative tracing (which can effectively 11866 * cache an EPID for an arbitrary amount of time). 11867 * 11868 * (b) The principal buffers have been switched twice since the 11869 * provider became defunct. 11870 * 11871 * (c) The aggregation buffers are of zero size or have been 11872 * switched twice since the provider became defunct. 11873 * 11874 * We use dts_speculates to determine (a) and call a function 11875 * (dtrace_buffer_consumed()) to determine (b) and (c). Note 11876 * that as soon as we've been unable to destroy one of the ECBs 11877 * associated with the probe, we quit trying -- reaping is only 11878 * fruitful in as much as we can destroy all ECBs associated 11879 * with the defunct provider's probes. 11880 */ 11881 while ((ecb = probe->dtpr_ecb) != NULL) { 11882 dtrace_state_t *state = ecb->dte_state; 11883 dtrace_buffer_t *buf = state->dts_buffer; 11884 dtrace_buffer_t *aggbuf = state->dts_aggbuffer; 11885 11886 if (state->dts_speculates) 11887 break; 11888 11889 if (!dtrace_buffer_consumed(buf, when)) 11890 break; 11891 11892 if (!dtrace_buffer_consumed(aggbuf, when)) 11893 break; 11894 11895 dtrace_ecb_disable(ecb); 11896 ASSERT(probe->dtpr_ecb != ecb); 11897 dtrace_ecb_destroy(ecb); 11898 } 11899 } 11900 11901 mutex_exit(&dtrace_lock); 11902 mutex_exit(&cpu_lock); 11903 } 11904 11905 /* 11906 * DTrace DOF Functions 11907 */ 11908 /*ARGSUSED*/ 11909 static void 11910 dtrace_dof_error(dof_hdr_t *dof, const char *str) 11911 { 11912 if (dtrace_err_verbose) 11913 cmn_err(CE_WARN, "failed to process DOF: %s", str); 11914 11915 #ifdef DTRACE_ERRDEBUG 11916 dtrace_errdebug(str); 11917 #endif 11918 } 11919 11920 /* 11921 * Create DOF out of a currently enabled state. Right now, we only create 11922 * DOF containing the run-time options -- but this could be expanded to create 11923 * complete DOF representing the enabled state. 11924 */ 11925 static dof_hdr_t * 11926 dtrace_dof_create(dtrace_state_t *state) 11927 { 11928 dof_hdr_t *dof; 11929 dof_sec_t *sec; 11930 dof_optdesc_t *opt; 11931 int i, len = sizeof (dof_hdr_t) + 11932 roundup(sizeof (dof_sec_t), sizeof (uint64_t)) + 11933 sizeof (dof_optdesc_t) * DTRACEOPT_MAX; 11934 11935 ASSERT(MUTEX_HELD(&dtrace_lock)); 11936 11937 dof = kmem_zalloc(len, KM_SLEEP); 11938 dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0; 11939 dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1; 11940 dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2; 11941 dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3; 11942 11943 dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE; 11944 dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE; 11945 dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION; 11946 dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION; 11947 dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS; 11948 dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS; 11949 11950 dof->dofh_flags = 0; 11951 dof->dofh_hdrsize = sizeof (dof_hdr_t); 11952 dof->dofh_secsize = sizeof (dof_sec_t); 11953 dof->dofh_secnum = 1; /* only DOF_SECT_OPTDESC */ 11954 dof->dofh_secoff = sizeof (dof_hdr_t); 11955 dof->dofh_loadsz = len; 11956 dof->dofh_filesz = len; 11957 dof->dofh_pad = 0; 11958 11959 /* 11960 * Fill in the option section header... 11961 */ 11962 sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t)); 11963 sec->dofs_type = DOF_SECT_OPTDESC; 11964 sec->dofs_align = sizeof (uint64_t); 11965 sec->dofs_flags = DOF_SECF_LOAD; 11966 sec->dofs_entsize = sizeof (dof_optdesc_t); 11967 11968 opt = (dof_optdesc_t *)((uintptr_t)sec + 11969 roundup(sizeof (dof_sec_t), sizeof (uint64_t))); 11970 11971 sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof; 11972 sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX; 11973 11974 for (i = 0; i < DTRACEOPT_MAX; i++) { 11975 opt[i].dofo_option = i; 11976 opt[i].dofo_strtab = DOF_SECIDX_NONE; 11977 opt[i].dofo_value = state->dts_options[i]; 11978 } 11979 11980 return (dof); 11981 } 11982 11983 static dof_hdr_t * 11984 dtrace_dof_copyin(uintptr_t uarg, int *errp) 11985 { 11986 dof_hdr_t hdr, *dof; 11987 11988 ASSERT(!MUTEX_HELD(&dtrace_lock)); 11989 11990 /* 11991 * First, we're going to copyin() the sizeof (dof_hdr_t). 11992 */ 11993 if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) { 11994 dtrace_dof_error(NULL, "failed to copyin DOF header"); 11995 *errp = EFAULT; 11996 return (NULL); 11997 } 11998 11999 /* 12000 * Now we'll allocate the entire DOF and copy it in -- provided 12001 * that the length isn't outrageous. 12002 */ 12003 if (hdr.dofh_loadsz >= dtrace_dof_maxsize) { 12004 dtrace_dof_error(&hdr, "load size exceeds maximum"); 12005 *errp = E2BIG; 12006 return (NULL); 12007 } 12008 12009 if (hdr.dofh_loadsz < sizeof (hdr)) { 12010 dtrace_dof_error(&hdr, "invalid load size"); 12011 *errp = EINVAL; 12012 return (NULL); 12013 } 12014 12015 dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP); 12016 12017 if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0) { 12018 kmem_free(dof, hdr.dofh_loadsz); 12019 *errp = EFAULT; 12020 return (NULL); 12021 } 12022 12023 return (dof); 12024 } 12025 12026 #if !defined(sun) 12027 static __inline uchar_t 12028 dtrace_dof_char(char c) { 12029 switch (c) { 12030 case '0': 12031 case '1': 12032 case '2': 12033 case '3': 12034 case '4': 12035 case '5': 12036 case '6': 12037 case '7': 12038 case '8': 12039 case '9': 12040 return (c - '0'); 12041 case 'A': 12042 case 'B': 12043 case 'C': 12044 case 'D': 12045 case 'E': 12046 case 'F': 12047 return (c - 'A' + 10); 12048 case 'a': 12049 case 'b': 12050 case 'c': 12051 case 'd': 12052 case 'e': 12053 case 'f': 12054 return (c - 'a' + 10); 12055 } 12056 /* Should not reach here. */ 12057 return (0); 12058 } 12059 #endif 12060 12061 static dof_hdr_t * 12062 dtrace_dof_property(const char *name) 12063 { 12064 uchar_t *buf; 12065 uint64_t loadsz; 12066 unsigned int len, i; 12067 dof_hdr_t *dof; 12068 12069 #if defined(sun) 12070 /* 12071 * Unfortunately, array of values in .conf files are always (and 12072 * only) interpreted to be integer arrays. We must read our DOF 12073 * as an integer array, and then squeeze it into a byte array. 12074 */ 12075 if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0, 12076 (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS) 12077 return (NULL); 12078 12079 for (i = 0; i < len; i++) 12080 buf[i] = (uchar_t)(((int *)buf)[i]); 12081 12082 if (len < sizeof (dof_hdr_t)) { 12083 ddi_prop_free(buf); 12084 dtrace_dof_error(NULL, "truncated header"); 12085 return (NULL); 12086 } 12087 12088 if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) { 12089 ddi_prop_free(buf); 12090 dtrace_dof_error(NULL, "truncated DOF"); 12091 return (NULL); 12092 } 12093 12094 if (loadsz >= dtrace_dof_maxsize) { 12095 ddi_prop_free(buf); 12096 dtrace_dof_error(NULL, "oversized DOF"); 12097 return (NULL); 12098 } 12099 12100 dof = kmem_alloc(loadsz, KM_SLEEP); 12101 bcopy(buf, dof, loadsz); 12102 ddi_prop_free(buf); 12103 #else 12104 char *p; 12105 char *p_env; 12106 12107 if ((p_env = getenv(name)) == NULL) 12108 return (NULL); 12109 12110 len = strlen(p_env) / 2; 12111 12112 buf = kmem_alloc(len, KM_SLEEP); 12113 12114 dof = (dof_hdr_t *) buf; 12115 12116 p = p_env; 12117 12118 for (i = 0; i < len; i++) { 12119 buf[i] = (dtrace_dof_char(p[0]) << 4) | 12120 dtrace_dof_char(p[1]); 12121 p += 2; 12122 } 12123 12124 freeenv(p_env); 12125 12126 if (len < sizeof (dof_hdr_t)) { 12127 kmem_free(buf, 0); 12128 dtrace_dof_error(NULL, "truncated header"); 12129 return (NULL); 12130 } 12131 12132 if (len < (loadsz = dof->dofh_loadsz)) { 12133 kmem_free(buf, 0); 12134 dtrace_dof_error(NULL, "truncated DOF"); 12135 return (NULL); 12136 } 12137 12138 if (loadsz >= dtrace_dof_maxsize) { 12139 kmem_free(buf, 0); 12140 dtrace_dof_error(NULL, "oversized DOF"); 12141 return (NULL); 12142 } 12143 #endif 12144 12145 return (dof); 12146 } 12147 12148 static void 12149 dtrace_dof_destroy(dof_hdr_t *dof) 12150 { 12151 kmem_free(dof, dof->dofh_loadsz); 12152 } 12153 12154 /* 12155 * Return the dof_sec_t pointer corresponding to a given section index. If the 12156 * index is not valid, dtrace_dof_error() is called and NULL is returned. If 12157 * a type other than DOF_SECT_NONE is specified, the header is checked against 12158 * this type and NULL is returned if the types do not match. 12159 */ 12160 static dof_sec_t * 12161 dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i) 12162 { 12163 dof_sec_t *sec = (dof_sec_t *)(uintptr_t) 12164 ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize); 12165 12166 if (i >= dof->dofh_secnum) { 12167 dtrace_dof_error(dof, "referenced section index is invalid"); 12168 return (NULL); 12169 } 12170 12171 if (!(sec->dofs_flags & DOF_SECF_LOAD)) { 12172 dtrace_dof_error(dof, "referenced section is not loadable"); 12173 return (NULL); 12174 } 12175 12176 if (type != DOF_SECT_NONE && type != sec->dofs_type) { 12177 dtrace_dof_error(dof, "referenced section is the wrong type"); 12178 return (NULL); 12179 } 12180 12181 return (sec); 12182 } 12183 12184 static dtrace_probedesc_t * 12185 dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc) 12186 { 12187 dof_probedesc_t *probe; 12188 dof_sec_t *strtab; 12189 uintptr_t daddr = (uintptr_t)dof; 12190 uintptr_t str; 12191 size_t size; 12192 12193 if (sec->dofs_type != DOF_SECT_PROBEDESC) { 12194 dtrace_dof_error(dof, "invalid probe section"); 12195 return (NULL); 12196 } 12197 12198 if (sec->dofs_align != sizeof (dof_secidx_t)) { 12199 dtrace_dof_error(dof, "bad alignment in probe description"); 12200 return (NULL); 12201 } 12202 12203 if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) { 12204 dtrace_dof_error(dof, "truncated probe description"); 12205 return (NULL); 12206 } 12207 12208 probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset); 12209 strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab); 12210 12211 if (strtab == NULL) 12212 return (NULL); 12213 12214 str = daddr + strtab->dofs_offset; 12215 size = strtab->dofs_size; 12216 12217 if (probe->dofp_provider >= strtab->dofs_size) { 12218 dtrace_dof_error(dof, "corrupt probe provider"); 12219 return (NULL); 12220 } 12221 12222 (void) strncpy(desc->dtpd_provider, 12223 (char *)(str + probe->dofp_provider), 12224 MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider)); 12225 12226 if (probe->dofp_mod >= strtab->dofs_size) { 12227 dtrace_dof_error(dof, "corrupt probe module"); 12228 return (NULL); 12229 } 12230 12231 (void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod), 12232 MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod)); 12233 12234 if (probe->dofp_func >= strtab->dofs_size) { 12235 dtrace_dof_error(dof, "corrupt probe function"); 12236 return (NULL); 12237 } 12238 12239 (void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func), 12240 MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func)); 12241 12242 if (probe->dofp_name >= strtab->dofs_size) { 12243 dtrace_dof_error(dof, "corrupt probe name"); 12244 return (NULL); 12245 } 12246 12247 (void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name), 12248 MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name)); 12249 12250 return (desc); 12251 } 12252 12253 static dtrace_difo_t * 12254 dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 12255 cred_t *cr) 12256 { 12257 dtrace_difo_t *dp; 12258 size_t ttl = 0; 12259 dof_difohdr_t *dofd; 12260 uintptr_t daddr = (uintptr_t)dof; 12261 size_t max = dtrace_difo_maxsize; 12262 int i, l, n; 12263 12264 static const struct { 12265 int section; 12266 int bufoffs; 12267 int lenoffs; 12268 int entsize; 12269 int align; 12270 const char *msg; 12271 } difo[] = { 12272 { DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf), 12273 offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t), 12274 sizeof (dif_instr_t), "multiple DIF sections" }, 12275 12276 { DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab), 12277 offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t), 12278 sizeof (uint64_t), "multiple integer tables" }, 12279 12280 { DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab), 12281 offsetof(dtrace_difo_t, dtdo_strlen), 0, 12282 sizeof (char), "multiple string tables" }, 12283 12284 { DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab), 12285 offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t), 12286 sizeof (uint_t), "multiple variable tables" }, 12287 12288 { DOF_SECT_NONE, 0, 0, 0, 0, NULL } 12289 }; 12290 12291 if (sec->dofs_type != DOF_SECT_DIFOHDR) { 12292 dtrace_dof_error(dof, "invalid DIFO header section"); 12293 return (NULL); 12294 } 12295 12296 if (sec->dofs_align != sizeof (dof_secidx_t)) { 12297 dtrace_dof_error(dof, "bad alignment in DIFO header"); 12298 return (NULL); 12299 } 12300 12301 if (sec->dofs_size < sizeof (dof_difohdr_t) || 12302 sec->dofs_size % sizeof (dof_secidx_t)) { 12303 dtrace_dof_error(dof, "bad size in DIFO header"); 12304 return (NULL); 12305 } 12306 12307 dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); 12308 n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1; 12309 12310 dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); 12311 dp->dtdo_rtype = dofd->dofd_rtype; 12312 12313 for (l = 0; l < n; l++) { 12314 dof_sec_t *subsec; 12315 void **bufp; 12316 uint32_t *lenp; 12317 12318 if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE, 12319 dofd->dofd_links[l])) == NULL) 12320 goto err; /* invalid section link */ 12321 12322 if (ttl + subsec->dofs_size > max) { 12323 dtrace_dof_error(dof, "exceeds maximum size"); 12324 goto err; 12325 } 12326 12327 ttl += subsec->dofs_size; 12328 12329 for (i = 0; difo[i].section != DOF_SECT_NONE; i++) { 12330 if (subsec->dofs_type != difo[i].section) 12331 continue; 12332 12333 if (!(subsec->dofs_flags & DOF_SECF_LOAD)) { 12334 dtrace_dof_error(dof, "section not loaded"); 12335 goto err; 12336 } 12337 12338 if (subsec->dofs_align != difo[i].align) { 12339 dtrace_dof_error(dof, "bad alignment"); 12340 goto err; 12341 } 12342 12343 bufp = (void **)((uintptr_t)dp + difo[i].bufoffs); 12344 lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs); 12345 12346 if (*bufp != NULL) { 12347 dtrace_dof_error(dof, difo[i].msg); 12348 goto err; 12349 } 12350 12351 if (difo[i].entsize != subsec->dofs_entsize) { 12352 dtrace_dof_error(dof, "entry size mismatch"); 12353 goto err; 12354 } 12355 12356 if (subsec->dofs_entsize != 0 && 12357 (subsec->dofs_size % subsec->dofs_entsize) != 0) { 12358 dtrace_dof_error(dof, "corrupt entry size"); 12359 goto err; 12360 } 12361 12362 *lenp = subsec->dofs_size; 12363 *bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP); 12364 bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset), 12365 *bufp, subsec->dofs_size); 12366 12367 if (subsec->dofs_entsize != 0) 12368 *lenp /= subsec->dofs_entsize; 12369 12370 break; 12371 } 12372 12373 /* 12374 * If we encounter a loadable DIFO sub-section that is not 12375 * known to us, assume this is a broken program and fail. 12376 */ 12377 if (difo[i].section == DOF_SECT_NONE && 12378 (subsec->dofs_flags & DOF_SECF_LOAD)) { 12379 dtrace_dof_error(dof, "unrecognized DIFO subsection"); 12380 goto err; 12381 } 12382 } 12383 12384 if (dp->dtdo_buf == NULL) { 12385 /* 12386 * We can't have a DIF object without DIF text. 12387 */ 12388 dtrace_dof_error(dof, "missing DIF text"); 12389 goto err; 12390 } 12391 12392 /* 12393 * Before we validate the DIF object, run through the variable table 12394 * looking for the strings -- if any of their size are under, we'll set 12395 * their size to be the system-wide default string size. Note that 12396 * this should _not_ happen if the "strsize" option has been set -- 12397 * in this case, the compiler should have set the size to reflect the 12398 * setting of the option. 12399 */ 12400 for (i = 0; i < dp->dtdo_varlen; i++) { 12401 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 12402 dtrace_diftype_t *t = &v->dtdv_type; 12403 12404 if (v->dtdv_id < DIF_VAR_OTHER_UBASE) 12405 continue; 12406 12407 if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0) 12408 t->dtdt_size = dtrace_strsize_default; 12409 } 12410 12411 if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0) 12412 goto err; 12413 12414 dtrace_difo_init(dp, vstate); 12415 return (dp); 12416 12417 err: 12418 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); 12419 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); 12420 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); 12421 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); 12422 12423 kmem_free(dp, sizeof (dtrace_difo_t)); 12424 return (NULL); 12425 } 12426 12427 static dtrace_predicate_t * 12428 dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 12429 cred_t *cr) 12430 { 12431 dtrace_difo_t *dp; 12432 12433 if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL) 12434 return (NULL); 12435 12436 return (dtrace_predicate_create(dp)); 12437 } 12438 12439 static dtrace_actdesc_t * 12440 dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 12441 cred_t *cr) 12442 { 12443 dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next; 12444 dof_actdesc_t *desc; 12445 dof_sec_t *difosec; 12446 size_t offs; 12447 uintptr_t daddr = (uintptr_t)dof; 12448 uint64_t arg; 12449 dtrace_actkind_t kind; 12450 12451 if (sec->dofs_type != DOF_SECT_ACTDESC) { 12452 dtrace_dof_error(dof, "invalid action section"); 12453 return (NULL); 12454 } 12455 12456 if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) { 12457 dtrace_dof_error(dof, "truncated action description"); 12458 return (NULL); 12459 } 12460 12461 if (sec->dofs_align != sizeof (uint64_t)) { 12462 dtrace_dof_error(dof, "bad alignment in action description"); 12463 return (NULL); 12464 } 12465 12466 if (sec->dofs_size < sec->dofs_entsize) { 12467 dtrace_dof_error(dof, "section entry size exceeds total size"); 12468 return (NULL); 12469 } 12470 12471 if (sec->dofs_entsize != sizeof (dof_actdesc_t)) { 12472 dtrace_dof_error(dof, "bad entry size in action description"); 12473 return (NULL); 12474 } 12475 12476 if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) { 12477 dtrace_dof_error(dof, "actions exceed dtrace_actions_max"); 12478 return (NULL); 12479 } 12480 12481 for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) { 12482 desc = (dof_actdesc_t *)(daddr + 12483 (uintptr_t)sec->dofs_offset + offs); 12484 kind = (dtrace_actkind_t)desc->dofa_kind; 12485 12486 if ((DTRACEACT_ISPRINTFLIKE(kind) && 12487 (kind != DTRACEACT_PRINTA || 12488 desc->dofa_strtab != DOF_SECIDX_NONE)) || 12489 (kind == DTRACEACT_DIFEXPR && 12490 desc->dofa_strtab != DOF_SECIDX_NONE)) { 12491 dof_sec_t *strtab; 12492 char *str, *fmt; 12493 uint64_t i; 12494 12495 /* 12496 * The argument to these actions is an index into the 12497 * DOF string table. For printf()-like actions, this 12498 * is the format string. For print(), this is the 12499 * CTF type of the expression result. 