1#include <darwintest.h> 2#include <darwintest_utils.h> 3#include <darwintest_multiprocess.h> 4#include <kern/debug.h> 5#include <kern/kern_cdata.h> 6#include <kern/block_hint.h> 7#include <kdd.h> 8#include <libproc.h> 9#include <os/atomic_private.h> 10#include <mach-o/dyld.h> 11#include <mach-o/dyld_images.h> 12#include <mach-o/dyld_priv.h> 13#include <sys/syscall.h> 14#include <sys/stackshot.h> 15#include <uuid/uuid.h> 16#include <servers/bootstrap.h> 17#include <pthread/workqueue_private.h> 18#include <dispatch/private.h> 19#include <stdalign.h> 20#include <TargetConditionals.h> 21 22#import <zlib.h> 23#import <IOKit/IOKitLib.h> 24#import <IOKit/IOKitLibPrivate.h> 25#import <IOKit/IOKitKeysPrivate.h> 26#import "test_utils.h" 27 28 29 30T_GLOBAL_META( 31 T_META_NAMESPACE("xnu.stackshot"), 32 T_META_RADAR_COMPONENT_NAME("xnu"), 33 T_META_RADAR_COMPONENT_VERSION("stackshot"), 34 T_META_OWNER("jonathan_w_adams"), 35 T_META_CHECK_LEAKS(false), 36 T_META_ASROOT(true), 37 XNU_T_META_SOC_SPECIFIC 38 ); 39 40static const char *current_process_name(void); 41static void verify_stackshot_sharedcache_layout(struct dyld_uuid_info_64 *uuids, uint32_t uuid_count); 42static void parse_stackshot(uint64_t stackshot_parsing_flags, void *ssbuf, size_t sslen, NSDictionary *extra); 43static void parse_thread_group_stackshot(void **sbuf, size_t sslen); 44static uint64_t stackshot_timestamp(void *ssbuf, size_t sslen); 45static void initialize_thread(void); 46 47static uint64_t global_flags = 0; 48 49#define DEFAULT_STACKSHOT_BUFFER_SIZE (1024 * 1024) 50#define MAX_STACKSHOT_BUFFER_SIZE (6 * 1024 * 1024) 51 52#define SRP_SERVICE_NAME "com.apple.xnu.test.stackshot.special_reply_port" 53 54/* bit flags for parse_stackshot */ 55#define PARSE_STACKSHOT_DELTA 0x01 56#define PARSE_STACKSHOT_ZOMBIE 0x02 57#define PARSE_STACKSHOT_SHAREDCACHE_LAYOUT 0x04 58#define PARSE_STACKSHOT_DISPATCH_QUEUE_LABEL 0x08 59#define PARSE_STACKSHOT_TURNSTILEINFO 0x10 60#define PARSE_STACKSHOT_POSTEXEC 0x20 61#define PARSE_STACKSHOT_WAITINFO_CSEG 0x40 62#define PARSE_STACKSHOT_WAITINFO_SRP 0x80 63#define PARSE_STACKSHOT_TRANSLATED 0x100 64#define PARSE_STACKSHOT_SHAREDCACHE_FLAGS 0x200 65#define PARSE_STACKSHOT_EXEC_INPROGRESS 0x400 66#define PARSE_STACKSHOT_TRANSITIONING 0x800 67#define PARSE_STACKSHOT_ASYNCSTACK 0x1000 68#define PARSE_STACKSHOT_COMPACTINFO 0x2000 /* TODO: rdar://88789261 */ 69#define PARSE_STACKSHOT_DRIVERKIT 0x4000 70#define PARSE_STACKSHOT_THROTTLED_SP 0x8000 71#define PARSE_STACKSHOT_SUSPENDINFO 0x10000 72#define PARSE_STACKSHOT_TARGETPID 0x20000 73 74/* keys for 'extra' dictionary for parse_stackshot */ 75static const NSString* zombie_child_pid_key = @"zombie_child_pid"; // -> @(pid), required for PARSE_STACKSHOT_ZOMBIE 76static const NSString* postexec_child_unique_pid_key = @"postexec_child_unique_pid"; // -> @(unique_pid), required for PARSE_STACKSHOT_POSTEXEC 77static const NSString* cseg_expected_threadid_key = @"cseg_expected_threadid"; // -> @(tid), required for PARSE_STACKSHOT_WAITINFO_CSEG 78static const NSString* srp_expected_threadid_key = @"srp_expected_threadid"; // -> @(tid), this or ..._pid required for PARSE_STACKSHOT_WAITINFO_SRP 79static const NSString* srp_expected_pid_key = @"srp_expected_pid"; // -> @(pid), this or ..._threadid required for PARSE_STACKSHOT_WAITINFO_SRP 80static const NSString* translated_child_pid_key = @"translated_child_pid"; // -> @(pid), required for PARSE_STACKSHOT_TRANSLATED 81static const NSString* sharedcache_child_pid_key = @"sharedcache_child_pid"; // @(pid), required for PARSE_STACKSHOT_SHAREDCACHE_FLAGS 82static const NSString* sharedcache_child_sameaddr_key = @"sharedcache_child_sameaddr"; // @(0 or 1), required for PARSE_STACKSHOT_SHAREDCACHE_FLAGS 83static const NSString* exec_inprogress_pid_key = @"exec_inprogress_pid"; 84static const NSString* exec_inprogress_found_key = @"exec_inprogress_found"; // callback when inprogress is found 85static const NSString* transitioning_pid_key = @"transitioning_task_pid"; // -> @(pid), required for PARSE_STACKSHOT_TRANSITIONING 86static const NSString* asyncstack_expected_threadid_key = @"asyncstack_expected_threadid"; // -> @(tid), required for PARSE_STACKSHOT_ASYNCSTACK 87static const NSString* asyncstack_expected_stack_key = @"asyncstack_expected_stack"; // -> @[pc...]), expected PCs for asyncstack 88static const NSString* driverkit_found_key = @"driverkit_found_key"; // callback when driverkit process is found. argument is the process pid. 89static const NSString* sp_throttled_expected_ctxt_key = @"sp_throttled_expected_ctxt_key"; // -> @(ctxt), required for PARSE_STACKSHOT_THROTTLED_SP 90static const NSString* sp_throttled_expect_flag = @"sp_throttled_expect_flag"; // -> @(is_throttled), required for PARSE_STACKSHOT_THROTTLED_SP 91static const NSString* no_exclaves_key = @"no_exclaves"; 92 93#define TEST_STACKSHOT_QUEUE_LABEL "houston.we.had.a.problem" 94#define TEST_STACKSHOT_QUEUE_LABEL_LENGTH sizeof(TEST_STACKSHOT_QUEUE_LABEL) 95 96#define THROTTLED_SERVICE_NAME "com.apple.xnu.test.stackshot.throttled_service" 97 98static int64_t 99run_sysctl_test(const char *t, int64_t value) 100{ 101 char name[1024]; 102 int64_t result = 0; 103 size_t s = sizeof(value); 104 int rc; 105 106 snprintf(name, sizeof(name), "debug.test.%s", t); 107 rc = sysctlbyname(name, &result, &s, &value, s); 108 T_QUIET; T_ASSERT_POSIX_SUCCESS(rc, "sysctlbyname(%s)", name); 109 return result; 110} 111 112T_DECL(microstackshots, "test the microstackshot syscall", T_META_TAG_VM_PREFERRED) 113{ 114 void *buf = NULL; 115 unsigned int size = DEFAULT_STACKSHOT_BUFFER_SIZE; 116 117 while (1) { 118 buf = malloc(size); 119 T_QUIET; T_ASSERT_NOTNULL(buf, "allocated stackshot buffer"); 120 121#pragma clang diagnostic push 122#pragma clang diagnostic ignored "-Wdeprecated-declarations" 123 int len = syscall(SYS_microstackshot, buf, size, 124 (uint32_t) STACKSHOT_GET_MICROSTACKSHOT); 125#pragma clang diagnostic pop 126 if (len == ENOSYS) { 127 T_SKIP("microstackshot syscall failed, likely not compiled with CONFIG_TELEMETRY"); 128 } 129 if (len == -1 && errno == ENOSPC) { 130 /* syscall failed because buffer wasn't large enough, try again */ 131 free(buf); 132 buf = NULL; 133 size *= 2; 134 T_ASSERT_LE(size, (unsigned int)MAX_STACKSHOT_BUFFER_SIZE, 135 "growing stackshot buffer to sane size"); 136 continue; 137 } 138 T_ASSERT_POSIX_SUCCESS(len, "called microstackshot syscall"); 139 break; 140 } 141 142 T_EXPECT_EQ(*(uint32_t *)buf, 143 (uint32_t)STACKSHOT_MICRO_SNAPSHOT_MAGIC, 144 "magic value for microstackshot matches"); 145 146 free(buf); 147} 148 149struct scenario { 150 const char *name; 151 uint64_t flags; 152 bool quiet; 153 bool should_fail; 154 bool maybe_unsupported; 155 bool maybe_enomem; 156 bool no_recordfile; 157 pid_t target_pid; 158 bool target_kernel; 159 uint64_t since_timestamp; 160 uint32_t size_hint; 161 dt_stat_time_t timer; 162}; 163 164static void 165quiet(struct scenario *scenario) 166{ 167 if (scenario->timer || scenario->quiet) { 168 T_QUIET; 169 } 170} 171 172static void 173take_stackshot(struct scenario *scenario, bool compress_ok, void (^cb)(void *buf, size_t size)) 174{ 175start: 176 initialize_thread(); 177 178 void *config = stackshot_config_create(); 179 quiet(scenario); 180 T_ASSERT_NOTNULL(config, "created stackshot config"); 181 182 int ret = stackshot_config_set_flags(config, scenario->flags | global_flags); 183 quiet(scenario); 184 T_ASSERT_POSIX_ZERO(ret, "set flags %#llx on stackshot config", scenario->flags); 185 186 if (scenario->size_hint > 0) { 187 ret = stackshot_config_set_size_hint(config, scenario->size_hint); 188 quiet(scenario); 189 T_ASSERT_POSIX_ZERO(ret, "set size hint %" PRIu32 " on stackshot config", 190 scenario->size_hint); 191 } 192 193 if (scenario->target_pid > 0) { 194 ret = stackshot_config_set_pid(config, scenario->target_pid); 195 quiet(scenario); 196 T_ASSERT_POSIX_ZERO(ret, "set target pid %d on stackshot config", 197 scenario->target_pid); 198 } else if (scenario->target_kernel) { 199 ret = stackshot_config_set_pid(config, 0); 200 quiet(scenario); 201 T_ASSERT_POSIX_ZERO(ret, "set kernel target on stackshot config"); 202 } 203 204 if (scenario->since_timestamp > 0) { 205 ret = stackshot_config_set_delta_timestamp(config, scenario->since_timestamp); 206 quiet(scenario); 207 T_ASSERT_POSIX_ZERO(ret, "set since timestamp %" PRIu64 " on stackshot config", 208 scenario->since_timestamp); 209 } 210 211 int retries_remaining = 5; 212 213retry: ; 214 uint64_t start_time = mach_absolute_time(); 215 ret = stackshot_capture_with_config(config); 216 uint64_t end_time = mach_absolute_time(); 217 218 if (scenario->should_fail) { 219 T_EXPECTFAIL; 220 T_ASSERT_POSIX_ZERO(ret, "called stackshot_capture_with_config"); 221 return; 222 } 223 224 if (ret == EBUSY || ret == ETIMEDOUT) { 225 if (retries_remaining > 0) { 226 if (!scenario->timer) { 227 T_LOG("stackshot_capture_with_config failed with %s (%d), retrying", 228 strerror(ret), ret); 229 } 230 231 retries_remaining--; 232 goto retry; 233 } else { 234 T_ASSERT_POSIX_ZERO(ret, 235 "called stackshot_capture_with_config (no retries remaining)"); 236 } 237 } else if ((ret == ENOTSUP) && scenario->maybe_unsupported) { 238 T_SKIP("kernel indicated this stackshot configuration is not supported"); 239 } else if ((ret == ENOMEM) && scenario->maybe_enomem) { 240 T_SKIP("insufficient available memory to run test"); 241 } else { 242 quiet(scenario); 243 T_ASSERT_POSIX_ZERO(ret, "called stackshot_capture_with_config"); 244 } 245 246 if (scenario->timer) { 247 dt_stat_mach_time_add(scenario->timer, end_time - start_time); 248 } 249 void *buf = stackshot_config_get_stackshot_buffer(config); 250 size_t size = stackshot_config_get_stackshot_size(config); 251 if (scenario->name && !scenario->no_recordfile) { 252 char sspath[MAXPATHLEN]; 253 strlcpy(sspath, scenario->name, sizeof(sspath)); 254 strlcat(sspath, ".kcdata", sizeof(sspath)); 255 T_QUIET; T_ASSERT_POSIX_ZERO(dt_resultfile(sspath, sizeof(sspath)), 256 "create result file path"); 257 258 if (!scenario->quiet) { 259 T_LOG("writing stackshot to %s", sspath); 260 } 261 262 FILE *f = fopen(sspath, "w"); 263 T_WITH_ERRNO; T_QUIET; T_ASSERT_NOTNULL(f, 264 "open stackshot output file"); 265 266 size_t written = fwrite(buf, size, 1, f); 267 T_QUIET; T_ASSERT_POSIX_SUCCESS(written, "wrote stackshot to file"); 268 269 fclose(f); 270 } 271 cb(buf, size); 272 if (compress_ok) { 273 if (global_flags == 0) { 274 T_LOG("Restarting test with compression"); 275 global_flags |= STACKSHOT_DO_COMPRESS; 276 goto start; 277 } else { 278 global_flags = 0; 279 } 280 } 281 282 ret = stackshot_config_dealloc(config); 283 T_QUIET; T_EXPECT_POSIX_ZERO(ret, "deallocated stackshot config"); 284} 285 286T_DECL(simple_compressed, "take a simple compressed stackshot", T_META_TAG_VM_PREFERRED) 287{ 288 struct scenario scenario = { 289 .name = "kcdata_compressed", 290 .flags = (STACKSHOT_DO_COMPRESS | STACKSHOT_SAVE_LOADINFO | STACKSHOT_THREAD_WAITINFO | STACKSHOT_GET_GLOBAL_MEM_STATS | 291 STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 292 }; 293 294 T_LOG("taking compressed kcdata stackshot"); 295 take_stackshot(&scenario, true, ^(void *ssbuf, size_t sslen) { 296 parse_stackshot(0, ssbuf, sslen, nil); 297 }); 298} 299 300T_DECL(panic_compressed, "take a compressed stackshot with the same flags as a panic stackshot", T_META_TAG_VM_PREFERRED) 301{ 302 uint64_t stackshot_flags = (STACKSHOT_SAVE_KEXT_LOADINFO | 303 STACKSHOT_SAVE_LOADINFO | 304 STACKSHOT_KCDATA_FORMAT | 305 STACKSHOT_ENABLE_BT_FAULTING | 306 STACKSHOT_ENABLE_UUID_FAULTING | 307 STACKSHOT_DO_COMPRESS | 308 STACKSHOT_NO_IO_STATS | 309 STACKSHOT_THREAD_WAITINFO | 310#if TARGET_OS_MAC 311 STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT | 312#endif 313 STACKSHOT_DISABLE_LATENCY_INFO); 314 315 struct scenario scenario = { 316 .name = "kcdata_panic_compressed", 317 .flags = stackshot_flags, 318 }; 319 320 T_LOG("taking compressed kcdata stackshot with panic flags"); 321 take_stackshot(&scenario, true, ^(void *ssbuf, size_t sslen) { 322 parse_stackshot(0, ssbuf, sslen, nil); 323 }); 324} 325 326T_DECL(kcdata, "test that kcdata stackshots can be taken and parsed", T_META_TAG_VM_PREFERRED) 327{ 328 struct scenario scenario = { 329 .name = "kcdata", 330 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS | 331 STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 332 }; 333 334 T_LOG("taking kcdata stackshot"); 335 take_stackshot(&scenario, true, ^(void *ssbuf, size_t sslen) { 336 parse_stackshot(0, ssbuf, sslen, nil); 337 }); 338} 339 340static void 341get_stats(stackshot_stats_t *_Nonnull out) 342{ 343 size_t oldlen = sizeof (*out); 344 bzero(out, oldlen); 345 int result = sysctlbyname("kern.stackshot_stats", out, &oldlen, NULL, 0); 346 T_WITH_ERRNO; T_ASSERT_POSIX_SUCCESS(result, "reading \"kern.stackshot_stats\" sysctl should succeed"); 347 T_EXPECT_EQ(oldlen, sizeof (*out), "kernel should update full stats structure"); 348} 349 350static void 351log_stats(mach_timebase_info_data_t timebase, uint64_t now, const char *name, stackshot_stats_t stat) 352{ 353 uint64_t last_ago = (now - stat.ss_last_start) * timebase.numer / timebase.denom; 354 uint64_t last_duration = (stat.ss_last_end - stat.ss_last_start) * timebase.numer / timebase.denom; 355 uint64_t total_duration = (stat.ss_duration) * timebase.numer / timebase.denom; 356 357 uint64_t nanosec = 1000000000llu; 358 T_LOG("%s: %8lld stackshots, %10lld.%09lld total nsecs, last %lld.%09lld secs ago, %lld.%09lld secs long", 359 name, stat.ss_count, 360 total_duration / nanosec, total_duration % nanosec, 361 last_ago / nanosec, last_ago % nanosec, 362 last_duration / nanosec, last_duration % nanosec); 363} 364 365T_DECL(stats, "test that stackshot stats can be read out and change when a stackshot occurs", T_META_TAG_VM_PREFERRED) 366{ 367 mach_timebase_info_data_t timebase = {0, 0}; 368 mach_timebase_info(&timebase); 369 370 struct scenario scenario = { 371 .name = "kcdata", 372 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT), 373 }; 374 375 stackshot_stats_t pre, post; 376 377 get_stats(&pre); 378 379 T_LOG("taking kcdata stackshot"); 380 take_stackshot(&scenario, true, ^(__unused void *ssbuf, __unused size_t sslen) { 381 (void)0; 382 }); 383 384 get_stats(&post); 385 386 uint64_t now = mach_absolute_time(); 387 388 log_stats(timebase, now, " pre", pre); 389 log_stats(timebase, now, " post", post); 390 391 int64_t delta_stackshots = (int64_t)(post.ss_count - pre.ss_count); 392 int64_t delta_duration = (int64_t)(post.ss_duration - pre.ss_duration) * (int64_t)timebase.numer / (int64_t)timebase.denom; 393 int64_t delta_nsec = delta_duration % 1000000000ll; 394 if (delta_nsec < 0) { 395 delta_nsec += 1000000000ll; 396 } 397 T_LOG("delta: %+8lld stackshots, %+10lld.%09lld total nsecs", delta_stackshots, delta_duration / 1000000000ll, delta_nsec); 398 399 T_EXPECT_LT(pre.ss_last_start, pre.ss_last_end, "pre: stackshot should take time"); 400 T_EXPECT_LT(pre.ss_count, post.ss_count, "stackshot count should increase when a stackshot is taken"); 401 T_EXPECT_LT(pre.ss_duration, post.ss_duration, "stackshot duration should increase when a stackshot is taken"); 402 T_EXPECT_LT(pre.ss_last_end, post.ss_last_start, "previous end should be less than new start after a stackshot"); 403 T_EXPECT_LT(post.ss_last_start, post.ss_last_end, "post: stackshot should take time"); 404} 405 406T_DECL(kcdata_faulting, "test that kcdata stackshots while faulting can be taken and parsed", T_META_TAG_VM_PREFERRED) 407{ 408 struct scenario scenario = { 409 .name = "faulting", 410 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 411 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT 412 | STACKSHOT_ENABLE_BT_FAULTING | STACKSHOT_ENABLE_UUID_FAULTING), 413 }; 414 415 T_LOG("taking faulting stackshot"); 416 take_stackshot(&scenario, true, ^(void *ssbuf, size_t sslen) { 417 parse_stackshot(0, ssbuf, sslen, nil); 418 }); 419} 420 421T_DECL(bad_flags, "test a poorly-formed stackshot syscall", T_META_TAG_VM_PREFERRED) 422{ 423 struct scenario scenario = { 424 .flags = STACKSHOT_SAVE_IN_KERNEL_BUFFER /* not allowed from user space */, 425 .should_fail = true, 426 }; 427 428 T_LOG("attempting to take stackshot with kernel-only flag"); 429 take_stackshot(&scenario, true, ^(__unused void *ssbuf, __unused size_t sslen) { 430 T_ASSERT_FAIL("stackshot data callback called"); 431 }); 432} 433 434T_DECL(delta, "test delta stackshots", T_META_TAG_VM_PREFERRED) 435{ 436 struct scenario scenario = { 437 .name = "delta", 438 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 439 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 440 }; 441 442 T_LOG("taking full stackshot"); 443 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 444 uint64_t stackshot_time = stackshot_timestamp(ssbuf, sslen); 445 446 T_LOG("taking delta stackshot since time %" PRIu64, stackshot_time); 447 448 parse_stackshot(0, ssbuf, sslen, nil); 449 450 struct scenario delta_scenario = { 451 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 452 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT 453 | STACKSHOT_COLLECT_DELTA_SNAPSHOT), 454 .since_timestamp = stackshot_time 455 }; 456 457 take_stackshot(&delta_scenario, false, ^(void *dssbuf, size_t dsslen) { 458 parse_stackshot(PARSE_STACKSHOT_DELTA, dssbuf, dsslen, nil); 459 }); 460 }); 461} 462 463T_DECL(shared_cache_layout, "test stackshot inclusion of shared cache layout", T_META_TAG_VM_PREFERRED) 464{ 465 struct scenario scenario = { 466 .name = "shared_cache_layout", 467 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 468 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT | 469 STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT), 470 }; 471 472 size_t shared_cache_length; 473 const void *cache_header = _dyld_get_shared_cache_range(&shared_cache_length); 474 if (cache_header == NULL) { 475 T_SKIP("Device not running with shared cache, skipping test..."); 476 } 477 478 if (shared_cache_length == 0) { 479 T_SKIP("dyld reports that currently running shared cache has zero length"); 480 } 481 482 T_LOG("taking stackshot with STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT set"); 