1 /* 2 * Copyright (C) 2008 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 #define LOG_TAG "dumpstate" 18 19 #include "dumpstate.h" 20 21 #include <dirent.h> 22 #include <fcntl.h> 23 #include <libgen.h> 24 #include <math.h> 25 #include <poll.h> 26 #include <signal.h> 27 #include <stdarg.h> 28 #include <stdio.h> 29 #include <stdlib.h> 30 #include <string.h> 31 #include <sys/capability.h> 32 #include <sys/inotify.h> 33 #include <sys/klog.h> 34 #include <sys/prctl.h> 35 #include <sys/stat.h> 36 #include <sys/time.h> 37 #include <sys/wait.h> 38 #include <time.h> 39 #include <unistd.h> 40 41 #include <memory> 42 #include <set> 43 #include <string> 44 #include <vector> 45 46 #include <android-base/file.h> 47 #include <android-base/properties.h> 48 #include <android-base/stringprintf.h> 49 #include <android-base/strings.h> 50 #include <android-base/unique_fd.h> 51 #include <cutils/properties.h> 52 #include <cutils/sockets.h> 53 #include <debuggerd/client.h> 54 #include <dumputils/dump_utils.h> 55 #include <log/log.h> 56 #include <private/android_filesystem_config.h> 57 58 #include "DumpstateInternal.h" 59 60 // TODO: remove once moved to namespace 61 using android::os::dumpstate::CommandOptions; 62 using android::os::dumpstate::DumpFileToFd; 63 using android::os::dumpstate::PropertiesHelper; 64 65 // Keep in sync with 66 // frameworks/base/services/core/java/com/android/server/am/ActivityManagerService.java 67 static const int TRACE_DUMP_TIMEOUT_MS = 10000; // 10 seconds 68 69 /* Most simple commands have 10 as timeout, so 5 is a good estimate */ 70 static const int32_t WEIGHT_FILE = 5; 71 72 // TODO: temporary variables and functions used during C++ refactoring 73 static Dumpstate& ds = Dumpstate::GetInstance(); 74 static int RunCommand(const std::string& title, const std::vector<std::string>& full_command, 75 const CommandOptions& options = CommandOptions::DEFAULT) { 76 return ds.RunCommand(title, full_command, options); 77 } 78 79 // Reasonable value for max stats. 80 static const int STATS_MAX_N_RUNS = 1000; 81 static const long STATS_MAX_AVERAGE = 100000; 82 83 CommandOptions Dumpstate::DEFAULT_DUMPSYS = CommandOptions::WithTimeout(30).Build(); 84 85 Dumpstate::Dumpstate(const std::string& version) 86 : pid_(getpid()), version_(version), now_(time(nullptr)) { 87 } 88 89 Dumpstate& Dumpstate::GetInstance() { 90 static Dumpstate singleton_(android::base::GetProperty("dumpstate.version", VERSION_CURRENT)); 91 return singleton_; 92 } 93 94 DurationReporter::DurationReporter(const std::string& title, bool log_only) 95 : title_(title), log_only_(log_only) { 96 if (!title_.empty()) { 97 started_ = Nanotime(); 98 } 99 } 100 101 DurationReporter::~DurationReporter() { 102 if (!title_.empty()) { 103 uint64_t elapsed = Nanotime() - started_; 104 if (log_only_) { 105 MYLOGD("Duration of '%s': %.3fs\n", title_.c_str(), (float)elapsed / NANOS_PER_SEC); 106 } else { 107 // Use "Yoda grammar" to make it easier to grep|sort sections. 108 printf("------ %.3fs was the duration of '%s' ------\n", (float)elapsed / NANOS_PER_SEC, 109 title_.c_str()); 110 } 111 } 112 } 113 114 const int32_t Progress::kDefaultMax = 5000; 115 116 Progress::Progress(const std::string& path) : Progress(Progress::kDefaultMax, 1.1, path) { 117 } 118 119 Progress::Progress(int32_t initial_max, int32_t progress, float growth_factor) 120 : Progress(initial_max, growth_factor, "") { 121 progress_ = progress; 122 } 123 124 Progress::Progress(int32_t initial_max, float growth_factor, const std::string& path) 125 : initial_max_(initial_max), 126 progress_(0), 127 max_(initial_max), 128 growth_factor_(growth_factor), 129 n_runs_(0), 130 average_max_(0), 131 path_(path) { 132 if (!path_.empty()) { 133 Load(); 134 } 135 } 136 137 void Progress::Load() { 138 MYLOGD("Loading stats from %s\n", path_.c_str()); 139 std::string content; 140 if (!android::base::ReadFileToString(path_, &content)) { 141 MYLOGI("Could not read stats from %s; using max of %d\n", path_.c_str(), max_); 142 return; 143 } 144 if (content.empty()) { 145 MYLOGE("No stats (empty file) on %s; using max of %d\n", path_.c_str(), max_); 146 return; 147 } 148 std::vector<std::string> lines = android::base::Split(content, "\n"); 149 150 if (lines.size() < 1) { 151 MYLOGE("Invalid stats on file %s: not enough lines (%d). Using max of %d\n", path_.c_str(), 152 (int)lines.size(), max_); 153 return; 154 } 155 char* ptr; 156 n_runs_ = strtol(lines[0].c_str(), &ptr, 10); 157 average_max_ = strtol(ptr, nullptr, 10); 158 if (n_runs_ <= 0 || average_max_ <= 0 || n_runs_ > STATS_MAX_N_RUNS || 159 average_max_ > STATS_MAX_AVERAGE) { 160 MYLOGE("Invalid stats line on file %s: %s\n", path_.c_str(), lines[0].c_str()); 161 initial_max_ = Progress::kDefaultMax; 162 } else { 163 initial_max_ = average_max_; 164 } 165 max_ = initial_max_; 166 167 MYLOGI("Average max progress: %d in %d runs; estimated max: %d\n", average_max_, n_runs_, max_); 168 } 169 170 void Progress::Save() { 171 int32_t total = n_runs_ * average_max_ + progress_; 172 int32_t runs = n_runs_ + 1; 173 int32_t average = floor(((float)total) / runs); 174 MYLOGI("Saving stats (total=%d, runs=%d, average=%d) on %s\n", total, runs, average, 175 path_.c_str()); 176 if (path_.empty()) { 177 return; 178 } 179 180 std::string content = android::base::StringPrintf("%d %d\n", runs, average); 181 if (!android::base::WriteStringToFile(content, path_)) { 182 MYLOGE("Could not save stats on %s\n", path_.c_str()); 183 } 184 } 185 186 int32_t Progress::Get() const { 187 return progress_; 188 } 189 190 bool Progress::Inc(int32_t delta_sec) { 191 bool changed = false; 192 if (delta_sec >= 0) { 193 progress_ += delta_sec; 194 if (progress_ > max_) { 195 int32_t old_max = max_; 196 max_ = floor((float)progress_ * growth_factor_); 197 MYLOGD("Adjusting max progress from %d to %d\n", old_max, max_); 198 changed = true; 199 } 200 } 201 return changed; 202 } 203 204 int32_t Progress::GetMax() const { 205 return max_; 206 } 207 208 int32_t Progress::GetInitialMax() const { 209 return initial_max_; 210 } 211 212 void Progress::Dump(int fd, const std::string& prefix) const { 213 const char* pr = prefix.c_str(); 214 dprintf(fd, "%sprogress: %d\n", pr, progress_); 215 dprintf(fd, "%smax: %d\n", pr, max_); 216 dprintf(fd, "%sinitial_max: %d\n", pr, initial_max_); 217 dprintf(fd, "%sgrowth_factor: %0.2f\n", pr, growth_factor_); 218 dprintf(fd, "%spath: %s\n", pr, path_.c_str()); 219 dprintf(fd, "%sn_runs: %d\n", pr, n_runs_); 220 dprintf(fd, "%saverage_max: %d\n", pr, average_max_); 221 } 222 223 bool Dumpstate::IsZipping() const { 224 return zip_writer_ != nullptr; 225 } 226 227 std::string Dumpstate::GetPath(const std::string& suffix) const { 228 return android::base::StringPrintf("%s/%s-%s%s", bugreport_dir_.c_str(), base_name_.c_str(), 229 name_.c_str(), suffix.c_str()); 230 } 231 232 void Dumpstate::SetProgress(std::unique_ptr<Progress> progress) { 233 progress_ = std::move(progress); 234 } 235 236 void for_each_userid(void (*func)(int), const char *header) { 237 std::string title = header == nullptr ? "for_each_userid" : android::base::StringPrintf( 238 "for_each_userid(%s)", header); 239 DurationReporter duration_reporter(title); 240 if (PropertiesHelper::IsDryRun()) return; 241 242 DIR *d; 243 struct dirent *de; 244 245 if (header) printf("\n------ %s ------\n", header); 246 func(0); 247 248 if (!(d = opendir("/data/system/users"))) { 249 printf("Failed to open /data/system/users (%s)\n", strerror(errno)); 250 return; 251 } 252 253 while ((de = readdir(d))) { 254 int userid; 255 if (de->d_type != DT_DIR || !(userid = atoi(de->d_name))) { 256 continue; 257 } 258 func(userid); 259 } 260 261 closedir(d); 262 } 263 264 static void __for_each_pid(void (*helper)(int, const char *, void *), const char *header, void *arg) { 265 DIR *d; 266 struct dirent *de; 267 268 if (!(d = opendir("/proc"))) { 269 printf("Failed to open /proc (%s)\n", strerror(errno)); 270 return; 271 } 272 273 if (header) printf("\n------ %s ------\n", header); 274 while ((de = readdir(d))) { 275 int pid; 276 int fd; 277 char cmdpath[255]; 278 char cmdline[255]; 279 280 if (!(pid = atoi(de->d_name))) { 281 continue; 282 } 283 284 memset(cmdline, 0, sizeof(cmdline)); 285 286 snprintf(cmdpath, sizeof(cmdpath), "/proc/%d/cmdline", pid); 287 if ((fd = TEMP_FAILURE_RETRY(open(cmdpath, O_RDONLY | O_CLOEXEC))) >= 0) { 288 TEMP_FAILURE_RETRY(read(fd, cmdline, sizeof(cmdline) - 2)); 289 close(fd); 290 if (cmdline[0]) { 291 helper(pid, cmdline, arg); 292 continue; 293 } 294 } 295 296 // if no cmdline, a kernel thread has comm 297 snprintf(cmdpath, sizeof(cmdpath), "/proc/%d/comm", pid); 298 if ((fd = TEMP_FAILURE_RETRY(open(cmdpath, O_RDONLY | O_CLOEXEC))) >= 0) { 299 TEMP_FAILURE_RETRY(read(fd, cmdline + 1, sizeof(cmdline) - 4)); 300 close(fd); 301 if (cmdline[1]) { 302 cmdline[0] = '['; 303 size_t len = strcspn(cmdline, "\f\b\r\n"); 304 cmdline[len] = ']'; 305 cmdline[len+1] = '\0'; 306 } 307 } 308 if (!cmdline[0]) { 309 strcpy(cmdline, "N/A"); 310 } 311 helper(pid, cmdline, arg); 312 } 313 314 closedir(d); 315 } 316 317 static void for_each_pid_helper(int pid, const char *cmdline, void *arg) { 318 for_each_pid_func *func = (for_each_pid_func*) arg; 319 func(pid, cmdline); 320 } 321 322 void for_each_pid(for_each_pid_func func, const char *header) { 323 std::string title = header == nullptr ? "for_each_pid" 324 : android::base::StringPrintf("for_each_pid(%s)", header); 325 DurationReporter duration_reporter(title); 326 if (PropertiesHelper::IsDryRun()) return; 327 328 __for_each_pid(for_each_pid_helper, header, (void *) func); 329 } 330 331 static void for_each_tid_helper(int pid, const char *cmdline, void *arg) { 332 DIR *d; 333 struct dirent *de; 334 char taskpath[255]; 335 for_each_tid_func *func = (for_each_tid_func *) arg; 336 337 snprintf(taskpath, sizeof(taskpath), "/proc/%d/task", pid); 338 339 if (!