12500 */ 12501 if ((strtab = dtrace_dof_sect(dof, 12502 DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL) 12503 goto err; 12504 12505 str = (char *)((uintptr_t)dof + 12506 (uintptr_t)strtab->dofs_offset); 12507 12508 for (i = desc->dofa_arg; i < strtab->dofs_size; i++) { 12509 if (str[i] == '\0') 12510 break; 12511 } 12512 12513 if (i >= strtab->dofs_size) { 12514 dtrace_dof_error(dof, "bogus format string"); 12515 goto err; 12516 } 12517 12518 if (i == desc->dofa_arg) { 12519 dtrace_dof_error(dof, "empty format string"); 12520 goto err; 12521 } 12522 12523 i -= desc->dofa_arg; 12524 fmt = kmem_alloc(i + 1, KM_SLEEP); 12525 bcopy(&str[desc->dofa_arg], fmt, i + 1); 12526 arg = (uint64_t)(uintptr_t)fmt; 12527 } else { 12528 if (kind == DTRACEACT_PRINTA) { 12529 ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE); 12530 arg = 0; 12531 } else { 12532 arg = desc->dofa_arg; 12533 } 12534 } 12535 12536 act = dtrace_actdesc_create(kind, desc->dofa_ntuple, 12537 desc->dofa_uarg, arg); 12538 12539 if (last != NULL) { 12540 last->dtad_next = act; 12541 } else { 12542 first = act; 12543 } 12544 12545 last = act; 12546 12547 if (desc->dofa_difo == DOF_SECIDX_NONE) 12548 continue; 12549 12550 if ((difosec = dtrace_dof_sect(dof, 12551 DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL) 12552 goto err; 12553 12554 act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr); 12555 12556 if (act->dtad_difo == NULL) 12557 goto err; 12558 } 12559 12560 ASSERT(first != NULL); 12561 return (first); 12562 12563 err: 12564 for (act = first; act != NULL; act = next) { 12565 next = act->dtad_next; 12566 dtrace_actdesc_release(act, vstate); 12567 } 12568 12569 return (NULL); 12570 } 12571 12572 static dtrace_ecbdesc_t * 12573 dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 12574 cred_t *cr) 12575 { 12576 dtrace_ecbdesc_t *ep; 12577 dof_ecbdesc_t *ecb; 12578 dtrace_probedesc_t *desc; 12579 dtrace_predicate_t *pred = NULL; 12580 12581 if (sec->dofs_size < sizeof (dof_ecbdesc_t)) { 12582 dtrace_dof_error(dof, "truncated ECB description"); 12583 return (NULL); 12584 } 12585 12586 if (sec->dofs_align != sizeof (uint64_t)) { 12587 dtrace_dof_error(dof, "bad alignment in ECB description"); 12588 return (NULL); 12589 } 12590 12591 ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset); 12592 sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes); 12593 12594 if (sec == NULL) 12595 return (NULL); 12596 12597 ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); 12598 ep->dted_uarg = ecb->dofe_uarg; 12599 desc = &ep->dted_probe; 12600 12601 if (dtrace_dof_probedesc(dof, sec, desc) == NULL) 12602 goto err; 12603 12604 if (ecb->dofe_pred != DOF_SECIDX_NONE) { 12605 if ((sec = dtrace_dof_sect(dof, 12606 DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL) 12607 goto err; 12608 12609 if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL) 12610 goto err; 12611 12612 ep->dted_pred.dtpdd_predicate = pred; 12613 } 12614 12615 if (ecb->dofe_actions != DOF_SECIDX_NONE) { 12616 if ((sec = dtrace_dof_sect(dof, 12617 DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL) 12618 goto err; 12619 12620 ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr); 12621 12622 if (ep->dted_action == NULL) 12623 goto err; 12624 } 12625 12626 return (ep); 12627 12628 err: 12629 if (pred != NULL) 12630 dtrace_predicate_release(pred, vstate); 12631 kmem_free(ep, sizeof (dtrace_ecbdesc_t)); 12632 return (NULL); 12633 } 12634 12635 /* 12636 * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the 12637 * specified DOF. At present, this amounts to simply adding 'ubase' to the 12638 * site of any user SETX relocations to account for load object base address. 12639 * In the future, if we need other relocations, this function can be extended. 12640 */ 12641 static int 12642 dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase) 12643 { 12644 uintptr_t daddr = (uintptr_t)dof; 12645 dof_relohdr_t *dofr = 12646 (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); 12647 dof_sec_t *ss, *rs, *ts; 12648 dof_relodesc_t *r; 12649 uint_t i, n; 12650 12651 if (sec->dofs_size < sizeof (dof_relohdr_t) || 12652 sec->dofs_align != sizeof (dof_secidx_t)) { 12653 dtrace_dof_error(dof, "invalid relocation header"); 12654 return (-1); 12655 } 12656 12657 ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab); 12658 rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec); 12659 ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec); 12660 12661 if (ss == NULL || rs == NULL || ts == NULL) 12662 return (-1); /* dtrace_dof_error() has been called already */ 12663 12664 if (rs->dofs_entsize < sizeof (dof_relodesc_t) || 12665 rs->dofs_align != sizeof (uint64_t)) { 12666 dtrace_dof_error(dof, "invalid relocation section"); 12667 return (-1); 12668 } 12669 12670 r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset); 12671 n = rs->dofs_size / rs->dofs_entsize; 12672 12673 for (i = 0; i < n; i++) { 12674 uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset; 12675 12676 switch (r->dofr_type) { 12677 case DOF_RELO_NONE: 12678 break; 12679 case DOF_RELO_SETX: 12680 if (r->dofr_offset >= ts->dofs_size || r->dofr_offset + 12681 sizeof (uint64_t) > ts->dofs_size) { 12682 dtrace_dof_error(dof, "bad relocation offset"); 12683 return (-1); 12684 } 12685 12686 if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) { 12687 dtrace_dof_error(dof, "misaligned setx relo"); 12688 return (-1); 12689 } 12690 12691 *(uint64_t *)taddr += ubase; 12692 break; 12693 default: 12694 dtrace_dof_error(dof, "invalid relocation type"); 12695 return (-1); 12696 } 12697 12698 r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize); 12699 } 12700 12701 return (0); 12702 } 12703 12704 /* 12705 * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated 12706 * header: it should be at the front of a memory region that is at least 12707 * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in 12708 * size. It need not be validated in any other way. 12709 */ 12710 static int 12711 dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr, 12712 dtrace_enabling_t **enabp, uint64_t ubase, int noprobes) 12713 { 12714 uint64_t len = dof->dofh_loadsz, seclen; 12715 uintptr_t daddr = (uintptr_t)dof; 12716 dtrace_ecbdesc_t *ep; 12717 dtrace_enabling_t *enab; 12718 uint_t i; 12719 12720 ASSERT(MUTEX_HELD(&dtrace_lock)); 12721 ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t)); 12722 12723 /* 12724 * Check the DOF header identification bytes. In addition to checking 12725 * valid settings, we also verify that unused bits/bytes are zeroed so 12726 * we can use them later without fear of regressing existing binaries. 12727 */ 12728 if (bcmp(&dof->dofh_ident[DOF_ID_MAG0], 12729 DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) { 12730 dtrace_dof_error(dof, "DOF magic string mismatch"); 12731 return (-1); 12732 } 12733 12734 if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 && 12735 dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) { 12736 dtrace_dof_error(dof, "DOF has invalid data model"); 12737 return (-1); 12738 } 12739 12740 if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) { 12741 dtrace_dof_error(dof, "DOF encoding mismatch"); 12742 return (-1); 12743 } 12744 12745 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && 12746 dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) { 12747 dtrace_dof_error(dof, "DOF version mismatch"); 12748 return (-1); 12749 } 12750 12751 if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) { 12752 dtrace_dof_error(dof, "DOF uses unsupported instruction set"); 12753 return (-1); 12754 } 12755 12756 if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) { 12757 dtrace_dof_error(dof, "DOF uses too many integer registers"); 12758 return (-1); 12759 } 12760 12761 if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) { 12762 dtrace_dof_error(dof, "DOF uses too many tuple registers"); 12763 return (-1); 12764 } 12765 12766 for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) { 12767 if (dof->dofh_ident[i] != 0) { 12768 dtrace_dof_error(dof, "DOF has invalid ident byte set"); 12769 return (-1); 12770 } 12771 } 12772 12773 if (dof->dofh_flags & ~DOF_FL_VALID) { 12774 dtrace_dof_error(dof, "DOF has invalid flag bits set"); 12775 return (-1); 12776 } 12777 12778 if (dof->dofh_secsize == 0) { 12779 dtrace_dof_error(dof, "zero section header size"); 12780 return (-1); 12781 } 12782 12783 /* 12784 * Check that the section headers don't exceed the amount of DOF 12785 * data. Note that we cast the section size and number of sections 12786 * to uint64_t's to prevent possible overflow in the multiplication. 12787 */ 12788 seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize; 12789 12790 if (dof->dofh_secoff > len || seclen > len || 12791 dof->dofh_secoff + seclen > len) { 12792 dtrace_dof_error(dof, "truncated section headers"); 12793 return (-1); 12794 } 12795 12796 if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) { 12797 dtrace_dof_error(dof, "misaligned section headers"); 12798 return (-1); 12799 } 12800 12801 if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) { 12802 dtrace_dof_error(dof, "misaligned section size"); 12803 return (-1); 12804 } 12805 12806 /* 12807 * Take an initial pass through the section headers to be sure that 12808 * the headers don't have stray offsets. If the 'noprobes' flag is 12809 * set, do not permit sections relating to providers, probes, or args. 12810 */ 12811 for (i = 0; i < dof->dofh_secnum; i++) { 12812 dof_sec_t *sec = (dof_sec_t *)(daddr + 12813 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12814 12815 if (noprobes) { 12816 switch (sec->dofs_type) { 12817 case DOF_SECT_PROVIDER: 12818 case DOF_SECT_PROBES: 12819 case DOF_SECT_PRARGS: 12820 case DOF_SECT_PROFFS: 12821 dtrace_dof_error(dof, "illegal sections " 12822 "for enabling"); 12823 return (-1); 12824 } 12825 } 12826 12827 if (!(sec->dofs_flags & DOF_SECF_LOAD)) 12828 continue; /* just ignore non-loadable sections */ 12829 12830 if (sec->dofs_align & (sec->dofs_align - 1)) { 12831 dtrace_dof_error(dof, "bad section alignment"); 12832 return (-1); 12833 } 12834 12835 if (sec->dofs_offset & (sec->dofs_align - 1)) { 12836 dtrace_dof_error(dof, "misaligned section"); 12837 return (-1); 12838 } 12839 12840 if (sec->dofs_offset > len || sec->dofs_size > len || 12841 sec->dofs_offset + sec->dofs_size > len) { 12842 dtrace_dof_error(dof, "corrupt section header"); 12843 return (-1); 12844 } 12845 12846 if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr + 12847 sec->dofs_offset + sec->dofs_size - 1) != '\0') { 12848 dtrace_dof_error(dof, "non-terminating string table"); 12849 return (-1); 12850 } 12851 } 12852 12853 /* 12854 * Take a second pass through the sections and locate and perform any 12855 * relocations that are present. We do this after the first pass to 12856 * be sure that all sections have had their headers validated. 12857 */ 12858 for (i = 0; i < dof->dofh_secnum; i++) { 12859 dof_sec_t *sec = (dof_sec_t *)(daddr + 12860 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12861 12862 if (!(sec->dofs_flags & DOF_SECF_LOAD)) 12863 continue; /* skip sections that are not loadable */ 12864 12865 switch (sec->dofs_type) { 12866 case DOF_SECT_URELHDR: 12867 if (dtrace_dof_relocate(dof, sec, ubase) != 0) 12868 return (-1); 12869 break; 12870 } 12871 } 12872 12873 if ((enab = *enabp) == NULL) 12874 enab = *enabp = dtrace_enabling_create(vstate); 12875 12876 for (i = 0; i < dof->dofh_secnum; i++) { 12877 dof_sec_t *sec = (dof_sec_t *)(daddr + 12878 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12879 12880 if (sec->dofs_type != DOF_SECT_ECBDESC) 12881 continue; 12882 12883 if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) { 12884 dtrace_enabling_destroy(enab); 12885 *enabp = NULL; 12886 return (-1); 12887 } 12888 12889 dtrace_enabling_add(enab, ep); 12890 } 12891 12892 return (0); 12893 } 12894 12895 /* 12896 * Process DOF for any options. This routine assumes that the DOF has been 12897 * at least processed by dtrace_dof_slurp(). 12898 */ 12899 static int 12900 dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state) 12901 { 12902 int i, rval; 12903 uint32_t entsize; 12904 size_t offs; 12905 dof_optdesc_t *desc; 12906 12907 for (i = 0; i < dof->dofh_secnum; i++) { 12908 dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof + 12909 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12910 12911 if (sec->dofs_type != DOF_SECT_OPTDESC) 12912 continue; 12913 12914 if (sec->dofs_align != sizeof (uint64_t)) { 12915 dtrace_dof_error(dof, "bad alignment in " 12916 "option description"); 12917 return (EINVAL); 12918 } 12919 12920 if ((entsize = sec->dofs_entsize) == 0) { 12921 dtrace_dof_error(dof, "zeroed option entry size"); 12922 return (EINVAL); 12923 } 12924 12925 if (entsize < sizeof (dof_optdesc_t)) { 12926 dtrace_dof_error(dof, "bad option entry size"); 12927 return (EINVAL); 12928 } 12929 12930 for (offs = 0; offs < sec->dofs_size; offs += entsize) { 12931 desc = (dof_optdesc_t *)((uintptr_t)dof + 12932 (uintptr_t)sec->dofs_offset + offs); 12933 12934 if (desc->dofo_strtab != DOF_SECIDX_NONE) { 12935 dtrace_dof_error(dof, "non-zero option string"); 12936 return (EINVAL); 12937 } 12938 12939 if (desc->dofo_value == DTRACEOPT_UNSET) { 12940 dtrace_dof_error(dof, "unset option"); 12941 return (EINVAL); 12942 } 12943 12944 if ((rval = dtrace_state_option(state, 12945 desc->dofo_option, desc->dofo_value)) != 0) { 12946 dtrace_dof_error(dof, "rejected option"); 12947 return (rval); 12948 } 12949 } 12950 } 12951 12952 return (0); 12953 } 12954 12955 /* 12956 * DTrace Consumer State Functions 12957 */ 12958 static int 12959 dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size) 12960 { 12961 size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize; 12962 void *base; 12963 uintptr_t limit; 12964 dtrace_dynvar_t *dvar, *next, *start; 12965 int i; 12966 12967 ASSERT(MUTEX_HELD(&dtrace_lock)); 12968 ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL); 12969 12970 bzero(dstate, sizeof (dtrace_dstate_t)); 12971 12972 if ((dstate->dtds_chunksize = chunksize) == 0) 12973 dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE; 12974 12975 if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t))) 12976 size = min; 12977 12978 if ((base = kmem_zalloc(size, KM_NOSLEEP | KM_NORMALPRI)) == NULL) 12979 return (ENOMEM); 12980 12981 dstate->dtds_size = size; 12982 dstate->dtds_base = base; 12983 dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP); 12984 bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t)); 12985 12986 hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)); 12987 12988 if (hashsize != 1 && (hashsize & 1)) 12989 hashsize--; 12990 12991 dstate->dtds_hashsize = hashsize; 12992 dstate->dtds_hash = dstate->dtds_base; 12993 12994 /* 12995 * Set all of our hash buckets to point to the single sink, and (if 12996 * it hasn't already been set), set the sink's hash value to be the 12997 * sink sentinel value. The sink is needed for dynamic variable 12998 * lookups to know that they have iterated over an entire, valid hash 12999 * chain. 13000 */ 13001 for (i = 0; i < hashsize; i++) 13002 dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink; 13003 13004 if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK) 13005 dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK; 13006 13007 /* 13008 * Determine number of active CPUs. Divide free list evenly among 13009 * active CPUs. 13010 */ 13011 start = (dtrace_dynvar_t *) 13012 ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t)); 13013 limit = (uintptr_t)base + size; 13014 13015 maxper = (limit - (uintptr_t)start) / NCPU; 13016 maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize; 13017 13018 #if !defined(sun) 13019 CPU_FOREACH(i) { 13020 #else 13021 for (i = 0; i < NCPU; i++) { 13022 #endif 13023 dstate->dtds_percpu[i].dtdsc_free = dvar = start; 13024 13025 /* 13026 * If we don't even have enough chunks to make it once through 13027 * NCPUs, we're just going to allocate everything to the first 13028 * CPU. And if we're on the last CPU, we're going to allocate 13029 * whatever is left over. In either case, we set the limit to 13030 * be the limit of the dynamic variable space. 13031 */ 13032 if (maxper == 0 || i == NCPU - 1) { 13033 limit = (uintptr_t)base + size; 13034 start = NULL; 13035 } else { 13036 limit = (uintptr_t)start + maxper; 13037 start = (dtrace_dynvar_t *)limit; 13038 } 13039 13040 ASSERT(limit <= (uintptr_t)base + size); 13041 13042 for (;;) { 13043 next = (dtrace_dynvar_t *)((uintptr_t)dvar + 13044 dstate->dtds_chunksize); 13045 13046 if ((uintptr_t)next + dstate->dtds_chunksize >= limit) 13047 break; 13048 13049 dvar->dtdv_next = next; 13050 dvar = next; 13051 } 13052 13053 if (maxper == 0) 13054 break; 13055 } 13056 13057 return (0); 13058 } 13059 13060 static void 13061 dtrace_dstate_fini(dtrace_dstate_t *dstate) 13062 { 13063 ASSERT(MUTEX_HELD(&cpu_lock)); 13064 13065 if (dstate->dtds_base == NULL) 13066 return; 13067 13068 kmem_free(dstate->dtds_base, dstate->dtds_size); 13069 kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu); 13070 } 13071 13072 static void 13073 dtrace_vstate_fini(dtrace_vstate_t *vstate) 13074 { 13075 /* 13076 * Logical XOR, where are you? 13077 */ 13078 ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL)); 13079 13080 if (vstate->dtvs_nglobals > 0) { 13081 kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals * 13082 sizeof (dtrace_statvar_t *)); 13083 } 13084 13085 if (vstate->dtvs_ntlocals > 0) { 13086 kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals * 13087 sizeof (dtrace_difv_t)); 13088 } 13089 13090 ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL)); 13091 13092 if (vstate->dtvs_nlocals > 0) { 13093 kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals * 13094 sizeof (dtrace_statvar_t *)); 13095 } 13096 } 13097 13098 #if defined(sun) 13099 static void 13100 dtrace_state_clean(dtrace_state_t *state) 13101 { 13102 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) 13103 return; 13104 13105 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); 13106 dtrace_speculation_clean(state); 13107 } 13108 13109 static void 13110 dtrace_state_deadman(dtrace_state_t *state) 13111 { 13112 hrtime_t now; 13113 13114 dtrace_sync(); 13115 13116 now = dtrace_gethrtime(); 13117 13118 if (state != dtrace_anon.dta_state && 13119 now - state->dts_laststatus >= dtrace_deadman_user) 13120 return; 13121 13122 /* 13123 * We must be sure that dts_alive never appears to be less than the 13124 * value upon entry to dtrace_state_deadman(), and because we lack a 13125 * dtrace_cas64(), we cannot store to it atomically. We thus instead 13126 * store INT64_MAX to it, followed by a memory barrier, followed by 13127 * the new value. This assures that dts_alive never appears to be 13128 * less than its true value, regardless of the order in which the 13129 * stores to the underlying storage are issued. 