1 /*
2 * Copyright (C) 2016 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 // The bootstat command provides options to persist boot events with the current
18 // timestamp, dump the persisted events, and log all events to EventLog to be
19 // uploaded to Android log storage via Tron.
20
21 #include <getopt.h>
22 #include <sys/klog.h>
23 #include <unistd.h>
24
25 #include <chrono>
26 #include <cmath>
27 #include <cstddef>
28 #include <cstdio>
29 #include <ctime>
30 #include <iterator>
31 #include <map>
32 #include <memory>
33 #include <regex>
34 #include <string>
35 #include <string_view>
36 #include <unordered_map>
37 #include <utility>
38 #include <vector>
39
40 #include <android-base/chrono_utils.h>
41 #include <android-base/file.h>
42 #include <android-base/logging.h>
43 #include <android-base/parseint.h>
44 #include <android-base/properties.h>
45 #include <android-base/strings.h>
46 #include <android/log.h>
47 #include <cutils/android_reboot.h>
48 #include <cutils/properties.h>
49 #include <statslog.h>
50
51 #include "boot_event_record_store.h"
52
53 namespace {
54
55 struct AtomInfo {
56 int32_t atom;
57 int32_t event;
58 };
59
60 // Maps BootEvent used inside bootstat into statsd atom defined in
61 // frameworks/base/cmds/statsd/src/atoms.proto.
62 const std::unordered_map<std::string_view, AtomInfo> kBootEventToAtomInfo = {
63 // ELAPSED_TIME
64 {"ro.boottime.init",
65 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
66 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__ANDROID_INIT_STAGE_1}},
67 {"boot_complete",
68 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
69 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__BOOT_COMPLETE}},
70 {"boot_decryption_complete",
71 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
72 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__BOOT_COMPLETE_ENCRYPTION}},
73 {"boot_complete_no_encryption",
74 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
75 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__BOOT_COMPLETE_NO_ENCRYPTION}},
76 {"boot_complete_post_decrypt",
77 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
78 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__BOOT_COMPLETE_POST_DECRYPT}},
79 {"factory_reset_boot_complete",
80 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
81 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__FACTORY_RESET_BOOT_COMPLETE}},
82 {"factory_reset_boot_complete_no_encryption",
83 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
84 android::util::
85 BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__FACTORY_RESET_BOOT_COMPLETE_NO_ENCRYPTION}},
86 {"factory_reset_boot_complete_post_decrypt",
87 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
88 android::util::
89 BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__FACTORY_RESET_BOOT_COMPLETE_POST_DECRYPT}},
90 {"ota_boot_complete",
91 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
92 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__OTA_BOOT_COMPLETE}},
93 {"ota_boot_complete_no_encryption",
94 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
95 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__OTA_BOOT_COMPLETE_NO_ENCRYPTION}},
96 {"ota_boot_complete_post_decrypt",
97 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
98 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__OTA_BOOT_COMPLETE_POST_DECRYPT}},
99 {"post_decrypt_time_elapsed",
100 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
101 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__POST_DECRYPT}},
102 // DURATION
103 {"absolute_boot_time",
104 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
105 android::util::BOOT_TIME_EVENT_DURATION__EVENT__ABSOLUTE_BOOT_TIME}},
106 {"boottime.bootloader.1BLE",
107 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
108 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_FIRST_STAGE_EXEC}},
109 {"boottime.bootloader.1BLL",
110 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
111 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_FIRST_STAGE_LOAD}},
112 {"boottime.bootloader.KL",
113 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
114 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_KERNEL_LOAD}},
115 {"boottime.bootloader.2BLE",
116 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
117 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_SECOND_STAGE_EXEC}},
118 {"boottime.bootloader.2BLL",
119 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
120 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_SECOND_STAGE_LOAD}},
121 {"boottime.bootloader.SW",
122 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
123 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_UI_WAIT}},
124 {"boottime.bootloader.total",
125 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
126 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_TOTAL}},
127 {"boottime.init.cold_boot_wait",
128 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
129 android::util::BOOT_TIME_EVENT_DURATION__EVENT__COLDBOOT_WAIT}},
130 {"time_since_factory_reset",
131 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
132 android::util::BOOT_TIME_EVENT_DURATION__EVENT__FACTORY_RESET_TIME_SINCE_RESET}},
133 {"ro.boottime.init.first_stage",
134 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
135 android::util::BOOT_TIME_EVENT_DURATION__EVENT__ANDROID_INIT_STAGE_1}},
136 {"ro.boottime.init.selinux",
137 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
138 android::util::BOOT_TIME_EVENT_DURATION__EVENT__SELINUX_INIT}},
139 // UTC_TIME
140 {"factory_reset",
141 {android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED,
142 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_RESET_TIME}},
143 {"factory_reset_current_time",
144 {android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED,
145 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_CURRENT_TIME}},
146 {"factory_reset_record_value",
147 {android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED,
148 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_RECORD_VALUE}},
149 // ERROR_CODE
150 {"factory_reset_current_time_failure",
151 {android::util::BOOT_TIME_EVENT_ERROR_CODE_REPORTED,
152 android::util::BOOT_TIME_EVENT_ERROR_CODE__EVENT__FACTORY_RESET_CURRENT_TIME_FAILURE}},
153 };
154
155 // Scans the boot event record store for record files and logs each boot event
156 // via EventLog.
LogBootEvents()157 void LogBootEvents() {
158 BootEventRecordStore boot_event_store;
159 auto events = boot_event_store.GetAllBootEvents();
160 std::vector<std::string_view> notSupportedEvents;
161 for (const auto& event : events) {
162 const auto& name = event.first;
163 const auto& info = kBootEventToAtomInfo.find(name);
164 if (info != kBootEventToAtomInfo.end()) {
165 if (info->second.atom == android::util::BOOT_TIME_EVENT_ERROR_CODE_REPORTED) {
166 android::util::stats_write(static_cast<int32_t>(info->second.atom),
167 static_cast<int32_t>(info->second.event),
168 static_cast<int32_t>(event.second));
169 } else {
170 android::util::stats_write(static_cast<int32_t>(info->second.atom),
171 static_cast<int32_t>(info->second.event),
172 static_cast<int64_t>(event.second));
173 }
174 } else {
175 notSupportedEvents.push_back(name);
176 }
177 }
178 if (!notSupportedEvents.empty()) {
179 LOG(WARNING) << "LogBootEvents, atomInfo not defined for events:"
180 << android::base::Join(notSupportedEvents, ',');
181 }
182 }
183
184 // Records the named boot |event| to the record store. If |value| is non-empty
185 // and is a proper string representation of an integer value, the converted
186 // integer value is associated with the boot event.
RecordBootEventFromCommandLine(const std::string & event,const std::string & value_str)187 void RecordBootEventFromCommandLine(const std::string& event, const std::string& value_str) {
188 BootEventRecordStore boot_event_store;
189 if (!value_str.empty()) {
190 int32_t value = 0;
191 if (android::base::ParseInt(value_str, &value)) {
192 boot_event_store.AddBootEventWithValue(event, value);
193 }
194 } else {
195 boot_event_store.AddBootEvent(event);
196 }
197 }
198
PrintBootEvents()199 void PrintBootEvents() {
200 printf("Boot events:\n");
201 printf("------------\n");
202
203 BootEventRecordStore boot_event_store;
204 auto events = boot_event_store.GetAllBootEvents();
205 for (auto i = events.cbegin(); i != events.cend(); ++i) {
206 printf("%s\t%d\n", i->first.c_str(), i->second);
207 }
208 }
209
ShowHelp(const char * cmd)210 void ShowHelp(const char* cmd) {
211 fprintf(stderr, "Usage: %s [options]...\n", cmd);
212 fprintf(stderr,
213 "options include:\n"
214 " -h, --help Show this help\n"
215 " -l, --log Log all metrics to logstorage\n"
216 " -p, --print Dump the boot event records to the console\n"
217 " -r, --record Record the timestamp of a named boot event\n"
218 " --value Optional value to associate with the boot event\n"
219 " --record_boot_complete Record metrics related to the time for the device boot\n"
220 " --record_boot_reason Record the reason why the device booted\n"
221 " --record_time_since_factory_reset Record the time since the device was reset\n"
222 " --boot_reason_enum=<reason> Report the match to the kBootReasonMap table\n");
223 }
224
225 // Constructs a readable, printable string from the givencommand line
226 // arguments.