483 take_stackshot(&scenario, true, ^(void *ssbuf, size_t sslen) { 484 parse_stackshot(PARSE_STACKSHOT_SHAREDCACHE_LAYOUT, ssbuf, sslen, nil); 485 }); 486} 487 488T_DECL(stress, "test that taking stackshots for 60 seconds doesn't crash the system", T_META_TAG_VM_PREFERRED) 489{ 490 uint64_t max_diff_time = 60ULL /* seconds */ * 1000000000ULL; 491 uint64_t start_time; 492 493 struct scenario scenario = { 494 .name = "stress", 495 .quiet = true, 496 .flags = (STACKSHOT_KCDATA_FORMAT | 497 STACKSHOT_THREAD_WAITINFO | 498 STACKSHOT_SAVE_LOADINFO | 499 STACKSHOT_SAVE_KEXT_LOADINFO | 500 STACKSHOT_GET_GLOBAL_MEM_STATS | 501 STACKSHOT_SAVE_IMP_DONATION_PIDS | 502 STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT | 503 STACKSHOT_THREAD_GROUP | 504 STACKSHOT_SAVE_JETSAM_COALITIONS | 505 STACKSHOT_ASID | 506 STACKSHOT_EXCLAVES | 507 0), 508 }; 509 510 start_time = clock_gettime_nsec_np(CLOCK_MONOTONIC); 511 while (clock_gettime_nsec_np(CLOCK_MONOTONIC) - start_time < max_diff_time) { 512 take_stackshot(&scenario, false, ^(void * __unused ssbuf, 513 size_t __unused sslen) { 514 printf("."); 515 fflush(stdout); 516 }); 517 518 /* 519 * After the first stackshot, there's no point in continuing to 520 * write them to disk, and it wears down the SSDs. 521 */ 522 scenario.no_recordfile = true; 523 524 /* Leave some time for the testing infrastructure to catch up */ 525 usleep(10000); 526 527 } 528 printf("\n"); 529} 530 531T_DECL(dispatch_queue_label, "test that kcdata stackshots contain libdispatch queue labels", T_META_TAG_VM_PREFERRED) 532{ 533 struct scenario scenario = { 534 .name = "kcdata", 535 .flags = (STACKSHOT_GET_DQ | STACKSHOT_KCDATA_FORMAT), 536 }; 537 dispatch_semaphore_t child_ready_sem, parent_done_sem; 538 dispatch_queue_t dq; 539 540#if TARGET_OS_WATCH 541 T_SKIP("This test is flaky on watches: 51663346"); 542#endif 543 544 child_ready_sem = dispatch_semaphore_create(0); 545 T_QUIET; T_ASSERT_NOTNULL(child_ready_sem, "dqlabel child semaphore"); 546 547 parent_done_sem = dispatch_semaphore_create(0); 548 T_QUIET; T_ASSERT_NOTNULL(parent_done_sem, "dqlabel parent semaphore"); 549 550 dq = dispatch_queue_create(TEST_STACKSHOT_QUEUE_LABEL, NULL); 551 T_QUIET; T_ASSERT_NOTNULL(dq, "dispatch queue"); 552 553 /* start the helper thread */ 554 dispatch_async(dq, ^{ 555 dispatch_semaphore_signal(child_ready_sem); 556 557 dispatch_semaphore_wait(parent_done_sem, DISPATCH_TIME_FOREVER); 558 }); 559 560 /* block behind the child starting up */ 561 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 562 563 T_LOG("taking kcdata stackshot with libdispatch queue labels"); 564 take_stackshot(&scenario, true, ^(void *ssbuf, size_t sslen) { 565 parse_stackshot(PARSE_STACKSHOT_DISPATCH_QUEUE_LABEL, ssbuf, sslen, nil); 566 }); 567 568 dispatch_semaphore_signal(parent_done_sem); 569} 570 571#define CACHEADDR_ENV "STACKSHOT_TEST_DYLDADDR" 572T_HELPER_DECL(spawn_reslide_child, "child process to spawn with alternate slide") 573{ 574 size_t shared_cache_len; 575 const void *addr, *prevaddr; 576 uintmax_t v; 577 char *endptr; 578 579 const char *cacheaddr_env = getenv(CACHEADDR_ENV); 580 T_QUIET; T_ASSERT_NOTNULL(cacheaddr_env, "getenv("CACHEADDR_ENV")"); 581 errno = 0; 582 endptr = NULL; 583 v = strtoumax(cacheaddr_env, &endptr, 16); /* read hex value */ 584 T_WITH_ERRNO; T_QUIET; T_ASSERT_NE(v, 0l, "getenv(%s) = \"%s\" should be a non-zero hex number", CACHEADDR_ENV, cacheaddr_env); 585 T_QUIET; T_ASSERT_EQ(*endptr, 0, "getenv(%s) = \"%s\" endptr \"%s\" should be empty", CACHEADDR_ENV, cacheaddr_env, endptr); 586 587 prevaddr = (const void *)v; 588 addr = _dyld_get_shared_cache_range(&shared_cache_len); 589 T_QUIET; T_ASSERT_NOTNULL(addr, "shared cache address"); 590 591 T_QUIET; T_ASSERT_POSIX_SUCCESS(kill(getppid(), (addr == prevaddr) ? SIGUSR2 : SIGUSR1), "signaled parent to take stackshot"); 592 for (;;) { 593 (void) pause(); /* parent will kill -9 us */ 594 } 595} 596 597T_DECL(shared_cache_flags, "tests stackshot's task_ss_flags for the shared cache", T_META_TAG_VM_PREFERRED) 598{ 599 posix_spawnattr_t attr; 600 char *env_addr; 601 char path[PATH_MAX]; 602 __block bool child_same_addr = false; 603 604 uint32_t path_size = sizeof(path); 605 T_QUIET; T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 606 char *args[] = { path, "-n", "spawn_reslide_child", NULL }; 607 pid_t pid; 608 size_t shared_cache_len; 609 const void *addr; 610 611 dispatch_source_t child_diffsig_src, child_samesig_src; 612 dispatch_semaphore_t child_ready_sem = dispatch_semaphore_create(0); 613 T_QUIET; T_ASSERT_NOTNULL(child_ready_sem, "shared_cache child semaphore"); 614 615 dispatch_queue_t signal_processing_q = dispatch_queue_create("signal processing queue", NULL); 616 T_QUIET; T_ASSERT_NOTNULL(signal_processing_q, "signal processing queue"); 617 618 signal(SIGUSR1, SIG_IGN); 619 signal(SIGUSR2, SIG_IGN); 620 child_samesig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, signal_processing_q); 621 T_QUIET; T_ASSERT_NOTNULL(child_samesig_src, "dispatch_source_create (child_samesig_src)"); 622 child_diffsig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR2, 0, signal_processing_q); 623 T_QUIET; T_ASSERT_NOTNULL(child_diffsig_src, "dispatch_source_create (child_diffsig_src)"); 624 625 /* child will signal us depending on if their addr is the same or different */ 626 dispatch_source_set_event_handler(child_samesig_src, ^{ child_same_addr = false; dispatch_semaphore_signal(child_ready_sem); }); 627 dispatch_source_set_event_handler(child_diffsig_src, ^{ child_same_addr = true; dispatch_semaphore_signal(child_ready_sem); }); 628 dispatch_activate(child_samesig_src); 629 dispatch_activate(child_diffsig_src); 630 631 addr = _dyld_get_shared_cache_range(&shared_cache_len); 632 T_QUIET; T_ASSERT_NOTNULL(addr, "shared cache address"); 633 634 T_QUIET; T_ASSERT_POSIX_SUCCESS(asprintf(&env_addr, "%p", addr), "asprintf of env_addr succeeded"); 635 T_QUIET; T_ASSERT_POSIX_SUCCESS(setenv(CACHEADDR_ENV, env_addr, true), "setting "CACHEADDR_ENV" to %s", env_addr); 636 637 T_QUIET; T_ASSERT_POSIX_ZERO(posix_spawnattr_init(&attr), "posix_spawnattr_init"); 638 T_QUIET; T_ASSERT_POSIX_ZERO(posix_spawnattr_setflags(&attr, _POSIX_SPAWN_RESLIDE), "posix_spawnattr_setflags"); 639 int sp_ret = posix_spawn(&pid, path, NULL, &attr, args, environ); 640 T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", args[0], pid); 641 642 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 643 T_LOG("received signal from child (%s), capturing stackshot", child_same_addr ? "same shared cache addr" : "different shared cache addr"); 644 645 struct scenario scenario = { 646 .name = "shared_cache_flags", 647 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 648 | STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT 649 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 650 }; 651 652 take_stackshot(&scenario, false, ^( void *ssbuf, size_t sslen) { 653 int status; 654 /* First kill the child so we can reap it */ 655 T_QUIET; T_ASSERT_POSIX_SUCCESS(kill(pid, SIGKILL), "killing spawned process"); 656 T_QUIET; T_ASSERT_POSIX_SUCCESS(waitpid(pid, &status, 0), "waitpid on spawned child"); 657 T_QUIET; T_ASSERT_EQ(!!WIFSIGNALED(status), 1, "waitpid status should be signalled"); 658 T_QUIET; T_ASSERT_EQ(WTERMSIG(status), SIGKILL, "waitpid status should be SIGKILLed"); 659 660 parse_stackshot(PARSE_STACKSHOT_SHAREDCACHE_FLAGS, ssbuf, sslen, 661 @{sharedcache_child_pid_key: @(pid), sharedcache_child_sameaddr_key: @(child_same_addr ? 1 : 0)}); 662 }); 663} 664 665T_DECL(transitioning_tasks, "test that stackshot contains transitioning task info", T_META_BOOTARGS_SET("enable_proc_exit_lpexit_spin=1"), T_META_TAG_VM_PREFERRED) 666{ 667 int32_t sysctlValue = -1, numAttempts =0; 668 char path[PATH_MAX]; 669 uint32_t path_size = sizeof(path); 670 T_QUIET; T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 671 char *args[] = { path, "-n", "exec_child_preexec", NULL }; 672 673 dispatch_source_t child_sig_src; 674 dispatch_semaphore_t child_ready_sem = dispatch_semaphore_create(0); 675 T_QUIET; T_ASSERT_NOTNULL(child_ready_sem, "exec child semaphore"); 676 677 dispatch_queue_t signal_processing_q = dispatch_queue_create("signal processing queue", NULL); 678 T_QUIET; T_ASSERT_NOTNULL(signal_processing_q, "signal processing queue"); 679 680 pid_t pid; 681 682 signal(SIGUSR1, SIG_IGN); 683 child_sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, signal_processing_q); 684 T_QUIET; T_ASSERT_NOTNULL(child_sig_src, "dispatch_source_create (child_sig_src)"); 685 686 dispatch_source_set_event_handler(child_sig_src, ^{ dispatch_semaphore_signal(child_ready_sem); }); 687 dispatch_activate(child_sig_src); 688 689 T_ASSERT_POSIX_SUCCESS(sysctlbyname("debug.proc_exit_lpexit_spin_pid", NULL, NULL, &sysctlValue, sizeof(sysctlValue)), "set debug.proc_exit_lpexit_spin_pid=-1"); 690 691 int proc_exit_spin_pos = 0 ; 692 693 while (0 == sysctlbyname("debug.proc_exit_lpexit_spin_pos", NULL, NULL, &proc_exit_spin_pos, sizeof(proc_exit_spin_pos))) { 694 695 T_LOG(" ##### Testing while spinning in proc_exit at position %d ##### ", proc_exit_spin_pos); 696 697 int sp_ret = posix_spawn(&pid, args[0], NULL, NULL, args, NULL); 698 T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", args[0], pid); 699 700 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 701 702 struct proc_uniqidentifierinfo proc_info_data = { }; 703 int retval = proc_pidinfo(getpid(), PROC_PIDUNIQIDENTIFIERINFO, 0, &proc_info_data, sizeof(proc_info_data)); 704 T_QUIET; T_EXPECT_POSIX_SUCCESS(retval, "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO"); 705 T_QUIET; T_ASSERT_EQ_INT(retval, (int) sizeof(proc_info_data), "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO returned data"); 706 707 T_ASSERT_POSIX_SUCCESS(kill(pid, SIGUSR1), "signaled pre-exec child to exec"); 708 709 /* wait for a signal from post-exec child */ 710 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 711 712 T_ASSERT_POSIX_SUCCESS(sysctlbyname("debug.proc_exit_lpexit_spin_pid", NULL, NULL, &pid, sizeof(pid)), "set debug.proc_exit_lpexit_spin_pid = %d, ", pid); 713 714 T_ASSERT_POSIX_SUCCESS(kill(pid, SIGKILL), "kill post-exec child %d", pid); 715 716 sysctlValue = 0; 717 size_t len = sizeof(sysctlValue); 718 while (numAttempts < 5) { 719 T_ASSERT_POSIX_SUCCESS(sysctlbyname("debug.proc_exit_lpexit_spinning", &sysctlValue, &len, NULL, 0), "retrieve debug.proc_exit_lpexit_spinning"); 720 if (sysctlValue != 1) numAttempts++; 721 else break; 722 sleep(1); 723 } 724 725 T_ASSERT_EQ_UINT(sysctlValue, 1, "find spinning task in proc_exit()"); 726 727 struct scenario scenario = { 728 .name = "transitioning_tasks", 729 .flags = (STACKSHOT_KCDATA_FORMAT) 730 }; 731 732 take_stackshot(&scenario, false, ^( void *ssbuf, size_t sslen) { 733 parse_stackshot(PARSE_STACKSHOT_TRANSITIONING, ssbuf, sslen, @{transitioning_pid_key: @(pid)}); 734 735 // Kill the child 736 int sysctlValueB = -1; 737 T_ASSERT_POSIX_SUCCESS(sysctlbyname("debug.proc_exit_lpexit_spin_pid", NULL, NULL, &sysctlValueB, sizeof(sysctlValueB)), "set debug.proc_exit_lpexit_spin_pid=-1"); 738 sleep(1); 739 size_t blen = sizeof(sysctlValueB); 740 T_ASSERT_POSIX_SUCCESS(sysctlbyname("debug.proc_exit_lpexit_spinning", &sysctlValueB, &blen, NULL, 0), "retrieve debug.proc_exit_lpexit_spinning"); 741 T_ASSERT_EQ_UINT(sysctlValueB, 0, "make sure nothing is spining in proc_exit()"); 742 int status; 743 T_ASSERT_POSIX_SUCCESS(waitpid(pid, &status, 0), "waitpid on post-exec child"); 744 }); 745 746 proc_exit_spin_pos++; 747 } 748 749} 750 751static void *stuck_sysctl_thread(void *arg) { 752 int val = 1; 753 dispatch_semaphore_t child_thread_started = *(dispatch_semaphore_t *)arg; 754 755 dispatch_semaphore_signal(child_thread_started); 756 T_ASSERT_POSIX_SUCCESS(sysctlbyname("kern.wedge_thread", NULL, NULL, &val, sizeof(val)), "wedge child thread"); 757 758 return NULL; 759} 760 761T_HELPER_DECL(zombie_child, "child process to sample as a zombie") 762{ 763 pthread_t pthread; 764 dispatch_semaphore_t child_thread_started = dispatch_semaphore_create(0); 765 T_QUIET; T_ASSERT_NOTNULL(child_thread_started, "zombie child thread semaphore"); 766 767 /* spawn another thread to get stuck in the kernel, then call exit() to become a zombie */ 768 T_QUIET; T_ASSERT_POSIX_SUCCESS(pthread_create(&pthread, NULL, stuck_sysctl_thread, &child_thread_started), "pthread_create"); 769 770 dispatch_semaphore_wait(child_thread_started, DISPATCH_TIME_FOREVER); 771 772 /* sleep for a bit in the hope of ensuring that the other thread has called the sysctl before we signal the parent */ 773 usleep(100); 774 T_ASSERT_POSIX_SUCCESS(kill(getppid(), SIGUSR1), "signaled parent to take stackshot"); 775 776 exit(0); 777} 778 779T_DECL(zombie, "tests a stackshot of a zombie task with a thread stuck in the kernel", 780 T_META_ENABLED(false), /* test is too flaky to run by default, but transitioning_tasks covers this case as well */ 781 T_META_TAG_VM_PREFERRED) 782{ 783 char path[PATH_MAX]; 784 uint32_t path_size = sizeof(path); 785 T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 786 char *args[] = { path, "-n", "zombie_child", NULL }; 787 788 dispatch_source_t child_sig_src; 789 dispatch_semaphore_t child_ready_sem = dispatch_semaphore_create(0); 790 T_QUIET; T_ASSERT_NOTNULL(child_ready_sem, "zombie child semaphore"); 791 792 dispatch_queue_t signal_processing_q = dispatch_queue_create("signal processing queue", NULL); 793 T_QUIET; T_ASSERT_NOTNULL(signal_processing_q, "signal processing queue"); 794 795 pid_t pid; 796 797 T_LOG("spawning a child"); 798 799 signal(SIGUSR1, SIG_IGN); 800 child_sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, signal_processing_q); 801 T_QUIET; T_ASSERT_NOTNULL(child_sig_src, "dispatch_source_create (child_sig_src)"); 802 803 dispatch_source_set_event_handler(child_sig_src, ^{ dispatch_semaphore_signal(child_ready_sem); }); 804 dispatch_activate(child_sig_src); 805 806 int sp_ret = posix_spawn(&pid, args[0], NULL, NULL, args, NULL); 807 T_QUIET; T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", args[0], pid); 808 809 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 810 811 T_LOG("received signal from child, capturing stackshot"); 812 813 struct proc_bsdshortinfo bsdshortinfo; 814 int retval, iterations_to_wait = 10; 815 816 while (iterations_to_wait > 0) { 817 retval = proc_pidinfo(pid, PROC_PIDT_SHORTBSDINFO, 0, &bsdshortinfo, sizeof(bsdshortinfo)); 818 if ((retval == 0) && errno == ESRCH) { 819 T_LOG("unable to find child using proc_pidinfo, assuming zombie"); 820 break; 821 } 822 823 T_QUIET; T_WITH_ERRNO; T_ASSERT_GT(retval, 0, "proc_pidinfo(PROC_PIDT_SHORTBSDINFO) returned a value > 0"); 824 T_QUIET; T_ASSERT_EQ(retval, (int)sizeof(bsdshortinfo), "proc_pidinfo call for PROC_PIDT_SHORTBSDINFO returned expected size"); 825 826 if (bsdshortinfo.pbsi_flags & PROC_FLAG_INEXIT) { 827 T_LOG("child proc info marked as in exit"); 828 break; 829 } 830 831 iterations_to_wait--; 832 if (iterations_to_wait == 0) { 833 /* 834 * This will mark the test as failed but let it continue so we 835 * don't leave a process stuck in the kernel. 836 */ 837 T_FAIL("unable to discover that child is marked as exiting"); 838 } 839 840 /* Give the child a few more seconds to make it to exit */ 841 sleep(5); 842 } 843 844 /* Give the child some more time to make it through exit */ 845 sleep(10); 846 847 struct scenario scenario = { 848 .name = "zombie", 849 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 850 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 851 }; 852 853 take_stackshot(&scenario, false, ^( void *ssbuf, size_t sslen) { 854 /* First unwedge the child so we can reap it */ 855 int val = 1, status; 856 T_ASSERT_POSIX_SUCCESS(sysctlbyname("kern.unwedge_thread", NULL, NULL, &val, sizeof(val)), "unwedge child"); 857 858 T_QUIET; T_ASSERT_POSIX_SUCCESS(waitpid(pid, &status, 0), "waitpid on zombie child"); 859 860 parse_stackshot(PARSE_STACKSHOT_ZOMBIE, ssbuf, sslen, @{zombie_child_pid_key: @(pid)}); 861 }); 862} 863 864T_HELPER_DECL(exec_child_preexec, "child process pre-exec") 865{ 866 dispatch_queue_t signal_processing_q = dispatch_queue_create("signal processing queue", NULL); 867 T_QUIET; T_ASSERT_NOTNULL(signal_processing_q, "signal processing queue"); 868 869 signal(SIGUSR1, SIG_IGN); 870 dispatch_source_t parent_sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, signal_processing_q); 871 T_QUIET; T_ASSERT_NOTNULL(parent_sig_src, "dispatch_source_create (child_sig_src)"); 872 dispatch_source_set_event_handler(parent_sig_src, ^{ 873 874 // Parent took a timestamp then signaled us: exec into the next process 875 876 char path[PATH_MAX]; 877 uint32_t path_size = sizeof(path); 878 T_QUIET; T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 879 char *args[] = { path, "-n", "exec_child_postexec", NULL }; 880 881 T_QUIET; T_ASSERT_POSIX_ZERO(execve(args[0], args, NULL), "execing into exec_child_postexec"); 882 }); 883 dispatch_activate(parent_sig_src); 884 885 T_ASSERT_POSIX_SUCCESS(kill(getppid(), SIGUSR1), "signaled parent to take timestamp"); 886 887 sleep(100); 888 // Should never get here 889 T_FAIL("Received signal to exec from parent"); 890} 891 892T_HELPER_DECL(exec_child_postexec, "child process post-exec to sample") 893{ 894 T_ASSERT_POSIX_SUCCESS(kill(getppid(), SIGUSR1), "signaled parent to take stackshot"); 895 sleep(100); 896 // Should never get here 897 T_FAIL("Killed by parent"); 898} 899 900T_DECL(exec, "test getting full task snapshots for a task that execs", T_META_TAG_VM_PREFERRED) 901{ 902 char path[PATH_MAX]; 903 uint32_t path_size = sizeof(path); 904 T_QUIET; T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 905 char *args[] = { path, "-n", "exec_child_preexec", NULL }; 906 907 dispatch_source_t child_sig_src; 908 dispatch_semaphore_t child_ready_sem = dispatch_semaphore_create(0); 909 T_QUIET; T_ASSERT_NOTNULL(child_ready_sem, "exec child semaphore"); 910 911 dispatch_queue_t signal_processing_q = dispatch_queue_create("signal processing queue", NULL); 912 T_QUIET; T_ASSERT_NOTNULL(signal_processing_q, "signal processing queue"); 913 914 pid_t pid; 915 916 T_LOG("spawning a child"); 917 918 