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 };
440
441 // Converts a string value representing the reason the system booted to an
442 // integer representation. This is necessary for logging the boot_reason metric
443 // via Tron, which does not accept non-integer buckets in histograms.
BootReasonStrToEnum(const std::string & boot_reason)444 int32_t BootReasonStrToEnum(const std::string& boot_reason) {
445 auto mapping = kBootReasonMap.find(boot_reason);
446 if (mapping != kBootReasonMap.end()) {
447 return mapping->second;
448 }
449
450 if (boot_reason.empty()) {
451 return kEmptyBootReason;
452 }
453
454 for (const auto& [match, id] : kBootReasonMap) {
455 // Regex matches as a minimum require either [, \ or * to be present.
456 if (match.find_first_of("[\\*") == match.npos) continue;
457 // enforce match from beginning to end
458 auto exact = match;
459 if (exact[0] != '^') exact = "^" + exact;
460 if (exact[exact.size() - 1] != '$') exact = exact + "$";
461 if (std::regex_search(boot_reason, std::regex(exact))) return id;
462 }
463
464 LOG(INFO) << "Unknown boot reason: " << boot_reason;
465 return kUnknownBootReason;
466 }
467
468 // Canonical list of supported primary reboot reasons.
469 const std::vector<const std::string> knownReasons = {
470 // clang-format off
471 // kernel
472 "watchdog",
473 "kernel_panic",
474 // strong
475 "recovery", // Should not happen from ro.boot.bootreason
476 "bootloader", // Should not happen from ro.boot.bootreason
477 // blunt
478 "cold",
479 "hard",
480 "warm",
481 // super blunt
482 "shutdown", // Can not happen from ro.boot.bootreason
483 "reboot", // Default catch-all for anything unknown
484 // clang-format on
485 };
486
487 // Returns true if the supplied reason prefix is considered detailed enough.
isStrongRebootReason(const std::string & r)488 bool isStrongRebootReason(const std::string& r) {
489 for (auto& s : knownReasons) {
490 if (s == "cold") break;
491 // Prefix defined as terminated by a nul or comma (,).
492 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
493 return true;
494 }
495 }
496 return false;
497 }
498
499 // Returns true if the supplied reason prefix is associated with the kernel.
isKernelRebootReason(const std::string & r)500 bool isKernelRebootReason(const std::string& r) {
501 for (auto& s : knownReasons) {
502 if (s == "recovery") break;
503 // Prefix defined as terminated by a nul or comma (,).
504 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
505 return true;
506 }
507 }
508 return false;
509 }
510
511 // Returns true if the supplied reason prefix is considered known.
isKnownRebootReason(const std::string & r)512 bool isKnownRebootReason(const std::string& r) {
513 for (auto& s : knownReasons) {
514 // Prefix defined as terminated by a nul or comma (,).
515 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
516 return true;
517 }
518 }
519 return false;
520 }
521
522 // If the reboot reason should be improved, report true if is too blunt.
isBluntRebootReason(const std::string & r)523 bool isBluntRebootReason(const std::string& r) {
524 if (isStrongRebootReason(r)) return false;
525
526 if (!isKnownRebootReason(r)) return true; // Can not support unknown as detail
527
528 size_t pos = 0;
529 while ((pos = r.find(',', pos)) != std::string::npos) {
530 ++pos;
531 std::string next(r.substr(pos));
532 if (next.length() == 0) break;
533 if (next[0] == ',') continue;
534 if (!isKnownRebootReason(next)) return false; // Unknown subreason is good.
535 if (isStrongRebootReason(next)) return false; // eg: reboot,reboot
536 }
537 return true;
538 }
539
readPstoreConsole(std::string & console)540 bool readPstoreConsole(std::string& console) {
541 if (android::base::ReadFileToString("/sys/fs/pstore/console-ramoops-0", &console)) {
542 return true;
543 }
544 return android::base::ReadFileToString("/sys/fs/pstore/console-ramoops", &console);
545 }
546
547 // Implement a variant of std::string::rfind that is resilient to errors in
548 // the data stream being inspected.
549 class pstoreConsole {
550 private:
551 const size_t kBitErrorRate = 8; // number of bits per error
552 const std::string& console;
553
554 // Number of bits that differ between the two arguments l and r.
555 // Returns zero if the values for l and r are identical.
numError(uint8_t l,uint8_t r) const556 size_t numError(uint8_t l, uint8_t r) const { return std::bitset<8>(l ^ r).count(); }
557
558 // A string comparison function, reports the number of errors discovered
559 // in the match to a maximum of the bitLength / kBitErrorRate, at that
560 // point returning npos to indicate match is too poor.
561 //
562 // Since called in rfind which works backwards, expect cache locality will
563 // help if we check in reverse here as well for performance.
564 //
565 // Assumption: l (from console.c_str() + pos) is long enough to house
566 // _r.length(), checked in rfind caller below.
567 //
numError(size_t pos,const std::string & _r) const568 size_t numError(size_t pos, const std::string& _r) const {
569 const char* l = console.c_str() + pos;
570 const char* r = _r.c_str();
571 size_t n = _r.length();
572 const uint8_t* le = reinterpret_cast<const uint8_t*>(l) + n;
573 const uint8_t* re = reinterpret_cast<const uint8_t*>(r) + n;
574 size_t count = 0;
575 n = 0;
576 do {
577 // individual character bit error rate > threshold + slop
578 size_t num = numError(*--le, *--re);
579 if (num > ((8 + kBitErrorRate) / kBitErrorRate)) return std::string::npos;
580 // total bit error rate > threshold + slop
581 count += num;
582 ++n;
583 if (count > ((n * 8 + kBitErrorRate - (n > 2)) / kBitErrorRate)) {
584 return std::string::npos;
585 }
586 } while (le != reinterpret_cast<const uint8_t*>(l));
587 return count;
588 }
589
590 public:
pstoreConsole(const std::string & console)591 explicit pstoreConsole(const std::string& console) : console(console) {}
592 // scope of argument must be equal to or greater than scope of pstoreConsole
593 explicit pstoreConsole(const std::string&& console) = delete;
594 explicit pstoreConsole(std::string&& console) = delete;
595
596 // Our implementation of rfind, use exact match first, then resort to fuzzy.
rfind(const std::string & needle) const597 size_t rfind(const std::string& needle) const {
598 size_t pos = console.rfind(needle); // exact match?
