| 1 | /* |
| 2 | * CDDL HEADER START |
| 3 | * |
| 4 | * The contents of this file are subject to the terms of the |
| 5 | * Common Development and Distribution License (the "License"). |
| 6 | * You may not use this file except in compliance with the License. |
| 7 | * |
| 8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE |
| 9 | * or http://www.opensolaris.org/os/licensing. |
| 10 | * See the License for the specific language governing permissions |
| 11 | * and limitations under the License. |
| 12 | * |
| 13 | * When distributing Covered Code, include this CDDL HEADER in each |
| 14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. |
| 15 | * If applicable, add the following below this CDDL HEADER, with the |
| 16 | * fields enclosed by brackets "[]" replaced with your own identifying |
| 17 | * information: Portions Copyright [yyyy] [name of copyright owner] |
| 18 | * |
| 19 | * CDDL HEADER END |
| 20 | */ |
| 21 | /* |
| 22 | * Copyright 2009 Sun Microsystems, Inc. All rights reserved. |
| 23 | * Use is subject to license terms. |
| 24 | */ |
| 25 | |
| 26 | #include <sys/param.h> |
| 27 | #include <sys/systm.h> |
| 28 | #include <sys/errno.h> |
| 29 | #include <sys/stat.h> |
| 30 | #include <sys/ioctl.h> |
| 31 | #include <sys/conf.h> |
| 32 | #include <sys/fcntl.h> |
| 33 | #include <miscfs/devfs/devfs.h> |
| 34 | |
| 35 | #if defined(__arm64__) |
| 36 | #include <arm/caches_internal.h> |
| 37 | #endif /* defined(__arm64__) */ |
| 38 | |
| 39 | #include <sys/dtrace.h> |
| 40 | #include <sys/dtrace_impl.h> |
| 41 | |
| 42 | #include <sys/dtrace_glue.h> |
| 43 | |
| 44 | #include <sys/sdt_impl.h> |
| 45 | extern int dtrace_kernel_symbol_mode; |
| 46 | |
| 47 | #include <ptrauth.h> |
| 48 | |
| 49 | /* #include <machine/trap.h */ |
| 50 | struct savearea_t; /* Used anonymously */ |
| 51 | |
| 52 | #if defined(__arm64__) |
| 53 | typedef kern_return_t (*perfCallback)(int, struct savearea_t *, __unused int, __unused int); |
| 54 | extern perfCallback tempDTraceTrapHook; |
| 55 | extern kern_return_t fbt_perfCallback(int, struct savearea_t *, __unused int, __unused int); |
| 56 | #define SDT_PATCHVAL 0xe7eeee7e |
| 57 | #define SDT_AFRAMES 7 |
| 58 | #elif defined(__x86_64__) |
| 59 | typedef kern_return_t (*perfCallback)(int, struct savearea_t *, uintptr_t *, int); |
| 60 | extern perfCallback tempDTraceTrapHook; |
| 61 | extern kern_return_t fbt_perfCallback(int, struct savearea_t *, uintptr_t *, int); |
| 62 | #define SDT_PATCHVAL 0xf0 |
| 63 | #define SDT_AFRAMES 6 |
| 64 | #else |
| 65 | #error Unknown architecture |
| 66 | #endif |
| 67 | |
| 68 | #define SDT_PROBETAB_SIZE 0x1000 /* 4k entries -- 16K total */ |
| 69 | |
| 70 | #define SDT_UNKNOWN_FUNCNAME "." /* function symbol name when not found in symbol table */ |
| 71 | |
| 72 | |
| 73 | static int sdt_verbose = 0; |
| 74 | sdt_probe_t **sdt_probetab; |
| 75 | int sdt_probetab_size; |
| 76 | int sdt_probetab_mask; |
| 77 | |
| 78 | /*ARGSUSED*/ |
| 79 | static void |
| 80 | __sdt_provide_module(void *arg, struct modctl *ctl) |
| 81 | { |
| 82 | #pragma unused(arg) |
| 83 | char *modname = ctl->mod_modname; |
| 84 | sdt_probedesc_t *sdpd; |
| 85 | sdt_probe_t *sdp, *old; |
| 86 | sdt_provider_t *prov; |
| 87 | |
| 88 | /* |
| 89 | * One for all, and all for one: if we haven't yet registered all of |
| 90 | * our providers, we'll refuse to provide anything. |
| 91 | */ |
| 92 | for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { |
| 93 | if (prov->sdtp_id == DTRACE_PROVNONE) { |