signal(SIGUSR1, SIG_IGN); 919 child_sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, signal_processing_q); 920 T_QUIET; T_ASSERT_NOTNULL(child_sig_src, "dispatch_source_create (child_sig_src)"); 921 922 dispatch_source_set_event_handler(child_sig_src, ^{ dispatch_semaphore_signal(child_ready_sem); }); 923 dispatch_activate(child_sig_src); 924 925 int sp_ret = posix_spawn(&pid, args[0], NULL, NULL, args, NULL); 926 T_QUIET; T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", args[0], pid); 927 928 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 929 uint64_t start_time = mach_absolute_time(); 930 931 struct proc_uniqidentifierinfo proc_info_data = { }; 932 int retval = proc_pidinfo(getpid(), PROC_PIDUNIQIDENTIFIERINFO, 0, &proc_info_data, sizeof(proc_info_data)); 933 T_QUIET; T_EXPECT_POSIX_SUCCESS(retval, "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO"); 934 T_QUIET; T_ASSERT_EQ_INT(retval, (int) sizeof(proc_info_data), "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO returned data"); 935 uint64_t unique_pid = proc_info_data.p_uniqueid; 936 937 T_LOG("received signal from pre-exec child, unique_pid is %llu, timestamp is %llu", unique_pid, start_time); 938 939 T_ASSERT_POSIX_SUCCESS(kill(pid, SIGUSR1), "signaled pre-exec child to exec"); 940 941 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 942 943 T_LOG("received signal from post-exec child, capturing stackshot"); 944 945 struct scenario scenario = { 946 .name = "exec", 947 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 948 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT 949 | STACKSHOT_COLLECT_DELTA_SNAPSHOT), 950 .since_timestamp = start_time 951 }; 952 953 take_stackshot(&scenario, false, ^( void *ssbuf, size_t sslen) { 954 // Kill the child 955 int status; 956 T_ASSERT_POSIX_SUCCESS(kill(pid, SIGKILL), "kill post-exec child %d", pid); 957 T_ASSERT_POSIX_SUCCESS(waitpid(pid, &status, 0), "waitpid on post-exec child"); 958 959 parse_stackshot(PARSE_STACKSHOT_POSTEXEC | PARSE_STACKSHOT_DELTA, ssbuf, sslen, @{postexec_child_unique_pid_key: @(unique_pid)}); 960 }); 961} 962 963T_DECL( 964 exec_inprogress, 965 "test stackshots of processes in the middle of exec", 966 T_META_ENABLED(false), /* rdar://111691318 */ 967 T_META_TAG_VM_PREFERRED) 968{ 969 pid_t pid; 970 /* a BASH quine which execs itself as long as the parent doesn't exit */ 971 char *bash_prog = "[[ $PPID -ne 1 ]] && exec /bin/bash -c \"$0\" \"$0\""; 972 char *args[] = { "/bin/bash", "-c", bash_prog, bash_prog, NULL }; 973 974 posix_spawnattr_t sattr; 975 T_ASSERT_POSIX_ZERO(posix_spawnattr_init(&sattr), "posix_spawnattr_init"); 976 T_ASSERT_POSIX_ZERO(posix_spawn(&pid, args[0], NULL, &sattr, args, NULL), "spawn exec_inprogress_child"); 977 978 struct scenario scenario = { 979 .name = "exec_inprogress", 980 .flags = (STACKSHOT_KCDATA_FORMAT), 981 .target_pid = pid, 982 }; 983 984 int tries = 0; 985 int tries_limit = 30; 986 __block bool found = false; 987 __block uint64_t cid1 = 0, cid2 = 0; 988 989 for (tries = 0; !found && tries < tries_limit; tries++) { 990 take_stackshot(&scenario, false, 991 ^( void *ssbuf, size_t sslen) { 992 parse_stackshot(PARSE_STACKSHOT_EXEC_INPROGRESS | PARSE_STACKSHOT_TARGETPID, 993 ssbuf, sslen, @{ 994 exec_inprogress_pid_key: @(pid), 995 exec_inprogress_found_key: ^(uint64_t id1, uint64_t id2) { found = true; cid1 = id1; cid2 = id2; }}); 996 }); 997 } 998 T_QUIET; T_ASSERT_POSIX_SUCCESS(kill(pid, SIGKILL), "killing exec loop"); 999 T_ASSERT_TRUE(found, "able to find our execing process mid-exec in %d tries", tries); 1000 T_ASSERT_NE(cid1, cid2, "container IDs for in-progress exec are unique"); 1001 T_PASS("found mid-exec process in %d tries", tries); 1002} 1003 1004#ifdef _LP64 1005#if __has_feature(ptrauth_calls) 1006#define __ptrauth_swift_async_context_parent \ 1007 __ptrauth(ptrauth_key_process_independent_data, 1, 0xbda2) 1008#define __ptrauth_swift_async_context_resume \ 1009 __ptrauth(ptrauth_key_function_pointer, 1, 0xd707) 1010#else 1011#define __ptrauth_swift_async_context_parent 1012#define __ptrauth_swift_async_context_resume 1013#endif 1014// Add 1 to match the symbolication aid added by the stackshot backtracer. 1015#define asyncstack_frame(x) ((uintptr_t)(void *)ptrauth_strip((void *)(x), ptrauth_key_function_pointer) + 1) 1016 1017// This struct fakes the Swift AsyncContext struct which is used by 1018// the Swift concurrency runtime. We only care about the first 2 fields. 1019struct fake_async_context { 1020 struct fake_async_context* __ptrauth_swift_async_context_parent next; 1021 void(*__ptrauth_swift_async_context_resume resume_pc)(void); 1022}; 1023 1024static void 1025level1_func() 1026{ 1027} 1028static void 1029level2_func() 1030{ 1031} 1032 1033// Create a chain of fake async contexts; sync with asyncstack_expected_stack below 1034static alignas(16) struct fake_async_context level1 = { 0, level1_func }; 1035static alignas(16) struct fake_async_context level2 = { &level1, level2_func }; 1036 1037struct async_test_semaphores { 1038 dispatch_semaphore_t child_ready_sem; /* signal parent we're ready */ 1039 dispatch_semaphore_t child_exit_sem; /* parent tells us to go away */ 1040}; 1041 1042#define ASYNCSTACK_THREAD_NAME "asyncstack_thread" 1043 1044static void __attribute__((noinline, not_tail_called)) 1045expect_asyncstack(void *arg) 1046{ 1047 struct async_test_semaphores *async_ts = arg; 1048 1049 T_QUIET; T_ASSERT_POSIX_ZERO(pthread_setname_np(ASYNCSTACK_THREAD_NAME), 1050 "set thread name to %s", ASYNCSTACK_THREAD_NAME); 1051 1052 /* Tell the main thread we're all set up, then wait for permission to exit */ 1053 dispatch_semaphore_signal(async_ts->child_ready_sem); 1054 dispatch_semaphore_wait(async_ts->child_exit_sem, DISPATCH_TIME_FOREVER); 1055 usleep(1); /* make sure we don't tailcall semaphore_wait */ 1056} 1057 1058static void * 1059asyncstack_thread(void *arg) 1060{ 1061 uint64_t *fp = __builtin_frame_address(0); 1062 // We cannot use a variable of pointer type, because this ABI is valid 1063 // on arm64_32 where pointers are 32bits, but the context pointer will 1064 // still be stored in a 64bits slot on the stack. 1065#if __has_feature(ptrauth_calls) 1066#define __stack_context_auth __ptrauth(ptrauth_key_process_dependent_data, 1, \ 1067 0xc31a) 1068 struct fake_async_context * __stack_context_auth ctx = &level2; 1069#else // __has_feature(ptrauth_calls) 1070 /* struct fake_async_context * */uint64_t ctx = (uintptr_t)&level2; 1071#endif // !__has_feature(ptrauth_calls) 1072 1073 // The signature of an async frame on the OS stack is: 1074 // [ <AsyncContext address>, <Saved FP | (1<<60)>, <return address> ] 1075 // The Async context must be right before the saved FP on the stack. This 1076 // should happen naturally in an optimized build as it is the only 1077 // variable on the stack. 1078 // This function cannot use T_ASSERT_* becuse it changes the stack 1079 // layout. 1080 assert((uintptr_t)fp - (uintptr_t)&ctx == 8); 1081 1082 // Modify the saved FP on the stack to include the async frame marker 1083 *fp |= (0x1ULL << 60); 1084 expect_asyncstack(arg); 1085 return NULL; 1086} 1087 1088T_DECL(asyncstack, "test swift async stack entries", T_META_TAG_VM_PREFERRED) 1089{ 1090 struct scenario scenario = { 1091 .name = "asyncstack", 1092 .flags = STACKSHOT_KCDATA_FORMAT | STACKSHOT_SAVE_LOADINFO, 1093 }; 1094 struct async_test_semaphores async_ts = { 1095 .child_ready_sem = dispatch_semaphore_create(0), 1096 .child_exit_sem = dispatch_semaphore_create(0), 1097 }; 1098 T_QUIET; T_ASSERT_NOTNULL(async_ts.child_ready_sem, "child_ready_sem alloc"); 1099 T_QUIET; T_ASSERT_NOTNULL(async_ts.child_exit_sem, "child_exit_sem alloc"); 1100 1101 pthread_t pthread; 1102 __block uint64_t threadid = 0; 1103 T_QUIET; T_ASSERT_POSIX_ZERO(pthread_create(&pthread, NULL, asyncstack_thread, &async_ts), "pthread_create"); 1104 T_QUIET; T_ASSERT_POSIX_ZERO(pthread_threadid_np(pthread, &threadid), "pthread_threadid_np"); 1105 1106 dispatch_semaphore_wait(async_ts.child_ready_sem, DISPATCH_TIME_FOREVER); 1107 1108 take_stackshot(&scenario, true, ^( void *ssbuf, size_t sslen) { 1109 parse_stackshot(PARSE_STACKSHOT_ASYNCSTACK, ssbuf, sslen, @{ 1110 asyncstack_expected_threadid_key: @(threadid), 1111 asyncstack_expected_stack_key: @[ @(asyncstack_frame(level2_func)), @(asyncstack_frame(level1_func)) ], 1112 }); 1113 }); 1114 1115 dispatch_semaphore_signal(async_ts.child_exit_sem); 1116 T_QUIET; T_ASSERT_POSIX_ZERO(pthread_join(pthread, NULL), "wait for thread"); 1117 1118} 1119#endif /* #ifdef _LP64 */ 1120 1121static uint32_t 1122get_user_promotion_basepri(void) 1123{ 1124 mach_msg_type_number_t count = THREAD_POLICY_STATE_COUNT; 1125 struct thread_policy_state thread_policy; 1126 boolean_t get_default = FALSE; 1127 mach_port_t thread_port = pthread_mach_thread_np(pthread_self()); 1128 1129 kern_return_t kr = thread_policy_get(thread_port, THREAD_POLICY_STATE, 1130 (thread_policy_t)&thread_policy, &count, &get_default); 1131 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "thread_policy_get"); 1132 return thread_policy.thps_user_promotion_basepri; 1133} 1134 1135static int 1136get_pri(thread_t thread_port) 1137{ 1138 kern_return_t kr; 1139 1140 thread_extended_info_data_t extended_info; 1141 mach_msg_type_number_t count = THREAD_EXTENDED_INFO_COUNT; 1142 kr = thread_info(thread_port, THREAD_EXTENDED_INFO, 1143 (thread_info_t)&extended_info, &count); 1144 1145 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "thread_info"); 1146 1147 return extended_info.pth_curpri; 1148} 1149 1150 1151T_DECL(turnstile_singlehop, "turnstile single hop test", T_META_TAG_VM_PREFERRED) 1152{ 1153 dispatch_queue_t dq1, dq2; 1154 dispatch_semaphore_t sema_x; 1155 dispatch_queue_attr_t dq1_attr, dq2_attr; 1156 __block qos_class_t main_qos = 0; 1157 __block int main_relpri = 0, main_relpri2 = 0, main_afterpri = 0; 1158 struct scenario scenario = { 1159 .name = "turnstile_singlehop", 1160 .flags = (STACKSHOT_THREAD_WAITINFO | STACKSHOT_KCDATA_FORMAT), 1161 }; 1162 dq1_attr = dispatch_queue_attr_make_with_qos_class(DISPATCH_QUEUE_SERIAL, QOS_CLASS_UTILITY, 0); 1163 dq2_attr = dispatch_queue_attr_make_with_qos_class(DISPATCH_QUEUE_SERIAL, QOS_CLASS_USER_INITIATED, 0); 1164 pthread_mutex_t lock_a = PTHREAD_MUTEX_INITIALIZER; 1165 pthread_mutex_t lock_b = PTHREAD_MUTEX_INITIALIZER; 1166 1167 pthread_mutex_t *lockap = &lock_a, *lockbp = &lock_b; 1168 1169 dq1 = dispatch_queue_create("q1", dq1_attr); 1170 dq2 = dispatch_queue_create("q2", dq2_attr); 1171 sema_x = dispatch_semaphore_create(0); 1172 1173 pthread_mutex_lock(lockap); 1174 dispatch_async(dq1, ^{ 1175 pthread_mutex_lock(lockbp); 1176 T_ASSERT_POSIX_SUCCESS(pthread_get_qos_class_np(pthread_self(), &main_qos, &main_relpri), "get qos class"); 1177 T_LOG("The priority of q1 is %d\n", get_pri(mach_thread_self())); 1178 dispatch_semaphore_signal(sema_x); 1179 pthread_mutex_lock(lockap); 1180 }); 1181 dispatch_semaphore_wait(sema_x, DISPATCH_TIME_FOREVER); 1182 1183 T_LOG("Async1 completed"); 1184 1185 pthread_set_qos_class_self_np(QOS_CLASS_UTILITY, 0); 1186 T_ASSERT_POSIX_SUCCESS(pthread_get_qos_class_np(pthread_self(), &main_qos, &main_relpri), "get qos class"); 1187 T_LOG("The priority of main is %d\n", get_pri(mach_thread_self())); 1188 main_relpri = get_pri(mach_thread_self()); 1189 1190 dispatch_async(dq2, ^{ 1191 T_ASSERT_POSIX_SUCCESS(pthread_get_qos_class_np(pthread_self(), &main_qos, &main_relpri2), "get qos class"); 1192 T_LOG("The priority of q2 is %d\n", get_pri(mach_thread_self())); 1193 dispatch_semaphore_signal(sema_x); 1194 pthread_mutex_lock(lockbp); 1195 }); 1196 dispatch_semaphore_wait(sema_x, DISPATCH_TIME_FOREVER); 1197 1198 T_LOG("Async2 completed"); 1199 1200 while (1) { 1201 main_afterpri = (int) get_user_promotion_basepri(); 1202 if (main_relpri != main_afterpri) { 1203 T_LOG("Success with promotion pri is %d", main_afterpri); 1204 break; 1205 } 1206 1207 usleep(100); 1208 } 1209 1210 take_stackshot(&scenario, true, ^( void *ssbuf, size_t sslen) { 1211 parse_stackshot(PARSE_STACKSHOT_TURNSTILEINFO, ssbuf, sslen, nil); 1212 }); 1213} 1214 1215 1216static void 1217expect_instrs_cycles_in_stackshot(void *ssbuf, size_t sslen) 1218{ 1219 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 1220 1221 bool in_task = false; 1222 bool in_thread = false; 1223 bool saw_instrs_cycles = false; 1224 iter = kcdata_iter_next(iter); 1225 1226 KCDATA_ITER_FOREACH(iter) { 1227 switch (kcdata_iter_type(iter)) { 1228 case KCDATA_TYPE_CONTAINER_BEGIN: 1229 switch (kcdata_iter_container_type(iter)) { 1230 case STACKSHOT_KCCONTAINER_TASK: 1231 in_task = true; 1232 saw_instrs_cycles = false; 1233 break; 1234 1235 case STACKSHOT_KCCONTAINER_THREAD: 1236 in_thread = true; 1237 saw_instrs_cycles = false; 1238 break; 1239 1240 default: 1241 break; 1242 } 1243 break; 1244 1245 case STACKSHOT_KCTYPE_INSTRS_CYCLES: 1246 saw_instrs_cycles = true; 1247 break; 1248 1249 case KCDATA_TYPE_CONTAINER_END: 1250 if (in_thread) { 1251 T_QUIET; T_EXPECT_TRUE(saw_instrs_cycles, 1252 "saw instructions and cycles in thread"); 1253 in_thread = false; 1254 } else if (in_task) { 1255 T_QUIET; T_EXPECT_TRUE(saw_instrs_cycles, 1256 "saw instructions and cycles in task"); 1257 in_task = false; 1258 } 1259 1260 default: 1261 break; 1262 } 1263 } 1264} 1265 1266static void 1267skip_if_monotonic_unsupported(void) 1268{ 1269 int supported = 0; 1270 size_t supported_size = sizeof(supported); 1271 int ret = sysctlbyname("kern.monotonic.supported", &supported, 1272 &supported_size, 0, 0); 1273 if (ret < 0 || !supported) { 1274 T_SKIP("monotonic is unsupported"); 1275 } 1276} 1277 1278T_DECL(instrs_cycles, "test a getting instructions and cycles in stackshot", T_META_TAG_VM_PREFERRED) 1279{ 1280 skip_if_monotonic_unsupported(); 1281 1282 struct scenario scenario = { 1283 .name = "instrs-cycles", 1284 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_INSTRS_CYCLES 1285 | STACKSHOT_KCDATA_FORMAT), 1286 }; 1287 1288 T_LOG("attempting to take stackshot with instructions and cycles"); 1289 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1290 parse_stackshot(0, ssbuf, sslen, nil); 1291 expect_instrs_cycles_in_stackshot(ssbuf, sslen); 1292 }); 1293} 1294 1295T_DECL(delta_instrs_cycles, 1296 "test delta stackshots with instructions and cycles", T_META_TAG_VM_PREFERRED) 1297{ 1298 skip_if_monotonic_unsupported(); 1299 1300 struct scenario scenario = { 1301 .name = "delta-instrs-cycles", 1302 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_INSTRS_CYCLES 1303 | STACKSHOT_KCDATA_FORMAT), 1304 }; 1305 1306 T_LOG("taking full stackshot"); 1307 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1308 uint64_t stackshot_time = stackshot_timestamp(ssbuf, sslen); 1309 1310 T_LOG("taking delta stackshot since time %" PRIu64, stackshot_time); 1311 1312 parse_stackshot(0, ssbuf, sslen, nil); 1313 expect_instrs_cycles_in_stackshot(ssbuf, sslen); 1314 1315 struct scenario delta_scenario = { 1316 .name = "delta-instrs-cycles-next", 1317 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_INSTRS_CYCLES 1318 | STACKSHOT_KCDATA_FORMAT 1319 | STACKSHOT_COLLECT_DELTA_SNAPSHOT), 1320 .since_timestamp = stackshot_time, 1321 }; 1322 1323 take_stackshot(&delta_scenario, false, ^(void *dssbuf, size_t dsslen) { 1324 parse_stackshot(PARSE_STACKSHOT_DELTA, dssbuf, dsslen, nil); 1325 expect_instrs_cycles_in_stackshot(dssbuf, dsslen); 1326 }); 1327 }); 1328} 1329 1330static void 1331check_thread_groups_supported() 1332{ 1333 int err; 1334 int supported = 0; 1335 size_t supported_size = sizeof(supported); 1336 err = sysctlbyname("kern.thread_groups_supported", &supported, &supported_size, NULL, 0); 1337 1338 if (err || !supported) 1339 T_SKIP("thread groups not supported on this system"); 1340} 1341 1342T_DECL(thread_groups, "test getting thread groups in stackshot", T_META_TAG_VM_PREFERRED) 1343{ 1344 check_thread_groups_supported(); 1345 1346 struct scenario scenario = { 1347 .name = "thread-groups", 1348 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_THREAD_GROUP 1349 | STACKSHOT_KCDATA_FORMAT), 1350 }; 1351 1352 T_LOG("attempting to take stackshot with thread group flag"); 1353 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1354 parse_thread_group_stackshot(ssbuf, sslen); 1355 }); 1356} 1357 1358T_DECL(compactinfo, "test compactinfo inclusion", T_META_TAG_VM_PREFERRED) 1359{ 1360 struct scenario scenario = { 1361 .name = "compactinfo", 1362 .target_pid = getpid(), 1363 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_SAVE_DYLD_COMPACTINFO 1364 | STACKSHOT_KCDATA_FORMAT), 1365 }; 1366 1367 T_LOG("attempting to take stackshot with compactinfo flag"); 1368 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1369 parse_stackshot(PARSE_STACKSHOT_COMPACTINFO | PARSE_STACKSHOT_TARGETPID, ssbuf, sslen, nil); 1370 }); 1371} 1372 1373T_DECL(suspendinfo, "test task suspend info inclusion", T_META_TAG_VM_PREFERRED) 1374{ 1375 struct scenario scenario = { 1376 .name = "suspendinfo", 1377 .target_pid = getpid(), 1378 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT), 1379 }; 1380 1381 T_LOG("attempting to take stackshot with suspendinfo flag"); 1382 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1383 parse_stackshot(PARSE_STACKSHOT_SUSPENDINFO | PARSE_STACKSHOT_TARGETPID, ssbuf, sslen, nil); 1384 }); 1385} 1386 1387static NSMutableSet * find_driverkit_pids(io_registry_entry_t root) { 1388 NSMutableSet * driverkit_pids = [NSMutableSet setWithCapacity:3]; 1389 io_registry_entry_t current = IO_OBJECT_NULL; 1390 io_iterator_t iter = IO_OBJECT_NULL; 1391 1392 T_EXPECT_MACH_SUCCESS(IORegistryEntryGetChildIterator(root, kIOServicePlane, &iter), "get registry iterator"); 1393 1394 while ((current = IOIteratorNext(iter)) != IO_OBJECT_NULL) { 1395 if (_IOObjectConformsTo(current, "IOUserServer", kIOClassNameOverrideNone)) { 1396 CFMutableDictionaryRef cfProperties = NULL; 1397 NSMutableDictionary * properties; 1398 NSString * client_creator_info; 1399 NSArray<NSString *> *creator_info_array; 1400 pid_t pid; 1401 1402 T_QUIET; T_EXPECT_MACH_SUCCESS(IORegistryEntryCreateCFProperties(current, &cfProperties, kCFAllocatorDefault, kNilOptions), "get properties"); 1403 properties = CFBridgingRelease(cfProperties); 1404 T_QUIET; T_ASSERT_NOTNULL(properties, "properties is not null"); 1405 client_creator_info = properties[@kIOUserClientCreatorKey]; 1406 creator_info_array = [client_creator_info componentsSeparatedByString:@","]; 1407 if ([creator_info_array[0] hasPrefix:@"pid"]) { 1408 NSArray<NSString *> *pid_info = [creator_info_array[0] componentsSeparatedByString:@" "]; 1409 T_QUIET; T_ASSERT_EQ(pid_info.count, 2UL, "Get pid info components from %s", creator_info_array[0].UTF8String); 1410 pid = pid_info[1].intValue; 1411 } else { 1412 T_ASSERT_FAIL("No pid info in client creator info: %s", client_creator_info.UTF8String); 1413 } 1414 T_LOG("Found driver pid %d", pid); 1415 [driverkit_pids addObject:[NSNumber numberWithInt:pid]]; 1416 } else { 1417 [driverkit_pids unionSet:find_driverkit_pids(current)]; 1418 } 1419 IOObjectRelease(current); 1420 } 1421 1422 IOObjectRelease(iter); 1423 return driverkit_pids; 1424} 1425 1426T_DECL(driverkit, "test driverkit inclusion", T_META_TAG_VM_PREFERRED) 1427{ 1428 struct scenario scenario = { 1429 .name = "driverkit", 1430 .target_kernel = true, 1431 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT 1432 | STACKSHOT_INCLUDE_DRIVER_THREADS_IN_KERNEL), 1433 }; 1434 1435 io_registry_entry_t root = IORegistryGetRootEntry(kIOMainPortDefault); 1436 NSMutableSet * driverkit_pids = find_driverkit_pids(root); 1437 IOObjectRelease(root); 1438 1439 T_LOG("expecting to find %lu driverkit processes", [driverkit_pids count]); 1440 T_LOG("attempting to take stackshot with STACKSHOT_INCLUDE_DRIVER_THREADS_IN_KERNEL flag"); 1441 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1442 parse_stackshot(PARSE_STACKSHOT_DRIVERKIT | PARSE_STACKSHOT_TARGETPID, ssbuf, sslen, @{ 1443 driverkit_found_key: ^(pid_t pid) { 1444 [driverkit_pids removeObject:[NSNumber numberWithInt:pid]]; 1445 }}); 1446 }); 1447 1448 T_EXPECT_EQ([driverkit_pids count], (NSUInteger)0, "found expected number of driverkit processes"); 1449} 1450 1451static void 1452parse_page_table_asid_stackshot(void **ssbuf, size_t sslen) 1453{ 1454 bool seen_asid = false; 1455 bool seen_page_table_snapshot = false; 1456 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 1457 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, 1458 "buffer provided is a stackshot"); 1459 1460 iter = kcdata_iter_next(iter); 1461 KCDATA_ITER_FOREACH(iter) { 1462 switch (kcdata_iter_type(iter)) { 1463 case KCDATA_TYPE_ARRAY: { 1464 T_QUIET; 1465 T_ASSERT_TRUE(kcdata_iter_array_valid(iter), 1466 "checked that array is valid"); 1467 1468 if (kcdata_iter_array_elem_type(iter) != STACKSHOT_KCTYPE_PAGE_TABLES) { 1469 continue; 1470 } 1471 1472 T_ASSERT_FALSE(seen_page_table_snapshot, "check that we haven't yet seen a page table snapshot"); 1473 seen_page_table_snapshot = true; 1474 1475 T_ASSERT_EQ((size_t) kcdata_iter_array_elem_size(iter), sizeof(uint64_t), 1476 "check that each element of the pagetable dump is the expected size"); 1477 1478 uint64_t *pt_array = kcdata_iter_payload(iter); 1479 uint32_t elem_count = kcdata_iter_array_elem_count(iter); 1480 uint32_t j; 1481 bool nonzero_tte = false; 1482 for (j = 0; j < elem_count;) { 1483 T_QUIET; T_ASSERT_LE(j + 4, elem_count, "check for valid page table segment header"); 1484 uint64_t pa = pt_array[j]; 1485 uint64_t num_entries = pt_array[j + 1]; 1486 uint64_t start_va = pt_array[j + 2]; 1487 uint64_t end_va = pt_array[j + 3]; 1488 1489 T_QUIET; T_ASSERT_NE(pa, (uint64_t) 0, "check that the pagetable physical address is non-zero"); 1490 T_QUIET; T_ASSERT_EQ(pa % (num_entries * sizeof(uint64_t)), (uint64_t) 0, "check that the pagetable physical address is correctly aligned"); 1491 T_QUIET; T_ASSERT_NE(num_entries, (uint64_t) 0, "check that a pagetable region has more than 0 entries"); 1492 T_QUIET; T_ASSERT_LE(j + 4 + num_entries, (uint64_t) elem_count, "check for sufficient space in page table array"); 1493 T_QUIET; T_ASSERT_GT(end_va, start_va, "check for valid VA bounds in page table segment header"); 1494 1495 for (uint32_t k = j + 4; k < (j + 4 + num_entries); ++k) { 1496 if (pt_array[k] != 0) { 1497 nonzero_tte = true; 1498 T_QUIET; T_ASSERT_EQ((pt_array[k] >> 48) & 0xf, (uint64_t) 0, "check that bits[48:51] of arm64 TTE are clear"); 1499 // L0-L2 table and non-compressed L3 block entries should always have bit 1 set; assumes L0-L2 blocks will not be used outside the kernel 1500 bool table = ((pt_array[k] & 0x2) != 0); 1501 if (table) { 1502 T_QUIET; T_ASSERT_NE(pt_array[k] & ((1ULL << 48) - 1) & ~((1ULL << 12) - 1), (uint64_t) 0, "check that arm64 TTE physical address is non-zero"); 1503 } else { // should be a compressed PTE 1504 T_QUIET; T_ASSERT_NE(pt_array[k] & 0xC000000000000000ULL, (uint64_t) 0, "check that compressed PTE has at least one of bits [63:62] set"); 1505 T_QUIET; T_ASSERT_EQ(pt_array[k] & ~0xC000000000000000ULL, (uint64_t) 0, "check that compressed PTE has no other bits besides [63:62] set"); 1506 } 1507 } 1508 } 1509 1510 j += (4 + num_entries); 1511 } 1512 T_ASSERT_TRUE(nonzero_tte, "check that we saw at least one non-empty TTE"); 1513 T_ASSERT_EQ(j, elem_count, "check that page table dump size matches extent of last header"); 1514 break; 1515 } 1516 case STACKSHOT_KCTYPE_ASID: { 1517 T_ASSERT_FALSE(seen_asid, "check that we haven't yet seen an ASID"); 1518 seen_asid = true; 1519 } 1520 } 1521 } 1522 T_ASSERT_TRUE(seen_page_table_snapshot, "check that we have seen a page table snapshot"); 1523 T_ASSERT_TRUE(seen_asid, "check that we have seen an ASID"); 1524} 1525 1526T_DECL(dump_page_tables, "test stackshot page table dumping support", T_META_TAG_VM_PREFERRED) 1527{ 1528 struct scenario scenario = { 1529 .name = "asid-page-tables", 1530 .flags = (STACKSHOT_KCDATA_FORMAT | STACKSHOT_ASID | STACKSHOT_PAGE_TABLES), 1531 .size_hint = (9ull << 20), // 9 MB 1532 .target_pid = getpid(), 1533 .maybe_unsupported = true, 1534 .maybe_enomem = true, 1535 }; 1536 1537 T_LOG("attempting to take stackshot with ASID and page table flags"); 1538 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1539 parse_page_table_asid_stackshot(ssbuf, sslen); 1540 }); 1541} 1542 1543static void stackshot_verify_current_proc_uuid_info(void **ssbuf, size_t sslen, uint64_t expected_offset, const struct proc_uniqidentifierinfo *proc_info_data) 1544{ 1545 const uuid_t *current_uuid = (const uuid_t *)(&proc_info_data->p_uuid); 1546 1547 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 1548 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, "buffer provided is a stackshot"); 1549 1550 iter = kcdata_iter_next(iter); 1551 1552 KCDATA_ITER_FOREACH(iter) { 1553 switch (kcdata_iter_type(iter)) { 1554 case KCDATA_TYPE_ARRAY: { 1555 T_QUIET; T_ASSERT_TRUE(kcdata_iter_array_valid(iter), "checked that array is valid"); 1556 if (kcdata_iter_array_elem_type(iter) == KCDATA_TYPE_LIBRARY_LOADINFO64) { 1557 struct user64_dyld_uuid_info *info = (struct user64_dyld_uuid_info *) kcdata_iter_payload(iter); 1558 if (uuid_compare(*current_uuid, info->imageUUID) == 0) { 1559 T_ASSERT_EQ(expected_offset, info->imageLoadAddress, "found matching UUID with matching binary offset"); 1560 return; 1561 } 1562 } else if (kcdata_iter_array_elem_type(iter) == KCDATA_TYPE_LIBRARY_LOADINFO) { 1563 struct user32_dyld_uuid_info *info = (struct user32_dyld_uuid_info *) kcdata_iter_payload(iter); 1564 if (uuid_compare(*current_uuid, info->imageUUID) == 0) { 1565 T_ASSERT_EQ(expected_offset, ((uint64_t) info->imageLoadAddress), "found matching UUID with matching binary offset"); 1566 return; 1567 } 1568 } 1569 break; 1570 } 1571 default: 1572 break; 1573 } 1574 } 1575 1576 T_FAIL("failed to find matching UUID in stackshot data"); 1577} 1578 1579T_DECL(translated, "tests translated bit is set correctly", T_META_TAG_VM_PREFERRED) 1580{ 1581#if !(TARGET_OS_OSX && TARGET_CPU_ARM64) 1582 T_SKIP("Only valid on Apple silicon Macs") 1583#endif 1584 // Get path of stackshot_translated_child helper binary 1585 char path[PATH_MAX]; 1586 uint32_t path_size = sizeof(path); 1587 T_QUIET; T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 1588 char* binary_name = strrchr(path, '/'); 1589 if (binary_name) binary_name++; 1590 T_QUIET; T_ASSERT_NOTNULL(binary_name, "Find basename in path '%s'", path); 1591 strlcpy(binary_name, "stackshot_translated_child", path_size - (binary_name - path)); 1592 char *args[] = { path, NULL }; 1593 1594 dispatch_source_t child_sig_src; 1595 dispatch_semaphore_t child_ready_sem = dispatch_semaphore_create(0); 1596 T_QUIET; T_ASSERT_NOTNULL(child_ready_sem, "exec child semaphore"); 1597 1598 dispatch_queue_t signal_processing_q = dispatch_queue_create("signal processing queue", NULL); 1599 T_QUIET; T_ASSERT_NOTNULL(signal_processing_q, "signal processing queue"); 1600 1601 signal(SIGUSR1, SIG_IGN); 1602 child_sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, signal_processing_q); 1603 T_QUIET; T_ASSERT_NOTNULL(child_sig_src, "dispatch_source_create (child_sig_src)"); 1604 1605 dispatch_source_set_event_handler(child_sig_src, ^{ dispatch_semaphore_signal(child_ready_sem); }); 1606 dispatch_activate(child_sig_src); 1607 1608 // Spawn child 1609 pid_t pid; 1610 T_LOG("spawning translated child"); 1611 T_QUIET; T_ASSERT_POSIX_ZERO(posix_spawn(&pid, args[0], NULL, NULL, args, NULL), "spawned process '%s' with PID %d", args[0], pid); 1612 1613 // Wait for the the child to spawn up 1614 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 1615 1616 // Make sure the child is running and is translated 1617 int mib[] = { CTL_KERN, KERN_PROC, KERN_PROC_PID, pid }; 1618 struct kinfo_proc process_info; 1619 size_t bufsize = sizeof(process_info); 1620 T_QUIET; T_ASSERT_POSIX_SUCCESS(sysctl(mib, (unsigned)(sizeof(mib)/sizeof(int)), &process_info, &bufsize, NULL, 0), "get translated child process info"); 1621 T_QUIET; T_ASSERT_GT(bufsize, (size_t)0, "process info is not empty"); 1622 T_QUIET; T_ASSERT_TRUE((process_info.kp_proc.p_flag & P_TRANSLATED), "KERN_PROC_PID reports child is translated"); 1623 1624 T_LOG("capturing stackshot"); 1625 1626 struct scenario scenario = { 1627 .name = "translated", 1628 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 1629 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 1630 }; 1631 1632 take_stackshot(&scenario, true, ^( void *ssbuf, size_t sslen) { 1633 parse_stackshot(PARSE_STACKSHOT_TRANSLATED, ssbuf, sslen, @{translated_child_pid_key: @(pid)}); 1634 }); 1635 1636 // Kill the child 1637 int status; 1638 T_QUIET; T_ASSERT_POSIX_SUCCESS(kill(pid, SIGTERM), "kill translated child"); 1639 T_QUIET; T_ASSERT_POSIX_SUCCESS(waitpid(pid, &status, 0), "waitpid on translated child"); 1640 1641} 1642 1643T_DECL(proc_uuid_info, "tests that the main binary UUID for a proc is always populated", T_META_TAG_VM_PREFERRED) 1644{ 1645 struct proc_uniqidentifierinfo proc_info_data = { }; 1646 mach_msg_type_number_t count; 1647 kern_return_t kernel_status; 1648 task_dyld_info_data_t task_dyld_info; 1649 struct dyld_all_image_infos *target_infos; 1650 int retval; 1651 bool found_image_in_image_infos = false; 1652 uint64_t expected_mach_header_offset = 0; 1653 1654 /* Find the UUID of our main binary */ 1655 retval = proc_pidinfo(getpid(), PROC_PIDUNIQIDENTIFIERINFO, 0, &proc_info_data, sizeof(proc_info_data)); 1656 T_QUIET; T_EXPECT_POSIX_SUCCESS(retval, "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO"); 1657 T_QUIET; T_ASSERT_EQ_INT(retval, (int) sizeof(proc_info_data), "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO returned data"); 1658 1659 uuid_string_t str = {}; 1660 uuid_unparse(*(uuid_t*)&proc_info_data.p_uuid, str); 1661 T_LOG("Found current UUID is %s", str); 1662 1663 /* Find the location of the dyld image info metadata */ 1664 count = TASK_DYLD_INFO_COUNT; 1665 kernel_status = task_info(mach_task_self(), TASK_DYLD_INFO, (task_info_t)&task_dyld_info, &count); 1666 T_QUIET; T_ASSERT_EQ(kernel_status, KERN_SUCCESS, "retrieve task_info for TASK_DYLD_INFO"); 1667 1668 target_infos = (struct dyld_all_image_infos *)task_dyld_info.all_image_info_addr; 1669 1670 /* Find our binary in the dyld image info array */ 1671 for (int i = 0; i < (int) target_infos->uuidArrayCount; i++) { 1672 if (uuid_compare(target_infos->uuidArray[i].imageUUID, *(uuid_t*)&proc_info_data.p_uuid) == 0) { 1673 expected_mach_header_offset = (uint64_t) target_infos->uuidArray[i].imageLoadAddress; 1674 found_image_in_image_infos = true; 1675 } 1676 } 1677 1678 T_ASSERT_TRUE(found_image_in_image_infos, "found binary image in dyld image info list"); 1679 1680 /* Overwrite the dyld image info data so the kernel has to fallback to the UUID stored in the proc structure */ 1681 target_infos->uuidArrayCount = 0; 1682 1683 struct scenario scenario = { 1684 .name = "proc_uuid_info", 1685 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT), 1686 .target_pid = getpid(), 1687 }; 1688 1689 T_LOG("attempting to take stackshot for current PID"); 1690 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1691 stackshot_verify_current_proc_uuid_info(ssbuf, sslen, expected_mach_header_offset, &proc_info_data); 1692 }); 1693} 1694 1695T_DECL(cseg_waitinfo, "test that threads stuck in the compressor report correct waitinfo", T_META_TAG_VM_PREFERRED) 1696{ 1697 struct scenario scenario = { 1698 .name = "cseg_waitinfo", 1699 .quiet = false, 1700 .flags = (STACKSHOT_THREAD_WAITINFO | STACKSHOT_KCDATA_FORMAT), 1701 }; 1702 __block uint64_t thread_id = 0; 1703 1704 dispatch_queue_t dq = dispatch_queue_create("com.apple.stackshot.cseg_waitinfo", NULL); 1705 dispatch_semaphore_t child_ok = dispatch_semaphore_create(0); 1706 1707 dispatch_async(dq, ^{ 1708 pthread_threadid_np(NULL, &thread_id); 1709 dispatch_semaphore_signal(child_ok); 1710 int val = 1; 1711 T_ASSERT_POSIX_SUCCESS(sysctlbyname("kern.cseg_wedge_thread", NULL, NULL, &val, sizeof(val)), "wedge child thread"); 1712 }); 1713 1714 dispatch_semaphore_wait(child_ok, DISPATCH_TIME_FOREVER); 1715 sleep(1); 1716 1717 T_LOG("taking stackshot"); 1718 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1719 int val = 1; 1720 T_ASSERT_POSIX_SUCCESS(sysctlbyname("kern.cseg_unwedge_thread", NULL, NULL, &val, sizeof(val)), "unwedge child thread"); 1721 parse_stackshot(PARSE_STACKSHOT_WAITINFO_CSEG, ssbuf, sslen, @{cseg_expected_threadid_key: @(thread_id)}); 1722 }); 1723} 1724 1725static void 1726srp_send( 1727 mach_port_t send_port, 1728 mach_port_t reply_port, 1729 mach_port_t msg_port) 1730{ 1731 kern_return_t ret = 0; 1732 1733 struct test_msg { 1734 mach_msg_header_t header; 1735 mach_msg_body_t body; 1736 mach_msg_port_descriptor_t port_descriptor; 1737 }; 1738 struct test_msg send_msg = { 1739 .header = { 1740 .msgh_remote_port = send_port, 1741 .msgh_local_port = reply_port, 1742 .msgh_bits = MACH_MSGH_BITS_SET(MACH_MSG_TYPE_COPY_SEND, 1743 reply_port ? MACH_MSG_TYPE_MAKE_SEND_ONCE : 0, 1744 MACH_MSG_TYPE_MOVE_SEND, 1745 MACH_MSGH_BITS_COMPLEX), 1746 .msgh_id = 0x100, 1747 .msgh_size = sizeof(send_msg), 1748 }, 1749 .body = { 1750 .msgh_descriptor_count = 1, 1751 }, 1752 .port_descriptor = { 1753 .name = msg_port, 1754 .disposition = MACH_MSG_TYPE_MOVE_RECEIVE, 1755 .type = MACH_MSG_PORT_DESCRIPTOR, 1756 }, 1757 }; 1758 1759 if (msg_port == MACH_PORT_NULL) { 1760 send_msg.body.msgh_descriptor_count = 0; 1761 } 1762 1763 ret = mach_msg(&(send_msg.header), 1764 MACH_SEND_MSG | 1765 MACH_SEND_TIMEOUT | 1766 MACH_SEND_OVERRIDE, 1767 send_msg.header.msgh_size, 1768 0, 1769 MACH_PORT_NULL, 1770 10000, 1771 0); 1772 1773 T_ASSERT_MACH_SUCCESS(ret, "client mach_msg"); 1774} 1775 1776T_HELPER_DECL(srp_client, 1777 "Client used for the special_reply_port test") 1778{ 1779 pid_t ppid = getppid(); 1780 dispatch_semaphore_t can_continue = dispatch_semaphore_create(0); 1781 dispatch_queue_t dq = dispatch_queue_create("client_signalqueue", NULL); 1782 dispatch_source_t sig_src; 1783 1784 mach_msg_return_t mr; 1785 mach_port_t service_port; 1786 mach_port_t conn_port; 1787 mach_port_t special_reply_port; 1788 mach_port_options_t opts = { 1789 .flags = MPO_INSERT_SEND_RIGHT, 1790 }; 1791 1792 signal(SIGUSR1, SIG_IGN); 1793 sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, dq); 1794 1795 dispatch_source_set_event_handler(sig_src, ^{ 1796 dispatch_semaphore_signal(can_continue); 1797 }); 1798 dispatch_activate(sig_src); 1799 1800 /* lookup the mach service port for the parent */ 1801 kern_return_t kr = bootstrap_look_up(bootstrap_port, 1802 SRP_SERVICE_NAME, &service_port); 1803 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up"); 1804 1805 /* create the send-once right (special reply port) and message to send to the server */ 1806 kr = mach_port_construct(mach_task_self(), &opts, 0ull, &conn_port); 1807 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct"); 1808 1809 special_reply_port = thread_get_special_reply_port(); 1810 T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port"); 1811 1812 /* send the message with the special reply port */ 1813 srp_send(service_port, special_reply_port, conn_port); 1814 1815 /* signal the parent to continue */ 1816 kill(ppid, SIGUSR1); 1817 1818 struct { 1819 mach_msg_header_t header; 1820 mach_msg_body_t body; 1821 mach_msg_port_descriptor_t port_descriptor; 1822 } rcv_msg = { 1823 .header = 1824 { 1825 .msgh_remote_port = MACH_PORT_NULL, 1826 .msgh_local_port = special_reply_port, 1827 .msgh_size = sizeof(rcv_msg), 1828 }, 1829 }; 1830 1831 /* wait on the reply from the parent (that we will never receive) */ 1832 mr = mach_msg(&(rcv_msg.header), 1833 (MACH_RCV_MSG | MACH_RCV_SYNC_WAIT), 1834 0, 1835 rcv_msg.header.msgh_size, 1836 special_reply_port, 1837 MACH_MSG_TIMEOUT_NONE, 1838 service_port); 1839 1840 /* not expected to execute as parent will SIGKILL client... */ 1841 T_LOG("client process exiting after sending message to parent (server)"); 1842} 1843 1844enum srp_test_type { 1845 SRP_TEST_THREAD, /* expect waiter on current thread */ 1846 SRP_TEST_PID, /* expect waiter on current PID */ 1847 SRP_TEST_EITHER, /* waiter could be on either */ 1848}; 1849 1850static void 1851check_srp_test(const char *name, enum srp_test_type ty) 1852{ 1853 struct scenario scenario = { 1854 .name = name, 1855 .quiet = false, 1856 .flags = (STACKSHOT_THREAD_WAITINFO | STACKSHOT_KCDATA_FORMAT), 1857 }; 1858 uint64_t thread_id = 0; 1859 pthread_threadid_np(NULL, &thread_id); 1860 if (ty == SRP_TEST_THREAD) { 1861 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1862 parse_stackshot(PARSE_STACKSHOT_WAITINFO_SRP, ssbuf, sslen, 1863 @{srp_expected_threadid_key: @(thread_id)}); 1864 }); 1865 } else if (ty == SRP_TEST_PID) { 1866 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1867 parse_stackshot(PARSE_STACKSHOT_WAITINFO_SRP, ssbuf, sslen, 1868 @{srp_expected_pid_key: @(getpid())}); 1869 }); 1870 } else { 1871 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1872 parse_stackshot(PARSE_STACKSHOT_WAITINFO_SRP, ssbuf, sslen, 1873 @{srp_expected_pid_key: @(getpid()), srp_expected_threadid_key: @(thread_id)}); 1874 }); 1875 } 1876 1877} 1878 1879 1880/* 1881 * Tests the stackshot wait info plumbing for synchronous IPC that doesn't use kevent on the server. 