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/proto_logging/stats/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_complete_no_encryption",
71 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
72 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__BOOT_COMPLETE_NO_ENCRYPTION}},
73 {"factory_reset_boot_complete",
74 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
75 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__FACTORY_RESET_BOOT_COMPLETE}},
76 {"factory_reset_boot_complete_no_encryption",
77 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
78 android::util::
79 BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__FACTORY_RESET_BOOT_COMPLETE_NO_ENCRYPTION}},
80 {"ota_boot_complete",
81 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
82 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__OTA_BOOT_COMPLETE}},
83 {"ota_boot_complete_no_encryption",
84 {android::util::BOOT_TIME_EVENT_ELAPSED_TIME_REPORTED,
85 android::util::BOOT_TIME_EVENT_ELAPSED_TIME__EVENT__OTA_BOOT_COMPLETE_NO_ENCRYPTION}},
86 // DURATION
87 {"absolute_boot_time",
88 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
89 android::util::BOOT_TIME_EVENT_DURATION__EVENT__ABSOLUTE_BOOT_TIME}},
90 {"boottime.bootloader.1BLE",
91 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
92 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_FIRST_STAGE_EXEC}},
93 {"boottime.bootloader.1BLL",
94 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
95 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_FIRST_STAGE_LOAD}},
96 {"boottime.bootloader.KL",
97 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
98 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_KERNEL_LOAD}},
99 {"boottime.bootloader.2BLE",
100 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
101 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_SECOND_STAGE_EXEC}},
102 {"boottime.bootloader.2BLL",
103 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
104 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_SECOND_STAGE_LOAD}},
105 {"boottime.bootloader.SW",
106 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
107 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_UI_WAIT}},
108 {"boottime.bootloader.total",
109 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
110 android::util::BOOT_TIME_EVENT_DURATION__EVENT__BOOTLOADER_TOTAL}},
111 {"boottime.init.cold_boot_wait",
112 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
113 android::util::BOOT_TIME_EVENT_DURATION__EVENT__COLDBOOT_WAIT}},
114 {"time_since_factory_reset",
115 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
116 android::util::BOOT_TIME_EVENT_DURATION__EVENT__FACTORY_RESET_TIME_SINCE_RESET}},
117 {"ro.boottime.init.first_stage",
118 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
119 android::util::BOOT_TIME_EVENT_DURATION__EVENT__ANDROID_INIT_STAGE_1}},
120 {"ro.boottime.init.selinux",
121 {android::util::BOOT_TIME_EVENT_DURATION_REPORTED,
122 android::util::BOOT_TIME_EVENT_DURATION__EVENT__SELINUX_INIT}},
123 // UTC_TIME
124 {"factory_reset",
125 {android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED,
126 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_RESET_TIME}},
127 {"factory_reset_current_time",
128 {android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED,
129 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_CURRENT_TIME}},
130 {"factory_reset_record_value",
131 {android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED,
132 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_RECORD_VALUE}},
133 // ERROR_CODE
134 {"factory_reset_current_time_failure",
135 {android::util::BOOT_TIME_EVENT_ERROR_CODE_REPORTED,
136 android::util::BOOT_TIME_EVENT_ERROR_CODE__EVENT__FACTORY_RESET_CURRENT_TIME_FAILURE}},
137 };
138
139 // Scans the boot event record store for record files and logs each boot event
140 // via EventLog.
LogBootEvents()141 void LogBootEvents() {
142 BootEventRecordStore boot_event_store;
143 auto events = boot_event_store.GetAllBootEvents();
144 std::vector<std::string_view> notSupportedEvents;
145 for (const auto& event : events) {
146 const auto& name = event.first;
147 const auto& info = kBootEventToAtomInfo.find(name);
148 if (info != kBootEventToAtomInfo.end()) {
149 if (info->second.atom == android::util::BOOT_TIME_EVENT_ERROR_CODE_REPORTED) {
150 android::util::stats_write(static_cast<int32_t>(info->second.atom),
151 static_cast<int32_t>(info->second.event),
152 static_cast<int32_t>(event.second));
153 } else {
154 android::util::stats_write(static_cast<int32_t>(info->second.atom),
155 static_cast<int32_t>(info->second.event),
156 static_cast<int64_t>(event.second));
157 }
158 } else {
159 notSupportedEvents.push_back(name);
160 }
161 }
162 if (!notSupportedEvents.empty()) {
163 LOG(WARNING) << "LogBootEvents, atomInfo not defined for events:"
164 << android::base::Join(notSupportedEvents, ',');
165 }
166 }
167
168 // Records the named boot |event| to the record store. If |value| is non-empty
169 // and is a proper string representation of an integer value, the converted
170 // integer value is associated with the boot event.
RecordBootEventFromCommandLine(const std::string & event,const std::string & value_str)171 void RecordBootEventFromCommandLine(const std::string& event, const std::string& value_str) {
172 BootEventRecordStore boot_event_store;
173 if (!value_str.empty()) {
174 int32_t value = 0;
175 if (android::base::ParseInt(value_str, &value)) {
176 boot_event_store.AddBootEventWithValue(event, value);
177 }
178 } else {
179 boot_event_store.AddBootEvent(event);
180 }
181 }
182
PrintBootEvents()183 void PrintBootEvents() {
184 printf("Boot events:\n");
185 printf("------------\n");
186
187 BootEventRecordStore boot_event_store;
188 auto events = boot_event_store.GetAllBootEvents();
189 for (auto i = events.cbegin(); i != events.cend(); ++i) {
190 printf("%s\t%d\n", i->first.c_str(), i->second);
191 }
192 }
193
ShowHelp(const char * cmd)194 void ShowHelp(const char* cmd) {
195 fprintf(stderr, "Usage: %s [options]...\n", cmd);
196 fprintf(stderr,
197 "options include:\n"
198 " -h, --help Show this help\n"
199 " -l, --log Log all metrics to logstorage\n"
200 " -p, --print Dump the boot event records to the console\n"
201 " -r, --record Record the timestamp of a named boot event\n"
202 " --value Optional value to associate with the boot event\n"
203 " --record_boot_complete Record metrics related to the time for the device boot\n"
204 " --record_boot_reason Record the reason why the device booted\n"
205 " --record_time_since_factory_reset Record the time since the device was reset\n"
206 " --boot_reason_enum=<reason> Report the match to the kBootReasonMap table\n");
207 }
208
209 // Constructs a readable, printable string from the givencommand line
210 // arguments.
GetCommandLine(int argc,char ** argv)211 std::string GetCommandLine(int argc, char** argv) {
212 std::string cmd;
213 for (int i = 0; i < argc; ++i) {
214 cmd += argv[i];
215 cmd += " ";
216 }
217
218 return cmd;
219 }
220
221 constexpr int32_t kEmptyBootReason = 0;
222 constexpr int32_t kUnknownBootReason = 1;
223
224 // A mapping from boot reason string, as read from the ro.boot.bootreason
225 // system property, to a unique integer ID. Viewers of log data dashboards for
226 // the boot_reason metric may refer to this mapping to discern the histogram
227 // values. Regex matching, to manage the scale, as a minimum require either
228 // [, \ or * to be present in the string to switch to checking.
229 const std::map<std::string, int32_t> kBootReasonMap = {
230 {"reboot,[empty]", kEmptyBootReason},
231 {"__BOOTSTAT_UNKNOWN__", kUnknownBootReason},
232 {"normal", 2},
233 {"recovery", 3},
234 {"reboot", 4},
235 {"PowerKey", 5},
236 {"hard_reset", 6},
237 {"kernel_panic", 7},
238 {"rpm_err", 8},
239 {"hw_reset", 9},
240 {"tz_err", 10},
241 {"adsp_err", 11},
242 {"modem_err", 12},
243 {"mba_err", 13},
244 {"Watchdog", 14},
245 {"Panic", 15},
246 {"power_key", 16}, // aliasReasons to cold,powerkey (Mediatek)
247 {"power_on", 17}, // aliasReasons to cold,powerkey
248 {"Reboot", 18},
249 {"rtc", 19},
250 {"edl", 20},
251 {"oem_pon1", 21},
252 {"oem_powerkey", 22}, // aliasReasons to cold,powerkey
253 {"oem_unknown_reset", 23},
254 {"srto: HWWDT reset SC", 24},
255 {"srto: HWWDT reset platform", 25},
256 {"srto: bootloader", 26},
257 {"srto: kernel panic", 27},
258 {"srto: kernel watchdog reset", 28},
259 {"srto: normal", 29},
260 {"srto: reboot", 30},
261 {"srto: reboot-bootloader", 31},
262 {"srto: security watchdog reset", 32},
263 {"srto: wakesrc", 33},
264 {"srto: watchdog", 34},
265 {"srto:1-1", 35},
266 {"srto:omap_hsmm", 36},
267 {"srto:phy0", 37},
268 {"srto:rtc0", 38},
269 {"srto:touchpad", 39},
270 {"watchdog", 40},
271 {"watchdogr", 41},
272 {"wdog_bark", 42},
273 {"wdog_bite", 43},
274 {"wdog_reset", 44},
275 {"shutdown,", 45}, // Trailing comma is intentional. Do NOT use.
