1 /*
2 * Copyright (C) 2017 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 #include "reboot.h"
18
19 #include <dirent.h>
20 #include <fcntl.h>
21 #include <linux/fs.h>
22 #include <linux/loop.h>
23 #include <mntent.h>
24 #include <semaphore.h>
25 #include <stdlib.h>
26 #include <sys/cdefs.h>
27 #include <sys/ioctl.h>
28 #include <sys/mount.h>
29 #include <sys/stat.h>
30 #include <sys/swap.h>
31 #include <sys/syscall.h>
32 #include <sys/types.h>
33 #include <sys/wait.h>
34
35 #include <chrono>
36 #include <memory>
37 #include <set>
38 #include <thread>
39 #include <vector>
40
41 #include <InitProperties.sysprop.h>
42 #include <android-base/chrono_utils.h>
43 #include <android-base/file.h>
44 #include <android-base/logging.h>
45 #include <android-base/macros.h>
46 #include <android-base/properties.h>
47 #include <android-base/scopeguard.h>
48 #include <android-base/strings.h>
49 #include <android-base/unique_fd.h>
50 #include <bootloader_message/bootloader_message.h>
51 #include <cutils/android_reboot.h>
52 #include <fs_mgr.h>
53 #include <logwrap/logwrap.h>
54 #include <private/android_filesystem_config.h>
55 #include <selinux/selinux.h>
56
57 #include "action.h"
58 #include "action_manager.h"
59 #include "builtin_arguments.h"
60 #include "init.h"
61 #include "mount_namespace.h"
62 #include "property_service.h"
63 #include "reboot_utils.h"
64 #include "service.h"
65 #include "service_list.h"
66 #include "sigchld_handler.h"
67 #include "util.h"
68
69 using namespace std::literals;
70
71 using android::base::boot_clock;
72 using android::base::GetBoolProperty;
73 using android::base::GetUintProperty;
74 using android::base::SetProperty;
75 using android::base::Split;
76 using android::base::Timer;
77 using android::base::unique_fd;
78 using android::base::WaitForProperty;
79 using android::base::WriteStringToFile;
80
81 namespace android {
82 namespace init {
83
84 static bool shutting_down = false;
85
86 static const std::set<std::string> kDebuggingServices{"tombstoned", "logd", "adbd", "console"};
87
GetPostDataDebuggingServices()88 static std::set<std::string> GetPostDataDebuggingServices() {
89 std::set<std::string> ret;
90 for (const auto& s : ServiceList::GetInstance()) {
91 if (kDebuggingServices.count(s->name()) && s->is_post_data()) {
92 ret.insert(s->name());
93 }
94 }
95 return ret;
96 }
97
PersistRebootReason(const char * reason,bool write_to_property)98 static void PersistRebootReason(const char* reason, bool write_to_property) {
99 if (write_to_property) {
100 SetProperty(LAST_REBOOT_REASON_PROPERTY, reason);
101 }
102 auto fd = unique_fd(TEMP_FAILURE_RETRY(open(
103 LAST_REBOOT_REASON_FILE, O_WRONLY | O_CREAT | O_TRUNC | O_CLOEXEC | O_BINARY, 0666)));
104 if (!fd.ok()) {
105 PLOG(ERROR) << "Could not open '" << LAST_REBOOT_REASON_FILE
106 << "' to persist reboot reason";
107 return;
108 }
109 WriteStringToFd(reason, fd);
110 fsync(fd.get());
111 }
112
113 // represents umount status during reboot / shutdown.