1882 * 1883 * (part 1): tests the scenario where a client sends a request that includes a special reply port 1884 * to a server that doesn't receive the message and doesn't copy the send-once right 1885 * into its address space as a result. for this case the special reply port is enqueued 1886 * in a port and we check which task has that receive right and use that info. (rdar://60440338) 1887 * (part 2): tests the scenario where a client sends a request that includes a special reply port 1888 * to a server that receives the message and copies in the send-once right, but doesn't 1889 * reply to the client. for this case the special reply port is copied out and the kernel 1890 * stashes the info about which task copied out the send once right. (rdar://60440592) 1891 * (part 3): tests the same as part 2, but uses kevents, which allow for 1892 * priority inheritance 1893 */ 1894T_DECL(special_reply_port, "test that tasks using special reply ports have correct waitinfo", T_META_TAG_VM_PREFERRED) 1895{ 1896 dispatch_semaphore_t can_continue = dispatch_semaphore_create(0); 1897 dispatch_queue_t dq = dispatch_queue_create("signalqueue", NULL); 1898 dispatch_queue_t machdq = dispatch_queue_create("machqueue", NULL); 1899 dispatch_source_t sig_src; 1900 char path[PATH_MAX]; 1901 uint32_t path_size = sizeof(path); 1902 T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 1903 char *client_args[] = { path, "-n", "srp_client", NULL }; 1904 pid_t client_pid; 1905 int sp_ret; 1906 kern_return_t kr; 1907 mach_port_t port; 1908 1909 /* setup the signal handler in the parent (server) */ 1910 T_LOG("setup sig handlers"); 1911 signal(SIGUSR1, SIG_IGN); 1912 sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, dq); 1913 1914 dispatch_source_set_event_handler(sig_src, ^{ 1915 dispatch_semaphore_signal(can_continue); 1916 }); 1917 dispatch_activate(sig_src); 1918 1919 /* register with the mach service name so the client can lookup and send a message to the parent (server) */ 1920 T_LOG("Server about to check in"); 1921 kr = bootstrap_check_in(bootstrap_port, SRP_SERVICE_NAME, &port); 1922 T_ASSERT_MACH_SUCCESS(kr, "server bootstrap_check_in"); 1923 1924 T_LOG("Launching client"); 1925 sp_ret = posix_spawn(&client_pid, client_args[0], NULL, NULL, client_args, NULL); 1926 T_QUIET; T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", client_args[0], client_pid); 1927 T_LOG("Spawned client as PID %d", client_pid); 1928 1929 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 1930 T_LOG("Ready to take stackshot, but waiting 1s for the coast to clear"); 1931 1932 /* 1933 * can_continue indicates the client has signaled us, but we want to make 1934 * sure they've actually blocked sending their mach message. It's cheesy, but 1935 * sleep() works for this. 1936 */ 1937 sleep(1); 1938 1939 /* 1940 * take the stackshot without calling receive to verify that the stackshot wait 1941 * info shows our (the server) thread for the scenario where the server has yet to 1942 * receive the message. 1943 */ 1944 T_LOG("Taking stackshot for part 1 coverage"); 1945 check_srp_test("srp", SRP_TEST_THREAD); 1946 1947 /* 1948 * receive the message from the client (which should copy the send once right into 1949 * our address space). 1950 */ 1951 struct { 1952 mach_msg_header_t header; 1953 mach_msg_body_t body; 1954 mach_msg_port_descriptor_t port_descriptor; 1955 } rcv_msg = { 1956 .header = 1957 { 1958 .msgh_remote_port = MACH_PORT_NULL, 1959 .msgh_local_port = port, 1960 .msgh_size = sizeof(rcv_msg), 1961 }, 1962 }; 1963 1964 T_LOG("server: starting sync receive\n"); 1965 1966 mach_msg_return_t mr; 1967 mr = mach_msg(&(rcv_msg.header), 1968 (MACH_RCV_MSG | MACH_RCV_TIMEOUT), 1969 0, 1970 4096, 1971 port, 1972 10000, 1973 MACH_PORT_NULL); 1974 T_QUIET; T_ASSERT_MACH_SUCCESS(mr, "mach_msg() recieve of message from client"); 1975 1976 /* 1977 * take the stackshot to verify that the stackshot wait info shows our (the server) PID 1978 * for the scenario where the server has received the message and copied in the send-once right. 1979 */ 1980 T_LOG("Taking stackshot for part 2 coverage"); 1981 check_srp_test("srp", SRP_TEST_PID); 1982 1983 /* cleanup - kill the client */ 1984 T_ASSERT_POSIX_SUCCESS(kill(client_pid, SIGKILL), "killing client"); 1985 T_ASSERT_POSIX_SUCCESS(waitpid(client_pid, NULL, 0), "waiting for the client to exit"); 1986 1987 // do it again, but using kevents 1988 T_LOG("Launching client"); 1989 sp_ret = posix_spawn(&client_pid, client_args[0], NULL, NULL, client_args, NULL); 1990 T_QUIET; T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", client_args[0], client_pid); 1991 T_LOG("Spawned client as PID %d", client_pid); 1992 1993 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 1994 T_LOG("Ready to take stackshot, but waiting 1s for the coast to clear"); 1995 1996 /* 1997 * can_continue indicates the client has signaled us, but we want to make 1998 * sure they've actually blocked sending their mach message. It's cheesy, but 1999 * sleep() works for this. 2000 */ 2001 sleep(1); 2002 2003 dispatch_mach_t dispatch_mach = dispatch_mach_create(SRP_SERVICE_NAME, machdq, 2004 ^(dispatch_mach_reason_t reason, 2005 dispatch_mach_msg_t message, 2006 mach_error_t error __unused) { 2007 switch (reason) { 2008 case DISPATCH_MACH_MESSAGE_RECEIVED: { 2009 size_t size = 0; 2010 mach_msg_header_t *msg __unused = dispatch_mach_msg_get_msg(message, &size); 2011 T_LOG("server: recieved %ld byte message", size); 2012 check_srp_test("turnstile_port_thread", SRP_TEST_THREAD); 2013 T_LOG("server: letting client go"); 2014 // drop the message on the ground, we'll kill the client later 2015 dispatch_semaphore_signal(can_continue); 2016 break; 2017 } 2018 default: 2019 break; 2020 } 2021 }); 2022 2023 dispatch_mach_connect(dispatch_mach, port, MACH_PORT_NULL, NULL); 2024 2025 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 2026 2027 /* cleanup - kill the client */ 2028 T_ASSERT_POSIX_SUCCESS(kill(client_pid, SIGKILL), "killing client"); 2029 T_ASSERT_POSIX_SUCCESS(waitpid(client_pid, NULL, 0), "waiting for the client to exit"); 2030} 2031 2032T_HELPER_DECL(throtlled_sp_client, 2033 "client that uses a connection port to send a message to a server") 2034{ 2035 mach_port_t conn_port, service_port, reply_port, *stash; 2036 mach_msg_type_number_t stash_cnt = 0; 2037 2038 kern_return_t kr = mach_ports_lookup(mach_task_self(), &stash, &stash_cnt); 2039 T_ASSERT_MACH_SUCCESS(kr, "mach_ports_lookup"); 2040 2041 service_port = stash[0]; 2042 T_ASSERT_TRUE(MACH_PORT_VALID(service_port), "valid service port"); 2043 mig_deallocate((vm_address_t)stash, stash_cnt * sizeof(stash[0])); 2044 2045 mach_port_options_t opts = { 2046 .flags = MPO_INSERT_SEND_RIGHT 2047 | MPO_CONNECTION_PORT, 2048 .service_port_name = service_port, 2049 }; 2050 2051 kr = mach_port_construct(mach_task_self(), &opts, 0ull, &conn_port); 2052 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct"); 2053 2054 mach_port_options_t opts2 = { 2055 .flags = MPO_REPLY_PORT 2056 }; 2057 kr = mach_port_construct(mach_task_self(), &opts2, 0ull, &reply_port); 2058 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct"); 2059 2060 /* XPC-like check-in message */ 2061 struct { 2062 mach_msg_header_t header; 2063 mach_msg_port_descriptor_t recvp; 2064 mach_msg_port_descriptor_t sendp; 2065 } checkin_message = { 2066 .header = 2067 { 2068 .msgh_remote_port = service_port, 2069 .msgh_local_port = MACH_PORT_NULL, 2070 .msgh_size = sizeof(checkin_message), 2071 .msgh_bits = MACH_MSGH_BITS(MACH_MSG_TYPE_COPY_SEND, 0), 2072 }, 2073 .recvp = 2074 { 2075 .type = MACH_MSG_PORT_DESCRIPTOR, 2076 .name = conn_port, 2077 .disposition = MACH_MSG_TYPE_MOVE_RECEIVE, 2078 }, 2079 .sendp = 2080 { 2081 .type = MACH_MSG_PORT_DESCRIPTOR, 2082 .name = reply_port, 2083 .disposition = MACH_MSG_TYPE_MAKE_SEND, 2084 } 2085 }; 2086 dispatch_mach_msg_t dmsg = dispatch_mach_msg_create((mach_msg_header_t *)&checkin_message, sizeof(checkin_message), 2087 DISPATCH_MACH_MSG_DESTRUCTOR_DEFAULT, NULL); 2088 2089 dispatch_queue_t machdq = dispatch_queue_create("machqueue", NULL); 2090 dispatch_mach_t dchannel = dispatch_mach_create(THROTTLED_SERVICE_NAME, machdq, 2091 ^(dispatch_mach_reason_t reason, 2092 dispatch_mach_msg_t message __unused, 2093 mach_error_t error __unused) { 2094 switch (reason) { 2095 case DISPATCH_MACH_CONNECTED: 2096 T_LOG("mach channel connected"); 2097 break; 2098 case DISPATCH_MACH_MESSAGE_SENT: 2099 T_LOG("sent mach message"); 2100 break; 2101 default: 2102 T_ASSERT_FAIL("Unexpected reply to channel reason %lu", reason); 2103 } 2104 }); 2105 dispatch_mach_connect(dchannel, reply_port, service_port, dmsg); 2106 dispatch_release(dmsg); 2107 2108 struct { 2109 mach_msg_header_t header; 2110 uint64_t request_id; 2111 } request = { 2112 .header = 2113 { 2114 .msgh_size = sizeof(request), 2115 .msgh_bits = MACH_MSGH_BITS(MACH_MSG_TYPE_COPY_SEND, MACH_MSG_TYPE_MAKE_SEND_ONCE), 2116 }, 2117 .request_id = 1, 2118 }; 2119 dispatch_mach_msg_t dmsg2 = dispatch_mach_msg_create((mach_msg_header_t *)&request, sizeof(request), 2120 DISPATCH_MACH_MSG_DESTRUCTOR_DEFAULT, NULL); 2121 2122 dispatch_mach_reason_t reason; 2123 mach_error_t error; 2124 2125 /* send the check-in message and the request message */ 2126 dispatch_mach_msg_t dreply = dispatch_mach_send_with_result_and_wait_for_reply(dchannel, 2127 dmsg2, 0, DISPATCH_MACH_SEND_DEFAULT, &reason, &error); 2128 dispatch_release(dmsg2); 2129 2130 /* not expected to execute as parent will SIGKILL client */ 2131 T_ASSERT_FAIL("client process exiting after receiving %s reply", dreply ? "non-null" : "null"); 2132} 2133 2134static void 2135check_throttled_sp(const char *test_name, uint64_t context, bool is_throttled) 2136{ 2137 struct scenario scenario = { 2138 .name = test_name, 2139 .quiet = false, 2140 .flags = (STACKSHOT_THREAD_WAITINFO | STACKSHOT_KCDATA_FORMAT), 2141 }; 2142 2143 T_LOG("taking stackshot %s", test_name); 2144 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2145 parse_stackshot(PARSE_STACKSHOT_THROTTLED_SP, ssbuf, sslen, 2146 @{sp_throttled_expected_ctxt_key: @(context), 2147 sp_throttled_expect_flag: @(is_throttled)}); 2148 }); 2149} 2150 2151/* Take stackshot when a client is blocked on the service port of a process, in the scenario when 2152 * the process with the receive right for the service port is: 2153 * (a) Monitoring the service port using kevents 2154 * (b) Not monitoring the service port 2155 */ 2156T_DECL(throttled_sp, 2157 "test that service port throttled flag is propagated to the stackshot correctly", T_META_TAG_VM_PREFERRED) 2158{ 2159 mach_port_t service_port; 2160 __block dispatch_semaphore_t can_continue = dispatch_semaphore_create(0); 2161 2162 char path[PATH_MAX]; 2163 uint32_t path_size = sizeof(path); 2164 T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 2165 char *client_args[] = { path, "-n", "throtlled_sp_client", NULL }; 2166 2167 __block uint64_t thread_id = 0; 2168 pid_t client_pid; 2169 int mark_throttled; 2170 2171 struct mach_service_port_info sp_info = {}; 2172 strcpy(sp_info.mspi_string_name, THROTTLED_SERVICE_NAME); 2173 sp_info.mspi_domain_type = (uint8_t)1; 2174 kern_return_t kr; 2175 2176 mach_port_options_t opts = { 2177 .flags = MPO_SERVICE_PORT | MPO_INSERT_SEND_RIGHT | MPO_CONTEXT_AS_GUARD | MPO_STRICT | MPO_TEMPOWNER, 2178 .service_port_info = &sp_info, 2179 }; 2180 2181 kr = mach_port_construct(mach_task_self(), &opts, 0ull, &service_port); 2182 T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct %u", service_port); 2183 2184 /* Setup a dispatch source to monitor the service port similar to how launchd does. */ 2185 dispatch_queue_t machdq = dispatch_queue_create("machqueue", NULL); 2186 dispatch_source_t mach_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_MACH_RECV, service_port, 2187 DISPATCH_MACH_RECV_SYNC_PEEK, machdq); 2188 dispatch_source_set_event_handler(mach_src, ^{ 2189 pthread_threadid_np(NULL, &thread_id); 2190 dispatch_semaphore_signal(can_continue); 2191 }); 2192 dispatch_activate(mach_src); 2193 2194 /* Stash the port in task to make sure child also gets it */ 2195 kr = mach_ports_register(mach_task_self(), &service_port, 1); 2196 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_ports_register service port"); 2197 2198 mark_throttled = 1; 2199 kr = mach_port_set_attributes(mach_task_self(), service_port, MACH_PORT_SERVICE_THROTTLED, (mach_port_info_t)(&mark_throttled), 2200 MACH_PORT_SERVICE_THROTTLED_COUNT); 2201 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mark service port as throttled"); 2202 2203 int rc = posix_spawn(&client_pid, client_args[0], NULL, NULL, client_args, NULL); 2204 T_QUIET; T_ASSERT_POSIX_ZERO(rc, "spawned process '%s' with PID %d", client_args[0], client_pid); 2205 T_LOG("Spawned client as PID %d", client_pid); 2206 2207 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 2208 2209 /* The service port has received the check-in message. Take stackshot for scenario (a). */ 2210 check_throttled_sp("throttled_service_port_monitored", thread_id, true); 2211 2212 /* This simulates a throttled spawn when the service port is no longer monitored. */ 2213 dispatch_source_cancel(mach_src); 2214 2215 /* Take stackshot for scenario (b) */ 2216 check_throttled_sp("throttled_service_port_unmonitored", (uint64_t)getpid(), true); 2217 2218 mark_throttled = 0; 2219 kr = mach_port_set_attributes(mach_task_self(), service_port, MACH_PORT_SERVICE_THROTTLED, (mach_port_info_t)(&mark_throttled), 2220 MACH_PORT_SERVICE_THROTTLED_COUNT); 2221 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "unmark service port as throttled"); 2222 2223 /* Throttled flag should not be set when the port is not throttled. */ 2224 check_throttled_sp("unthrottled_service_port_unmonitored", (uint64_t)getpid(), false); 2225 2226 /* cleanup - kill the client */ 2227 T_ASSERT_POSIX_SUCCESS(kill(client_pid, SIGKILL), "killing client"); 2228 T_ASSERT_POSIX_SUCCESS(waitpid(client_pid, NULL, 0), "waiting for the client to exit"); 2229} 2230 2231 2232char *const clpcctrl_path = "/usr/local/bin/clpcctrl"; 2233 2234static void 2235run_clpcctrl(char *const argv[]) { 2236 posix_spawnattr_t sattr; 2237 pid_t pid; 2238 int wstatus; 2239 2240 T_QUIET; T_ASSERT_POSIX_ZERO(posix_spawn(&pid, argv[0], NULL, NULL, argv, NULL), "spawn clpcctrl"); 2241 T_QUIET; T_ASSERT_POSIX_SUCCESS(waitpid(pid, &wstatus, 0), "wait for clpcctrl"); 2242 T_QUIET; T_ASSERT_TRUE(WIFEXITED(wstatus), "clpcctrl exited normally"); 2243 T_QUIET; T_ASSERT_POSIX_ZERO(WEXITSTATUS(wstatus), "clpcctrl exited successfully"); 2244 2245 uint64_t sched_recommended_cores = 1; 2246 size_t sched_recommended_cores_sz = sizeof(uint64_t); 2247 T_QUIET; T_ASSERT_POSIX_SUCCESS( 2248 sysctlbyname("kern.sched_recommended_cores", &sched_recommended_cores, &sched_recommended_cores_sz, NULL, 0), 2249 "get kern.sched_recommended_cores"); 2250 T_LOG("Recommended cores: 0x%llx", sched_recommended_cores); 2251} 2252 2253static void 2254restore_clpcctrl() { 2255 run_clpcctrl((char *const []) { clpcctrl_path, "-d", NULL }); 2256} 2257 2258#define CLUSTER_TYPE_SMP 0 2259#define CLUSTER_TYPE_E 1 2260#define CLUSTER_TYPE_P 2 2261 2262void test_stackshot_cpu_info(void *ssbuf, size_t sslen, int exp_cpus, NSArray *exp_cluster_types) { 2263 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2264 bool seen = false; 2265 int singlethread_override = 0; 2266 size_t singlethread_override_sz = sizeof(singlethread_override); 2267 T_QUIET; T_ASSERT_POSIX_SUCCESS( 2268 sysctlbyname("kern.stackshot_single_thread", &singlethread_override, &singlethread_override_sz, NULL, 0), 2269 "get kern.stackshot_single_thread"); 2270 if (singlethread_override) { 2271 T_LOG("skipping cpu count/type check due to single-thread override (kern.stackshot_single_thread=1)"); 2272 return; 2273 } 2274 2275 KCDATA_ITER_FOREACH(iter) { 2276 if ((kcdata_iter_type(iter) != KCDATA_TYPE_ARRAY) || (kcdata_iter_array_elem_type(iter) != STACKSHOT_KCTYPE_LATENCY_INFO_CPU)) { 2277 continue; 2278 } 2279 2280 seen = true; 2281 2282 /* Check ncpus */ 2283 int ncpus = kcdata_iter_array_elem_count(iter); 2284 if (exp_cpus != -1) { 2285 T_QUIET; T_ASSERT_EQ(exp_cpus, ncpus, "Expected number of CPUs matches number of CPUs used for stackshot"); 2286 } 2287 2288 if (exp_cluster_types == nil) { 2289 continue; 2290 } 2291 2292 /* Check cluster types */ 2293 struct stackshot_latency_cpu *latencies = (struct stackshot_latency_cpu *) kcdata_iter_payload(iter); 2294 for (int i = 0; i < ncpus; i++) { 2295 NSNumber *cluster_type = [NSNumber numberWithInt:latencies[i].cluster_type]; 2296 T_QUIET; T_ASSERT_TRUE([exp_cluster_types containsObject:cluster_type], "Type of CPU cluster in expected CPU cluster types"); 2297 } 2298 } 2299 2300 T_QUIET; T_ASSERT_TRUE(seen || !is_development_kernel(), "Seen CPU latency info or is release kernel"); 2301} 2302 2303void test_stackshot_with_clpcctrl(char *const name, char *const argv[], int exp_cpus, NSArray *exp_cluster_types) { 2304 T_LOG("Stackshot CLPC scenario %s", name); 2305 run_clpcctrl(argv); 2306 struct scenario scenario = { 2307 .name = name, 2308 .flags = (STACKSHOT_KCDATA_FORMAT | STACKSHOT_SAVE_LOADINFO | 2309 STACKSHOT_THREAD_WAITINFO | STACKSHOT_GET_GLOBAL_MEM_STATS) 2310 }; 2311 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2312 