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, 1580 "tests translated bit is set correctly", 1581 T_META_TAG_VM_PREFERRED, 1582 T_META_ENABLED(false /* rdar://133956022 */)) 1583{ 1584#if !(TARGET_OS_OSX && TARGET_CPU_ARM64) 1585 T_SKIP("Only valid on Apple silicon Macs") 1586#endif 1587 // Get path of stackshot_translated_child helper binary 1588 char path[PATH_MAX]; 1589 uint32_t path_size = sizeof(path); 1590 T_QUIET; T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 1591 char* binary_name = strrchr(path, '/'); 1592 if (binary_name) binary_name++; 1593 T_QUIET; T_ASSERT_NOTNULL(binary_name, "Find basename in path '%s'", path); 1594 strlcpy(binary_name, "stackshot_translated_child", path_size - (binary_name - path)); 1595 char *args[] = { path, NULL }; 1596 1597 dispatch_source_t child_sig_src; 1598 dispatch_semaphore_t child_ready_sem = dispatch_semaphore_create(0); 1599 T_QUIET; T_ASSERT_NOTNULL(child_ready_sem, "exec child semaphore"); 1600 1601 dispatch_queue_t signal_processing_q = dispatch_queue_create("signal processing queue", NULL); 1602 T_QUIET; T_ASSERT_NOTNULL(signal_processing_q, "signal processing queue"); 1603 1604 signal(SIGUSR1, SIG_IGN); 1605 child_sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, signal_processing_q); 1606 T_QUIET; T_ASSERT_NOTNULL(child_sig_src, "dispatch_source_create (child_sig_src)"); 1607 1608 dispatch_source_set_event_handler(child_sig_src, ^{ dispatch_semaphore_signal(child_ready_sem); }); 1609 dispatch_activate(child_sig_src); 1610 1611 // Spawn child 1612 pid_t pid; 1613 T_LOG("spawning translated child"); 1614 T_QUIET; T_ASSERT_POSIX_ZERO(posix_spawn(&pid, args[0], NULL, NULL, args, NULL), "spawned process '%s' with PID %d", args[0], pid); 1615 1616 // Wait for the the child to spawn up 1617 dispatch_semaphore_wait(child_ready_sem, DISPATCH_TIME_FOREVER); 1618 1619 // Make sure the child is running and is translated 1620 int mib[] = { CTL_KERN, KERN_PROC, KERN_PROC_PID, pid }; 1621 struct kinfo_proc process_info; 1622 size_t bufsize = sizeof(process_info); 1623 T_QUIET; T_ASSERT_POSIX_SUCCESS(sysctl(mib, (unsigned)(sizeof(mib)/sizeof(int)), &process_info, &bufsize, NULL, 0), "get translated child process info"); 1624 T_QUIET; T_ASSERT_GT(bufsize, (size_t)0, "process info is not empty"); 1625 T_QUIET; T_ASSERT_TRUE((process_info.kp_proc.p_flag & P_TRANSLATED), "KERN_PROC_PID reports child is translated"); 1626 1627 T_LOG("capturing stackshot"); 1628 1629 struct scenario scenario = { 1630 .name = "translated", 1631 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 1632 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 1633 }; 1634 1635 take_stackshot(&scenario, true, ^( void *ssbuf, size_t sslen) { 1636 parse_stackshot(PARSE_STACKSHOT_TRANSLATED, ssbuf, sslen, @{translated_child_pid_key: @(pid)}); 1637 }); 1638 1639 // Kill the child 1640 int status; 1641 T_QUIET; T_ASSERT_POSIX_SUCCESS(kill(pid, SIGTERM), "kill translated child"); 1642 T_QUIET; T_ASSERT_POSIX_SUCCESS(waitpid(pid, &status, 0), "waitpid on translated child"); 1643 1644} 1645 1646T_DECL(proc_uuid_info, "tests that the main binary UUID for a proc is always populated", T_META_TAG_VM_PREFERRED) 1647{ 1648 struct proc_uniqidentifierinfo proc_info_data = { }; 1649 mach_msg_type_number_t count; 1650 kern_return_t kernel_status; 1651 task_dyld_info_data_t task_dyld_info; 1652 struct dyld_all_image_infos *target_infos; 1653 int retval; 1654 bool found_image_in_image_infos = false; 1655 uint64_t expected_mach_header_offset = 0; 1656 1657 /* Find the UUID of our main binary */ 1658 retval = proc_pidinfo(getpid(), PROC_PIDUNIQIDENTIFIERINFO, 0, &proc_info_data, sizeof(proc_info_data)); 1659 T_QUIET; T_EXPECT_POSIX_SUCCESS(retval, "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO"); 1660 T_QUIET; T_ASSERT_EQ_INT(retval, (int) sizeof(proc_info_data), "proc_pidinfo PROC_PIDUNIQIDENTIFIERINFO returned data"); 1661 1662 uuid_string_t str = {}; 1663 uuid_unparse(*(uuid_t*)&proc_info_data.p_uuid, str); 1664 T_LOG("Found current UUID is %s", str); 1665 1666 /* Find the location of the dyld image info metadata */ 1667 count = TASK_DYLD_INFO_COUNT; 1668 kernel_status = task_info(mach_task_self(), TASK_DYLD_INFO, (task_info_t)&task_dyld_info, &count); 1669 T_QUIET; T_ASSERT_EQ(kernel_status, KERN_SUCCESS, "retrieve task_info for TASK_DYLD_INFO"); 1670 1671 target_infos = (struct dyld_all_image_infos *)task_dyld_info.all_image_info_addr; 1672 1673 /* Find our binary in the dyld image info array */ 1674 for (int i = 0; i < (int) target_infos->uuidArrayCount; i++) { 1675 if (uuid_compare(target_infos->uuidArray[i].imageUUID, *(uuid_t*)&proc_info_data.p_uuid) == 0) { 1676 expected_mach_header_offset = (uint64_t) target_infos->uuidArray[i].imageLoadAddress; 1677 found_image_in_image_infos = true; 1678 } 1679 } 1680 1681 T_ASSERT_TRUE(found_image_in_image_infos, "found binary image in dyld image info list"); 1682 1683 /* Overwrite the dyld image info data so the kernel has to fallback to the UUID stored in the proc structure */ 1684 target_infos->uuidArrayCount = 0; 1685 1686 struct scenario scenario = { 1687 .name = "proc_uuid_info", 1688 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT), 1689 .target_pid = getpid(), 1690 }; 1691 1692 T_LOG("attempting to take stackshot for current PID"); 1693 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1694 stackshot_verify_current_proc_uuid_info(ssbuf, sslen, expected_mach_header_offset, &proc_info_data); 1695 }); 1696} 1697 1698T_DECL(cseg_waitinfo, "test that threads stuck in the compressor report correct waitinfo", T_META_TAG_VM_PREFERRED) 1699{ 1700 struct scenario scenario = { 1701 .name = "cseg_waitinfo", 1702 .quiet = false, 1703 .flags = (STACKSHOT_THREAD_WAITINFO | STACKSHOT_KCDATA_FORMAT), 1704 }; 1705 __block uint64_t thread_id = 0; 1706 1707 dispatch_queue_t dq = dispatch_queue_create("com.apple.stackshot.cseg_waitinfo", NULL); 1708 dispatch_semaphore_t child_ok = dispatch_semaphore_create(0); 1709 1710 dispatch_async(dq, ^{ 1711 pthread_threadid_np(NULL, &thread_id); 1712 dispatch_semaphore_signal(child_ok); 1713 int val = 1; 1714 T_ASSERT_POSIX_SUCCESS(sysctlbyname("kern.cseg_wedge_thread", NULL, NULL, &val, sizeof(val)), "wedge child thread"); 1715 }); 1716 1717 dispatch_semaphore_wait(child_ok, DISPATCH_TIME_FOREVER); 1718 sleep(1); 1719 1720 T_LOG("taking stackshot"); 1721 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1722 int val = 1; 1723 T_ASSERT_POSIX_SUCCESS(sysctlbyname("kern.cseg_unwedge_thread", NULL, NULL, &val, sizeof(val)), "unwedge child thread"); 1724 parse_stackshot(PARSE_STACKSHOT_WAITINFO_CSEG, ssbuf, sslen, @{cseg_expected_threadid_key: @(thread_id)}); 1725 }); 1726} 1727 1728static void 1729srp_send( 1730 mach_port_t send_port, 1731 mach_port_t reply_port, 1732 mach_port_t msg_port) 1733{ 1734 kern_return_t ret = 0; 1735 1736 struct test_msg { 1737 mach_msg_header_t header; 1738 mach_msg_body_t body; 1739 mach_msg_port_descriptor_t port_descriptor; 1740 }; 1741 struct test_msg send_msg = { 1742 .header = { 1743 .msgh_remote_port = send_port, 1744 .msgh_local_port = reply_port, 1745 .msgh_bits = MACH_MSGH_BITS_SET(MACH_MSG_TYPE_COPY_SEND, 1746 reply_port ? MACH_MSG_TYPE_MAKE_SEND_ONCE : 0, 1747 MACH_MSG_TYPE_MOVE_SEND, 1748 MACH_MSGH_BITS_COMPLEX), 1749 .msgh_id = 0x100, 1750 .msgh_size = sizeof(send_msg), 1751 }, 1752 .body = { 1753 .msgh_descriptor_count = 1, 1754 }, 1755 .port_descriptor = { 1756 .name = msg_port, 1757 .disposition = MACH_MSG_TYPE_MOVE_RECEIVE, 1758 .type = MACH_MSG_PORT_DESCRIPTOR, 1759 }, 1760 }; 1761 1762 if (msg_port == MACH_PORT_NULL) { 1763 send_msg.body.msgh_descriptor_count = 0; 1764 } 1765 1766 ret = mach_msg(&(send_msg.header), 1767 MACH_SEND_MSG | 1768 MACH_SEND_TIMEOUT | 1769 MACH_SEND_OVERRIDE, 1770 send_msg.header.msgh_size, 1771 0, 1772 MACH_PORT_NULL, 1773 10000, 1774 0); 1775 1776 T_ASSERT_MACH_SUCCESS(ret, "client mach_msg"); 1777} 1778 1779T_HELPER_DECL(srp_client, 1780 "Client used for the special_reply_port test") 1781{ 1782 pid_t ppid = getppid(); 1783 dispatch_semaphore_t can_continue = dispatch_semaphore_create(0); 1784 dispatch_queue_t dq = dispatch_queue_create("client_signalqueue", NULL); 1785 dispatch_source_t sig_src; 1786 1787 mach_msg_return_t mr; 1788 mach_port_t service_port; 1789 mach_port_t conn_port; 1790 mach_port_t special_reply_port; 1791 mach_port_options_t opts = { 1792 .flags = MPO_INSERT_SEND_RIGHT, 1793 }; 1794 1795 signal(SIGUSR1, SIG_IGN); 1796 sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, dq); 1797 1798 dispatch_source_set_event_handler(sig_src, ^{ 1799 dispatch_semaphore_signal(can_continue); 1800 }); 1801 dispatch_activate(sig_src); 1802 1803 /* lookup the mach service port for the parent */ 1804 kern_return_t kr = bootstrap_look_up(bootstrap_port, 1805 SRP_SERVICE_NAME, &service_port); 1806 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "client bootstrap_look_up"); 1807 1808 /* create the send-once right (special reply port) and message to send to the server */ 1809 kr = mach_port_construct(mach_task_self(), &opts, 0ull, &conn_port); 1810 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct"); 1811 1812 special_reply_port = thread_get_special_reply_port(); 1813 T_QUIET; T_ASSERT_TRUE(MACH_PORT_VALID(special_reply_port), "get_thread_special_reply_port"); 1814 1815 /* send the message with the special reply port */ 1816 srp_send(service_port, special_reply_port, conn_port); 1817 1818 /* signal the parent to continue */ 1819 kill(ppid, SIGUSR1); 1820 1821 struct { 1822 mach_msg_header_t header; 1823 mach_msg_body_t body; 1824 mach_msg_port_descriptor_t port_descriptor; 1825 } rcv_msg = { 1826 .header = 1827 { 1828 .msgh_remote_port = MACH_PORT_NULL, 1829 .msgh_local_port = special_reply_port, 1830 .msgh_size = sizeof(rcv_msg), 1831 }, 1832 }; 1833 1834 /* wait on the reply from the parent (that we will never receive) */ 1835 mr = mach_msg(&(rcv_msg.header), 1836 (MACH_RCV_MSG | MACH_RCV_SYNC_WAIT), 1837 0, 1838 rcv_msg.header.msgh_size, 1839 special_reply_port, 1840 MACH_MSG_TIMEOUT_NONE, 1841 service_port); 1842 1843 /* not expected to execute as parent will SIGKILL client... */ 1844 T_LOG("client process exiting after sending message to parent (server)"); 1845} 1846 1847enum srp_test_type { 1848 SRP_TEST_THREAD, /* expect waiter on current thread */ 1849 SRP_TEST_PID, /* expect waiter on current PID */ 1850 SRP_TEST_EITHER, /* waiter could be on either */ 1851}; 1852 1853static void 1854check_srp_test(const char *name, enum srp_test_type ty) 1855{ 1856 struct scenario scenario = { 1857 .name = name, 1858 .quiet = false, 1859 .flags = (STACKSHOT_THREAD_WAITINFO | STACKSHOT_KCDATA_FORMAT), 1860 }; 1861 uint64_t thread_id = 0; 1862 pthread_threadid_np(NULL, &thread_id); 1863 if (ty == SRP_TEST_THREAD) { 1864 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1865 parse_stackshot(PARSE_STACKSHOT_WAITINFO_SRP, ssbuf, sslen, 1866 @{srp_expected_threadid_key: @(thread_id)}); 1867 }); 1868 } else if (ty == SRP_TEST_PID) { 1869 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1870 parse_stackshot(PARSE_STACKSHOT_WAITINFO_SRP, ssbuf, sslen, 1871 @{srp_expected_pid_key: @(getpid())}); 1872 }); 1873 } else { 1874 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 1875 parse_stackshot(PARSE_STACKSHOT_WAITINFO_SRP, ssbuf, sslen, 1876 @{srp_expected_pid_key: @(getpid()), srp_expected_threadid_key: @(thread_id)}); 1877 }); 1878 } 1879 1880} 1881 1882 1883/* 1884 * Tests the stackshot wait info plumbing for synchronous IPC that doesn't use kevent on the server. 