114 enum UmountStat {
115 /* umount succeeded. */
116 UMOUNT_STAT_SUCCESS = 0,
117 /* umount was not run. */
118 UMOUNT_STAT_SKIPPED = 1,
119 /* umount failed with timeout. */
120 UMOUNT_STAT_TIMEOUT = 2,
121 /* could not run due to error */
122 UMOUNT_STAT_ERROR = 3,
123 /* not used by init but reserved for other part to use this to represent the
124 the state where umount status before reboot is not found / available. */
125 UMOUNT_STAT_NOT_AVAILABLE = 4,
126 };
127
128 // Utility for struct mntent
129 class MountEntry {
130 public:
MountEntry(const mntent & entry)131 explicit MountEntry(const mntent& entry)
132 : mnt_fsname_(entry.mnt_fsname),
133 mnt_dir_(entry.mnt_dir),
134 mnt_type_(entry.mnt_type),
135 mnt_opts_(entry.mnt_opts) {}
136
Umount(bool force)137 bool Umount(bool force) {
138 LOG(INFO) << "Unmounting " << mnt_fsname_ << ":" << mnt_dir_ << " opts " << mnt_opts_;
139 int r = umount2(mnt_dir_.c_str(), force ? MNT_FORCE : 0);
140 if (r == 0) {
141 LOG(INFO) << "Umounted " << mnt_fsname_ << ":" << mnt_dir_ << " opts " << mnt_opts_;
142 return true;
143 } else {
144 PLOG(WARNING) << "Cannot umount " << mnt_fsname_ << ":" << mnt_dir_ << " opts "
145 << mnt_opts_;
146 return false;
147 }
148 }
149
DoFsck()150 void DoFsck() {
151 int st;
152 if (IsF2Fs()) {
153 const char* f2fs_argv[] = {
154 "/system/bin/fsck.f2fs",
155 "-a",
156 mnt_fsname_.c_str(),
157 };
158 logwrap_fork_execvp(arraysize(f2fs_argv), f2fs_argv, &st, false, LOG_KLOG, true,
159 nullptr);
160 } else if (IsExt4()) {
161 const char* ext4_argv[] = {
162 "/system/bin/e2fsck",
163 "-y",
164 mnt_fsname_.c_str(),
165 };
166 logwrap_fork_execvp(arraysize(ext4_argv), ext4_argv, &st, false, LOG_KLOG, true,
167 nullptr);
168 }
169 }
170
IsBlockDevice(const struct mntent & mntent)171 static bool IsBlockDevice(const struct mntent& mntent) {
172 return android::base::StartsWith(mntent.mnt_fsname, "/dev/block");
173 }
174
IsEmulatedDevice(const struct mntent & mntent)175 static bool IsEmulatedDevice(const struct mntent& mntent) {
176 return android::base::StartsWith(mntent.mnt_fsname, "/data/");
177 }
178
179 private:
IsF2Fs() const180 bool IsF2Fs() const { return mnt_type_ == "f2fs"; }
181
IsExt4() const182 bool IsExt4() const { return mnt_type_ == "ext4"; }
183
184 std::string mnt_fsname_;
185 std::string mnt_dir_;
186 std::string mnt_type_;
187 std::string mnt_opts_;
188 };
189
190 // Turn off backlight while we are performing power down cleanup activities.
TurnOffBacklight()191 static void TurnOffBacklight() {
192 Service* service = ServiceList::GetInstance().FindService("blank_screen");
193 if (service == nullptr) {
194 LOG(WARNING) << "cannot find blank_screen in TurnOffBacklight";
195 return;
196 }
197 if (auto result = service->Start(); !result.ok()) {
198 LOG(WARNING) << "Could not start blank_screen service: " << result.error();
199 }
200 }
201
CallVdc(const std::string & system,const std::string & cmd)202 static Result<void> CallVdc(const std::string& system, const std::string& cmd) {
203 LOG(INFO) << "Calling /system/bin/vdc " << system << " " << cmd;
204 const char* vdc_argv[] = {"/system/bin/vdc", system.c_str(), cmd.c_str()};
205 int status;
206 if (logwrap_fork_execvp(arraysize(vdc_argv), vdc_argv, &status, false, LOG_KLOG, true,
207 nullptr) != 0) {
208 return ErrnoError() << "Failed to call '/system/bin/vdc " << system << " " << cmd << "'";
209 }
210 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
211 return {};
212 }
213 return Error() << "'/system/bin/vdc " << system << " " << cmd << "' failed : " << status;
214 }
215
LogShutdownTime(UmountStat stat,Timer * t)216 static void LogShutdownTime(UmountStat stat, Timer* t) {
217 LOG(WARNING) << "powerctl_shutdown_time_ms:" << std::to_string(t->duration().count()) << ":"
218 << stat;
219 }
220
IsDataMounted()221 static bool IsDataMounted() {
222 std::unique_ptr<std::FILE, int (*)(std::FILE*)> fp(setmntent("/proc/mounts", "re"), endmntent);
223 if (fp == nullptr) {
224 PLOG(ERROR) << "Failed to open /proc/mounts";
225 return false;
226 }
227 mntent* mentry;
228 while ((mentry = getmntent(fp.get())) != nullptr) {
229 if (mentry->mnt_dir == "/data"s) {
230 return true;
231 }
232 }
233 return false;
234 }
235
236 // Find all read+write block devices and emulated devices in /proc/mounts and add them to
237 // the correpsponding list.
FindPartitionsToUmount(std::vector<MountEntry> * block_dev_partitions,std::vector<MountEntry> * emulated_partitions,bool dump)238 static bool FindPartitionsToUmount(std::vector<MountEntry>* block_dev_partitions,
239 std::vector<MountEntry>* emulated_partitions, bool dump) {
240 std::unique_ptr<std::FILE, int (*)(std::FILE*)> fp(setmntent("/proc/mounts", "re"), endmntent);
241 if (fp == nullptr) {
242 PLOG(ERROR) << "Failed to open /proc/mounts";
243 return false;
244 }
245 mntent* mentry;
246 while ((mentry = getmntent(fp.get())) != nullptr) {
247 if (dump) {
248 LOG(INFO) << "mount entry " << mentry->mnt_fsname << ":" << mentry->mnt_dir << " opts "
249 << mentry->mnt_opts << " type " << mentry->mnt_type;
250 } else if (MountEntry::IsBlockDevice(*mentry) && hasmntopt(mentry, "rw")) {
251 std::string mount_dir(mentry->mnt_dir);
252 // These are R/O partitions changed to R/W after adb remount.