parse_stackshot(0, ssbuf, sslen, nil); 2313 test_stackshot_cpu_info(ssbuf, sslen, exp_cpus, exp_cluster_types); 2314 }); 2315} 2316 2317T_DECL(core_masks, 2318 "test that stackshot works under various core masks on ARM systems", 2319 T_META_REQUIRES_SYSCTL_EQ("hw.optional.arm64", 1), 2320 T_META_REQUIRES_SYSCTL_NE("kern.kasan.available", 1), /* rdar://115577993 */ 2321 XNU_T_META_REQUIRES_DEVELOPMENT_KERNEL, 2322 T_META_REQUIRE_NOT_VIRTUALIZED, 2323 T_META_RUN_CONCURRENTLY(false), 2324 T_META_TAG_VM_NOT_ELIGIBLE, 2325 T_META_ENABLED(!TARGET_OS_VISION)) // disable for visionOS: device may not be stable with many cores masked off (127904530) 2326{ 2327 /* 2328 * Make sure we're not in a release kernel 2329 * (cannot check with T_META; only one sysctl T_META at a time will work) 2330 */ 2331 if (!is_development_kernel()) { 2332 T_SKIP("test was not run because kernel is release; cannot set core masks"); 2333 return; 2334 } 2335 2336 /* 2337 * rdar://115577993 - CLPC compiles as release in KASAN-variant builds, 2338 * preventing clpcctrl from working. For now, skip this. (Cannot check 2339 * with T_META; only one sysctl T_META at a time will work) 2340 */ 2341 int kasan_avail = 0; 2342 size_t kasan_avail_sz = sizeof(kasan_avail); 2343 sysctlbyname("kern.kasan.available", &kasan_avail, &kasan_avail_sz, NULL, 0); 2344 if (kasan_avail) { 2345 T_SKIP("test was not run because kernel is KASAN; cannot set core masks (see rdar://115577993)"); 2346 return; 2347 } 2348 2349 2350 T_ATEND(restore_clpcctrl); 2351 2352 /* Test with 1 and 2 CPUs for basic functionality */ 2353 test_stackshot_with_clpcctrl( 2354 "core_masks_1cpu", (char *const[]) {clpcctrl_path, "-c", "1", NULL}, 2355 1, nil); 2356 2357 test_stackshot_with_clpcctrl( 2358 "core_masks_2cpus", (char *const[]) {clpcctrl_path, "-c", "2", NULL}, 2359 2, nil); 2360 2361 /* Check nperflevels to see if we're on an AMP system */ 2362 int nperflevels = 1; 2363 size_t nperflevels_sz = sizeof(int); 2364 T_ASSERT_POSIX_SUCCESS( 2365 sysctlbyname("hw.nperflevels", &nperflevels, &nperflevels_sz, NULL, 0), 2366 "get hw.nperflevels"); 2367 if (nperflevels == 1) { 2368 T_LOG("On SMP system, skipping stackshot core_masks AMP tests"); 2369 return; 2370 } 2371 2372 T_QUIET; T_ASSERT_EQ(nperflevels, 2, "nperflevels is 1 or 2"); 2373 T_LOG("On AMP system, performing stackshot core_masks AMP tests"); 2374 2375 /* Perform AMP tests with different cluster types active */ 2376 test_stackshot_with_clpcctrl( 2377 "core_masks_amp_allcpus", 2378 (char *const[]) {clpcctrl_path, "-C", "all", NULL}, 2379 -1, @[@CLUSTER_TYPE_E, @CLUSTER_TYPE_P]); 2380 2381 test_stackshot_with_clpcctrl( 2382 "core_masks_amp_ecpus", 2383 (char *const[]) {clpcctrl_path, "-C", "e", NULL}, 2384 -1, @[@CLUSTER_TYPE_E]); 2385 2386 test_stackshot_with_clpcctrl( 2387 "core_masks_amp_pcpus", 2388 (char *const[]) {clpcctrl_path, "-C", "p", NULL}, 2389 -1, @[@CLUSTER_TYPE_P]); 2390} 2391 2392#pragma mark performance tests 2393 2394#define SHOULD_REUSE_SIZE_HINT 0x01 2395#define SHOULD_USE_DELTA 0x02 2396#define SHOULD_TARGET_SELF 0x04 2397 2398static void 2399stackshot_perf(unsigned int options) 2400{ 2401 struct scenario scenario = { 2402 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 2403 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 2404 }; 2405 2406 dt_stat_t size = dt_stat_create("bytes", "size"); 2407 dt_stat_time_t duration = dt_stat_time_create("duration"); 2408 scenario.timer = duration; 2409 2410 if (options & SHOULD_TARGET_SELF) { 2411 scenario.target_pid = getpid(); 2412 } 2413 2414 while (!dt_stat_stable(duration) || !dt_stat_stable(size)) { 2415 __block uint64_t last_time = 0; 2416 __block uint32_t size_hint = 0; 2417 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2418 dt_stat_add(size, (double)sslen); 2419 last_time = stackshot_timestamp(ssbuf, sslen); 2420 size_hint = (uint32_t)sslen; 2421 }); 2422 if (options & SHOULD_USE_DELTA) { 2423 scenario.since_timestamp = last_time; 2424 scenario.flags |= STACKSHOT_COLLECT_DELTA_SNAPSHOT; 2425 } 2426 if (options & SHOULD_REUSE_SIZE_HINT) { 2427 scenario.size_hint = size_hint; 2428 } 2429 } 2430 2431 dt_stat_finalize(duration); 2432 dt_stat_finalize(size); 2433} 2434 2435static void 2436stackshot_flag_perf_noclobber(uint64_t flag, char *flagname) 2437{ 2438 struct scenario scenario = { 2439 .quiet = true, 2440 .flags = (flag | STACKSHOT_KCDATA_FORMAT), 2441 }; 2442 2443 dt_stat_t duration = dt_stat_create("nanoseconds per thread", "%s_duration", flagname); 2444 dt_stat_t size = dt_stat_create("bytes per thread", "%s_size", flagname); 2445 T_LOG("Testing \"%s\" = 0x%" PRIx64, flagname, flag); 2446 2447 while (!dt_stat_stable(duration) || !dt_stat_stable(size)) { 2448 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2449 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2450 unsigned long no_threads = 0; 2451 mach_timebase_info_data_t timebase = {0, 0}; 2452 uint64_t stackshot_duration = 0; 2453 int found = 0; 2454 T_QUIET; T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, "stackshot buffer"); 2455 2456 KCDATA_ITER_FOREACH(iter) { 2457 switch(kcdata_iter_type(iter)) { 2458 case STACKSHOT_KCTYPE_THREAD_SNAPSHOT: { 2459 found |= 1; 2460 no_threads ++; 2461 break; 2462 } 2463 case STACKSHOT_KCTYPE_STACKSHOT_DURATION: { 2464 struct stackshot_duration *ssd = kcdata_iter_payload(iter); 2465 stackshot_duration = ssd->stackshot_duration; 2466 found |= 2; 2467 break; 2468 } 2469 case KCDATA_TYPE_TIMEBASE: { 2470 found |= 4; 2471 mach_timebase_info_data_t *tb = kcdata_iter_payload(iter); 2472 memcpy(&timebase, tb, sizeof(timebase)); 2473 break; 2474 } 2475 } 2476 } 2477 2478 T_QUIET; T_ASSERT_EQ(found, 0x7, "found everything needed"); 2479 2480 uint64_t ns = (stackshot_duration * timebase.numer) / timebase.denom; 2481 uint64_t per_thread_ns = ns / no_threads; 2482 uint64_t per_thread_size = sslen / no_threads; 2483 2484 dt_stat_add(duration, per_thread_ns); 2485 dt_stat_add(size, per_thread_size); 2486 }); 2487 } 2488 2489 dt_stat_finalize(duration); 2490 dt_stat_finalize(size); 2491} 2492 2493static void 2494stackshot_flag_perf(uint64_t flag, char *flagname) 2495{ 2496 /* 2497 * STACKSHOT_NO_IO_STATS disables data collection, so set it for 2498 * more accurate perfdata collection. 2499 */ 2500 flag |= STACKSHOT_NO_IO_STATS; 2501 2502 stackshot_flag_perf_noclobber(flag, flagname); 2503} 2504 2505 2506T_DECL(flag_perf, "test stackshot performance with different flags set", T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2507{ 2508 stackshot_flag_perf_noclobber(STACKSHOT_NO_IO_STATS, "baseline"); 2509 stackshot_flag_perf_noclobber(0, "io_stats"); 2510 2511 stackshot_flag_perf(STACKSHOT_THREAD_WAITINFO, "thread_waitinfo"); 2512 stackshot_flag_perf(STACKSHOT_GET_DQ, "get_dq"); 2513 stackshot_flag_perf(STACKSHOT_SAVE_LOADINFO, "save_loadinfo"); 2514 stackshot_flag_perf(STACKSHOT_GET_GLOBAL_MEM_STATS, "get_global_mem_stats"); 2515 stackshot_flag_perf(STACKSHOT_SAVE_KEXT_LOADINFO, "save_kext_loadinfo"); 2516 stackshot_flag_perf(STACKSHOT_SAVE_IMP_DONATION_PIDS, "save_imp_donation_pids"); 2517 stackshot_flag_perf(STACKSHOT_ENABLE_BT_FAULTING, "enable_bt_faulting"); 2518 stackshot_flag_perf(STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT, "collect_sharedcache_layout"); 2519 stackshot_flag_perf(STACKSHOT_ENABLE_UUID_FAULTING, "enable_uuid_faulting"); 2520 stackshot_flag_perf(STACKSHOT_THREAD_GROUP, "thread_group"); 2521 stackshot_flag_perf(STACKSHOT_SAVE_JETSAM_COALITIONS, "save_jetsam_coalitions"); 2522 stackshot_flag_perf(STACKSHOT_INSTRS_CYCLES, "instrs_cycles"); 2523 stackshot_flag_perf(STACKSHOT_ASID, "asid"); 2524 stackshot_flag_perf(STACKSHOT_EXCLAVES, "all_exclaves"); 2525 stackshot_flag_perf(STACKSHOT_EXCLAVES | STACKSHOT_ASID, "all_exclaves_and_asid"); 2526 stackshot_flag_perf(STACKSHOT_SKIP_EXCLAVES, "skip_exclaves"); 2527} 2528 2529T_DECL(perf_no_size_hint, "test stackshot performance with no size hint", 2530 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2531{ 2532 stackshot_perf(0); 2533} 2534 2535T_DECL(perf_size_hint, "test stackshot performance with size hint", 2536 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2537{ 2538 stackshot_perf(SHOULD_REUSE_SIZE_HINT); 2539} 2540 2541T_DECL(perf_process, "test stackshot performance targeted at process", 2542 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2543{ 2544 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_TARGET_SELF); 2545} 2546 2547T_DECL(perf_delta, "test delta stackshot performance", 2548 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2549{ 2550 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_USE_DELTA); 2551} 2552 2553T_DECL(perf_delta_no_exclaves, "test delta stackshot performance without Exclaves", 2554 T_META_REQUIRES_SYSCTL_EQ("kern.exclaves_status", 1), 2555 T_META_REQUIRES_SYSCTL_EQ("kern.exclaves_inspection_status", 1), 2556 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2557{ 2558 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_USE_DELTA | STACKSHOT_SKIP_EXCLAVES); 2559} 2560 2561T_DECL(perf_delta_process, "test delta stackshot performance targeted at a process", 2562 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2563{ 2564 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_USE_DELTA | SHOULD_TARGET_SELF); 2565} 2566 2567T_DECL(stackshot_entitlement_report_test, "test stackshot entitlement report", T_META_TAG_VM_PREFERRED) 2568{ 2569 int sysctlValue = 1; 2570 T_ASSERT_POSIX_SUCCESS( 2571 sysctlbyname("debug.stackshot_entitlement_send_batch", NULL, NULL, &sysctlValue, sizeof(sysctlValue)), 2572 "set debug.stackshot_entitlement_send_batch=1"); 2573 // having a way to verify that the coreanalytics event was received would be even better 2574 // See rdar://74197197 2575 T_PASS("entitlement test ran"); 2576} 2577 2578static void 2579expect_os_build_version_in_stackshot(void *ssbuf, size_t sslen) 2580{ 2581 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2582 2583 bool saw_os_build_version = false; 2584 iter = kcdata_iter_next(iter); 2585 2586 KCDATA_ITER_FOREACH(iter) { 2587 switch (kcdata_iter_type(iter)) { 2588 case STACKSHOT_KCTYPE_OS_BUILD_VERSION: 2589 saw_os_build_version = true; 2590 T_LOG("Found os build version in stackshot: %s", kcdata_iter_payload(iter)); 2591 return; 2592 2593 default: 2594 break; 2595 } 2596 } 2597 2598 T_ASSERT_FAIL("didn't see os build version in stackshot"); 2599} 2600 2601T_DECL(os_build_version, "test stackshot contains os build version", T_META_TAG_VM_PREFERRED) 2602{ 2603 2604 struct scenario scenario = { 2605 .name = "os-build-version", 2606 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT), 2607 }; 2608 2609 T_LOG("attempting to take stackshot with an os build version"); 2610 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2611 expect_os_build_version_in_stackshot(ssbuf, sslen); 2612 }); 2613} 2614 2615static uint64_t 2616stackshot_timestamp(void *ssbuf, size_t sslen) 2617{ 2618 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2619 2620 uint32_t type = kcdata_iter_type(iter); 2621 if (type != KCDATA_BUFFER_BEGIN_STACKSHOT && type != KCDATA_BUFFER_BEGIN_DELTA_STACKSHOT) { 2622 T_ASSERT_FAIL("invalid kcdata type %u", kcdata_iter_type(iter)); 2623 } 2624 2625 iter = kcdata_iter_find_type(iter, KCDATA_TYPE_MACH_ABSOLUTE_TIME); 2626 T_QUIET; 2627 T_ASSERT_TRUE(kcdata_iter_valid(iter), "timestamp found in stackshot"); 2628 2629 return *(uint64_t *)kcdata_iter_payload(iter); 2630} 2631 2632#define TEST_THREAD_NAME "stackshot_test_thread" 2633 2634static void 2635parse_thread_group_stackshot(void **ssbuf, size_t sslen) 2636{ 2637 bool seen_thread_group_snapshot = false; 2638 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2639 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, 2640 "buffer provided is a stackshot"); 2641 2642 NSMutableSet *thread_groups = [[NSMutableSet alloc] init]; 2643 2644 iter = kcdata_iter_next(iter); 2645 KCDATA_ITER_FOREACH(iter) { 2646 switch (kcdata_iter_type(iter)) { 2647 case KCDATA_TYPE_ARRAY: { 2648 T_QUIET; 2649 T_ASSERT_TRUE(kcdata_iter_array_valid(iter), 2650 "checked that array is valid"); 2651 2652 if (kcdata_iter_array_elem_type(iter) != STACKSHOT_KCTYPE_THREAD_GROUP_SNAPSHOT) { 2653 continue; 2654 } 2655 2656 seen_thread_group_snapshot = true; 2657 2658 if (kcdata_iter_array_elem_size(iter) >= sizeof(struct thread_group_snapshot_v3)) { 2659 struct thread_group_snapshot_v3 *tgs_array = kcdata_iter_payload(iter); 2660 for (uint32_t j = 0; j < kcdata_iter_array_elem_count(iter); j++) { 2661 struct thread_group_snapshot_v3 *tgs = tgs_array + j; 2662 [thread_groups addObject:@(tgs->tgs_id)]; 2663 } 2664 } 2665 else { 2666 struct thread_group_snapshot *tgs_array = kcdata_iter_payload(iter); 2667 for (uint32_t j = 0; j < kcdata_iter_array_elem_count(iter); j++) { 2668 struct thread_group_snapshot *tgs = tgs_array + j; 2669 [thread_groups addObject:@(tgs->tgs_id)]; 2670 } 2671 } 2672 break; 2673 } 2674 } 2675 } 2676 KCDATA_ITER_FOREACH(iter) { 2677 NSError *error = nil; 2678 2679 switch (kcdata_iter_type(iter)) { 2680 2681 case KCDATA_TYPE_CONTAINER_BEGIN: { 2682 T_QUIET; 2683 T_ASSERT_TRUE(kcdata_iter_container_valid(iter), 2684 "checked that container is valid"); 2685 2686 if (kcdata_iter_container_type(iter) != STACKSHOT_KCCONTAINER_THREAD) { 2687 break; 2688 } 2689 2690 NSDictionary *container = parseKCDataContainer(&iter, &error); 2691 T_QUIET; T_ASSERT_NOTNULL(container, "parsed thread container from stackshot"); 2692 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing container"); 2693 2694 int tg = [container[@"thread_snapshots"][@"thread_group"] intValue]; 2695 2696 T_ASSERT_TRUE([thread_groups containsObject:@(tg)], "check that the thread group the thread is in exists"); 2697 2698 break; 2699 }; 2700 2701 } 2702 } 2703 T_ASSERT_TRUE(seen_thread_group_snapshot, "check that we have seen a thread group snapshot"); 2704} 2705 2706static void 2707verify_stackshot_sharedcache_layout(struct dyld_uuid_info_64 *uuids, uint32_t uuid_count) 2708{ 2709 uuid_t cur_shared_cache_uuid; 2710 __block uint32_t lib_index = 0, libs_found = 0; 2711 2712 _dyld_get_shared_cache_uuid(cur_shared_cache_uuid); 2713 int result = dyld_shared_cache_iterate_text(cur_shared_cache_uuid, ^(const dyld_shared_cache_dylib_text_info* info) { 2714 T_QUIET; T_ASSERT_LT(lib_index, uuid_count, "dyld_shared_cache_iterate_text exceeded number of libraries returned by kernel"); 2715 2716 libs_found++; 2717 struct dyld_uuid_info_64 *cur_stackshot_uuid_entry = &uuids[lib_index]; 2718 T_QUIET; T_ASSERT_EQ(memcmp(info->dylibUuid, cur_stackshot_uuid_entry->imageUUID, sizeof(info->dylibUuid)), 0, 2719 "dyld returned UUID doesn't match kernel returned UUID"); 2720 T_QUIET; T_ASSERT_EQ(info->loadAddressUnslid, cur_stackshot_uuid_entry->imageLoadAddress, 2721 "dyld returned load address doesn't match kernel returned load address"); 2722 lib_index++; 2723 }); 2724 2725 T_ASSERT_EQ(result, 0, "iterate shared cache layout"); 2726 T_ASSERT_EQ(libs_found, uuid_count, "dyld iterator returned same number of libraries as kernel"); 2727 2728 T_LOG("verified %d libraries from dyld shared cache", libs_found); 2729} 2730 2731static void 2732check_shared_cache_uuid(uuid_t imageUUID) 2733{ 2734 static uuid_t shared_cache_uuid; 2735 static dispatch_once_t read_shared_cache_uuid; 2736 2737 dispatch_once(&read_shared_cache_uuid, ^{ 2738 T_QUIET; 2739 T_ASSERT_TRUE(_dyld_get_shared_cache_uuid(shared_cache_uuid), "retrieve current shared cache UUID"); 2740 }); 2741 T_QUIET; T_ASSERT_EQ(uuid_compare(shared_cache_uuid, imageUUID), 0, 2742 "dyld returned UUID doesn't match kernel returned UUID for system shared cache"); 2743} 2744 2745/* 2746 * extra dictionary contains data relevant for the given flags: 2747 * PARSE_STACKSHOT_ZOMBIE: zombie_child_pid_key -> @(pid) 2748 * PARSE_STACKSHOT_POSTEXEC: postexec_child_unique_pid_key -> @(unique_pid) 2749 */ 2750static void 2751parse_stackshot(uint64_t stackshot_parsing_flags, void *ssbuf, size_t sslen, NSDictionary *extra) 2752{ 2753 bool delta = (stackshot_parsing_flags & PARSE_STACKSHOT_DELTA); 2754 bool expect_sharedcache_child = (stackshot_parsing_flags & PARSE_STACKSHOT_SHAREDCACHE_FLAGS); 2755 bool expect_zombie_child = (stackshot_parsing_flags & PARSE_STACKSHOT_ZOMBIE); 2756 bool expect_postexec_child = (stackshot_parsing_flags & PARSE_STACKSHOT_POSTEXEC); 2757 bool expect_cseg_waitinfo = (stackshot_parsing_flags & PARSE_STACKSHOT_WAITINFO_CSEG); 2758 bool expect_translated_child = (stackshot_parsing_flags & PARSE_STACKSHOT_TRANSLATED); 2759 bool expect_shared_cache_layout = false; 2760 bool expect_shared_cache_uuid = !delta; 2761 bool expect_dispatch_queue_label = (stackshot_parsing_flags & PARSE_STACKSHOT_DISPATCH_QUEUE_LABEL); 2762 bool expect_turnstile_lock = (stackshot_parsing_flags & PARSE_STACKSHOT_TURNSTILEINFO); 2763 bool expect_srp_waitinfo = (stackshot_parsing_flags & PARSE_STACKSHOT_WAITINFO_SRP); 2764 bool expect_sp_throttled = (stackshot_parsing_flags & PARSE_STACKSHOT_THROTTLED_SP); 2765 bool expect_exec_inprogress = (stackshot_parsing_flags & PARSE_STACKSHOT_EXEC_INPROGRESS); 2766 bool expect_transitioning_task = (stackshot_parsing_flags & PARSE_STACKSHOT_TRANSITIONING); 2767 bool expect_asyncstack = (stackshot_parsing_flags & PARSE_STACKSHOT_ASYNCSTACK); 2768 bool expect_driverkit = (stackshot_parsing_flags & PARSE_STACKSHOT_DRIVERKIT); 2769 bool expect_suspendinfo = (stackshot_parsing_flags & PARSE_STACKSHOT_SUSPENDINFO); 2770 bool found_zombie_child = false, found_postexec_child = false, found_shared_cache_layout = false, found_shared_cache_uuid = false; 2771 bool found_translated_child = false, found_transitioning_task = false; 2772 bool found_dispatch_queue_label = false, found_turnstile_lock = false; 2773 bool found_cseg_waitinfo = false, found_srp_waitinfo = false; 2774 bool found_sharedcache_child = false, found_sharedcache_badflags = false, found_sharedcache_self = false; 2775 bool found_asyncstack = false; 2776 bool found_throttled_service = false; 2777 bool found_exclaves = false; 2778 bool expect_single_task = (stackshot_parsing_flags & PARSE_STACKSHOT_TARGETPID); 2779 uint64_t srp_expected_threadid = 0; 2780 pid_t zombie_child_pid = -1, srp_expected_pid = -1, sharedcache_child_pid = -1, throttled_service_ctx = -1; 2781 pid_t translated_child_pid = -1, transistioning_task_pid = -1; 2782 bool sharedcache_child_sameaddr = false, is_throttled = false; 2783 uint64_t postexec_child_unique_pid = 0, cseg_expected_threadid = 0; 2784 uint64_t sharedcache_child_flags = 0, sharedcache_self_flags = 0; 2785 uint64_t asyncstack_threadid = 0; 2786 NSArray *asyncstack_stack = nil; 2787 char *inflatedBufferBase = NULL; 2788 pid_t exec_inprogress_pid = -1; 2789 void (^exec_inprogress_cb)(uint64_t, uint64_t) = NULL; 2790 int exec_inprogress_found = 0; 2791 uint64_t exec_inprogress_containerid = 0; 2792 void (^driverkit_cb)(pid_t) = NULL; 2793 NSMutableDictionary *sharedCaches = [NSMutableDictionary new]; 2794 uint64_t expected_num_threads = 0, expected_num_tasks = 0, found_percpu_threads = 0, found_tasks = 0, found_percpu_tasks = 0; 2795 NSMutableSet *seen_tasks = [NSMutableSet new]; 2796 2797 if (expect_shared_cache_uuid) { 2798 uuid_t shared_cache_uuid; 2799 if (!