276 {"shutdown,userrequested", 46},
277 {"reboot,bootloader", 47},
278 {"reboot,cold", 48},
279 {"reboot,recovery", 49},
280 {"thermal_shutdown", 50},
281 {"s3_wakeup", 51},
282 {"kernel_panic,sysrq", 52},
283 {"kernel_panic,NULL", 53},
284 {"kernel_panic,null", 53},
285 {"kernel_panic,BUG", 54},
286 {"kernel_panic,bug", 54},
287 {"bootloader", 55},
288 {"cold", 56},
289 {"hard", 57},
290 {"warm", 58},
291 {"reboot,kernel_power_off_charging__reboot_system", 59}, // Can not happen
292 {"thermal-shutdown", 60},
293 {"shutdown,thermal", 61},
294 {"shutdown,battery", 62},
295 {"reboot,ota", 63},
296 {"reboot,factory_reset", 64},
297 {"reboot,", 65},
298 {"reboot,shell", 66},
299 {"reboot,adb", 67},
300 {"reboot,userrequested", 68},
301 {"shutdown,container", 69}, // Host OS asking Android Container to shutdown
302 {"cold,powerkey", 70},
303 {"warm,s3_wakeup", 71},
304 {"hard,hw_reset", 72},
305 {"shutdown,suspend", 73}, // Suspend to RAM
306 {"shutdown,hibernate", 74}, // Suspend to DISK
307 {"power_on_key", 75}, // aliasReasons to cold,powerkey
308 {"reboot_by_key", 76}, // translated to reboot,by_key
309 {"wdt_by_pass_pwk", 77}, // Mediatek
310 {"reboot_longkey", 78}, // translated to reboot,longkey
311 {"powerkey", 79}, // aliasReasons to cold,powerkey
312 {"usb", 80}, // aliasReasons to cold,charger (Mediatek)
313 {"wdt", 81}, // Mediatek
314 {"tool_by_pass_pwk", 82}, // aliasReasons to reboot,tool (Mediatek)
315 {"2sec_reboot", 83}, // aliasReasons to cold,rtc,2sec (Mediatek)
316 {"reboot,by_key", 84},
317 {"reboot,longkey", 85},
318 {"reboot,2sec", 86}, // Deprecate in two years, replaced with cold,rtc,2sec
319 {"shutdown,thermal,battery", 87},
320 {"reboot,its_just_so_hard", 88}, // produced by boot_reason_test
321 {"reboot,Its Just So Hard", 89}, // produced by boot_reason_test
322 {"reboot,rescueparty", 90},
323 {"charge", 91},
324 {"oem_tz_crash", 92},
325 {"uvlo", 93}, // aliasReasons to reboot,undervoltage
326 {"oem_ps_hold", 94},
327 {"abnormal_reset", 95},
328 {"oemerr_unknown", 96},
329 {"reboot_fastboot_mode", 97},
330 {"watchdog_apps_bite", 98},
331 {"xpu_err", 99},
332 {"power_on_usb", 100}, // aliasReasons to cold,charger
333 {"watchdog_rpm", 101},
334 {"watchdog_nonsec", 102},
335 {"watchdog_apps_bark", 103},
336 {"reboot_dmverity_corrupted", 104},
337 {"reboot_smpl", 105}, // aliasReasons to reboot,powerloss
338 {"watchdog_sdi_apps_reset", 106},
339 {"smpl", 107}, // aliasReasons to reboot,powerloss
340 {"oem_modem_failed_to_powerup", 108},
341 {"reboot_normal", 109},
342 {"oem_lpass_cfg", 110},
343 {"oem_xpu_ns_error", 111},
344 {"power_key_press", 112}, // aliasReasons to cold,powerkey
345 {"hardware_reset", 113},
346 {"reboot_by_powerkey", 114}, // aliasReasons to cold,powerkey (is this correct?)
347 {"reboot_verity", 115},
348 {"oem_rpm_undef_error", 116},
349 {"oem_crash_on_the_lk", 117},
350 {"oem_rpm_reset", 118},
351 {"reboot,powerloss", 119},
352 {"reboot,undervoltage", 120},
353 {"factory_cable", 121},
354 {"oem_ar6320_failed_to_powerup", 122},
355 {"watchdog_rpm_bite", 123},
356 {"power_on_cable", 124}, // aliasReasons to cold,charger
357 {"reboot_unknown", 125},
358 {"wireless_charger", 126},
359 {"0x776655ff", 127},
360 {"oem_thermal_bite_reset", 128},
361 {"charger", 129},
362 {"pon1", 130},
363 {"unknown", 131},
364 {"reboot_rtc", 132},
365 {"cold_boot", 133},
366 {"hard_rst", 134},
367 {"power-on", 135},
368 {"oem_adsp_resetting_the_soc", 136},
369 {"kpdpwr", 137},
370 {"oem_modem_timeout_waiting", 138},
371 {"usb_chg", 139},
372 {"warm_reset_0x02", 140},
373 {"warm_reset_0x80", 141},
374 {"pon_reason_0xb0", 142},
375 {"reboot_download", 143},
376 {"reboot_recovery_mode", 144},
377 {"oem_sdi_err_fatal", 145},
378 {"pmic_watchdog", 146},
379 {"software_master", 147},
380 {"cold,charger", 148},
381 {"cold,rtc", 149},
382 {"cold,rtc,2sec", 150}, // Mediatek
383 {"reboot,tool", 151}, // Mediatek
384 {"reboot,wdt", 152}, // Mediatek
385 {"reboot,unknown", 153}, // Mediatek
386 {"kernel_panic,audit", 154},
387 {"kernel_panic,atomic", 155},
388 {"kernel_panic,hung", 156},
389 {"kernel_panic,hung,rcu", 157},
390 {"kernel_panic,init", 158},
391 {"kernel_panic,oom", 159},
392 {"kernel_panic,stack", 160},
393 {"kernel_panic,sysrq,livelock,alarm", 161}, // llkd
394 {"kernel_panic,sysrq,livelock,driver", 162}, // llkd
395 {"kernel_panic,sysrq,livelock,zombie", 163}, // llkd
396 {"kernel_panic,modem", 164},
397 {"kernel_panic,adsp", 165},
398 {"kernel_panic,dsps", 166},
399 {"kernel_panic,wcnss", 167},
400 {"kernel_panic,_sde_encoder_phys_cmd_handle_ppdone_timeout", 168},
401 {"recovery,quiescent", 169},
402 {"reboot,quiescent", 170},
403 {"reboot,rtc", 171},
404 {"reboot,dm-verity_device_corrupted", 172},
405 {"reboot,dm-verity_enforcing", 173},
406 {"reboot,keys_clear", 174},
407 {"reboot,pmic_off_fault,.*", 175},
408 {"reboot,pmic_off_s3rst,.*", 176},
409 {"reboot,pmic_off_other,.*", 177},
410 {"reboot,userrequested,fastboot", 178},
411 {"reboot,userrequested,recovery", 179},
412 {"reboot,userrequested,recovery,ui", 180},
413 {"shutdown,userrequested,fastboot", 181},
414 {"shutdown,userrequested,recovery", 182},
415 {"reboot,unknown[0-9]*", 183},
416 {"reboot,longkey,.*", 184},
417 {"reboot,boringssl-self-check-failed", 185},
418 {"reboot,userspace_failed,shutdown_aborted", 186},
419 {"reboot,userspace_failed,watchdog_triggered", 187},
420 {"reboot,userspace_failed,watchdog_fork", 188},
421 {"reboot,userspace_failed,*", 189},
422 {"reboot,mount_userdata_failed", 190},
423 {"reboot,forcedsilent", 191},
424 {"reboot,forcednonsilent", 192},
425 {"reboot,thermal,tj", 193},
426 {"reboot,emergency", 194},
427 {"reboot,factory", 195},
428 {"reboot,fastboot", 196},
429 {"reboot,gsa,hard", 197},
430 {"reboot,gsa,soft", 198},
431 {"reboot,master_dc,fault_n", 199},
432 {"reboot,master_dc,reset", 200},
433 {"reboot,ocp", 201},
434 {"reboot,pin", 202},
435 {"reboot,rom_recovery", 203},
436 {"reboot,uvlo", 204},
437 {"reboot,uvlo,pmic,if", 205},
438 {"reboot,uvlo,pmic,main", 206},
439 {"reboot,uvlo,pmic,sub", 207},
440 {"reboot,warm", 208},
441 {"watchdog,aoc", 209},
442 {"watchdog,apc", 210},
443 {"watchdog,apc,bl,debug,early", 211},
444 {"watchdog,apc,bl,early", 212},
445 {"watchdog,apc,early", 213},
446 {"watchdog,apm", 214},
447 {"watchdog,gsa,hard", 215},
448 {"watchdog,gsa,soft", 216},
449 {"watchdog,pmucal", 217},
450 {"reboot,early,bl", 218},
451 {"watchdog,apc,gsa,crashed", 219},
452 {"watchdog,apc,bl31,crashed", 220},
453 {"watchdog,apc,pbl,crashed", 221},
454 {"reboot,memory_protect,hyp", 222},
455 {"reboot,tsd,pmic,main", 223},
456 {"reboot,tsd,pmic,sub", 224},
457 {"reboot,ocp,pmic,main", 225},
458 {"reboot,ocp,pmic,sub", 226},
459 {"reboot,sys_ldo_ok,pmic,main", 227},
460 {"reboot,sys_ldo_ok,pmic,sub", 228},
461 {"reboot,smpl_timeout,pmic,main", 229},
462 };
463
464 // Converts a string value representing the reason the system booted to an
465 // integer representation. This is necessary for logging the boot_reason metric
466 // via Tron, which does not accept non-integer buckets in histograms.