| 94 | return; |
| 95 | } |
| 96 | } |
| 97 | |
| 98 | /* Nothing to do. Module is either invalid or we haven't found any SDT probe descriptions. */ |
| 99 | if (!ctl || ctl->mod_sdtprobecnt != 0 || (sdpd = ctl->mod_sdtdesc) == NULL) { |
| 100 | return; |
| 101 | } |
| 102 | |
| 103 | for (sdpd = ctl->mod_sdtdesc; sdpd != NULL; sdpd = sdpd->sdpd_next) { |
| 104 | dtrace_id_t id; |
| 105 | |
| 106 | /* Validate probe's provider name. Do not provide probes for unknown providers. */ |
| 107 | for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { |
| 108 | if (strcmp(s1: prov->sdtp_prefix, s2: sdpd->sdpd_prov) == 0) { |
| 109 | break; |
| 110 | } |
| 111 | } |
| 112 | |
| 113 | if (prov->sdtp_name == NULL) { |
| 114 | printf("Ignoring probes from unsupported provider %s\n" , sdpd->sdpd_prov); |
| 115 | continue; |
| 116 | } |
| 117 | |
| 118 | if (sdpd->sdpd_func == NULL) { |
| 119 | /* |
| 120 | * Ignore probes for which we don't have any symbol. That's likely some problem with |
| 121 | * __sdt section processing. |
| 122 | */ |
| 123 | printf("Ignoring probe %s (no symbol name)\n" , sdpd->sdpd_name); |
| 124 | continue; |
| 125 | } |
| 126 | |
| 127 | sdp = kmem_zalloc(sizeof(sdt_probe_t), KM_SLEEP); |
| 128 | sdp->sdp_loadcnt = ctl->mod_loadcnt; |
| 129 | sdp->sdp_ctl = ctl; |
| 130 | sdp->sdp_name = kmem_alloc(strlen(sdpd->sdpd_name) + 1, KM_SLEEP); |
| 131 | (void) strlcpy(dst: sdp->sdp_name, src: sdpd->sdpd_name, n: strlen(s: sdpd->sdpd_name) + 1); |
| 132 | sdp->sdp_namelen = strlen(s: sdpd->sdpd_name) + 1; |
| 133 | sdp->sdp_provider = prov; |
| 134 | |
| 135 | /* |
| 136 | * We have our provider. Now create the probe. |
| 137 | */ |
| 138 | if ((id = dtrace_probe_lookup(prov->sdtp_id, modname, |
| 139 | sdpd->sdpd_func, sdp->sdp_name)) != DTRACE_IDNONE) { |
| 140 | old = dtrace_probe_arg(prov->sdtp_id, id); |
| 141 | ASSERT(old != NULL); |
| 142 | |
| 143 | sdp->sdp_next = old->sdp_next; |
| 144 | sdp->sdp_id = id; |
| 145 | old->sdp_next = sdp; |
| 146 | } else { |
| 147 | sdp->sdp_id = dtrace_probe_create(prov->sdtp_id, |
| 148 | modname, sdpd->sdpd_func, sdp->sdp_name, SDT_AFRAMES, sdp); |
| 149 | |
| 150 | ctl->mod_sdtprobecnt++; |
| 151 | } |
| 152 | |
| 153 | #if 0 |
| 154 | printf("__sdt_provide_module: sdpd=0x%p sdp=0x%p name=%s, id=%d\n" , sdpd, sdp, |
| 155 | sdp->sdp_name, sdp->sdp_id); |
| 156 | #endif |
| 157 | |
| 158 | sdp->sdp_hashnext = |
| 159 | sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)]; |
| 160 | sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)] = sdp; |
| 161 | |
| 162 | sdp->sdp_patchval = SDT_PATCHVAL; |
| 163 | sdp->sdp_patchpoint = (sdt_instr_t *)sdpd->sdpd_offset; |
| 164 | sdp->sdp_savedval = *sdp->sdp_patchpoint; |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | /*ARGSUSED*/ |
| 169 | static void |
| 170 | sdt_destroy(void *arg, dtrace_id_t id, void *parg) |
| 171 | { |
| 172 | #pragma unused(arg,id) |
| 173 | sdt_probe_t *sdp = parg, *old, *last, *hash; |
| 174 | int ndx; |
| 175 | |
| 176 | struct modctl *ctl = sdp->sdp_ctl; |
| 177 | |
| 178 | /* |
| 179 | * Decrement SDT probe counts only when a probe being destroyed belongs to the |
| 180 | * currently loaded version of a module and not the stale one. |
| 181 | */ |
| 182 | if (ctl != NULL && ctl->mod_loadcnt == sdp->sdp_loadcnt && ctl->mod_loaded) { |
| 183 | ctl->mod_sdtprobecnt--; |
| 184 | } |
| 185 | |
| 186 | while (sdp != NULL) { |
| 187 | old = sdp; |
| 188 | |
| 189 | /* |
| 190 | * Now we need to remove this probe from the sdt_probetab. |