_dyld_get_shared_cache_uuid(shared_cache_uuid)) { 2800 T_LOG("Skipping verifying shared cache UUID in stackshot data because not running with a shared cache"); 2801 expect_shared_cache_uuid = false; 2802 } 2803 } 2804 2805 if (stackshot_parsing_flags & PARSE_STACKSHOT_SHAREDCACHE_LAYOUT) { 2806 size_t shared_cache_length = 0; 2807 const void *cache_header = _dyld_get_shared_cache_range(&shared_cache_length); 2808 T_QUIET; T_ASSERT_NOTNULL(cache_header, "current process running with shared cache"); 2809 T_QUIET; T_ASSERT_GT(shared_cache_length, sizeof(struct _dyld_cache_header), "valid shared cache length populated by _dyld_get_shared_cache_range"); 2810 2811 if (_dyld_shared_cache_is_locally_built()) { 2812 T_LOG("device running with locally built shared cache, expect shared cache layout"); 2813 expect_shared_cache_layout = true; 2814 } else { 2815 T_LOG("device running with B&I built shared-cache, no shared cache layout expected"); 2816 } 2817 } 2818 2819 if (expect_sharedcache_child) { 2820 NSNumber* pid_num = extra[sharedcache_child_pid_key]; 2821 NSNumber* sameaddr_num = extra[sharedcache_child_sameaddr_key]; 2822 T_QUIET; T_ASSERT_NOTNULL(pid_num, "sharedcache child pid provided"); 2823 T_QUIET; T_ASSERT_NOTNULL(sameaddr_num, "sharedcache child addrsame provided"); 2824 sharedcache_child_pid = [pid_num intValue]; 2825 T_QUIET; T_ASSERT_GT(sharedcache_child_pid, 0, "sharedcache child pid greater than zero"); 2826 sharedcache_child_sameaddr = [sameaddr_num intValue]; 2827 T_QUIET; T_ASSERT_GE([sameaddr_num intValue], 0, "sharedcache child sameaddr is boolean (0 or 1)"); 2828 T_QUIET; T_ASSERT_LE([sameaddr_num intValue], 1, "sharedcache child sameaddr is boolean (0 or 1)"); 2829 } 2830 2831 if (expect_transitioning_task) { 2832 NSNumber* pid_num = extra[transitioning_pid_key]; 2833 T_ASSERT_NOTNULL(pid_num, "transitioning task pid provided"); 2834 transistioning_task_pid = [pid_num intValue]; 2835 } 2836 2837 if (expect_zombie_child) { 2838 NSNumber* pid_num = extra[zombie_child_pid_key]; 2839 T_QUIET; T_ASSERT_NOTNULL(pid_num, "zombie child pid provided"); 2840 zombie_child_pid = [pid_num intValue]; 2841 T_QUIET; T_ASSERT_GT(zombie_child_pid, 0, "zombie child pid greater than zero"); 2842 } 2843 2844 if (expect_postexec_child) { 2845 NSNumber* unique_pid_num = extra[postexec_child_unique_pid_key]; 2846 T_QUIET; T_ASSERT_NOTNULL(unique_pid_num, "postexec child unique pid provided"); 2847 postexec_child_unique_pid = [unique_pid_num unsignedLongLongValue]; 2848 T_QUIET; T_ASSERT_GT(postexec_child_unique_pid, 0ull, "postexec child unique pid greater than zero"); 2849 } 2850 2851 if (expect_cseg_waitinfo) { 2852 NSNumber* tid_num = extra[cseg_expected_threadid_key]; 2853 T_QUIET; T_ASSERT_NOTNULL(tid_num, "cseg's expected thread id provided"); 2854 cseg_expected_threadid = tid_num.unsignedLongValue; 2855 T_QUIET; T_ASSERT_GT(cseg_expected_threadid, UINT64_C(0), "compressor segment thread is present"); 2856 } 2857 2858 if (expect_srp_waitinfo) { 2859 NSNumber* threadid_num = extra[srp_expected_threadid_key]; 2860 NSNumber* pid_num = extra[srp_expected_pid_key]; 2861 T_QUIET; T_ASSERT_TRUE(threadid_num != nil || pid_num != nil, "expected SRP threadid or pid"); 2862 if (threadid_num != nil) { 2863 srp_expected_threadid = [threadid_num unsignedLongLongValue]; 2864 T_QUIET; T_ASSERT_GT(srp_expected_threadid, 0ull, "srp_expected_threadid greater than zero"); 2865 } 2866 if (pid_num != nil) { 2867 srp_expected_pid = [pid_num intValue]; 2868 T_QUIET; T_ASSERT_GT(srp_expected_pid, 0, "srp_expected_pid greater than zero"); 2869 } 2870 T_LOG("looking for SRP pid: %d threadid: %llu", srp_expected_pid, srp_expected_threadid); 2871 } 2872 2873 if (expect_sp_throttled) { 2874 NSNumber* ctx = extra[sp_throttled_expected_ctxt_key]; 2875 T_QUIET; T_ASSERT_TRUE(ctx != nil, "expected pid"); 2876 throttled_service_ctx = [ctx intValue]; 2877 T_QUIET; T_ASSERT_GT(throttled_service_ctx, 0, "expected pid greater than zero"); 2878 2879 NSNumber *throttled = extra[sp_throttled_expect_flag]; 2880 T_QUIET; T_ASSERT_TRUE(throttled != nil, "expected flag value"); 2881 is_throttled = ([throttled intValue] != 0); 2882 2883 T_LOG("Looking for service with ctxt: %d, thottled:%d", throttled_service_ctx, is_throttled); 2884 } 2885 2886 if (expect_translated_child) { 2887 NSNumber* pid_num = extra[translated_child_pid_key]; 2888 T_QUIET; T_ASSERT_NOTNULL(pid_num, "translated child pid provided"); 2889 translated_child_pid = [pid_num intValue]; 2890 T_QUIET; T_ASSERT_GT(translated_child_pid, 0, "translated child pid greater than zero"); 2891 } 2892 if (expect_exec_inprogress) { 2893 NSNumber* pid_num = extra[exec_inprogress_pid_key]; 2894 T_QUIET; T_ASSERT_NOTNULL(pid_num, "exec inprogress pid provided"); 2895 exec_inprogress_pid = [pid_num intValue]; 2896 T_QUIET; T_ASSERT_GT(exec_inprogress_pid, 0, "exec inprogress pid greater than zero"); 2897 2898 exec_inprogress_cb = extra[exec_inprogress_found_key]; 2899 T_QUIET; T_ASSERT_NOTNULL(exec_inprogress_cb, "exec inprogress found callback provided"); 2900 } 2901 if (expect_driverkit) { 2902 driverkit_cb = extra[driverkit_found_key]; 2903 T_QUIET; T_ASSERT_NOTNULL(driverkit_cb, "driverkit found callback provided"); 2904 } 2905 2906 if (expect_asyncstack) { 2907 NSNumber* threadid_id = extra[asyncstack_expected_threadid_key]; 2908 T_QUIET; T_ASSERT_NOTNULL(threadid_id, "asyncstack threadid provided"); 2909 asyncstack_threadid = [threadid_id unsignedLongLongValue]; 2910 asyncstack_stack = extra[asyncstack_expected_stack_key]; 2911 T_QUIET; T_ASSERT_NOTNULL(asyncstack_stack, "asyncstack expected stack provided"); 2912 } 2913 2914 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2915 if (delta) { 2916 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_DELTA_STACKSHOT, 2917 "buffer provided is a delta stackshot"); 2918 2919 iter = kcdata_iter_next(iter); 2920 } else { 2921 if (kcdata_iter_type(iter) != KCDATA_BUFFER_BEGIN_COMPRESSED) { 2922 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, 2923 "buffer provided is a stackshot"); 2924 2925 iter = kcdata_iter_next(iter); 2926 } else { 2927 /* we are dealing with a compressed buffer */ 2928 iter = kcdata_iter_next(iter); 2929 uint64_t compression_type = 0, totalout = 0, totalin = 0; 2930 2931 uint64_t *data; 2932 char *desc; 2933 for (int i = 0; i < 3; i ++) { 2934 kcdata_iter_get_data_with_desc(iter, &desc, (void **)&data, NULL); 2935 if (strcmp(desc, "kcd_c_type") == 0) { 2936 compression_type = *data; 2937 } else if (strcmp(desc, "kcd_c_totalout") == 0){ 2938 totalout = *data; 2939 } else if (strcmp(desc, "kcd_c_totalin") == 0){ 2940 totalin = *data; 2941 } 2942 2943 iter = kcdata_iter_next(iter); 2944 } 2945 2946 T_ASSERT_EQ(compression_type, UINT64_C(1), "zlib compression is used"); 2947 T_ASSERT_GT(totalout, UINT64_C(0), "successfully gathered how long the compressed buffer is"); 2948 T_ASSERT_GT(totalin, UINT64_C(0), "successfully gathered how long the uncompressed buffer will be at least"); 2949 2950 /* progress to the next kcdata item */ 2951 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, "compressed stackshot found"); 2952 2953 char *bufferBase = kcdata_iter_payload(iter); 2954 2955 /* 2956 * zlib is used, allocate a buffer based on the metadata, plus 2957 * extra scratch space (+12.5%) in case totalin was inconsistent 2958 */ 2959 size_t inflatedBufferSize = totalin + (totalin >> 3); 2960 inflatedBufferBase = malloc(inflatedBufferSize); 2961 T_QUIET; T_WITH_ERRNO; T_ASSERT_NOTNULL(inflatedBufferBase, "allocated temporary output buffer"); 2962 2963 z_stream zs; 2964 memset(&zs, 0, sizeof(zs)); 2965 T_QUIET; T_ASSERT_EQ(inflateInit(&zs), Z_OK, "inflateInit OK"); 2966 zs.next_in = (unsigned char *)bufferBase; 2967 T_QUIET; T_ASSERT_LE(totalout, (uint64_t)UINT_MAX, "stackshot is not too large"); 2968 zs.avail_in = (uInt)totalout; 2969 zs.next_out = (unsigned char *)inflatedBufferBase; 2970 T_QUIET; T_ASSERT_LE(inflatedBufferSize, (size_t)UINT_MAX, "output region is not too large"); 2971 zs.avail_out = (uInt)inflatedBufferSize; 2972 T_ASSERT_EQ(inflate(&zs, Z_FINISH), Z_STREAM_END, "inflated buffer"); 2973 inflateEnd(&zs); 2974 2975 T_ASSERT_EQ((uint64_t)zs.total_out, totalin, "expected number of bytes inflated"); 2976 2977 /* copy the data after the compressed area */ 2978 T_QUIET; T_ASSERT_GE((void *)bufferBase, ssbuf, 2979 "base of compressed stackshot is after the returned stackshot buffer"); 2980 size_t header_size = (size_t)(bufferBase - (char *)ssbuf); 2981 size_t data_after_compressed_size = sslen - totalout - header_size; 2982 T_QUIET; T_ASSERT_LE(data_after_compressed_size, 2983 inflatedBufferSize - zs.total_out, 2984 "footer fits in the buffer"); 2985 memcpy(inflatedBufferBase + zs.total_out, 2986 bufferBase + totalout, 2987 data_after_compressed_size); 2988 2989 iter = kcdata_iter(inflatedBufferBase, inflatedBufferSize); 2990 } 2991 } 2992 2993 KCDATA_ITER_FOREACH(iter) { 2994 NSError *error = nil; 2995 2996 switch (kcdata_iter_type(iter)) { 2997 case KCDATA_TYPE_ARRAY: { 2998 T_QUIET; 2999 T_ASSERT_TRUE(kcdata_iter_array_valid(iter), 3000 "checked that array is valid"); 3001 3002 NSMutableDictionary *array = parseKCDataArray(iter, &error); 3003 T_QUIET; T_ASSERT_NOTNULL(array, "parsed array from stackshot"); 3004 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing array"); 3005 3006 if (kcdata_iter_array_elem_type(iter) == STACKSHOT_KCTYPE_SYS_SHAREDCACHE_LAYOUT) { 3007 struct dyld_uuid_info_64 *shared_cache_uuids = kcdata_iter_payload(iter); 3008 uint32_t uuid_count = kcdata_iter_array_elem_count(iter); 3009 T_ASSERT_NOTNULL(shared_cache_uuids, "parsed shared cache layout array"); 3010 T_ASSERT_GT(uuid_count, 0, "returned valid number of UUIDs from shared cache"); 3011 verify_stackshot_sharedcache_layout(shared_cache_uuids, uuid_count); 3012 found_shared_cache_layout = true; 3013 } 3014 3015 break; 3016 } 3017 case KCDATA_TYPE_CONTAINER_BEGIN: { 3018 T_QUIET; 3019 T_ASSERT_TRUE(kcdata_iter_container_valid(iter), 3020 "checked that container is valid"); 3021 3022 uint64_t containerid = kcdata_iter_container_id(iter); 3023 uint32_t container_type = kcdata_iter_container_type(iter); 3024 3025 if (container_type == STACKSHOT_KCCONTAINER_SHAREDCACHE) { 3026 NSDictionary *container = parseKCDataContainer(&iter, &error); 3027 T_QUIET; T_ASSERT_NOTNULL(container, "parsed sharedcache container from stackshot"); 3028 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing sharedcache container"); 3029 T_QUIET; T_EXPECT_EQ(sharedCaches[@(containerid)], nil, "sharedcache containerid %lld should be unique", containerid); 3030 sharedCaches[@(containerid)] = container; 3031 break; 3032 } 3033 3034 if (container_type == STACKSHOT_KCCONTAINER_EXCLAVES) { 3035 found_exclaves = true; 3036 break; 3037 } 3038 3039 /* 3040 * treat containers other than tasks/transitioning_tasks 3041 * as expanded in-line. 3042 */ 3043 if (container_type != STACKSHOT_KCCONTAINER_TASK && 3044 container_type != STACKSHOT_KCCONTAINER_TRANSITIONING_TASK) { 3045 T_LOG("container skipped: %d", container_type); 3046 break; 3047 } 3048 NSDictionary *container = parseKCDataContainer(&iter, &error); 3049 T_QUIET; T_ASSERT_NOTNULL(container, "parsed task/transitioning_task container from stackshot"); 3050 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing container"); 3051 3052 found_tasks++; 3053 3054 NSDictionary* task_snapshot = container[@"task_snapshots"][@"task_snapshot"]; 3055 NSDictionary* task_delta_snapshot = container[@"task_snapshots"][@"task_delta_snapshot"]; 3056 NSDictionary* transitioning_task_snapshot = container[@"transitioning_task_snapshots"][@"transitioning_task_snapshot"]; 3057 3058 NSNumber *task_pid = NULL; 3059 if (task_snapshot) { 3060 task_pid = task_snapshot[@"ts_unique_pid"]; 3061 } else if(task_delta_snapshot) { 3062 task_pid = task_snapshot[@"tds_unique_pid"]; 3063 } else if(transitioning_task_snapshot) { 3064 task_pid = transitioning_task_snapshot[@"tts_pid"]; 3065 } 3066 3067 if (task_pid && [seen_tasks containsObject:task_pid]) { 3068 T_QUIET; T_ASSERT_FALSE([seen_tasks containsObject:task_pid], "No duplicate PIDs in stackshot"); 3069 [seen_tasks addObject:task_pid]; 3070 } 3071 3072 /* 3073 * Having processed the container, we now only check it 3074 * if it's the correct type. 3075 */ 3076 if ((!expect_transitioning_task && (container_type != STACKSHOT_KCCONTAINER_TASK)) || 3077 (expect_transitioning_task && (container_type != STACKSHOT_KCCONTAINER_TRANSITIONING_TASK))) { 3078 break; 3079 } 3080 if (!expect_transitioning_task) { 3081 T_QUIET; T_ASSERT_TRUE(!!task_snapshot != !!task_delta_snapshot, "Either task_snapshot xor task_delta_snapshot provided"); 3082 } 3083 3084 if (expect_dispatch_queue_label && !found_dispatch_queue_label) { 3085 for (id thread_key in container[@"task_snapshots"][@"thread_snapshots"]) { 3086 NSMutableDictionary *thread = container[@"task_snapshots"][@"thread_snapshots"][thread_key]; 3087 NSString *dql = thread[@"dispatch_queue_label"]; 3088 3089 if ([dql isEqualToString:@TEST_STACKSHOT_QUEUE_LABEL]) { 3090 found_dispatch_queue_label = true; 3091 break; 3092 } 3093 } 3094 } 3095 3096 if (expect_transitioning_task && !found_transitioning_task) { 3097 if (transitioning_task_snapshot) { 3098 uint64_t the_pid = [transitioning_task_snapshot[@"tts_pid"] unsignedLongLongValue]; 3099 if (the_pid == (uint64_t)transistioning_task_pid) { 3100 found_transitioning_task = true; 3101 3102 T_PASS("FOUND Transitioning task %llu has a transitioning task snapshot", (uint64_t) transistioning_task_pid); 3103 break; 3104 } 3105 } 3106 } 3107 3108 if (expect_postexec_child && !found_postexec_child) { 3109 if (task_snapshot) { 3110 uint64_t unique_pid = [task_snapshot[@"ts_unique_pid"] unsignedLongLongValue]; 3111 if (unique_pid == postexec_child_unique_pid) { 3112 found_postexec_child = true; 3113 3114 T_PASS("post-exec child %llu has a task snapshot", postexec_child_unique_pid); 3115 3116 break; 3117 } 3118 } 3119 3120 if (task_delta_snapshot) { 3121 uint64_t unique_pid = [task_delta_snapshot[@"tds_unique_pid"] unsignedLongLongValue]; 3122 if (unique_pid == postexec_child_unique_pid) { 3123 found_postexec_child = true; 3124 3125 T_FAIL("post-exec child %llu shouldn't have a delta task snapshot", postexec_child_unique_pid); 3126 3127 break; 3128 } 3129 } 3130 } 3131 3132 int pid = [task_snapshot[@"ts_pid"] intValue]; 3133 3134 if (pid && expect_shared_cache_uuid && !found_shared_cache_uuid) { 3135 id ptr = container[@"task_snapshots"][@"shared_cache_dyld_load_info"]; 3136 if (ptr) { 3137 id uuid = ptr[@"imageUUID"]; 3138 3139 uint8_t uuid_p[16]; 3140 for (unsigned int i = 0; i < 16; i ++) { 3141 NSNumber *uuidByte = uuid[i]; 3142 uuid_p[i] = (uint8_t)uuidByte.charValue; 3143 } 3144 3145 check_shared_cache_uuid(uuid_p); 3146 3147 uint64_t baseAddress = (uint64_t)((NSNumber *)ptr[@"imageSlidBaseAddress"]).longLongValue; 3148 uint64_t firstMapping = (uint64_t)((NSNumber *)ptr[@"sharedCacheSlidFirstMapping"]).longLongValue; 3149 3150 T_EXPECT_LE(baseAddress, firstMapping, 3151 "in per-task shared_cache_dyld_load_info, " 3152 "baseAddress <= firstMapping"); 3153 T_EXPECT_GE(baseAddress + (7ull << 32) + (1ull << 29), 3154 firstMapping, 3155 "in per-task shared_cache_dyld_load_info, " 3156 "baseAddress + 28.5gig >= firstMapping"); 3157 3158 size_t shared_cache_len; 3159 const void *addr = _dyld_get_shared_cache_range(&shared_cache_len); 3160 T_EXPECT_EQ((uint64_t)addr, firstMapping, 3161 "SlidFirstMapping should match shared_cache_range"); 3162 3163 /* 3164 * check_shared_cache_uuid() will assert on failure, so if 3165 * we get here, then we have found the shared cache UUID 3166 * and it's correct 3167 */ 3168 found_shared_cache_uuid = true; 3169 } 3170 } 3171 3172 if (expect_sharedcache_child) { 3173 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3174 uint64_t sharedregion_flags = (task_flags & (kTaskSharedRegionNone | kTaskSharedRegionSystem | kTaskSharedRegionOther)); 3175 id sharedregion_info = container[@"task_snapshots"][@"shared_cache_dyld_load_info"]; 3176 id sharedcache_id = container[@"task_snapshots"][@"sharedCacheID"]; 3177 if (!found_sharedcache_badflags) { 3178 T_QUIET; T_EXPECT_NE(sharedregion_flags, 0ll, "one of the kTaskSharedRegion flags should be set on all tasks"); 3179 bool multiple = (sharedregion_flags & (sharedregion_flags - 1)) != 0; 3180 T_QUIET; T_EXPECT_FALSE(multiple, "only one kTaskSharedRegion flag should be set on each task"); 3181 found_sharedcache_badflags = (sharedregion_flags == 0 || multiple); 3182 } 3183 if (pid == 0) { 3184 T_ASSERT_EQ(sharedregion_flags, (uint64_t)kTaskSharedRegionNone, "Kernel proc (pid 0) should have no shared region"); 3185 } else if (pid == sharedcache_child_pid) { 3186 found_sharedcache_child = true; 3187 sharedcache_child_flags = sharedregion_flags; 3188 } else if (pid == getpid()) { 3189 found_sharedcache_self = true; 3190 sharedcache_self_flags = sharedregion_flags; 3191 } 3192 if (sharedregion_flags == kTaskSharedRegionOther && !