13130 */ 13131 state->dts_alive = INT64_MAX; 13132 dtrace_membar_producer(); 13133 state->dts_alive = now; 13134 } 13135 #else 13136 static void 13137 dtrace_state_clean(void *arg) 13138 { 13139 dtrace_state_t *state = arg; 13140 dtrace_optval_t *opt = state->dts_options; 13141 13142 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) 13143 return; 13144 13145 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); 13146 dtrace_speculation_clean(state); 13147 13148 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC, 13149 dtrace_state_clean, state); 13150 } 13151 13152 static void 13153 dtrace_state_deadman(void *arg) 13154 { 13155 dtrace_state_t *state = arg; 13156 hrtime_t now; 13157 13158 dtrace_sync(); 13159 13160 dtrace_debug_output(); 13161 13162 now = dtrace_gethrtime(); 13163 13164 if (state != dtrace_anon.dta_state && 13165 now - state->dts_laststatus >= dtrace_deadman_user) 13166 return; 13167 13168 /* 13169 * We must be sure that dts_alive never appears to be less than the 13170 * value upon entry to dtrace_state_deadman(), and because we lack a 13171 * dtrace_cas64(), we cannot store to it atomically. We thus instead 13172 * store INT64_MAX to it, followed by a memory barrier, followed by 13173 * the new value. This assures that dts_alive never appears to be 13174 * less than its true value, regardless of the order in which the 13175 * stores to the underlying storage are issued. 13176 */ 13177 state->dts_alive = INT64_MAX; 13178 dtrace_membar_producer(); 13179 state->dts_alive = now; 13180 13181 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC, 13182 dtrace_state_deadman, state); 13183 } 13184 #endif 13185 13186 static dtrace_state_t * 13187 #if defined(sun) 13188 dtrace_state_create(dev_t *devp, cred_t *cr) 13189 #else 13190 dtrace_state_create(struct cdev *dev) 13191 #endif 13192 { 13193 #if defined(sun) 13194 minor_t minor; 13195 major_t major; 13196 #else 13197 cred_t *cr = NULL; 13198 int m = 0; 13199 #endif 13200 char c[30]; 13201 dtrace_state_t *state; 13202 dtrace_optval_t *opt; 13203 int bufsize = NCPU * sizeof (dtrace_buffer_t), i; 13204 13205 ASSERT(MUTEX_HELD(&dtrace_lock)); 13206 ASSERT(MUTEX_HELD(&cpu_lock)); 13207 13208 #if defined(sun) 13209 minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1, 13210 VM_BESTFIT | VM_SLEEP); 13211 13212 if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) { 13213 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); 13214 return (NULL); 13215 } 13216 13217 state = ddi_get_soft_state(dtrace_softstate, minor); 13218 #else 13219 if (dev != NULL) { 13220 cr = dev->si_cred; 13221 m = dev2unit(dev); 13222 } 13223 13224 /* Allocate memory for the state. */ 13225 state = kmem_zalloc(sizeof(dtrace_state_t), KM_SLEEP); 13226 #endif 13227 13228 state->dts_epid = DTRACE_EPIDNONE + 1; 13229 13230 (void) snprintf(c, sizeof (c), "dtrace_aggid_%d", m); 13231 #if defined(sun) 13232 state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1, 13233 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); 13234 13235 if (devp != NULL) { 13236 major = getemajor(*devp); 13237 } else { 13238 major = ddi_driver_major(dtrace_devi); 13239 } 13240 13241 state->dts_dev = makedevice(major, minor); 13242 13243 if (devp != NULL) 13244 *devp = state->dts_dev; 13245 #else 13246 state->dts_aggid_arena = new_unrhdr(1, INT_MAX, &dtrace_unr_mtx); 13247 state->dts_dev = dev; 13248 #endif 13249 13250 /* 13251 * We allocate NCPU buffers. On the one hand, this can be quite 13252 * a bit of memory per instance (nearly 36K on a Starcat). On the 13253 * other hand, it saves an additional memory reference in the probe 13254 * path. 13255 */ 13256 state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP); 13257 state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP); 13258 13259 #if defined(sun) 13260 state->dts_cleaner = CYCLIC_NONE; 13261 state->dts_deadman = CYCLIC_NONE; 13262 #else 13263 callout_init(&state->dts_cleaner, CALLOUT_MPSAFE); 13264 callout_init(&state->dts_deadman, CALLOUT_MPSAFE); 13265 #endif 13266 state->dts_vstate.dtvs_state = state; 13267 13268 for (i = 0; i < DTRACEOPT_MAX; i++) 13269 state->dts_options[i] = DTRACEOPT_UNSET; 13270 13271 /* 13272 * Set the default options. 13273 */ 13274 opt = state->dts_options; 13275 opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH; 13276 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO; 13277 opt[DTRACEOPT_NSPEC] = dtrace_nspec_default; 13278 opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default; 13279 opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL; 13280 opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default; 13281 opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default; 13282 opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default; 13283 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default; 13284 opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default; 13285 opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default; 13286 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default; 13287 opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default; 13288 opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default; 13289 13290 state->dts_activity = DTRACE_ACTIVITY_INACTIVE; 13291 13292 /* 13293 * Depending on the user credentials, we set flag bits which alter probe 13294 * visibility or the amount of destructiveness allowed. In the case of 13295 * actual anonymous tracing, or the possession of all privileges, all of 13296 * the normal checks are bypassed. 13297 */ 13298 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { 13299 state->dts_cred.dcr_visible = DTRACE_CRV_ALL; 13300 state->dts_cred.dcr_action = DTRACE_CRA_ALL; 13301 } else { 13302 /* 13303 * Set up the credentials for this instantiation. We take a 13304 * hold on the credential to prevent it from disappearing on 13305 * us; this in turn prevents the zone_t referenced by this 13306 * credential from disappearing. This means that we can 13307 * examine the credential and the zone from probe context. 13308 */ 13309 crhold(cr); 13310 state->dts_cred.dcr_cred = cr; 13311 13312 /* 13313 * CRA_PROC means "we have *some* privilege for dtrace" and 13314 * unlocks the use of variables like pid, zonename, etc. 13315 */ 13316 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) || 13317 PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { 13318 state->dts_cred.dcr_action |= DTRACE_CRA_PROC; 13319 } 13320 13321 /* 13322 * dtrace_user allows use of syscall and profile providers. 13323 * If the user also has proc_owner and/or proc_zone, we 13324 * extend the scope to include additional visibility and 13325 * destructive power. 13326 */ 13327 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) { 13328 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) { 13329 state->dts_cred.dcr_visible |= 13330 DTRACE_CRV_ALLPROC; 13331 13332 state->dts_cred.dcr_action |= 13333 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; 13334 } 13335 13336 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) { 13337 state->dts_cred.dcr_visible |= 13338 DTRACE_CRV_ALLZONE; 13339 13340 state->dts_cred.dcr_action |= 13341 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; 13342 } 13343 13344 /* 13345 * If we have all privs in whatever zone this is, 13346 * we can do destructive things to processes which 13347 * have altered credentials. 13348 */ 13349 #if defined(sun) 13350 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), 13351 cr->cr_zone->zone_privset)) { 13352 state->dts_cred.dcr_action |= 13353 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; 13354 } 13355 #endif 13356 } 13357 13358 /* 13359 * Holding the dtrace_kernel privilege also implies that 13360 * the user has the dtrace_user privilege from a visibility 13361 * perspective. But without further privileges, some 13362 * destructive actions are not available. 13363 */ 13364 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) { 13365 /* 13366 * Make all probes in all zones visible. However, 13367 * this doesn't mean that all actions become available 13368 * to all zones. 13369 */ 13370 state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL | 13371 DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE; 13372 13373 state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL | 13374 DTRACE_CRA_PROC; 13375 /* 13376 * Holding proc_owner means that destructive actions 13377 * for *this* zone are allowed. 13378 */ 13379 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) 13380 state->dts_cred.dcr_action |= 13381 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; 13382 13383 /* 13384 * Holding proc_zone means that destructive actions 13385 * for this user/group ID in all zones is allowed. 13386 */ 13387 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) 13388 state->dts_cred.dcr_action |= 13389 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; 13390 13391 #if defined(sun) 13392 /* 13393 * If we have all privs in whatever zone this is, 13394 * we can do destructive things to processes which 13395 * have altered credentials. 13396 */ 13397 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), 13398 cr->cr_zone->zone_privset)) { 13399 state->dts_cred.dcr_action |= 13400 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; 13401 } 13402 #endif 13403 } 13404 13405 /* 13406 * Holding the dtrace_proc privilege gives control over fasttrap 13407 * and pid providers. We need to grant wider destructive 13408 * privileges in the event that the user has proc_owner and/or 13409 * proc_zone. 13410 */ 13411 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { 13412 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) 13413 state->dts_cred.dcr_action |= 13414 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; 13415 13416 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) 13417 state->dts_cred.dcr_action |= 13418 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; 13419 } 13420 } 13421 13422 return (state); 13423 } 13424 13425 static int 13426 dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which) 13427 { 13428 dtrace_optval_t *opt = state->dts_options, size; 13429 processorid_t cpu = 0;; 13430 int flags = 0, rval, factor, divisor = 1; 13431 13432 ASSERT(MUTEX_HELD(&dtrace_lock)); 13433 ASSERT(MUTEX_HELD(&cpu_lock)); 13434 ASSERT(which < DTRACEOPT_MAX); 13435 ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE || 13436 (state == dtrace_anon.dta_state && 13437 state->dts_activity == DTRACE_ACTIVITY_ACTIVE)); 13438 13439 if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0) 13440 return (0); 13441 13442 if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET) 13443 cpu = opt[DTRACEOPT_CPU]; 13444 13445 if (which == DTRACEOPT_SPECSIZE) 13446 flags |= DTRACEBUF_NOSWITCH; 13447 13448 if (which == DTRACEOPT_BUFSIZE) { 13449 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING) 13450 flags |= DTRACEBUF_RING; 13451 13452 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL) 13453 flags |= DTRACEBUF_FILL; 13454 13455 if (state != dtrace_anon.dta_state || 13456 state->dts_activity != DTRACE_ACTIVITY_ACTIVE) 13457 flags |= DTRACEBUF_INACTIVE; 13458 } 13459 13460 for (size = opt[which]; size >= sizeof (uint64_t); size /= divisor) { 13461 /* 13462 * The size must be 8-byte aligned. If the size is not 8-byte 13463 * aligned, drop it down by the difference. 13464 */ 13465 if (size & (sizeof (uint64_t) - 1)) 13466 size -= size & (sizeof (uint64_t) - 1); 13467 13468 if (size < state->dts_reserve) { 13469 /* 13470 * Buffers always must be large enough to accommodate 13471 * their prereserved space. We return E2BIG instead 13472 * of ENOMEM in this case to allow for user-level 13473 * software to differentiate the cases. 13474 */ 13475 return (E2BIG); 13476 } 13477 13478 rval = dtrace_buffer_alloc(buf, size, flags, cpu, &factor); 13479 13480 if (rval != ENOMEM) { 13481 opt[which] = size; 13482 return (rval); 13483 } 13484 13485 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) 13486 return (rval); 13487 13488 for (divisor = 2; divisor < factor; divisor <<= 1) 13489 continue; 13490 } 13491 13492 return (ENOMEM); 13493 } 13494 13495 static int 13496 dtrace_state_buffers(dtrace_state_t *state) 13497 { 13498 dtrace_speculation_t *spec = state->dts_speculations; 13499 int rval, i; 13500 13501 if ((rval = dtrace_state_buffer(state, state->dts_buffer, 13502 DTRACEOPT_BUFSIZE)) != 0) 13503 return (rval); 13504 13505 if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer, 13506 DTRACEOPT_AGGSIZE)) != 0) 13507 return (rval); 13508 13509 for (i = 0; i < state->dts_nspeculations; i++) { 13510 if ((rval = dtrace_state_buffer(state, 13511 spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0) 13512 return (rval); 13513 } 13514 13515 return (0); 13516 } 13517 13518 static void 13519 dtrace_state_prereserve(dtrace_state_t *state) 13520 { 13521 dtrace_ecb_t *ecb; 13522 dtrace_probe_t *probe; 13523 13524 state->dts_reserve = 0; 13525 13526 if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL) 13527 return; 13528 13529 /* 13530 * If our buffer policy is a "fill" buffer policy, we need to set the 13531 * prereserved space to be the space required by the END probes. 13532 */ 13533 probe = dtrace_probes[dtrace_probeid_end - 1]; 13534 ASSERT(probe != NULL); 13535 13536 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { 13537 if (ecb->dte_state != state) 13538 continue; 13539 13540 state->dts_reserve += ecb->dte_needed + ecb->dte_alignment; 13541 } 13542 } 13543 13544 static int 13545 dtrace_state_go(dtrace_state_t *state, processorid_t *cpu) 13546 { 13547 dtrace_optval_t *opt = state->dts_options, sz, nspec; 13548 dtrace_speculation_t *spec; 13549 dtrace_buffer_t *buf; 13550 #if defined(sun) 13551 cyc_handler_t hdlr; 13552 cyc_time_t when; 13553 #endif 13554 int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t); 13555 dtrace_icookie_t cookie; 13556 13557 mutex_enter(&cpu_lock); 13558 mutex_enter(&dtrace_lock); 13559 13560 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { 13561 rval = EBUSY; 13562 goto out; 13563 } 13564 13565 /* 13566 * Before we can perform any checks, we must prime all of the 13567 * retained enablings that correspond to this state. 13568 */ 13569 dtrace_enabling_prime(state); 13570 13571 if (state->dts_destructive && !state->dts_cred.dcr_destructive) { 13572 rval = EACCES; 13573 goto out; 13574 } 13575 13576 dtrace_state_prereserve(state); 13577 13578 /* 13579 * Now we want to do is try to allocate our speculations. 13580 * We do not automatically resize the number of speculations; if 13581 * this fails, we will fail the operation. 13582 */ 13583 nspec = opt[DTRACEOPT_NSPEC]; 13584 ASSERT(nspec != DTRACEOPT_UNSET); 13585 13586 if (nspec > INT_MAX) { 13587 rval = ENOMEM; 13588 goto out; 13589 } 13590 13591 spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t), 13592 KM_NOSLEEP | KM_NORMALPRI); 13593 13594 if (spec == NULL) { 13595 rval = ENOMEM; 13596 goto out; 13597 } 13598 13599 state->dts_speculations = spec; 13600 state->dts_nspeculations = (int)nspec; 13601 13602 for (i = 0; i < nspec; i++) { 13603 if ((buf = kmem_zalloc(bufsize, 13604 KM_NOSLEEP | KM_NORMALPRI)) == NULL) { 13605 rval = ENOMEM; 13606 goto err; 13607 } 13608 13609 spec[i].dtsp_buffer = buf; 13610 } 13611 13612 if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) { 13613 if (dtrace_anon.dta_state == NULL) { 13614 rval = ENOENT; 13615 goto out; 13616 } 13617 13618 if (state->dts_necbs != 0) { 13619 rval = EALREADY; 13620 goto out; 13621 } 13622 13623 state->dts_anon = dtrace_anon_grab(); 13624 ASSERT(state->dts_anon != NULL); 13625 state = state->dts_anon; 13626 13627 /* 13628 * We want "grabanon" to be set in the grabbed state, so we'll 13629 * copy that option value from the grabbing state into the 13630 * grabbed state. 13631 */ 13632 state->dts_options[DTRACEOPT_GRABANON] = 13633 opt[DTRACEOPT_GRABANON]; 13634 13635 *cpu = dtrace_anon.dta_beganon; 13636 13637 /* 13638 * If the anonymous state is active (as it almost certainly 13639 * is if the anonymous enabling ultimately matched anything), 13640 * we don't allow any further option processing -- but we 13641 * don't return failure. 13642 */ 13643 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) 13644 goto out; 13645 } 13646 13647 if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET && 13648 opt[DTRACEOPT_AGGSIZE] != 0) { 13649 if (state->dts_aggregations == NULL) { 13650 /* 13651 * We're not going to create an aggregation buffer 13652 * because we don't have any ECBs that contain 13653 * aggregations -- set this option to 0. 13654 */ 13655 opt[DTRACEOPT_AGGSIZE] = 0; 13656 } else { 13657 /* 13658 * If we have an aggregation buffer, we must also have 13659 * a buffer to use as scratch. 13660 */ 13661 if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET || 13662 opt[DTRACEOPT_BUFSIZE] < state->dts_needed) { 13663 opt[DTRACEOPT_BUFSIZE] = state->dts_needed; 13664 } 13665 } 13666 } 13667 13668 if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET && 13669 opt[DTRACEOPT_SPECSIZE] != 0) { 13670 if (!state->dts_speculates) { 13671 /* 13672 * We're not going to create speculation buffers 13673 * because we don't have any ECBs that actually 13674 * speculate -- set the speculation size to 0. 13675 */ 13676 opt[DTRACEOPT_SPECSIZE] = 0; 13677 } 13678 } 13679 13680 /* 13681 * The bare minimum size for any buffer that we're actually going to 13682 * do anything to is sizeof (uint64_t). 13683 */ 13684 sz = sizeof (uint64_t); 13685 13686 if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) || 13687 (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) || 13688 (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) { 13689 /* 13690 * A buffer size has been explicitly set to 0 (or to a size 13691 * that will be adjusted to 0) and we need the space -- we 13692 * need to return failure. We return ENOSPC to differentiate 13693 * it from failing to allocate a buffer due to failure to meet 13694 * the reserve (for which we return E2BIG). 