| 191 | */ |
| 192 | ndx = SDT_ADDR2NDX(sdp->sdp_patchpoint); |
| 193 | last = NULL; |
| 194 | hash = sdt_probetab[ndx]; |
| 195 | |
| 196 | while (hash != sdp) { |
| 197 | ASSERT(hash != NULL); |
| 198 | last = hash; |
| 199 | hash = hash->sdp_hashnext; |
| 200 | } |
| 201 | |
| 202 | if (last != NULL) { |
| 203 | last->sdp_hashnext = sdp->sdp_hashnext; |
| 204 | } else { |
| 205 | sdt_probetab[ndx] = sdp->sdp_hashnext; |
| 206 | } |
| 207 | |
| 208 | kmem_free(sdp->sdp_name, sdp->sdp_namelen); |
| 209 | sdp = sdp->sdp_next; |
| 210 | kmem_free(old, sizeof(sdt_probe_t)); |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | /*ARGSUSED*/ |
| 215 | static int |
| 216 | sdt_enable(void *arg, dtrace_id_t id, void *parg) |
| 217 | { |
| 218 | #pragma unused(arg,id) |
| 219 | sdt_probe_t *sdp = parg; |
| 220 | struct modctl *ctl = sdp->sdp_ctl; |
| 221 | |
| 222 | ctl->mod_nenabled++; |
| 223 | |
| 224 | /* |
| 225 | * If this module has disappeared since we discovered its probes, |
| 226 | * refuse to enable it. |
| 227 | */ |
| 228 | if (!ctl->mod_loaded) { |
| 229 | if (sdt_verbose) { |
| 230 | cmn_err(CE_NOTE, "sdt is failing for probe %s " |
| 231 | "(module %s unloaded)" , |
| 232 | sdp->sdp_name, ctl->mod_modname); |
| 233 | } |
| 234 | goto err; |
| 235 | } |
| 236 | |
| 237 | /* |
| 238 | * Now check that our modctl has the expected load count. If it |
| 239 | * doesn't, this module must have been unloaded and reloaded -- and |
| 240 | * we're not going to touch it. |
| 241 | */ |
| 242 | if (ctl->mod_loadcnt != sdp->sdp_loadcnt) { |
| 243 | if (sdt_verbose) { |
| 244 | cmn_err(CE_NOTE, "sdt is failing for probe %s " |
| 245 | "(module %s reloaded)" , |
| 246 | sdp->sdp_name, ctl->mod_modname); |
| 247 | } |
| 248 | goto err; |
| 249 | } |
| 250 | |
| 251 | dtrace_casptr(&tempDTraceTrapHook, NULL, ptrauth_nop_cast(void *, &fbt_perfCallback)); |
| 252 | if (tempDTraceTrapHook != (perfCallback)fbt_perfCallback) { |
| 253 | if (sdt_verbose) { |
| 254 | cmn_err(CE_NOTE, "sdt_enable is failing for probe %s " |
| 255 | "in module %s: tempDTraceTrapHook already occupied." , |
| 256 | sdp->sdp_name, ctl->mod_modname); |
| 257 | } |
| 258 | return 0; |
| 259 | } |
| 260 | |
| 261 | while (sdp != NULL) { |
| 262 | (void)ml_nofault_copy(virtsrc: (vm_offset_t)&sdp->sdp_patchval, virtdst: (vm_offset_t)sdp->sdp_patchpoint, |
| 263 | size: (vm_size_t)sizeof(sdp->sdp_patchval)); |
| 264 | |
| 265 | /* |
| 266 | * Make the patched instruction visible via a data + instruction |
| 267 | * cache fush on platforms that need it |
| 268 | */ |
| 269 | flush_dcache((vm_offset_t)sdp->sdp_patchpoint, (vm_size_t)sizeof(sdp->sdp_patchval), 0); |
| 270 | invalidate_icache((vm_offset_t)sdp->sdp_patchpoint, (vm_size_t)sizeof(sdp->sdp_patchval), 0); |
| 271 | |
| 272 | sdp = sdp->sdp_next; |
| 273 | } |
| 274 | |
| 275 | err: |
| 276 | return 0; |
| 277 | } |
| 278 | |
| 279 | /*ARGSUSED*/ |
| 280 | static void |
| 281 | sdt_disable(void *arg, dtrace_id_t id, void *parg) |
| 282 | { |
| 283 | #pragma unused(arg,id) |
| 284 | sdt_probe_t *sdp = parg; |
| 285 | struct modctl *ctl = sdp->sdp_ctl; |
| 286 | |
| 287 | ctl->mod_nenabled--; |
| 288 | |
| 289 | if (!ctl->mod_loaded || ctl->mod_loadcnt != sdp->sdp_loadcnt) { |
| 290 | goto err; |
| 291 | } |
| 292 | |
| 293 | while (sdp != NULL) { |
| 294 | (void)ml_nofault_copy(virtsrc: (vm_offset_t)&sdp->sdp_savedval, virtdst: (vm_offset_t)sdp->sdp_patchpoint, |
| 295 | size: (vm_size_t)sizeof(sdp->sdp_savedval)); |