BootReasonStrToEnum(const std::string & boot_reason)467 int32_t BootReasonStrToEnum(const std::string& boot_reason) {
468 auto mapping = kBootReasonMap.find(boot_reason);
469 if (mapping != kBootReasonMap.end()) {
470 return mapping->second;
471 }
472
473 if (boot_reason.empty()) {
474 return kEmptyBootReason;
475 }
476
477 for (const auto& [match, id] : kBootReasonMap) {
478 // Regex matches as a minimum require either [, \ or * to be present.
479 if (match.find_first_of("[\\*") == match.npos) continue;
480 // enforce match from beginning to end
481 auto exact = match;
482 if (exact[0] != '^') exact = "^" + exact;
483 if (exact[exact.size() - 1] != '$') exact = exact + "$";
484 if (std::regex_search(boot_reason, std::regex(exact))) return id;
485 }
486
487 LOG(INFO) << "Unknown boot reason: " << boot_reason;
488 return kUnknownBootReason;
489 }
490
491 // Canonical list of supported primary reboot reasons.
492 const std::vector<const std::string> knownReasons = {
493 // clang-format off
494 // kernel
495 "watchdog",
496 "kernel_panic",
497 // strong
498 "recovery", // Should not happen from ro.boot.bootreason
499 "bootloader", // Should not happen from ro.boot.bootreason
500 // blunt
501 "cold",
502 "hard",
503 "warm",
504 // super blunt
505 "shutdown", // Can not happen from ro.boot.bootreason
506 "reboot", // Default catch-all for anything unknown
507 // clang-format on
508 };
509
510 // Returns true if the supplied reason prefix is considered detailed enough.
isStrongRebootReason(const std::string & r)511 bool isStrongRebootReason(const std::string& r) {
512 for (auto& s : knownReasons) {
513 if (s == "cold") break;
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 // Returns true if the supplied reason prefix is associated with the kernel.
isKernelRebootReason(const std::string & r)523 bool isKernelRebootReason(const std::string& r) {
524 for (auto& s : knownReasons) {
525 if (s == "recovery") break;
526 // Prefix defined as terminated by a nul or comma (,).
527 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
528 return true;
529 }
530 }
531 return false;
532 }
533
534 // Returns true if the supplied reason prefix is considered known.
isKnownRebootReason(const std::string & r)535 bool isKnownRebootReason(const std::string& r) {
536 for (auto& s : knownReasons) {
537 // Prefix defined as terminated by a nul or comma (,).
538 if (android::base::StartsWith(r, s) && ((r.length() == s.length()) || (r[s.length()] == ','))) {
539 return true;
540 }
541 }
542 return false;
543 }
544
545 // If the reboot reason should be improved, report true if is too blunt.
isBluntRebootReason(const std::string & r)546 bool isBluntRebootReason(const std::string& r) {
547 if (isStrongRebootReason(r)) return false;
548
549 if (!isKnownRebootReason(r)) return true; // Can not support unknown as detail
550
551 size_t pos = 0;
552 while ((pos = r.find(',', pos)) != std::string::npos) {
553 ++pos;
554 std::string next(r.substr(pos));
555 if (next.length() == 0) break;
556 if (next[0] == ',') continue;
557 if (!isKnownRebootReason(next)) return false; // Unknown subreason is good.
558 if (isStrongRebootReason(next)) return false; // eg: reboot,reboot
559 }
560 return true;
561 }
562
readPstoreConsole(std::string & console)563 bool readPstoreConsole(std::string& console) {
564 if (android::base::ReadFileToString("/sys/fs/pstore/console-ramoops-0", &console)) {
565 return true;
566 }
567 return android::base::ReadFileToString("/sys/fs/pstore/console-ramoops", &console);
568 }
569
570 // Implement a variant of std::string::rfind that is resilient to errors in
571 // the data stream being inspected.
572 class pstoreConsole {
573 private:
574 const size_t kBitErrorRate = 8; // number of bits per error
575 const std::string& console;
576
577 // Number of bits that differ between the two arguments l and r.
578 // Returns zero if the values for l and r are identical.
numError(uint8_t l,uint8_t r) const579 size_t numError(uint8_t l, uint8_t r) const { return std::bitset<8>(l ^ r).count(); }
580
581 // A string comparison function, reports the number of errors discovered
582 // in the match to a maximum of the bitLength / kBitErrorRate, at that
583 // point returning npos to indicate match is too poor.
584 //
585 // Since called in rfind which works backwards, expect cache locality will
586 // help if we check in reverse here as well for performance.
587 //
588 // Assumption: l (from console.c_str() + pos) is long enough to house
589 // _r.length(), checked in rfind caller below.
590 //
numError(size_t pos,const std::string & _r) const591 size_t numError(size_t pos, const std::string& _r) const {
592 const char* l = console.c_str() + pos;
593 const char* r = _r.c_str();
594 size_t n = _r.length();
595 const uint8_t* le = reinterpret_cast<const uint8_t*>(l) + n;
596 const uint8_t* re = reinterpret_cast<const uint8_t*>(r) + n;
597 size_t count = 0;
598 n = 0;
599 do {
600 // individual character bit error rate > threshold + slop
601 size_t num = numError(*--le, *--re);
602 if (num > ((8 + kBitErrorRate) / kBitErrorRate)) return std::string::npos;
603 // total bit error rate > threshold + slop
604 count += num;
605 ++n;
606 if (count > ((n * 8 + kBitErrorRate - (n > 2)) / kBitErrorRate)) {
607 return std::string::npos;
608 }
609 } while (le != reinterpret_cast<const uint8_t*>(l));
610 return count;
611 }
612
613 public:
pstoreConsole(const std::string & console)614 explicit pstoreConsole(const std::string& console) : console(console) {}
615 // scope of argument must be equal to or greater than scope of pstoreConsole
616 explicit pstoreConsole(const std::string&& console) = delete;
617 explicit pstoreConsole(std::string&& console) = delete;
618
619 // Our implementation of rfind, use exact match first, then resort to fuzzy.
rfind(const std::string & needle) const620 size_t rfind(const std::string& needle) const {
621 size_t pos = console.rfind(needle); // exact match?
622 if (pos != std::string::npos) return pos;
623
624 // Check to make sure needle fits in console string.
625 pos = console.length();
626 if (needle.length() > pos) return std::string::npos;
627 pos -= needle.length();
628 // fuzzy match to maximum kBitErrorRate
629 for (;;) {
630 if (numError(pos, needle) != std::string::npos) return pos;
631 if (pos == 0) break;
632 --pos;
633 }
634 return std::string::npos;
635 }
636
637 // Our implementation of find, use only fuzzy match.
find(const std::string & needle,size_t start=0) const638 size_t find(const std::string& needle, size_t start = 0) const {
639 // Check to make sure needle fits in console string.