1885 * 1886 * (part 1): tests the scenario where a client sends a request that includes a special reply port 1887 * to a server that doesn't receive the message and doesn't copy the send-once right 1888 * into its address space as a result. for this case the special reply port is enqueued 1889 * in a port and we check which task has that receive right and use that info. (rdar://60440338) 1890 * (part 2): tests the scenario where a client sends a request that includes a special reply port 1891 * to a server that receives the message and copies in the send-once right, but doesn't 1892 * reply to the client. for this case the special reply port is copied out and the kernel 1893 * stashes the info about which task copied out the send once right. (rdar://60440592) 1894 * (part 3): tests the same as part 2, but uses kevents, which allow for 1895 * priority inheritance 1896 */ 1897T_DECL(special_reply_port, "test that tasks using special reply ports have correct waitinfo", T_META_TAG_VM_PREFERRED) 1898{ 1899 dispatch_semaphore_t can_continue = dispatch_semaphore_create(0); 1900 dispatch_queue_t dq = dispatch_queue_create("signalqueue", NULL); 1901 dispatch_queue_t machdq = dispatch_queue_create("machqueue", NULL); 1902 dispatch_source_t sig_src; 1903 char path[PATH_MAX]; 1904 uint32_t path_size = sizeof(path); 1905 T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 1906 char *client_args[] = { path, "-n", "srp_client", NULL }; 1907 pid_t client_pid; 1908 int sp_ret; 1909 kern_return_t kr; 1910 mach_port_t port; 1911 1912 /* setup the signal handler in the parent (server) */ 1913 T_LOG("setup sig handlers"); 1914 signal(SIGUSR1, SIG_IGN); 1915 sig_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_SIGNAL, SIGUSR1, 0, dq); 1916 1917 dispatch_source_set_event_handler(sig_src, ^{ 1918 dispatch_semaphore_signal(can_continue); 1919 }); 1920 dispatch_activate(sig_src); 1921 1922 /* register with the mach service name so the client can lookup and send a message to the parent (server) */ 1923 T_LOG("Server about to check in"); 1924 kr = bootstrap_check_in(bootstrap_port, SRP_SERVICE_NAME, &port); 1925 T_ASSERT_MACH_SUCCESS(kr, "server bootstrap_check_in"); 1926 1927 T_LOG("Launching client"); 1928 sp_ret = posix_spawn(&client_pid, client_args[0], NULL, NULL, client_args, NULL); 1929 T_QUIET; T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", client_args[0], client_pid); 1930 T_LOG("Spawned client as PID %d", client_pid); 1931 1932 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 1933 T_LOG("Ready to take stackshot, but waiting 1s for the coast to clear"); 1934 1935 /* 1936 * can_continue indicates the client has signaled us, but we want to make 1937 * sure they've actually blocked sending their mach message. It's cheesy, but 1938 * sleep() works for this. 1939 */ 1940 sleep(1); 1941 1942 /* 1943 * take the stackshot without calling receive to verify that the stackshot wait 1944 * info shows our (the server) thread for the scenario where the server has yet to 1945 * receive the message. 1946 */ 1947 T_LOG("Taking stackshot for part 1 coverage"); 1948 check_srp_test("srp", SRP_TEST_THREAD); 1949 1950 /* 1951 * receive the message from the client (which should copy the send once right into 1952 * our address space). 1953 */ 1954 struct { 1955 mach_msg_header_t header; 1956 mach_msg_body_t body; 1957 mach_msg_port_descriptor_t port_descriptor; 1958 } rcv_msg = { 1959 .header = 1960 { 1961 .msgh_remote_port = MACH_PORT_NULL, 1962 .msgh_local_port = port, 1963 .msgh_size = sizeof(rcv_msg), 1964 }, 1965 }; 1966 1967 T_LOG("server: starting sync receive\n"); 1968 1969 mach_msg_return_t mr; 1970 mr = mach_msg(&(rcv_msg.header), 1971 (MACH_RCV_MSG | MACH_RCV_TIMEOUT), 1972 0, 1973 4096, 1974 port, 1975 10000, 1976 MACH_PORT_NULL); 1977 T_QUIET; T_ASSERT_MACH_SUCCESS(mr, "mach_msg() recieve of message from client"); 1978 1979 /* 1980 * take the stackshot to verify that the stackshot wait info shows our (the server) PID 1981 * for the scenario where the server has received the message and copied in the send-once right. 1982 */ 1983 T_LOG("Taking stackshot for part 2 coverage"); 1984 check_srp_test("srp", SRP_TEST_PID); 1985 1986 /* cleanup - kill the client */ 1987 T_ASSERT_POSIX_SUCCESS(kill(client_pid, SIGKILL), "killing client"); 1988 T_ASSERT_POSIX_SUCCESS(waitpid(client_pid, NULL, 0), "waiting for the client to exit"); 1989 1990 // do it again, but using kevents 1991 T_LOG("Launching client"); 1992 sp_ret = posix_spawn(&client_pid, client_args[0], NULL, NULL, client_args, NULL); 1993 T_QUIET; T_ASSERT_POSIX_ZERO(sp_ret, "spawned process '%s' with PID %d", client_args[0], client_pid); 1994 T_LOG("Spawned client as PID %d", client_pid); 1995 1996 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 1997 T_LOG("Ready to take stackshot, but waiting 1s for the coast to clear"); 1998 1999 /* 2000 * can_continue indicates the client has signaled us, but we want to make 2001 * sure they've actually blocked sending their mach message. It's cheesy, but 2002 * sleep() works for this. 2003 */ 2004 sleep(1); 2005 2006 dispatch_mach_t dispatch_mach = dispatch_mach_create(SRP_SERVICE_NAME, machdq, 2007 ^(dispatch_mach_reason_t reason, 2008 dispatch_mach_msg_t message, 2009 mach_error_t error __unused) { 2010 switch (reason) { 2011 case DISPATCH_MACH_MESSAGE_RECEIVED: { 2012 size_t size = 0; 2013 mach_msg_header_t *msg __unused = dispatch_mach_msg_get_msg(message, &size); 2014 T_LOG("server: recieved %ld byte message", size); 2015 check_srp_test("turnstile_port_thread", SRP_TEST_THREAD); 2016 T_LOG("server: letting client go"); 2017 // drop the message on the ground, we'll kill the client later 2018 dispatch_semaphore_signal(can_continue); 2019 break; 2020 } 2021 default: 2022 break; 2023 } 2024 }); 2025 2026 dispatch_mach_connect(dispatch_mach, port, MACH_PORT_NULL, NULL); 2027 2028 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 2029 2030 /* cleanup - kill the client */ 2031 T_ASSERT_POSIX_SUCCESS(kill(client_pid, SIGKILL), "killing client"); 2032 T_ASSERT_POSIX_SUCCESS(waitpid(client_pid, NULL, 0), "waiting for the client to exit"); 2033} 2034 2035T_HELPER_DECL(throtlled_sp_client, 2036 "client that uses a connection port to send a message to a server") 2037{ 2038 mach_port_t conn_port, service_port, reply_port, *stash; 2039 mach_msg_type_number_t stash_cnt = 0; 2040 2041 kern_return_t kr = mach_ports_lookup(mach_task_self(), &stash, &stash_cnt); 2042 T_ASSERT_MACH_SUCCESS(kr, "mach_ports_lookup"); 2043 2044 service_port = stash[0]; 2045 T_ASSERT_TRUE(MACH_PORT_VALID(service_port), "valid service port"); 2046 mig_deallocate((vm_address_t)stash, stash_cnt * sizeof(stash[0])); 2047 2048 mach_port_options_t opts = { 2049 .flags = MPO_INSERT_SEND_RIGHT 2050 | MPO_CONNECTION_PORT, 2051 .service_port_name = service_port, 2052 }; 2053 2054 kr = mach_port_construct(mach_task_self(), &opts, 0ull, &conn_port); 2055 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct"); 2056 2057 mach_port_options_t opts2 = { 2058 .flags = MPO_REPLY_PORT 2059 }; 2060 kr = mach_port_construct(mach_task_self(), &opts2, 0ull, &reply_port); 2061 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct"); 2062 2063 /* XPC-like check-in message */ 2064 struct { 2065 mach_msg_header_t header; 2066 mach_msg_port_descriptor_t recvp; 2067 mach_msg_port_descriptor_t sendp; 2068 } checkin_message = { 2069 .header = 2070 { 2071 .msgh_remote_port = service_port, 2072 .msgh_local_port = MACH_PORT_NULL, 2073 .msgh_size = sizeof(checkin_message), 2074 .msgh_bits = MACH_MSGH_BITS(MACH_MSG_TYPE_COPY_SEND, 0), 2075 }, 2076 .recvp = 2077 { 2078 .type = MACH_MSG_PORT_DESCRIPTOR, 2079 .name = conn_port, 2080 .disposition = MACH_MSG_TYPE_MOVE_RECEIVE, 2081 }, 2082 .sendp = 2083 { 2084 .type = MACH_MSG_PORT_DESCRIPTOR, 2085 .name = reply_port, 2086 .disposition = MACH_MSG_TYPE_MAKE_SEND, 2087 } 2088 }; 2089 dispatch_mach_msg_t dmsg = dispatch_mach_msg_create((mach_msg_header_t *)&checkin_message, sizeof(checkin_message), 2090 DISPATCH_MACH_MSG_DESTRUCTOR_DEFAULT, NULL); 2091 2092 dispatch_queue_t machdq = dispatch_queue_create("machqueue", NULL); 2093 dispatch_mach_t dchannel = dispatch_mach_create(THROTTLED_SERVICE_NAME, machdq, 2094 ^(dispatch_mach_reason_t reason, 2095 dispatch_mach_msg_t message __unused, 2096 mach_error_t error __unused) { 2097 switch (reason) { 2098 case DISPATCH_MACH_CONNECTED: 2099 T_LOG("mach channel connected"); 2100 break; 2101 case DISPATCH_MACH_MESSAGE_SENT: 2102 T_LOG("sent mach message"); 2103 break; 2104 default: 2105 T_ASSERT_FAIL("Unexpected reply to channel reason %lu", reason); 2106 } 2107 }); 2108 dispatch_mach_connect(dchannel, reply_port, service_port, dmsg); 2109 dispatch_release(dmsg); 2110 2111 struct { 2112 mach_msg_header_t header; 2113 uint64_t request_id; 2114 } request = { 2115 .header = 2116 { 2117 .msgh_size = sizeof(request), 2118 .msgh_bits = MACH_MSGH_BITS(MACH_MSG_TYPE_COPY_SEND, MACH_MSG_TYPE_MAKE_SEND_ONCE), 2119 }, 2120 .request_id = 1, 2121 }; 2122 dispatch_mach_msg_t dmsg2 = dispatch_mach_msg_create((mach_msg_header_t *)&request, sizeof(request), 2123 DISPATCH_MACH_MSG_DESTRUCTOR_DEFAULT, NULL); 2124 2125 dispatch_mach_reason_t reason; 2126 mach_error_t error; 2127 2128 /* send the check-in message and the request message */ 2129 dispatch_mach_msg_t dreply = dispatch_mach_send_with_result_and_wait_for_reply(dchannel, 2130 dmsg2, 0, DISPATCH_MACH_SEND_DEFAULT, &reason, &error); 2131 dispatch_release(dmsg2); 2132 2133 /* not expected to execute as parent will SIGKILL client */ 2134 T_ASSERT_FAIL("client process exiting after receiving %s reply", dreply ? "non-null" : "null"); 2135} 2136 2137static void 2138check_throttled_sp(const char *test_name, uint64_t context, bool is_throttled) 2139{ 2140 struct scenario scenario = { 2141 .name = test_name, 2142 .quiet = false, 2143 .flags = (STACKSHOT_THREAD_WAITINFO | STACKSHOT_KCDATA_FORMAT), 2144 }; 2145 2146 T_LOG("taking stackshot %s", test_name); 2147 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2148 parse_stackshot(PARSE_STACKSHOT_THROTTLED_SP, ssbuf, sslen, 2149 @{sp_throttled_expected_ctxt_key: @(context), 2150 sp_throttled_expect_flag: @(is_throttled)}); 2151 }); 2152} 2153 2154/* Take stackshot when a client is blocked on the service port of a process, in the scenario when 2155 * the process with the receive right for the service port is: 2156 * (a) Monitoring the service port using kevents 2157 * (b) Not monitoring the service port 2158 */ 2159T_DECL(throttled_sp, 2160 "test that service port throttled flag is propagated to the stackshot correctly", T_META_TAG_VM_PREFERRED) 2161{ 2162 mach_port_t service_port; 2163 __block dispatch_semaphore_t can_continue = dispatch_semaphore_create(0); 2164 2165 char path[PATH_MAX]; 2166 uint32_t path_size = sizeof(path); 2167 T_ASSERT_POSIX_ZERO(_NSGetExecutablePath(path, &path_size), "_NSGetExecutablePath"); 2168 char *client_args[] = { path, "-n", "throtlled_sp_client", NULL }; 2169 2170 __block uint64_t thread_id = 0; 2171 pid_t client_pid; 2172 int mark_throttled; 2173 2174 struct mach_service_port_info sp_info = {}; 2175 strcpy(sp_info.mspi_string_name, THROTTLED_SERVICE_NAME); 2176 sp_info.mspi_domain_type = (uint8_t)1; 2177 kern_return_t kr; 2178 2179 mach_port_options_t opts = { 2180 .flags = MPO_SERVICE_PORT | MPO_INSERT_SEND_RIGHT | MPO_CONTEXT_AS_GUARD | MPO_STRICT | MPO_TEMPOWNER, 2181 .service_port_info = &sp_info, 2182 }; 2183 2184 kr = mach_port_construct(mach_task_self(), &opts, 0ull, &service_port); 2185 T_ASSERT_MACH_SUCCESS(kr, "mach_port_construct %u", service_port); 2186 2187 /* Setup a dispatch source to monitor the service port similar to how launchd does. */ 2188 dispatch_queue_t machdq = dispatch_queue_create("machqueue", NULL); 2189 dispatch_source_t mach_src = dispatch_source_create(DISPATCH_SOURCE_TYPE_MACH_RECV, service_port, 2190 DISPATCH_MACH_RECV_SYNC_PEEK, machdq); 2191 dispatch_source_set_event_handler(mach_src, ^{ 2192 pthread_threadid_np(NULL, &thread_id); 2193 dispatch_semaphore_signal(can_continue); 2194 }); 2195 dispatch_activate(mach_src); 2196 2197 /* Stash the port in task to make sure child also gets it */ 2198 kr = mach_ports_register(mach_task_self(), &service_port, 1); 2199 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mach_ports_register service port"); 2200 2201 mark_throttled = 1; 2202 kr = mach_port_set_attributes(mach_task_self(), service_port, MACH_PORT_SERVICE_THROTTLED, (mach_port_info_t)(&mark_throttled), 2203 MACH_PORT_SERVICE_THROTTLED_COUNT); 2204 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "mark service port as throttled"); 2205 2206 int rc = posix_spawn(&client_pid, client_args[0], NULL, NULL, client_args, NULL); 2207 T_QUIET; T_ASSERT_POSIX_ZERO(rc, "spawned process '%s' with PID %d", client_args[0], client_pid); 2208 T_LOG("Spawned client as PID %d", client_pid); 2209 2210 dispatch_semaphore_wait(can_continue, DISPATCH_TIME_FOREVER); 2211 2212 /* The service port has received the check-in message. Take stackshot for scenario (a). */ 2213 check_throttled_sp("throttled_service_port_monitored", thread_id, true); 2214 2215 /* This simulates a throttled spawn when the service port is no longer monitored. */ 2216 dispatch_source_cancel(mach_src); 2217 2218 /* Take stackshot for scenario (b) */ 2219 check_throttled_sp("throttled_service_port_unmonitored", (uint64_t)getpid(), true); 2220 2221 mark_throttled = 0; 2222 kr = mach_port_set_attributes(mach_task_self(), service_port, MACH_PORT_SERVICE_THROTTLED, (mach_port_info_t)(&mark_throttled), 2223 MACH_PORT_SERVICE_THROTTLED_COUNT); 2224 T_QUIET; T_ASSERT_MACH_SUCCESS(kr, "unmark service port as throttled"); 2225 2226 /* Throttled flag should not be set when the port is not throttled. */ 2227 check_throttled_sp("unthrottled_service_port_unmonitored", (uint64_t)getpid(), false); 2228 2229 /* cleanup - kill the client */ 2230 T_ASSERT_POSIX_SUCCESS(kill(client_pid, SIGKILL), "killing client"); 2231 T_ASSERT_POSIX_SUCCESS(waitpid(client_pid, NULL, 0), "waiting for the client to exit"); 2232} 2233 2234 2235char *const clpcctrl_path = "/usr/local/bin/clpcctrl"; 2236 2237static void 2238run_clpcctrl(char *const argv[]) { 2239 posix_spawnattr_t sattr; 2240 pid_t pid; 2241 int wstatus; 2242 2243 T_QUIET; T_ASSERT_POSIX_ZERO(posix_spawn(&pid, argv[0], NULL, NULL, argv, NULL), "spawn clpcctrl"); 2244 T_QUIET; T_ASSERT_POSIX_SUCCESS(waitpid(pid, &wstatus, 0), "wait for clpcctrl"); 2245 T_QUIET; T_ASSERT_TRUE(WIFEXITED(wstatus), "clpcctrl exited normally"); 2246 T_QUIET; T_ASSERT_POSIX_ZERO(WEXITSTATUS(wstatus), "clpcctrl exited successfully"); 2247 2248 uint64_t sched_recommended_cores = 1; 2249 size_t sched_recommended_cores_sz = sizeof(uint64_t); 2250 T_QUIET; T_ASSERT_POSIX_SUCCESS( 2251 sysctlbyname("kern.sched_recommended_cores", &sched_recommended_cores, &sched_recommended_cores_sz, NULL, 0), 2252 "get kern.sched_recommended_cores"); 2253 T_LOG("Recommended cores: 0x%llx", sched_recommended_cores); 2254} 2255 2256static void 2257restore_clpcctrl() { 2258 run_clpcctrl((char *const []) { clpcctrl_path, "-d", NULL }); 2259} 2260 2261#define CLUSTER_TYPE_SMP 0 2262#define CLUSTER_TYPE_E 1 2263#define CLUSTER_TYPE_P 2 2264 2265void test_stackshot_cpu_info(void *ssbuf, size_t sslen, int exp_cpus, NSArray *exp_cluster_types) { 2266 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2267 bool seen = false; 2268 int singlethread_override = 0; 2269 size_t singlethread_override_sz = sizeof(singlethread_override); 2270 T_QUIET; T_ASSERT_POSIX_SUCCESS( 2271 sysctlbyname("kern.stackshot_single_thread", &singlethread_override, &singlethread_override_sz, NULL, 0), 2272 "get kern.stackshot_single_thread"); 2273 if (singlethread_override) { 2274 T_LOG("skipping cpu count/type check due to single-thread override (kern.stackshot_single_thread=1)"); 2275 return; 2276 } 2277 2278 KCDATA_ITER_FOREACH(iter) { 2279 if ((kcdata_iter_type(iter) != KCDATA_TYPE_ARRAY) || (kcdata_iter_array_elem_type(iter) != STACKSHOT_KCTYPE_LATENCY_INFO_CPU)) { 2280 continue; 2281 } 2282 2283 seen = true; 2284 2285 /* Check ncpus */ 2286 int ncpus = kcdata_iter_array_elem_count(iter); 2287 if (exp_cpus != -1) { 2288 T_QUIET; T_ASSERT_EQ(exp_cpus, ncpus, "Expected number of CPUs matches number of CPUs used for stackshot"); 2289 } 2290 2291 if (exp_cluster_types == nil) { 2292 continue; 2293 } 2294 2295 /* Check cluster types */ 2296 struct stackshot_latency_cpu *latencies = (struct stackshot_latency_cpu *) kcdata_iter_payload(iter); 2297 for (int i = 0; i < ncpus; i++) { 2298 NSNumber *cluster_type = [NSNumber numberWithInt:latencies[i].cluster_type]; 2299 T_QUIET; T_ASSERT_TRUE([exp_cluster_types containsObject:cluster_type], "Type of CPU cluster in expected CPU cluster types"); 2300 } 2301 } 2302 2303 T_QUIET; T_ASSERT_TRUE(seen || !is_development_kernel(), "Seen CPU latency info or is release kernel"); 2304} 2305 2306void test_stackshot_with_clpcctrl(char *const name, char *const argv[], int exp_cpus, NSArray *exp_cluster_types) { 2307 T_LOG("Stackshot CLPC scenario %s", name); 2308 run_clpcctrl(argv); 2309 struct scenario scenario = { 2310 .name = name, 2311 .flags = (STACKSHOT_KCDATA_FORMAT | STACKSHOT_SAVE_LOADINFO | 2312 STACKSHOT_THREAD_WAITINFO | STACKSHOT_GET_GLOBAL_MEM_STATS) 2313 }; 2314 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2315 parse_stackshot(0, ssbuf, sslen, nil); 2316 test_stackshot_cpu_info(ssbuf, sslen, exp_cpus, exp_cluster_types); 2317 }); 2318} 2319 2320T_DECL(core_masks, 2321 "test that stackshot works under various core masks on ARM systems", 2322 T_META_REQUIRES_SYSCTL_EQ("hw.optional.arm64", 1), 2323 T_META_REQUIRES_SYSCTL_NE("kern.kasan.available", 1), /* rdar://115577993 */ 2324 XNU_T_META_REQUIRES_DEVELOPMENT_KERNEL, 2325 T_META_REQUIRE_NOT_VIRTUALIZED, 2326 T_META_RUN_CONCURRENTLY(false), 2327 T_META_TAG_VM_NOT_ELIGIBLE, 2328 T_META_ENABLED(!TARGET_OS_VISION)) // disable for visionOS: device may not be stable with many cores masked off (127904530) 2329{ 2330 /* 2331 * Make sure we're not in a release kernel 2332 * (cannot check with T_META; only one sysctl T_META at a time will work) 2333 */ 2334 if (!is_development_kernel()) { 2335 T_SKIP("test was not run because kernel is release; cannot set core masks"); 2336 return; 2337 } 2338 2339 /* 2340 * rdar://115577993 - CLPC compiles as release in KASAN-variant builds, 2341 * preventing clpcctrl from working. For now, skip this. (Cannot check 2342 * with T_META; only one sysctl T_META at a time will work) 2343 */ 2344 int kasan_avail = 0; 2345 size_t kasan_avail_sz = sizeof(kasan_avail); 2346 sysctlbyname("kern.kasan.available", &kasan_avail, &kasan_avail_sz, NULL, 0); 2347 if (kasan_avail) { 2348 T_SKIP("test was not run because kernel is KASAN; cannot set core masks (see rdar://115577993)"); 2349 return; 2350 } 2351 2352 2353 T_ATEND(restore_clpcctrl); 2354 2355 /* Test with 1 and 2 CPUs for basic functionality */ 2356 test_stackshot_with_clpcctrl( 2357 "core_masks_1cpu", (char *const[]) {clpcctrl_path, "-c", "1", NULL}, 2358 1, nil); 2359 2360 test_stackshot_with_clpcctrl( 2361 "core_masks_2cpus", (char *const[]) {clpcctrl_path, "-c", "2", NULL}, 2362 2, nil); 2363 2364 /* Check nperflevels to see if we're on an AMP system */ 2365 int nperflevels = 1; 2366 size_t nperflevels_sz = sizeof(int); 2367 T_ASSERT_POSIX_SUCCESS( 2368 sysctlbyname("hw.nperflevels", &nperflevels, &nperflevels_sz, NULL, 0), 2369 "get hw.nperflevels"); 2370 if (nperflevels == 1) { 2371 T_LOG("On SMP system, skipping stackshot core_masks AMP tests"); 2372 return; 2373 } 2374 2375 T_QUIET; T_ASSERT_EQ(nperflevels, 2, "nperflevels is 1 or 2"); 2376 T_LOG("On AMP system, performing stackshot core_masks AMP tests"); 2377 2378 /* Perform AMP tests with different cluster types