| 296 | /* |
| 297 | * Make the patched instruction visible via a data + instruction |
| 298 | * cache flush on platforms that need it |
| 299 | */ |
| 300 | flush_dcache((vm_offset_t)sdp->sdp_patchpoint, (vm_size_t)sizeof(sdp->sdp_savedval), 0); |
| 301 | invalidate_icache((vm_offset_t)sdp->sdp_patchpoint, (vm_size_t)sizeof(sdp->sdp_savedval), 0); |
| 302 | sdp = sdp->sdp_next; |
| 303 | } |
| 304 | |
| 305 | err: |
| 306 | ; |
| 307 | } |
| 308 | |
| 309 | static dtrace_pops_t sdt_pops = { |
| 310 | .dtps_provide = NULL, |
| 311 | .dtps_provide_module = sdt_provide_module, |
| 312 | .dtps_enable = sdt_enable, |
| 313 | .dtps_disable = sdt_disable, |
| 314 | .dtps_suspend = NULL, |
| 315 | .dtps_resume = NULL, |
| 316 | .dtps_getargdesc = sdt_getargdesc, |
| 317 | .dtps_getargval = sdt_getarg, |
| 318 | .dtps_usermode = NULL, |
| 319 | .dtps_destroy = sdt_destroy, |
| 320 | }; |
| 321 | |
| 322 | /*ARGSUSED*/ |
| 323 | static int |
| 324 | sdt_attach(dev_info_t *devi) |
| 325 | { |
| 326 | sdt_provider_t *prov; |
| 327 | |
| 328 | if (ddi_create_minor_node(devi, "sdt" , S_IFCHR, |
| 329 | 0, DDI_PSEUDO, 0) == DDI_FAILURE) { |
| 330 | cmn_err(CE_NOTE, "/dev/sdt couldn't create minor node" ); |
| 331 | ddi_remove_minor_node(devi, NULL); |
| 332 | return DDI_FAILURE; |
| 333 | } |
| 334 | |
| 335 | if (sdt_probetab_size == 0) { |
| 336 | sdt_probetab_size = SDT_PROBETAB_SIZE; |
| 337 | } |
| 338 | |
| 339 | sdt_probetab_mask = sdt_probetab_size - 1; |
| 340 | sdt_probetab = |
| 341 | kmem_zalloc(sdt_probetab_size * sizeof(sdt_probe_t *), KM_SLEEP); |
| 342 | dtrace_invop_add(sdt_invop); |
| 343 | |
| 344 | for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { |
| 345 | if (dtrace_register(prov->sdtp_name, prov->sdtp_attr, |
| 346 | DTRACE_PRIV_KERNEL, NULL, |
| 347 | &sdt_pops, prov, &prov->sdtp_id) != 0) { |
| 348 | cmn_err(CE_WARN, "failed to register sdt provider %s" , |
| 349 | prov->sdtp_name); |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | return DDI_SUCCESS; |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * APPLE NOTE: sdt_detach not implemented |
| 358 | */ |
| 359 | #if !defined(__APPLE__) |
| 360 | /*ARGSUSED*/ |
| 361 | static int |
| 362 | sdt_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) |
| 363 | { |
| 364 | sdt_provider_t *prov; |
| 365 | |
| 366 | switch (cmd) { |
| 367 | case DDI_DETACH: |
| 368 | break; |
| 369 | |
| 370 | case DDI_SUSPEND: |
| 371 | return DDI_SUCCESS; |
| 372 | |
| 373 | default: |
| 374 | return DDI_FAILURE; |
| 375 | } |
| 376 | |
| 377 | for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { |
| 378 | if (prov->sdtp_id != DTRACE_PROVNONE) { |
| 379 | if (dtrace_unregister(prov->sdtp_id) != 0) { |
| 380 | return DDI_FAILURE; |
| 381 | } |
| 382 | |
| 383 | prov->sdtp_id = DTRACE_PROVNONE; |
| 384 | } |
| 385 | } |
| 386 | |
| 387 | dtrace_invop_remove(sdt_invop); |
| 388 | kmem_free(sdt_probetab, sdt_probetab_size * sizeof(sdt_probe_t *)); |
| 389 | |
| 390 | return DDI_SUCCESS; |
| 391 | } |
| 392 | #endif /* __APPLE__ */ |
| 393 | |
| 394 | d_open_t _sdt_open; |
| 395 | |
| 396 | int |
| 397 | _sdt_open(dev_t dev, int flags, int devtype, struct proc *p) |
| 398 | { |
| 399 | #pragma unused(dev,flags,devtype,p) |
| 400 | return 0; |
| 401 | } |
| 402 | |
| 403 | #define SDT_MAJOR -24 /* let the kernel pick the device number */ |
| 404 | |
| 405 | static const struct cdevsw sdt_cdevsw = |
| 406 | { |
| 407 | .d_open = _sdt_open, |
| 408 | .d_close = eno_opcl, |