640 if (needle.length() > console.length()) return std::string::npos;
641 const size_t last_pos = console.length() - needle.length();
642 // fuzzy match to maximum kBitErrorRate
643 for (size_t pos = start; pos <= last_pos; ++pos) {
644 if (numError(pos, needle) != std::string::npos) return pos;
645 }
646 return std::string::npos;
647 }
648
operator const std::string&() const649 operator const std::string&() const { return console; }
650 };
651
652 // If bit error match to needle, correct it.
653 // Return true if any corrections were discovered and applied.
correctForBitError(std::string & reason,const std::string & needle)654 bool correctForBitError(std::string& reason, const std::string& needle) {
655 bool corrected = false;
656 if (reason.length() < needle.length()) return corrected;
657 const pstoreConsole console(reason);
658 const size_t last_pos = reason.length() - needle.length();
659 for (size_t pos = 0; pos <= last_pos; pos += needle.length()) {
660 pos = console.find(needle, pos);
661 if (pos == std::string::npos) break;
662
663 // exact match has no malice
664 if (needle == reason.substr(pos, needle.length())) continue;
665
666 corrected = true;
667 reason = reason.substr(0, pos) + needle + reason.substr(pos + needle.length());
668 }
669 return corrected;
670 }
671
672 // If bit error match to needle, correct it.
673 // Return true if any corrections were discovered and applied.
674 // Try again if we can replace underline with spaces.
correctForBitErrorOrUnderline(std::string & reason,const std::string & needle)675 bool correctForBitErrorOrUnderline(std::string& reason, const std::string& needle) {
676 bool corrected = correctForBitError(reason, needle);
677 std::string _needle(needle);
678 std::transform(_needle.begin(), _needle.end(), _needle.begin(),
679 [](char c) { return (c == '_') ? ' ' : c; });
680 if (needle != _needle) {
681 corrected |= correctForBitError(reason, _needle);
682 }
683 return corrected;
684 }
685
686 // Converts a string value representing the reason the system booted to a
687 // string complying with Android system standard reason.
transformReason(std::string & reason)688 void transformReason(std::string& reason) {
689 std::transform(reason.begin(), reason.end(), reason.begin(), ::tolower);
690 std::transform(reason.begin(), reason.end(), reason.begin(),
691 [](char c) { return ::isblank(c) ? '_' : c; });
692 std::transform(reason.begin(), reason.end(), reason.begin(),
693 [](char c) { return ::isprint(c) ? c : '?'; });
694 }
695
696 // Check subreasons for reboot,<subreason> kernel_panic,sysrq,<subreason> or
697 // kernel_panic,<subreason>.
698 //
699 // If quoted flag is set, pull out and correct single quoted ('), newline (\n)
700 // or unprintable character terminated subreason, pos is supplied just beyond
701 // first quote. if quoted false, pull out and correct newline (\n) or
702 // unprintable character terminated subreason.
703 //
704 // Heuristics to find termination is painted into a corner:
705
706 // single bit error for quote ' that we can block. It is acceptable for
707 // the others 7, g in reason. 2/9 chance will miss the terminating quote,
708 // but there is always the terminating newline that usually immediately
709 // follows to fortify our chances.
likely_single_quote(char c)710 bool likely_single_quote(char c) {
711 switch (static_cast<uint8_t>(c)) {
712 case '\'': // '\''
713 case '\'' ^ 0x01: // '&'
714 case '\'' ^ 0x02: // '%'
715 case '\'' ^ 0x04: // '#'
716 case '\'' ^ 0x08: // '/'
717 return true;
718 case '\'' ^ 0x10: // '7'
719 break;
720 case '\'' ^ 0x20: // '\a' (unprintable)
721 return true;
722 case '\'' ^ 0x40: // 'g'
723 break;
724 case '\'' ^ 0x80: // 0xA7 (unprintable)
725 return true;
726 }
727 return false;
728 }
729
730 // ::isprint(c) and likely_space() will prevent us from being called for
731 // fundamentally printable entries, except for '\r' and '\b'.
732 //
733 // Except for * and J, single bit errors for \n, all others are non-
734 // printable so easy catch. It is _acceptable_ for *, J or j to exist in
735 // the reason string, so 2/9 chance we will miss the terminating newline.
736 //
737 // NB: J might not be acceptable, except if at the beginning or preceded
738 // with a space, '(' or any of the quotes and their BER aliases.
739 // NB: * might not be acceptable, except if at the beginning or preceded
740 // with a space, another *, or any of the quotes or their BER aliases.
741 //
742 // To reduce the chances to closer to 1/9 is too complicated for the gain.
likely_newline(char c)743 bool likely_newline(char c) {
744 switch (static_cast<uint8_t>(c)) {
745 case '\n': // '\n' (unprintable)
746 case '\n' ^ 0x01: // '\r' (unprintable)
747 case '\n' ^ 0x02: // '\b' (unprintable)
748 case '\n' ^ 0x04: // 0x0E (unprintable)
749 case '\n' ^ 0x08: // 0x02 (unprintable)
750 case '\n' ^ 0x10: // 0x1A (unprintable)
751 return true;
752 case '\n' ^ 0x20: // '*'
753 case '\n' ^ 0x40: // 'J'
754 break;
755 case '\n' ^ 0x80: // 0x8A (unprintable)
756 return true;
757 }
758 return false;
759 }
760
761 // ::isprint(c) will prevent us from being called for all the printable
762 // matches below. If we let unprintables through because of this, they
763 // get converted to underscore (_) by the validation phase.
likely_space(char c)764 bool likely_space(char c) {
765 switch (static_cast<uint8_t>(c)) {
766 case ' ': // ' '
767 case ' ' ^ 0x01: // '!'
768 case ' ' ^ 0x02: // '"'
769 case ' ' ^ 0x04: // '$'
770 case ' ' ^ 0x08: // '('
771 case ' ' ^ 0x10: // '0'
772 case ' ' ^ 0x20: // '\0' (unprintable)
773 case ' ' ^ 0x40: // 'P'
774 case ' ' ^ 0x80: // 0xA0 (unprintable)
775 case '\t': // '\t'
776 case '\t' ^ 0x01: // '\b' (unprintable) (likely_newline counters)
777 case '\t' ^ 0x02: // '\v' (unprintable)
778 case '\t' ^ 0x04: // '\r' (unprintable) (likely_newline counters)
779 case '\t' ^ 0x08: // 0x01 (unprintable)
780 case '\t' ^ 0x10: // 0x19 (unprintable)
781 case '\t' ^ 0x20: // ')'
782 case '\t' ^ 0x40: // '1'
783 case '\t' ^ 0x80: // 0x89 (unprintable)
784 return true;
785 }
786 return false;
787 }
788
getSubreason(const std::string & content,size_t pos,bool quoted)789 std::string getSubreason(const std::string& content, size_t pos, bool quoted) {
790 static constexpr size_t max_reason_length = 256;
791
792 std::string subReason(content.substr(pos, max_reason_length));
793 // Correct against any known strings that Bit Error Match
794 for (const auto& s : knownReasons) {
795 correctForBitErrorOrUnderline(subReason, s);
796 }
797 std::string terminator(quoted ? "'" : "");
798 for (const auto& m : kBootReasonMap) {
799 if (m.first.length() <= strlen("cold")) continue; // too short?
800 if (correctForBitErrorOrUnderline(subReason, m.first + terminator)) continue;
801 if (m.first.length() <= strlen("reboot,cold")) continue; // short?
802 if (android::base::StartsWith(m.first, "reboot,")) {
803 correctForBitErrorOrUnderline(subReason, m.first.substr(strlen("reboot,")) + terminator);
804 } else if (android::base::StartsWith(m.first, "kernel_panic,sysrq,")) {
805 correctForBitErrorOrUnderline(subReason,
806 m.first.substr(strlen("kernel_panic,sysrq,")) + terminator);
807 } else if (android::base::StartsWith(m.first, "kernel_panic,")) {
808 correctForBitErrorOrUnderline(subReason, m.first.substr(strlen("kernel_panic,")) + terminator);
809 }
810 }
811 for (pos = 0; pos < subReason.length(); ++pos) {
812 char c = subReason[pos];
813 if (!(::isprint(c) || likely_space(c)) || likely_newline(c) ||
814 (quoted && likely_single_quote(c))) {
815 subReason.erase(pos);
816 break;
817 }
818 }
819 transformReason(subReason);
820 return subReason;
821 }
822
addKernelPanicSubReason(const pstoreConsole & console,std::string & ret)823 bool addKernelPanicSubReason(const pstoreConsole& console, std::string& ret) {
824 // Check for kernel panic types to refine information
825 if ((console.rfind("SysRq : Trigger a crash") != std::string::npos) ||
826 (console.rfind("PC is at sysrq_handle_crash+") != std::string::npos)) {
827 ret = "kernel_panic,sysrq";
828 // Invented for Android to allow daemons that specifically trigger sysrq
829 // to communicate more accurate boot subreasons via last console messages.