(task_flags & kTaskSharedRegionInfoUnavailable)) { 3193 T_QUIET; T_EXPECT_NOTNULL(sharedregion_info, "kTaskSharedRegionOther should have a shared_cache_dyld_load_info struct"); 3194 T_QUIET; T_EXPECT_NOTNULL(sharedcache_id, "kTaskSharedRegionOther should have a sharedCacheID"); 3195 if (sharedcache_id != nil) { 3196 T_QUIET; T_EXPECT_NOTNULL(sharedCaches[sharedcache_id], "sharedCacheID %d should exist", [sharedcache_id intValue]); 3197 } 3198 } else { 3199 T_QUIET; T_EXPECT_NULL(sharedregion_info, "non-kTaskSharedRegionOther should have no shared_cache_dyld_load_info struct"); 3200 T_QUIET; T_EXPECT_NULL(sharedcache_id, "non-kTaskSharedRegionOther should have no sharedCacheID"); 3201 } 3202 } 3203 3204 if (expect_zombie_child && (pid == zombie_child_pid)) { 3205 found_zombie_child = true; 3206 3207 expected_num_tasks += 1; 3208 3209 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3210 T_ASSERT_TRUE((task_flags & kTerminatedSnapshot) == kTerminatedSnapshot, "child zombie marked as terminated"); 3211 3212 continue; 3213 } 3214 3215 if (expect_translated_child && (pid == translated_child_pid)) { 3216 found_translated_child = true; 3217 3218 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3219 T_EXPECT_BITS_SET(task_flags, kTaskIsTranslated, "child marked as translated"); 3220 3221 continue; 3222 } 3223 if (expect_exec_inprogress && (pid == exec_inprogress_pid || pid == -exec_inprogress_pid)) { 3224 exec_inprogress_found++; 3225 T_LOG("found exec task with pid %d, instance %d", pid, exec_inprogress_found); 3226 T_QUIET; T_ASSERT_LE(exec_inprogress_found, 2, "no more than two with the expected pid"); 3227 if (exec_inprogress_found == 2) { 3228 T_LOG("found 2 tasks with pid %d", exec_inprogress_pid); 3229 exec_inprogress_cb(containerid, exec_inprogress_containerid); 3230 } else { 3231 exec_inprogress_containerid = containerid; 3232 } 3233 } 3234 if (expect_driverkit && driverkit_cb != NULL) { 3235 driverkit_cb(pid); 3236 } 3237 if (expect_cseg_waitinfo) { 3238 NSArray *winfos = container[@"task_snapshots"][@"thread_waitinfo"]; 3239 3240 for (id i in winfos) { 3241 NSNumber *waitType = i[@"wait_type"]; 3242 NSNumber *owner = i[@"owner"]; 3243 if (waitType.intValue == kThreadWaitCompressor && 3244 owner.unsignedLongValue == cseg_expected_threadid) { 3245 found_cseg_waitinfo = true; 3246 break; 3247 } 3248 } 3249 } 3250 3251 if (expect_srp_waitinfo) { 3252 NSArray *tinfos = container[@"task_snapshots"][@"thread_turnstileinfo"]; 3253 NSArray *winfos = container[@"task_snapshots"][@"thread_waitinfo"]; 3254 for (id i in tinfos) { 3255 if (!found_srp_waitinfo) { 3256 bool found_thread = false; 3257 bool found_pid = false; 3258 if (([i[@"turnstile_flags"] intValue] & STACKSHOT_TURNSTILE_STATUS_THREAD) && 3259 [i[@"turnstile_context"] unsignedLongLongValue] == srp_expected_threadid && 3260 srp_expected_threadid != 0) { 3261 found_thread = true; 3262 } 3263 if (([i[@"turnstile_flags"] intValue] & STACKSHOT_TURNSTILE_STATUS_BLOCKED_ON_TASK) && 3264 [i[@"turnstile_context"] intValue] == srp_expected_pid && 3265 srp_expected_pid != -1) { 3266 found_pid = true; 3267 } 3268 if (found_pid || found_thread) { 3269 T_LOG("found SRP %s %lld waiter: %d", (found_thread ? "thread" : "pid"), 3270 [i[@"turnstile_context"] unsignedLongLongValue], [i[@"waiter"] intValue]); 3271 /* we found something that is blocking the correct threadid */ 3272 for (id j in winfos) { 3273 if ([j[@"waiter"] intValue] == [i[@"waiter"] intValue] && 3274 [j[@"wait_type"] intValue] == kThreadWaitPortReceive) { 3275 found_srp_waitinfo = true; 3276 T_EXPECT_EQ([j[@"wait_flags"] intValue], STACKSHOT_WAITINFO_FLAGS_SPECIALREPLY, 3277 "SRP waitinfo should be marked as a special reply"); 3278 break; 3279 } 3280 } 3281 3282 if (found_srp_waitinfo) { 3283 break; 3284 } 3285 } 3286 } 3287 } 3288 } 3289 3290 if (expect_sp_throttled) { 3291 NSArray *tinfos = container[@"task_snapshots"][@"thread_turnstileinfo"]; 3292 for (id i in tinfos) { 3293 if (([i[@"turnstile_flags"] intValue] & STACKSHOT_TURNSTILE_STATUS_PORTFLAGS) 3294 && [i[@"turnstile_context"] intValue] == throttled_service_ctx) { 3295 int portlabel_id = [i[@"portlabel_id"] intValue]; 3296 T_LOG("[pid:%d] Turnstile (flags = 0x%x, ctx = %d, portlabel_id = %d)", pid, 3297 [i[@"turnstile_flags"] intValue], [i[@"turnstile_context"] intValue], portlabel_id); 3298 for (id portid in container[@"task_snapshots"][@"portlabels"]) { 3299 if (portlabel_id != [portid intValue]) { 3300 continue; 3301 } 3302 3303 NSMutableDictionary *portlabel = container[@"task_snapshots"][@"portlabels"][portid]; 3304 T_ASSERT_TRUE(portlabel != nil, "Found portlabel id: %d", [portid intValue]); 3305 NSString *portlabel_name = portlabel[@"portlabel_name"]; 3306 T_EXPECT_TRUE(portlabel_name != nil, "Found portlabel %s", portlabel_name.UTF8String); 3307 T_EXPECT_EQ_STR(portlabel_name.UTF8String, THROTTLED_SERVICE_NAME, "throttled service port name matches"); 3308 T_EXPECT_EQ(([portlabel[@"portlabel_flags"] intValue] & STACKSHOT_PORTLABEL_THROTTLED) != 0, 3309 is_throttled, "Port %s throttled", is_throttled ? "is" : "isn't"); 3310 found_throttled_service = true; 3311 break; 3312 } 3313 } 3314 3315 if (found_throttled_service) { 3316 break; 3317 } 3318 } 3319 } 3320 3321 if (expect_suspendinfo) { 3322 // TODO: rdar://112563110 3323 } 3324 3325 3326 if (pid != getpid()) { 3327 break; 3328 } 3329 3330 T_EXPECT_EQ_STR(current_process_name(), 3331 [task_snapshot[@"ts_p_comm"] UTF8String], 3332 "current process name matches in stackshot"); 3333 3334 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3335 T_ASSERT_BITS_NOTSET(task_flags, kTerminatedSnapshot, "current process not marked as terminated"); 3336 T_ASSERT_BITS_NOTSET(task_flags, kTaskIsTranslated, "current process not marked as translated"); 3337 3338 T_QUIET; 3339 T_EXPECT_LE(pid, [task_snapshot[@"ts_unique_pid"] intValue], 3340 "unique pid is greater than pid"); 3341 3342 NSDictionary* task_cpu_architecture = container[@"task_snapshots"][@"task_cpu_architecture"]; 3343 T_QUIET; T_ASSERT_NOTNULL(task_cpu_architecture[@"cputype"], "have cputype"); 3344 T_QUIET; T_ASSERT_NOTNULL(task_cpu_architecture[@"cpusubtype"], "have cputype"); 3345 int cputype = [task_cpu_architecture[@"cputype"] intValue]; 3346 int cpusubtype = [task_cpu_architecture[@"cpusubtype"] intValue]; 3347 3348 struct proc_archinfo archinfo; 3349 int retval = proc_pidinfo(pid, PROC_PIDARCHINFO, 0, &archinfo, sizeof(archinfo)); 3350 T_QUIET; T_WITH_ERRNO; T_ASSERT_GT(retval, 0, "proc_pidinfo(PROC_PIDARCHINFO) returned a value > 0"); 3351 T_QUIET; T_ASSERT_EQ(retval, (int)sizeof(struct proc_archinfo), "proc_pidinfo call for PROC_PIDARCHINFO returned expected size"); 3352 T_QUIET; T_EXPECT_EQ(cputype, archinfo.p_cputype, "cpu type is correct"); 3353 T_QUIET; T_EXPECT_EQ(cpusubtype, archinfo.p_cpusubtype, "cpu subtype is correct"); 3354 3355 NSDictionary * codesigning_info = container[@"task_snapshots"][@"stackshot_task_codesigning_info"]; 3356 T_QUIET; T_ASSERT_NOTNULL(codesigning_info[@"csflags"], "have csflags"); 3357 uint64_t flags = [codesigning_info[@"csflags"] unsignedLongLongValue]; 3358 T_QUIET; T_EXPECT_GT(flags, 0, "nonzero csflags"); 3359 3360 T_QUIET; T_ASSERT_NOTNULL(container[@"task_snapshots"][@"jetsam_coalition"], "have jetsam coalition"); 3361 uint64_t jetsam_coalition = [container[@"task_snapshots"][@"jetsam_coalition"] unsignedLongLongValue]; 3362 T_QUIET; T_EXPECT_GT(jetsam_coalition, 0, "nonzero jetsam coalition"); 3363 3364 bool found_main_thread = false; 3365 uint64_t main_thread_id = -1ULL; 3366 bool found_null_kernel_frame = false; 3367 for (id thread_key in container[@"task_snapshots"][@"thread_snapshots"]) { 3368 NSMutableDictionary *thread = container[@"task_snapshots"][@"thread_snapshots"][thread_key]; 3369 NSDictionary *thread_snap = thread[@"thread_snapshot"]; 3370 3371 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_thread_id"] intValue], 0, 3372 "thread ID of thread in current task is valid"); 3373 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_base_priority"] intValue], 0, 3374 "base priority of thread in current task is valid"); 3375 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_sched_priority"] intValue], 0, 3376 "scheduling priority of thread in current task is valid"); 3377 3378 NSString *pth_name = thread[@"pth_name"]; 3379 if (pth_name != nil && [pth_name isEqualToString:@TEST_THREAD_NAME]) { 3380 found_main_thread = true; 3381 main_thread_id = [thread_snap[@"ths_thread_id"] unsignedLongLongValue]; 3382 3383 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_total_syscalls"] intValue], 0, 3384 "total syscalls of current thread is valid"); 3385 3386 NSDictionary *cpu_times = thread[@"cpu_times"]; 3387 T_EXPECT_GE([cpu_times[@"runnable_time"] intValue], 3388 [cpu_times[@"system_time"] intValue] + 3389 [cpu_times[@"user_time"] intValue], 3390 "runnable time of current thread is valid"); 3391 } 3392 if (!found_null_kernel_frame) { 3393 for (NSNumber *frame in thread[@"kernel_frames"]) { 3394 if (frame.unsignedLongValue == 0) { 3395 found_null_kernel_frame = true; 3396 break; 3397 } 3398 } 3399 } 3400 if (expect_asyncstack && !found_asyncstack && 3401 asyncstack_threadid == [thread_snap[@"ths_thread_id"] unsignedLongLongValue]) { 3402 found_asyncstack = true; 3403 NSArray* async_stack = thread[@"user_async_stack_frames"]; 3404 NSNumber* start_idx = thread[@"user_async_start_index"]; 3405 NSArray* user_stack = thread[@"user_stack_frames"]; 3406 T_QUIET; T_ASSERT_NOTNULL(async_stack, "async thread %#llx has user_async_stack_frames", asyncstack_threadid); 3407 T_QUIET; T_ASSERT_NOTNULL(start_idx, "async thread %#llx has user_async_start_index", asyncstack_threadid); 3408 T_QUIET; T_ASSERT_NOTNULL(user_stack, "async thread %#llx has user_stack_frames", asyncstack_threadid); 3409 T_QUIET; T_ASSERT_EQ(async_stack.count, asyncstack_stack.count, 3410 "actual async_stack count == expected async_stack count"); 3411 for (size_t i = 0; i < async_stack.count; i++) { 3412 T_EXPECT_EQ([async_stack[i][@"lr"] unsignedLongLongValue], 3413 [asyncstack_stack[i] unsignedLongLongValue], "frame %zu matches", i); 3414 } 3415 } 3416 } 3417 T_EXPECT_TRUE(found_main_thread, "found main thread for current task in stackshot"); 3418 T_EXPECT_FALSE(found_null_kernel_frame, "should not see any NULL kernel frames"); 3419 3420 if (expect_turnstile_lock && !found_turnstile_lock) { 3421 NSArray *tsinfos = container[@"task_snapshots"][@"thread_turnstileinfo"]; 3422 3423 for (id i in tsinfos) { 3424 if ([i[@"turnstile_context"] unsignedLongLongValue] == main_thread_id) { 3425 found_turnstile_lock = true; 3426 break; 3427 } 3428 } 3429 } 3430 break; 3431 } 3432 case STACKSHOT_KCTYPE_SHAREDCACHE_LOADINFO: { 3433 // Legacy shared cache info 3434 struct dyld_shared_cache_loadinfo *payload = kcdata_iter_payload(iter); 3435 T_ASSERT_EQ((size_t)kcdata_iter_size(iter), sizeof(*payload), "valid dyld_shared_cache_loadinfo struct"); 3436 3437 check_shared_cache_uuid(payload->sharedCacheUUID); 3438 3439 T_EXPECT_LE(payload->sharedCacheUnreliableSlidBaseAddress, 3440 payload->sharedCacheSlidFirstMapping, 3441 "SlidBaseAddress <= SlidFirstMapping"); 3442 T_EXPECT_GE(payload->sharedCacheUnreliableSlidBaseAddress + (7ull << 32) + (1ull << 29), 3443 payload->sharedCacheSlidFirstMapping, 3444 "SlidFirstMapping should be within 28.5gigs of SlidBaseAddress"); 3445 3446 size_t shared_cache_len; 3447 const void *addr = _dyld_get_shared_cache_range(&shared_cache_len); 3448 T_EXPECT_EQ((uint64_t)addr, payload->sharedCacheSlidFirstMapping, 3449 "SlidFirstMapping should match shared_cache_range"); 3450 3451 /* 3452 * check_shared_cache_uuid() asserts on failure, so we must have 3453 * found the shared cache UUID to be correct. 3454 */ 3455 found_shared_cache_uuid = true; 3456 break; 3457 } 3458 case KCDATA_TYPE_UINT64_DESC: { 3459 char *desc; 3460 uint64_t *data; 3461 uint32_t size; 3462 kcdata_iter_get_data_with_desc(iter, &desc, &data, &size); 3463 3464 if (strcmp(desc, "stackshot_tasks_count") == 0) { 3465 expected_num_tasks = *data; 3466 } else if (strcmp(desc, "stackshot_threads_count") == 0) { 3467 expected_num_threads = *data; 3468 } 3469 3470 break; 3471 } 3472 case STACKSHOT_KCTYPE_LATENCY_INFO_CPU: { 3473 struct stackshot_latency_cpu *cpu_latency = kcdata_iter_payload(iter); 3474 found_percpu_tasks += cpu_latency->tasks_processed; 3475 found_percpu_threads += cpu_latency->threads_processed; 3476 break; 3477 } 3478 } 3479 } 3480 3481 if (expect_sharedcache_child) { 3482 T_QUIET; T_ASSERT_TRUE(found_sharedcache_child, "found sharedcache child in kcdata"); 3483 T_QUIET; T_ASSERT_TRUE(found_sharedcache_self, "found self in kcdata"); 3484 if (found_sharedcache_child && found_sharedcache_self) { 3485 T_QUIET; T_ASSERT_NE(sharedcache_child_flags, (uint64_t)kTaskSharedRegionNone, "sharedcache child should have shared region"); 3486 T_QUIET; T_ASSERT_NE(sharedcache_self_flags, (uint64_t)kTaskSharedRegionNone, "sharedcache: self should have shared region"); 3487 if (sharedcache_self_flags == kTaskSharedRegionSystem && !sharedcache_child_sameaddr) { 3488 /* If we're in the system shared region, and the child has a different address, child must have an Other shared region */ 3489 T_ASSERT_EQ(sharedcache_child_flags, (uint64_t)kTaskSharedRegionOther, 3490 "sharedcache child should have Other shared region"); 3491 } 3492 } 3493 } 3494 3495 if (expect_transitioning_task) { 3496 T_QUIET; T_ASSERT_TRUE(found_transitioning_task, "found transitioning_task child in kcdata"); 3497 } 3498 3499 if (expect_exec_inprogress) { 3500 T_QUIET; T_ASSERT_GT(exec_inprogress_found, 0, "found at least 1 task for execing process"); 3501 } 3502 3503 if (expect_zombie_child) { 3504 T_QUIET; T_ASSERT_TRUE(found_zombie_child, "found zombie child in kcdata"); 3505 } 3506 3507 if (expect_postexec_child) { 3508 T_QUIET; T_ASSERT_TRUE(found_postexec_child, "found post-exec child in kcdata"); 3509 } 3510 3511 if (expect_translated_child) { 3512 T_QUIET; T_ASSERT_TRUE(found_translated_child, "found translated child in kcdata"); 3513 } 3514 3515 if (expect_shared_cache_layout) { 3516 T_QUIET; T_ASSERT_TRUE(found_shared_cache_layout, "shared cache layout found in kcdata"); 3517 } 3518 3519 if (expect_shared_cache_uuid) { 3520 T_QUIET; T_ASSERT_TRUE(found_shared_cache_uuid, "shared cache UUID found in kcdata"); 3521 } 3522 3523 if (expect_dispatch_queue_label) { 3524 T_QUIET; T_ASSERT_TRUE(found_dispatch_queue_label, "dispatch queue label found in kcdata"); 3525 } 3526 3527 if (expect_turnstile_lock) { 3528 T_QUIET; T_ASSERT_TRUE(found_turnstile_lock, "found expected deadlock"); 3529 } 3530 3531 if (expect_cseg_waitinfo) { 3532 T_QUIET; T_ASSERT_TRUE(found_cseg_waitinfo, "found c_seg waitinfo"); 3533 } 3534 3535 if (expect_srp_waitinfo) { 3536 T_QUIET; T_ASSERT_TRUE(found_srp_waitinfo, "found special reply port waitinfo"); 3537 } 3538 3539 if (expect_sp_throttled) { 3540 T_QUIET; T_ASSERT_TRUE(found_throttled_service, "found the throttled service"); 3541 } 3542 3543 if (expect_asyncstack) { 3544 T_QUIET; T_ASSERT_TRUE(found_asyncstack, "found async stack threadid"); 3545 } 3546 3547 if ([extra objectForKey:no_exclaves_key] != nil) { 3548 T_QUIET; T_ASSERT_FALSE(found_exclaves, "did not find any Exclaves data"); 3549 } 3550 3551 3552 bool check_counts = !delta && !found_transitioning_task && !expect_single_task && !expect_driverkit; 3553 3554 if (check_counts && (expected_num_threads != 0) && (found_percpu_threads != 0)) { 3555 /* If the task counts below check out, we can be sure that the per-cpu reported thread counts are accurate. */ 3556 T_QUIET; T_ASSERT_EQ_ULLONG(found_percpu_threads, expected_num_threads, "number of threads reported by CPUs matches expected count"); 3557 } 3558 3559 if (check_counts && (expected_num_tasks != 0)) { 3560 T_QUIET; T_ASSERT_EQ_ULLONG(found_tasks, expected_num_tasks, "number of tasks in kcdata matches expected count"); 3561 if (found_percpu_tasks != 0) { 3562 T_QUIET; T_ASSERT_EQ_ULLONG(found_percpu_tasks, expected_num_tasks, "number of tasks reported by CPUs matches expected count"); 3563 } 3564 } 3565 3566 T_ASSERT_FALSE(KCDATA_ITER_FOREACH_FAILED(iter), "successfully iterated kcdata"); 3567 3568 free(inflatedBufferBase); 3569} 3570 3571static const char * 3572current_process_name(void) 3573{ 3574 static char name[64]; 3575 3576 if (!name[0]) { 3577 int ret = proc_name(getpid(), name, sizeof(name)); 3578 T_QUIET; 3579 T_ASSERT_POSIX_SUCCESS(ret, "proc_name failed for current process"); 3580 } 3581 3582 return name; 3583} 3584 3585static void 3586initialize_thread(void) 3587{ 3588 int ret = pthread_setname_np(TEST_THREAD_NAME); 3589 T_QUIET; 3590 T_ASSERT_POSIX_ZERO(ret, "set thread name to %s", TEST_THREAD_NAME); 3591} 3592 3593T_DECL(dirty_buffer, "test that stackshot works with a dirty input buffer from kernel", T_META_TAG_VM_PREFERRED) 3594{ 3595 const char *test_sysctl = "stackshot_dirty_buffer"; 3596 int64_t result; 3597 3598 T_LOG("running sysctl to trigger kernel-driven stackshot"); 3599 result = run_sysctl_test(test_sysctl, 0); 3600 T_ASSERT_EQ_LLONG(result, 1, "sysctl result indicated success"); 3601} 3602 3603T_DECL(kernel_initiated, "smoke test that stackshot works with kernel-initiated stackshots", T_META_TAG_VM_PREFERRED) 3604{ 3605 const char *test_sysctl = "stackshot_kernel_initiator"; 3606 int64_t result; 3607 __block bool did_get_stackshot = false; 3608 3609 initialize_thread(); // must run before the stackshots to keep parse_stackshot happy 3610 3611 T_LOG("running sysctl to trigger kernel-driven stackshot type 1"); 3612 result = run_sysctl_test(test_sysctl, 1); 3613 T_ASSERT_EQ_LLONG(result, 1, "sysctl result indicated success"); 3614 3615 T_LOG("running sysctl to trigger kernel-driven stackshot type 2"); 3616 result = run_sysctl_test(test_sysctl, 2); 3617 T_ASSERT_EQ_LLONG(result, 1, "sysctl result indicated success"); 3618 3619 struct scenario scenario = { 3620 .name = "from_kernel_initiated", 3621 .flags = STACKSHOT_RETRIEVE_EXISTING_BUFFER, 3622 }; 3623 3624 T_LOG("attempting to fetch stored in-kernel stackshot"); 3625 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 3626 T_ASSERT_NOTNULL(ssbuf, "non-null kernel stackshot"); 3627 T_ASSERT_GT(sslen, 0, "non-zero stackshot size"); 3628 parse_stackshot(0, ssbuf, sslen, nil); 3629 did_get_stackshot = true; 3630 }); 3631 3632 T_ASSERT_TRUE(did_get_stackshot, "got stackshot from kernel type 2"); 3633} 3634