GetCommandLine(int argc,char ** argv)227 std::string GetCommandLine(int argc, char** argv) {
228 std::string cmd;
229 for (int i = 0; i < argc; ++i) {
230 cmd += argv[i];
231 cmd += " ";
232 }
233
234 return cmd;
235 }
236
237 constexpr int32_t kEmptyBootReason = 0;
238 constexpr int32_t kUnknownBootReason = 1;
239
240 // A mapping from boot reason string, as read from the ro.boot.bootreason
241 // system property, to a unique integer ID. Viewers of log data dashboards for
242 // the boot_reason metric may refer to this mapping to discern the histogram
243 // values. Regex matching, to manage the scale, as a minimum require either
244 // [, \ or * to be present in the string to switch to checking.
245 const std::map<std::string, int32_t> kBootReasonMap = {
246 {"reboot,[empty]", kEmptyBootReason},
247 {"__BOOTSTAT_UNKNOWN__", kUnknownBootReason},
248 {"normal", 2},
249 {"recovery", 3},
250 {"reboot", 4},
251 {"PowerKey", 5},
252 {"hard_reset", 6},
253 {"kernel_panic", 7},
254 {"rpm_err", 8},
255 {"hw_reset", 9},
256 {"tz_err", 10},
257 {"adsp_err", 11},
258 {"modem_err", 12},
259 {"mba_err", 13},
260 {"Watchdog", 14},
261 {"Panic", 15},
262 {"power_key", 16}, // aliasReasons to cold,powerkey (Mediatek)
263 {"power_on", 17}, // aliasReasons to cold,powerkey
264 {"Reboot", 18},
265 {"rtc", 19},
266 {"edl", 20},
267 {"oem_pon1", 21},
268 {"oem_powerkey", 22}, // aliasReasons to cold,powerkey
269 {"oem_unknown_reset", 23},
270 {"srto: HWWDT reset SC", 24},
271 {"srto: HWWDT reset platform", 25},
272 {"srto: bootloader", 26},
273 {"srto: kernel panic", 27},
274 {"srto: kernel watchdog reset", 28},
275 {"srto: normal", 29},
276 {"srto: reboot", 30},
277 {"srto: reboot-bootloader", 31},
278 {"srto: security watchdog reset", 32},
279 {"srto: wakesrc", 33},
280 {"srto: watchdog", 34},
281 {"srto:1-1", 35},
282 {"srto:omap_hsmm", 36},
283 {"srto:phy0", 37},
284 {"srto:rtc0", 38},
285 {"srto:touchpad", 39},
286 {"watchdog", 40},
287 {"watchdogr", 41},
288 {"wdog_bark", 42},
289 {"wdog_bite", 43},
290 {"wdog_reset", 44},
291 {"shutdown,", 45}, // Trailing comma is intentional. Do NOT use.
292 {"shutdown,userrequested", 46},
293 {"reboot,bootloader", 47},
294 {"reboot,cold", 48},
295 {"reboot,recovery", 49},
296 {"thermal_shutdown", 50},
297 {"s3_wakeup", 51},
298 {"kernel_panic,sysrq", 52},
299 {"kernel_panic,NULL", 53},
300 {"kernel_panic,null", 53},
301 {"kernel_panic,BUG", 54},
302 {"kernel_panic,bug", 54},
303 {"bootloader", 55},
304 {"cold", 56},
305 {"hard", 57},
306 {"warm", 58},
307 {"reboot,kernel_power_off_charging__reboot_system", 59}, // Can not happen
308 {"thermal-shutdown", 60},
309 {"shutdown,thermal", 61},
310 {"shutdown,battery", 62},
311 {"reboot,ota", 63},
312 {"reboot,factory_reset", 64},
313 {"reboot,", 65},
314 {"reboot,shell", 66},
315 {"reboot,adb", 67},
316 {"reboot,userrequested", 68},
317 {"shutdown,container", 69}, // Host OS asking Android Container to shutdown
318 {"cold,powerkey", 70},
319 {"warm,s3_wakeup", 71},
320 {"hard,hw_reset", 72},
321 {"shutdown,suspend", 73}, // Suspend to RAM
322 {"shutdown,hibernate", 74}, // Suspend to DISK
323 {"power_on_key", 75}, // aliasReasons to cold,powerkey
324 {"reboot_by_key", 76}, // translated to reboot,by_key
325 {"wdt_by_pass_pwk", 77}, // Mediatek
326 {"reboot_longkey", 78}, // translated to reboot,longkey
327 {"powerkey", 79}, // aliasReasons to cold,powerkey
328 {"usb", 80}, // aliasReasons to cold,charger (Mediatek)
329 {"wdt", 81}, // Mediatek
330 {"tool_by_pass_pwk", 82}, // aliasReasons to reboot,tool (Mediatek)
331 {"2sec_reboot", 83}, // aliasReasons to cold,rtc,2sec (Mediatek)
332 {"reboot,by_key", 84},
333 {"reboot,longkey", 85},
334 {"reboot,2sec", 86}, // Deprecate in two years, replaced with cold,rtc,2sec
335 {"shutdown,thermal,battery", 87},
336 {"reboot,its_just_so_hard", 88}, // produced by boot_reason_test
337 {"reboot,Its Just So Hard", 89}, // produced by boot_reason_test
338 {"reboot,rescueparty", 90},
339 {"charge", 91},
340 {"oem_tz_crash", 92},
341 {"uvlo", 93}, // aliasReasons to reboot,undervoltage
342 {"oem_ps_hold", 94},
343 {"abnormal_reset", 95},
344 {"oemerr_unknown", 96},
345 {"reboot_fastboot_mode", 97},
346 {"watchdog_apps_bite", 98},
347 {"xpu_err", 99},
348 {"power_on_usb", 100}, // aliasReasons to cold,charger
349 {"watchdog_rpm", 101},
350 {"watchdog_nonsec", 102},
351 {"watchdog_apps_bark", 103},
352 {"reboot_dmverity_corrupted", 104},
353 {"reboot_smpl", 105}, // aliasReasons to reboot,powerloss
354 {"watchdog_sdi_apps_reset", 106},
355 {"smpl", 107}, // aliasReasons to reboot,powerloss
356 {"oem_modem_failed_to_powerup", 108},
357 {"reboot_normal", 109},
358 {"oem_lpass_cfg", 110},
359 {"oem_xpu_ns_error", 111},
360 {"power_key_press", 112}, // aliasReasons to cold,powerkey
361 {"hardware_reset", 113},
362 {"reboot_by_powerkey", 114}, // aliasReasons to cold,powerkey (is this correct?)
363 {"reboot_verity", 115},
364 {"oem_rpm_undef_error", 116},
365 {"oem_crash_on_the_lk", 117},
366 {"oem_rpm_reset", 118},
367 {"reboot,powerloss", 119},
368 {"reboot,undervoltage", 120},
369 {"factory_cable", 121},
370 {"oem_ar6320_failed_to_powerup", 122},
371 {"watchdog_rpm_bite", 123},
372 {"power_on_cable", 124}, // aliasReasons to cold,charger
373 {"reboot_unknown", 125},
374 {"wireless_charger", 126},
375 {"0x776655ff", 127},
376 {"oem_thermal_bite_reset", 128},
377 {"charger", 129},
378 {"pon1", 130},
379 {"unknown", 131},
380 {"reboot_rtc", 132},
381 {"cold_boot", 133},
382 {"hard_rst", 134},
383 {"power-on", 135},
384 {"oem_adsp_resetting_the_soc", 136},
385 {"kpdpwr", 137},
386 {"oem_modem_timeout_waiting", 138},
387 {"usb_chg", 139},
388 {"warm_reset_0x02", 140},
389 {"warm_reset_0x80", 141},
390 {"pon_reason_0xb0", 142},
391 {"reboot_download", 143},
392 {"reboot_recovery_mode", 144},
393 {"oem_sdi_err_fatal", 145},
394 {"pmic_watchdog", 146},
395 {"software_master", 147},
396 {"cold,charger", 148},
397 {"cold,rtc", 149},
398 {"cold,rtc,2sec", 150}, // Mediatek
399 {"reboot,tool", 151}, // Mediatek
400 {"reboot,wdt", 152}, // Mediatek
401 {"reboot,unknown", 153}, // Mediatek
402 {"kernel_panic,audit", 154},
403 {"kernel_panic,atomic", 155},
404 {"kernel_panic,hung", 156},
405 {"kernel_panic,hung,rcu", 157},
406 {"kernel_panic,init", 158},
407 {"kernel_panic,oom", 159},
408 {"kernel_panic,stack", 160},
409 {"kernel_panic,sysrq,livelock,alarm", 161}, // llkd
410 {"kernel_panic,sysrq,livelock,driver", 162}, // llkd
411 {"kernel_panic,sysrq,livelock,zombie", 163}, // llkd
412 {"kernel_panic,modem", 164},
413 {"kernel_panic,adsp", 165},
414 {"kernel_panic,dsps", 166},
415 {"kernel_panic,wcnss", 167},
416 {"kernel_panic,_sde_encoder_phys_cmd_handle_ppdone_timeout", 168},
417 {"recovery,quiescent", 169},
418 {"reboot,quiescent", 170},
419 {"reboot,rtc", 171},
420 {"reboot,dm-verity_device_corrupted", 172},
421 {"reboot,dm-verity_enforcing", 173},
422 {"reboot,keys_clear", 174},
423 {"reboot,pmic_off_fault,.*", 175},
424 {"reboot,pmic_off_s3rst,.*", 176},
425 {"reboot,pmic_off_other,.*", 177},
426 {"reboot,userrequested,fastboot", 178},
427 {"reboot,userrequested,recovery", 179},
428 {"reboot,userrequested,recovery,ui", 180},
429 {"shutdown,userrequested,fastboot", 181},
430 {"shutdown,userrequested,recovery", 182},
431 {"reboot,unknown[0-9]*", 183},
432 {"reboot,longkey,.*", 184},
433 {"reboot,boringssl-self-check-failed", 185},
434 {"reboot,userspace_failed,shutdown_aborted", 186},
435 {"reboot,userspace_failed,watchdog_triggered", 187},
436 {"reboot,userspace_failed,watchdog_fork", 188},
437 {"reboot,userspace_failed,*", 189},
438 {"reboot,mount_userdata_failed", 190},
439 {"reboot,forcedsilent", 191},
440 {"reboot,forcednonsilent", 192},
441 };
442
443 // Converts a string value representing the reason the system booted to an
444 // integer representation. This is necessary for logging the boot_reason metric
445 // via Tron, which does not accept non-integer buckets in histograms.