| 409 | .d_read = eno_rdwrt, |
| 410 | .d_write = eno_rdwrt, |
| 411 | .d_ioctl = eno_ioctl, |
| 412 | .d_stop = eno_stop, |
| 413 | .d_reset = eno_reset, |
| 414 | .d_select = eno_select, |
| 415 | .d_mmap = eno_mmap, |
| 416 | .d_strategy = eno_strat, |
| 417 | .d_reserved_1 = eno_getc, |
| 418 | .d_reserved_2 = eno_putc, |
| 419 | }; |
| 420 | |
| 421 | |
| 422 | #include <mach-o/nlist.h> |
| 423 | #include <libkern/kernel_mach_header.h> |
| 424 | |
| 425 | /* |
| 426 | * Represents single record in __DATA_CONST,__sdt section. |
| 427 | */ |
| 428 | typedef struct dtrace_sdt_def { |
| 429 | uintptr_t dsd_addr; /* probe site location */ |
| 430 | const char *dsd_prov; /* provider's name */ |
| 431 | const char *dsd_name; /* probe's name */ |
| 432 | } __attribute__((__packed__)) dtrace_sdt_def_t; |
| 433 | |
| 434 | /* |
| 435 | * Creates a copy of name and unescapes '-' characters. |
| 436 | */ |
| 437 | static char * |
| 438 | sdt_strdup_name(const char *name) |
| 439 | { |
| 440 | size_t len = strlen(s: name) + 1; |
| 441 | size_t i, j; |
| 442 | char *nname = kmem_alloc(len, KM_SLEEP); |
| 443 | |
| 444 | for (i = 0, j = 0; name[j] != '\0'; i++) { |
| 445 | if (name[j] == '_' && name[j + 1] == '_') { |
| 446 | nname[i] = '-'; |
| 447 | j += 2; |
| 448 | } else { |
| 449 | nname[i] = name[j++]; |
| 450 | } |
| 451 | } |
| 452 | |
| 453 | nname[i] = '\0'; |
| 454 | return nname; |
| 455 | } |
| 456 | |
| 457 | /* |
| 458 | * Returns Mach-O header that should be used for given modctl. |
| 459 | */ |
| 460 | static kernel_mach_header_t * |
| 461 | sdt_get_module_mh(struct modctl *ctl) |
| 462 | { |
| 463 | kernel_mach_header_t *mh = (kernel_mach_header_t *)ctl->mod_address; |
| 464 | |
| 465 | /* Static KEXTs have their __sdt section merged into kernel's __sdt. */ |
| 466 | if (MOD_IS_STATIC_KEXT(ctl)) { |
| 467 | mh = &_mh_execute_header; |
| 468 | } |
| 469 | |
| 470 | if (mh->magic != MH_MAGIC_KERNEL) { |
| 471 | return NULL; |
| 472 | } |
| 473 | |
| 474 | return mh; |
| 475 | } |
| 476 | |
| 477 | /* |
| 478 | * Finds symbol table for given kernel module. |
| 479 | */ |
| 480 | static uint32_t |
| 481 | sdt_find_symbol_table(struct modctl *ctl, kernel_nlist_t **sym, char **strings) |
| 482 | { |
| 483 | kernel_mach_header_t *mh = sdt_get_module_mh(ctl); |
| 484 | struct load_command *cmd = (struct load_command *)&mh[1]; |
| 485 | kernel_segment_command_t *orig_le = NULL; |
| 486 | struct symtab_command *orig_st = NULL; |
| 487 | |
| 488 | for (int i = 0; i < mh->ncmds; i++) { |
| 489 | if (cmd->cmd == LC_SEGMENT_KERNEL) { |
| 490 | kernel_segment_command_t *orig_sg = (kernel_segment_command_t *) cmd; |
| 491 | |
| 492 | if (LIT_STRNEQL(orig_sg->segname, SEG_LINKEDIT)) { |
| 493 | orig_le = orig_sg; |
| 494 | } |
| 495 | } else if (cmd->cmd == LC_SYMTAB) { |
| 496 | orig_st = (struct symtab_command *) cmd; |
| 497 | } |
| 498 | |
| 499 | cmd = (struct load_command *) ((uintptr_t) cmd + cmd->cmdsize); |
| 500 | } |
| 501 | |
| 502 | if ((orig_st == NULL) || (orig_le == NULL)) { |
| 503 | return 0; |
| 504 | } |
| 505 | |
| 506 | *sym = (kernel_nlist_t *)(orig_le->vmaddr + orig_st->symoff - orig_le->fileoff); |
| 507 | *strings = (char *)(orig_le->vmaddr + orig_st->stroff - orig_le->fileoff); |
| 508 | |
| 509 | return orig_st->nsyms; |
| 510 | } |
| 511 | |
| 512 | /* Last kernel address. */ |
| 513 | static SECURITY_READ_ONLY_LATE(vm_address_t) kern_end = (vm_address_t)-1; |
| 514 | |
| 515 | void |
| 516 | sdt_early_init(void) |
| 517 | { |