13695 */ 13696 rval = ENOSPC; 13697 goto out; 13698 } 13699 13700 if ((rval = dtrace_state_buffers(state)) != 0) 13701 goto err; 13702 13703 if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET) 13704 sz = dtrace_dstate_defsize; 13705 13706 do { 13707 rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz); 13708 13709 if (rval == 0) 13710 break; 13711 13712 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) 13713 goto err; 13714 } while (sz >>= 1); 13715 13716 opt[DTRACEOPT_DYNVARSIZE] = sz; 13717 13718 if (rval != 0) 13719 goto err; 13720 13721 if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max) 13722 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max; 13723 13724 if (opt[DTRACEOPT_CLEANRATE] == 0) 13725 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; 13726 13727 if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min) 13728 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min; 13729 13730 if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max) 13731 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; 13732 13733 state->dts_alive = state->dts_laststatus = dtrace_gethrtime(); 13734 #if defined(sun) 13735 hdlr.cyh_func = (cyc_func_t)dtrace_state_clean; 13736 hdlr.cyh_arg = state; 13737 hdlr.cyh_level = CY_LOW_LEVEL; 13738 13739 when.cyt_when = 0; 13740 when.cyt_interval = opt[DTRACEOPT_CLEANRATE]; 13741 13742 state->dts_cleaner = cyclic_add(&hdlr, &when); 13743 13744 hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman; 13745 hdlr.cyh_arg = state; 13746 hdlr.cyh_level = CY_LOW_LEVEL; 13747 13748 when.cyt_when = 0; 13749 when.cyt_interval = dtrace_deadman_interval; 13750 13751 state->dts_deadman = cyclic_add(&hdlr, &when); 13752 #else 13753 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC, 13754 dtrace_state_clean, state); 13755 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC, 13756 dtrace_state_deadman, state); 13757 #endif 13758 13759 state->dts_activity = DTRACE_ACTIVITY_WARMUP; 13760 13761 /* 13762 * Now it's time to actually fire the BEGIN probe. We need to disable 13763 * interrupts here both to record the CPU on which we fired the BEGIN 13764 * probe (the data from this CPU will be processed first at user 13765 * level) and to manually activate the buffer for this CPU. 13766 */ 13767 cookie = dtrace_interrupt_disable(); 13768 *cpu = curcpu; 13769 ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE); 13770 state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE; 13771 13772 dtrace_probe(dtrace_probeid_begin, 13773 (uint64_t)(uintptr_t)state, 0, 0, 0, 0); 13774 dtrace_interrupt_enable(cookie); 13775 /* 13776 * We may have had an exit action from a BEGIN probe; only change our 13777 * state to ACTIVE if we're still in WARMUP. 13778 */ 13779 ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP || 13780 state->dts_activity == DTRACE_ACTIVITY_DRAINING); 13781 13782 if (state->dts_activity == DTRACE_ACTIVITY_WARMUP) 13783 state->dts_activity = DTRACE_ACTIVITY_ACTIVE; 13784 13785 /* 13786 * Regardless of whether or not now we're in ACTIVE or DRAINING, we 13787 * want each CPU to transition its principal buffer out of the 13788 * INACTIVE state. Doing this assures that no CPU will suddenly begin 13789 * processing an ECB halfway down a probe's ECB chain; all CPUs will 13790 * atomically transition from processing none of a state's ECBs to 13791 * processing all of them. 13792 */ 13793 dtrace_xcall(DTRACE_CPUALL, 13794 (dtrace_xcall_t)dtrace_buffer_activate, state); 13795 goto out; 13796 13797 err: 13798 dtrace_buffer_free(state->dts_buffer); 13799 dtrace_buffer_free(state->dts_aggbuffer); 13800 13801 if ((nspec = state->dts_nspeculations) == 0) { 13802 ASSERT(state->dts_speculations == NULL); 13803 goto out; 13804 } 13805 13806 spec = state->dts_speculations; 13807 ASSERT(spec != NULL); 13808 13809 for (i = 0; i < state->dts_nspeculations; i++) { 13810 if ((buf = spec[i].dtsp_buffer) == NULL) 13811 break; 13812 13813 dtrace_buffer_free(buf); 13814 kmem_free(buf, bufsize); 13815 } 13816 13817 kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); 13818 state->dts_nspeculations = 0; 13819 state->dts_speculations = NULL; 13820 13821 out: 13822 mutex_exit(&dtrace_lock); 13823 mutex_exit(&cpu_lock); 13824 13825 return (rval); 13826 } 13827 13828 static int 13829 dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu) 13830 { 13831 dtrace_icookie_t cookie; 13832 13833 ASSERT(MUTEX_HELD(&dtrace_lock)); 13834 13835 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE && 13836 state->dts_activity != DTRACE_ACTIVITY_DRAINING) 13837 return (EINVAL); 13838 13839 /* 13840 * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync 13841 * to be sure that every CPU has seen it. See below for the details 13842 * on why this is done. 13843 */ 13844 state->dts_activity = DTRACE_ACTIVITY_DRAINING; 13845 dtrace_sync(); 13846 13847 /* 13848 * By this point, it is impossible for any CPU to be still processing 13849 * with DTRACE_ACTIVITY_ACTIVE. We can thus set our activity to 13850 * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any 13851 * other CPU in dtrace_buffer_reserve(). This allows dtrace_probe() 13852 * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN 13853 * iff we're in the END probe. 13854 */ 13855 state->dts_activity = DTRACE_ACTIVITY_COOLDOWN; 13856 dtrace_sync(); 13857 ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN); 13858 13859 /* 13860 * Finally, we can release the reserve and call the END probe. We 13861 * disable interrupts across calling the END probe to allow us to 13862 * return the CPU on which we actually called the END probe. This 13863 * allows user-land to be sure that this CPU's principal buffer is 13864 * processed last. 13865 */ 13866 state->dts_reserve = 0; 13867 13868 cookie = dtrace_interrupt_disable(); 13869 *cpu = curcpu; 13870 dtrace_probe(dtrace_probeid_end, 13871 (uint64_t)(uintptr_t)state, 0, 0, 0, 0); 13872 dtrace_interrupt_enable(cookie); 13873 13874 state->dts_activity = DTRACE_ACTIVITY_STOPPED; 13875 dtrace_sync(); 13876 13877 return (0); 13878 } 13879 13880 static int 13881 dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option, 13882 dtrace_optval_t val) 13883 { 13884 ASSERT(MUTEX_HELD(&dtrace_lock)); 13885 13886 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) 13887 return (EBUSY); 13888 13889 if (option >= DTRACEOPT_MAX) 13890 return (EINVAL); 13891 13892 if (option != DTRACEOPT_CPU && val < 0) 13893 return (EINVAL); 13894 13895 switch (option) { 13896 case DTRACEOPT_DESTRUCTIVE: 13897 if (dtrace_destructive_disallow) 13898 return (EACCES); 13899 13900 state->dts_cred.dcr_destructive = 1; 13901 break; 13902 13903 case DTRACEOPT_BUFSIZE: 13904 case DTRACEOPT_DYNVARSIZE: 13905 case DTRACEOPT_AGGSIZE: 13906 case DTRACEOPT_SPECSIZE: 13907 case DTRACEOPT_STRSIZE: 13908 if (val < 0) 13909 return (EINVAL); 13910 13911 if (val >= LONG_MAX) { 13912 /* 13913 * If this is an otherwise negative value, set it to 13914 * the highest multiple of 128m less than LONG_MAX. 13915 * Technically, we're adjusting the size without 13916 * regard to the buffer resizing policy, but in fact, 13917 * this has no effect -- if we set the buffer size to 13918 * ~LONG_MAX and the buffer policy is ultimately set to 13919 * be "manual", the buffer allocation is guaranteed to 13920 * fail, if only because the allocation requires two 13921 * buffers. (We set the the size to the highest 13922 * multiple of 128m because it ensures that the size 13923 * will remain a multiple of a megabyte when 13924 * repeatedly halved -- all the way down to 15m.) 13925 */ 13926 val = LONG_MAX - (1 << 27) + 1; 13927 } 13928 } 13929 13930 state->dts_options[option] = val; 13931 13932 return (0); 13933 } 13934 13935 static void 13936 dtrace_state_destroy(dtrace_state_t *state) 13937 { 13938 dtrace_ecb_t *ecb; 13939 dtrace_vstate_t *vstate = &state->dts_vstate; 13940 #if defined(sun) 13941 minor_t minor = getminor(state->dts_dev); 13942 #endif 13943 int i, bufsize = NCPU * sizeof (dtrace_buffer_t); 13944 dtrace_speculation_t *spec = state->dts_speculations; 13945 int nspec = state->dts_nspeculations; 13946 uint32_t match; 13947 13948 ASSERT(MUTEX_HELD(&dtrace_lock)); 13949 ASSERT(MUTEX_HELD(&cpu_lock)); 13950 13951 /* 13952 * First, retract any retained enablings for this state. 13953 */ 13954 dtrace_enabling_retract(state); 13955 ASSERT(state->dts_nretained == 0); 13956 13957 if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE || 13958 state->dts_activity == DTRACE_ACTIVITY_DRAINING) { 13959 /* 13960 * We have managed to come into dtrace_state_destroy() on a 13961 * hot enabling -- almost certainly because of a disorderly 13962 * shutdown of a consumer. (That is, a consumer that is 13963 * exiting without having called dtrace_stop().) In this case, 13964 * we're going to set our activity to be KILLED, and then 13965 * issue a sync to be sure that everyone is out of probe 13966 * context before we start blowing away ECBs. 13967 */ 13968 state->dts_activity = DTRACE_ACTIVITY_KILLED; 13969 dtrace_sync(); 13970 } 13971 13972 /* 13973 * Release the credential hold we took in dtrace_state_create(). 13974 */ 13975 if (state->dts_cred.dcr_cred != NULL) 13976 crfree(state->dts_cred.dcr_cred); 13977 13978 /* 13979 * Now we can safely disable and destroy any enabled probes. Because 13980 * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress 13981 * (especially if they're all enabled), we take two passes through the 13982 * ECBs: in the first, we disable just DTRACE_PRIV_KERNEL probes, and 13983 * in the second we disable whatever is left over. 13984 */ 13985 for (match = DTRACE_PRIV_KERNEL; ; match = 0) { 13986 for (i = 0; i < state->dts_necbs; i++) { 13987 if ((ecb = state->dts_ecbs[i]) == NULL) 13988 continue; 13989 13990 if (match && ecb->dte_probe != NULL) { 13991 dtrace_probe_t *probe = ecb->dte_probe; 13992 dtrace_provider_t *prov = probe->dtpr_provider; 13993 13994 if (!(prov->dtpv_priv.dtpp_flags & match)) 13995 continue; 13996 } 13997 13998 dtrace_ecb_disable(ecb); 13999 dtrace_ecb_destroy(ecb); 14000 } 14001 14002 if (!match) 14003 break; 14004 } 14005 14006 /* 14007 * Before we free the buffers, perform one more sync to assure that 14008 * every CPU is out of probe context. 14009 */ 14010 dtrace_sync(); 14011 14012 dtrace_buffer_free(state->dts_buffer); 14013 dtrace_buffer_free(state->dts_aggbuffer); 14014 14015 for (i = 0; i < nspec; i++) 14016 dtrace_buffer_free(spec[i].dtsp_buffer); 14017 14018 #if defined(sun) 14019 if (state->dts_cleaner != CYCLIC_NONE) 14020 cyclic_remove(state->dts_cleaner); 14021 14022 if (state->dts_deadman != CYCLIC_NONE) 14023 cyclic_remove(state->dts_deadman); 14024 #else 14025 callout_stop(&state->dts_cleaner); 14026 callout_drain(&state->dts_cleaner); 14027 callout_stop(&state->dts_deadman); 14028 callout_drain(&state->dts_deadman); 14029 #endif 14030 14031 dtrace_dstate_fini(&vstate->dtvs_dynvars); 14032 dtrace_vstate_fini(vstate); 14033 if (state->dts_ecbs != NULL) 14034 kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *)); 14035 14036 if (state->dts_aggregations != NULL) { 14037 #ifdef DEBUG 14038 for (i = 0; i < state->dts_naggregations; i++) 14039 ASSERT(state->dts_aggregations[i] == NULL); 14040 #endif 14041 ASSERT(state->dts_naggregations > 0); 14042 kmem_free(state->dts_aggregations, 14043 state->dts_naggregations * sizeof (dtrace_aggregation_t *)); 14044 } 14045 14046 kmem_free(state->dts_buffer, bufsize); 14047 kmem_free(state->dts_aggbuffer, bufsize); 14048 14049 for (i = 0; i < nspec; i++) 14050 kmem_free(spec[i].dtsp_buffer, bufsize); 14051 14052 if (spec != NULL) 14053 kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); 14054 14055 dtrace_format_destroy(state); 14056 14057 if (state->dts_aggid_arena != NULL) { 14058 #if defined(sun) 14059 vmem_destroy(state->dts_aggid_arena); 14060 #else 14061 delete_unrhdr(state->dts_aggid_arena); 14062 #endif 14063 state->dts_aggid_arena = NULL; 14064 } 14065 #if defined(sun) 14066 ddi_soft_state_free(dtrace_softstate, minor); 14067 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); 14068 #endif 14069 } 14070 14071 /* 14072 * DTrace Anonymous Enabling Functions 14073 */ 14074 static dtrace_state_t * 14075 dtrace_anon_grab(void) 14076 { 14077 dtrace_state_t *state; 14078 14079 ASSERT(MUTEX_HELD(&dtrace_lock)); 14080 14081 if ((state = dtrace_anon.dta_state) == NULL) { 14082 ASSERT(dtrace_anon.dta_enabling == NULL); 14083 return (NULL); 14084 } 14085 14086 ASSERT(dtrace_anon.dta_enabling != NULL); 14087 ASSERT(dtrace_retained != NULL); 14088 14089 dtrace_enabling_destroy(dtrace_anon.dta_enabling); 14090 dtrace_anon.dta_enabling = NULL; 14091 dtrace_anon.dta_state = NULL; 14092 14093 return (state); 14094 } 14095 14096 static void 14097 dtrace_anon_property(void) 14098 { 14099 int i, rv; 14100 dtrace_state_t *state; 14101 dof_hdr_t *dof; 14102 char c[32]; /* enough for "dof-data-" + digits */ 14103 14104 ASSERT(MUTEX_HELD(&dtrace_lock)); 14105 ASSERT(MUTEX_HELD(&cpu_lock)); 14106 14107 for (i = 0; ; i++) { 14108 (void) snprintf(c, sizeof (c), "dof-data-%d", i); 14109 14110 dtrace_err_verbose = 1; 14111 14112 if ((dof = dtrace_dof_property(c)) == NULL) { 14113 dtrace_err_verbose = 0; 14114 break; 14115 } 14116 14117 #if defined(sun) 14118 /* 14119 * We want to create anonymous state, so we need to transition 14120 * the kernel debugger to indicate that DTrace is active. If 14121 * this fails (e.g. because the debugger has modified text in 14122 * some way), we won't continue with the processing. 14123 */ 14124 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { 14125 cmn_err(CE_NOTE, "kernel debugger active; anonymous " 14126 "enabling ignored."); 14127 dtrace_dof_destroy(dof); 14128 break; 14129 } 14130 #endif 14131 14132 /* 14133 * If we haven't allocated an anonymous state, we'll do so now. 14134 */ 14135 if ((state = dtrace_anon.dta_state) == NULL) { 14136 #if defined(sun) 14137 state = dtrace_state_create(NULL, NULL); 14138 #else 14139 state = dtrace_state_create(NULL); 14140 #endif 14141 dtrace_anon.dta_state = state; 14142 14143 if (state == NULL) { 14144 /* 14145 * This basically shouldn't happen: the only 14146 * failure mode from dtrace_state_create() is a 14147 * failure of ddi_soft_state_zalloc() that 14148 * itself should never happen. Still, the 14149 * interface allows for a failure mode, and 14150 * we want to fail as gracefully as possible: 14151 * we'll emit an error message and cease 14152 * processing anonymous state in this case. 14153 */ 14154 cmn_err(CE_WARN, "failed to create " 14155 "anonymous state"); 14156 dtrace_dof_destroy(dof); 14157 break; 14158 } 14159 } 14160 14161 rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(), 14162 &dtrace_anon.dta_enabling, 0, B_TRUE); 14163 14164 if (rv == 0) 14165 rv = dtrace_dof_options(dof, state); 14166 14167 dtrace_err_verbose = 0; 14168 dtrace_dof_destroy(dof); 14169 14170 if (rv != 0) { 14171 /* 14172 * This is malformed DOF; chuck any anonymous state 14173 * that we created. 14174 */ 14175 ASSERT(dtrace_anon.dta_enabling == NULL); 14176 dtrace_state_destroy(state); 14177 dtrace_anon.dta_state = NULL; 14178 break; 14179 } 14180 14181 ASSERT(dtrace_anon.dta_enabling != NULL); 14182 } 14183 14184 if (dtrace_anon.dta_enabling != NULL) { 14185 int rval; 14186 14187 /* 14188 * dtrace_enabling_retain() can only fail because we are 14189 * trying to retain more enablings than are allowed -- but 14190 * we only have one anonymous enabling, and we are guaranteed 14191 * to be allowed at least one retained enabling; we assert 14192 * that dtrace_enabling_retain() returns success. 14193 */ 14194 rval = dtrace_enabling_retain(dtrace_anon.dta_enabling); 14195 ASSERT(rval == 0); 14196 14197 dtrace_enabling_dump(dtrace_anon.dta_enabling); 14198 } 14199 } 14200 14201 /* 14202 * DTrace Helper Functions 14203 */ 14204 static void 14205 dtrace_helper_trace(dtrace_helper_action_t *helper, 14206 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where) 14207 { 14208 uint32_t size, next, nnext, i; 14209 dtrace_helptrace_t *ent; 14210 uint16_t flags = cpu_core[curcpu].cpuc_dtrace_flags; 14211 14212 if (!dtrace_helptrace_enabled) 14213 return; 14214 14215 ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals); 14216 14217 /* 14218 * What would a tracing framework be without its own tracing 14219 * framework? (Well, a hell of a lot simpler, for starters...) 14220 */ 14221 size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals * 14222 sizeof (uint64_t) - sizeof (uint64_t); 14223 14224 /* 14225 * Iterate until we can allocate a slot in the trace buffer. 14226 */ 14227 do { 14228 next = dtrace_helptrace_next; 14229 14230 if (next + size < dtrace_helptrace_bufsize) { 14231 nnext = next + size; 14232 } else { 14233 nnext = size; 14234 } 14235 } while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next); 14236 14237 /* 14238 * We have our slot; fill it in. 14239 */ 14240 if (nnext == size) 14241 next = 0; 14242 14243 ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next]; 14244 ent->dtht_helper = helper; 14245 ent->dtht_where = where; 14246 ent->dtht_nlocals = vstate->dtvs_nlocals; 14247 14248 ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ? 14249 mstate->dtms_fltoffs : -1; 14250 ent->dtht_fault = DTRACE_FLAGS2FLT(flags); 14251 ent->dtht_illval = cpu_core[curcpu].cpuc_dtrace_illval; 14252 14253 for (i = 0; i < vstate->dtvs_nlocals; i++) { 14254 dtrace_statvar_t *svar; 14255 14256 if ((svar = vstate->dtvs_locals[i]) == NULL) 14257 continue; 14258 14259 ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t)); 14260 ent->dtht_locals[i] = 14261 ((uint64_t *)(uintptr_t)svar->dtsv_data)[curcpu]; 14262 } 14263 } 14264 14265 static uint64_t 14266 dtrace_helper(int which, dtrace_mstate_t *mstate, 14267 dtrace_state_t *state, uint64_t arg0, uint64_t arg1) 14268 { 14269 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 14270 uint64_t sarg0 = mstate->dtms_arg[0]; 14271 uint64_t sarg1 = mstate->dtms_arg[1]; 14272 uint64_t rval = 0; 14273 dtrace_helpers_t *helpers = curproc->p_dtrace_helpers; 14274 dtrace_helper_action_t *helper; 14275 dtrace_vstate_t *vstate; 14276 dtrace_difo_t *pred; 14277 int i, trace = dtrace_helptrace_enabled; 14278 14279 ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS); 14280 14281 if (helpers == NULL) 14282 return (0); 14283 14284 if ((helper = helpers->dthps_actions[which]) == NULL) 14285 return (0); 14286 14287 vstate = &helpers->dthps_vstate; 14288 mstate->dtms_arg[0] = arg0; 14289 mstate->dtms_arg[1] = arg1; 14290 14291 /* 14292 * Now iterate over each helper. If its predicate evaluates to 'true', 14293 * we'll call the corresponding actions. Note that the below calls 14294 * to dtrace_dif_emulate() may set faults in machine state. This is 14295 * okay: our caller (the outer dtrace_dif_emulate()) will simply plow 14296 * the stored DIF offset with its own (which is the desired behavior). 14297 * Also, note the calls to dtrace_dif_emulate() may allocate scratch 14298 * from machine state; this is okay, too. 14299 */ 14300 for (; helper != NULL; helper = helper->dtha_next) { 14301 if ((pred = helper->dtha_predicate) != NULL) { 14302 if (trace) 14303 dtrace_helper_trace(helper, mstate, vstate, 0); 14304 14305 if (!dtrace_dif_emulate(pred, mstate, vstate, state)) 14306 goto next; 14307 14308 if (*flags & CPU_DTRACE_FAULT) 14309 goto err; 14310 } 14311 14312 for (i = 0; i < helper->dtha_nactions; i++) { 14313 if (trace) 14314 dtrace_helper_trace(helper, 14315 mstate, vstate, i + 1); 14316 14317 rval = dtrace_dif_emulate(helper->dtha_actions[i], 14318 mstate, vstate, state); 14319 14320 if (*flags & CPU_DTRACE_FAULT) 14321 goto err; 14322 } 14323 14324 next: 14325 if (trace) 14326 dtrace_helper_trace(helper, mstate, vstate, 14327 DTRACE_HELPTRACE_NEXT); 14328 } 14329 14330 if (trace) 14331 dtrace_helper_trace(helper, mstate, vstate, 14332 DTRACE_HELPTRACE_DONE); 14333 14334 /* 14335 * Restore the arg0 that we saved upon entry. 