830 static constexpr char sysrqSubreason[] = "SysRq : Trigger a crash : '";
831 auto pos = console.rfind(sysrqSubreason);
832 if (pos != std::string::npos) {
833 ret += "," + getSubreason(console, pos + strlen(sysrqSubreason), /* quoted */ true);
834 }
835 return true;
836 }
837 if (console.rfind("Unable to handle kernel NULL pointer dereference at virtual address") !=
838 std::string::npos) {
839 ret = "kernel_panic,null";
840 return true;
841 }
842 if (console.rfind("Kernel BUG at ") != std::string::npos) {
843 ret = "kernel_panic,bug";
844 return true;
845 }
846
847 std::string panic("Kernel panic - not syncing: ");
848 auto pos = console.rfind(panic);
849 if (pos != std::string::npos) {
850 static const std::vector<std::pair<const std::string, const std::string>> panicReasons = {
851 {"Out of memory", "oom"},
852 {"out of memory", "oom"},
853 {"Oh boy, that early out of memory", "oom"}, // omg
854 {"BUG!", "bug"},
855 {"hung_task: blocked tasks", "hung"},
856 {"audit: ", "audit"},
857 {"scheduling while atomic", "atomic"},
858 {"Attempted to kill init!", "init"},
859 {"Requested init", "init"},
860 {"No working init", "init"},
861 {"Could not decompress init", "init"},
862 {"RCU Stall", "hung,rcu"},
863 {"stack-protector", "stack"},
864 {"kernel stack overflow", "stack"},
865 {"Corrupt kernel stack", "stack"},
866 {"low stack detected", "stack"},
867 {"corrupted stack end", "stack"},
868 {"subsys-restart: Resetting the SoC - modem crashed.", "modem"},
869 {"subsys-restart: Resetting the SoC - adsp crashed.", "adsp"},
870 {"subsys-restart: Resetting the SoC - dsps crashed.", "dsps"},
871 {"subsys-restart: Resetting the SoC - wcnss crashed.", "wcnss"},
872 };
873
874 ret = "kernel_panic";
875 for (auto& s : panicReasons) {
876 if (console.find(panic + s.first, pos) != std::string::npos) {
877 ret += "," + s.second;
878 return true;
879 }
880 }
881 auto reason = getSubreason(console, pos + panic.length(), /* newline */ false);
882 if (reason.length() > 3) {
883 ret += "," + reason;
884 }
885 return true;
886 }
887 return false;
888 }
889
addKernelPanicSubReason(const std::string & content,std::string & ret)890 bool addKernelPanicSubReason(const std::string& content, std::string& ret) {
891 return addKernelPanicSubReason(pstoreConsole(content), ret);
892 }
893
894 const char system_reboot_reason_property[] = "sys.boot.reason";
895 const char last_reboot_reason_property[] = LAST_REBOOT_REASON_PROPERTY;
896 const char last_reboot_reason_file[] = LAST_REBOOT_REASON_FILE;
897 const char last_last_reboot_reason_property[] = "sys.boot.reason.last";
898 constexpr size_t history_reboot_reason_size = 4;
899 const char history_reboot_reason_property[] = LAST_REBOOT_REASON_PROPERTY ".history";
900 const char bootloader_reboot_reason_property[] = "ro.boot.bootreason";
901
902 // Land system_boot_reason into system_reboot_reason_property.
903 // Shift system_boot_reason into history_reboot_reason_property.
BootReasonAddToHistory(const std::string & system_boot_reason)904 void BootReasonAddToHistory(const std::string& system_boot_reason) {
905 if (system_boot_reason.empty()) return;
906 LOG(INFO) << "Canonical boot reason: " << system_boot_reason;
907 auto old_system_boot_reason = android::base::GetProperty(system_reboot_reason_property, "");
908 if (!android::base::SetProperty(system_reboot_reason_property, system_boot_reason)) {
909 android::base::SetProperty(system_reboot_reason_property,
910 system_boot_reason.substr(0, PROPERTY_VALUE_MAX - 1));
911 }
912 auto reason_history =
913 android::base::Split(android::base::GetProperty(history_reboot_reason_property, ""), "\n");
914 static auto mark = time(nullptr);
915 auto mark_str = std::string(",") + std::to_string(mark);
916 auto marked_system_boot_reason = system_boot_reason + mark_str;
917 if (!reason_history.empty()) {
918 // delete any entries that we just wrote in a previous
919 // call and leveraging duplicate line handling
920 auto last = old_system_boot_reason + mark_str;
921 // trim the list to (history_reboot_reason_size - 1)
922 ssize_t max = history_reboot_reason_size;
923 for (auto it = reason_history.begin(); it != reason_history.end();) {
924 if (it->empty() || (last == *it) || (marked_system_boot_reason == *it) || (--max <= 0)) {
925 it = reason_history.erase(it);
926 } else {
927 last = *it;
928 ++it;
929 }
930 }
931 }
932 // insert at the front, concatenating mark (<epoch time>) detail to the value.
933 reason_history.insert(reason_history.begin(), marked_system_boot_reason);
934 // If the property string is too long ( > PROPERTY_VALUE_MAX)
935 // we get an error, so trim out last entry and try again.
936 while (!android::base::SetProperty(history_reboot_reason_property,
937 android::base::Join(reason_history, '\n'))) {
938 auto it = std::prev(reason_history.end());
939 if (it == reason_history.end()) break;
940 reason_history.erase(it);
941 }
942 }
943
944 // Scrub, Sanitize, Standardize and Enhance the boot reason string supplied.
BootReasonStrToReason(const std::string & boot_reason)945 std::string BootReasonStrToReason(const std::string& boot_reason) {
946 auto ret = android::base::GetProperty(system_reboot_reason_property, "");
947 std::string reason(boot_reason);
948 // If sys.boot.reason == ro.boot.bootreason, let's re-evaluate
949 if (reason == ret) ret = "";
950
951 transformReason(reason);
952
953 // Is the current system boot reason sys.boot.reason valid?
954 if (!isKnownRebootReason(ret)) ret = "";
955
956 if (ret == "") {
957 // Is the bootloader boot reason ro.boot.bootreason known?
958 std::vector<std::string> words(android::base::Split(reason, ",_-"));
959 for (auto& s : knownReasons) {
960 std::string blunt;
961 for (auto& r : words) {
962 if (r == s) {
963 if (isBluntRebootReason(s)) {
964 blunt = s;
965 } else {
966 ret = s;
967 break;
968 }
969 }
970 }
971 if (ret == "") ret = blunt;
972 if (ret != "") break;
973 }
974 }
975
976 if (ret == "") {
977 // A series of checks to take some officially unsupported reasons
978 // reported by the bootloader and find some logical and canonical
979 // sense. In an ideal world, we would require those bootloaders
980 // to behave and follow our CTS standards.
981 //
982 // first member is the output
983 // second member is an unanchored regex for an alias
984 //
985 // If output has a prefix of <bang> '!', we do not use it as a
986 // match needle (and drop the <bang> prefix when landing in output),
987 // otherwise look for it as well. This helps keep the scale of the
988 // following table smaller.
989 static const std::vector<std::pair<const std::string, const std::string>> aliasReasons = {
990 {"watchdog", "wdog"},
991 {"kernel_panic", "panic"},
992 {"shutdown,thermal", "thermal"},
993 {"warm,s3_wakeup", "s3_wakeup"},
994 {"hard,hw_reset", "hw_reset"},
995 {"cold,charger", "usb|power_on_cable"},
996 {"cold,powerkey", "powerkey|power_key|PowerKey|power_on"},
997 {"cold,rtc", "rtc"},
998 {"cold,rtc,2sec", "2sec_reboot"},
999 {"!warm", "wdt_by_pass_pwk"}, // change flavour of blunt
1000 {"!reboot", "^wdt$"}, // change flavour of blunt
1001 {"reboot,tool", "tool_by_pass_pwk"},
1002 {"!reboot,longkey", "reboot_longkey"},
1003 {"!reboot,longkey", "kpdpwr"},
1004 {"!reboot,undervoltage", "uvlo"},
1005 {"!reboot,powerloss", "smpl"},
1006 {"bootloader", ""},
1007 };
1008
1009 for (auto& s : aliasReasons) {
1010 size_t firstHasNot = s.first[0] == '!';
1011 if (!firstHasNot && (reason.find(s.first) != std::string::npos)) {
1012 ret = s.first;
1013 break;
1014 }
1015 if (s.second.size() && std::regex_search(reason, std::regex(s.second))) {
1016 ret = s.first.substr(firstHasNot);
1017 break;
1018 }
1019 }
1020 }
1021
1022 // If watchdog is the reason, see if there is a security angle?