BootReasonStrToEnum(const std::string & boot_reason)446 int32_t BootReasonStrToEnum(const std::string& boot_reason) {
447 auto mapping = kBootReasonMap.find(boot_reason);
448 if (mapping != kBootReasonMap.end()) {
449 return mapping->second;
450 }
451
452 if (boot_reason.empty()) {
453 return kEmptyBootReason;
454 }
455
456 for (const auto& [match, id] : kBootReasonMap) {
457 // Regex matches as a minimum require either [, \ or * to be present.
458 if (match.find_first_of("[\\*") == match.npos) continue;
459 // enforce match from beginning to end
460 auto exact = match;
461 if (exact[0] != '^') exact = "^" + exact;
462 if (exact[exact.size() - 1] != '$') exact = exact + "$";
463 if (std::regex_search(boot_reason, std::regex(exact))) return id;
464 }
465
466 LOG(INFO) << "Unknown boot reason: " << boot_reason;
467 return kUnknownBootReason;
468 }
469
470 // Canonical list of supported primary reboot reasons.
471 const std::vector<const std::string> knownReasons = {
472 // clang-format off
473 // kernel
474 "watchdog",
475 "kernel_panic",
476 // strong
477 "recovery", // Should not happen from ro.boot.bootreason
478 "bootloader", // Should not happen from ro.boot.bootreason
479 // blunt
480 "cold",
481 "hard",
482 "warm",
483 // super blunt
484 "shutdown", // Can not happen from ro.boot.bootreason
485 "reboot", // Default catch-all for anything unknown
486 // clang-format on
487 };
488
489 // Returns true if the supplied reason prefix is considered detailed enough.
isStrongRebootReason(const std::string & r)490 bool isStrongRebootReason(const std::string& r) {
491 for (auto& s : knownReasons) {
492 if (s == "cold") break;
493 // Prefix defined as terminated by a nul or comma (,).
494 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
495 return true;
496 }
497 }
498 return false;
499 }
500
501 // Returns true if the supplied reason prefix is associated with the kernel.
isKernelRebootReason(const std::string & r)502 bool isKernelRebootReason(const std::string& r) {
503 for (auto& s : knownReasons) {
504 if (s == "recovery") break;
505 // Prefix defined as terminated by a nul or comma (,).
506 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
507 return true;
508 }
509 }
510 return false;
511 }
512
513 // Returns true if the supplied reason prefix is considered known.
isKnownRebootReason(const std::string & r)514 bool isKnownRebootReason(const std::string& r) {
515 for (auto& s : knownReasons) {
516 // Prefix defined as terminated by a nul or comma (,).
517 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
518 return true;
519 }
520 }
521 return false;
522 }
523
524 // If the reboot reason should be improved, report true if is too blunt.
isBluntRebootReason(const std::string & r)525 bool isBluntRebootReason(const std::string& r) {
526 if (isStrongRebootReason(r)) return false;
527
528 if (!isKnownRebootReason(r)) return true; // Can not support unknown as detail
529
530 size_t pos = 0;
531 while ((pos = r.find(',', pos)) != std::string::npos) {
532 ++pos;
533 std::string next(r.substr(pos));
534 if (next.length() == 0) break;
535 if (next[0] == ',') continue;
536 if (!isKnownRebootReason(next)) return false; // Unknown subreason is good.
537 if (isStrongRebootReason(next)) return false; // eg: reboot,reboot
538 }
539 return true;
540 }
541
readPstoreConsole(std::string & console)542 bool readPstoreConsole(std::string& console) {
543 if (android::base::ReadFileToString("/sys/fs/pstore/console-ramoops-0", &console)) {
544 return true;
545 }
546 return android::base::ReadFileToString("/sys/fs/pstore/console-ramoops", &console);
547 }
548
549 // Implement a variant of std::string::rfind that is resilient to errors in
550 // the data stream being inspected.
551 class pstoreConsole {
552 private:
553 const size_t kBitErrorRate = 8; // number of bits per error
554 const std::string& console;
555
556 // Number of bits that differ between the two arguments l and r.
557 // Returns zero if the values for l and r are identical.
numError(uint8_t l,uint8_t r) const558 size_t numError(uint8_t l, uint8_t r) const { return std::bitset<8>(l ^ r).count(); }
559
560 // A string comparison function, reports the number of errors discovered
561 // in the match to a maximum of the bitLength / kBitErrorRate, at that
562 // point returning npos to indicate match is too poor.
563 //
564 // Since called in rfind which works backwards, expect cache locality will
565 // help if we check in reverse here as well for performance.
566 //
567 // Assumption: l (from console.c_str() + pos) is long enough to house
568 // _r.length(), checked in rfind caller below.
569 //
numError(size_t pos,const std::string & _r) const570 size_t numError(size_t pos, const std::string& _r) const {
571 const char* l = console.c_str() + pos;
572 const char* r = _r.c_str();
573 size_t n = _r.length();
574 const uint8_t* le = reinterpret_cast<const uint8_t*>(l) + n;
575 const uint8_t* re = reinterpret_cast<const uint8_t*>(r) + n;
576 size_t count = 0;
577 n = 0;
578 do {
579 // individual character bit error rate > threshold + slop
580 size_t num = numError(*--le, *--re);
581 if (num > ((8 + kBitErrorRate) / kBitErrorRate)) return std::string::npos;
582 // total bit error rate > threshold + slop
583 count += num;
584 ++n;
585 if (count > ((n * 8 + kBitErrorRate - (n > 2)) / kBitErrorRate)) {
586 return std::string::npos;
587 }
588 } while (le != reinterpret_cast<const uint8_t*>(l));
589 return count;
590 }
591
592 public:
pstoreConsole(const std::string & console)593 explicit pstoreConsole(const std::string& console) : console(console) {}
594 // scope of argument must be equal to or greater than scope of pstoreConsole
595 explicit pstoreConsole(const std::string&& console) = delete;
596 explicit pstoreConsole(std::string&& console) = delete;
597
598 // Our implementation of rfind, use exact match first, then resort to fuzzy.
rfind(const std::string & needle) const599 size_t rfind(const std::string& needle) const {
600 size_t pos = console.rfind(needle); // exact match?
601 if (pos != std::string::npos) return pos;
602
603 // Check to make sure needle fits in console string.
604 pos = console.length();
605 if (needle.length() > pos) return std::string::npos;
606 pos -= needle.length();
607 // fuzzy match to maximum kBitErrorRate
608 for (;;) {
609 if (numError(pos, needle) != std::string::npos) return pos;
610 if (pos == 0) break;
611 --pos;
612 }
613 return std::string::npos;
614 }
615
616 // Our implementation of find, use only fuzzy match.
find(const std::string & needle,size_t start=0) const617 size_t find(const std::string& needle, size_t start = 0) const {
618 // Check to make sure needle fits in console string.