active */ 2379 test_stackshot_with_clpcctrl( 2380 "core_masks_amp_allcpus", 2381 (char *const[]) {clpcctrl_path, "-C", "all", NULL}, 2382 -1, @[@CLUSTER_TYPE_E, @CLUSTER_TYPE_P]); 2383 2384 test_stackshot_with_clpcctrl( 2385 "core_masks_amp_ecpus", 2386 (char *const[]) {clpcctrl_path, "-C", "e", NULL}, 2387 -1, @[@CLUSTER_TYPE_E]); 2388 2389 test_stackshot_with_clpcctrl( 2390 "core_masks_amp_pcpus", 2391 (char *const[]) {clpcctrl_path, "-C", "p", NULL}, 2392 -1, @[@CLUSTER_TYPE_P]); 2393} 2394 2395#pragma mark performance tests 2396 2397#define SHOULD_REUSE_SIZE_HINT 0x01 2398#define SHOULD_USE_DELTA 0x02 2399#define SHOULD_TARGET_SELF 0x04 2400 2401static void 2402stackshot_perf(unsigned int options) 2403{ 2404 struct scenario scenario = { 2405 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS 2406 | STACKSHOT_SAVE_IMP_DONATION_PIDS | STACKSHOT_KCDATA_FORMAT), 2407 }; 2408 2409 dt_stat_t size = dt_stat_create("bytes", "size"); 2410 dt_stat_time_t duration = dt_stat_time_create("duration"); 2411 scenario.timer = duration; 2412 2413 if (options & SHOULD_TARGET_SELF) { 2414 scenario.target_pid = getpid(); 2415 } 2416 2417 while (!dt_stat_stable(duration) || !dt_stat_stable(size)) { 2418 __block uint64_t last_time = 0; 2419 __block uint32_t size_hint = 0; 2420 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2421 dt_stat_add(size, (double)sslen); 2422 last_time = stackshot_timestamp(ssbuf, sslen); 2423 size_hint = (uint32_t)sslen; 2424 }); 2425 if (options & SHOULD_USE_DELTA) { 2426 scenario.since_timestamp = last_time; 2427 scenario.flags |= STACKSHOT_COLLECT_DELTA_SNAPSHOT; 2428 } 2429 if (options & SHOULD_REUSE_SIZE_HINT) { 2430 scenario.size_hint = size_hint; 2431 } 2432 } 2433 2434 dt_stat_finalize(duration); 2435 dt_stat_finalize(size); 2436} 2437 2438static void 2439stackshot_flag_perf_noclobber(uint64_t flag, char *flagname) 2440{ 2441 struct scenario scenario = { 2442 .quiet = true, 2443 .flags = (flag | STACKSHOT_KCDATA_FORMAT), 2444 }; 2445 2446 dt_stat_t duration = dt_stat_create("nanoseconds per thread", "%s_duration", flagname); 2447 dt_stat_t size = dt_stat_create("bytes per thread", "%s_size", flagname); 2448 T_LOG("Testing \"%s\" = 0x%" PRIx64, flagname, flag); 2449 2450 while (!dt_stat_stable(duration) || !dt_stat_stable(size)) { 2451 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2452 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2453 unsigned long no_threads = 0; 2454 mach_timebase_info_data_t timebase = {0, 0}; 2455 uint64_t stackshot_duration = 0; 2456 int found = 0; 2457 T_QUIET; T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, "stackshot buffer"); 2458 2459 KCDATA_ITER_FOREACH(iter) { 2460 switch(kcdata_iter_type(iter)) { 2461 case STACKSHOT_KCTYPE_THREAD_SNAPSHOT: { 2462 found |= 1; 2463 no_threads ++; 2464 break; 2465 } 2466 case STACKSHOT_KCTYPE_STACKSHOT_DURATION: { 2467 struct stackshot_duration *ssd = kcdata_iter_payload(iter); 2468 stackshot_duration = ssd->stackshot_duration; 2469 found |= 2; 2470 break; 2471 } 2472 case KCDATA_TYPE_TIMEBASE: { 2473 found |= 4; 2474 mach_timebase_info_data_t *tb = kcdata_iter_payload(iter); 2475 memcpy(&timebase, tb, sizeof(timebase)); 2476 break; 2477 } 2478 } 2479 } 2480 2481 T_QUIET; T_ASSERT_EQ(found, 0x7, "found everything needed"); 2482 2483 uint64_t ns = (stackshot_duration * timebase.numer) / timebase.denom; 2484 uint64_t per_thread_ns = ns / no_threads; 2485 uint64_t per_thread_size = sslen / no_threads; 2486 2487 dt_stat_add(duration, per_thread_ns); 2488 dt_stat_add(size, per_thread_size); 2489 }); 2490 } 2491 2492 dt_stat_finalize(duration); 2493 dt_stat_finalize(size); 2494} 2495 2496static void 2497stackshot_flag_perf(uint64_t flag, char *flagname) 2498{ 2499 /* 2500 * STACKSHOT_NO_IO_STATS disables data collection, so set it for 2501 * more accurate perfdata collection. 2502 */ 2503 flag |= STACKSHOT_NO_IO_STATS; 2504 2505 stackshot_flag_perf_noclobber(flag, flagname); 2506} 2507 2508 2509T_DECL(flag_perf, "test stackshot performance with different flags set", T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2510{ 2511 stackshot_flag_perf_noclobber(STACKSHOT_NO_IO_STATS, "baseline"); 2512 stackshot_flag_perf_noclobber(0, "io_stats"); 2513 2514 stackshot_flag_perf(STACKSHOT_THREAD_WAITINFO, "thread_waitinfo"); 2515 stackshot_flag_perf(STACKSHOT_GET_DQ, "get_dq"); 2516 stackshot_flag_perf(STACKSHOT_SAVE_LOADINFO, "save_loadinfo"); 2517 stackshot_flag_perf(STACKSHOT_GET_GLOBAL_MEM_STATS, "get_global_mem_stats"); 2518 stackshot_flag_perf(STACKSHOT_SAVE_KEXT_LOADINFO, "save_kext_loadinfo"); 2519 stackshot_flag_perf(STACKSHOT_SAVE_IMP_DONATION_PIDS, "save_imp_donation_pids"); 2520 stackshot_flag_perf(STACKSHOT_ENABLE_BT_FAULTING, "enable_bt_faulting"); 2521 stackshot_flag_perf(STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT, "collect_sharedcache_layout"); 2522 stackshot_flag_perf(STACKSHOT_ENABLE_UUID_FAULTING, "enable_uuid_faulting"); 2523 stackshot_flag_perf(STACKSHOT_THREAD_GROUP, "thread_group"); 2524 stackshot_flag_perf(STACKSHOT_SAVE_JETSAM_COALITIONS, "save_jetsam_coalitions"); 2525 stackshot_flag_perf(STACKSHOT_INSTRS_CYCLES, "instrs_cycles"); 2526 stackshot_flag_perf(STACKSHOT_ASID, "asid"); 2527 stackshot_flag_perf(STACKSHOT_EXCLAVES, "all_exclaves"); 2528 stackshot_flag_perf(STACKSHOT_EXCLAVES | STACKSHOT_ASID, "all_exclaves_and_asid"); 2529 stackshot_flag_perf(STACKSHOT_SKIP_EXCLAVES, "skip_exclaves"); 2530} 2531 2532T_DECL(perf_no_size_hint, "test stackshot performance with no size hint", 2533 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2534{ 2535 stackshot_perf(0); 2536} 2537 2538T_DECL(perf_size_hint, "test stackshot performance with size hint", 2539 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2540{ 2541 stackshot_perf(SHOULD_REUSE_SIZE_HINT); 2542} 2543 2544T_DECL(perf_process, "test stackshot performance targeted at process", 2545 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2546{ 2547 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_TARGET_SELF); 2548} 2549 2550T_DECL(perf_delta, "test delta stackshot performance", 2551 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2552{ 2553 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_USE_DELTA); 2554} 2555 2556T_DECL(perf_delta_no_exclaves, "test delta stackshot performance without Exclaves", 2557 T_META_REQUIRES_SYSCTL_EQ("kern.exclaves_status", 1), 2558 T_META_REQUIRES_SYSCTL_EQ("kern.exclaves_inspection_status", 1), 2559 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2560{ 2561 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_USE_DELTA | STACKSHOT_SKIP_EXCLAVES); 2562} 2563 2564T_DECL(perf_delta_process, "test delta stackshot performance targeted at a process", 2565 T_META_TAG_PERF, T_META_TAG_VM_NOT_ELIGIBLE) 2566{ 2567 stackshot_perf(SHOULD_REUSE_SIZE_HINT | SHOULD_USE_DELTA | SHOULD_TARGET_SELF); 2568} 2569 2570T_DECL(stackshot_entitlement_report_test, "test stackshot entitlement report", T_META_TAG_VM_PREFERRED) 2571{ 2572 int sysctlValue = 1; 2573 T_ASSERT_POSIX_SUCCESS( 2574 sysctlbyname("debug.stackshot_entitlement_send_batch", NULL, NULL, &sysctlValue, sizeof(sysctlValue)), 2575 "set debug.stackshot_entitlement_send_batch=1"); 2576 // having a way to verify that the coreanalytics event was received would be even better 2577 // See rdar://74197197 2578 T_PASS("entitlement test ran"); 2579} 2580 2581static void 2582expect_os_build_version_in_stackshot(void *ssbuf, size_t sslen) 2583{ 2584 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2585 2586 bool saw_os_build_version = false; 2587 iter = kcdata_iter_next(iter); 2588 2589 KCDATA_ITER_FOREACH(iter) { 2590 switch (kcdata_iter_type(iter)) { 2591 case STACKSHOT_KCTYPE_OS_BUILD_VERSION: 2592 saw_os_build_version = true; 2593 T_LOG("Found os build version in stackshot: %s", kcdata_iter_payload(iter)); 2594 return; 2595 2596 default: 2597 break; 2598 } 2599 } 2600 2601 T_ASSERT_FAIL("didn't see os build version in stackshot"); 2602} 2603 2604T_DECL(os_build_version, "test stackshot contains os build version", T_META_TAG_VM_PREFERRED) 2605{ 2606 2607 struct scenario scenario = { 2608 .name = "os-build-version", 2609 .flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT), 2610 }; 2611 2612 T_LOG("attempting to take stackshot with an os build version"); 2613 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 2614 expect_os_build_version_in_stackshot(ssbuf, sslen); 2615 }); 2616} 2617 2618static uint64_t 2619stackshot_timestamp(void *ssbuf, size_t sslen) 2620{ 2621 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2622 2623 uint32_t type = kcdata_iter_type(iter); 2624 if (type != KCDATA_BUFFER_BEGIN_STACKSHOT && type != KCDATA_BUFFER_BEGIN_DELTA_STACKSHOT) { 2625 T_ASSERT_FAIL("invalid kcdata type %u", kcdata_iter_type(iter)); 2626 } 2627 2628 iter = kcdata_iter_find_type(iter, KCDATA_TYPE_MACH_ABSOLUTE_TIME); 2629 T_QUIET; 2630 T_ASSERT_TRUE(kcdata_iter_valid(iter), "timestamp found in stackshot"); 2631 2632 return *(uint64_t *)kcdata_iter_payload(iter); 2633} 2634 2635#define TEST_THREAD_NAME "stackshot_test_thread" 2636 2637static void 2638parse_thread_group_stackshot(void **ssbuf, size_t sslen) 2639{ 2640 bool seen_thread_group_snapshot = false; 2641 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2642 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, 2643 "buffer provided is a stackshot"); 2644 2645 NSMutableSet *thread_groups = [[NSMutableSet alloc] init]; 2646 2647 iter = kcdata_iter_next(iter); 2648 KCDATA_ITER_FOREACH(iter) { 2649 switch (kcdata_iter_type(iter)) { 2650 case KCDATA_TYPE_ARRAY: { 2651 T_QUIET; 2652 T_ASSERT_TRUE(kcdata_iter_array_valid(iter), 2653 "checked that array is valid"); 2654 2655 if (kcdata_iter_array_elem_type(iter) != STACKSHOT_KCTYPE_THREAD_GROUP_SNAPSHOT) { 2656 continue; 2657 } 2658 2659 seen_thread_group_snapshot = true; 2660 2661 if (kcdata_iter_array_elem_size(iter) >= sizeof(struct thread_group_snapshot_v3)) { 2662 struct thread_group_snapshot_v3 *tgs_array = kcdata_iter_payload(iter); 2663 for (uint32_t j = 0; j < kcdata_iter_array_elem_count(iter); j++) { 2664 struct thread_group_snapshot_v3 *tgs = tgs_array + j; 2665 [thread_groups addObject:@(tgs->tgs_id)]; 2666 } 2667 } 2668 else { 2669 struct thread_group_snapshot *tgs_array = kcdata_iter_payload(iter); 2670 for (uint32_t j = 0; j < kcdata_iter_array_elem_count(iter); j++) { 2671 struct thread_group_snapshot *tgs = tgs_array + j; 2672 [thread_groups