599 if (pos != std::string::npos) return pos;
600
601 // Check to make sure needle fits in console string.
602 pos = console.length();
603 if (needle.length() > pos) return std::string::npos;
604 pos -= needle.length();
605 // fuzzy match to maximum kBitErrorRate
606 for (;;) {
607 if (numError(pos, needle) != std::string::npos) return pos;
608 if (pos == 0) break;
609 --pos;
610 }
611 return std::string::npos;
612 }
613
614 // Our implementation of find, use only fuzzy match.
find(const std::string & needle,size_t start=0) const615 size_t find(const std::string& needle, size_t start = 0) const {
616 // Check to make sure needle fits in console string.
617 if (needle.length() > console.length()) return std::string::npos;
618 const size_t last_pos = console.length() - needle.length();
619 // fuzzy match to maximum kBitErrorRate
620 for (size_t pos = start; pos <= last_pos; ++pos) {
621 if (numError(pos, needle) != std::string::npos) return pos;
622 }
623 return std::string::npos;
624 }
625
operator const std::string&() const626 operator const std::string&() const { return console; }
627 };
628
629 // If bit error match to needle, correct it.
630 // Return true if any corrections were discovered and applied.
correctForBitError(std::string & reason,const std::string & needle)631 bool correctForBitError(std::string& reason, const std::string& needle) {
632 bool corrected = false;
633 if (reason.length() < needle.length()) return corrected;
634 const pstoreConsole console(reason);
635 const size_t last_pos = reason.length() - needle.length();
636 for (size_t pos = 0; pos <= last_pos; pos += needle.length()) {
637 pos = console.find(needle, pos);
638 if (pos == std::string::npos) break;
639
640 // exact match has no malice
641 if (needle == reason.substr(pos, needle.length())) continue;
642
643 corrected = true;
644 reason = reason.substr(0, pos) + needle + reason.substr(pos + needle.length());
645 }
646 return corrected;
647 }
648
649 // If bit error match to needle, correct it.
650 // Return true if any corrections were discovered and applied.
651 // Try again if we can replace underline with spaces.
correctForBitErrorOrUnderline(std::string & reason,const std::string & needle)652 bool correctForBitErrorOrUnderline(std::string& reason, const std::string& needle) {
653 bool corrected = correctForBitError(reason, needle);
654 std::string _needle(needle);
655 std::transform(_needle.begin(), _needle.end(), _needle.begin(),
656 [](char c) { return (c == '_') ? ' ' : c; });
657 if (needle != _needle) {
658 corrected |= correctForBitError(reason, _needle);
659 }
660 return corrected;
661 }
662
663 // Converts a string value representing the reason the system booted to a
664 // string complying with Android system standard reason.
transformReason(std::string & reason)665 void transformReason(std::string& reason) {
666 std::transform(reason.begin(), reason.end(), reason.begin(), ::tolower);
667 std::transform(reason.begin(), reason.end(), reason.begin(),
668 [](char c) { return ::isblank(c) ? '_' : c; });
669 std::transform(reason.begin(), reason.end(), reason.begin(),
670 [](char c) { return ::isprint(c) ? c : '?'; });
671 }
672
673 // Check subreasons for reboot,<subreason> kernel_panic,sysrq,<subreason> or
674 // kernel_panic,<subreason>.
675 //
676 // If quoted flag is set, pull out and correct single quoted ('), newline (\n)
677 // or unprintable character terminated subreason, pos is supplied just beyond
678 // first quote. if quoted false, pull out and correct newline (\n) or
679 // unprintable character terminated subreason.
680 //
681 // Heuristics to find termination is painted into a corner:
682
683 // single bit error for quote ' that we can block. It is acceptable for
684 // the others 7, g in reason. 2/9 chance will miss the terminating quote,
685 // but there is always the terminating newline that usually immediately
686 // follows to fortify our chances.
likely_single_quote(char c)687 bool likely_single_quote(char c) {
688 switch (static_cast<uint8_t>(c)) {
689 case '\'': // '\''
690 case '\'' ^ 0x01: // '&'
691 case '\'' ^ 0x02: // '%'
692 case '\'' ^ 0x04: // '#'
693 case '\'' ^ 0x08: // '/'
694 return true;
695 case '\'' ^ 0x10: // '7'
696 break;
697 case '\'' ^ 0x20: // '\a' (unprintable)
698 return true;
699 case '\'' ^ 0x40: // 'g'
700 break;
701 case '\'' ^ 0x80: // 0xA7 (unprintable)
702 return true;
703 }
704 return false;
705 }
706
707 // ::isprint(c) and likely_space() will prevent us from being called for
708 // fundamentally printable entries, except for '\r' and '\b'.
709 //
710 // Except for * and J, single bit errors for \n, all others are non-
711 // printable so easy catch. It is _acceptable_ for *, J or j to exist in
712 // the reason string, so 2/9 chance we will miss the terminating newline.
713 //
714 // NB: J might not be acceptable, except if at the beginning or preceded
715 // with a space, '(' or any of the quotes and their BER aliases.
716 // NB: * might not be acceptable, except if at the beginning or preceded
717 // with a space, another *, or any of the quotes or their BER aliases.
718 //
719 // To reduce the chances to closer to 1/9 is too complicated for the gain.
likely_newline(char c)720 bool likely_newline(char c) {
721 switch (static_cast<uint8_t>(c)) {
722 case '\n': // '\n' (unprintable)
723 case '\n' ^ 0x01: // '\r' (unprintable)
724 case '\n' ^ 0x02: // '\b' (unprintable)
725 case '\n' ^ 0x04: // 0x0E (unprintable)
726 case '\n' ^ 0x08: // 0x02 (unprintable)
727 case '\n' ^ 0x10: // 0x1A (unprintable)
728 return true;
729 case '\n' ^ 0x20: // '*'
730 case '\n' ^ 0x40: // 'J'
731 break;
732 case '\n' ^ 0x80: // 0x8A (unprintable)
733 return true;
734 }
735 return false;
736 }
737
738 // ::isprint(c) will prevent us from being called for all the printable
739 // matches below. If we let unprintables through because of this, they
740 // get converted to underscore (_) by the validation phase.
likely_space(char c)741 bool likely_space(char c) {
742 switch (static_cast<uint8_t>(c)) {
743 case ' ': // ' '
744 case ' ' ^ 0x01: // '!'