619 if (needle.length() > console.length()) return std::string::npos;
620 const size_t last_pos = console.length() - needle.length();
621 // fuzzy match to maximum kBitErrorRate
622 for (size_t pos = start; pos <= last_pos; ++pos) {
623 if (numError(pos, needle) != std::string::npos) return pos;
624 }
625 return std::string::npos;
626 }
627
operator const std::string&() const628 operator const std::string&() const { return console; }
629 };
630
631 // If bit error match to needle, correct it.
632 // Return true if any corrections were discovered and applied.
correctForBitError(std::string & reason,const std::string & needle)633 bool correctForBitError(std::string& reason, const std::string& needle) {
634 bool corrected = false;
635 if (reason.length() < needle.length()) return corrected;
636 const pstoreConsole console(reason);
637 const size_t last_pos = reason.length() - needle.length();
638 for (size_t pos = 0; pos <= last_pos; pos += needle.length()) {
639 pos = console.find(needle, pos);
640 if (pos == std::string::npos) break;
641
642 // exact match has no malice
643 if (needle == reason.substr(pos, needle.length())) continue;
644
645 corrected = true;
646 reason = reason.substr(0, pos) + needle + reason.substr(pos + needle.length());
647 }
648 return corrected;
649 }
650
651 // If bit error match to needle, correct it.
652 // Return true if any corrections were discovered and applied.
653 // Try again if we can replace underline with spaces.
correctForBitErrorOrUnderline(std::string & reason,const std::string & needle)654 bool correctForBitErrorOrUnderline(std::string& reason, const std::string& needle) {
655 bool corrected = correctForBitError(reason, needle);
656 std::string _needle(needle);
657 std::transform(_needle.begin(), _needle.end(), _needle.begin(),
658 [](char c) { return (c == '_') ? ' ' : c; });
659 if (needle != _needle) {
660 corrected |= correctForBitError(reason, _needle);
661 }
662 return corrected;
663 }
664
665 // Converts a string value representing the reason the system booted to a
666 // string complying with Android system standard reason.
transformReason(std::string & reason)667 void transformReason(std::string& reason) {
668 std::transform(reason.begin(), reason.end(), reason.begin(), ::tolower);
669 std::transform(reason.begin(), reason.end(), reason.begin(),
670 [](char c) { return ::isblank(c) ? '_' : c; });
671 std::transform(reason.begin(), reason.end(), reason.begin(),
672 [](char c) { return ::isprint(c) ? c : '?'; });
673 }
674
675 // Check subreasons for reboot,<subreason> kernel_panic,sysrq,<subreason> or
676 // kernel_panic,<subreason>.
677 //
678 // If quoted flag is set, pull out and correct single quoted ('), newline (\n)
679 // or unprintable character terminated subreason, pos is supplied just beyond
680 // first quote. if quoted false, pull out and correct newline (\n) or
681 // unprintable character terminated subreason.
682 //
683 // Heuristics to find termination is painted into a corner:
684
685 // single bit error for quote ' that we can block. It is acceptable for
686 // the others 7, g in reason. 2/9 chance will miss the terminating quote,
687 // but there is always the terminating newline that usually immediately
688 // follows to fortify our chances.
likely_single_quote(char c)689 bool likely_single_quote(char c) {
690 switch (static_cast<uint8_t>(c)) {
691 case '\'': // '\''
692 case '\'' ^ 0x01: // '&'
693 case '\'' ^ 0x02: // '%'
694 case '\'' ^ 0x04: // '#'
695 case '\'' ^ 0x08: // '/'
696 return true;
697 case '\'' ^ 0x10: // '7'
698 break;
699 case '\'' ^ 0x20: // '\a' (unprintable)
700 return true;
701 case '\'' ^ 0x40: // 'g'
702 break;
703 case '\'' ^ 0x80: // 0xA7 (unprintable)
704 return true;
705 }
706 return false;
707 }
708
709 // ::isprint(c) and likely_space() will prevent us from being called for
710 // fundamentally printable entries, except for '\r' and '\b'.
711 //
712 // Except for * and J, single bit errors for \n, all others are non-
713 // printable so easy catch. It is _acceptable_ for *, J or j to exist in
714 // the reason string, so 2/9 chance we will miss the terminating newline.
715 //
716 // NB: J might not be acceptable, except if at the beginning or preceded
717 // with a space, '(' or any of the quotes and their BER aliases.
718 // NB: * might not be acceptable, except if at the beginning or preceded
719 // with a space, another *, or any of the quotes or their BER aliases.
720 //
721 // To reduce the chances to closer to 1/9 is too complicated for the gain.
likely_newline(char c)722 bool likely_newline(char c) {
723 switch (static_cast<uint8_t>(c)) {
724 case '\n': // '\n' (unprintable)
725 case '\n' ^ 0x01: // '\r' (unprintable)
726 case '\n' ^ 0x02: // '\b' (unprintable)
727 case '\n' ^ 0x04: // 0x0E (unprintable)
728 case '\n' ^ 0x08: // 0x02 (unprintable)
729 case '\n' ^ 0x10: // 0x1A (unprintable)
730 return true;
731 case '\n' ^ 0x20: // '*'
732 case '\n' ^ 0x40: // 'J'
733 break;
734 case '\n' ^ 0x80: // 0x8A (unprintable)
735 return true;
736 }
737 return false;
738 }
739
740 // ::isprint(c) will prevent us from being called for all the printable
741 // matches below. If we let unprintables through because of this, they
742 // get converted to underscore (_) by the validation phase.
likely_space(char c)743 bool likely_space(char c) {
744 switch (static_cast<uint8_t>(c)) {
745 case ' ': // ' '
746 case ' ' ^ 0x01: // '!'
747 case ' ' ^ 0x02: // '"'
748 case ' ' ^ 0x04: // '$'
749 case ' ' ^ 0x08: // '('
750 case ' ' ^ 0x10: // '0'
751 case ' ' ^ 0x20: // '\0' (unprintable)
752 case ' ' ^ 0x40: // 'P'
753 case ' ' ^ 0x80: // 0xA0 (unprintable)
754 case '\t': // '\t'
755 case '\t' ^ 0x01: // '\b' (unprintable) (likely_newline counters)
756 case '\t' ^ 0x02: // '\v' (unprintable)
757 case '\t' ^ 0x04: // '\r' (unprintable) (likely_newline counters)
758 case '\t' ^ 0x08: // 0x01 (unprintable)
759 case '\t' ^ 0x10: // 0x19 (unprintable)
760 case '\t' ^ 0x20: // ')'
761 case '\t' ^ 0x40: // '1'
762 case '\t' ^ 0x80: // 0x89 (unprintable)
763 return true;
764 }
765 return false;
766 }
767
getSubreason(const std::string & content,size_t pos,bool quoted)768 std::string getSubreason(const std::string& content, size_t pos, bool quoted) {
769 static constexpr size_t max_reason_length = 256;
770
771 std::string subReason(content.substr(pos, max_reason_length));
772 // Correct against any known strings that Bit Error Match
773 for (const auto& s : knownReasons) {
774 correctForBitErrorOrUnderline(subReason, s);
775 }
776 std::string terminator(quoted ? "'" : "");
777 for (const auto& m : kBootReasonMap) {
778 if (m.first.length() <= strlen("cold")) continue; // too short?
779 if (correctForBitErrorOrUnderline(subReason, m.first + terminator)) continue;
780 if (m.first.length() <= strlen("reboot,cold")) continue; // short?
781 if (android::base::StartsWith(m.first, "reboot,")) {
782 correctForBitErrorOrUnderline(subReason, m.first.substr(strlen("reboot,")) + terminator);
783 } else if (android::base::StartsWith(m.first, "kernel_panic,sysrq,")) {
784 correctForBitErrorOrUnderline(subReason,
785 m.first.substr(strlen("kernel_panic,sysrq,")) + terminator);
786 } else if (android::base::StartsWith(m.first, "kernel_panic,")) {
787 correctForBitErrorOrUnderline(subReason, m.first.substr(strlen("kernel_panic,")) + terminator);
788 }
789 }
790 for (pos = 0; pos < subReason.length(); ++pos) {
791 char c = subReason[pos];
792 if (!(::isprint(c) || likely_space(c)) || likely_newline(c) ||
793 (quoted && likely_single_quote(c))) {
794 subReason.erase(pos);
795 break;
796 }
797 }
798 transformReason(subReason);
799 return subReason;
800 }
801
addKernelPanicSubReason(const pstoreConsole & console,std::string & ret)802 bool addKernelPanicSubReason(const pstoreConsole& console, std::string& ret) {
803 // Check for kernel panic types to refine information
804 if ((console.rfind("SysRq : Trigger a crash") != std::string::npos) ||
805 (console.rfind("PC is at sysrq_handle_crash+") != std::string::npos)) {
806 ret = "kernel_panic,sysrq";
807 // Invented for Android to allow daemons that specifically trigger sysrq
808 // to communicate more accurate boot subreasons via last console messages.