| 518 | kernel_mach_header_t *mh = &_mh_execute_header; |
| 519 | kernel_section_t *sec_ks = NULL; |
| 520 | kc_format_t kc_format; |
| 521 | |
| 522 | if (!PE_get_primary_kc_format(type: &kc_format)) { |
| 523 | kc_format = KCFormatUnknown; |
| 524 | } |
| 525 | |
| 526 | /* |
| 527 | * Detects end of kernel's text in static kernel cache. It is the last text address before |
| 528 | * the first kext text section start. |
| 529 | */ |
| 530 | if (kc_format == KCFormatStatic) { |
| 531 | if ((sec_ks = getsectbynamefromheader(header: mh, seg_name: "__PRELINK_INFO" , sect_name: "__kmod_start" )) == NULL) { |
| 532 | printf("SDT: unable to find prelink info\n" ); |
| 533 | return; |
| 534 | } |
| 535 | |
| 536 | /* find the MIN(start_address) of all kexts in this image. */ |
| 537 | const uint64_t *start_addr = (const uint64_t *)sec_ks->addr; |
| 538 | for (int i = 0; i < sec_ks->size / sizeof(uint64_t); i++) { |
| 539 | if (kern_end > start_addr[i]) { |
| 540 | kern_end = start_addr[i]; |
| 541 | } |
| 542 | } |
| 543 | } |
| 544 | } |
| 545 | |
| 546 | /* |
| 547 | * Finds TEXT range that belongs to given module. |
| 548 | */ |
| 549 | static int |
| 550 | sdt_find_module_text_range(struct modctl *ctl, vm_address_t *start, vm_address_t *end) |
| 551 | { |
| 552 | kc_format_t kc_format; |
| 553 | |
| 554 | if (!PE_get_primary_kc_format(type: &kc_format)) { |
| 555 | kc_format = KCFormatUnknown; |
| 556 | } |
| 557 | |
| 558 | /* Adjust kernel region for static kernel cache. */ |
| 559 | *start = ctl->mod_address; |
| 560 | |
| 561 | if (MOD_IS_MACH_KERNEL(ctl) && kc_format == KCFormatStatic) { |
| 562 | *end = kern_end; |
| 563 | } else { |
| 564 | *end = ctl->mod_address + ctl->mod_size; |
| 565 | } |
| 566 | |
| 567 | return 1; |
| 568 | } |
| 569 | |
| 570 | /* |
| 571 | * Processes SDT section in given Mach-O header |
| 572 | */ |
| 573 | void |
| 574 | sdt_load_machsect(struct modctl *ctl) |
| 575 | { |
| 576 | kernel_mach_header_t *mh = sdt_get_module_mh(ctl); |
| 577 | kernel_section_t *sec_sdt = NULL; |
| 578 | char *strings = NULL; |
| 579 | kernel_nlist_t *sym = NULL; |
| 580 | vm_address_t text_start, text_end; |
| 581 | unsigned int len; |
| 582 | uint32_t nsyms = 0; |
| 583 | |
| 584 | if (mh == NULL) { |
| 585 | return; |
| 586 | } |
| 587 | |
| 588 | /* Ignore SDT definitions if we don't know where they belong. */ |
| 589 | if (!sdt_find_module_text_range(ctl, start: &text_start, end: &text_end)) { |
| 590 | printf("SDT: Unable to determine text range for %s\n" , ctl->mod_modname); |
| 591 | return; |
| 592 | } |
| 593 | |
| 594 | /* Do not load SDTs when asked to use kernel symbols but symbol table is not available. */ |
| 595 | if (MOD_HAS_KERNEL_SYMBOLS(ctl) && (nsyms = sdt_find_symbol_table(ctl, sym: &sym, strings: &strings)) == 0) { |
| 596 | printf("SDT: No kernel symbols for %s\n" , ctl->mod_modname); |
| 597 | return; |
| 598 | } |
| 599 | |
| 600 | /* Locate DTrace SDT section in the object. */ |
| 601 | if ((sec_sdt = getsectbynamefromheader(header: mh, seg_name: "__DATA_CONST" , sect_name: "__sdt" )) == NULL) { |
| 602 | return; |
| 603 | } |
| 604 | |
| 605 | /* |
| 606 | * Iterate over SDT section and establish all SDT probe descriptions. |
| 607 | */ |
| 608 | dtrace_sdt_def_t *sdtdef = (dtrace_sdt_def_t *)(sec_sdt->addr); |
| 609 | for (size_t k = 0; k < sec_sdt->size / sizeof(dtrace_sdt_def_t); k++, sdtdef++) { |
| 610 | unsigned long best = 0; |
| 611 | |
| 612 | /* |
| 613 | * Static KEXTs share __sdt section with kernel after linking. It is required |