(d = opendir(taskpath))) { 340 printf("Failed to open %s (%s)\n", taskpath, strerror(errno)); 341 return; 342 } 343 344 func(pid, pid, cmdline); 345 346 while ((de = readdir(d))) { 347 int tid; 348 int fd; 349 char commpath[255]; 350 char comm[255]; 351 352 if (!(tid = atoi(de->d_name))) { 353 continue; 354 } 355 356 if (tid == pid) 357 continue; 358 359 snprintf(commpath, sizeof(commpath), "/proc/%d/comm", tid); 360 memset(comm, 0, sizeof(comm)); 361 if ((fd = TEMP_FAILURE_RETRY(open(commpath, O_RDONLY | O_CLOEXEC))) < 0) { 362 strcpy(comm, "N/A"); 363 } else { 364 char *c; 365 TEMP_FAILURE_RETRY(read(fd, comm, sizeof(comm) - 2)); 366 close(fd); 367 368 c = strrchr(comm, '\n'); 369 if (c) { 370 *c = '\0'; 371 } 372 } 373 func(pid, tid, comm); 374 } 375 376 closedir(d); 377 } 378 379 void for_each_tid(for_each_tid_func func, const char *header) { 380 std::string title = header == nullptr ? "for_each_tid" 381 : android::base::StringPrintf("for_each_tid(%s)", header); 382 DurationReporter duration_reporter(title); 383 384 if (PropertiesHelper::IsDryRun()) return; 385 386 __for_each_pid(for_each_tid_helper, header, (void *) func); 387 } 388 389 void show_wchan(int pid, int tid, const char *name) { 390 if (PropertiesHelper::IsDryRun()) return; 391 392 char path[255]; 393 char buffer[255]; 394 int fd, ret, save_errno; 395 char name_buffer[255]; 396 397 memset(buffer, 0, sizeof(buffer)); 398 399 snprintf(path, sizeof(path), "/proc/%d/wchan", tid); 400 if ((fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC))) < 0) { 401 printf("Failed to open '%s' (%s)\n", path, strerror(errno)); 402 return; 403 } 404 405 ret = TEMP_FAILURE_RETRY(read(fd, buffer, sizeof(buffer))); 406 save_errno = errno; 407 close(fd); 408 409 if (ret < 0) { 410 printf("Failed to read '%s' (%s)\n", path, strerror(save_errno)); 411 return; 412 } 413 414 snprintf(name_buffer, sizeof(name_buffer), "%*s%s", 415 pid == tid ? 0 : 3, "", name); 416 417 printf("%-7d %-32s %s\n", tid, name_buffer, buffer); 418 419 return; 420 } 421 422 // print time in centiseconds 423 static void snprcent(char *buffer, size_t len, size_t spc, 424 unsigned long long time) { 425 static long hz; // cache discovered hz 426 427 if (hz <= 0) { 428 hz = sysconf(_SC_CLK_TCK); 429 if (hz <= 0) { 430 hz = 1000; 431 } 432 } 433 434 // convert to centiseconds 435 time = (time * 100 + (hz / 2)) / hz; 436 437 char str[16]; 438 439 snprintf(str, sizeof(str), " %llu.%02u", 440 time / 100, (unsigned)(time % 100)); 441 size_t offset = strlen(buffer); 442 snprintf(buffer + offset, (len > offset) ? len - offset : 0, 443 "%*s", (spc > offset) ? (int)(spc - offset) : 0, str); 444 } 445 446 // print permille as a percent 447 static void snprdec(char *buffer, size_t len, size_t spc, unsigned permille) { 448 char str[16]; 449 450 snprintf(str, sizeof(str), " %u.%u%%", permille / 10, permille % 10); 451 size_t offset = strlen(buffer); 452 snprintf(buffer + offset, (len > offset) ? len - offset : 0, 453 "%*s", (spc > offset) ? (int)(spc - offset) : 0, str); 454 } 455 456 void show_showtime(int pid, const char *name) { 457 if (PropertiesHelper::IsDryRun()) return; 458 459 char path[255]; 460 char buffer[1023]; 461 int fd, ret, save_errno; 462 463 memset(buffer, 0, sizeof(buffer)); 464 465 snprintf(path, sizeof(path), "/proc/%d/stat", pid); 466 if ((fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC))) < 0) { 467 printf("Failed to open '%s' (%s)\n", path, strerror(errno)); 468 return; 469 } 470 471 ret = TEMP_FAILURE_RETRY(read(fd, buffer, sizeof(buffer))); 472 save_errno = errno; 473 close(fd); 474 475 if (ret < 0) { 476 printf("Failed to read '%s' (%s)\n", path, strerror(save_errno)); 477 return; 478 } 479 480 // field 14 is utime 481 // field 15 is stime 482 // field 42 is iotime 483 unsigned long long utime = 0, stime = 0, iotime = 0; 484 if (sscanf(buffer, 485 "%*u %*s %*s %*d %*d %*d %*d %*d %*d %*d %*d " 486 "%*d %*d %llu %llu %*d %*d %*d %*d %*d %*d " 487 "%*d %*d %*d %*d %*d %*d %*d %*d %*d %*d " 488 "%*d %*d %*d %*d %*d %*d %*d %*d %*d %llu ", 489 &utime, &stime, &iotime) != 3) { 490 return; 491 } 492 493 unsigned long long total = utime + stime; 494 if (!total) { 495 return; 496 } 497 498 unsigned permille = (iotime * 1000 + (total / 2)) / total; 499 if (permille > 1000) { 500 permille = 1000; 501 } 502 503 // try to beautify and stabilize columns at <80 characters 504 snprintf(buffer, sizeof(buffer), "%-6d%s", pid, name); 505 if ((name[0] != '[') || utime) { 506 snprcent(buffer, sizeof(buffer), 57, utime); 507 } 508 snprcent(buffer, sizeof(buffer), 65, stime); 509 if ((name[0] != '[') || iotime) { 510 snprcent(buffer, sizeof(buffer), 73, iotime); 511 } 512 if (iotime) { 513 snprdec(buffer, sizeof(buffer), 79, permille); 514 } 515 puts(buffer); // adds a trailing newline 516 517 