809 static constexpr char sysrqSubreason[] = "SysRq : Trigger a crash : '";
810 auto pos = console.rfind(sysrqSubreason);
811 if (pos != std::string::npos) {
812 ret += "," + getSubreason(console, pos + strlen(sysrqSubreason), /* quoted */ true);
813 }
814 return true;
815 }
816 if (console.rfind("Unable to handle kernel NULL pointer dereference at virtual address") !=
817 std::string::npos) {
818 ret = "kernel_panic,null";
819 return true;
820 }
821 if (console.rfind("Kernel BUG at ") != std::string::npos) {
822 ret = "kernel_panic,bug";
823 return true;
824 }
825
826 std::string panic("Kernel panic - not syncing: ");
827 auto pos = console.rfind(panic);
828 if (pos != std::string::npos) {
829 static const std::vector<std::pair<const std::string, const std::string>> panicReasons = {
830 {"Out of memory", "oom"},
831 {"out of memory", "oom"},
832 {"Oh boy, that early out of memory", "oom"}, // omg
833 {"BUG!", "bug"},
834 {"hung_task: blocked tasks", "hung"},
835 {"audit: ", "audit"},
836 {"scheduling while atomic", "atomic"},
837 {"Attempted to kill init!", "init"},
838 {"Requested init", "init"},
839 {"No working init", "init"},
840 {"Could not decompress init", "init"},
841 {"RCU Stall", "hung,rcu"},
842 {"stack-protector", "stack"},
843 {"kernel stack overflow", "stack"},
844 {"Corrupt kernel stack", "stack"},
845 {"low stack detected", "stack"},
846 {"corrupted stack end", "stack"},
847 {"subsys-restart: Resetting the SoC - modem crashed.", "modem"},
848 {"subsys-restart: Resetting the SoC - adsp crashed.", "adsp"},
849 {"subsys-restart: Resetting the SoC - dsps crashed.", "dsps"},
850 {"subsys-restart: Resetting the SoC - wcnss crashed.", "wcnss"},
851 };
852
853 ret = "kernel_panic";
854 for (auto& s : panicReasons) {
855 if (console.find(panic + s.first, pos) != std::string::npos) {
856 ret += "," + s.second;
857 return true;
858 }
859 }
860 auto reason = getSubreason(console, pos + panic.length(), /* newline */ false);
861 if (reason.length() > 3) {
862 ret += "," + reason;
863 }
864 return true;
865 }
866 return false;
867 }
868
addKernelPanicSubReason(const std::string & content,std::string & ret)869 bool addKernelPanicSubReason(const std::string& content, std::string& ret) {
870 return addKernelPanicSubReason(pstoreConsole(content), ret);
871 }
872
873 const char system_reboot_reason_property[] = "sys.boot.reason";
874 const char last_reboot_reason_property[] = LAST_REBOOT_REASON_PROPERTY;
875 const char last_reboot_reason_file[] = LAST_REBOOT_REASON_FILE;
876 const char last_last_reboot_reason_property[] = "sys.boot.reason.last";
877 constexpr size_t history_reboot_reason_size = 4;
878 const char history_reboot_reason_property[] = LAST_REBOOT_REASON_PROPERTY ".history";
879 const char bootloader_reboot_reason_property[] = "ro.boot.bootreason";
880
881 // Land system_boot_reason into system_reboot_reason_property.
882 // Shift system_boot_reason into history_reboot_reason_property.
BootReasonAddToHistory(const std::string & system_boot_reason)883 void BootReasonAddToHistory(const std::string& system_boot_reason) {
884 if (system_boot_reason.empty()) return;
885 LOG(INFO) << "Canonical boot reason: " << system_boot_reason;
886 auto old_system_boot_reason = android::base::GetProperty(system_reboot_reason_property, "");
887 if (!android::base::SetProperty(system_reboot_reason_property, system_boot_reason)) {
888 android::base::SetProperty(system_reboot_reason_property,
889 system_boot_reason.substr(0, PROPERTY_VALUE_MAX - 1));
890 }
891 auto reason_history =
892 android::base::Split(android::base::GetProperty(history_reboot_reason_property, ""), "\n");
893 static auto mark = time(nullptr);
894 auto mark_str = std::string(",") + std::to_string(mark);
895 auto marked_system_boot_reason = system_boot_reason + mark_str;
896 if (!reason_history.empty()) {
897 // delete any entries that we just wrote in a previous
898 // call and leveraging duplicate line handling
899 auto last = old_system_boot_reason + mark_str;
900 // trim the list to (history_reboot_reason_size - 1)
901 ssize_t max = history_reboot_reason_size;
902 for (auto it = reason_history.begin(); it != reason_history.end();) {
903 if (it->empty() || (last == *it) || (marked_system_boot_reason == *it) || (--max <= 0)) {
904 it = reason_history.erase(it);
905 } else {
906 last = *it;
907 ++it;
908 }
909 }
910 }
911 // insert at the front, concatenating mark (<epoch time>) detail to the value.
912 reason_history.insert(reason_history.begin(), marked_system_boot_reason);
913 // If the property string is too long ( > PROPERTY_VALUE_MAX)
914 // we get an error, so trim out last entry and try again.
915 while (!android::base::SetProperty(history_reboot_reason_property,
916 android::base::Join(reason_history, '\n'))) {
917 auto it = std::prev(reason_history.end());
918 if (it == reason_history.end()) break;
919 reason_history.erase(it);
920 }
921 }
922
923 // Scrub, Sanitize, Standardize and Enhance the boot reason string supplied.
BootReasonStrToReason(const std::string & boot_reason)924 std::string BootReasonStrToReason(const std::string& boot_reason) {
925 auto ret = android::base::GetProperty(system_reboot_reason_property, "");
926 std::string reason(boot_reason);
927 // If sys.boot.reason == ro.boot.bootreason, let's re-evaluate
928 if (reason == ret) ret = "";
929
930 transformReason(reason);
931
932 // Is the current system boot reason sys.boot.reason valid?
933 if (!isKnownRebootReason(ret)) ret = "";
934
935 if (ret == "") {
936 // Is the bootloader boot reason ro.boot.bootreason known?
937 std::vector<std::string> words(android::base::Split(reason, ",_-"));
938 for (auto& s : knownReasons) {
939 std::string blunt;
940 for (auto& r : words) {
941 if (r == s) {
942 if (isBluntRebootReason(s)) {
943 blunt = s;
944 } else {
945 ret = s;
946 break;
947 }
948 }
949 }
950 if (ret == "") ret = blunt;
951 if (ret != "") break;
952 }
953 }
954
955 if (ret == "") {
956 // A series of checks to take some officially unsupported reasons
957 // reported by the bootloader and find some logical and canonical
958 // sense. In an ideal world, we would require those bootloaders
959 // to behave and follow our CTS standards.
960 //
961 // first member is the output
962 // second member is an unanchored regex for an alias
963 //
964 // If output has a prefix of <bang> '!', we do not use it as a
965 // match needle (and drop the <bang> prefix when landing in output),
966 // otherwise look for it as well. This helps keep the scale of the
967 // following table smaller.
968 static const std::vector<std::pair<const std::string, const std::string>> aliasReasons = {
969 {"watchdog", "wdog"},
970 {"kernel_panic", "panic"},
971 {"shutdown,thermal", "thermal"},
972 {"warm,s3_wakeup", "s3_wakeup"},
973 {"hard,hw_reset", "hw_reset"},
974 {"cold,charger", "usb|power_on_cable"},
975 {"cold,powerkey", "powerkey|power_key|PowerKey|power_on"},
976 {"cold,rtc", "rtc"},
977 {"cold,rtc,2sec", "2sec_reboot"},
978 {"!warm", "wdt_by_pass_pwk"}, // change flavour of blunt
979 {"!reboot", "^wdt$"}, // change flavour of blunt
980 {"reboot,tool", "tool_by_pass_pwk"},
981 {"!reboot,longkey", "reboot_longkey"},
982 {"!reboot,longkey", "kpdpwr"},
983 {"!reboot,undervoltage", "uvlo"},
984 {"!reboot,powerloss", "smpl"},
985 {"bootloader", ""},
986 };
987
988 for (auto& s : aliasReasons) {
989 size_t firstHasNot = s.first[0] == '!';
990 if (!firstHasNot && (reason.find(s.first) != std::string::npos)) {
991 ret = s.first;
992 break;
993 }
994 if (s.second.size() && std::regex_search(reason, std::regex(s.second))) {
995 ret = s.first.substr(firstHasNot);
996 break;
997 }
998 }
999 }
1000
1001 // If watchdog is the reason, see if there is a security angle?