745 case ' ' ^ 0x02: // '"'
746 case ' ' ^ 0x04: // '$'
747 case ' ' ^ 0x08: // '('
748 case ' ' ^ 0x10: // '0'
749 case ' ' ^ 0x20: // '\0' (unprintable)
750 case ' ' ^ 0x40: // 'P'
751 case ' ' ^ 0x80: // 0xA0 (unprintable)
752 case '\t': // '\t'
753 case '\t' ^ 0x01: // '\b' (unprintable) (likely_newline counters)
754 case '\t' ^ 0x02: // '\v' (unprintable)
755 case '\t' ^ 0x04: // '\r' (unprintable) (likely_newline counters)
756 case '\t' ^ 0x08: // 0x01 (unprintable)
757 case '\t' ^ 0x10: // 0x19 (unprintable)
758 case '\t' ^ 0x20: // ')'
759 case '\t' ^ 0x40: // '1'
760 case '\t' ^ 0x80: // 0x89 (unprintable)
761 return true;
762 }
763 return false;
764 }
765
getSubreason(const std::string & content,size_t pos,bool quoted)766 std::string getSubreason(const std::string& content, size_t pos, bool quoted) {
767 static constexpr size_t max_reason_length = 256;
768
769 std::string subReason(content.substr(pos, max_reason_length));
770 // Correct against any known strings that Bit Error Match
771 for (const auto& s : knownReasons) {
772 correctForBitErrorOrUnderline(subReason, s);
773 }
774 std::string terminator(quoted ? "'" : "");
775 for (const auto& m : kBootReasonMap) {
776 if (m.first.length() <= strlen("cold")) continue; // too short?
777 if (correctForBitErrorOrUnderline(subReason, m.first + terminator)) continue;
778 if (m.first.length() <= strlen("reboot,cold")) continue; // short?
779 if (android::base::StartsWith(m.first, "reboot,")) {
780 correctForBitErrorOrUnderline(subReason, m.first.substr(strlen("reboot,")) + terminator);
781 } else if (android::base::StartsWith(m.first, "kernel_panic,sysrq,")) {
782 correctForBitErrorOrUnderline(subReason,
783 m.first.substr(strlen("kernel_panic,sysrq,")) + terminator);
784 } else if (android::base::StartsWith(m.first, "kernel_panic,")) {
785 correctForBitErrorOrUnderline(subReason, m.first.substr(strlen("kernel_panic,")) + terminator);
786 }
787 }
788 for (pos = 0; pos < subReason.length(); ++pos) {
789 char c = subReason[pos];
790 if (!(::isprint(c) || likely_space(c)) || likely_newline(c) ||
791 (quoted && likely_single_quote(c))) {
792 subReason.erase(pos);
793 break;
794 }
795 }
796 transformReason(subReason);
797 return subReason;
798 }
799
addKernelPanicSubReason(const pstoreConsole & console,std::string & ret)800 bool addKernelPanicSubReason(const pstoreConsole& console, std::string& ret) {
801 // Check for kernel panic types to refine information
802 if ((console.rfind("SysRq : Trigger a crash") != std::string::npos) ||
803 (console.rfind("PC is at sysrq_handle_crash+") != std::string::npos)) {
804 ret = "kernel_panic,sysrq";
805 // Invented for Android to allow daemons that specifically trigger sysrq
806 // to communicate more accurate boot subreasons via last console messages.
807 static constexpr char sysrqSubreason[] = "SysRq : Trigger a crash : '";
808 auto pos = console.rfind(sysrqSubreason);
809 if (pos != std::string::npos) {
810 ret += "," + getSubreason(console, pos + strlen(sysrqSubreason), /* quoted */ true);
811 }
812 return true;
813 }
814 if (console.rfind("Unable to handle kernel NULL pointer dereference at virtual address") !=
815 std::string::npos) {
816 ret = "kernel_panic,null";
817 return true;
818 }
819 if (console.rfind("Kernel BUG at ") != std::string::npos) {
820 ret = "kernel_panic,bug";
821 return true;
822 }
823
824 std::string panic("Kernel panic - not syncing: ");
825 auto pos = console.rfind(panic);
826 if (pos != std::string::npos) {
827 static const std::vector<std::pair<const std::string, const std::string>> panicReasons = {
828 {"Out of memory", "oom"},
829 {"out of memory", "oom"},
830 {"Oh boy, that early out of memory", "oom"}, // omg
831 {"BUG!", "bug"},
832 {"hung_task: blocked tasks", "hung"},
833 {"audit: ", "audit"},
834 {"scheduling while atomic", "atomic"},
835 {"Attempted to kill init!", "init"},
836 {"Requested init", "init"},
837 {"No working init", "init"},
838 {"Could not decompress init", "init"},
839 {"RCU Stall", "hung,rcu"},
840 {"stack-protector", "stack"},
841 {"kernel stack overflow", "stack"},
842 {"Corrupt kernel stack", "stack"},
843 {"low stack detected", "stack"},
844 {"corrupted stack end", "stack"},
845 {"subsys-restart: Resetting the SoC - modem crashed.", "modem"},
846 {"subsys-restart: Resetting the SoC - adsp crashed.", "adsp"},
847 {"subsys-restart: Resetting the SoC - dsps crashed.", "dsps"},
848 {"subsys-restart: Resetting the SoC - wcnss crashed.", "wcnss"},
849 };
850
851 ret = "kernel_panic";
852 for (auto& s : panicReasons) {
853 if (console.find(panic + s.first, pos) != std::string::npos) {
854 ret += "," + s.second;
855 return true;
856 }
857 }
858 auto reason = getSubreason(console, pos + panic.length(), /* newline */ false);
859 if (reason.length() > 3) {
860 ret += "," + reason;
861 }
862 return true;
863 }
864 return false;
865 }
866
addKernelPanicSubReason(const std::string & content,std::string & ret)867 bool addKernelPanicSubReason(const std::string& content, std::string& ret) {
868 return addKernelPanicSubReason(pstoreConsole(content), ret);
869 }
870
871 const char system_reboot_reason_property[] = "sys.boot.reason";
872 const char last_reboot_reason_property[] = LAST_REBOOT_REASON_PROPERTY;
873 const char last_reboot_reason_file[] = LAST_REBOOT_REASON_FILE;
874 const char last_last_reboot_reason_property[] = "sys.boot.reason.last";
875 constexpr size_t history_reboot_reason_size = 4;
876 const char history_reboot_reason_property[] = LAST_REBOOT_REASON_PROPERTY ".history";
877 const char bootloader_reboot_reason_property[] = "ro.boot.bootreason";
878
879 // Land system_boot_reason into system_reboot_reason_property.