14336 */ 14337 mstate->dtms_arg[0] = sarg0; 14338 mstate->dtms_arg[1] = sarg1; 14339 14340 return (rval); 14341 14342 err: 14343 if (trace) 14344 dtrace_helper_trace(helper, mstate, vstate, 14345 DTRACE_HELPTRACE_ERR); 14346 14347 /* 14348 * Restore the arg0 that we saved upon entry. 14349 */ 14350 mstate->dtms_arg[0] = sarg0; 14351 mstate->dtms_arg[1] = sarg1; 14352 14353 return (0); 14354 } 14355 14356 static void 14357 dtrace_helper_action_destroy(dtrace_helper_action_t *helper, 14358 dtrace_vstate_t *vstate) 14359 { 14360 int i; 14361 14362 if (helper->dtha_predicate != NULL) 14363 dtrace_difo_release(helper->dtha_predicate, vstate); 14364 14365 for (i = 0; i < helper->dtha_nactions; i++) { 14366 ASSERT(helper->dtha_actions[i] != NULL); 14367 dtrace_difo_release(helper->dtha_actions[i], vstate); 14368 } 14369 14370 kmem_free(helper->dtha_actions, 14371 helper->dtha_nactions * sizeof (dtrace_difo_t *)); 14372 kmem_free(helper, sizeof (dtrace_helper_action_t)); 14373 } 14374 14375 static int 14376 dtrace_helper_destroygen(int gen) 14377 { 14378 proc_t *p = curproc; 14379 dtrace_helpers_t *help = p->p_dtrace_helpers; 14380 dtrace_vstate_t *vstate; 14381 int i; 14382 14383 ASSERT(MUTEX_HELD(&dtrace_lock)); 14384 14385 if (help == NULL || gen > help->dthps_generation) 14386 return (EINVAL); 14387 14388 vstate = &help->dthps_vstate; 14389 14390 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { 14391 dtrace_helper_action_t *last = NULL, *h, *next; 14392 14393 for (h = help->dthps_actions[i]; h != NULL; h = next) { 14394 next = h->dtha_next; 14395 14396 if (h->dtha_generation == gen) { 14397 if (last != NULL) { 14398 last->dtha_next = next; 14399 } else { 14400 help->dthps_actions[i] = next; 14401 } 14402 14403 dtrace_helper_action_destroy(h, vstate); 14404 } else { 14405 last = h; 14406 } 14407 } 14408 } 14409 14410 /* 14411 * Interate until we've cleared out all helper providers with the 14412 * given generation number. 14413 */ 14414 for (;;) { 14415 dtrace_helper_provider_t *prov; 14416 14417 /* 14418 * Look for a helper provider with the right generation. We 14419 * have to start back at the beginning of the list each time 14420 * because we drop dtrace_lock. It's unlikely that we'll make 14421 * more than two passes. 14422 */ 14423 for (i = 0; i < help->dthps_nprovs; i++) { 14424 prov = help->dthps_provs[i]; 14425 14426 if (prov->dthp_generation == gen) 14427 break; 14428 } 14429 14430 /* 14431 * If there were no matches, we're done. 14432 */ 14433 if (i == help->dthps_nprovs) 14434 break; 14435 14436 /* 14437 * Move the last helper provider into this slot. 14438 */ 14439 help->dthps_nprovs--; 14440 help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs]; 14441 help->dthps_provs[help->dthps_nprovs] = NULL; 14442 14443 mutex_exit(&dtrace_lock); 14444 14445 /* 14446 * If we have a meta provider, remove this helper provider. 14447 */ 14448 mutex_enter(&dtrace_meta_lock); 14449 if (dtrace_meta_pid != NULL) { 14450 ASSERT(dtrace_deferred_pid == NULL); 14451 dtrace_helper_provider_remove(&prov->dthp_prov, 14452 p->p_pid); 14453 } 14454 mutex_exit(&dtrace_meta_lock); 14455 14456 dtrace_helper_provider_destroy(prov); 14457 14458 mutex_enter(&dtrace_lock); 14459 } 14460 14461 return (0); 14462 } 14463 14464 static int 14465 dtrace_helper_validate(dtrace_helper_action_t *helper) 14466 { 14467 int err = 0, i; 14468 dtrace_difo_t *dp; 14469 14470 if ((dp = helper->dtha_predicate) != NULL) 14471 err += dtrace_difo_validate_helper(dp); 14472 14473 for (i = 0; i < helper->dtha_nactions; i++) 14474 err += dtrace_difo_validate_helper(helper->dtha_actions[i]); 14475 14476 return (err == 0); 14477 } 14478 14479 static int 14480 dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep) 14481 { 14482 dtrace_helpers_t *help; 14483 dtrace_helper_action_t *helper, *last; 14484 dtrace_actdesc_t *act; 14485 dtrace_vstate_t *vstate; 14486 dtrace_predicate_t *pred; 14487 int count = 0, nactions = 0, i; 14488 14489 if (which < 0 || which >= DTRACE_NHELPER_ACTIONS) 14490 return (EINVAL); 14491 14492 help = curproc->p_dtrace_helpers; 14493 last = help->dthps_actions[which]; 14494 vstate = &help->dthps_vstate; 14495 14496 for (count = 0; last != NULL; last = last->dtha_next) { 14497 count++; 14498 if (last->dtha_next == NULL) 14499 break; 14500 } 14501 14502 /* 14503 * If we already have dtrace_helper_actions_max helper actions for this 14504 * helper action type, we'll refuse to add a new one. 14505 */ 14506 if (count >= dtrace_helper_actions_max) 14507 return (ENOSPC); 14508 14509 helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP); 14510 helper->dtha_generation = help->dthps_generation; 14511 14512 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) { 14513 ASSERT(pred->dtp_difo != NULL); 14514 dtrace_difo_hold(pred->dtp_difo); 14515 helper->dtha_predicate = pred->dtp_difo; 14516 } 14517 14518 for (act = ep->dted_action; act != NULL; act = act->dtad_next) { 14519 if (act->dtad_kind != DTRACEACT_DIFEXPR) 14520 goto err; 14521 14522 if (act->dtad_difo == NULL) 14523 goto err; 14524 14525 nactions++; 14526 } 14527 14528 helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) * 14529 (helper->dtha_nactions = nactions), KM_SLEEP); 14530 14531 for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) { 14532 dtrace_difo_hold(act->dtad_difo); 14533 helper->dtha_actions[i++] = act->dtad_difo; 14534 } 14535 14536 if (!dtrace_helper_validate(helper)) 14537 goto err; 14538 14539 if (last == NULL) { 14540 help->dthps_actions[which] = helper; 14541 } else { 14542 last->dtha_next = helper; 14543 } 14544 14545 if (vstate->dtvs_nlocals > dtrace_helptrace_nlocals) { 14546 dtrace_helptrace_nlocals = vstate->dtvs_nlocals; 14547 dtrace_helptrace_next = 0; 14548 } 14549 14550 return (0); 14551 err: 14552 dtrace_helper_action_destroy(helper, vstate); 14553 return (EINVAL); 14554 } 14555 14556 static void 14557 dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help, 14558 dof_helper_t *dofhp) 14559 { 14560 ASSERT(MUTEX_NOT_HELD(&dtrace_lock)); 14561 14562 mutex_enter(&dtrace_meta_lock); 14563 mutex_enter(&dtrace_lock); 14564 14565 if (!dtrace_attached() || dtrace_meta_pid == NULL) { 14566 /* 14567 * If the dtrace module is loaded but not attached, or if 14568 * there aren't isn't a meta provider registered to deal with 14569 * these provider descriptions, we need to postpone creating 14570 * the actual providers until later. 14571 */ 14572 14573 if (help->dthps_next == NULL && help->dthps_prev == NULL && 14574 dtrace_deferred_pid != help) { 14575 help->dthps_deferred = 1; 14576 help->dthps_pid = p->p_pid; 14577 help->dthps_next = dtrace_deferred_pid; 14578 help->dthps_prev = NULL; 14579 if (dtrace_deferred_pid != NULL) 14580 dtrace_deferred_pid->dthps_prev = help; 14581 dtrace_deferred_pid = help; 14582 } 14583 14584 mutex_exit(&dtrace_lock); 14585 14586 } else if (dofhp != NULL) { 14587 /* 14588 * If the dtrace module is loaded and we have a particular 14589 * helper provider description, pass that off to the 14590 * meta provider. 14591 */ 14592 14593 mutex_exit(&dtrace_lock); 14594 14595 dtrace_helper_provide(dofhp, p->p_pid); 14596 14597 } else { 14598 /* 14599 * Otherwise, just pass all the helper provider descriptions 14600 * off to the meta provider. 14601 */ 14602 14603 int i; 14604 mutex_exit(&dtrace_lock); 14605 14606 for (i = 0; i < help->dthps_nprovs; i++) { 14607 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, 14608 p->p_pid); 14609 } 14610 } 14611 14612 mutex_exit(&dtrace_meta_lock); 14613 } 14614 14615 static int 14616 dtrace_helper_provider_add(dof_helper_t *dofhp, int gen) 14617 { 14618 dtrace_helpers_t *help; 14619 dtrace_helper_provider_t *hprov, **tmp_provs; 14620 uint_t tmp_maxprovs, i; 14621 14622 ASSERT(MUTEX_HELD(&dtrace_lock)); 14623 14624 help = curproc->p_dtrace_helpers; 14625 ASSERT(help != NULL); 14626 14627 /* 14628 * If we already have dtrace_helper_providers_max helper providers, 14629 * we're refuse to add a new one. 14630 */ 14631 if (help->dthps_nprovs >= dtrace_helper_providers_max) 14632 return (ENOSPC); 14633 14634 /* 14635 * Check to make sure this isn't a duplicate. 14636 */ 14637 for (i = 0; i < help->dthps_nprovs; i++) { 14638 if (dofhp->dofhp_dof == 14639 help->dthps_provs[i]->dthp_prov.dofhp_dof) 14640 return (EALREADY); 14641 } 14642 14643 hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP); 14644 hprov->dthp_prov = *dofhp; 14645 hprov->dthp_ref = 1; 14646 hprov->dthp_generation = gen; 14647 14648 /* 14649 * Allocate a bigger table for helper providers if it's already full. 14650 */ 14651 if (help->dthps_maxprovs == help->dthps_nprovs) { 14652 tmp_maxprovs = help->dthps_maxprovs; 14653 tmp_provs = help->dthps_provs; 14654 14655 if (help->dthps_maxprovs == 0) 14656 help->dthps_maxprovs = 2; 14657 else 14658 help->dthps_maxprovs *= 2; 14659 if (help->dthps_maxprovs > dtrace_helper_providers_max) 14660 help->dthps_maxprovs = dtrace_helper_providers_max; 14661 14662 ASSERT(tmp_maxprovs < help->dthps_maxprovs); 14663 14664 help->dthps_provs = kmem_zalloc(help->dthps_maxprovs * 14665 sizeof (dtrace_helper_provider_t *), KM_SLEEP); 14666 14667 if (tmp_provs != NULL) { 14668 bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs * 14669 sizeof (dtrace_helper_provider_t *)); 14670 kmem_free(tmp_provs, tmp_maxprovs * 14671 sizeof (dtrace_helper_provider_t *)); 14672 } 14673 } 14674 14675 help->dthps_provs[help->dthps_nprovs] = hprov; 14676 help->dthps_nprovs++; 14677 14678 return (0); 14679 } 14680 14681 static void 14682 dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov) 14683 { 14684 mutex_enter(&dtrace_lock); 14685 14686 if (--hprov->dthp_ref == 0) { 14687 dof_hdr_t *dof; 14688 mutex_exit(&dtrace_lock); 14689 dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof; 14690 dtrace_dof_destroy(dof); 14691 kmem_free(hprov, sizeof (dtrace_helper_provider_t)); 14692 } else { 14693 mutex_exit(&dtrace_lock); 14694 } 14695 } 14696 14697 static int 14698 dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec) 14699 { 14700 uintptr_t daddr = (uintptr_t)dof; 14701 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; 14702 dof_provider_t *provider; 14703 dof_probe_t *probe; 14704 uint8_t *arg; 14705 char *strtab, *typestr; 14706 dof_stridx_t typeidx; 14707 size_t typesz; 14708 uint_t nprobes, j, k; 14709 14710 ASSERT(sec->dofs_type == DOF_SECT_PROVIDER); 14711 14712 if (sec->dofs_offset & (sizeof (uint_t) - 1)) { 14713 dtrace_dof_error(dof, "misaligned section offset"); 14714 return (-1); 14715 } 14716 14717 /* 14718 * The section needs to be large enough to contain the DOF provider 14719 * structure appropriate for the given version. 14720 */ 14721 if (sec->dofs_size < 14722 ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ? 14723 offsetof(dof_provider_t, dofpv_prenoffs) : 14724 sizeof (dof_provider_t))) { 14725 dtrace_dof_error(dof, "provider section too small"); 14726 return (-1); 14727 } 14728 14729 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); 14730 str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab); 14731 prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes); 14732 arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs); 14733 off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs); 14734 14735 if (str_sec == NULL || prb_sec == NULL || 14736 arg_sec == NULL || off_sec == NULL) 14737 return (-1); 14738 14739 enoff_sec = NULL; 14740 14741 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && 14742 provider->dofpv_prenoffs != DOF_SECT_NONE && 14743 (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS, 14744 provider->dofpv_prenoffs)) == NULL) 14745 return (-1); 14746 14747 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); 14748 14749 if (provider->dofpv_name >= str_sec->dofs_size || 14750 strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) { 14751 dtrace_dof_error(dof, "invalid provider name"); 14752 return (-1); 14753 } 14754 14755 if (prb_sec->dofs_entsize == 0 || 14756 prb_sec->dofs_entsize > prb_sec->dofs_size) { 14757 dtrace_dof_error(dof, "invalid entry size"); 14758 return (-1); 14759 } 14760 14761 if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) { 14762 dtrace_dof_error(dof, "misaligned entry size"); 14763 return (-1); 14764 } 14765 14766 if (off_sec->dofs_entsize != sizeof (uint32_t)) { 14767 dtrace_dof_error(dof, "invalid entry size"); 14768 return (-1); 14769 } 14770 14771 if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) { 14772 dtrace_dof_error(dof, "misaligned section offset"); 14773 return (-1); 14774 } 14775 14776 if (arg_sec->dofs_entsize != sizeof (uint8_t)) { 14777 dtrace_dof_error(dof, "invalid entry size"); 14778 return (-1); 14779 } 14780 14781 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); 14782 14783 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; 14784 14785 /* 14786 * Take a pass through the probes to check for errors. 14787 */ 14788 for (j = 0; j < nprobes; j++) { 14789 probe = (dof_probe_t *)(uintptr_t)(daddr + 14790 prb_sec->dofs_offset + j * prb_sec->dofs_entsize); 14791 14792 if (probe->dofpr_func >= str_sec->dofs_size) { 14793 dtrace_dof_error(dof, "invalid function name"); 14794 return (-1); 14795 } 14796 14797 if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) { 14798 dtrace_dof_error(dof, "function name too long"); 14799 return (-1); 14800 } 14801 14802 if (probe->dofpr_name >= str_sec->dofs_size || 14803 strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) { 14804 dtrace_dof_error(dof, "invalid probe name"); 14805 return (-1); 14806 } 14807 14808 /* 14809 * The offset count must not wrap the index, and the offsets 14810 * must also not overflow the section's data. 14811 */ 14812 if (probe->dofpr_offidx + probe->dofpr_noffs < 14813 probe->dofpr_offidx || 14814 (probe->dofpr_offidx + probe->dofpr_noffs) * 14815 off_sec->dofs_entsize > off_sec->dofs_size) { 14816 dtrace_dof_error(dof, "invalid probe offset"); 14817 return (-1); 14818 } 14819 14820 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) { 14821 /* 14822 * If there's no is-enabled offset section, make sure 14823 * there aren't any is-enabled offsets. Otherwise 14824 * perform the same checks as for probe offsets 14825 * (immediately above). 14826 */ 14827 if (enoff_sec == NULL) { 14828 if (probe->dofpr_enoffidx != 0 || 14829 probe->dofpr_nenoffs != 0) { 14830 dtrace_dof_error(dof, "is-enabled " 14831 "offsets with null section"); 14832 return (-1); 14833 } 14834 } else if (probe->dofpr_enoffidx + 14835 probe->dofpr_nenoffs < probe->dofpr_enoffidx || 14836 (probe->dofpr_enoffidx + probe->dofpr_nenoffs) * 14837 enoff_sec->dofs_entsize > enoff_sec->dofs_size) { 14838 dtrace_dof_error(dof, "invalid is-enabled " 14839 "offset"); 14840 return (-1); 14841 } 14842 14843 if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) { 14844 dtrace_dof_error(dof, "zero probe and " 14845 "is-enabled offsets"); 14846 return (-1); 14847 } 14848 } else if (probe->dofpr_noffs == 0) { 14849 dtrace_dof_error(dof, "zero probe offsets"); 14850 return (-1); 14851 } 14852 14853 if (probe->dofpr_argidx + probe->dofpr_xargc < 14854 probe->dofpr_argidx || 14855 (probe->dofpr_argidx + probe->dofpr_xargc) * 14856 arg_sec->dofs_entsize > arg_sec->dofs_size) { 14857 dtrace_dof_error(dof, "invalid args"); 14858 return (-1); 14859 } 14860 14861 typeidx = probe->dofpr_nargv; 14862 typestr = strtab + probe->dofpr_nargv; 14863 for (k = 0; k < probe->dofpr_nargc; k++) { 14864 if (typeidx >= str_sec->dofs_size) { 14865 dtrace_dof_error(dof, "bad " 14866 "native argument type"); 14867 return (-1); 14868 } 14869 14870 typesz = strlen(typestr) + 1; 14871 if (typesz > DTRACE_ARGTYPELEN) { 14872 dtrace_dof_error(dof, "native " 14873 "argument type too long"); 14874 return (-1); 14875 } 14876 typeidx += typesz; 14877 typestr += typesz; 14878 } 14879 14880 typeidx = probe->dofpr_xargv; 14881 typestr = strtab + probe->dofpr_xargv; 14882 for (k = 0; k < probe->dofpr_xargc; k++) { 14883 if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) { 14884 dtrace_dof_error(dof, "bad " 14885 "native argument index"); 14886 return (-1); 14887 } 14888 14889 if (typeidx >= str_sec->dofs_size) { 14890 dtrace_dof_error(dof, "bad " 14891 "translated argument type"); 14892 return (-1); 14893 } 14894 14895 typesz = strlen(typestr) + 1; 14896 if (typesz > DTRACE_ARGTYPELEN) { 14897 dtrace_dof_error(dof, "translated argument " 14898 "type too long"); 14899 return (-1); 14900 } 14901 14902 typeidx += typesz; 14903 typestr += typesz; 14904 } 14905 } 14906 14907 return (0); 14908 } 14909 14910 static int 14911 dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp) 14912 { 14913 dtrace_helpers_t *help; 14914 dtrace_vstate_t *vstate; 14915 dtrace_enabling_t *enab = NULL; 14916 int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1; 14917 uintptr_t daddr = (uintptr_t)dof; 14918 14919 ASSERT(MUTEX_HELD(&dtrace_lock)); 14920 14921 if ((help = curproc->p_dtrace_helpers) == NULL) 14922 help = dtrace_helpers_create(curproc); 14923 14924 vstate = &help->dthps_vstate; 14925 14926 if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab, 14927 dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) { 14928 dtrace_dof_destroy(dof); 14929 return (rv); 14930 } 14931 14932 /* 14933 * Look for helper providers and validate their descriptions. 14934 */ 14935 if (dhp != NULL) { 14936 for (i = 0; i < dof->dofh_secnum; i++) { 14937 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + 14938 dof->dofh_secoff + i * dof->dofh_secsize); 14939 14940 if (sec->dofs_type != DOF_SECT_PROVIDER) 14941 continue; 14942 14943 if (dtrace_helper_provider_validate(dof, sec) != 0) { 14944 dtrace_enabling_destroy(enab); 14945 dtrace_dof_destroy(dof); 14946 return (-1); 14947 } 14948 14949 nprovs++; 14950 } 14951 } 14952 14953 /* 14954 * Now we need to walk through the ECB descriptions in the enabling. 14955 */ 14956 for (i = 0; i < enab->dten_ndesc; i++) { 14957 dtrace_ecbdesc_t *ep = enab->dten_desc[i]; 14958 dtrace_probedesc_t *desc = &ep->dted_probe; 14959 14960 if (strcmp(desc->dtpd_provider, "dtrace") != 0) 14961 continue; 14962 14963 if (strcmp(desc->dtpd_mod, "helper") != 0) 14964 continue; 14965 14966 if (strcmp(desc->dtpd_func, "ustack") != 0) 14967 continue; 14968 14969 if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK, 14970 ep)) != 0) { 14971 /* 14972 * Adding this helper action failed -- we are now going 14973 * to rip out the entire generation and return failure. 