1023 if (ret == "watchdog") {
1024 if (reason.find("sec") != std::string::npos) {
1025 ret += ",security";
1026 }
1027 }
1028
1029 if (ret == "kernel_panic") {
1030 // Check to see if last klog has some refinement hints.
1031 std::string content;
1032 if (readPstoreConsole(content)) {
1033 addKernelPanicSubReason(content, ret);
1034 }
1035 } else if (isBluntRebootReason(ret)) {
1036 // Check the other available reason resources if the reason is still blunt.
1037
1038 // Check to see if last klog has some refinement hints.
1039 std::string content;
1040 if (readPstoreConsole(content)) {
1041 const pstoreConsole console(content);
1042 // The toybox reboot command used directly (unlikely)? But also
1043 // catches init's response to Android's more controlled reboot command.
1044 if (console.rfind("reboot: Power down") != std::string::npos) {
1045 ret = "shutdown"; // Still too blunt, but more accurate.
1046 // ToDo: init should record the shutdown reason to kernel messages ala:
1047 // init: shutdown system with command 'last_reboot_reason'
1048 // so that if pstore has persistence we can get some details
1049 // that could be missing in last_reboot_reason_property.
1050 }
1051
1052 static const char cmd[] = "reboot: Restarting system with command '";
1053 size_t pos = console.rfind(cmd);
1054 if (pos != std::string::npos) {
1055 std::string subReason(getSubreason(content, pos + strlen(cmd), /* quoted */ true));
1056 if (subReason != "") { // Will not land "reboot" as that is too blunt.
1057 if (isKernelRebootReason(subReason)) {
1058 ret = "reboot," + subReason; // User space can't talk kernel reasons.
1059 } else if (isKnownRebootReason(subReason)) {
1060 ret = subReason;
1061 } else {
1062 ret = "reboot," + subReason; // legitimize unknown reasons
1063 }
1064 }
1065 // Some bootloaders shutdown results record in last kernel message.
1066 if (!strcmp(ret.c_str(), "reboot,kernel_power_off_charging__reboot_system")) {
1067 ret = "shutdown";
1068 }
1069 }
1070
1071 // Check for kernel panics, allowed to override reboot command.
1072 if (!addKernelPanicSubReason(console, ret) &&
1073 // check for long-press power down
1074 ((console.rfind("Power held for ") != std::string::npos) ||
1075 (console.rfind("charger: [") != std::string::npos))) {
1076 ret = "cold";
1077 }
1078 }
1079
1080 // TODO: use the HAL to get battery level (http://b/77725702).
1081
1082 // Is there a controlled shutdown hint in last_reboot_reason_property?
1083 if (isBluntRebootReason(ret)) {
1084 // Content buffer no longer will have console data. Beware if more
1085 // checks added below, that depend on parsing console content.
1086 if (!android::base::ReadFileToString(last_reboot_reason_file, &content)) {
1087 content = android::base::GetProperty(last_reboot_reason_property, "");
1088 }
1089 transformReason(content);
1090
1091 // Anything in last is better than 'super-blunt' reboot or shutdown.
1092 if ((ret == "") || (ret == "reboot") || (ret == "shutdown") || !isBluntRebootReason(content)) {
1093 ret = content;
1094 }
1095 }
1096
1097 // Other System Health HAL reasons?
1098
1099 // ToDo: /proc/sys/kernel/boot_reason needs a HAL interface to
1100 // possibly offer hardware-specific clues from the PMIC.
1101 }
1102
1103 // If unknown left over from above, make it "reboot,<boot_reason>"
1104 if (ret == "") {
1105 ret = "reboot";
1106 if (android::base::StartsWith(reason, "reboot")) {
1107 reason = reason.substr(strlen("reboot"));
1108 while ((reason[0] == ',') || (reason[0] == '_')) {
1109 reason = reason.substr(1);
1110 }
1111 }
1112 if (reason != "") {
1113 ret += ",";
1114 ret += reason;
1115 }
1116 }
1117
1118 LOG(INFO) << "Canonical boot reason: " << ret;
1119 return ret;
1120 }
1121
1122 // Returns the appropriate metric key prefix for the boot_complete metric such
1123 // that boot metrics after a system update are labeled as ota_boot_complete;
1124 // otherwise, they are labeled as boot_complete. This method encapsulates the
1125 // bookkeeping required to track when a system update has occurred by storing
1126 // the UTC timestamp of the system build date and comparing against the current
1127 // system build date.
CalculateBootCompletePrefix()1128 std::string CalculateBootCompletePrefix() {
1129 static const std::string kBuildDateKey = "build_date";
1130 std::string boot_complete_prefix = "boot_complete";
1131
1132 auto build_date_str = android::base::GetProperty("ro.build.date.utc", "");
1133 int32_t build_date;
1134 if (!android::base::ParseInt(build_date_str, &build_date)) {
1135 return std::string();
1136 }
1137
1138 BootEventRecordStore boot_event_store;
1139 BootEventRecordStore::BootEventRecord record;
1140 if (!boot_event_store.GetBootEvent(kBuildDateKey, &record)) {
1141 boot_complete_prefix = "factory_reset_" + boot_complete_prefix;
1142 boot_event_store.AddBootEventWithValue(kBuildDateKey, build_date);
1143 BootReasonAddToHistory("reboot,factory_reset");
1144 } else if (build_date != record.second) {
1145 boot_complete_prefix = "ota_" + boot_complete_prefix;
1146 boot_event_store.AddBootEventWithValue(kBuildDateKey, build_date);
1147 BootReasonAddToHistory("reboot,ota");
1148 }
1149
1150 return boot_complete_prefix;
1151 }
1152
1153 // Records the value of a given ro.boottime.init property in milliseconds.
RecordInitBootTimeProp(BootEventRecordStore * boot_event_store,const char * property)1154 void RecordInitBootTimeProp(BootEventRecordStore* boot_event_store, const char* property) {
1155 auto value = android::base::GetProperty(property, "");
1156
1157 int32_t time_in_ms;
1158 if (android::base::ParseInt(value, &time_in_ms)) {
1159 boot_event_store->AddBootEventWithValue(property, time_in_ms);
1160 }
1161 }
1162
1163 // A map from bootloader timing stage to the time that stage took during boot.
1164 typedef std::map<std::string, int32_t> BootloaderTimingMap;
1165
1166 // Returns a mapping from bootloader stage names to the time those stages
1167 // took to boot.
GetBootLoaderTimings()1168 const BootloaderTimingMap GetBootLoaderTimings() {
1169 BootloaderTimingMap timings;
1170
1171 // |ro.boot.boottime| is of the form 'stage1:time1,...,stageN:timeN',
1172 // where timeN is in milliseconds.
1173 auto value = android::base::GetProperty("ro.boot.boottime", "");
1174 if (value.empty()) {
1175 // ro.boot.boottime is not reported on all devices.
1176 return BootloaderTimingMap();
1177 }
1178
1179 auto stages = android::base::Split(value, ",");
1180 for (const auto& stageTiming : stages) {
1181 // |stageTiming| is of the form 'stage:time'.
1182 auto stageTimingValues = android::base::Split(stageTiming, ":");
1183 DCHECK_EQ(2U, stageTimingValues.size());
1184
1185 if (stageTimingValues.size() < 2) continue;
1186 std::string stageName = stageTimingValues[0];
1187 int32_t time_ms;
1188 if (android::base::ParseInt(stageTimingValues[1], &time_ms)) {
1189 timings[stageName] = time_ms;
1190 }
1191 }
1192
1193 return timings;
1194 }
1195
1196 // Returns the total bootloader boot time from the ro.boot.boottime system property.
GetBootloaderTime(const BootloaderTimingMap & bootloader_timings)1197 int32_t GetBootloaderTime(const BootloaderTimingMap& bootloader_timings) {
1198 int32_t total_time = 0;
1199 for (const auto& timing : bootloader_timings) {
1200 total_time += timing.second;
1201 }
1202
1203 return total_time;
1204 }
1205
1206 // Parses and records the set of bootloader stages and associated boot times
1207 // from the ro.boot.boottime system property.