1002 if (ret == "watchdog") {
1003 if (reason.find("sec") != std::string::npos) {
1004 ret += ",security";
1005 }
1006 }
1007
1008 if (ret == "kernel_panic") {
1009 // Check to see if last klog has some refinement hints.
1010 std::string content;
1011 if (readPstoreConsole(content)) {
1012 addKernelPanicSubReason(content, ret);
1013 }
1014 } else if (isBluntRebootReason(ret)) {
1015 // Check the other available reason resources if the reason is still blunt.
1016
1017 // Check to see if last klog has some refinement hints.
1018 std::string content;
1019 if (readPstoreConsole(content)) {
1020 const pstoreConsole console(content);
1021 // The toybox reboot command used directly (unlikely)? But also
1022 // catches init's response to Android's more controlled reboot command.
1023 if (console.rfind("reboot: Power down") != std::string::npos) {
1024 ret = "shutdown"; // Still too blunt, but more accurate.
1025 // ToDo: init should record the shutdown reason to kernel messages ala:
1026 // init: shutdown system with command 'last_reboot_reason'
1027 // so that if pstore has persistence we can get some details
1028 // that could be missing in last_reboot_reason_property.
1029 }
1030
1031 static const char cmd[] = "reboot: Restarting system with command '";
1032 size_t pos = console.rfind(cmd);
1033 if (pos != std::string::npos) {
1034 std::string subReason(getSubreason(content, pos + strlen(cmd), /* quoted */ true));
1035 if (subReason != "") { // Will not land "reboot" as that is too blunt.
1036 if (isKernelRebootReason(subReason)) {
1037 ret = "reboot," + subReason; // User space can't talk kernel reasons.
1038 } else if (isKnownRebootReason(subReason)) {
1039 ret = subReason;
1040 } else {
1041 ret = "reboot," + subReason; // legitimize unknown reasons
1042 }
1043 }
1044 // Some bootloaders shutdown results record in last kernel message.
1045 if (!strcmp(ret.c_str(), "reboot,kernel_power_off_charging__reboot_system")) {
1046 ret = "shutdown";
1047 }
1048 }
1049
1050 // Check for kernel panics, allowed to override reboot command.
1051 if (!addKernelPanicSubReason(console, ret) &&
1052 // check for long-press power down
1053 ((console.rfind("Power held for ") != std::string::npos) ||
1054 (console.rfind("charger: [") != std::string::npos))) {
1055 ret = "cold";
1056 }
1057 }
1058
1059 // TODO: use the HAL to get battery level (http://b/77725702).
1060
1061 // Is there a controlled shutdown hint in last_reboot_reason_property?
1062 if (isBluntRebootReason(ret)) {
1063 // Content buffer no longer will have console data. Beware if more
1064 // checks added below, that depend on parsing console content.
1065 if (!android::base::ReadFileToString(last_reboot_reason_file, &content)) {
1066 content = android::base::GetProperty(last_reboot_reason_property, "");
1067 }
1068 transformReason(content);
1069
1070 // Anything in last is better than 'super-blunt' reboot or shutdown.
1071 if ((ret == "") || (ret == "reboot") || (ret == "shutdown") || !isBluntRebootReason(content)) {
1072 ret = content;
1073 }
1074 }
1075
1076 // Other System Health HAL reasons?
1077
1078 // ToDo: /proc/sys/kernel/boot_reason needs a HAL interface to
1079 // possibly offer hardware-specific clues from the PMIC.
1080 }
1081
1082 // If unknown left over from above, make it "reboot,<boot_reason>"
1083 if (ret == "") {
1084 ret = "reboot";
1085 if (android::base::StartsWith(reason, "reboot")) {
1086 reason = reason.substr(strlen("reboot"));
1087 while ((reason[0] == ',') || (reason[0] == '_')) {
1088 reason = reason.substr(1);
1089 }
1090 }
1091 if (reason != "") {
1092 ret += ",";
1093 ret += reason;
1094 }
1095 }
1096
1097 LOG(INFO) << "Canonical boot reason: " << ret;
1098 return ret;
1099 }
1100
1101 // Returns the appropriate metric key prefix for the boot_complete metric such
1102 // that boot metrics after a system update are labeled as ota_boot_complete;
1103 // otherwise, they are labeled as boot_complete. This method encapsulates the
1104 // bookkeeping required to track when a system update has occurred by storing
1105 // the UTC timestamp of the system build date and comparing against the current
1106 // system build date.
CalculateBootCompletePrefix()1107 std::string CalculateBootCompletePrefix() {
1108 static const std::string kBuildDateKey = "build_date";
1109 std::string boot_complete_prefix = "boot_complete";
1110
1111 auto build_date_str = android::base::GetProperty("ro.build.date.utc", "");
1112 int32_t build_date;
1113 if (!android::base::ParseInt(build_date_str, &build_date)) {
1114 return std::string();
1115 }
1116
1117 BootEventRecordStore boot_event_store;
1118 BootEventRecordStore::BootEventRecord record;
1119 if (!boot_event_store.GetBootEvent(kBuildDateKey, &record)) {
1120 boot_complete_prefix = "factory_reset_" + boot_complete_prefix;
1121 boot_event_store.AddBootEventWithValue(kBuildDateKey, build_date);
1122 BootReasonAddToHistory("reboot,factory_reset");
1123 } else if (build_date != record.second) {
1124 boot_complete_prefix = "ota_" + boot_complete_prefix;
1125 boot_event_store.AddBootEventWithValue(kBuildDateKey, build_date);
1126 BootReasonAddToHistory("reboot,ota");
1127 }
1128
1129 return boot_complete_prefix;
1130 }
1131
1132 // Records the value of a given ro.boottime.init property in milliseconds.
RecordInitBootTimeProp(BootEventRecordStore * boot_event_store,const char * property)1133 void RecordInitBootTimeProp(BootEventRecordStore* boot_event_store, const char* property) {
1134 auto value = android::base::GetProperty(property, "");
1135
1136 int32_t time_in_ms;
1137 if (android::base::ParseInt(value, &time_in_ms)) {
1138 boot_event_store->AddBootEventWithValue(property, time_in_ms);
1139 }
1140 }
1141
1142 // A map from bootloader timing stage to the time that stage took during boot.
1143 typedef std::map<std::string, int32_t> BootloaderTimingMap;
1144
1145 // Returns a mapping from bootloader stage names to the time those stages
1146 // took to boot.
GetBootLoaderTimings()1147 const BootloaderTimingMap GetBootLoaderTimings() {
1148 BootloaderTimingMap timings;
1149
1150 // |ro.boot.boottime| is of the form 'stage1:time1,...,stageN:timeN',
1151 // where timeN is in milliseconds.
1152 auto value = android::base::GetProperty("ro.boot.boottime", "");
1153 if (value.empty()) {
1154 // ro.boot.boottime is not reported on all devices.
1155 return BootloaderTimingMap();
1156 }
1157
1158 auto stages = android::base::Split(value, ",");
1159 for (const auto& stageTiming : stages) {
1160 // |stageTiming| is of the form 'stage:time'.
1161 auto stageTimingValues = android::base::Split(stageTiming, ":");
1162 DCHECK_EQ(2U, stageTimingValues.size());
1163
1164 if (stageTimingValues.size() < 2) continue;
1165 std::string stageName = stageTimingValues[0];
1166 int32_t time_ms;
1167 if (android::base::ParseInt(stageTimingValues[1], &time_ms)) {
1168 timings[stageName] = time_ms;
1169 }
1170 }
1171
1172 return timings;
1173 }
1174
1175 // Returns the total bootloader boot time from the ro.boot.boottime system property.
GetBootloaderTime(const BootloaderTimingMap & bootloader_timings)1176 int32_t GetBootloaderTime(const BootloaderTimingMap& bootloader_timings) {
1177 int32_t total_time = 0;
1178 for (const auto& timing : bootloader_timings) {
1179 total_time += timing.second;
1180 }
1181
1182 return total_time;
1183 }
1184
1185 // Parses and records the set of bootloader stages and associated boot times
1186 // from the ro.boot.boottime system property.