14974 */ 14975 (void) dtrace_helper_destroygen(help->dthps_generation); 14976 dtrace_enabling_destroy(enab); 14977 dtrace_dof_destroy(dof); 14978 return (-1); 14979 } 14980 14981 nhelpers++; 14982 } 14983 14984 if (nhelpers < enab->dten_ndesc) 14985 dtrace_dof_error(dof, "unmatched helpers"); 14986 14987 gen = help->dthps_generation++; 14988 dtrace_enabling_destroy(enab); 14989 14990 if (dhp != NULL && nprovs > 0) { 14991 dhp->dofhp_dof = (uint64_t)(uintptr_t)dof; 14992 if (dtrace_helper_provider_add(dhp, gen) == 0) { 14993 mutex_exit(&dtrace_lock); 14994 dtrace_helper_provider_register(curproc, help, dhp); 14995 mutex_enter(&dtrace_lock); 14996 14997 destroy = 0; 14998 } 14999 } 15000 15001 if (destroy) 15002 dtrace_dof_destroy(dof); 15003 15004 return (gen); 15005 } 15006 15007 static dtrace_helpers_t * 15008 dtrace_helpers_create(proc_t *p) 15009 { 15010 dtrace_helpers_t *help; 15011 15012 ASSERT(MUTEX_HELD(&dtrace_lock)); 15013 ASSERT(p->p_dtrace_helpers == NULL); 15014 15015 help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP); 15016 help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) * 15017 DTRACE_NHELPER_ACTIONS, KM_SLEEP); 15018 15019 p->p_dtrace_helpers = help; 15020 dtrace_helpers++; 15021 15022 return (help); 15023 } 15024 15025 #if defined(sun) 15026 static 15027 #endif 15028 void 15029 dtrace_helpers_destroy(proc_t *p) 15030 { 15031 dtrace_helpers_t *help; 15032 dtrace_vstate_t *vstate; 15033 #if defined(sun) 15034 proc_t *p = curproc; 15035 #endif 15036 int i; 15037 15038 mutex_enter(&dtrace_lock); 15039 15040 ASSERT(p->p_dtrace_helpers != NULL); 15041 ASSERT(dtrace_helpers > 0); 15042 15043 help = p->p_dtrace_helpers; 15044 vstate = &help->dthps_vstate; 15045 15046 /* 15047 * We're now going to lose the help from this process. 15048 */ 15049 p->p_dtrace_helpers = NULL; 15050 dtrace_sync(); 15051 15052 /* 15053 * Destory the helper actions. 15054 */ 15055 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { 15056 dtrace_helper_action_t *h, *next; 15057 15058 for (h = help->dthps_actions[i]; h != NULL; h = next) { 15059 next = h->dtha_next; 15060 dtrace_helper_action_destroy(h, vstate); 15061 h = next; 15062 } 15063 } 15064 15065 mutex_exit(&dtrace_lock); 15066 15067 /* 15068 * Destroy the helper providers. 15069 */ 15070 if (help->dthps_maxprovs > 0) { 15071 mutex_enter(&dtrace_meta_lock); 15072 if (dtrace_meta_pid != NULL) { 15073 ASSERT(dtrace_deferred_pid == NULL); 15074 15075 for (i = 0; i < help->dthps_nprovs; i++) { 15076 dtrace_helper_provider_remove( 15077 &help->dthps_provs[i]->dthp_prov, p->p_pid); 15078 } 15079 } else { 15080 mutex_enter(&dtrace_lock); 15081 ASSERT(help->dthps_deferred == 0 || 15082 help->dthps_next != NULL || 15083 help->dthps_prev != NULL || 15084 help == dtrace_deferred_pid); 15085 15086 /* 15087 * Remove the helper from the deferred list. 15088 */ 15089 if (help->dthps_next != NULL) 15090 help->dthps_next->dthps_prev = help->dthps_prev; 15091 if (help->dthps_prev != NULL) 15092 help->dthps_prev->dthps_next = help->dthps_next; 15093 if (dtrace_deferred_pid == help) { 15094 dtrace_deferred_pid = help->dthps_next; 15095 ASSERT(help->dthps_prev == NULL); 15096 } 15097 15098 mutex_exit(&dtrace_lock); 15099 } 15100 15101 mutex_exit(&dtrace_meta_lock); 15102 15103 for (i = 0; i < help->dthps_nprovs; i++) { 15104 dtrace_helper_provider_destroy(help->dthps_provs[i]); 15105 } 15106 15107 kmem_free(help->dthps_provs, help->dthps_maxprovs * 15108 sizeof (dtrace_helper_provider_t *)); 15109 } 15110 15111 mutex_enter(&dtrace_lock); 15112 15113 dtrace_vstate_fini(&help->dthps_vstate); 15114 kmem_free(help->dthps_actions, 15115 sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS); 15116 kmem_free(help, sizeof (dtrace_helpers_t)); 15117 15118 --dtrace_helpers; 15119 mutex_exit(&dtrace_lock); 15120 } 15121 15122 #if defined(sun) 15123 static 15124 #endif 15125 void 15126 dtrace_helpers_duplicate(proc_t *from, proc_t *to) 15127 { 15128 dtrace_helpers_t *help, *newhelp; 15129 dtrace_helper_action_t *helper, *new, *last; 15130 dtrace_difo_t *dp; 15131 dtrace_vstate_t *vstate; 15132 int i, j, sz, hasprovs = 0; 15133 15134 mutex_enter(&dtrace_lock); 15135 ASSERT(from->p_dtrace_helpers != NULL); 15136 ASSERT(dtrace_helpers > 0); 15137 15138 help = from->p_dtrace_helpers; 15139 newhelp = dtrace_helpers_create(to); 15140 ASSERT(to->p_dtrace_helpers != NULL); 15141 15142 newhelp->dthps_generation = help->dthps_generation; 15143 vstate = &newhelp->dthps_vstate; 15144 15145 /* 15146 * Duplicate the helper actions. 15147 */ 15148 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { 15149 if ((helper = help->dthps_actions[i]) == NULL) 15150 continue; 15151 15152 for (last = NULL; helper != NULL; helper = helper->dtha_next) { 15153 new = kmem_zalloc(sizeof (dtrace_helper_action_t), 15154 KM_SLEEP); 15155 new->dtha_generation = helper->dtha_generation; 15156 15157 if ((dp = helper->dtha_predicate) != NULL) { 15158 dp = dtrace_difo_duplicate(dp, vstate); 15159 new->dtha_predicate = dp; 15160 } 15161 15162 new->dtha_nactions = helper->dtha_nactions; 15163 sz = sizeof (dtrace_difo_t *) * new->dtha_nactions; 15164 new->dtha_actions = kmem_alloc(sz, KM_SLEEP); 15165 15166 for (j = 0; j < new->dtha_nactions; j++) { 15167 dtrace_difo_t *dp = helper->dtha_actions[j]; 15168 15169 ASSERT(dp != NULL); 15170 dp = dtrace_difo_duplicate(dp, vstate); 15171 new->dtha_actions[j] = dp; 15172 } 15173 15174 if (last != NULL) { 15175 last->dtha_next = new; 15176 } else { 15177 newhelp->dthps_actions[i] = new; 15178 } 15179 15180 last = new; 15181 } 15182 } 15183 15184 /* 15185 * Duplicate the helper providers and register them with the 15186 * DTrace framework. 15187 */ 15188 if (help->dthps_nprovs > 0) { 15189 newhelp->dthps_nprovs = help->dthps_nprovs; 15190 newhelp->dthps_maxprovs = help->dthps_nprovs; 15191 newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs * 15192 sizeof (dtrace_helper_provider_t *), KM_SLEEP); 15193 for (i = 0; i < newhelp->dthps_nprovs; i++) { 15194 newhelp->dthps_provs[i] = help->dthps_provs[i]; 15195 newhelp->dthps_provs[i]->dthp_ref++; 15196 } 15197 15198 hasprovs = 1; 15199 } 15200 15201 mutex_exit(&dtrace_lock); 15202 15203 if (hasprovs) 15204 dtrace_helper_provider_register(to, newhelp, NULL); 15205 } 15206 15207 /* 15208 * DTrace Hook Functions 15209 */ 15210 static void 15211 dtrace_module_loaded(modctl_t *ctl) 15212 { 15213 dtrace_provider_t *prv; 15214 15215 mutex_enter(&dtrace_provider_lock); 15216 #if defined(sun) 15217 mutex_enter(&mod_lock); 15218 #endif 15219 15220 #if defined(sun) 15221 ASSERT(ctl->mod_busy); 15222 #endif 15223 15224 /* 15225 * We're going to call each providers per-module provide operation 15226 * specifying only this module. 15227 */ 15228 for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next) 15229 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); 15230 15231 #if defined(sun) 15232 mutex_exit(&mod_lock); 15233 #endif 15234 mutex_exit(&dtrace_provider_lock); 15235 15236 /* 15237 * If we have any retained enablings, we need to match against them. 15238 * Enabling probes requires that cpu_lock be held, and we cannot hold 15239 * cpu_lock here -- it is legal for cpu_lock to be held when loading a 15240 * module. (In particular, this happens when loading scheduling 15241 * classes.) So if we have any retained enablings, we need to dispatch 15242 * our task queue to do the match for us. 15243 */ 15244 mutex_enter(&dtrace_lock); 15245 15246 if (dtrace_retained == NULL) { 15247 mutex_exit(&dtrace_lock); 15248 return; 15249 } 15250 15251 (void) taskq_dispatch(dtrace_taskq, 15252 (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP); 15253 15254 mutex_exit(&dtrace_lock); 15255 15256 /* 15257 * And now, for a little heuristic sleaze: in general, we want to 15258 * match modules as soon as they load. However, we cannot guarantee 15259 * this, because it would lead us to the lock ordering violation 15260 * outlined above. The common case, of course, is that cpu_lock is 15261 * _not_ held -- so we delay here for a clock tick, hoping that that's 15262 * long enough for the task queue to do its work. If it's not, it's 15263 * not a serious problem -- it just means that the module that we 15264 * just loaded may not be immediately instrumentable. 15265 */ 15266 delay(1); 15267 } 15268 15269 static void 15270 #if defined(sun) 15271 dtrace_module_unloaded(modctl_t *ctl) 15272 #else 15273 dtrace_module_unloaded(modctl_t *ctl, int *error) 15274 #endif 15275 { 15276 dtrace_probe_t template, *probe, *first, *next; 15277 dtrace_provider_t *prov; 15278 #if !defined(sun) 15279 char modname[DTRACE_MODNAMELEN]; 15280 size_t len; 15281 #endif 15282 15283 #if defined(sun) 15284 template.dtpr_mod = ctl->mod_modname; 15285 #else 15286 /* Handle the fact that ctl->filename may end in ".ko". */ 15287 strlcpy(modname, ctl->filename, sizeof(modname)); 15288 len = strlen(ctl->filename); 15289 if (len > 3 && strcmp(modname + len - 3, ".ko") == 0) 15290 modname[len - 3] = '\0'; 15291 template.dtpr_mod = modname; 15292 #endif 15293 15294 mutex_enter(&dtrace_provider_lock); 15295 #if defined(sun) 15296 mutex_enter(&mod_lock); 15297 #endif 15298 mutex_enter(&dtrace_lock); 15299 15300 #if !defined(sun) 15301 if (ctl->nenabled > 0) { 15302 /* Don't allow unloads if a probe is enabled. */ 15303 mutex_exit(&dtrace_provider_lock); 15304 mutex_exit(&dtrace_lock); 15305 *error = -1; 15306 printf( 15307 "kldunload: attempt to unload module that has DTrace probes enabled\n"); 15308 return; 15309 } 15310 #endif 15311 15312 if (dtrace_bymod == NULL) { 15313 /* 15314 * The DTrace module is loaded (obviously) but not attached; 15315 * we don't have any work to do. 15316 */ 15317 mutex_exit(&dtrace_provider_lock); 15318 #if defined(sun) 15319 mutex_exit(&mod_lock); 15320 #endif 15321 mutex_exit(&dtrace_lock); 15322 return; 15323 } 15324 15325 for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template); 15326 probe != NULL; probe = probe->dtpr_nextmod) { 15327 if (probe->dtpr_ecb != NULL) { 15328 mutex_exit(&dtrace_provider_lock); 15329 #if defined(sun) 15330 mutex_exit(&mod_lock); 15331 #endif 15332 mutex_exit(&dtrace_lock); 15333 15334 /* 15335 * This shouldn't _actually_ be possible -- we're 15336 * unloading a module that has an enabled probe in it. 15337 * (It's normally up to the provider to make sure that 15338 * this can't happen.) However, because dtps_enable() 15339 * doesn't have a failure mode, there can be an 15340 * enable/unload race. Upshot: we don't want to 15341 * assert, but we're not going to disable the 15342 * probe, either. 15343 */ 15344 if (dtrace_err_verbose) { 15345 #if defined(sun) 15346 cmn_err(CE_WARN, "unloaded module '%s' had " 15347 "enabled probes", ctl->mod_modname); 15348 #else 15349 cmn_err(CE_WARN, "unloaded module '%s' had " 15350 "enabled probes", modname); 15351 #endif 15352 } 15353 15354 return; 15355 } 15356 } 15357 15358 probe = first; 15359 15360 for (first = NULL; probe != NULL; probe = next) { 15361 ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe); 15362 15363 dtrace_probes[probe->dtpr_id - 1] = NULL; 15364 15365 next = probe->dtpr_nextmod; 15366 dtrace_hash_remove(dtrace_bymod, probe); 15367 dtrace_hash_remove(dtrace_byfunc, probe); 15368 dtrace_hash_remove(dtrace_byname, probe); 15369 15370 if (first == NULL) { 15371 first = probe; 15372 probe->dtpr_nextmod = NULL; 15373 } else { 15374 probe->dtpr_nextmod = first; 15375 first = probe; 15376 } 15377 } 15378 15379 /* 15380 * We've removed all of the module's probes from the hash chains and 15381 * from the probe array. Now issue a dtrace_sync() to be sure that 15382 * everyone has cleared out from any probe array processing. 15383 */ 15384 dtrace_sync(); 15385 15386 for (probe = first; probe != NULL; probe = first) { 15387 first = probe->dtpr_nextmod; 15388 prov = probe->dtpr_provider; 15389 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id, 15390 probe->dtpr_arg); 15391 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); 15392 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); 15393 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); 15394 #if defined(sun) 15395 vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1); 15396 #else 15397 free_unr(dtrace_arena, probe->dtpr_id); 15398 #endif 15399 kmem_free(probe, sizeof (dtrace_probe_t)); 15400 } 15401 15402 mutex_exit(&dtrace_lock); 15403 #if defined(sun) 15404 mutex_exit(&mod_lock); 15405 #endif 15406 mutex_exit(&dtrace_provider_lock); 15407 } 15408 15409 #if !defined(sun) 15410 static void 15411 dtrace_kld_load(void *arg __unused, linker_file_t lf) 15412 { 15413 15414 dtrace_module_loaded(lf); 15415 } 15416 15417 static void 15418 dtrace_kld_unload_try(void *arg __unused, linker_file_t lf, int *error) 15419 { 15420 15421 if (*error != 0) 15422 /* We already have an error, so don't do anything. */ 15423 return; 15424 dtrace_module_unloaded(lf, error); 15425 } 15426 #endif 15427 15428 #if defined(sun) 15429 static void 15430 dtrace_suspend(void) 15431 { 15432 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend)); 15433 } 15434 15435 static void 15436 dtrace_resume(void) 15437 { 15438 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume)); 15439 } 15440 #endif 15441 15442 static int 15443 dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu) 15444 { 15445 ASSERT(MUTEX_HELD(&cpu_lock)); 15446 mutex_enter(&dtrace_lock); 15447 15448 switch (what) { 15449 case CPU_CONFIG: { 15450 dtrace_state_t *state; 15451 dtrace_optval_t *opt, rs, c; 15452 15453 /* 15454 * For now, we only allocate a new buffer for anonymous state. 15455 */ 15456 if ((state = dtrace_anon.dta_state) == NULL) 15457 break; 15458 15459 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) 15460 break; 15461 15462 opt = state->dts_options; 15463 c = opt[DTRACEOPT_CPU]; 15464 15465 if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu) 15466 break; 15467 15468 /* 15469 * Regardless of what the actual policy is, we're going to 15470 * temporarily set our resize policy to be manual. We're 15471 * also going to temporarily set our CPU option to denote 15472 * the newly configured CPU. 15473 */ 15474 rs = opt[DTRACEOPT_BUFRESIZE]; 15475 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL; 15476 opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu; 15477 15478 (void) dtrace_state_buffers(state); 15479 15480 opt[DTRACEOPT_BUFRESIZE] = rs; 15481 opt[DTRACEOPT_CPU] = c; 15482 15483 break; 15484 } 15485 15486 case CPU_UNCONFIG: 15487 /* 15488 * We don't free the buffer in the CPU_UNCONFIG case. (The 15489 * buffer will be freed when the consumer exits.) 15490 */ 15491 break; 15492 15493 default: 15494 break; 15495 } 15496 15497 mutex_exit(&dtrace_lock); 15498 return (0); 15499 } 15500 15501 #if defined(sun) 15502 static void 15503 dtrace_cpu_setup_initial(processorid_t cpu) 15504 { 15505 (void) dtrace_cpu_setup(CPU_CONFIG, cpu); 15506 } 15507 #endif 15508 15509 static void 15510 dtrace_toxrange_add(uintptr_t base, uintptr_t limit) 15511 { 15512 if (dtrace_toxranges >= dtrace_toxranges_max) { 15513 int osize, nsize; 15514 dtrace_toxrange_t *range; 15515 15516 osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); 15517 15518 if (osize == 0) { 15519 ASSERT(dtrace_toxrange == NULL); 15520 ASSERT(dtrace_toxranges_max == 0); 15521 dtrace_toxranges_max = 1; 15522 } else { 15523 dtrace_toxranges_max <<= 1; 15524 } 15525 15526 nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); 15527 range = kmem_zalloc(nsize, KM_SLEEP); 15528 15529 if (dtrace_toxrange != NULL) { 15530 ASSERT(osize != 0); 15531 bcopy(dtrace_toxrange, range, osize); 15532 kmem_free(dtrace_toxrange, osize); 15533 } 15534 15535 dtrace_toxrange = range; 15536 } 15537 15538 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == 0); 15539 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == 0); 15540 15541 dtrace_toxrange[dtrace_toxranges].dtt_base = base; 15542 dtrace_toxrange[dtrace_toxranges].dtt_limit = limit; 15543 dtrace_toxranges++; 15544 } 15545 15546 /* 15547 * DTrace Driver Cookbook Functions 15548 */ 15549 #if defined(sun) 15550 /*ARGSUSED*/ 15551 static int 15552 dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) 15553 { 15554 dtrace_provider_id_t id; 15555 dtrace_state_t *state = NULL; 15556 dtrace_enabling_t *enab; 15557 15558 mutex_enter(&cpu_lock); 15559 mutex_enter(&dtrace_provider_lock); 15560 mutex_enter(&dtrace_lock); 15561 15562 if (ddi_soft_state_init(&dtrace_softstate, 15563 sizeof (dtrace_state_t), 0) != 0) { 15564 cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state"); 15565 mutex_exit(&cpu_lock); 15566 mutex_exit(&dtrace_provider_lock); 15567 mutex_exit(&dtrace_lock); 15568 return (DDI_FAILURE); 15569 } 15570 15571 if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR, 15572 DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE || 15573 ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR, 15574 DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) { 15575 cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes"); 15576 ddi_remove_minor_node(devi, NULL); 15577 ddi_soft_state_fini(&dtrace_softstate); 15578 mutex_exit(&cpu_lock); 15579 mutex_exit(&dtrace_provider_lock); 15580 mutex_exit(&dtrace_lock); 15581 return (DDI_FAILURE); 15582 } 15583 15584 ddi_report_dev(devi); 15585 dtrace_devi = devi; 15586 15587 dtrace_modload = dtrace_module_loaded; 15588 dtrace_modunload = dtrace_module_unloaded; 15589 dtrace_cpu_init = dtrace_cpu_setup_initial; 15590 dtrace_helpers_cleanup = dtrace_helpers_destroy; 15591 dtrace_helpers_fork = dtrace_helpers_duplicate; 15592 dtrace_cpustart_init = dtrace_suspend; 15593 dtrace_cpustart_fini = dtrace_resume; 15594 dtrace_debugger_init = dtrace_suspend; 15595 dtrace_debugger_fini = dtrace_resume; 15596 15597 register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); 15598 15599 ASSERT(MUTEX_HELD(&cpu_lock)); 15600 15601 dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1, 15602 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); 15603 dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE, 15604 UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0, 15605 VM_SLEEP | VMC_IDENTIFIER); 15606 dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri, 15607 1, INT_MAX, 0); 15608 15609 dtrace_state_cache = kmem_cache_create("dtrace_state_cache", 15610 sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN, 15611 NULL, NULL, NULL, NULL, NULL, 0); 15612 15613 ASSERT(MUTEX_HELD(&cpu_lock)); 15614 dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod), 15615 offsetof(dtrace_probe_t, dtpr_nextmod), 15616 offsetof(dtrace_probe_t, dtpr_prevmod)); 15617 15618 dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func), 15619 offsetof(dtrace_probe_t, dtpr_nextfunc), 15620 offsetof(dtrace_probe_t, dtpr_prevfunc)); 15621 15622 dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name), 15623 offsetof(dtrace_probe_t, dtpr_nextname), 15624 offsetof(dtrace_probe_t, dtpr_prevname)); 15625 15626 if (dtrace_retain_max < 1) { 15627 cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; " 15628 "setting to 1", dtrace_retain_max); 15629 dtrace_retain_max = 1; 15630 } 15631 15632 /* 15633 * Now discover our toxic ranges. 