return; 518 } 519 520 void do_dmesg() { 521 const char *title = "KERNEL LOG (dmesg)"; 522 DurationReporter duration_reporter(title); 523 printf("------ %s ------\n", title); 524 525 if (PropertiesHelper::IsDryRun()) return; 526 527 /* Get size of kernel buffer */ 528 int size = klogctl(KLOG_SIZE_BUFFER, NULL, 0); 529 if (size <= 0) { 530 printf("Unexpected klogctl return value: %d\n\n", size); 531 return; 532 } 533 char *buf = (char *) malloc(size + 1); 534 if (buf == NULL) { 535 printf("memory allocation failed\n\n"); 536 return; 537 } 538 int retval = klogctl(KLOG_READ_ALL, buf, size); 539 if (retval < 0) { 540 printf("klogctl failure\n\n"); 541 free(buf); 542 return; 543 } 544 buf[retval] = '\0'; 545 printf("%s\n\n", buf); 546 free(buf); 547 return; 548 } 549 550 void do_showmap(int pid, const char *name) { 551 char title[255]; 552 char arg[255]; 553 554 snprintf(title, sizeof(title), "SHOW MAP %d (%s)", pid, name); 555 snprintf(arg, sizeof(arg), "%d", pid); 556 RunCommand(title, {"showmap", "-q", arg}, CommandOptions::AS_ROOT); 557 } 558 559 int Dumpstate::DumpFile(const std::string& title, const std::string& path) { 560 DurationReporter duration_reporter(title); 561 562 int status = DumpFileToFd(STDOUT_FILENO, title, path); 563 564 UpdateProgress(WEIGHT_FILE); 565 566 return status; 567 } 568 569 int read_file_as_long(const char *path, long int *output) { 570 int fd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_NONBLOCK | O_CLOEXEC)); 571 if (fd < 0) { 572 int err = errno; 573 MYLOGE("Error opening file descriptor for %s: %s\n", path, strerror(err)); 574 return -1; 575 } 576 char buffer[50]; 577 ssize_t bytes_read = TEMP_FAILURE_RETRY(read(fd, buffer, sizeof(buffer))); 578 if (bytes_read == -1) { 579 MYLOGE("Error reading file %s: %s\n", path, strerror(errno)); 580 return -2; 581 } 582 if (bytes_read == 0) { 583 MYLOGE("File %s is empty\n", path); 584 return -3; 585 } 586 *output = atoi(buffer); 587 return 0; 588 } 589 590 /* calls skip to gate calling dump_from_fd recursively 591 * in the specified directory. dump_from_fd defaults to 592 * dump_file_from_fd above when set to NULL. skip defaults 593 * to false when set to NULL. dump_from_fd will always be 594 * called with title NULL. 595 */ 596 int dump_files(const std::string& title, const char* dir, bool (*skip)(const char* path), 597 int (*dump_from_fd)(const char* title, const char* path, int fd)) { 598 DurationReporter duration_reporter(title); 599 DIR *dirp; 600 struct dirent *d; 601 char *newpath = NULL; 602 const char *slash = "/"; 603 int retval = 0; 604 605 if (!title.empty()) { 606 printf("------ %s (%s) ------\n", title.c_str(), dir); 607 } 608 if (PropertiesHelper::IsDryRun()) return 0; 609 610 if (dir[strlen(dir) - 1] == '/') { 611 ++slash; 612 } 613 dirp = opendir(dir); 614 if (dirp == NULL) { 615 retval = -errno; 616 MYLOGE("%s: %s\n", dir, strerror(errno)); 617 return retval; 618 } 619 620 if (!dump_from_fd) { 621 dump_from_fd = dump_file_from_fd; 622 } 623 for (; ((d = readdir(dirp))); free(newpath), newpath = NULL) { 624 if ((d->d_name[0] == '.') 625 && (((d->d_name[1] == '.') && (d->d_name[2] == '\0')) 626 || (d->d_name[1] == '\0'))) { 627 continue; 628 } 629 asprintf(&newpath, "%s%s%s%s", dir, slash, d->d_name, 630 (d->d_type == DT_DIR) ? "/" : ""); 631 if (!newpath) { 632 retval = -errno; 633 continue; 634 } 635 if (skip && (*skip)(newpath)) { 636 continue; 637 } 638 if (d->d_type == DT_DIR) { 639 int ret = dump_files("", newpath, skip, dump_from_fd); 640 if (ret < 0) { 641 retval = ret; 642 } 643 continue; 644 } 645 android::base::unique_fd fd(TEMP_FAILURE_RETRY(open(newpath, O_RDONLY | O_NONBLOCK | O_CLOEXEC))); 646 if (fd.get() < 0) { 647 retval = -1; 648 printf("*** %s: %s\n", newpath, strerror(errno)); 649 continue; 650 } 651 (*dump_from_fd)(NULL, newpath, fd.get()); 652 } 653 closedir(dirp); 654 if (!title.empty()) { 655 printf("\n"); 656 } 657 return retval; 658 } 659 660 /* fd must have been opened with the flag O_NONBLOCK. With this flag set, 661 * it's possible to avoid issues where opening the file itself can get 662 * stuck. 663 */ 664 int dump_file_from_fd(const char *title, const char *path, int fd) { 665 if (PropertiesHelper::IsDryRun()) return 0; 666 667 int flags = fcntl(fd, F_GETFL); 668 if (flags == -1) { 669 printf("*** %s: failed to get flags on fd %d: %s\n", path, fd, strerror(errno)); 670 return -1; 671 } else if (!(flags & O_NONBLOCK)) { 672 printf("*** %s: fd must have O_NONBLOCK set.\n", path); 673 return -1; 674 } 675 return DumpFileFromFdToFd(title, path, fd, STDOUT_FILENO, PropertiesHelper::IsDryRun()); 676 } 677 678 int Dumpstate::RunCommand(const std::string& title, const std::vector<std::string>& full_command, 679 const CommandOptions& options) { 680 DurationReporter duration_reporter(title); 681 682 int status = RunCommandToFd(STDOUT_FILENO, title, full_command, options); 683 684 /* TODO: for now we're simplifying the progress calculation by using the 685 * timeout as the weight. It's a good approximation for most cases, except when calling dumpsys, 686 * where its weight should be much higher proportionally to its timeout. 