880 // Shift system_boot_reason into history_reboot_reason_property.
BootReasonAddToHistory(const std::string & system_boot_reason)881 void BootReasonAddToHistory(const std::string& system_boot_reason) {
882 if (system_boot_reason.empty()) return;
883 LOG(INFO) << "Canonical boot reason: " << system_boot_reason;
884 auto old_system_boot_reason = android::base::GetProperty(system_reboot_reason_property, "");
885 if (!android::base::SetProperty(system_reboot_reason_property, system_boot_reason)) {
886 android::base::SetProperty(system_reboot_reason_property,
887 system_boot_reason.substr(0, PROPERTY_VALUE_MAX - 1));
888 }
889 auto reason_history =
890 android::base::Split(android::base::GetProperty(history_reboot_reason_property, ""), "\n");
891 static auto mark = time(nullptr);
892 auto mark_str = std::string(",") + std::to_string(mark);
893 auto marked_system_boot_reason = system_boot_reason + mark_str;
894 if (!reason_history.empty()) {
895 // delete any entries that we just wrote in a previous
896 // call and leveraging duplicate line handling
897 auto last = old_system_boot_reason + mark_str;
898 // trim the list to (history_reboot_reason_size - 1)
899 ssize_t max = history_reboot_reason_size;
900 for (auto it = reason_history.begin(); it != reason_history.end();) {
901 if (it->empty() || (last == *it) || (marked_system_boot_reason == *it) || (--max <= 0)) {
902 it = reason_history.erase(it);
903 } else {
904 last = *it;
905 ++it;
906 }
907 }
908 }
909 // insert at the front, concatenating mark (<epoch time>) detail to the value.
910 reason_history.insert(reason_history.begin(), marked_system_boot_reason);
911 // If the property string is too long ( > PROPERTY_VALUE_MAX)
912 // we get an error, so trim out last entry and try again.
913 while (!android::base::SetProperty(history_reboot_reason_property,
914 android::base::Join(reason_history, '\n'))) {
915 auto it = std::prev(reason_history.end());
916 if (it == reason_history.end()) break;
917 reason_history.erase(it);
918 }
919 }
920
921 // Scrub, Sanitize, Standardize and Enhance the boot reason string supplied.
BootReasonStrToReason(const std::string & boot_reason)922 std::string BootReasonStrToReason(const std::string& boot_reason) {
923 auto ret = android::base::GetProperty(system_reboot_reason_property, "");
924 std::string reason(boot_reason);
925 // If sys.boot.reason == ro.boot.bootreason, let's re-evaluate
926 if (reason == ret) ret = "";
927
928 transformReason(reason);
929
930 // Is the current system boot reason sys.boot.reason valid?
931 if (!isKnownRebootReason(ret)) ret = "";
932
933 if (ret == "") {
934 // Is the bootloader boot reason ro.boot.bootreason known?
935 std::vector<std::string> words(android::base::Split(reason, ",_-"));
936 for (auto& s : knownReasons) {
937 std::string blunt;
938 for (auto& r : words) {
939 if (r == s) {
940 if (isBluntRebootReason(s)) {
941 blunt = s;
942 } else {
943 ret = s;
944 break;
945 }
946 }
947 }
948 if (ret == "") ret = blunt;
949 if (ret != "") break;
950 }
951 }
952
953 if (ret == "") {
954 // A series of checks to take some officially unsupported reasons
955 // reported by the bootloader and find some logical and canonical
956 // sense. In an ideal world, we would require those bootloaders
957 // to behave and follow our CTS standards.
958 //
959 // first member is the output
960 // second member is an unanchored regex for an alias
961 //
962 // If output has a prefix of <bang> '!', we do not use it as a
963 // match needle (and drop the <bang> prefix when landing in output),
964 // otherwise look for it as well. This helps keep the scale of the
965 // following table smaller.
966 static const std::vector<std::pair<const std::string, const std::string>> aliasReasons = {
967 {"watchdog", "wdog"},
968 {"kernel_panic", "panic"},
969 {"shutdown,thermal", "thermal"},
970 {"warm,s3_wakeup", "s3_wakeup"},
971 {"hard,hw_reset", "hw_reset"},
972 {"cold,charger", "usb|power_on_cable"},
973 {"cold,powerkey", "powerkey|power_key|PowerKey|power_on"},
974 {"cold,rtc", "rtc"},
975 {"cold,rtc,2sec", "2sec_reboot"},
976 {"!warm", "wdt_by_pass_pwk"}, // change flavour of blunt
977 {"!reboot", "^wdt$"}, // change flavour of blunt
978 {"reboot,tool", "tool_by_pass_pwk"},
979 {"!reboot,longkey", "reboot_longkey"},
980 {"!reboot,longkey", "kpdpwr"},
981 {"!reboot,undervoltage", "uvlo"},
982 {"!reboot,powerloss", "smpl"},
983 {"bootloader", ""},
984 };
985
986 for (auto& s : aliasReasons) {
987 size_t firstHasNot = s.first[0] == '!';
988 if (!firstHasNot && (reason.find(s.first) != std::string::npos)) {
989 ret = s.first;
990 break;
991 }
992 if (s.second.size() && std::regex_search(reason, std::regex(s.second))) {
993 ret = s.first.substr(firstHasNot);
994 break;
995 }
996 }
997 }
998
999 // If watchdog is the reason, see if there is a security angle?
1000 if (ret == "watchdog") {
1001 if (reason.find("sec") != std::string::npos) {
1002 ret += ",security";
1003 }
1004 }
1005
1006 if (ret == "kernel_panic") {
1007 // Check to see if last klog has some refinement hints.