15634 */ 15635 dtrace_toxic_ranges(dtrace_toxrange_add); 15636 15637 /* 15638 * Before we register ourselves as a provider to our own framework, 15639 * we would like to assert that dtrace_provider is NULL -- but that's 15640 * not true if we were loaded as a dependency of a DTrace provider. 15641 * Once we've registered, we can assert that dtrace_provider is our 15642 * pseudo provider. 15643 */ 15644 (void) dtrace_register("dtrace", &dtrace_provider_attr, 15645 DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id); 15646 15647 ASSERT(dtrace_provider != NULL); 15648 ASSERT((dtrace_provider_id_t)dtrace_provider == id); 15649 15650 dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t) 15651 dtrace_provider, NULL, NULL, "BEGIN", 0, NULL); 15652 dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t) 15653 dtrace_provider, NULL, NULL, "END", 0, NULL); 15654 dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t) 15655 dtrace_provider, NULL, NULL, "ERROR", 1, NULL); 15656 15657 dtrace_anon_property(); 15658 mutex_exit(&cpu_lock); 15659 15660 /* 15661 * If DTrace helper tracing is enabled, we need to allocate the 15662 * trace buffer and initialize the values. 15663 */ 15664 if (dtrace_helptrace_enabled) { 15665 ASSERT(dtrace_helptrace_buffer == NULL); 15666 dtrace_helptrace_buffer = 15667 kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP); 15668 dtrace_helptrace_next = 0; 15669 } 15670 15671 /* 15672 * If there are already providers, we must ask them to provide their 15673 * probes, and then match any anonymous enabling against them. Note 15674 * that there should be no other retained enablings at this time: 15675 * the only retained enablings at this time should be the anonymous 15676 * enabling. 15677 */ 15678 if (dtrace_anon.dta_enabling != NULL) { 15679 ASSERT(dtrace_retained == dtrace_anon.dta_enabling); 15680 15681 dtrace_enabling_provide(NULL); 15682 state = dtrace_anon.dta_state; 15683 15684 /* 15685 * We couldn't hold cpu_lock across the above call to 15686 * dtrace_enabling_provide(), but we must hold it to actually 15687 * enable the probes. We have to drop all of our locks, pick 15688 * up cpu_lock, and regain our locks before matching the 15689 * retained anonymous enabling. 15690 */ 15691 mutex_exit(&dtrace_lock); 15692 mutex_exit(&dtrace_provider_lock); 15693 15694 mutex_enter(&cpu_lock); 15695 mutex_enter(&dtrace_provider_lock); 15696 mutex_enter(&dtrace_lock); 15697 15698 if ((enab = dtrace_anon.dta_enabling) != NULL) 15699 (void) dtrace_enabling_match(enab, NULL); 15700 15701 mutex_exit(&cpu_lock); 15702 } 15703 15704 mutex_exit(&dtrace_lock); 15705 mutex_exit(&dtrace_provider_lock); 15706 15707 if (state != NULL) { 15708 /* 15709 * If we created any anonymous state, set it going now. 15710 */ 15711 (void) dtrace_state_go(state, &dtrace_anon.dta_beganon); 15712 } 15713 15714 return (DDI_SUCCESS); 15715 } 15716 #endif 15717 15718 #if !defined(sun) 15719 #if __FreeBSD_version >= 800039 15720 static void dtrace_dtr(void *); 15721 #endif 15722 #endif 15723 15724 /*ARGSUSED*/ 15725 static int 15726 #if defined(sun) 15727 dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) 15728 #else 15729 dtrace_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 15730 #endif 15731 { 15732 dtrace_state_t *state; 15733 uint32_t priv; 15734 uid_t uid; 15735 zoneid_t zoneid; 15736 15737 #if defined(sun) 15738 if (getminor(*devp) == DTRACEMNRN_HELPER) 15739 return (0); 15740 15741 /* 15742 * If this wasn't an open with the "helper" minor, then it must be 15743 * the "dtrace" minor. 15744 */ 15745 ASSERT(getminor(*devp) == DTRACEMNRN_DTRACE); 15746 #else 15747 cred_t *cred_p = NULL; 15748 15749 #if __FreeBSD_version < 800039 15750 /* 15751 * The first minor device is the one that is cloned so there is 15752 * nothing more to do here. 15753 */ 15754 if (dev2unit(dev) == 0) 15755 return 0; 15756 15757 /* 15758 * Devices are cloned, so if the DTrace state has already 15759 * been allocated, that means this device belongs to a 15760 * different client. Each client should open '/dev/dtrace' 15761 * to get a cloned device. 15762 */ 15763 if (dev->si_drv1 != NULL) 15764 return (EBUSY); 15765 #endif 15766 15767 cred_p = dev->si_cred; 15768 #endif 15769 15770 /* 15771 * If no DTRACE_PRIV_* bits are set in the credential, then the 15772 * caller lacks sufficient permission to do anything with DTrace. 15773 */ 15774 dtrace_cred2priv(cred_p, &priv, &uid, &zoneid); 15775 if (priv == DTRACE_PRIV_NONE) { 15776 #if !defined(sun) 15777 #if __FreeBSD_version < 800039 15778 /* Destroy the cloned device. */ 15779 destroy_dev(dev); 15780 #endif 15781 #endif 15782 15783 return (EACCES); 15784 } 15785 15786 /* 15787 * Ask all providers to provide all their probes. 15788 */ 15789 mutex_enter(&dtrace_provider_lock); 15790 dtrace_probe_provide(NULL, NULL); 15791 mutex_exit(&dtrace_provider_lock); 15792 15793 mutex_enter(&cpu_lock); 15794 mutex_enter(&dtrace_lock); 15795 dtrace_opens++; 15796 dtrace_membar_producer(); 15797 15798 #if defined(sun) 15799 /* 15800 * If the kernel debugger is active (that is, if the kernel debugger 15801 * modified text in some way), we won't allow the open. 15802 */ 15803 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { 15804 dtrace_opens--; 15805 mutex_exit(&cpu_lock); 15806 mutex_exit(&dtrace_lock); 15807 return (EBUSY); 15808 } 15809 15810 state = dtrace_state_create(devp, cred_p); 15811 #else 15812 state = dtrace_state_create(dev); 15813 #if __FreeBSD_version < 800039 15814 dev->si_drv1 = state; 15815 #else 15816 devfs_set_cdevpriv(state, dtrace_dtr); 15817 #endif 15818 #endif 15819 15820 mutex_exit(&cpu_lock); 15821 15822 if (state == NULL) { 15823 #if defined(sun) 15824 if (--dtrace_opens == 0) 15825 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); 15826 #else 15827 --dtrace_opens; 15828 #endif 15829 mutex_exit(&dtrace_lock); 15830 #if !defined(sun) 15831 #if __FreeBSD_version < 800039 15832 /* Destroy the cloned device. */ 15833 destroy_dev(dev); 15834 #endif 15835 #endif 15836 return (EAGAIN); 15837 } 15838 15839 mutex_exit(&dtrace_lock); 15840 15841 return (0); 15842 } 15843 15844 /*ARGSUSED*/ 15845 #if defined(sun) 15846 static int 15847 dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p) 15848 #elif __FreeBSD_version < 800039 15849 static int 15850 dtrace_close(struct cdev *dev, int flags, int fmt __unused, struct thread *td) 15851 #else 15852 static void 15853 dtrace_dtr(void *data) 15854 #endif 15855 { 15856 #if defined(sun) 15857 minor_t minor = getminor(dev); 15858 dtrace_state_t *state; 15859 15860 if (minor == DTRACEMNRN_HELPER) 15861 return (0); 15862 15863 state = ddi_get_soft_state(dtrace_softstate, minor); 15864 #else 15865 #if __FreeBSD_version < 800039 15866 dtrace_state_t *state = dev->si_drv1; 15867 15868 /* Check if this is not a cloned device. */ 15869 if (dev2unit(dev) == 0) 15870 return (0); 15871 #else 15872 dtrace_state_t *state = data; 15873 #endif 15874 15875 #endif 15876 15877 mutex_enter(&cpu_lock); 15878 mutex_enter(&dtrace_lock); 15879 15880 if (state != NULL) { 15881 if (state->dts_anon) { 15882 /* 15883 * There is anonymous state. Destroy that first. 15884 */ 15885 ASSERT(dtrace_anon.dta_state == NULL); 15886 dtrace_state_destroy(state->dts_anon); 15887 } 15888 15889 dtrace_state_destroy(state); 15890 15891 #if !defined(sun) 15892 kmem_free(state, 0); 15893 #if __FreeBSD_version < 800039 15894 dev->si_drv1 = NULL; 15895 #endif 15896 #endif 15897 } 15898 15899 ASSERT(dtrace_opens > 0); 15900 #if defined(sun) 15901 if (--dtrace_opens == 0) 15902 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); 15903 #else 15904 --dtrace_opens; 15905 #endif 15906 15907 mutex_exit(&dtrace_lock); 15908 mutex_exit(&cpu_lock); 15909 15910 #if __FreeBSD_version < 800039 15911 /* Schedule this cloned device to be destroyed. */ 15912 destroy_dev_sched(dev); 15913 #endif 15914 15915 #if defined(sun) || __FreeBSD_version < 800039 15916 return (0); 15917 #endif 15918 } 15919 15920 #if defined(sun) 15921 /*ARGSUSED*/ 15922 static int 15923 dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv) 15924 { 15925 int rval; 15926 dof_helper_t help, *dhp = NULL; 15927 15928 switch (cmd) { 15929 case DTRACEHIOC_ADDDOF: 15930 if (copyin((void *)arg, &help, sizeof (help)) != 0) { 15931 dtrace_dof_error(NULL, "failed to copyin DOF helper"); 15932 return (EFAULT); 15933 } 15934 15935 dhp = &help; 15936 arg = (intptr_t)help.dofhp_dof; 15937 /*FALLTHROUGH*/ 15938 15939 case DTRACEHIOC_ADD: { 15940 dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval); 15941 15942 if (dof == NULL) 15943 return (rval); 15944 15945 mutex_enter(&dtrace_lock); 15946 15947 /* 15948 * dtrace_helper_slurp() takes responsibility for the dof -- 15949 * it may free it now or it may save it and free it later. 15950 */ 15951 if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) { 15952 *rv = rval; 15953 rval = 0; 15954 } else { 15955 rval = EINVAL; 15956 } 15957 15958 mutex_exit(&dtrace_lock); 15959 return (rval); 15960 } 15961 15962 case DTRACEHIOC_REMOVE: { 15963 mutex_enter(&dtrace_lock); 15964 rval = dtrace_helper_destroygen(arg); 15965 mutex_exit(&dtrace_lock); 15966 15967 return (rval); 15968 } 15969 15970 default: 15971 break; 15972 } 15973 15974 return (ENOTTY); 15975 } 15976 15977 /*ARGSUSED*/ 15978 static int 15979 dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv) 15980 { 15981 minor_t minor = getminor(dev); 15982 dtrace_state_t *state; 15983 int rval; 15984 15985 if (minor == DTRACEMNRN_HELPER) 15986 return (dtrace_ioctl_helper(cmd, arg, rv)); 15987 15988 state = ddi_get_soft_state(dtrace_softstate, minor); 15989 15990 if (state->dts_anon) { 15991 ASSERT(dtrace_anon.dta_state == NULL); 15992 state = state->dts_anon; 15993 } 15994 15995 switch (cmd) { 15996 case DTRACEIOC_PROVIDER: { 15997 dtrace_providerdesc_t pvd; 15998 dtrace_provider_t *pvp; 15999 16000 if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0) 16001 return (EFAULT); 16002 16003 pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0'; 16004 mutex_enter(&dtrace_provider_lock); 16005 16006 for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) { 16007 if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0) 16008 break; 16009 } 16010 16011 mutex_exit(&dtrace_provider_lock); 16012 16013 if (pvp == NULL) 16014 return (ESRCH); 16015 16016 bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t)); 16017 bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t)); 16018 16019 if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0) 16020 return (EFAULT); 16021 16022 return (0); 16023 } 16024 16025 case DTRACEIOC_EPROBE: { 16026 dtrace_eprobedesc_t epdesc; 16027 dtrace_ecb_t *ecb; 16028 dtrace_action_t *act; 16029 void *buf; 16030 size_t size; 16031 uintptr_t dest; 16032 int nrecs; 16033 16034 if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0) 16035 return (EFAULT); 16036 16037 mutex_enter(&dtrace_lock); 16038 16039 if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) { 16040 mutex_exit(&dtrace_lock); 16041 return (EINVAL); 16042 } 16043 16044 if (ecb->dte_probe == NULL) { 16045 mutex_exit(&dtrace_lock); 16046 return (EINVAL); 16047 } 16048 16049 epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id; 16050 epdesc.dtepd_uarg = ecb->dte_uarg; 16051 epdesc.dtepd_size = ecb->dte_size; 16052 16053 nrecs = epdesc.dtepd_nrecs; 16054 epdesc.dtepd_nrecs = 0; 16055 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 16056 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) 16057 continue; 16058 16059 epdesc.dtepd_nrecs++; 16060 } 16061 16062 /* 16063 * Now that we have the size, we need to allocate a temporary 16064 * buffer in which to store the complete description. We need 16065 * the temporary buffer to be able to drop dtrace_lock() 16066 * across the copyout(), below. 16067 */ 16068 size = sizeof (dtrace_eprobedesc_t) + 16069 (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t)); 16070 16071 buf = kmem_alloc(size, KM_SLEEP); 16072 dest = (uintptr_t)buf; 16073 16074 bcopy(&epdesc, (void *)dest, sizeof (epdesc)); 16075 dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]); 16076 16077 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 16078 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) 16079 continue; 16080 16081 if (nrecs-- == 0) 16082 break; 16083 16084 bcopy(&act->dta_rec, (void *)dest, 16085 sizeof (dtrace_recdesc_t)); 16086 dest += sizeof (dtrace_recdesc_t); 16087 } 16088 16089 mutex_exit(&dtrace_lock); 16090 16091 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { 16092 kmem_free(buf, size); 16093 return (EFAULT); 16094 } 16095 16096 kmem_free(buf, size); 16097 return (0); 16098 } 16099 16100 case DTRACEIOC_AGGDESC: { 16101 dtrace_aggdesc_t aggdesc; 16102 dtrace_action_t *act; 16103 dtrace_aggregation_t *agg; 16104 int nrecs; 16105 uint32_t offs; 16106 dtrace_recdesc_t *lrec; 16107 void *buf; 16108 size_t size; 16109 uintptr_t dest; 16110 16111 if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0) 16112 return (EFAULT); 16113 16114 mutex_enter(&dtrace_lock); 16115 16116 if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) { 16117 mutex_exit(&dtrace_lock); 16118 return (EINVAL); 16119 } 16120 16121 aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid; 16122 16123 nrecs = aggdesc.dtagd_nrecs; 16124 aggdesc.dtagd_nrecs = 0; 16125 16126 offs = agg->dtag_base; 16127 lrec = &agg->dtag_action.dta_rec; 16128 aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs; 16129 16130 for (act = agg->dtag_first; ; act = act->dta_next) { 16131 ASSERT(act->dta_intuple || 16132 DTRACEACT_ISAGG(act->dta_kind)); 16133 16134 /* 16135 * If this action has a record size of zero, it 16136 * denotes an argument to the aggregating action. 16137 * Because the presence of this record doesn't (or 16138 * shouldn't) affect the way the data is interpreted, 16139 * we don't copy it out to save user-level the 16140 * confusion of dealing with a zero-length record. 16141 */ 16142 if (act->dta_rec.dtrd_size == 0) { 16143 ASSERT(agg->dtag_hasarg); 16144 continue; 16145 } 16146 16147 aggdesc.dtagd_nrecs++; 16148 16149 if (act == &agg->dtag_action) 16150 break; 16151 } 16152 16153 /* 16154 * Now that we have the size, we need to allocate a temporary 16155 * buffer in which to store the complete description. We need 16156 * the temporary buffer to be able to drop dtrace_lock() 16157 * across the copyout(), below. 16158 */ 16159 size = sizeof (dtrace_aggdesc_t) + 16160 (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t)); 16161 16162 buf = kmem_alloc(size, KM_SLEEP); 16163 dest = (uintptr_t)buf; 16164 16165 bcopy(&aggdesc, (void *)dest, sizeof (aggdesc)); 16166 dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]); 16167 16168 for (act = agg->dtag_first; ; act = act->dta_next) { 16169 dtrace_recdesc_t rec = act->dta_rec; 16170 16171 /* 16172 * See the comment in the above loop for why we pass 16173 * over zero-length records. 16174 */ 16175 if (rec.dtrd_size == 0) { 16176 ASSERT(agg->dtag_hasarg); 16177 continue; 16178 } 16179 16180 if (nrecs-- == 0) 16181 break; 16182 16183 rec.dtrd_offset -= offs; 16184 bcopy(&rec, (void *)dest, sizeof (rec)); 16185 dest += sizeof (dtrace_recdesc_t); 16186 16187 if (act == &agg->dtag_action) 16188 break; 16189 } 16190 16191 mutex_exit(&dtrace_lock); 16192 16193 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { 16194 kmem_free(buf, size); 16195 return (EFAULT); 16196 } 16197 16198 kmem_free(buf, size); 16199 return (0); 16200 } 16201 16202 case DTRACEIOC_ENABLE: { 16203 dof_hdr_t *dof; 16204 dtrace_enabling_t *enab = NULL; 16205 dtrace_vstate_t *vstate; 16206 int err = 0; 16207 16208 *rv = 0; 16209 16210 /* 16211 * If a NULL argument has been passed, we take this as our 16212 * cue to reevaluate our enablings. 16213 */ 16214 if (arg == NULL) { 16215 dtrace_enabling_matchall(); 16216 16217 return (0); 16218 } 16219 16220 if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL) 16221 return (rval); 16222 16223 mutex_enter(&cpu_lock); 16224 mutex_enter(&dtrace_lock); 16225 vstate = &state->dts_vstate; 16226 16227 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { 16228 mutex_exit(&dtrace_lock); 16229 mutex_exit(&cpu_lock); 16230 dtrace_dof_destroy(dof); 16231 return (EBUSY); 16232 } 16233 16234 if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) { 16235 mutex_exit(&dtrace_lock); 16236 mutex_exit(&cpu_lock); 16237 dtrace_dof_destroy(dof); 16238 return (EINVAL); 16239 } 16240 16241 if ((rval = dtrace_dof_options(dof, state)) != 0) { 16242 dtrace_enabling_destroy(enab); 16243 mutex_exit(&dtrace_lock); 16244 mutex_exit(&cpu_lock); 16245 dtrace_dof_destroy(dof); 16246 return (rval); 16247 } 16248 16249 if ((err = dtrace_enabling_match(enab, rv)) == 0) { 16250 err = dtrace_enabling_retain(enab); 16251 } else { 16252 dtrace_enabling_destroy(enab); 16253 } 16254 16255 mutex_exit(&cpu_lock); 16256 mutex_exit(&dtrace_lock); 16257 dtrace_dof_destroy(dof); 16258 16259 return (err); 16260 } 16261 16262 case DTRACEIOC_REPLICATE: { 16263 dtrace_repldesc_t desc; 16264 dtrace_probedesc_t *match = &desc.dtrpd_match; 16265 dtrace_probedesc_t *create = &desc.dtrpd_create; 16266 int err; 16267 16268 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 16269 return (EFAULT); 16270 16271 match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; 16272 match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; 16273 match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; 16274 match->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; 16275 16276 create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; 16277 create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; 16278 create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; 16279 create->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; 16280 16281 mutex_enter(&dtrace_lock); 16282 err = dtrace_enabling_replicate(state, match, create); 16283 mutex_exit(&dtrace_lock); 16284 16285 return (err); 16286 } 16287 16288 case DTRACEIOC_PROBEMATCH: 16289 case DTRACEIOC_PROBES: { 16290 dtrace_probe_t *probe = NULL; 16291 dtrace_probedesc_t desc; 16292 dtrace_probekey_t pkey; 16293 dtrace_id_t i; 16294 int m = 0; 16295 uint32_t priv; 16296 uid_t uid; 16297 zoneid_t zoneid; 16298 16299 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 16300 return (EFAULT); 16301 16302 desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; 16303 desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; 16304 desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; 16305 desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0'; 16306 16307 /* 16308 * Before we attempt to match this probe, we want to give 16309 * all providers the opportunity to provide it. 16310 */ 16311 if (desc.dtpd_id == DTRACE_IDNONE) { 16312 mutex_enter(&dtrace_provider_lock); 16313 dtrace_probe_provide(&desc, NULL); 16314 mutex_exit(&dtrace_provider_lock); 16315 desc.dtpd_id++; 16316 } 16317 16318 if (cmd == DTRACEIOC_PROBEMATCH) { 16319 dtrace_probekey(&desc, &pkey); 16320 pkey.dtpk_id = DTRACE_IDNONE; 16321 } 16322 16323 dtrace_cred2priv(cr, &priv, &uid, &zoneid); 16324 16325 mutex_enter(&dtrace_lock); 16326 16327 if (cmd == DTRACEIOC_PROBEMATCH) { 16328 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { 16329 if ((probe = dtrace_probes[i - 1]) != NULL && 16330 (m = dtrace_match_probe(probe, &pkey, 16331 priv, uid, zoneid)) != 0) 16332 break; 16333 } 16334 16335 if (m < 0) { 16336 mutex_exit(&dtrace_lock); 16337 return (EINVAL); 16338 } 16339 16340 } else { 16341 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { 16342 if ((probe = dtrace_probes[i - 1]) != NULL && 16343 dtrace_match_priv(probe, priv, uid, zoneid)) 16344 break; 16345 } 16346 } 16347 16348 if (probe == NULL) { 16349 mutex_exit(&dtrace_lock); 16350 return (ESRCH); 16351 } 16352 16353 dtrace_probe_description(probe, &desc); 16354 mutex_exit(&dtrace_lock); 16355 16356 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 16357 return (EFAULT); 16358 16359 return (0); 16360 } 16361 16362 case DTRACEIOC_PROBEARG: { 16363 dtrace_argdesc_t desc; 16364 dtrace_probe_t *probe; 16365 dtrace_provider_t *prov; 16366 16367 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 16368 return (EFAULT); 16369 16370 if (desc.dtargd_id == DTRACE_IDNONE) 16371 return (EINVAL); 16372 16373 if (desc.dtargd_ndx == DTRACE_ARGNONE) 16374 return (EINVAL); 16375 16376 mutex_enter(&dtrace_provider_lock); 16377 mutex_enter(&mod_lock); 16378 mutex_enter(&dtrace_lock); 16379 16380 if (desc.dtargd_id > dtrace_nprobes) { 16381 mutex_exit(&dtrace_lock); 16382 mutex_exit(&mod_lock); 16383 mutex_exit(&dtrace_provider_lock); 16384 return (EINVAL); 16385 } 16386 16387 if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) { 16388 mutex_exit(&dtrace_lock); 16389 mutex_exit(&mod_lock); 16390 mutex_exit(&dtrace_provider_lock); 16391 return (EINVAL); 16392 } 16393 16394 mutex_exit(&dtrace_lock); 16395 16396 prov = probe->dtpr_provider; 16397 16398 if (prov->dtpv_pops.dtps_getargdesc == NULL) { 16399 /* 16400 * There isn't any typed information for this probe. 