687 * Ideally, it should use a options.EstimatedDuration() instead...*/ 688 UpdateProgress(options.Timeout()); 689 690 return status; 691 } 692 693 void Dumpstate::RunDumpsys(const std::string& title, const std::vector<std::string>& dumpsys_args, 694 const CommandOptions& options, long dumpsysTimeoutMs) { 695 long timeout_ms = dumpsysTimeoutMs > 0 ? dumpsysTimeoutMs : options.TimeoutInMs(); 696 std::vector<std::string> dumpsys = {"/system/bin/dumpsys", "-T", std::to_string(timeout_ms)}; 697 dumpsys.insert(dumpsys.end(), dumpsys_args.begin(), dumpsys_args.end()); 698 RunCommand(title, dumpsys, options); 699 } 700 701 int open_socket(const char *service) { 702 int s = android_get_control_socket(service); 703 if (s < 0) { 704 MYLOGE("android_get_control_socket(%s): %s\n", service, strerror(errno)); 705 exit(1); 706 } 707 fcntl(s, F_SETFD, FD_CLOEXEC); 708 if (listen(s, 4) < 0) { 709 MYLOGE("listen(control socket): %s\n", strerror(errno)); 710 exit(1); 711 } 712 713 struct sockaddr addr; 714 socklen_t alen = sizeof(addr); 715 int fd = accept(s, &addr, &alen); 716 if (fd < 0) { 717 MYLOGE("accept(control socket): %s\n", strerror(errno)); 718 exit(1); 719 } 720 721 return fd; 722 } 723 724 /* redirect output to a service control socket */ 725 void redirect_to_socket(FILE *redirect, const char *service) { 726 int fd = open_socket(service); 727 fflush(redirect); 728 dup2(fd, fileno(redirect)); 729 close(fd); 730 } 731 732 // TODO: should call is_valid_output_file and/or be merged into it. 733 void create_parent_dirs(const char *path) { 734 char *chp = const_cast<char *> (path); 735 736 /* skip initial slash */ 737 if (chp[0] == '/') 738 chp++; 739 740 /* create leading directories, if necessary */ 741 struct stat dir_stat; 742 while (chp && chp[0]) { 743 chp = strchr(chp, '/'); 744 if (chp) { 745 *chp = 0; 746 if (stat(path, &dir_stat) == -1 || !S_ISDIR(dir_stat.st_mode)) { 747 MYLOGI("Creating directory %s\n", path); 748 if (mkdir(path, 0770)) { /* drwxrwx--- */ 749 MYLOGE("Unable to create directory %s: %s\n", path, strerror(errno)); 750 } else if (chown(path, AID_SHELL, AID_SHELL)) { 751 MYLOGE("Unable to change ownership of dir %s: %s\n", path, strerror(errno)); 752 } 753 } 754 *chp++ = '/'; 755 } 756 } 757 } 758 759 void _redirect_to_file(FILE *redirect, char *path, int truncate_flag) { 760 create_parent_dirs(path); 761 762 int fd = TEMP_FAILURE_RETRY(open(path, 763 O_WRONLY | O_CREAT | truncate_flag | O_CLOEXEC | O_NOFOLLOW, 764 S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH)); 765 if (fd < 0) { 766 MYLOGE("%s: %s\n", path, strerror(errno)); 767 exit(1); 768 } 769 770 TEMP_FAILURE_RETRY(dup2(fd, fileno(redirect))); 771 close(fd); 772 } 773 774 void redirect_to_file(FILE *redirect, char *path) { 775 _redirect_to_file(redirect, path, O_TRUNC); 776 } 777 778 void redirect_to_existing_file(FILE *redirect, char *path) { 779 _redirect_to_file(redirect, path, O_APPEND); 780 } 781 782 const char* DumpTraces(const std::string& traces_path); 783 const char* DumpTracesTombstoned(const std::string& traces_dir); 784 785 /* dump Dalvik and native stack traces, return the trace file location (NULL if none) */ 786 const char *dump_traces() { 787 DurationReporter duration_reporter("DUMP TRACES"); 788 789 const std::string traces_dir = android::base::GetProperty("dalvik.vm.stack-trace-dir", ""); 790 if (!traces_dir.empty()) { 791 return DumpTracesTombstoned(traces_dir); 792 } 793 794 const std::string traces_file = android::base::GetProperty("dalvik.vm.stack-trace-file", ""); 795 if (!traces_file.empty()) { 796 return DumpTraces(traces_file); 797 } 798 799 return nullptr; 800 } 801 802 const char* DumpTracesTombstoned(const std::string& traces_dir) { 803 const std::string temp_file_pattern = traces_dir + "/dumptrace_XXXXXX"; 804 805 const size_t buf_size = temp_file_pattern.length() + 1; 806 std::unique_ptr<char[]> file_name_buf(new char[buf_size]); 807 memcpy(file_name_buf.get(), temp_file_pattern.c_str(), buf_size); 808 809 // Create a new, empty file to receive all trace dumps. 810 // 811 // TODO: This can be simplified once we remove support for the old style 812 // dumps. We can have a file descriptor passed in to dump_traces instead 813 // of creating a file, closing it and then reopening it again. 