RecordBootloaderTimings(BootEventRecordStore * boot_event_store,const BootloaderTimingMap & bootloader_timings)1208 void RecordBootloaderTimings(BootEventRecordStore* boot_event_store,
1209 const BootloaderTimingMap& bootloader_timings) {
1210 int32_t total_time = 0;
1211 for (const auto& timing : bootloader_timings) {
1212 total_time += timing.second;
1213 boot_event_store->AddBootEventWithValue("boottime.bootloader." + timing.first, timing.second);
1214 }
1215
1216 boot_event_store->AddBootEventWithValue("boottime.bootloader.total", total_time);
1217 }
1218
1219 // Returns the closest estimation to the absolute device boot time, i.e.,
1220 // from power on to boot_complete, including bootloader times.
GetAbsoluteBootTime(const BootloaderTimingMap & bootloader_timings,std::chrono::milliseconds uptime)1221 std::chrono::milliseconds GetAbsoluteBootTime(const BootloaderTimingMap& bootloader_timings,
1222 std::chrono::milliseconds uptime) {
1223 int32_t bootloader_time_ms = 0;
1224
1225 for (const auto& timing : bootloader_timings) {
1226 if (timing.first.compare("SW") != 0) {
1227 bootloader_time_ms += timing.second;
1228 }
1229 }
1230
1231 auto bootloader_duration = std::chrono::milliseconds(bootloader_time_ms);
1232 return bootloader_duration + uptime;
1233 }
1234
1235 // Records the closest estimation to the absolute device boot time in seconds.
1236 // i.e. from power on to boot_complete, including bootloader times.
RecordAbsoluteBootTime(BootEventRecordStore * boot_event_store,std::chrono::milliseconds absolute_total)1237 void RecordAbsoluteBootTime(BootEventRecordStore* boot_event_store,
1238 std::chrono::milliseconds absolute_total) {
1239 auto absolute_total_sec = std::chrono::duration_cast<std::chrono::seconds>(absolute_total);
1240 boot_event_store->AddBootEventWithValue("absolute_boot_time", absolute_total_sec.count());
1241 }
1242
1243 // 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)1244 void LogBootInfoToStatsd(std::chrono::milliseconds end_time,
1245 std::chrono::milliseconds total_duration, int32_t bootloader_duration_ms,
1246 double time_since_last_boot_sec) {
1247 auto reason = android::base::GetProperty(bootloader_reboot_reason_property, "<EMPTY>");
1248 auto system_reason = android::base::GetProperty(system_reboot_reason_property, "<EMPTY>");
1249 android::util::stats_write(android::util::BOOT_SEQUENCE_REPORTED, reason.c_str(),
1250 system_reason.c_str(), end_time.count(), total_duration.count(),
1251 (int64_t)bootloader_duration_ms,
1252 (int64_t)time_since_last_boot_sec * 1000);
1253 }
1254
SetSystemBootReason()1255 void SetSystemBootReason() {
1256 const auto bootloader_boot_reason =
1257 android::base::GetProperty(bootloader_reboot_reason_property, "");
1258 const std::string system_boot_reason(BootReasonStrToReason(bootloader_boot_reason));
1259 // Record the scrubbed system_boot_reason to the property
1260 BootReasonAddToHistory(system_boot_reason);
1261 // Shift last_reboot_reason_property to last_last_reboot_reason_property
1262 std::string last_boot_reason;
1263 if (!android::base::ReadFileToString(last_reboot_reason_file, &last_boot_reason)) {
1264 PLOG(ERROR) << "Failed to read " << last_reboot_reason_file;
1265 last_boot_reason = android::base::GetProperty(last_reboot_reason_property, "");
1266 LOG(INFO) << "Value of " << last_reboot_reason_property << " : " << last_boot_reason;
1267 } else {
1268 LOG(INFO) << "Last reboot reason read from " << last_reboot_reason_file << " : "
1269 << last_boot_reason << ". Last reboot reason read from "
1270 << last_reboot_reason_property << " : "
1271 << android::base::GetProperty(last_reboot_reason_property, "");
1272 }
1273 if (last_boot_reason.empty() || isKernelRebootReason(system_boot_reason)) {
1274 last_boot_reason = system_boot_reason;
1275 } else {
1276 transformReason(last_boot_reason);
1277 }
1278 LOG(INFO) << "Normalized last reboot reason : " << last_boot_reason;
1279 android::base::SetProperty(last_last_reboot_reason_property, last_boot_reason);
1280 android::base::SetProperty(last_reboot_reason_property, "");
1281 if (unlink(last_reboot_reason_file) != 0) {
1282 PLOG(ERROR) << "Failed to unlink " << last_reboot_reason_file;
1283 }
1284 }
1285
1286 // Gets the boot time offset. This is useful when Android is running in a
1287 // container, because the boot_clock is not reset when Android reboots.
GetBootTimeOffset()1288 std::chrono::nanoseconds GetBootTimeOffset() {
1289 static const int64_t boottime_offset =
1290 android::base::GetIntProperty<int64_t>("ro.boot.boottime_offset", 0);
1291 return std::chrono::nanoseconds(boottime_offset);
1292 }
1293
1294 // Returns the current uptime, accounting for any offset in the CLOCK_BOOTTIME
1295 // clock.
GetUptime()1296 android::base::boot_clock::duration GetUptime() {
1297 return android::base::boot_clock::now().time_since_epoch() - GetBootTimeOffset();
1298 }
1299
1300 // Records several metrics related to the time it takes to boot the device.
RecordBootComplete()1301 void RecordBootComplete() {
1302 BootEventRecordStore boot_event_store;
1303 BootEventRecordStore::BootEventRecord record;
1304
1305 auto uptime_ns = GetUptime();
1306 auto uptime_s = std::chrono::duration_cast<std::chrono::seconds>(uptime_ns);
1307 time_t current_time_utc = time(nullptr);
1308 time_t time_since_last_boot = 0;
1309
1310 if (boot_event_store.GetBootEvent("last_boot_time_utc", &record)) {
1311 time_t last_boot_time_utc = record.second;
1312 time_since_last_boot = difftime(current_time_utc, last_boot_time_utc);
1313 boot_event_store.AddBootEventWithValue("time_since_last_boot", time_since_last_boot);
1314 }
1315
1316 boot_event_store.AddBootEventWithValue("last_boot_time_utc", current_time_utc);
1317
1318 // The boot_complete metric has two variants: boot_complete and
1319 // ota_boot_complete. The latter signifies that the device is booting after
1320 // a system update.
1321 std::string boot_complete_prefix = CalculateBootCompletePrefix();
1322 if (boot_complete_prefix.empty()) {
1323 // The system is hosed because the build date property could not be read.
1324 return;
1325 }
1326
1327 // The *_no_encryption events are emitted unconditionally, since they are left
1328 // over from a time when encryption meant "full-disk encryption". But Android
1329 // now always uses file-based encryption instead of full-disk encryption. At
1330 // some point, these misleading and redundant events should be removed.
1331 boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_no_encryption",
1332 uptime_s.count());
1333
1334 // Record the total time from device startup to boot complete. Note: we are
1335 // recording seconds here even though the field in statsd atom specifies
1336 // milliseconds.
1337 boot_event_store.AddBootEventWithValue(boot_complete_prefix, uptime_s.count());
1338
1339 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init");
1340 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.first_stage");
1341 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.selinux");
1342 RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.cold_boot_wait");
1343
1344 const BootloaderTimingMap bootloader_timings = GetBootLoaderTimings();
1345 int32_t bootloader_boot_duration = GetBootloaderTime(bootloader_timings);
1346 RecordBootloaderTimings(&boot_event_store, bootloader_timings);
1347
1348 auto uptime_ms = std::chrono::duration_cast<std::chrono::milliseconds>(uptime_ns);
1349 auto absolute_boot_time = GetAbsoluteBootTime(bootloader_timings, uptime_ms);
1350 RecordAbsoluteBootTime(&boot_event_store, absolute_boot_time);
1351
1352 auto boot_end_time_point = std::chrono::system_clock::now().time_since_epoch();
1353 auto boot_end_time = std::chrono::duration_cast<std::chrono::milliseconds>(boot_end_time_point);
1354
1355 LogBootInfoToStatsd(boot_end_time, absolute_boot_time, bootloader_boot_duration,
1356 time_since_last_boot);
1357 }
1358
1359 // Records the boot_reason metric by querying the ro.boot.bootreason system
1360 // property.