| 614 | * to filter out description and pick only those that belong to requested |
| 615 | * module or kernel itself. |
| 616 | */ |
| 617 | if (MOD_IS_STATIC_KEXT(ctl) || MOD_IS_MACH_KERNEL(ctl)) { |
| 618 | if ((sdtdef->dsd_addr < text_start) || (sdtdef->dsd_addr > text_end)) { |
| 619 | continue; |
| 620 | } |
| 621 | } else { |
| 622 | /* Skip over probe descripton that do not belong to current module. */ |
| 623 | if (!dtrace_addr_in_module((void *)sdtdef->dsd_addr, ctl)) { |
| 624 | continue; |
| 625 | } |
| 626 | } |
| 627 | |
| 628 | sdt_probedesc_t *sdpd = kmem_alloc(sizeof(sdt_probedesc_t), KM_SLEEP); |
| 629 | |
| 630 | /* Unescape probe name and keep a note of the size of original memory allocation. */ |
| 631 | sdpd->sdpd_name = sdt_strdup_name(name: sdtdef->dsd_name); |
| 632 | sdpd->sdpd_namelen = strlen(s: sdtdef->dsd_name) + 1; |
| 633 | |
| 634 | /* Used only for provider structure lookup so there is no need to make dynamic copy. */ |
| 635 | sdpd->sdpd_prov = sdtdef->dsd_prov; |
| 636 | |
| 637 | /* |
| 638 | * Find the symbol immediately preceding the sdt probe site just discovered, |
| 639 | * that symbol names the function containing the sdt probe. |
| 640 | */ |
| 641 | sdpd->sdpd_func = NULL; |
| 642 | |
| 643 | if (MOD_HAS_KERNEL_SYMBOLS(ctl)) { |
| 644 | const char *funcname = SDT_UNKNOWN_FUNCNAME; |
| 645 | |
| 646 | for (int i = 0; i < nsyms; i++) { |
| 647 | uint8_t jn_type = sym[i].n_type & N_TYPE; |
| 648 | char *jname = strings + sym[i].n_un.n_strx; |
| 649 | |
| 650 | if ((N_SECT != jn_type && N_ABS != jn_type)) { |
| 651 | continue; |
| 652 | } |
| 653 | |
| 654 | if (0 == sym[i].n_un.n_strx) { /* iff a null, "", name. */ |
| 655 | continue; |
| 656 | } |
| 657 | |
| 658 | if (*jname == '_') { |
| 659 | jname += 1; |
| 660 | } |
| 661 | |
| 662 | if (sdtdef->dsd_addr <= (unsigned long)sym[i].n_value) { |
| 663 | continue; |
| 664 | } |
| 665 | |
| 666 | if ((unsigned long)sym[i].n_value > best) { |
| 667 | best = (unsigned long)sym[i].n_value; |
| 668 | funcname = jname; |
| 669 | } |
| 670 | } |
| 671 | |
| 672 | len = strlen(s: funcname) + 1; |
| 673 | sdpd->sdpd_func = kmem_alloc(len, KM_SLEEP); |
| 674 | (void) strlcpy(dst: sdpd->sdpd_func, src: funcname, n: len); |
| 675 | } |
| 676 | |
| 677 | #if defined(__arm64__) |
| 678 | sdpd->sdpd_offset = sdtdef->dsd_addr & ~0x1LU; |
| 679 | #else |
| 680 | sdpd->sdpd_offset = sdtdef->dsd_addr; |
| 681 | #endif /* __arm64__ */ |
| 682 | |
| 683 | sdpd->sdpd_next = (sdt_probedesc_t *)ctl->mod_sdtdesc; |
| 684 | ctl->mod_sdtdesc = sdpd; |
| 685 | } |
| 686 | } |
| 687 | |
| 688 | void |
| 689 | sdt_init( void ) |
| 690 | { |
| 691 | int majdevno = cdevsw_add(SDT_MAJOR, &sdt_cdevsw); |
| 692 | |
| 693 | if (majdevno < 0) { |
| 694 | printf("sdt_init: failed to allocate a major number!\n" ); |
| 695 | return; |
| 696 | } |
| 697 | |
| 698 | if (dtrace_sdt_probes_restricted()) { |
| 699 | return; |
| 700 | } |
| 701 | |
| 702 | sdt_attach(devi: (dev_info_t*)(uintptr_t)majdevno); |
| 703 | } |
| 704 | |
| 705 | #undef SDT_MAJOR |
| 706 | |
| 707 | /* |
| 708 | * Provide SDT modules with userspace symbols. |
| 709 | * |
| 710 | * A module contains only partially filled in SDT probe descriptions because symbols were |
| 711 | * not available at the time when __sdt section was loaded. Fixup descriptons before providing |
| 712 | * the probes. |
| 713 | */ |
| 714 | static void |
| 715 | sdt_provide_module_user_syms(void *arg, struct modctl *ctl) |
| 716 | { |