814 android::base::unique_fd fd(mkostemp(file_name_buf.get(), O_APPEND | O_CLOEXEC)); 815 if (fd < 0) { 816 MYLOGE("mkostemp on pattern %s: %s\n", file_name_buf.get(), strerror(errno)); 817 return nullptr; 818 } 819 820 // Nobody should have access to this temporary file except dumpstate, but we 821 // temporarily grant 'read' to 'others' here because this file is created 822 // when tombstoned is still running as root, but dumped after dropping. This 823 // can go away once support for old style dumping has. 824 const int chmod_ret = fchmod(fd, 0666); 825 if (chmod_ret < 0) { 826 MYLOGE("fchmod on %s failed: %s\n", file_name_buf.get(), strerror(errno)); 827 return nullptr; 828 } 829 830 std::unique_ptr<DIR, decltype(&closedir)> proc(opendir("/proc"), closedir); 831 if (proc.get() == nullptr) { 832 MYLOGE("opendir /proc failed: %s\n", strerror(errno)); 833 return nullptr; 834 } 835 836 // Number of times process dumping has timed out. If we encounter too many 837 // failures, we'll give up. 838 int timeout_failures = 0; 839 bool dalvik_found = false; 840 841 const std::set<int> hal_pids = get_interesting_hal_pids(); 842 843 struct dirent* d; 844 while ((d = readdir(proc.get()))) { 845 int pid = atoi(d->d_name); 846 if (pid <= 0) { 847 continue; 848 } 849 850 const std::string link_name = android::base::StringPrintf("/proc/%d/exe", pid); 851 std::string exe; 852 if (!android::base::Readlink(link_name, &exe)) { 853 continue; 854 } 855 856 bool is_java_process; 857 if (exe == "/system/bin/app_process32" || exe == "/system/bin/app_process64") { 858 // Don't bother dumping backtraces for the zygote. 859 if (IsZygote(pid)) { 860 continue; 861 } 862 863 dalvik_found = true; 864 is_java_process = true; 865 } else if (should_dump_native_traces(exe.c_str()) || hal_pids.find(pid) != hal_pids.end()) { 866 is_java_process = false; 867 } else { 868 // Probably a native process we don't care about, continue. 869 continue; 870 } 871 872 // If 3 backtrace dumps fail in a row, consider debuggerd dead. 873 if (timeout_failures == 3) { 874 dprintf(fd, "ERROR: Too many stack dump failures, exiting.\n"); 875 break; 876 } 877 878 const uint64_t start = Nanotime(); 879 const int ret = dump_backtrace_to_file_timeout( 880 pid, is_java_process ? kDebuggerdJavaBacktrace : kDebuggerdNativeBacktrace, 881 is_java_process ? 5 : 20, fd); 882 883 if (ret == -1) { 884 dprintf(fd, "dumping failed, likely due to a timeout\n"); 885 timeout_failures++; 886 continue; 887 } 888 889 // We've successfully dumped stack traces, reset the failure count 890 // and write a summary of the elapsed time to the file and continue with the 891 // next process. 892 timeout_failures = 0; 893 894 dprintf(fd, "[dump %s stack %d: %.3fs elapsed]\n", is_java_process ? "dalvik" : "native", 895 pid, (float)(Nanotime() - start) / NANOS_PER_SEC); 896 } 897 898 if (!dalvik_found) { 899 MYLOGE("Warning: no Dalvik processes found to dump stacks\n"); 900 } 901 902 return file_name_buf.release(); 903 } 904 905 const char* DumpTraces(const std::string& traces_path) { 906 const char* result = NULL; 907 /* move the old traces.txt (if any) out of the way temporarily */ 908 std::string anrtraces_path = traces_path + ".anr"; 909 if (rename(traces_path.c_str(), anrtraces_path.c_str()) && errno != ENOENT) { 910 MYLOGE("rename(%s, %s): %s\n", traces_path.c_str(), anrtraces_path.c_str(), strerror(errno)); 911 return nullptr; // Can't rename old traces.txt -- no permission? -- leave it alone instead 912 } 913 914 /* create a new, empty traces.txt file to receive stack dumps */ 915 int fd = TEMP_FAILURE_RETRY( 916 open(traces_path.c_str(), O_CREAT | O_WRONLY | O_APPEND | O_TRUNC | O_NOFOLLOW | O_CLOEXEC, 917 0666)); /* -rw-rw-rw- */ 918 if (fd < 0) { 919 MYLOGE("%s: %s\n", traces_path.c_str(), strerror(errno)); 920 return nullptr; 921 } 922 int chmod_ret = fchmod(fd, 0666); 923 if (chmod_ret < 0) { 924 MYLOGE("fchmod on %s failed: %s\n", traces_path.c_str(), strerror(errno)); 925 close(fd); 926 return nullptr; 927 } 928 929 /* Variables below must be initialized before 'goto' statements */ 930 int dalvik_found = 0; 931 int ifd, wfd = -1; 932 std::set<int> hal_pids = get_interesting_hal_pids(); 933 934 /* walk /proc and kill -QUIT all Dalvik processes */ 935 DIR *proc = opendir("/proc"); 936 if (proc == NULL) { 937 MYLOGE("/proc: %s\n", strerror(errno)); 938 goto error_close_fd; 939 } 940 941 /* use inotify to find when processes are done dumping */ 942 ifd = inotify_init(); 943 if (ifd < 0) { 944 MYLOGE("inotify_init: %s\n", strerror(errno)); 945 goto error_close_fd; 946 } 947 948 wfd = inotify_add_watch(ifd, traces_path.c_str(), IN_CLOSE_WRITE); 949 if (wfd < 0) { 950 MYLOGE("inotify_add_watch(%s): %s\n", traces_path.c_str(), strerror(errno)); 951 goto error_close_ifd; 952 } 953 954 struct dirent *d; 955 while ((d = readdir(proc))) { 956 int pid = atoi(d->d_name); 957 if (pid <= 0) continue; 958 959 char path[PATH_MAX]; 960 char data[PATH_MAX]; 961 snprintf(path, sizeof(path), "/proc/%d/exe", pid); 962 ssize_t len = readlink(path, data, sizeof(data) - 1); 963 if (len <= 0) { 964 continue; 965 } 966 data[len] = '\0'; 967 968 if (!strncmp(data, "/system/bin/app_process", strlen("/system/bin/app_process"))) { 969 /* skip zygote -- it won't dump its stack