1008 std::string content;
1009 if (readPstoreConsole(content)) {
1010 addKernelPanicSubReason(content, ret);
1011 }
1012 } else if (isBluntRebootReason(ret)) {
1013 // Check the other available reason resources if the reason is still blunt.
1014
1015 // Check to see if last klog has some refinement hints.
1016 std::string content;
1017 if (readPstoreConsole(content)) {
1018 const pstoreConsole console(content);
1019 // The toybox reboot command used directly (unlikely)? But also
1020 // catches init's response to Android's more controlled reboot command.
1021 if (console.rfind("reboot: Power down") != std::string::npos) {
1022 ret = "shutdown"; // Still too blunt, but more accurate.
1023 // ToDo: init should record the shutdown reason to kernel messages ala:
1024 // init: shutdown system with command 'last_reboot_reason'
1025 // so that if pstore has persistence we can get some details
1026 // that could be missing in last_reboot_reason_property.
1027 }
1028
1029 static const char cmd[] = "reboot: Restarting system with command '";
1030 size_t pos = console.rfind(cmd);
1031 if (pos != std::string::npos) {
1032 std::string subReason(getSubreason(content, pos + strlen(cmd), /* quoted */ true));
1033 if (subReason != "") { // Will not land "reboot" as that is too blunt.
1034 if (isKernelRebootReason(subReason)) {
1035 ret = "reboot," + subReason; // User space can't talk kernel reasons.
1036 } else if (isKnownRebootReason(subReason)) {
1037 ret = subReason;
1038 } else {
1039 ret = "reboot," + subReason; // legitimize unknown reasons
1040 }
1041 }
1042 // Some bootloaders shutdown results record in last kernel message.
1043 if (!strcmp(ret.c_str(), "reboot,kernel_power_off_charging__reboot_system")) {
1044 ret = "shutdown";
1045 }
1046 }
1047
1048 // Check for kernel panics, allowed to override reboot command.
1049 if (!addKernelPanicSubReason(console, ret) &&
1050 // check for long-press power down
1051 ((console.rfind("Power held for ") != std::string::npos) ||
1052 (console.rfind("charger: [") != std::string::npos))) {
1053 ret = "cold";
1054 }
1055 }
1056
1057 // TODO: use the HAL to get battery level (http://b/77725702).
1058
1059 // Is there a controlled shutdown hint in last_reboot_reason_property?
1060 if (isBluntRebootReason(ret)) {
1061 // Content buffer no longer will have console data. Beware if more
1062 // checks added below, that depend on parsing console content.
1063 if (!android::base::ReadFileToString(last_reboot_reason_file, &content)) {
1064 content = android::base::GetProperty(last_reboot_reason_property, "");
1065 }
1066 transformReason(content);
1067
1068 // Anything in last is better than 'super-blunt' reboot or shutdown.
1069 if ((ret == "") || (ret == "reboot") || (ret == "shutdown") || !isBluntRebootReason(content)) {
1070 ret = content;
1071 }
1072 }
1073
1074 // Other System Health HAL reasons?
1075
1076 // ToDo: /proc/sys/kernel/boot_reason needs a HAL interface to
1077 // possibly offer hardware-specific clues from the PMIC.
1078 }
1079
1080 // If unknown left over from above, make it "reboot,<boot_reason>"
1081 if (ret == "") {
1082 ret = "reboot";
1083 if (android::base::StartsWith(reason, "reboot")) {
1084 reason = reason.substr(strlen("reboot"));
1085 while ((reason[0] == ',') || (reason[0] == '_')) {
1086 reason = reason.substr(1);
1087 }
1088 }
1089 if (reason != "") {
1090 ret += ",";
1091 ret += reason;
1092 }
1093 }
1094
1095 LOG(INFO) << "Canonical boot reason: " << ret;
1096 return ret;
1097 }
1098
1099 // Returns the appropriate metric key prefix for the boot_complete metric such
1100 // that boot metrics after a system update are labeled as ota_boot_complete;
1101 // otherwise, they are labeled as boot_complete. This method encapsulates the
1102 // bookkeeping required to track when a system update has occurred by storing
1103 // the UTC timestamp of the system build date and comparing against the current
1104 // system build date.
CalculateBootCompletePrefix()1105 std::string CalculateBootCompletePrefix() {
1106 static const std::string kBuildDateKey = "build_date";
1107 std::string boot_complete_prefix = "boot_complete";
1108
1109 auto build_date_str = android::base::GetProperty("ro.build.date.utc", "");
1110 int32_t build_date;
1111 if (!android::base::ParseInt(build_date_str, &build_date)) {
1112 return std::string();
1113 }
1114
1115 BootEventRecordStore boot_event_store;
1116 BootEventRecordStore::BootEventRecord record;
1117 if (!boot_event_store.GetBootEvent(kBuildDateKey, &record)) {
1118 boot_complete_prefix = "factory_reset_" + boot_complete_prefix;
1119 boot_event_store.AddBootEventWithValue(kBuildDateKey, build_date);
1120 BootReasonAddToHistory("reboot,factory_reset");
1121 } else if (build_date != record.second) {
1122 boot_complete_prefix = "ota_" + boot_complete_prefix;
1123 boot_event_store.AddBootEventWithValue(kBuildDateKey, build_date);
1124 BootReasonAddToHistory("reboot,ota");
1125 }
1126
1127 return boot_complete_prefix;
1128 }
1129
1130 // Records the value of a given ro.boottime.init property in milliseconds.
RecordInitBootTimeProp(BootEventRecordStore * boot_event_store,const char * property)1131 void RecordInitBootTimeProp(BootEventRecordStore* boot_event_store, const char* property) {
1132 auto value = android::base::GetProperty(property, "");
1133
1134 int32_t time_in_ms;
1135 if (android::base::ParseInt(value, &time_in_ms)) {
1136 boot_event_store->AddBootEventWithValue(property, time_in_ms);
1137 }
1138 }
1139
1140 // A map from bootloader timing stage to the time that stage took during boot.
1141 typedef std::map<std::string, int32_t> BootloaderTimingMap;
1142
1143 // Returns a mapping from bootloader stage names to the time those stages
1144 // took to boot.
GetBootLoaderTimings()1145 const BootloaderTimingMap GetBootLoaderTimings() {
1146 BootloaderTimingMap timings;
1147
1148 // |ro.boot.boottime| is of the form 'stage1:time1,...,stageN:timeN',
1149 // where timeN is in milliseconds.