addObject:@(tgs->tgs_id)]; 2673 } 2674 } 2675 break; 2676 } 2677 } 2678 } 2679 KCDATA_ITER_FOREACH(iter) { 2680 NSError *error = nil; 2681 2682 switch (kcdata_iter_type(iter)) { 2683 2684 case KCDATA_TYPE_CONTAINER_BEGIN: { 2685 T_QUIET; 2686 T_ASSERT_TRUE(kcdata_iter_container_valid(iter), 2687 "checked that container is valid"); 2688 2689 if (kcdata_iter_container_type(iter) != STACKSHOT_KCCONTAINER_THREAD) { 2690 break; 2691 } 2692 2693 NSDictionary *container = parseKCDataContainer(&iter, &error); 2694 T_QUIET; T_ASSERT_NOTNULL(container, "parsed thread container from stackshot"); 2695 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing container"); 2696 2697 int tg = [container[@"thread_snapshots"][@"thread_group"] intValue]; 2698 2699 T_ASSERT_TRUE([thread_groups containsObject:@(tg)], "check that the thread group the thread is in exists"); 2700 2701 break; 2702 }; 2703 2704 } 2705 } 2706 T_ASSERT_TRUE(seen_thread_group_snapshot, "check that we have seen a thread group snapshot"); 2707} 2708 2709static void 2710verify_stackshot_sharedcache_layout(struct dyld_uuid_info_64 *uuids, uint32_t uuid_count) 2711{ 2712 uuid_t cur_shared_cache_uuid; 2713 __block uint32_t lib_index = 0, libs_found = 0; 2714 2715 _dyld_get_shared_cache_uuid(cur_shared_cache_uuid); 2716 int result = dyld_shared_cache_iterate_text(cur_shared_cache_uuid, ^(const dyld_shared_cache_dylib_text_info* info) { 2717 T_QUIET; T_ASSERT_LT(lib_index, uuid_count, "dyld_shared_cache_iterate_text exceeded number of libraries returned by kernel"); 2718 2719 libs_found++; 2720 struct dyld_uuid_info_64 *cur_stackshot_uuid_entry = &uuids[lib_index]; 2721 T_QUIET; T_ASSERT_EQ(memcmp(info->dylibUuid, cur_stackshot_uuid_entry->imageUUID, sizeof(info->dylibUuid)), 0, 2722 "dyld returned UUID doesn't match kernel returned UUID"); 2723 T_QUIET; T_ASSERT_EQ(info->loadAddressUnslid, cur_stackshot_uuid_entry->imageLoadAddress, 2724 "dyld returned load address doesn't match kernel returned load address"); 2725 lib_index++; 2726 }); 2727 2728 T_ASSERT_EQ(result, 0, "iterate shared cache layout"); 2729 T_ASSERT_EQ(libs_found, uuid_count, "dyld iterator returned same number of libraries as kernel"); 2730 2731 T_LOG("verified %d libraries from dyld shared cache", libs_found); 2732} 2733 2734static void 2735check_shared_cache_uuid(uuid_t imageUUID) 2736{ 2737 static uuid_t shared_cache_uuid; 2738 static dispatch_once_t read_shared_cache_uuid; 2739 2740 dispatch_once(&read_shared_cache_uuid, ^{ 2741 T_QUIET; 2742 T_ASSERT_TRUE(_dyld_get_shared_cache_uuid(shared_cache_uuid), "retrieve current shared cache UUID"); 2743 }); 2744 T_QUIET; T_ASSERT_EQ(uuid_compare(shared_cache_uuid, imageUUID), 0, 2745 "dyld returned UUID doesn't match kernel returned UUID for system shared cache"); 2746} 2747 2748/* 2749 * extra dictionary contains data relevant for the given flags: 2750 * PARSE_STACKSHOT_ZOMBIE: zombie_child_pid_key -> @(pid) 2751 * PARSE_STACKSHOT_POSTEXEC: postexec_child_unique_pid_key -> @(unique_pid) 2752 */ 2753static void 2754parse_stackshot(uint64_t stackshot_parsing_flags, void *ssbuf, size_t sslen, NSDictionary *extra) 2755{ 2756 bool delta = (stackshot_parsing_flags & PARSE_STACKSHOT_DELTA); 2757 bool expect_sharedcache_child = (stackshot_parsing_flags & PARSE_STACKSHOT_SHAREDCACHE_FLAGS); 2758 bool expect_zombie_child = (stackshot_parsing_flags & PARSE_STACKSHOT_ZOMBIE); 2759 bool expect_postexec_child = (stackshot_parsing_flags & PARSE_STACKSHOT_POSTEXEC); 2760 bool expect_cseg_waitinfo = (stackshot_parsing_flags & PARSE_STACKSHOT_WAITINFO_CSEG); 2761 bool expect_translated_child = (stackshot_parsing_flags & PARSE_STACKSHOT_TRANSLATED); 2762 bool expect_shared_cache_layout = false; 2763 bool expect_shared_cache_uuid = !delta; 2764 bool expect_dispatch_queue_label = (stackshot_parsing_flags & PARSE_STACKSHOT_DISPATCH_QUEUE_LABEL); 2765 bool expect_turnstile_lock = (stackshot_parsing_flags & PARSE_STACKSHOT_TURNSTILEINFO); 2766 bool expect_srp_waitinfo = (stackshot_parsing_flags & PARSE_STACKSHOT_WAITINFO_SRP); 2767 bool expect_sp_throttled = (stackshot_parsing_flags & PARSE_STACKSHOT_THROTTLED_SP); 2768 bool expect_exec_inprogress = (stackshot_parsing_flags & PARSE_STACKSHOT_EXEC_INPROGRESS); 2769 bool expect_transitioning_task = (stackshot_parsing_flags & PARSE_STACKSHOT_TRANSITIONING); 2770 bool expect_asyncstack = (stackshot_parsing_flags & PARSE_STACKSHOT_ASYNCSTACK); 2771 bool expect_driverkit = (stackshot_parsing_flags & PARSE_STACKSHOT_DRIVERKIT); 2772 bool expect_suspendinfo = (stackshot_parsing_flags & PARSE_STACKSHOT_SUSPENDINFO); 2773 bool found_zombie_child = false, found_postexec_child = false, found_shared_cache_layout = false, found_shared_cache_uuid = false; 2774 bool found_translated_child = false, found_transitioning_task = false; 2775 bool found_dispatch_queue_label = false, found_turnstile_lock = false; 2776 bool found_cseg_waitinfo = false, found_srp_waitinfo = false; 2777 bool found_sharedcache_child = false, found_sharedcache_badflags = false, found_sharedcache_self = false; 2778 bool found_asyncstack = false; 2779 bool found_throttled_service = false; 2780 bool found_exclaves = false; 2781 bool expect_single_task = (stackshot_parsing_flags & PARSE_STACKSHOT_TARGETPID); 2782 uint64_t srp_expected_threadid = 0; 2783 pid_t zombie_child_pid = -1, srp_expected_pid = -1, sharedcache_child_pid = -1, throttled_service_ctx = -1; 2784 pid_t translated_child_pid = -1, transistioning_task_pid = -1; 2785 bool sharedcache_child_sameaddr = false, is_throttled = false; 2786 uint64_t postexec_child_unique_pid = 0, cseg_expected_threadid = 0; 2787 uint64_t sharedcache_child_flags = 0, sharedcache_self_flags = 0; 2788 uint64_t asyncstack_threadid = 0; 2789 NSArray *asyncstack_stack = nil; 2790 char *inflatedBufferBase = NULL; 2791 pid_t exec_inprogress_pid = -1; 2792 void (^exec_inprogress_cb)(uint64_t, uint64_t) = NULL; 2793 int exec_inprogress_found = 0; 2794 uint64_t exec_inprogress_containerid = 0; 2795 void (^driverkit_cb)(pid_t) = NULL; 2796 NSMutableDictionary *sharedCaches = [NSMutableDictionary new]; 2797 uint64_t expected_num_threads = 0, expected_num_tasks = 0, found_percpu_threads = 0, found_tasks = 0, found_percpu_tasks = 0; 2798 NSMutableSet *seen_tasks = [NSMutableSet new]; 2799 2800 if (expect_shared_cache_uuid) { 2801 uuid_t shared_cache_uuid; 2802 if (!_dyld_get_shared_cache_uuid(shared_cache_uuid)) { 2803 T_LOG("Skipping verifying shared cache UUID in stackshot data because not running with a shared cache"); 2804 expect_shared_cache_uuid = false; 2805 } 2806 } 2807 2808 if (stackshot_parsing_flags & PARSE_STACKSHOT_SHAREDCACHE_LAYOUT) { 2809 size_t shared_cache_length = 0; 2810 const void *cache_header = _dyld_get_shared_cache_range(&shared_cache_length); 2811 T_QUIET; T_ASSERT_NOTNULL(cache_header, "current process running with shared cache"); 2812 T_QUIET; T_ASSERT_GT(shared_cache_length, sizeof(struct _dyld_cache_header), "valid shared cache length populated by _dyld_get_shared_cache_range"); 2813 2814 if (_dyld_shared_cache_is_locally_built()) { 2815 T_LOG("device running with locally built shared cache, expect shared cache layout"); 2816 expect_shared_cache_layout = true; 2817 } else { 2818 T_LOG("device running with B&I built shared-cache, no shared cache layout expected"); 2819 } 2820 } 2821 2822 if (expect_sharedcache_child) { 2823 NSNumber* pid_num = extra[sharedcache_child_pid_key]; 2824 NSNumber* sameaddr_num = extra[sharedcache_child_sameaddr_key]; 2825 T_QUIET; T_ASSERT_NOTNULL(pid_num, "sharedcache child pid provided"); 2826 T_QUIET; T_ASSERT_NOTNULL(sameaddr_num, "sharedcache child addrsame provided"); 2827 sharedcache_child_pid = [pid_num intValue]; 2828 T_QUIET; T_ASSERT_GT(sharedcache_child_pid, 0, "sharedcache child pid greater than zero"); 2829 sharedcache_child_sameaddr = [sameaddr_num intValue]; 2830 T_QUIET; T_ASSERT_GE([sameaddr_num intValue], 0, "sharedcache child sameaddr is boolean (0 or 1)"); 2831 T_QUIET; T_ASSERT_LE([sameaddr_num intValue], 1, "sharedcache child sameaddr is boolean (0 or 1)"); 2832 } 2833 2834 if (expect_transitioning_task) { 2835 NSNumber* pid_num = extra[transitioning_pid_key]; 2836 T_ASSERT_NOTNULL(pid_num, "transitioning task pid provided"); 2837 transistioning_task_pid = [pid_num intValue]; 2838 } 2839 2840 if (expect_zombie_child) { 2841 NSNumber* pid_num = extra[zombie_child_pid_key]; 2842 T_QUIET; T_ASSERT_NOTNULL(pid_num, "zombie child pid provided"); 2843 zombie_child_pid = [pid_num intValue]; 2844 T_QUIET; T_ASSERT_GT(zombie_child_pid, 0, "zombie child pid greater than zero"); 2845 } 2846 2847 if (expect_postexec_child) { 2848 NSNumber* unique_pid_num = extra[postexec_child_unique_pid_key]; 2849 T_QUIET; T_ASSERT_NOTNULL(unique_pid_num, "postexec child unique pid provided"); 2850 postexec_child_unique_pid = [unique_pid_num unsignedLongLongValue]; 2851 T_QUIET; T_ASSERT_GT(postexec_child_unique_pid, 0ull, "postexec child unique pid greater than zero"); 2852 } 2853 2854 if (expect_cseg_waitinfo) { 2855 NSNumber* tid_num = extra[cseg_expected_threadid_key]; 2856 T_QUIET; T_ASSERT_NOTNULL(tid_num, "cseg's expected thread id provided"); 2857 cseg_expected_threadid = tid_num.unsignedLongValue; 2858 T_QUIET; T_ASSERT_GT(cseg_expected_threadid, UINT64_C(0), "compressor segment thread is present"); 2859 } 2860 2861 if (expect_srp_waitinfo) { 2862 NSNumber* threadid_num = extra[srp_expected_threadid_key]; 2863 NSNumber* pid_num = extra[srp_expected_pid_key]; 2864 T_QUIET; T_ASSERT_TRUE(threadid_num != nil || pid_num != nil, "expected SRP threadid or pid"); 2865 if (threadid_num != nil) { 2866 srp_expected_threadid = [threadid_num unsignedLongLongValue]; 2867 T_QUIET; T_ASSERT_GT(srp_expected_threadid, 0ull, "srp_expected_threadid greater than zero"); 2868 } 2869 if (pid_num != nil) { 2870 srp_expected_pid = [pid_num intValue]; 2871 T_QUIET; T_ASSERT_GT(srp_expected_pid, 0, "srp_expected_pid greater than zero"); 2872 } 2873 T_LOG("looking for SRP pid: %d threadid: %llu", srp_expected_pid, srp_expected_threadid); 2874 } 2875 2876 if (expect_sp_throttled) { 2877 NSNumber* ctx = extra[sp_throttled_expected_ctxt_key]; 2878 T_QUIET; T_ASSERT_TRUE(ctx != nil, "expected pid"); 2879 throttled_service_ctx = [ctx intValue]; 2880 T_QUIET; T_ASSERT_GT(throttled_service_ctx, 0, "expected pid greater than zero"); 2881 2882 NSNumber *throttled = extra[sp_throttled_expect_flag]; 2883 T_QUIET; T_ASSERT_TRUE(throttled != nil, "expected flag value"); 