anyway */ 970 snprintf(path, sizeof(path), "/proc/%d/cmdline", pid); 971 int cfd = TEMP_FAILURE_RETRY(open(path, O_RDONLY | O_CLOEXEC)); 972 len = read(cfd, data, sizeof(data) - 1); 973 close(cfd); 974 if (len <= 0) { 975 continue; 976 } 977 data[len] = '\0'; 978 if (!strncmp(data, "zygote", strlen("zygote"))) { 979 continue; 980 } 981 982 ++dalvik_found; 983 uint64_t start = Nanotime(); 984 if (kill(pid, SIGQUIT)) { 985 MYLOGE("kill(%d, SIGQUIT): %s\n", pid, strerror(errno)); 986 continue; 987 } 988 989 /* wait for the writable-close notification from inotify */ 990 struct pollfd pfd = { ifd, POLLIN, 0 }; 991 int ret = poll(&pfd, 1, TRACE_DUMP_TIMEOUT_MS); 992 if (ret < 0) { 993 MYLOGE("poll: %s\n", strerror(errno)); 994 } else if (ret == 0) { 995 MYLOGE("warning: timed out dumping pid %d\n", pid); 996 } else { 997 struct inotify_event ie; 998 read(ifd, &ie, sizeof(ie)); 999 } 1000 1001 if (lseek(fd, 0, SEEK_END) < 0) { 1002 MYLOGE("lseek: %s\n", strerror(errno)); 1003 } else { 1004 dprintf(fd, "[dump dalvik stack %d: %.3fs elapsed]\n", pid, 1005 (float)(Nanotime() - start) / NANOS_PER_SEC); 1006 } 1007 } else if (should_dump_native_traces(data) || 1008 hal_pids.find(pid) != hal_pids.end()) { 1009 /* dump native process if appropriate */ 1010 if (lseek(fd, 0, SEEK_END) < 0) { 1011 MYLOGE("lseek: %s\n", strerror(errno)); 1012 } else { 1013 static uint16_t timeout_failures = 0; 1014 uint64_t start = Nanotime(); 1015 1016 /* If 3 backtrace dumps fail in a row, consider debuggerd dead. */ 1017 if (timeout_failures == 3) { 1018 dprintf(fd, "too many stack dump failures, skipping...\n"); 1019 } else if (dump_backtrace_to_file_timeout( 1020 pid, kDebuggerdNativeBacktrace, 20, fd) == -1) { 1021 dprintf(fd, "dumping failed, likely due to a timeout\n"); 1022 timeout_failures++; 1023 } else { 1024 timeout_failures = 0; 1025 } 1026 dprintf(fd, "[dump native stack %d: %.3fs elapsed]\n", pid, 1027 (float)(Nanotime() - start) / NANOS_PER_SEC); 1028 } 1029 } 1030 } 1031 1032 if (dalvik_found == 0) { 1033 MYLOGE("Warning: no Dalvik processes found to dump stacks\n"); 1034 } 1035 1036 static std::string dumptraces_path = android::base::StringPrintf( 1037 "%s/bugreport-%s", dirname(traces_path.c_str()), basename(traces_path.c_str())); 1038 if (rename(traces_path.c_str(), dumptraces_path.c_str())) { 1039 MYLOGE("rename(%s, %s): %s\n", traces_path.c_str(), dumptraces_path.c_str(), 1040 strerror(errno)); 1041 goto error_close_ifd; 1042 } 1043 result = dumptraces_path.c_str(); 1044 1045 /* replace the saved [ANR] traces.txt file */ 1046 rename(anrtraces_path.c_str(), traces_path.c_str()); 1047 1048 error_close_ifd: 1049 close(ifd); 1050 error_close_fd: 1051 close(fd); 1052 return result; 1053 } 1054 1055 void dump_route_tables() { 1056 DurationReporter duration_reporter("DUMP ROUTE TABLES"); 1057 if (PropertiesHelper::IsDryRun()) return; 1058 const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables"; 1059 ds.DumpFile("RT_TABLES", RT_TABLES_PATH); 1060 FILE* fp = fopen(RT_TABLES_PATH, "re"); 1061 if (!fp) { 1062 printf("*** %s: %s\n", RT_TABLES_PATH, strerror(errno)); 1063 return; 1064 } 1065 char table[16]; 1066 // Each line has an integer (the table number), a space, and a string (the table name). We only 1067 // need the table number. It's a 32-bit unsigned number, so max 10 chars. Skip the table name. 1068 // Add a fixed max limit so this doesn't go awry. 1069 for (int i = 0; i < 64 && fscanf(fp, " %10s %*s", table) == 1; ++i) { 1070 RunCommand("ROUTE TABLE IPv4", {"ip", "-4", "route", "show", "table", table}); 1071 RunCommand("ROUTE TABLE IPv6", {"ip", "-6", "route", "show", "table", table}); 1072 } 1073 fclose(fp); 1074 } 1075 1076 // TODO: make this function thread safe if sections are generated in parallel. 1077 void Dumpstate::UpdateProgress(int32_t delta_sec) { 1078 if (progress_ == nullptr) { 1079 MYLOGE("UpdateProgress: progress_ not set\n"); 1080 return; 1081 } 1082 1083 // Always update progess so stats can be tuned... 1084 bool max_changed = progress_->Inc(delta_sec); 1085 1086 // ...but only notifiy listeners when necessary. 1087 if (!update_progress_) return; 1088 1089 int progress = progress_->Get(); 1090 int max = progress_->GetMax(); 1091 1092 // adjusts max on the fly 1093 if (max_changed && listener_ != nullptr) { 1094 listener_->onMaxProgressUpdated(max); 1095 } 1096 1097 int32_t last_update_delta = progress - last_updated_progress_; 1098 if (last_updated_progress_ > 0 && last_update_delta < update_progress_threshold_) { 1099 return; 1100 } 1101 last_updated_progress_ = progress; 1102 1103 if (control_socket_fd_ >= 0) { 1104 dprintf(control_socket_fd_, "PROGRESS:%d/%d\n", progress, max); 1105 fsync(control_socket_fd_); 1106 } 1107 1108 if (listener_ != nullptr) { 1109 if (progress % 100 == 0) { 1110 // We don't want to spam logcat, so only log multiples of 100. 