RecordBootloaderTimings(BootEventRecordStore * boot_event_store,const BootloaderTimingMap & bootloader_timings)1187 void RecordBootloaderTimings(BootEventRecordStore* boot_event_store,
1188 const BootloaderTimingMap& bootloader_timings) {
1189 int32_t total_time = 0;
1190 for (const auto& timing : bootloader_timings) {
1191 total_time += timing.second;
1192 boot_event_store->AddBootEventWithValue("boottime.bootloader." + timing.first, timing.second);
1193 }
1194
1195 boot_event_store->AddBootEventWithValue("boottime.bootloader.total", total_time);
1196 }
1197
1198 // Returns the closest estimation to the absolute device boot time, i.e.,
1199 // from power on to boot_complete, including bootloader times.
GetAbsoluteBootTime(const BootloaderTimingMap & bootloader_timings,std::chrono::milliseconds uptime)1200 std::chrono::milliseconds GetAbsoluteBootTime(const BootloaderTimingMap& bootloader_timings,
1201 std::chrono::milliseconds uptime) {
1202 int32_t bootloader_time_ms = 0;
1203
1204 for (const auto& timing : bootloader_timings) {
1205 if (timing.first.compare("SW") != 0) {
1206 bootloader_time_ms += timing.second;
1207 }
1208 }
1209
1210 auto bootloader_duration = std::chrono::milliseconds(bootloader_time_ms);
1211 return bootloader_duration + uptime;
1212 }
1213
1214 // Records the closest estimation to the absolute device boot time in seconds.
1215 // i.e. from power on to boot_complete, including bootloader times.
RecordAbsoluteBootTime(BootEventRecordStore * boot_event_store,std::chrono::milliseconds absolute_total)1216 void RecordAbsoluteBootTime(BootEventRecordStore* boot_event_store,
1217 std::chrono::milliseconds absolute_total) {
1218 auto absolute_total_sec = std::chrono::duration_cast<std::chrono::seconds>(absolute_total);
1219 boot_event_store->AddBootEventWithValue("absolute_boot_time", absolute_total_sec.count());
1220 }
1221
1222 // Logs the total boot time and reason to statsd.
LogBootInfoToStatsd(std::chrono::milliseconds end_time,std::chrono::milliseconds total_duration,int32_t bootloader_duration_ms,double time_since_last_boot_sec)1223 void LogBootInfoToStatsd(std::chrono::milliseconds end_time,
1224 std::chrono::milliseconds total_duration, int32_t bootloader_duration_ms,
1225 double time_since_last_boot_sec) {
1226 auto reason = android::base::GetProperty(bootloader_reboot_reason_property, "<EMPTY>");
1227 auto system_reason = android::base::GetProperty(system_reboot_reason_property, "<EMPTY>");
1228 android::util::stats_write(android::util::BOOT_SEQUENCE_REPORTED, reason.c_str(),
1229 system_reason.c_str(), end_time.count(), total_duration.count(),
1230 (int64_t)bootloader_duration_ms,
1231 (int64_t)time_since_last_boot_sec * 1000);
1232 }
1233
SetSystemBootReason()1234 void SetSystemBootReason() {
1235 const auto bootloader_boot_reason =
1236 android::base::GetProperty(bootloader_reboot_reason_property, "");
1237 const std::string system_boot_reason(BootReasonStrToReason(bootloader_boot_reason));
1238 // Record the scrubbed system_boot_reason to the property
1239 BootReasonAddToHistory(system_boot_reason);
1240 // Shift last_reboot_reason_property to last_last_reboot_reason_property
1241 std::string last_boot_reason;
1242 if (!android::base::ReadFileToString(last_reboot_reason_file, &last_boot_reason)) {
1243 PLOG(ERROR) << "Failed to read " << last_reboot_reason_file;
1244 last_boot_reason = android::base::GetProperty(last_reboot_reason_property, "");
1245 LOG(INFO) << "Value of " << last_reboot_reason_property << " : " << last_boot_reason;
1246 } else {
1247 LOG(INFO) << "Last reboot reason read from " << last_reboot_reason_file << " : "
1248 << last_boot_reason << ". Last reboot reason read from "
1249 << last_reboot_reason_property << " : "
1250 << android::base::GetProperty(last_reboot_reason_property, "");
1251 }
1252 if (last_boot_reason.empty() || isKernelRebootReason(system_boot_reason)) {
1253 last_boot_reason = system_boot_reason;
1254 } else {
1255 transformReason(last_boot_reason);
1256 }
1257 LOG(INFO) << "Normalized last reboot reason : " << last_boot_reason;
1258 android::base::SetProperty(last_last_reboot_reason_property, last_boot_reason);
1259 android::base::SetProperty(last_reboot_reason_property, "");
1260 if (unlink(last_reboot_reason_file) != 0) {
1261 PLOG(ERROR) << "Failed to unlink " << last_reboot_reason_file;
1262 }
1263 }
1264
1265 // Gets the boot time offset. This is useful when Android is running in a
1266 // container, because the boot_clock is not reset when Android reboots.
GetBootTimeOffset()1267 std::chrono::nanoseconds GetBootTimeOffset() {
1268 static const int64_t boottime_offset =
1269 android::base::GetIntProperty<int64_t>("ro.boot.boottime_offset", 0);
1270 return std::chrono::nanoseconds(boottime_offset);
1271 }
1272
1273 // Returns the current uptime, accounting for any offset in the CLOCK_BOOTTIME
1274 // clock.
GetUptime()1275 android::base::boot_clock::duration GetUptime() {
1276 return android::base::boot_clock::now().time_since_epoch() - GetBootTimeOffset();
1277 }
1278
1279 // Records several metrics related to the time it takes to boot the device,
1280 // including disambiguating boot time on encrypted or non-encrypted devices.
RecordBootComplete()1281 void RecordBootComplete() {
1282 BootEventRecordStore boot_event_store;
1283 BootEventRecordStore::BootEventRecord record;
1284
1285 auto uptime_ns = GetUptime();
1286 auto uptime_s = std::chrono::duration_cast<std::chrono::seconds>(uptime_ns);
1287 time_t current_time_utc = time(nullptr);
1288 time_t time_since_last_boot = 0;
1289
1290 if (boot_event_store.GetBootEvent("last_boot_time_utc", &record)) {
1291 time_t last_boot_time_utc = record.second;
1292 time_since_last_boot = difftime(current_time_utc, last_boot_time_utc);
1293 boot_event_store.AddBootEventWithValue("time_since_last_boot", time_since_last_boot);
1294 }
1295
1296 boot_event_store.AddBootEventWithValue("last_boot_time_utc", current_time_utc);
1297
1298 // The boot_complete metric has two variants: boot_complete and
1299 // ota_boot_complete. The latter signifies that the device is booting after
1300 // a system update.
1301 std::string boot_complete_prefix = CalculateBootCompletePrefix();
1302 if (boot_complete_prefix.empty()) {
1303 // The system is hosed because the build date property could not be read.
1304 return;
1305 }
1306
1307 // post_decrypt_time_elapsed is only logged on encrypted devices.
1308 if (boot_event_store.GetBootEvent("post_decrypt_time_elapsed", &record)) {
1309 // Log the amount of time elapsed until the device is decrypted, which
1310 // includes the variable amount of time the user takes to enter the
1311 // decryption password.
1312 boot_event_store.AddBootEventWithValue("boot_decryption_complete", uptime_s.count());
1313
1314 // Subtract the decryption time to normalize the boot cycle timing.
1315 std::chrono::seconds boot_complete = std::chrono::seconds(uptime_s.count() - record.second);
1316 boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_post_decrypt",
1317 boot_complete.count());
1318 } else {
1319 boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_no_encryption",
1320 uptime_s.count());
1321 }
1322
1323 // Record the total time from device startup to boot complete, regardless of
1324 // encryption state.
1325 boot_event_store.AddBootEventWithValue(boot_complete_prefix, uptime_s.count());
1326
1327 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init");
1328 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.first_stage");
1329 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.selinux");
1330 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.cold_boot_wait");
1331
1332 const BootloaderTimingMap bootloader_timings = GetBootLoaderTimings();
1333 int32_t bootloader_boot_duration = GetBootloaderTime(bootloader_timings);
1334 RecordBootloaderTimings(&boot_event_store, bootloader_timings);
1335
1336 auto uptime_ms = std::chrono::duration_cast<std::chrono::milliseconds>(uptime_ns);
1337 auto absolute_boot_time = GetAbsoluteBootTime(bootloader_timings, uptime_ms);
1338 RecordAbsoluteBootTime(&boot_event_store, absolute_boot_time);
1339
1340 auto boot_end_time_point = std::chrono::system_clock::now().time_since_epoch();
1341 auto boot_end_time = std::chrono::duration_cast<std::chrono::milliseconds>(boot_end_time_point);
1342
1343 LogBootInfoToStatsd(boot_end_time, absolute_boot_time, bootloader_boot_duration,
1344 time_since_last_boot);
1345 }
1346
1347 // Records the boot_reason metric by querying the ro.boot.bootreason system
1348 // property.