2884 is_throttled = ([throttled intValue] != 0); 2885 2886 T_LOG("Looking for service with ctxt: %d, thottled:%d", throttled_service_ctx, is_throttled); 2887 } 2888 2889 if (expect_translated_child) { 2890 NSNumber* pid_num = extra[translated_child_pid_key]; 2891 T_QUIET; T_ASSERT_NOTNULL(pid_num, "translated child pid provided"); 2892 translated_child_pid = [pid_num intValue]; 2893 T_QUIET; T_ASSERT_GT(translated_child_pid, 0, "translated child pid greater than zero"); 2894 } 2895 if (expect_exec_inprogress) { 2896 NSNumber* pid_num = extra[exec_inprogress_pid_key]; 2897 T_QUIET; T_ASSERT_NOTNULL(pid_num, "exec inprogress pid provided"); 2898 exec_inprogress_pid = [pid_num intValue]; 2899 T_QUIET; T_ASSERT_GT(exec_inprogress_pid, 0, "exec inprogress pid greater than zero"); 2900 2901 exec_inprogress_cb = extra[exec_inprogress_found_key]; 2902 T_QUIET; T_ASSERT_NOTNULL(exec_inprogress_cb, "exec inprogress found callback provided"); 2903 } 2904 if (expect_driverkit) { 2905 driverkit_cb = extra[driverkit_found_key]; 2906 T_QUIET; T_ASSERT_NOTNULL(driverkit_cb, "driverkit found callback provided"); 2907 } 2908 2909 if (expect_asyncstack) { 2910 NSNumber* threadid_id = extra[asyncstack_expected_threadid_key]; 2911 T_QUIET; T_ASSERT_NOTNULL(threadid_id, "asyncstack threadid provided"); 2912 asyncstack_threadid = [threadid_id unsignedLongLongValue]; 2913 asyncstack_stack = extra[asyncstack_expected_stack_key]; 2914 T_QUIET; T_ASSERT_NOTNULL(asyncstack_stack, "asyncstack expected stack provided"); 2915 } 2916 2917 kcdata_iter_t iter = kcdata_iter(ssbuf, sslen); 2918 if (delta) { 2919 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_DELTA_STACKSHOT, 2920 "buffer provided is a delta stackshot"); 2921 2922 iter = kcdata_iter_next(iter); 2923 } else { 2924 if (kcdata_iter_type(iter) != KCDATA_BUFFER_BEGIN_COMPRESSED) { 2925 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, 2926 "buffer provided is a stackshot"); 2927 2928 iter = kcdata_iter_next(iter); 2929 } else { 2930 /* we are dealing with a compressed buffer */ 2931 iter = kcdata_iter_next(iter); 2932 uint64_t compression_type = 0, totalout = 0, totalin = 0; 2933 2934 uint64_t *data; 2935 char *desc; 2936 for (int i = 0; i < 3; i ++) { 2937 kcdata_iter_get_data_with_desc(iter, &desc, (void **)&data, NULL); 2938 if (strcmp(desc, "kcd_c_type") == 0) { 2939 compression_type = *data; 2940 } else if (strcmp(desc, "kcd_c_totalout") == 0){ 2941 totalout = *data; 2942 } else if (strcmp(desc, "kcd_c_totalin") == 0){ 2943 totalin = *data; 2944 } 2945 2946 iter = kcdata_iter_next(iter); 2947 } 2948 2949 T_ASSERT_EQ(compression_type, UINT64_C(1), "zlib compression is used"); 2950 T_ASSERT_GT(totalout, UINT64_C(0), "successfully gathered how long the compressed buffer is"); 2951 T_ASSERT_GT(totalin, UINT64_C(0), "successfully gathered how long the uncompressed buffer will be at least"); 2952 2953 /* progress to the next kcdata item */ 2954 T_ASSERT_EQ(kcdata_iter_type(iter), KCDATA_BUFFER_BEGIN_STACKSHOT, "compressed stackshot found"); 2955 2956 char *bufferBase = kcdata_iter_payload(iter); 2957 2958 /* 2959 * zlib is used, allocate a buffer based on the metadata, plus 2960 * extra scratch space (+12.5%) in case totalin was inconsistent 2961 */ 2962 size_t inflatedBufferSize = totalin + (totalin >> 3); 2963 inflatedBufferBase = malloc(inflatedBufferSize); 2964 T_QUIET; T_WITH_ERRNO; T_ASSERT_NOTNULL(inflatedBufferBase, "allocated temporary output buffer"); 2965 2966 z_stream zs; 2967 memset(&zs, 0, sizeof(zs)); 2968 T_QUIET; T_ASSERT_EQ(inflateInit(&zs), Z_OK, "inflateInit OK"); 2969 zs.next_in = (unsigned char *)bufferBase; 2970 T_QUIET; T_ASSERT_LE(totalout, (uint64_t)UINT_MAX, "stackshot is not too large"); 2971 zs.avail_in = (uInt)totalout; 2972 zs.next_out = (unsigned char *)inflatedBufferBase; 2973 T_QUIET; T_ASSERT_LE(inflatedBufferSize, (size_t)UINT_MAX, "output region is not too large"); 2974 zs.avail_out = (uInt)inflatedBufferSize; 2975 T_ASSERT_EQ(inflate(&zs, Z_FINISH), Z_STREAM_END, "inflated buffer"); 2976 inflateEnd(&zs); 2977 2978 T_ASSERT_EQ((uint64_t)zs.total_out, totalin, "expected number of bytes inflated"); 2979 2980 /* copy the data after the compressed area */ 2981 T_QUIET; T_ASSERT_GE((void *)bufferBase, ssbuf, 2982 "base of compressed stackshot is after the returned stackshot buffer"); 2983 size_t header_size = (size_t)(bufferBase - (char *)ssbuf); 2984 size_t data_after_compressed_size = sslen - totalout - header_size; 2985 T_QUIET; T_ASSERT_LE(data_after_compressed_size, 2986 inflatedBufferSize - zs.total_out, 2987 "footer fits in the buffer"); 2988 memcpy(inflatedBufferBase + zs.total_out, 2989 bufferBase + totalout, 2990 data_after_compressed_size); 2991 2992 iter = kcdata_iter(inflatedBufferBase, inflatedBufferSize); 2993 } 2994 } 2995 2996 KCDATA_ITER_FOREACH(iter) { 2997 NSError *error = nil; 2998 2999 switch (kcdata_iter_type(iter)) { 3000 case KCDATA_TYPE_ARRAY: { 3001 T_QUIET; 3002 T_ASSERT_TRUE(kcdata_iter_array_valid(iter), 3003 "checked that array is valid"); 3004 3005 NSMutableDictionary *array = parseKCDataArray(iter, &error); 3006 T_QUIET; T_ASSERT_NOTNULL(array, "parsed array from stackshot"); 3007 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing array"); 3008 3009 if (kcdata_iter_array_elem_type(iter) == STACKSHOT_KCTYPE_SYS_SHAREDCACHE_LAYOUT) { 3010 struct dyld_uuid_info_64 *shared_cache_uuids = kcdata_iter_payload(iter); 3011 uint32_t uuid_count = kcdata_iter_array_elem_count(iter); 3012 T_ASSERT_NOTNULL(shared_cache_uuids, "parsed shared cache layout array"); 3013 T_ASSERT_GT(uuid_count, 0, "returned valid number of UUIDs from shared cache"); 3014 verify_stackshot_sharedcache_layout(shared_cache_uuids, uuid_count); 3015 found_shared_cache_layout = true; 3016 } 3017 3018 break; 3019 } 3020 case KCDATA_TYPE_CONTAINER_BEGIN: { 3021 T_QUIET; 3022 T_ASSERT_TRUE(kcdata_iter_container_valid(iter), 3023 "checked that container is valid"); 3024 3025 uint64_t containerid = kcdata_iter_container_id(iter); 3026 uint32_t container_type = kcdata_iter_container_type(iter); 3027 3028 if (container_type == STACKSHOT_KCCONTAINER_SHAREDCACHE) { 3029 NSDictionary *container = parseKCDataContainer(&iter, &error); 3030 T_QUIET; T_ASSERT_NOTNULL(container, "parsed sharedcache container from stackshot"); 3031 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing sharedcache container"); 3032 T_QUIET; T_EXPECT_EQ(sharedCaches[@(containerid)], nil, "sharedcache containerid %lld should be unique", containerid); 3033 sharedCaches[@(containerid)] = container; 3034 break; 3035 } 3036 3037 if (container_type == STACKSHOT_KCCONTAINER_EXCLAVES) { 3038 found_exclaves = true; 3039 break; 3040 } 3041 3042 /* 3043 * treat containers other than tasks/transitioning_tasks 3044 * as expanded in-line. 3045 */ 3046 if (container_type != STACKSHOT_KCCONTAINER_TASK && 3047 container_type != STACKSHOT_KCCONTAINER_TRANSITIONING_TASK) { 3048 T_LOG("container skipped: %d", container_type); 3049 break; 3050 } 3051 NSDictionary *container = parseKCDataContainer(&iter, &error); 3052 T_QUIET; T_ASSERT_NOTNULL(container, "parsed task/transitioning_task container from stackshot"); 3053 T_QUIET; T_ASSERT_NULL(error, "error unset after parsing container"); 3054 3055 found_tasks++; 3056 3057 NSDictionary* task_snapshot = container[@"task_snapshots"][@"task_snapshot"]; 3058 NSDictionary* task_delta_snapshot = container[@"task_snapshots"][@"task_delta_snapshot"]; 3059 NSDictionary* transitioning_task_snapshot = container[@"transitioning_task_snapshots"][@"transitioning_task_snapshot"]; 3060 3061 NSNumber *task_pid = NULL; 3062 if (task_snapshot) { 3063 task_pid = task_snapshot[@"ts_unique_pid"]; 3064 } else if(task_delta_snapshot) { 3065 task_pid = task_snapshot[@"tds_unique_pid"]; 3066 } else if(transitioning_task_snapshot) { 3067 task_pid = transitioning_task_snapshot[@"tts_pid"]; 3068 } 3069 3070 if (task_pid && [seen_tasks containsObject:task_pid]) { 3071 T_QUIET; T_ASSERT_FALSE([seen_tasks containsObject:task_pid], "No duplicate PIDs in stackshot"); 3072 [seen_tasks addObject:task_pid]; 3073 } 3074 3075 /* 3076 * Having processed the container, we now only check it 3077 * if it's the correct type. 3078 */ 3079 if ((!expect_transitioning_task && (container_type != STACKSHOT_KCCONTAINER_TASK)) || 3080 (expect_transitioning_task && (container_type != STACKSHOT_KCCONTAINER_TRANSITIONING_TASK))) { 3081 break; 3082 } 3083 if (!expect_transitioning_task) { 3084 T_QUIET; T_ASSERT_TRUE(!!task_snapshot != !!task_delta_snapshot, "Either task_snapshot xor task_delta_snapshot provided"); 3085 } 3086 3087 if (expect_dispatch_queue_label && !found_dispatch_queue_label) { 3088 for (id thread_key in container[@"task_snapshots"][@"thread_snapshots"]) { 3089 NSMutableDictionary *thread = container[@"task_snapshots"][@"thread_snapshots"][thread_key]; 3090 NSString *dql = thread[@"dispatch_queue_label"]; 3091 3092 if ([dql isEqualToString:@TEST_STACKSHOT_QUEUE_LABEL]) { 3093 found_dispatch_queue_label = true; 3094 break; 3095 } 3096 } 3097 } 3098 3099 if (expect_transitioning_task && !found_transitioning_task) { 3100 if (transitioning_task_snapshot) { 3101 uint64_t the_pid = [transitioning_task_snapshot[@"tts_pid"] unsignedLongLongValue]; 3102 if (the_pid == (uint64_t)transistioning_task_pid) { 3103 found_transitioning_task = true; 3104 3105 T_PASS("FOUND Transitioning task %llu has a transitioning task snapshot", (uint64_t) transistioning_task_pid); 3106 break; 3107 } 3108 } 3109 } 3110 3111 if (expect_postexec_child && !found_postexec_child) { 3112 if (task_snapshot) { 3113 uint64_t unique_pid = [task_snapshot[@"ts_unique_pid"] unsignedLongLongValue]; 3114 if (unique_pid == postexec_child_unique_pid) { 3115 found_postexec_child = true; 3116 3117 T_PASS("post-exec child %llu has a task snapshot", postexec_child_unique_pid); 3118 3119 break; 3120 } 3121 } 3122 3123 if (task_delta_snapshot) { 3124 uint64_t unique_pid = [task_delta_snapshot[@"tds_unique_pid"] unsignedLongLongValue]; 3125 if (unique_pid == postexec_child_unique_pid) { 3126 found_postexec_child = true; 3127 3128 T_FAIL("post-exec child %llu shouldn't have a delta task snapshot", postexec_child_unique_pid); 3129 3130 break; 3131 } 3132 } 3133 } 3134 3135 int pid = [task_snapshot[@"ts_pid"] intValue]; 3136 3137 if (pid && expect_shared_cache_uuid && !found_shared_cache_uuid) { 3138 id ptr = container[@"task_snapshots"][@"shared_cache_dyld_load_info"]; 3139 if (ptr) { 3140 id uuid = ptr[@"imageUUID"]; 3141 3142 uint8_t uuid_p[16]; 3143 for (unsigned int i = 0; i < 16; i ++) { 3144 NSNumber *uuidByte = uuid[i]; 3145 uuid_p[i] = (uint8_t)uuidByte.charValue; 3146 } 3147 3148 check_shared_cache_uuid(uuid_p); 3149 3150 uint64_t baseAddress = (uint64_t)((NSNumber *)ptr[@"imageSlidBaseAddress"]).longLongValue; 3151 uint64_t firstMapping = (uint64_t)((NSNumber *)ptr[@"sharedCacheSlidFirstMapping"]).longLongValue; 3152 3153 T_EXPECT_LE(baseAddress, firstMapping, 3154 "in per-task shared_cache_dyld_load_info, " 3155 "baseAddress <= firstMapping"); 3156 T_EXPECT_GE(baseAddress + (7ull << 32) + (1ull << 29), 3157 firstMapping, 3158 "in per-task shared_cache_dyld_load_info, " 3159 "baseAddress + 28.5gig >= firstMapping"); 3160 3161 size_t shared_cache_len; 3162 const void *addr = _dyld_get_shared_cache_range(&shared_cache_len); 3163 T_EXPECT_EQ((uint64_t)addr, firstMapping, 3164 "SlidFirstMapping should match shared_cache_range"); 3165 3166 /* 3167 * check_shared_cache_uuid() will assert on failure, so if 3168 * we get here, then we have found the shared cache UUID 3169 * and it's correct 3170 */ 3171 found_shared_cache_uuid = true; 3172 } 3173 } 3174 3175 if (expect_sharedcache_child) { 3176 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3177 uint64_t sharedregion_flags = (task_flags & (kTaskSharedRegionNone | kTaskSharedRegionSystem | kTaskSharedRegionOther)); 3178 id sharedregion_info = container[@"task_snapshots"][@"shared_cache_dyld_load_info"]; 3179 id sharedcache_id = container[@"task_snapshots"][@"sharedCacheID"]; 3180 if (!found_sharedcache_badflags) { 3181 T_QUIET; T_EXPECT_NE(sharedregion_flags, 0ll, "one of the kTaskSharedRegion flags should be set on all tasks"); 3182 bool multiple = (sharedregion_flags & (sharedregion_flags - 1)) != 0; 3183 T_QUIET; T_EXPECT_FALSE(multiple, "only one kTaskSharedRegion flag should be set on each task"); 3184 found_sharedcache_badflags = (sharedregion_flags == 0 || multiple); 3185 } 3186 if (pid == 0) { 3187 T_ASSERT_EQ(sharedregion_flags, (uint64_t)kTaskSharedRegionNone, "Kernel proc (pid 0) should have no shared region"); 3188 } else if (pid == sharedcache_child_pid) { 3189 found_sharedcache_child = true; 3190 sharedcache_child_flags = sharedregion_flags; 3191 } else if (pid == getpid()) { 3192 found_sharedcache_self = true; 3193 sharedcache_self_flags = sharedregion_flags; 3194 } 3195 if (sharedregion_flags == kTaskSharedRegionOther && !(task_flags & kTaskSharedRegionInfoUnavailable)) { 3196 T_QUIET; T_EXPECT_NOTNULL(sharedregion_info, "kTaskSharedRegionOther should have a shared_cache_dyld_load_info struct"); 3197 T_QUIET; T_EXPECT_NOTNULL(sharedcache_id, "kTaskSharedRegionOther should have a sharedCacheID"); 3198 if (sharedcache_id != nil) { 3199 T_QUIET; T_EXPECT_NOTNULL(sharedCaches[sharedcache_id], "sharedCacheID %d should exist", [sharedcache_id intValue]); 3200 } 3201 } else { 3202 T_QUIET; T_EXPECT_NULL(sharedregion_info, "non-kTaskSharedRegionOther should have no shared_cache_dyld_load_info struct"); 3203 T_QUIET; T_EXPECT_NULL(sharedcache_id, "non-kTaskSharedRegionOther should have no sharedCacheID"); 3204 } 3205 } 3206 3207 if (expect_zombie_child && (pid == zombie_child_pid)) { 3208 found_zombie_child = true; 3209 3210 expected_num_tasks += 1; 3211 3212 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3213 T_ASSERT_TRUE((task_flags & kTerminatedSnapshot) == kTerminatedSnapshot, "child zombie marked as terminated"); 3214 3215 continue; 3216 } 3217 3218 if (expect_translated_child && (pid == translated_child_pid)) { 3219 found_translated_child = true; 3220 3221 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3222 T_EXPECT_BITS_SET(task_flags, kTaskIsTranslated, "child marked as translated"); 3223 3224 continue; 3225 } 3226 if (expect_exec_inprogress && (pid == exec_inprogress_pid || pid == -exec_inprogress_pid)) { 3227 exec_inprogress_found++; 3228 T_LOG("found exec task with pid %d, instance %d", pid, exec_inprogress_found); 3229 T_QUIET; T_ASSERT_LE(exec_inprogress_found, 2, "no more than two with the expected pid"); 3230 if (exec_inprogress_found == 2) { 3231 T_LOG("found 2 tasks with pid %d", exec_inprogress_pid); 3232 exec_inprogress_cb(containerid, exec_inprogress_containerid); 3233 } else { 3234 exec_inprogress_containerid = containerid; 3235 } 3236 } 3237 if (expect_driverkit && driverkit_cb != NULL) { 3238 driverkit_cb(pid); 3239 } 3240 if (expect_cseg_waitinfo) { 3241 NSArray *winfos = container[@"task_snapshots"][@"thread_waitinfo"]; 3242 3243 for (id i in winfos) { 3244 NSNumber *waitType = i[@"wait_type"]; 3245 NSNumber *owner = i[@"owner"]; 3246 if (waitType.intValue == kThreadWaitCompressor && 3247 owner.unsignedLongValue == cseg_expected_threadid) { 3248 found_cseg_waitinfo = true; 3249 break; 3250 } 3251 } 3252 } 3253 3254 if (expect_srp_waitinfo) { 3255 NSArray *tinfos = container[@"task_snapshots"][@"thread_turnstileinfo"]; 3256 NSArray *winfos = container[@"task_snapshots"][@"thread_waitinfo"]; 3257 for (id i in tinfos) { 3258 if (!found_srp_waitinfo) { 3259 bool found_thread = false; 3260 bool found_pid = false; 3261 if (([i[@"turnstile_flags"] intValue] & STACKSHOT_TURNSTILE_STATUS_THREAD) && 3262 [i[@"turnstile_context"] unsignedLongLongValue] == srp_expected_threadid && 3263 srp_expected_threadid != 0) { 3264 found_thread = true; 3265 } 3266 if (([i[@"turnstile_flags"] intValue] & STACKSHOT_TURNSTILE_STATUS_BLOCKED_ON_TASK) && 3267 [i[@"turnstile_context"] intValue] == srp_expected_pid && 3268 srp_expected_pid != -1) { 3269 found_pid = true; 3270 } 3271 if (found_pid || found_thread) { 3272 T_LOG("found SRP %s %lld waiter: %d", (found_thread ? "thread" : "pid"), 3273 [i[@"turnstile_context"] unsignedLongLongValue], [i[@"waiter"] intValue]); 3274 /* we found something that is blocking the correct threadid */ 3275 for (id j in winfos) { 3276 if ([j[@"waiter"] intValue] == [i[@"waiter"] intValue] && 3277 [j[@"wait_type"] intValue] == kThreadWaitPortReceive) { 3278 found_srp_waitinfo = true; 3279 T_EXPECT_EQ([j[@"wait_flags"] intValue], STACKSHOT_WAITINFO_FLAGS_SPECIALREPLY, 3280 "SRP waitinfo should be marked as a special reply"); 3281 break; 3282 } 3283 } 3284 3285 if (found_srp_waitinfo) { 3286 break; 3287 } 3288 } 3289 } 3290 } 3291 } 3292 3293 if (expect_sp_throttled) { 3294 NSArray *tinfos = container[@"task_snapshots"][@"thread_turnstileinfo"]; 3295 for (id i in tinfos) { 3296 if (([i[@"turnstile_flags"] intValue] & STACKSHOT_TURNSTILE_STATUS_PORTFLAGS) 3297 && [i[@"turnstile_context"] intValue] == throttled_service_ctx) { 3298 int portlabel_id = [i[@"portlabel_id"] intValue]; 3299 T_LOG("[pid:%d] Turnstile (flags = 0x%x, ctx = %d, portlabel_id = %d)", pid, 3300 [i[@"turnstile_flags"] intValue], [i[@"turnstile_context"] intValue], portlabel_id); 3301 for (id portid in container[@"task_snapshots"][@"portlabels"]) { 3302 if (portlabel_id != [portid intValue]) { 3303 continue; 3304 } 3305 3306 NSMutableDictionary *portlabel = container[@"task_snapshots"][@"portlabels"][portid]; 3307 T_ASSERT_TRUE(portlabel != nil, "Found portlabel id: %d", [portid intValue]); 3308 NSString *portlabel_name = portlabel[@"portlabel_name"]; 3309 T_EXPECT_TRUE(portlabel_name != nil, "Found portlabel %s", portlabel_name.UTF8String); 3310 T_EXPECT_EQ_STR(portlabel_name.UTF8String, THROTTLED_SERVICE_NAME, "throttled service port name matches"); 3311 T_EXPECT_EQ(([portlabel[@"portlabel_flags"] intValue] & STACKSHOT_PORTLABEL_THROTTLED) != 0, 3312 is_throttled, "Port %s throttled", is_throttled ? "is" : "isn't"); 3313 found_throttled_service = true; 3314 break; 3315 } 3316 } 3317 3318 if (found_throttled_service) { 3319 break; 3320 } 3321 } 3322 } 3323 3324 if (expect_suspendinfo) { 3325 // TODO: rdar://112563110 3326 } 3327 3328 3329 if (pid != getpid()) { 3330 break; 3331 } 3332 3333 T_EXPECT_EQ_STR(current_process_name(), 3334 [task_snapshot[@"ts_p_comm"] UTF8String], 3335 "current process name matches in stackshot"); 3336 3337 uint64_t task_flags = [task_snapshot[@"ts_ss_flags"] unsignedLongLongValue]; 3338 T_ASSERT_BITS_NOTSET(task_flags, kTerminatedSnapshot, "current process not marked as terminated"); 3339 T_ASSERT_BITS_NOTSET(task_flags, kTaskIsTranslated, "current process not marked as translated"); 3340 3341 T_QUIET; 3342 T_EXPECT_LE(pid, [task_snapshot[@"ts_unique_pid"] intValue], 3343 "unique pid is greater than pid"); 3344 3345 NSDictionary* task_cpu_architecture = container[@"task_snapshots"][@"task_cpu_architecture"]; 3346 T_QUIET; T_ASSERT_NOTNULL(task_cpu_architecture[@"cputype"], "have cputype"); 3347 T_QUIET; T_ASSERT_NOTNULL(task_cpu_architecture[@"cpusubtype"], "have cputype"); 3348 int cputype = [task_cpu_architecture[@"cputype"] intValue]; 3349 int cpusubtype = [task_cpu_architecture[@"cpusubtype"] intValue]; 3350 3351 struct proc_archinfo archinfo; 3352 int retval = proc_pidinfo(pid, PROC_PIDARCHINFO, 0, &archinfo, sizeof(archinfo)); 3353 T_QUIET; T_WITH_ERRNO; T_ASSERT_GT(retval, 0, "proc_pidinfo(PROC_PIDARCHINFO) returned a value > 0"); 3354 T_QUIET; T_ASSERT_EQ(retval, (int)sizeof(struct proc_archinfo), "proc_pidinfo call for PROC_PIDARCHINFO returned expected size"); 3355 T_QUIET; T_EXPECT_EQ(cputype, archinfo.p_cputype, "cpu type is correct"); 3356 T_QUIET; T_EXPECT_EQ(cpusubtype, archinfo.p_cpusubtype, "cpu subtype is correct"); 3357 3358 NSDictionary * codesigning_info = container[@"task_snapshots"][@"stackshot_task_codesigning_info"]; 3359 T_QUIET; T_ASSERT_NOTNULL(codesigning_info[@"csflags"], "have csflags"); 3360 uint64_t flags = [codesigning_info[@"csflags"] unsignedLongLongValue]; 3361 T_QUIET; T_EXPECT_GT(flags, 0, "nonzero csflags"); 3362 3363 T_QUIET; T_ASSERT_NOTNULL(container[@"task_snapshots"][@"jetsam_coalition"], "have jetsam coalition"); 3364 uint64_t jetsam_coalition = [container[@"task_snapshots"][@"jetsam_coalition"] unsignedLongLongValue]; 3365 T_QUIET; T_EXPECT_GT(jetsam_coalition, 0, "nonzero jetsam coalition"); 3366 3367 bool found_main_thread = false; 3368 uint64_t main_thread_id = -1ULL; 3369 bool found_null_kernel_frame = false; 3370 for (id thread_key in container[@"task_snapshots"][@"thread_snapshots"]) { 3371 NSMutableDictionary *thread = container[@"task_snapshots"][@"thread_snapshots"][thread_key]; 3372 NSDictionary *thread_snap = thread[@"thread_snapshot"]; 3373 3374 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_thread_id"] intValue], 0, 3375 "thread ID of thread in current task is