1111 MYLOGD("Setting progress (%s): %d/%d\n", listener_name_.c_str(), progress, max); 1112 } else { 1113 // stderr is ignored on normal invocations, but useful when calling 1114 // /system/bin/dumpstate directly for debuggging. 1115 fprintf(stderr, "Setting progress (%s): %d/%d\n", listener_name_.c_str(), progress, max); 1116 } 1117 listener_->onProgressUpdated(progress); 1118 } 1119 } 1120 1121 void Dumpstate::TakeScreenshot(const std::string& path) { 1122 const std::string& real_path = path.empty() ? screenshot_path_ : path; 1123 int status = 1124 RunCommand("", {"/system/bin/screencap", "-p", real_path}, 1125 CommandOptions::WithTimeout(10).Always().DropRoot().RedirectStderr().Build()); 1126 if (status == 0) { 1127 MYLOGD("Screenshot saved on %s\n", real_path.c_str()); 1128 } else { 1129 MYLOGE("Failed to take screenshot on %s\n", real_path.c_str()); 1130 } 1131 } 1132 1133 bool is_dir(const char* pathname) { 1134 struct stat info; 1135 if (stat(pathname, &info) == -1) { 1136 return false; 1137 } 1138 return S_ISDIR(info.st_mode); 1139 } 1140 1141 time_t get_mtime(int fd, time_t default_mtime) { 1142 struct stat info; 1143 if (fstat(fd, &info) == -1) { 1144 return default_mtime; 1145 } 1146 return info.st_mtime; 1147 } 1148 1149 void dump_emmc_ecsd(const char *ext_csd_path) { 1150 // List of interesting offsets 1151 struct hex { 1152 char str[2]; 1153 }; 1154 static const size_t EXT_CSD_REV = 192 * sizeof(hex); 1155 static const size_t EXT_PRE_EOL_INFO = 267 * sizeof(hex); 1156 static const size_t EXT_DEVICE_LIFE_TIME_EST_TYP_A = 268 * sizeof(hex); 1157 static const size_t EXT_DEVICE_LIFE_TIME_EST_TYP_B = 269 * sizeof(hex); 1158 1159 std::string buffer; 1160 if (!android::base::ReadFileToString(ext_csd_path, &buffer)) { 1161 return; 1162 } 1163 1164 printf("------ %s Extended CSD ------\n", ext_csd_path); 1165 1166 if (buffer.length() < (EXT_CSD_REV + sizeof(hex))) { 1167 printf("*** %s: truncated content %zu\n\n", ext_csd_path, buffer.length()); 1168 return; 1169 } 1170 1171 int ext_csd_rev = 0; 1172 std::string sub = buffer.substr(EXT_CSD_REV, sizeof(hex)); 1173 if (sscanf(sub.c_str(), "%2x", &ext_csd_rev) != 1) { 1174 printf("*** %s: EXT_CSD_REV parse error \"%s\"\n\n", ext_csd_path, sub.c_str()); 1175 return; 1176 } 1177 1178 static const char *ver_str[] = { 1179 "4.0", "4.1", "4.2", "4.3", "Obsolete", "4.41", "4.5", "5.0" 1180 }; 1181 printf("rev 1.%d (MMC %s)\n", ext_csd_rev, 1182 (ext_csd_rev < (int)(sizeof(ver_str) / sizeof(ver_str[0]))) ? ver_str[ext_csd_rev] 1183 : "Unknown"); 1184 if (ext_csd_rev < 7) { 1185 printf("\n"); 1186 return; 1187 } 1188 1189 if (buffer.length() < (EXT_PRE_EOL_INFO + sizeof(hex))) { 1190 printf("*** %s: truncated content %zu\n\n", ext_csd_path, buffer.length()); 1191 return; 1192 } 1193 1194 int ext_pre_eol_info = 0; 1195 sub = buffer.substr(EXT_PRE_EOL_INFO, sizeof(hex)); 1196 if (sscanf(sub.c_str(), "%2x", &ext_pre_eol_info) != 1) { 1197 printf("*** %s: PRE_EOL_INFO parse error \"%s\"\n\n", ext_csd_path, sub.c_str()); 1198 return; 1199 } 1200 1201 static const char *eol_str[] = { 1202 "Undefined", 1203 "Normal", 1204 "Warning (consumed 80% of reserve)", 1205 "Urgent (consumed 90% of reserve)" 1206 }; 1207 printf( 1208 "PRE_EOL_INFO %d (MMC %s)\n", ext_pre_eol_info, 1209 eol_str[(ext_pre_eol_info < (int)(sizeof(eol_str) / sizeof(eol_str[0]))) ? ext_pre_eol_info 1210 : 0]); 1211 1212 for (size_t lifetime = EXT_DEVICE_LIFE_TIME_EST_TYP_A; 1213 lifetime <= EXT_DEVICE_LIFE_TIME_EST_TYP_B; 1214 lifetime += sizeof(hex)) { 1215 int ext_device_life_time_est; 1216 static const char *est_str[] = { 1217 "Undefined", 1218 "0-10% of device lifetime used", 1219 "10-20% of device lifetime used", 1220 "20-30% of device lifetime used", 1221 "30-40% of device lifetime used", 1222 "40-50% of device lifetime used", 1223 "50-60% of device lifetime used", 1224 "60-70% of device lifetime used", 1225 "70-80% of device lifetime used", 1226 "80-90% of device lifetime used", 1227 "90-100% of device lifetime used", 1228 "Exceeded the maximum estimated device lifetime", 1229 }; 1230 1231 if (buffer.length() < (lifetime + sizeof(hex))) { 1232 printf("*** %s: truncated content %zu\n", ext_csd_path, buffer.length()); 1233 break; 1234 } 1235 1236 ext_device_life_time_est = 0; 1237 sub = buffer.substr(lifetime, sizeof(hex)); 1238 if (sscanf(sub.c_str(), "%2x", &ext_device_life_time_est) != 1) { 1239 printf("*** %s: DEVICE_LIFE_TIME_EST_TYP_%c parse error \"%s\"\n", ext_csd_path, 1240 (unsigned)((lifetime - EXT_DEVICE_LIFE_TIME_EST_TYP_A) / sizeof(hex)) + 'A', 1241 sub.c_str()); 1242 continue; 1243 } 1244 printf("DEVICE_LIFE_TIME_EST_TYP_%c %d (MMC %s)\n", 1245 (unsigned)((lifetime - EXT_DEVICE_LIFE_TIME_EST_TYP_A) / sizeof(hex)) + 'A', 1246 ext_device_life_time_est, 1247 est_str[(ext_device_life_time_est < (int)(sizeof(est_str) / sizeof(est_str[0]))) 1248 ? ext_device_life_time_est 1249 : 0]); 1250 } 1251 1252 printf("\n"); 1253 } 1254