RecordBootReason()1349 void RecordBootReason() {
1350 const auto reason = android::base::GetProperty(bootloader_reboot_reason_property, "");
1351
1352 if (reason.empty()) {
1353 // TODO(b/148575354): Replace with statsd.
1354 // Log an empty boot reason value as '<EMPTY>' to ensure the value is intentional
1355 // (and not corruption anywhere else in the reporting pipeline).
1356 // android::metricslogger::LogMultiAction(android::metricslogger::ACTION_BOOT,
1357 // android::metricslogger::FIELD_PLATFORM_REASON,
1358 // "<EMPTY>");
1359 } else {
1360 // TODO(b/148575354): Replace with statsd.
1361 // android::metricslogger::LogMultiAction(android::metricslogger::ACTION_BOOT,
1362 // android::metricslogger::FIELD_PLATFORM_REASON,
1363 // reason);
1364 }
1365
1366 // Log the raw bootloader_boot_reason property value.
1367 int32_t boot_reason = BootReasonStrToEnum(reason);
1368 BootEventRecordStore boot_event_store;
1369 boot_event_store.AddBootEventWithValue("boot_reason", boot_reason);
1370
1371 // Log the scrubbed system_boot_reason.
1372 const auto system_reason = android::base::GetProperty(system_reboot_reason_property, "");
1373 int32_t system_boot_reason = BootReasonStrToEnum(system_reason);
1374 boot_event_store.AddBootEventWithValue("system_boot_reason", system_boot_reason);
1375
1376 if (reason == "") {
1377 android::base::SetProperty(bootloader_reboot_reason_property, system_reason);
1378 }
1379 }
1380
1381 // Records two metrics related to the user resetting a device: the time at
1382 // which the device is reset, and the time since the user last reset the
1383 // device. The former is only set once per-factory reset.
RecordFactoryReset()1384 void RecordFactoryReset() {
1385 BootEventRecordStore boot_event_store;
1386 BootEventRecordStore::BootEventRecord record;
1387
1388 time_t current_time_utc = time(nullptr);
1389
1390 if (current_time_utc < 0) {
1391 // UMA does not display negative values in buckets, so convert to positive.
1392 // Logging via BootEventRecordStore.
1393 android::util::stats_write(
1394 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_ERROR_CODE_REPORTED),
1395 static_cast<int32_t>(
1396 android::util::BOOT_TIME_EVENT_ERROR_CODE__EVENT__FACTORY_RESET_CURRENT_TIME_FAILURE),
1397 static_cast<int32_t>(std::abs(current_time_utc)));
1398
1399 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1400 // is losing records somehow.
1401 boot_event_store.AddBootEventWithValue("factory_reset_current_time_failure",
1402 std::abs(current_time_utc));
1403 return;
1404 } else {
1405 android::util::stats_write(
1406 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED),
1407 static_cast<int32_t>(
1408 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_CURRENT_TIME),
1409 static_cast<int64_t>(current_time_utc));
1410
1411 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1412 // is losing records somehow.
1413 boot_event_store.AddBootEventWithValue("factory_reset_current_time", current_time_utc);
1414 }
1415
1416 // The factory_reset boot event does not exist after the device is reset, so
1417 // use this signal to mark the time of the factory reset.
1418 if (!boot_event_store.GetBootEvent("factory_reset", &record)) {
1419 boot_event_store.AddBootEventWithValue("factory_reset", current_time_utc);
1420
1421 // Don't log the time_since_factory_reset until some time has elapsed.
1422 // The data is not meaningful yet and skews the histogram buckets.
1423 return;
1424 }
1425
1426 // Calculate and record the difference in time between now and the
1427 // factory_reset time.
1428 time_t factory_reset_utc = record.second;
1429 android::util::stats_write(
1430 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED),
1431 static_cast<int32_t>(
1432 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_RECORD_VALUE),
1433 static_cast<int64_t>(factory_reset_utc));
1434
1435 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1436 // is losing records somehow.
1437 boot_event_store.AddBootEventWithValue("factory_reset_record_value", factory_reset_utc);
1438
1439 time_t time_since_factory_reset = difftime(current_time_utc, factory_reset_utc);
1440 boot_event_store.AddBootEventWithValue("time_since_factory_reset", time_since_factory_reset);
1441 }
1442
1443 // List the associated boot reason(s), if arg is nullptr then all.
PrintBootReasonEnum(const char * arg)1444 void PrintBootReasonEnum(const char* arg) {
1445 int value = -1;
1446 if (arg != nullptr) {
1447 value = BootReasonStrToEnum(arg);
1448 }
1449 for (const auto& [match, id] : kBootReasonMap) {
1450 if ((value < 0) || (value == id)) {
1451 printf("%u\t%s\n", id, match.c_str());
1452 }
1453 }
1454 }
1455
1456 } // namespace
1457
main(int argc,char ** argv)1458 int main(int argc, char** argv) {
1459 android::base::InitLogging(argv);
1460
1461 const std::string cmd_line = GetCommandLine(argc, argv);
1462 LOG(INFO) << "Service started: " << cmd_line;
1463
1464 int option_index = 0;
1465 static const char value_str[] = "value";
1466 static const char system_boot_reason_str[] = "set_system_boot_reason";
1467 static const char boot_complete_str[] = "record_boot_complete";
1468 static const char boot_reason_str[] = "record_boot_reason";
1469 static const char factory_reset_str[] = "record_time_since_factory_reset";
1470 static const char boot_reason_enum_str[] = "boot_reason_enum";
1471 static const struct option long_options[] = {
1472 // clang-format off
1473 { "help", no_argument, NULL, 'h' },
1474 { "log", no_argument, NULL, 'l' },
1475 { "print", no_argument, NULL, 'p' },
1476 { "record", required_argument, NULL, 'r' },
1477 { value_str, required_argument, NULL, 0 },
1478 { system_boot_reason_str, no_argument, NULL, 0 },
1479 { boot_complete_str, no_argument, NULL, 0 },
1480 { boot_reason_str, no_argument, NULL, 0 },
1481 { factory_reset_str, no_argument, NULL, 0 },
1482 { boot_reason_enum_str, optional_argument, NULL, 0 },
1483 { NULL, 0, NULL, 0 }
1484 // clang-format on
1485 };
1486
1487 std::string boot_event;
1488 std::string value;
1489 int opt = 0;
1490 while ((opt = getopt_long(argc, argv, "hlpr:", long_options, &option_index)) != -1) {
1491 switch (opt) {
1492 // This case handles long options which have no single-character mapping.
1493 case 0: {
1494 const std::string option_name = long_options[option_index].name;
1495 if (option_name == value_str) {
1496 // |optarg| is an external variable set by getopt representing
1497 // the option argument.
1498 value = optarg;
1499 } else if (option_name == system_boot_reason_str) {
1500 SetSystemBootReason();
1501 } else if (option_name == boot_complete_str) {
1502 RecordBootComplete();
1503 } else if (option_name == boot_reason_str) {
1504 RecordBootReason();
1505 } else if (option_name == factory_reset_str) {
1506 RecordFactoryReset();
1507 } else if (option_name == boot_reason_enum_str) {
1508 PrintBootReasonEnum(optarg);
1509 } else {
1510 LOG(ERROR) << "Invalid option: " << option_name;
1511 }
1512 break;
1513 }
1514
1515 case 'h': {
1516 ShowHelp(argv[0]);
1517 break;
1518 }
1519
1520 case 'l': {
1521 LogBootEvents();
1522 break;
1523 }
1524
1525 case 'p': {
1526 PrintBootEvents();
1527 break;
1528 }
1529
1530 case 'r': {
1531 // |optarg| is an external variable set by getopt representing
1532 // the option argument.
1533 boot_event = optarg;
1534 break;
1535 }
1536
1537 default: {
1538 DCHECK_EQ(opt, '?');
1539
1540 // |optopt| is an external variable set by getopt representing
1541 // the value of the invalid option.
1542 LOG(ERROR) << "Invalid option: " << optopt;
1543 ShowHelp(argv[0]);
1544 return EXIT_FAILURE;
1545 }
1546 }
1547 }
1548
1549 if (!boot_event.empty()) {
1550 RecordBootEventFromCommandLine(boot_event, value);
1551 }
1552
1553 return 0;
1554 }
1555