1150 auto value = android::base::GetProperty("ro.boot.boottime", "");
1151 if (value.empty()) {
1152 // ro.boot.boottime is not reported on all devices.
1153 return BootloaderTimingMap();
1154 }
1155
1156 auto stages = android::base::Split(value, ",");
1157 for (const auto& stageTiming : stages) {
1158 // |stageTiming| is of the form 'stage:time'.
1159 auto stageTimingValues = android::base::Split(stageTiming, ":");
1160 DCHECK_EQ(2U, stageTimingValues.size());
1161
1162 if (stageTimingValues.size() < 2) continue;
1163 std::string stageName = stageTimingValues[0];
1164 int32_t time_ms;
1165 if (android::base::ParseInt(stageTimingValues[1], &time_ms)) {
1166 timings[stageName] = time_ms;
1167 }
1168 }
1169
1170 return timings;
1171 }
1172
1173 // Returns the total bootloader boot time from the ro.boot.boottime system property.
GetBootloaderTime(const BootloaderTimingMap & bootloader_timings)1174 int32_t GetBootloaderTime(const BootloaderTimingMap& bootloader_timings) {
1175 int32_t total_time = 0;
1176 for (const auto& timing : bootloader_timings) {
1177 total_time += timing.second;
1178 }
1179
1180 return total_time;
1181 }
1182
1183 // Parses and records the set of bootloader stages and associated boot times
1184 // from the ro.boot.boottime system property.
RecordBootloaderTimings(BootEventRecordStore * boot_event_store,const BootloaderTimingMap & bootloader_timings)1185 void RecordBootloaderTimings(BootEventRecordStore* boot_event_store,
1186 const BootloaderTimingMap& bootloader_timings) {
1187 int32_t total_time = 0;
1188 for (const auto& timing : bootloader_timings) {
1189 total_time += timing.second;
1190 boot_event_store->AddBootEventWithValue("boottime.bootloader." + timing.first, timing.second);
1191 }
1192
1193 boot_event_store->AddBootEventWithValue("boottime.bootloader.total", total_time);
1194 }
1195
1196 // Returns the closest estimation to the absolute device boot time, i.e.,
1197 // from power on to boot_complete, including bootloader times.
GetAbsoluteBootTime(const BootloaderTimingMap & bootloader_timings,std::chrono::milliseconds uptime)1198 std::chrono::milliseconds GetAbsoluteBootTime(const BootloaderTimingMap& bootloader_timings,
1199 std::chrono::milliseconds uptime) {
1200 int32_t bootloader_time_ms = 0;
1201
1202 for (const auto& timing : bootloader_timings) {
1203 if (timing.first.compare("SW") != 0) {
1204 bootloader_time_ms += timing.second;
1205 }
1206 }
1207
1208 auto bootloader_duration = std::chrono::milliseconds(bootloader_time_ms);
1209 return bootloader_duration + uptime;
1210 }
1211
1212 // Records the closest estimation to the absolute device boot time in seconds.
1213 // i.e. from power on to boot_complete, including bootloader times.
RecordAbsoluteBootTime(BootEventRecordStore * boot_event_store,std::chrono::milliseconds absolute_total)1214 void RecordAbsoluteBootTime(BootEventRecordStore* boot_event_store,
1215 std::chrono::milliseconds absolute_total) {
1216 auto absolute_total_sec = std::chrono::duration_cast<std::chrono::seconds>(absolute_total);
1217 boot_event_store->AddBootEventWithValue("absolute_boot_time", absolute_total_sec.count());
1218 }
1219
1220 // 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)1221 void LogBootInfoToStatsd(std::chrono::milliseconds end_time,
1222 std::chrono::milliseconds total_duration, int32_t bootloader_duration_ms,
1223 double time_since_last_boot_sec) {
1224 auto reason = android::base::GetProperty(bootloader_reboot_reason_property, "<EMPTY>");
1225 auto system_reason = android::base::GetProperty(system_reboot_reason_property, "<EMPTY>");
1226 android::util::stats_write(android::util::BOOT_SEQUENCE_REPORTED, reason.c_str(),
1227 system_reason.c_str(), end_time.count(), total_duration.count(),
1228 (int64_t)bootloader_duration_ms,
1229 (int64_t)time_since_last_boot_sec * 1000);
1230 }
1231
SetSystemBootReason()1232 void SetSystemBootReason() {
1233 const auto bootloader_boot_reason =
1234 android::base::GetProperty(bootloader_reboot_reason_property, "");
1235 const std::string system_boot_reason(BootReasonStrToReason(bootloader_boot_reason));
1236 // Record the scrubbed system_boot_reason to the property
1237 BootReasonAddToHistory(system_boot_reason);
1238 // Shift last_reboot_reason_property to last_last_reboot_reason_property
1239 std::string last_boot_reason;
1240 if (!android::base::ReadFileToString(last_reboot_reason_file, &last_boot_reason)) {
1241 PLOG(ERROR) << "Failed to read " << last_reboot_reason_file;
1242 last_boot_reason = android::base::GetProperty(last_reboot_reason_property, "");
1243 LOG(INFO) << "Value of " << last_reboot_reason_property << " : " << last_boot_reason;
1244 } else {
1245 LOG(INFO) << "Last reboot reason read from " << last_reboot_reason_file << " : "
1246 << last_boot_reason << ". Last reboot reason read from "
1247 << last_reboot_reason_property << " : "
1248 << android::base::GetProperty(last_reboot_reason_property, "");
1249 }
1250 if (last_boot_reason.empty() || isKernelRebootReason(system_boot_reason)) {
1251 last_boot_reason = system_boot_reason;
1252 } else {
1253 transformReason(last_boot_reason);
1254 }
1255 LOG(INFO) << "Normalized last reboot reason : " << last_boot_reason;
1256 android::base::SetProperty(last_last_reboot_reason_property, last_boot_reason);
1257 android::base::SetProperty(last_reboot_reason_property, "");
1258 if (unlink(last_reboot_reason_file) != 0) {
1259 PLOG(ERROR) << "Failed to unlink " << last_reboot_reason_file;
1260 }
1261 }
1262
1263 // Gets the boot time offset. This is useful when Android is running in a
1264 // container, because the boot_clock is not reset when Android reboots.