valid"); 3376 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_base_priority"] intValue], 0, 3377 "base priority of thread in current task is valid"); 3378 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_sched_priority"] intValue], 0, 3379 "scheduling priority of thread in current task is valid"); 3380 3381 NSString *pth_name = thread[@"pth_name"]; 3382 if (pth_name != nil && [pth_name isEqualToString:@TEST_THREAD_NAME]) { 3383 found_main_thread = true; 3384 main_thread_id = [thread_snap[@"ths_thread_id"] unsignedLongLongValue]; 3385 3386 T_QUIET; T_EXPECT_GT([thread_snap[@"ths_total_syscalls"] intValue], 0, 3387 "total syscalls of current thread is valid"); 3388 3389 NSDictionary *cpu_times = thread[@"cpu_times"]; 3390 T_EXPECT_GE([cpu_times[@"runnable_time"] intValue], 3391 [cpu_times[@"system_time"] intValue] + 3392 [cpu_times[@"user_time"] intValue], 3393 "runnable time of current thread is valid"); 3394 } 3395 if (!found_null_kernel_frame) { 3396 for (NSNumber *frame in thread[@"kernel_frames"]) { 3397 if (frame.unsignedLongValue == 0) { 3398 found_null_kernel_frame = true; 3399 break; 3400 } 3401 } 3402 } 3403 if (expect_asyncstack && !found_asyncstack && 3404 asyncstack_threadid == [thread_snap[@"ths_thread_id"] unsignedLongLongValue]) { 3405 found_asyncstack = true; 3406 NSArray* async_stack = thread[@"user_async_stack_frames"]; 3407 NSNumber* start_idx = thread[@"user_async_start_index"]; 3408 NSArray* user_stack = thread[@"user_stack_frames"]; 3409 T_QUIET; T_ASSERT_NOTNULL(async_stack, "async thread %#llx has user_async_stack_frames", asyncstack_threadid); 3410 T_QUIET; T_ASSERT_NOTNULL(start_idx, "async thread %#llx has user_async_start_index", asyncstack_threadid); 3411 T_QUIET; T_ASSERT_NOTNULL(user_stack, "async thread %#llx has user_stack_frames", asyncstack_threadid); 3412 T_QUIET; T_ASSERT_EQ(async_stack.count, asyncstack_stack.count, 3413 "actual async_stack count == expected async_stack count"); 3414 for (size_t i = 0; i < async_stack.count; i++) { 3415 T_EXPECT_EQ([async_stack[i][@"lr"] unsignedLongLongValue], 3416 [asyncstack_stack[i] unsignedLongLongValue], "frame %zu matches", i); 3417 } 3418 } 3419 } 3420 T_EXPECT_TRUE(found_main_thread, "found main thread for current task in stackshot"); 3421 T_EXPECT_FALSE(found_null_kernel_frame, "should not see any NULL kernel frames"); 3422 3423 if (expect_turnstile_lock && !found_turnstile_lock) { 3424 NSArray *tsinfos = container[@"task_snapshots"][@"thread_turnstileinfo"]; 3425 3426 for (id i in tsinfos) { 3427 if ([i[@"turnstile_context"] unsignedLongLongValue] == main_thread_id) { 3428 found_turnstile_lock = true; 3429 break; 3430 } 3431 } 3432 } 3433 break; 3434 } 3435 case STACKSHOT_KCTYPE_SHAREDCACHE_LOADINFO: { 3436 // Legacy shared cache info 3437 struct dyld_shared_cache_loadinfo *payload = kcdata_iter_payload(iter); 3438 T_ASSERT_EQ((size_t)kcdata_iter_size(iter), sizeof(*payload), "valid dyld_shared_cache_loadinfo struct"); 3439 3440 check_shared_cache_uuid(payload->sharedCacheUUID); 3441 3442 T_EXPECT_LE(payload->sharedCacheUnreliableSlidBaseAddress, 3443 payload->sharedCacheSlidFirstMapping, 3444 "SlidBaseAddress <= SlidFirstMapping"); 3445 T_EXPECT_GE(payload->sharedCacheUnreliableSlidBaseAddress + (7ull << 32) + (1ull << 29), 3446 payload->sharedCacheSlidFirstMapping, 3447 "SlidFirstMapping should be within 28.5gigs of SlidBaseAddress"); 3448 3449 size_t shared_cache_len; 3450 const void *addr = _dyld_get_shared_cache_range(&shared_cache_len); 3451 T_EXPECT_EQ((uint64_t)addr, payload->sharedCacheSlidFirstMapping, 3452 "SlidFirstMapping should match shared_cache_range"); 3453 3454 /* 3455 * check_shared_cache_uuid() asserts on failure, so we must have 3456 * found the shared cache UUID to be correct. 3457 */ 3458 found_shared_cache_uuid = true; 3459 break; 3460 } 3461 case KCDATA_TYPE_UINT64_DESC: { 3462 char *desc; 3463 uint64_t *data; 3464 uint32_t size; 3465 kcdata_iter_get_data_with_desc(iter, &desc, &data, &size); 3466 3467 if (strcmp(desc, "stackshot_tasks_count") == 0) { 3468 expected_num_tasks = *data; 3469 } else if (strcmp(desc, "stackshot_threads_count") == 0) { 3470 expected_num_threads = *data; 3471 } 3472 3473 break; 3474 } 3475 case STACKSHOT_KCTYPE_LATENCY_INFO_CPU: { 3476 struct stackshot_latency_cpu *cpu_latency = kcdata_iter_payload(iter); 3477 found_percpu_tasks += cpu_latency->tasks_processed; 3478 found_percpu_threads += cpu_latency->threads_processed; 3479 break; 3480 } 3481 } 3482 } 3483 3484 if (expect_sharedcache_child) { 3485 T_QUIET; T_ASSERT_TRUE(found_sharedcache_child, "found sharedcache child in kcdata"); 3486 T_QUIET; T_ASSERT_TRUE(found_sharedcache_self, "found self in kcdata"); 3487 if (found_sharedcache_child && found_sharedcache_self) { 3488 T_QUIET; T_ASSERT_NE(sharedcache_child_flags, (uint64_t)kTaskSharedRegionNone, "sharedcache child should have shared region"); 3489 T_QUIET; T_ASSERT_NE(sharedcache_self_flags, (uint64_t)kTaskSharedRegionNone, "sharedcache: self should have shared region"); 3490 if (sharedcache_self_flags == kTaskSharedRegionSystem && !sharedcache_child_sameaddr) { 3491 /* If we're in the system shared region, and the child has a different address, child must have an Other shared region */ 3492 T_ASSERT_EQ(sharedcache_child_flags, (uint64_t)kTaskSharedRegionOther, 3493 "sharedcache child should have Other shared region"); 3494 } 3495 } 3496 } 3497 3498 if (expect_transitioning_task) { 3499 T_QUIET; T_ASSERT_TRUE(found_transitioning_task, "found transitioning_task child in kcdata"); 3500 } 3501 3502 if (expect_exec_inprogress) { 3503 T_QUIET; T_ASSERT_GT(exec_inprogress_found, 0, "found at least 1 task for execing process"); 3504 } 3505 3506 if (expect_zombie_child) { 3507 T_QUIET; T_ASSERT_TRUE(found_zombie_child, "found zombie child in kcdata"); 3508 } 3509 3510 if (expect_postexec_child) { 3511 T_QUIET; T_ASSERT_TRUE(found_postexec_child, "found post-exec child in kcdata"); 3512 } 3513 3514 if (expect_translated_child) { 3515 T_QUIET; T_ASSERT_TRUE(found_translated_child, "found translated child in kcdata"); 3516 } 3517 3518 if (expect_shared_cache_layout) { 3519 T_QUIET; T_ASSERT_TRUE(found_shared_cache_layout, "shared cache layout found in kcdata"); 3520 } 3521 3522 if (expect_shared_cache_uuid) { 3523 T_QUIET; T_ASSERT_TRUE(found_shared_cache_uuid, "shared cache UUID found in kcdata"); 3524 } 3525 3526 if (expect_dispatch_queue_label) { 3527 T_QUIET; T_ASSERT_TRUE(found_dispatch_queue_label, "dispatch queue label found in kcdata"); 3528 } 3529 3530 if (expect_turnstile_lock) { 3531 T_QUIET; T_ASSERT_TRUE(found_turnstile_lock, "found expected deadlock"); 3532 } 3533 3534 if (expect_cseg_waitinfo) { 3535 T_QUIET; T_ASSERT_TRUE(found_cseg_waitinfo, "found c_seg waitinfo"); 3536 } 3537 3538 if (expect_srp_waitinfo) { 3539 T_QUIET; T_ASSERT_TRUE(found_srp_waitinfo, "found special reply port waitinfo"); 3540 } 3541 3542 if (expect_sp_throttled) { 3543 T_QUIET; T_ASSERT_TRUE(found_throttled_service, "found the throttled service"); 3544 } 3545 3546 if (expect_asyncstack) { 3547 T_QUIET; T_ASSERT_TRUE(found_asyncstack, "found async stack threadid"); 3548 } 3549 3550 if ([extra objectForKey:no_exclaves_key] != nil) { 3551 T_QUIET; T_ASSERT_FALSE(found_exclaves, "did not find any Exclaves data"); 3552 } 3553 3554 3555 bool check_counts = !delta && !found_transitioning_task && !expect_single_task && !expect_driverkit; 3556 3557 if (check_counts && (expected_num_threads != 0) && (found_percpu_threads != 0)) { 3558 /* If the task counts below check out, we can be sure that the per-cpu reported thread counts are accurate. */ 3559 T_QUIET; T_ASSERT_EQ_ULLONG(found_percpu_threads, expected_num_threads, "number of threads reported by CPUs matches expected count"); 3560 } 3561 3562 if (check_counts && (expected_num_tasks != 0)) { 3563 T_QUIET; T_ASSERT_EQ_ULLONG(found_tasks, expected_num_tasks, "number of tasks in kcdata matches expected count"); 3564 if (found_percpu_tasks != 0) { 3565 T_QUIET; T_ASSERT_EQ_ULLONG(found_percpu_tasks, expected_num_tasks, "number of tasks reported by CPUs matches expected count"); 3566 } 3567 } 3568 3569 T_ASSERT_FALSE(KCDATA_ITER_FOREACH_FAILED(iter), "successfully iterated kcdata"); 3570 3571 free(inflatedBufferBase); 3572} 3573 3574static const char * 3575current_process_name(void) 3576{ 3577 static char name[64]; 3578 3579 if (!name[0]) { 3580 int ret = proc_name(getpid(), name, sizeof(name)); 3581 T_QUIET; 3582 T_ASSERT_POSIX_SUCCESS(ret, "proc_name failed for current process"); 3583 } 3584 3585 return name; 3586} 3587 3588static void 3589initialize_thread(void) 3590{ 3591 int ret = pthread_setname_np(TEST_THREAD_NAME); 3592 T_QUIET; 3593 T_ASSERT_POSIX_ZERO(ret, "set thread name to %s", TEST_THREAD_NAME); 3594} 3595 3596T_DECL(dirty_buffer, "test that stackshot works with a dirty input buffer from kernel", T_META_TAG_VM_PREFERRED) 3597{ 3598 const char *test_sysctl = "stackshot_dirty_buffer"; 3599 int64_t result; 3600 3601 T_LOG("running sysctl to trigger kernel-driven stackshot"); 3602 result = run_sysctl_test(test_sysctl, 0); 3603 T_ASSERT_EQ_LLONG(result, 1, "sysctl result indicated success"); 3604} 3605 3606T_DECL(kernel_initiated, "smoke test that stackshot works with kernel-initiated stackshots", T_META_TAG_VM_PREFERRED) 3607{ 3608 const char *test_sysctl = "stackshot_kernel_initiator"; 3609 int64_t result; 3610 __block bool did_get_stackshot = false; 3611 3612 initialize_thread(); // must run before the stackshots to keep parse_stackshot happy 3613 3614 T_LOG("running sysctl to trigger kernel-driven stackshot type 1"); 3615 result = run_sysctl_test(test_sysctl, 1); 3616 T_ASSERT_EQ_LLONG(result, 1, "sysctl result indicated success"); 3617 3618 T_LOG("running sysctl to trigger kernel-driven stackshot type 2"); 3619 result = run_sysctl_test(test_sysctl, 2); 3620 T_ASSERT_EQ_LLONG(result, 1, "sysctl result indicated success"); 3621 3622 struct scenario scenario = { 3623 .name = "from_kernel_initiated", 3624 .flags = STACKSHOT_RETRIEVE_EXISTING_BUFFER, 3625 }; 3626 3627 T_LOG("attempting to fetch stored in-kernel stackshot"); 3628 take_stackshot(&scenario, false, ^(void *ssbuf, size_t sslen) { 3629 T_ASSERT_NOTNULL(ssbuf, "non-null kernel stackshot"); 3630 T_ASSERT_GT(sslen, 0, "non-zero stackshot size"); 3631 parse_stackshot(0, ssbuf, sslen, nil); 3632 did_get_stackshot = true; 3633 }); 3634 3635 T_ASSERT_TRUE(did_get_stackshot, "got stackshot from kernel type 2"); 3636} 3637