| 717 | sdt_probedesc_t *sdpd; |
| 718 | dtrace_module_symbols_t *mod_sym = ctl->mod_user_symbols; |
| 719 | |
| 720 | if (mod_sym == NULL) { |
| 721 | printf("DTrace missing userspace symbols for module %s\n" , ctl->mod_modname); |
| 722 | return; |
| 723 | } |
| 724 | |
| 725 | /* Fixup missing probe description parts. */ |
| 726 | for (sdpd = ctl->mod_sdtdesc; sdpd != NULL; sdpd = sdpd->sdpd_next) { |
| 727 | ASSERT(sdpd->sdpd_func == NULL); |
| 728 | const char *funcname = SDT_UNKNOWN_FUNCNAME; |
| 729 | |
| 730 | /* Look for symbol that contains SDT probe offset. */ |
| 731 | for (int i = 0; i < mod_sym->dtmodsyms_count; i++) { |
| 732 | dtrace_symbol_t *symbol = &mod_sym->dtmodsyms_symbols[i]; |
| 733 | char *name = symbol->dtsym_name; |
| 734 | |
| 735 | /* |
| 736 | * Every function symbol gets extra '_' prepended in the Mach-O symbol table. |
| 737 | * Strip it away to make a probe's function name match source code. |
| 738 | */ |
| 739 | if (*name == '_') { |
| 740 | name += 1; |
| 741 | } |
| 742 | |
| 743 | if (!symbol->dtsym_addr) { |
| 744 | continue; |
| 745 | } |
| 746 | |
| 747 | /* Ignore symbols that do not belong to this module. */ |
| 748 | if (!dtrace_addr_in_module((void *)symbol->dtsym_addr, ctl)) { |
| 749 | continue; |
| 750 | } |
| 751 | |
| 752 | /* Pick symbol name when we found match. */ |
| 753 | if ((symbol->dtsym_addr <= sdpd->sdpd_offset) && |
| 754 | (sdpd->sdpd_offset < symbol->dtsym_addr + symbol->dtsym_size)) { |
| 755 | funcname = name; |
| 756 | break; |
| 757 | } |
| 758 | } |
| 759 | |
| 760 | size_t len = strlen(s: funcname) + 1; |
| 761 | sdpd->sdpd_func = kmem_alloc(len, KM_SLEEP); |
| 762 | (void) strlcpy(dst: sdpd->sdpd_func, src: funcname, n: len); |
| 763 | } |
| 764 | |
| 765 | /* Probe descriptionds are now fixed up. Provide them as usual. */ |
| 766 | __sdt_provide_module(arg, ctl); |
| 767 | } |
| 768 | |
| 769 | /*ARGSUSED*/ |
| 770 | void |
| 771 | sdt_provide_module(void *arg, struct modctl *ctl) |
| 772 | { |
| 773 | ASSERT(ctl != NULL); |
| 774 | ASSERT(dtrace_kernel_symbol_mode != DTRACE_KERNEL_SYMBOLS_NEVER); |
| 775 | LCK_MTX_ASSERT(&mod_lock, LCK_MTX_ASSERT_OWNED); |
| 776 | |
| 777 | if (MOD_SDT_DONE(ctl)) { |
| 778 | return; |
| 779 | } |
| 780 | |
| 781 | if (MOD_HAS_KERNEL_SYMBOLS(ctl)) { |
| 782 | __sdt_provide_module(arg, ctl); |
| 783 | ctl->mod_flags |= MODCTL_SDT_PROBES_PROVIDED; |
| 784 | return; |
| 785 | } |
| 786 | |
| 787 | if (MOD_HAS_USERSPACE_SYMBOLS(ctl)) { |
| 788 | sdt_provide_module_user_syms(arg, ctl); |
| 789 | ctl->mod_flags |= MODCTL_SDT_PROBES_PROVIDED; |
| 790 | return; |
| 791 | } |
| 792 | |
| 793 | /* |
| 794 | * The SDT provider's module is not detachable so we don't have to re-provide SDT |
| 795 | * probes if that happens. After succesfull providing, the probe descriptions are |
| 796 | * no longer required. If module gets re-loaded it will get a new set of probe |
| 797 | * descriptions from its __sdt section. |
| 798 | */ |
| 799 | if (MOD_SDT_PROBES_PROVIDED(ctl)) { |
| 800 | sdt_probedesc_t *sdpd = ctl->mod_sdtdesc; |
| 801 | while (sdpd) { |
| 802 | sdt_probedesc_t *this_sdpd = sdpd; |
| 803 | kmem_free((void *)sdpd->sdpd_name, sdpd->sdpd_namelen); |
| 804 | if (sdpd->sdpd_func) { |
| 805 | kmem_free((void *)sdpd->sdpd_func, strlen(s: sdpd->sdpd_func) + 1); |
| 806 | } |
| 807 | sdpd = sdpd->sdpd_next; |
| 808 | kmem_free((void *)this_sdpd, sizeof(sdt_probedesc_t)); |
| 809 | } |
| 810 | ctl->mod_sdtdesc = NULL; |
| 811 | } |
| 812 | } |
| 813 | |