GetBootTimeOffset()1265 std::chrono::nanoseconds GetBootTimeOffset() {
1266 static const int64_t boottime_offset =
1267 android::base::GetIntProperty<int64_t>("ro.boot.boottime_offset", 0);
1268 return std::chrono::nanoseconds(boottime_offset);
1269 }
1270
1271 // Returns the current uptime, accounting for any offset in the CLOCK_BOOTTIME
1272 // clock.
GetUptime()1273 android::base::boot_clock::duration GetUptime() {
1274 return android::base::boot_clock::now().time_since_epoch() - GetBootTimeOffset();
1275 }
1276
1277 // Records several metrics related to the time it takes to boot the device,
1278 // including disambiguating boot time on encrypted or non-encrypted devices.
RecordBootComplete()1279 void RecordBootComplete() {
1280 BootEventRecordStore boot_event_store;
1281 BootEventRecordStore::BootEventRecord record;
1282
1283 auto uptime_ns = GetUptime();
1284 auto uptime_s = std::chrono::duration_cast<std::chrono::seconds>(uptime_ns);
1285 time_t current_time_utc = time(nullptr);
1286 time_t time_since_last_boot = 0;
1287
1288 if (boot_event_store.GetBootEvent("last_boot_time_utc", &record)) {
1289 time_t last_boot_time_utc = record.second;
1290 time_since_last_boot = difftime(current_time_utc, last_boot_time_utc);
1291 boot_event_store.AddBootEventWithValue("time_since_last_boot", time_since_last_boot);
1292 }
1293
1294 boot_event_store.AddBootEventWithValue("last_boot_time_utc", current_time_utc);
1295
1296 // The boot_complete metric has two variants: boot_complete and
1297 // ota_boot_complete. The latter signifies that the device is booting after
1298 // a system update.
1299 std::string boot_complete_prefix = CalculateBootCompletePrefix();
1300 if (boot_complete_prefix.empty()) {
1301 // The system is hosed because the build date property could not be read.
1302 return;
1303 }
1304
1305 // post_decrypt_time_elapsed is only logged on encrypted devices.
1306 if (boot_event_store.GetBootEvent("post_decrypt_time_elapsed", &record)) {
1307 // Log the amount of time elapsed until the device is decrypted, which
1308 // includes the variable amount of time the user takes to enter the
1309 // decryption password.
1310 boot_event_store.AddBootEventWithValue("boot_decryption_complete", uptime_s.count());
1311
1312 // Subtract the decryption time to normalize the boot cycle timing.
1313 std::chrono::seconds boot_complete = std::chrono::seconds(uptime_s.count() - record.second);
1314 boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_post_decrypt",
1315 boot_complete.count());
1316 } else {
1317 boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_no_encryption",
1318 uptime_s.count());
1319 }
1320
1321 // Record the total time from device startup to boot complete, regardless of
1322 // encryption state.
1323 boot_event_store.AddBootEventWithValue(boot_complete_prefix, uptime_s.count());
1324
1325 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init");
1326 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.first_stage");
1327 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.selinux");
1328 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.cold_boot_wait");
1329
1330 const BootloaderTimingMap bootloader_timings = GetBootLoaderTimings();
1331 int32_t bootloader_boot_duration = GetBootloaderTime(bootloader_timings);
1332 RecordBootloaderTimings(&boot_event_store, bootloader_timings);
1333
1334 auto uptime_ms = std::chrono::duration_cast<std::chrono::milliseconds>(uptime_ns);
1335 auto absolute_boot_time = GetAbsoluteBootTime(bootloader_timings, uptime_ms);
1336 RecordAbsoluteBootTime(&boot_event_store, absolute_boot_time);
1337
1338 auto boot_end_time_point = std::chrono::system_clock::now().time_since_epoch();
1339 auto boot_end_time = std::chrono::duration_cast<std::chrono::milliseconds>(boot_end_time_point);
1340
1341 LogBootInfoToStatsd(boot_end_time, absolute_boot_time, bootloader_boot_duration,
1342 time_since_last_boot);
1343 }
1344
1345 // Records the boot_reason metric by querying the ro.boot.bootreason system
1346 // property.
RecordBootReason()1347 void RecordBootReason() {
1348 const auto reason = android::base::GetProperty(bootloader_reboot_reason_property, "");
1349
1350 if (reason.empty()) {
1351 // TODO(b/148575354): Replace with statsd.
1352 // Log an empty boot reason value as '<EMPTY>' to ensure the value is intentional
1353 // (and not corruption anywhere else in the reporting pipeline).
1354 // android::metricslogger::LogMultiAction(android::metricslogger::ACTION_BOOT,
1355 // android::metricslogger::FIELD_PLATFORM_REASON,
1356 // "<EMPTY>");
1357 } else {
1358 // TODO(b/148575354): Replace with statsd.
1359 // android::metricslogger::LogMultiAction(android::metricslogger::ACTION_BOOT,
1360 // android::metricslogger::FIELD_PLATFORM_REASON,
1361 // reason);
1362 }
1363
1364 // Log the raw bootloader_boot_reason property value.
1365 int32_t boot_reason = BootReasonStrToEnum(reason);
1366 BootEventRecordStore boot_event_store;
1367 boot_event_store.AddBootEventWithValue("boot_reason", boot_reason);
1368
1369 // Log the scrubbed system_boot_reason.
1370 const auto system_reason = android::base::GetProperty(system_reboot_reason_property, "");
1371 int32_t system_boot_reason = BootReasonStrToEnum(system_reason);
1372 boot_event_store.AddBootEventWithValue("system_boot_reason", system_boot_reason);
1373
1374 if (reason == "") {
1375 android::base::SetProperty(bootloader_reboot_reason_property, system_reason);
1376 }
1377 }
1378
1379 // Records two metrics related to the user resetting a device: the time at
1380 // which the device is reset, and the time since the user last reset the
1381 // device. The former is only set once per-factory reset.
RecordFactoryReset()1382 void RecordFactoryReset() {
1383 BootEventRecordStore boot_event_store;
1384 BootEventRecordStore::BootEventRecord record;
1385
1386 time_t current_time_utc = time(nullptr);
1387
1388 if (current_time_utc < 0) {
1389 // UMA does not display negative values in buckets, so convert to positive.