16401 * Set the argument number to DTRACE_ARGNONE. 16402 */ 16403 desc.dtargd_ndx = DTRACE_ARGNONE; 16404 } else { 16405 desc.dtargd_native[0] = '\0'; 16406 desc.dtargd_xlate[0] = '\0'; 16407 desc.dtargd_mapping = desc.dtargd_ndx; 16408 16409 prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg, 16410 probe->dtpr_id, probe->dtpr_arg, &desc); 16411 } 16412 16413 mutex_exit(&mod_lock); 16414 mutex_exit(&dtrace_provider_lock); 16415 16416 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 16417 return (EFAULT); 16418 16419 return (0); 16420 } 16421 16422 case DTRACEIOC_GO: { 16423 processorid_t cpuid; 16424 rval = dtrace_state_go(state, &cpuid); 16425 16426 if (rval != 0) 16427 return (rval); 16428 16429 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) 16430 return (EFAULT); 16431 16432 return (0); 16433 } 16434 16435 case DTRACEIOC_STOP: { 16436 processorid_t cpuid; 16437 16438 mutex_enter(&dtrace_lock); 16439 rval = dtrace_state_stop(state, &cpuid); 16440 mutex_exit(&dtrace_lock); 16441 16442 if (rval != 0) 16443 return (rval); 16444 16445 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) 16446 return (EFAULT); 16447 16448 return (0); 16449 } 16450 16451 case DTRACEIOC_DOFGET: { 16452 dof_hdr_t hdr, *dof; 16453 uint64_t len; 16454 16455 if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0) 16456 return (EFAULT); 16457 16458 mutex_enter(&dtrace_lock); 16459 dof = dtrace_dof_create(state); 16460 mutex_exit(&dtrace_lock); 16461 16462 len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz); 16463 rval = copyout(dof, (void *)arg, len); 16464 dtrace_dof_destroy(dof); 16465 16466 return (rval == 0 ? 0 : EFAULT); 16467 } 16468 16469 case DTRACEIOC_AGGSNAP: 16470 case DTRACEIOC_BUFSNAP: { 16471 dtrace_bufdesc_t desc; 16472 caddr_t cached; 16473 dtrace_buffer_t *buf; 16474 16475 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 16476 return (EFAULT); 16477 16478 if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU) 16479 return (EINVAL); 16480 16481 mutex_enter(&dtrace_lock); 16482 16483 if (cmd == DTRACEIOC_BUFSNAP) { 16484 buf = &state->dts_buffer[desc.dtbd_cpu]; 16485 } else { 16486 buf = &state->dts_aggbuffer[desc.dtbd_cpu]; 16487 } 16488 16489 if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) { 16490 size_t sz = buf->dtb_offset; 16491 16492 if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) { 16493 mutex_exit(&dtrace_lock); 16494 return (EBUSY); 16495 } 16496 16497 /* 16498 * If this buffer has already been consumed, we're 16499 * going to indicate that there's nothing left here 16500 * to consume. 16501 */ 16502 if (buf->dtb_flags & DTRACEBUF_CONSUMED) { 16503 mutex_exit(&dtrace_lock); 16504 16505 desc.dtbd_size = 0; 16506 desc.dtbd_drops = 0; 16507 desc.dtbd_errors = 0; 16508 desc.dtbd_oldest = 0; 16509 sz = sizeof (desc); 16510 16511 if (copyout(&desc, (void *)arg, sz) != 0) 16512 return (EFAULT); 16513 16514 return (0); 16515 } 16516 16517 /* 16518 * If this is a ring buffer that has wrapped, we want 16519 * to copy the whole thing out. 16520 */ 16521 if (buf->dtb_flags & DTRACEBUF_WRAPPED) { 16522 dtrace_buffer_polish(buf); 16523 sz = buf->dtb_size; 16524 } 16525 16526 if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) { 16527 mutex_exit(&dtrace_lock); 16528 return (EFAULT); 16529 } 16530 16531 desc.dtbd_size = sz; 16532 desc.dtbd_drops = buf->dtb_drops; 16533 desc.dtbd_errors = buf->dtb_errors; 16534 desc.dtbd_oldest = buf->dtb_xamot_offset; 16535 desc.dtbd_timestamp = dtrace_gethrtime(); 16536 16537 mutex_exit(&dtrace_lock); 16538 16539 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 16540 return (EFAULT); 16541 16542 buf->dtb_flags |= DTRACEBUF_CONSUMED; 16543 16544 return (0); 16545 } 16546 16547 if (buf->dtb_tomax == NULL) { 16548 ASSERT(buf->dtb_xamot == NULL); 16549 mutex_exit(&dtrace_lock); 16550 return (ENOENT); 16551 } 16552 16553 cached = buf->dtb_tomax; 16554 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); 16555 16556 dtrace_xcall(desc.dtbd_cpu, 16557 (dtrace_xcall_t)dtrace_buffer_switch, buf); 16558 16559 state->dts_errors += buf->dtb_xamot_errors; 16560 16561 /* 16562 * If the buffers did not actually switch, then the cross call 16563 * did not take place -- presumably because the given CPU is 16564 * not in the ready set. If this is the case, we'll return 16565 * ENOENT. 16566 */ 16567 if (buf->dtb_tomax == cached) { 16568 ASSERT(buf->dtb_xamot != cached); 16569 mutex_exit(&dtrace_lock); 16570 return (ENOENT); 16571 } 16572 16573 ASSERT(cached == buf->dtb_xamot); 16574 16575 /* 16576 * We have our snapshot; now copy it out. 16577 */ 16578 if (copyout(buf->dtb_xamot, desc.dtbd_data, 16579 buf->dtb_xamot_offset) != 0) { 16580 mutex_exit(&dtrace_lock); 16581 return (EFAULT); 16582 } 16583 16584 desc.dtbd_size = buf->dtb_xamot_offset; 16585 desc.dtbd_drops = buf->dtb_xamot_drops; 16586 desc.dtbd_errors = buf->dtb_xamot_errors; 16587 desc.dtbd_oldest = 0; 16588 desc.dtbd_timestamp = buf->dtb_switched; 16589 16590 mutex_exit(&dtrace_lock); 16591 16592 /* 16593 * Finally, copy out the buffer description. 16594 */ 16595 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 16596 return (EFAULT); 16597 16598 return (0); 16599 } 16600 16601 case DTRACEIOC_CONF: { 16602 dtrace_conf_t conf; 16603 16604 bzero(&conf, sizeof (conf)); 16605 conf.dtc_difversion = DIF_VERSION; 16606 conf.dtc_difintregs = DIF_DIR_NREGS; 16607 conf.dtc_diftupregs = DIF_DTR_NREGS; 16608 conf.dtc_ctfmodel = CTF_MODEL_NATIVE; 16609 16610 if (copyout(&conf, (void *)arg, sizeof (conf)) != 0) 16611 return (EFAULT); 16612 16613 return (0); 16614 } 16615 16616 case DTRACEIOC_STATUS: { 16617 dtrace_status_t stat; 16618 dtrace_dstate_t *dstate; 16619 int i, j; 16620 uint64_t nerrs; 16621 16622 /* 16623 * See the comment in dtrace_state_deadman() for the reason 16624 * for setting dts_laststatus to INT64_MAX before setting 16625 * it to the correct value. 16626 */ 16627 state->dts_laststatus = INT64_MAX; 16628 dtrace_membar_producer(); 16629 state->dts_laststatus = dtrace_gethrtime(); 16630 16631 bzero(&stat, sizeof (stat)); 16632 16633 mutex_enter(&dtrace_lock); 16634 16635 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) { 16636 mutex_exit(&dtrace_lock); 16637 return (ENOENT); 16638 } 16639 16640 if (state->dts_activity == DTRACE_ACTIVITY_DRAINING) 16641 stat.dtst_exiting = 1; 16642 16643 nerrs = state->dts_errors; 16644 dstate = &state->dts_vstate.dtvs_dynvars; 16645 16646 for (i = 0; i < NCPU; i++) { 16647 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i]; 16648 16649 stat.dtst_dyndrops += dcpu->dtdsc_drops; 16650 stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops; 16651 stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops; 16652 16653 if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL) 16654 stat.dtst_filled++; 16655 16656 nerrs += state->dts_buffer[i].dtb_errors; 16657 16658 for (j = 0; j < state->dts_nspeculations; j++) { 16659 dtrace_speculation_t *spec; 16660 dtrace_buffer_t *buf; 16661 16662 spec = &state->dts_speculations[j]; 16663 buf = &spec->dtsp_buffer[i]; 16664 stat.dtst_specdrops += buf->dtb_xamot_drops; 16665 } 16666 } 16667 16668 stat.dtst_specdrops_busy = state->dts_speculations_busy; 16669 stat.dtst_specdrops_unavail = state->dts_speculations_unavail; 16670 stat.dtst_stkstroverflows = state->dts_stkstroverflows; 16671 stat.dtst_dblerrors = state->dts_dblerrors; 16672 stat.dtst_killed = 16673 (state->dts_activity == DTRACE_ACTIVITY_KILLED); 16674 stat.dtst_errors = nerrs; 16675 16676 mutex_exit(&dtrace_lock); 16677 16678 if (copyout(&stat, (void *)arg, sizeof (stat)) != 0) 16679 return (EFAULT); 16680 16681 return (0); 16682 } 16683 16684 case DTRACEIOC_FORMAT: { 16685 dtrace_fmtdesc_t fmt; 16686 char *str; 16687 int len; 16688 16689 if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0) 16690 return (EFAULT); 16691 16692 mutex_enter(&dtrace_lock); 16693 16694 if (fmt.dtfd_format == 0 || 16695 fmt.dtfd_format > state->dts_nformats) { 16696 mutex_exit(&dtrace_lock); 16697 return (EINVAL); 16698 } 16699 16700 /* 16701 * Format strings are allocated contiguously and they are 16702 * never freed; if a format index is less than the number 16703 * of formats, we can assert that the format map is non-NULL 16704 * and that the format for the specified index is non-NULL. 16705 */ 16706 ASSERT(state->dts_formats != NULL); 16707 str = state->dts_formats[fmt.dtfd_format - 1]; 16708 ASSERT(str != NULL); 16709 16710 len = strlen(str) + 1; 16711 16712 if (len > fmt.dtfd_length) { 16713 fmt.dtfd_length = len; 16714 16715 if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) { 16716 mutex_exit(&dtrace_lock); 16717 return (EINVAL); 16718 } 16719 } else { 16720 if (copyout(str, fmt.dtfd_string, len) != 0) { 16721 mutex_exit(&dtrace_lock); 16722 return (EINVAL); 16723 } 16724 } 16725 16726 mutex_exit(&dtrace_lock); 16727 return (0); 16728 } 16729 16730 default: 16731 break; 16732 } 16733 16734 return (ENOTTY); 16735 } 16736 16737 /*ARGSUSED*/ 16738 static int 16739 dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) 16740 { 16741 dtrace_state_t *state; 16742 16743 switch (cmd) { 16744 case DDI_DETACH: 16745 break; 16746 16747 case DDI_SUSPEND: 16748 return (DDI_SUCCESS); 16749 16750 default: 16751 return (DDI_FAILURE); 16752 } 16753 16754 mutex_enter(&cpu_lock); 16755 mutex_enter(&dtrace_provider_lock); 16756 mutex_enter(&dtrace_lock); 16757 16758 ASSERT(dtrace_opens == 0); 16759 16760 if (dtrace_helpers > 0) { 16761 mutex_exit(&dtrace_provider_lock); 16762 mutex_exit(&dtrace_lock); 16763 mutex_exit(&cpu_lock); 16764 return (DDI_FAILURE); 16765 } 16766 16767 if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) { 16768 mutex_exit(&dtrace_provider_lock); 16769 mutex_exit(&dtrace_lock); 16770 mutex_exit(&cpu_lock); 16771 return (DDI_FAILURE); 16772 } 16773 16774 dtrace_provider = NULL; 16775 16776 if ((state = dtrace_anon_grab()) != NULL) { 16777 /* 16778 * If there were ECBs on this state, the provider should 16779 * have not been allowed to detach; assert that there is 16780 * none. 16781 */ 16782 ASSERT(state->dts_necbs == 0); 16783 dtrace_state_destroy(state); 16784 16785 /* 16786 * If we're being detached with anonymous state, we need to 16787 * indicate to the kernel debugger that DTrace is now inactive. 16788 */ 16789 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); 16790 } 16791 16792 bzero(&dtrace_anon, sizeof (dtrace_anon_t)); 16793 unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); 16794 dtrace_cpu_init = NULL; 16795 dtrace_helpers_cleanup = NULL; 16796 dtrace_helpers_fork = NULL; 16797 dtrace_cpustart_init = NULL; 16798 dtrace_cpustart_fini = NULL; 16799 dtrace_debugger_init = NULL; 16800 dtrace_debugger_fini = NULL; 16801 dtrace_modload = NULL; 16802 dtrace_modunload = NULL; 16803 16804 mutex_exit(&cpu_lock); 16805 16806 if (dtrace_helptrace_enabled) { 16807 kmem_free(dtrace_helptrace_buffer, dtrace_helptrace_bufsize); 16808 dtrace_helptrace_buffer = NULL; 16809 } 16810 16811 kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *)); 16812 dtrace_probes = NULL; 16813 dtrace_nprobes = 0; 16814 16815 dtrace_hash_destroy(dtrace_bymod); 16816 dtrace_hash_destroy(dtrace_byfunc); 16817 dtrace_hash_destroy(dtrace_byname); 16818 dtrace_bymod = NULL; 16819 dtrace_byfunc = NULL; 16820 dtrace_byname = NULL; 16821 16822 kmem_cache_destroy(dtrace_state_cache); 16823 vmem_destroy(dtrace_minor); 16824 vmem_destroy(dtrace_arena); 16825 16826 if (dtrace_toxrange != NULL) { 16827 kmem_free(dtrace_toxrange, 16828 dtrace_toxranges_max * sizeof (dtrace_toxrange_t)); 16829 dtrace_toxrange = NULL; 16830 dtrace_toxranges = 0; 16831 dtrace_toxranges_max = 0; 16832 } 16833 16834 ddi_remove_minor_node(dtrace_devi, NULL); 16835 dtrace_devi = NULL; 16836 16837 ddi_soft_state_fini(&dtrace_softstate); 16838 16839 ASSERT(dtrace_vtime_references == 0); 16840 ASSERT(dtrace_opens == 0); 16841 ASSERT(dtrace_retained == NULL); 16842 16843 mutex_exit(&dtrace_lock); 16844 mutex_exit(&dtrace_provider_lock); 16845 16846 /* 16847 * We don't destroy the task queue until after we have dropped our 16848 * locks (taskq_destroy() may block on running tasks). To prevent 16849 * attempting to do work after we have effectively detached but before 16850 * the task queue has been destroyed, all tasks dispatched via the 16851 * task queue must check that DTrace is still attached before 16852 * performing any operation. 16853 */ 16854 taskq_destroy(dtrace_taskq); 16855 dtrace_taskq = NULL; 16856 16857 return (DDI_SUCCESS); 16858 } 16859 #endif 16860 16861 #if defined(sun) 16862 /*ARGSUSED*/ 16863 static int 16864 dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) 16865 { 16866 int error; 16867 16868 switch (infocmd) { 16869 case DDI_INFO_DEVT2DEVINFO: 16870 *result = (void *)dtrace_devi; 16871 error = DDI_SUCCESS; 16872 break; 16873 case DDI_INFO_DEVT2INSTANCE: 16874 *result = (void *)0; 16875 error = DDI_SUCCESS; 16876 break; 16877 default: 16878 error = DDI_FAILURE; 16879 } 16880 return (error); 16881 } 16882 #endif 16883 16884 #if defined(sun) 16885 static struct cb_ops dtrace_cb_ops = { 16886 dtrace_open, /* open */ 16887 dtrace_close, /* close */ 16888 nulldev, /* strategy */ 16889 nulldev, /* print */ 16890 nodev, /* dump */ 16891 nodev, /* read */ 16892 nodev, /* write */ 16893 dtrace_ioctl, /* ioctl */ 16894 nodev, /* devmap */ 16895 nodev, /* mmap */ 16896 nodev, /* segmap */ 16897 nochpoll, /* poll */ 16898 ddi_prop_op, /* cb_prop_op */ 16899 0, /* streamtab */ 16900 D_NEW | D_MP /* Driver compatibility flag */ 16901 }; 16902 16903 static struct dev_ops dtrace_ops = { 16904 DEVO_REV, /* devo_rev */ 16905 0, /* refcnt */ 16906 dtrace_info, /* get_dev_info */ 16907 nulldev, /* identify */ 16908 nulldev, /* probe */ 16909 dtrace_attach, /* attach */ 16910 dtrace_detach, /* detach */ 16911 nodev, /* reset */ 16912 &dtrace_cb_ops, /* driver operations */ 16913 NULL, /* bus operations */ 16914 nodev /* dev power */ 16915 }; 16916 16917 static struct modldrv modldrv = { 16918 &mod_driverops, /* module type (this is a pseudo driver) */ 16919 "Dynamic Tracing", /* name of module */ 16920 &dtrace_ops, /* driver ops */ 16921 }; 16922 16923 static struct modlinkage modlinkage = { 16924 MODREV_1, 16925 (void *)&modldrv, 16926 NULL 16927 }; 16928 16929 int 16930 _init(void) 16931 { 16932 return (mod_install(&modlinkage)); 16933 } 16934 16935 int 16936 _info(struct modinfo *modinfop) 16937 { 16938 return (mod_info(&modlinkage, modinfop)); 16939 } 16940 16941 int 16942 _fini(void) 16943 { 16944 return (mod_remove(&modlinkage)); 16945 } 16946 #else 16947 16948 static d_ioctl_t dtrace_ioctl; 16949 static d_ioctl_t dtrace_ioctl_helper; 16950 static void dtrace_load(void *); 16951 static int dtrace_unload(void); 16952 #if __FreeBSD_version < 800039 16953 static void dtrace_clone(void *, struct ucred *, char *, int , struct cdev **); 16954 static struct clonedevs *dtrace_clones; /* Ptr to the array of cloned devices. */ 16955 static eventhandler_tag eh_tag; /* Event handler tag. */ 16956 #else 16957 static struct cdev *dtrace_dev; 16958 static struct cdev *helper_dev; 16959 #endif 16960 16961 void dtrace_invop_init(void); 16962 void dtrace_invop_uninit(void); 16963 16964 static struct cdevsw dtrace_cdevsw = { 16965 .d_version = D_VERSION, 16966 #if __FreeBSD_version < 800039 16967 .d_flags = D_TRACKCLOSE | D_NEEDMINOR, 16968 .d_close = dtrace_close, 16969 #endif 16970 .d_ioctl = dtrace_ioctl, 16971 .d_open = dtrace_open, 16972 .d_name = "dtrace", 16973 }; 16974 16975 static struct cdevsw helper_cdevsw = { 16976 .d_version = D_VERSION, 16977 .d_ioctl = dtrace_ioctl_helper, 16978 .d_name = "helper", 16979 }; 16980 16981 #include <dtrace_anon.c> 16982 #if __FreeBSD_version < 800039 16983 #include <dtrace_clone.c> 16984 #endif 16985 #include <dtrace_ioctl.c> 16986 #include <dtrace_load.c> 16987 #include <dtrace_modevent.c> 16988 #include <dtrace_sysctl.c> 16989 #include <dtrace_unload.c> 16990 #include <dtrace_vtime.c> 16991 #include <dtrace_hacks.c> 16992 #include <dtrace_isa.c> 16993 16994 SYSINIT(dtrace_load, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_load, NULL); 16995 SYSUNINIT(dtrace_unload, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_unload, NULL); 16996 SYSINIT(dtrace_anon_init, SI_SUB_DTRACE_ANON, SI_ORDER_FIRST, dtrace_anon_init, NULL); 16997 16998 DEV_MODULE(dtrace, dtrace_modevent, NULL); 16999 MODULE_VERSION(dtrace, 1); 17000 MODULE_DEPEND(dtrace, cyclic, 1, 1, 1); 17001 MODULE_DEPEND(dtrace, opensolaris, 1, 1, 1); 17002 #endif 17003