RecordBootReason()1361 void RecordBootReason() {
1362 const auto reason = android::base::GetProperty(bootloader_reboot_reason_property, "");
1363
1364 if (reason.empty()) {
1365 // TODO(b/148575354): Replace with statsd.
1366 // Log an empty boot reason value as '<EMPTY>' to ensure the value is intentional
1367 // (and not corruption anywhere else in the reporting pipeline).
1368 // android::metricslogger::LogMultiAction(android::metricslogger::ACTION_BOOT,
1369 // android::metricslogger::FIELD_PLATFORM_REASON,
1370 // "<EMPTY>");
1371 } else {
1372 // TODO(b/148575354): Replace with statsd.
1373 // android::metricslogger::LogMultiAction(android::metricslogger::ACTION_BOOT,
1374 // android::metricslogger::FIELD_PLATFORM_REASON,
1375 // reason);
1376 }
1377
1378 // Log the raw bootloader_boot_reason property value.
1379 int32_t boot_reason = BootReasonStrToEnum(reason);
1380 BootEventRecordStore boot_event_store;
1381 boot_event_store.AddBootEventWithValue("boot_reason", boot_reason);
1382
1383 // Log the scrubbed system_boot_reason.
1384 const auto system_reason = android::base::GetProperty(system_reboot_reason_property, "");
1385 int32_t system_boot_reason = BootReasonStrToEnum(system_reason);
1386 boot_event_store.AddBootEventWithValue("system_boot_reason", system_boot_reason);
1387
1388 if (reason == "") {
1389 android::base::SetProperty(bootloader_reboot_reason_property, system_reason);
1390 }
1391 }
1392
1393 // Records two metrics related to the user resetting a device: the time at
1394 // which the device is reset, and the time since the user last reset the
1395 // device. The former is only set once per-factory reset.
RecordFactoryReset()1396 void RecordFactoryReset() {
1397 BootEventRecordStore boot_event_store;
1398 BootEventRecordStore::BootEventRecord record;
1399
1400 time_t current_time_utc = time(nullptr);
1401
1402 if (current_time_utc < 0) {
1403 // UMA does not display negative values in buckets, so convert to positive.
1404 // Logging via BootEventRecordStore.
1405 android::util::stats_write(
1406 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_ERROR_CODE_REPORTED),
1407 static_cast<int32_t>(
1408 android::util::BOOT_TIME_EVENT_ERROR_CODE__EVENT__FACTORY_RESET_CURRENT_TIME_FAILURE),
1409 static_cast<int32_t>(std::abs(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_failure",
1414 std::abs(current_time_utc));
1415 return;
1416 } else {
1417 android::util::stats_write(
1418 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED),
1419 static_cast<int32_t>(
1420 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_CURRENT_TIME),
1421 static_cast<int64_t>(current_time_utc));
1422
1423 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1424 // is losing records somehow.
1425 boot_event_store.AddBootEventWithValue("factory_reset_current_time", current_time_utc);
1426 }
1427
1428 // The factory_reset boot event does not exist after the device is reset, so
1429 // use this signal to mark the time of the factory reset.
1430 if (!boot_event_store.GetBootEvent("factory_reset", &record)) {
1431 boot_event_store.AddBootEventWithValue("factory_reset", current_time_utc);
1432
1433 // Don't log the time_since_factory_reset until some time has elapsed.
1434 // The data is not meaningful yet and skews the histogram buckets.
1435 return;
1436 }
1437
1438 // Calculate and record the difference in time between now and the
1439 // factory_reset time.
1440 time_t factory_reset_utc = record.second;
1441 android::util::stats_write(
1442 static_cast<int32_t>(android::util::BOOT_TIME_EVENT_UTC_TIME_REPORTED),
1443 static_cast<int32_t>(
1444 android::util::BOOT_TIME_EVENT_UTC_TIME__EVENT__FACTORY_RESET_RECORD_VALUE),
1445 static_cast<int64_t>(factory_reset_utc));
1446
1447 // Logging via BootEventRecordStore to see if using android::metricslogger::LogHistogram
1448 // is losing records somehow.
1449 boot_event_store.AddBootEventWithValue("factory_reset_record_value", factory_reset_utc);
1450
1451 time_t time_since_factory_reset = difftime(current_time_utc, factory_reset_utc);
1452 boot_event_store.AddBootEventWithValue("time_since_factory_reset", time_since_factory_reset);
1453 }
1454
1455 // List the associated boot reason(s), if arg is nullptr then all.
PrintBootReasonEnum(const char * arg)1456 void PrintBootReasonEnum(const char* arg) {
1457 int value = -1;
1458 if (arg != nullptr) {
1459 value = BootReasonStrToEnum(arg);
1460 }
1461 for (const auto& [match, id] : kBootReasonMap) {
1462 if ((value < 0) || (value == id)) {
1463 printf("%u\t%s\n", id, match.c_str());
1464 }
1465 }
1466 }
1467
1468 } // namespace
1469
main(int argc,char ** argv)1470 int main(int argc, char** argv) {
1471 android::base::InitLogging(argv);
1472
1473 const std::string cmd_line = GetCommandLine(argc, argv);
1474 LOG(INFO) << "Service started: " << cmd_line;
1475
1476 int option_index = 0;
1477 static const char value_str[] = "value";
1478 static const char system_boot_reason_str[] = "set_system_boot_reason";
1479 static const char boot_complete_str[] = "record_boot_complete";
1480 static const char boot_reason_str[] = "record_boot_reason";
1481 static const char factory_reset_str[] = "record_time_since_factory_reset";
1482 static const char boot_reason_enum_str[] = "boot_reason_enum";
1483 static const struct option long_options[] = {
1484 // clang-format off
1485 { "help", no_argument, NULL, 'h' },
1486 { "log", no_argument, NULL, 'l' },
1487 { "print", no_argument, NULL, 'p' },
1488 { "record", required_argument, NULL, 'r' },
1489 { value_str, required_argument, NULL, 0 },
1490 { system_boot_reason_str, no_argument, NULL, 0 },
1491 { boot_complete_str, no_argument, NULL, 0 },
1492 { boot_reason_str, no_argument, NULL, 0 },
1493 { factory_reset_str, no_argument, NULL, 0 },
1494 { boot_reason_enum_str, optional_argument, NULL, 0 },
1495 { NULL, 0, NULL, 0 }
1496 // clang-format on
1497 };
1498
1499 std::string boot_event;
1500 std::string value;
1501 int opt = 0;
1502 while ((opt = getopt_long(argc, argv, "hlpr:", long_options, &option_index)) != -1) {
1503 switch (opt) {
1504 // This case handles long options which have no single-character mapping.
1505 case 0: {
1506 const std::string option_name = long_options[option_index].name;
1507 if (option_name == value_str) {
1508 // |optarg| is an external variable set by getopt representing
1509 // the option argument.
1510 value = optarg;
1511 } else if (option_name == system_boot_reason_str) {
1512 SetSystemBootReason();
1513 } else if (option_name == boot_complete_str) {
1514 RecordBootComplete();
1515 } else if (option_name == boot_reason_str) {
1516 RecordBootReason();
1517 } else if (option_name == factory_reset_str) {
1518 RecordFactoryReset();
1519 } else if (option_name == boot_reason_enum_str) {
1520 PrintBootReasonEnum(optarg);
1521 } else {
1522 LOG(ERROR) << "Invalid option: " << option_name;
1523 }
1524 break;
1525 }
1526
1527 case 'h': {
1528 ShowHelp(argv[0]);
1529 break;
1530 }
1531
1532 case 'l': {
1533 LogBootEvents();
1534 break;
1535 }
1536
1537 case 'p': {
1538 PrintBootEvents();
1539 break;
1540 }
1541
1542 case 'r': {
1543 // |optarg| is an external variable set by getopt representing
1544 // the option argument.
1545 boot_event = optarg;
1546 break;
1547 }
1548
1549 default: {
1550 DCHECK_EQ(opt, '?');
1551
1552 // |optopt| is an external variable set by getopt representing
1553 // the value of the invalid option.
1554 LOG(ERROR) << "Invalid option: " << optopt;
1555 ShowHelp(argv[0]);
1556 return EXIT_FAILURE;
1557 }
1558 }
1559 }
1560
1561 if (!boot_event.empty()) {
1562 RecordBootEventFromCommandLine(boot_event, value);
1563 }
1564
1565 return 0;
1566 }
1567