1390 // Logging via BootEventRecordStore.
1391 android::util::stats_write(
1392 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_ERROR_CODE_REPORTED),
1393 static_cast<int32_t>(
1394 android::util::BOOT_TIME_EVENT_ERROR_CODE__EVENT__FACTORY_RESET_CURRENT_TIME_FAILURE),
1395 static_cast<int32_t>(std::abs(current_time_utc)));
1396
1397 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1398 // is losing records somehow.
1399 boot_event_store.AddBootEventWithValue("factory_reset_current_time_failure",
1400 std::abs(current_time_utc));
1401 return;
1402 } else {
1403 android::util::stats_write(
1404 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED),
1405 static_cast<int32_t>(
1406 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_CURRENT_TIME),
1407 static_cast<int64_t>(current_time_utc));
1408
1409 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1410 // is losing records somehow.
1411 boot_event_store.AddBootEventWithValue("factory_reset_current_time", current_time_utc);
1412 }
1413
1414 // The factory_reset boot event does not exist after the device is reset, so
1415 // use this signal to mark the time of the factory reset.
1416 if (!boot_event_store.GetBootEvent("factory_reset", &record)) {
1417 boot_event_store.AddBootEventWithValue("factory_reset", current_time_utc);
1418
1419 // Don't log the time_since_factory_reset until some time has elapsed.
1420 // The data is not meaningful yet and skews the histogram buckets.
1421 return;
1422 }
1423
1424 // Calculate and record the difference in time between now and the
1425 // factory_reset time.
1426 time_t factory_reset_utc = record.second;
1427 android::util::stats_write(
1428 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED),
1429 static_cast<int32_t>(
1430 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_RECORD_VALUE),
1431 static_cast<int64_t>(factory_reset_utc));
1432
1433 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1434 // is losing records somehow.
1435 boot_event_store.AddBootEventWithValue("factory_reset_record_value", factory_reset_utc);
1436
1437 time_t time_since_factory_reset = difftime(current_time_utc, factory_reset_utc);
1438 boot_event_store.AddBootEventWithValue("time_since_factory_reset", time_since_factory_reset);
1439 }
1440
1441 // List the associated boot reason(s), if arg is nullptr then all.
PrintBootReasonEnum(const char * arg)1442 void PrintBootReasonEnum(const char* arg) {
1443 int value = -1;
1444 if (arg != nullptr) {
1445 value = BootReasonStrToEnum(arg);
1446 }
1447 for (const auto& [match, id] : kBootReasonMap) {
1448 if ((value < 0) || (value == id)) {
1449 printf("%u\t%s\n", id, match.c_str());
1450 }
1451 }
1452 }
1453
1454 } // namespace
1455
main(int argc,char ** argv)1456 int main(int argc, char** argv) {
1457 android::base::InitLogging(argv);
1458
1459 const std::string cmd_line = GetCommandLine(argc, argv);
1460 LOG(INFO) << "Service started: " << cmd_line;
1461
1462 int option_index = 0;
1463 static const char value_str[] = "value";
1464 static const char system_boot_reason_str[] = "set_system_boot_reason";
1465 static const char boot_complete_str[] = "record_boot_complete";
1466 static const char boot_reason_str[] = "record_boot_reason";
1467 static const char factory_reset_str[] = "record_time_since_factory_reset";
1468 static const char boot_reason_enum_str[] = "boot_reason_enum";
1469 static const struct option long_options[] = {
1470 // clang-format off
1471 { "help", no_argument, NULL, 'h' },
1472 { "log", no_argument, NULL, 'l' },
1473 { "print", no_argument, NULL, 'p' },
1474 { "record", required_argument, NULL, 'r' },
1475 { value_str, required_argument, NULL, 0 },
1476 { system_boot_reason_str, no_argument, NULL, 0 },
1477 { boot_complete_str, no_argument, NULL, 0 },
1478 { boot_reason_str, no_argument, NULL, 0 },
1479 { factory_reset_str, no_argument, NULL, 0 },
1480 { boot_reason_enum_str, optional_argument, NULL, 0 },
1481 { NULL, 0, NULL, 0 }
1482 // clang-format on
1483 };
1484
1485 std::string boot_event;
1486 std::string value;
1487 int opt = 0;
1488 while ((opt = getopt_long(argc, argv, "hlpr:", long_options, &option_index)) != -1) {
1489 switch (opt) {
1490 // This case handles long options which have no single-character mapping.
1491 case 0: {
1492 const std::string option_name = long_options[option_index].name;
1493 if (option_name == value_str) {
1494 // |optarg| is an external variable set by getopt representing
1495 // the option argument.
1496 value = optarg;
1497 } else if (option_name == system_boot_reason_str) {
1498 SetSystemBootReason();
1499 } else if (option_name == boot_complete_str) {
1500 RecordBootComplete();
1501 } else if (option_name == boot_reason_str) {
1502 RecordBootReason();
1503 } else if (option_name == factory_reset_str) {
1504 RecordFactoryReset();
1505 } else if (option_name == boot_reason_enum_str) {
1506 PrintBootReasonEnum(optarg);
1507 } else {
1508 LOG(ERROR) << "Invalid option: " << option_name;
1509 }
1510 break;
1511 }
1512
1513 case 'h': {
1514 ShowHelp(argv[0]);
1515 break;
1516 }
1517
1518 case 'l': {
1519 LogBootEvents();
1520 break;
1521 }
1522
1523 case 'p': {
1524 PrintBootEvents();
1525 break;
1526 }
1527
1528 case 'r': {
1529 // |optarg| is an external variable set by getopt representing
1530 // the option argument.
1531 boot_event = optarg;
1532 break;
1533 }
1534
1535 default: {
1536 DCHECK_EQ(opt, '?');
1537
1538 // |optopt| is an external variable set by getopt representing
1539 // the value of the invalid option.
1540 LOG(ERROR) << "Invalid option: " << optopt;
1541 ShowHelp(argv[0]);
1542 return EXIT_FAILURE;
1543 }
1544 }
1545 }
1546
1547 if (!boot_event.empty()) {
1548 RecordBootEventFromCommandLine(boot_event, value);
1549 }
1550
1551 return 0;
1552 }
1553