253 // Do not umount them as shutdown critical services may rely on them.
254 if (mount_dir != "/" && mount_dir != "/system" && mount_dir != "/vendor" &&
255 mount_dir != "/oem") {
256 block_dev_partitions->emplace(block_dev_partitions->begin(), *mentry);
257 }
258 } else if (MountEntry::IsEmulatedDevice(*mentry)) {
259 emulated_partitions->emplace(emulated_partitions->begin(), *mentry);
260 }
261 }
262 return true;
263 }
264
DumpUmountDebuggingInfo()265 static void DumpUmountDebuggingInfo() {
266 int status;
267 if (!security_getenforce()) {
268 LOG(INFO) << "Run lsof";
269 const char* lsof_argv[] = {"/system/bin/lsof"};
270 logwrap_fork_execvp(arraysize(lsof_argv), lsof_argv, &status, false, LOG_KLOG, true,
271 nullptr);
272 }
273 FindPartitionsToUmount(nullptr, nullptr, true);
274 // dump current CPU stack traces and uninterruptible tasks
275 WriteStringToFile("l", PROC_SYSRQ);
276 WriteStringToFile("w", PROC_SYSRQ);
277 }
278
UmountPartitions(std::chrono::milliseconds timeout)279 static UmountStat UmountPartitions(std::chrono::milliseconds timeout) {
280 Timer t;
281 /* data partition needs all pending writes to be completed and all emulated partitions
282 * umounted.If the current waiting is not good enough, give
283 * up and leave it to e2fsck after reboot to fix it.
284 */
285 while (true) {
286 std::vector<MountEntry> block_devices;
287 std::vector<MountEntry> emulated_devices;
288 if (!FindPartitionsToUmount(&block_devices, &emulated_devices, false)) {
289 return UMOUNT_STAT_ERROR;
290 }
291 if (block_devices.size() == 0) {
292 return UMOUNT_STAT_SUCCESS;
293 }
294 bool unmount_done = true;
295 if (emulated_devices.size() > 0) {
296 for (auto& entry : emulated_devices) {
297 if (!entry.Umount(false)) unmount_done = false;
298 }
299 if (unmount_done) {
300 sync();
301 }
302 }
303 for (auto& entry : block_devices) {
304 if (!entry.Umount(timeout == 0ms)) unmount_done = false;
305 }
306 if (unmount_done) {
307 return UMOUNT_STAT_SUCCESS;
308 }
309 if ((timeout < t.duration())) { // try umount at least once
310 return UMOUNT_STAT_TIMEOUT;
311 }
312 std::this_thread::sleep_for(100ms);
313 }
314 }
315
KillAllProcesses()316 static void KillAllProcesses() {
317 WriteStringToFile("i", PROC_SYSRQ);
318 }
319
320 // Create reboot/shutdwon monitor thread
RebootMonitorThread(unsigned int cmd,const std::string & reboot_target,sem_t * reboot_semaphore,std::chrono::milliseconds shutdown_timeout,bool * reboot_monitor_run)321 void RebootMonitorThread(unsigned int cmd, const std::string& reboot_target,
322 sem_t* reboot_semaphore, std::chrono::milliseconds shutdown_timeout,
323 bool* reboot_monitor_run) {
324 unsigned int remaining_shutdown_time = 0;
325
326 // 300 seconds more than the timeout passed to the thread as there is a final Umount pass
327 // after the timeout is reached.
328 constexpr unsigned int shutdown_watchdog_timeout_default = 300;
329 auto shutdown_watchdog_timeout = android::base::GetUintProperty(
330 "ro.build.shutdown.watchdog.timeout", shutdown_watchdog_timeout_default);
331 remaining_shutdown_time = shutdown_watchdog_timeout + shutdown_timeout.count() / 1000;
332
333 while (*reboot_monitor_run == true) {
334 if (TEMP_FAILURE_RETRY(sem_wait(reboot_semaphore)) == -1) {
335 LOG(ERROR) << "sem_wait failed and exit RebootMonitorThread()";
336 return;
337 }
338
339 timespec shutdown_timeout_timespec;
340 if (clock_gettime(CLOCK_MONOTONIC, &shutdown_timeout_timespec) == -1) {
341 LOG(ERROR) << "clock_gettime() fail! exit RebootMonitorThread()";
342 return;
343 }
344
345 // If there are some remaining shutdown time left from previous round, we use
346 // remaining time here.
347 shutdown_timeout_timespec.tv_sec += remaining_shutdown_time;
348
349 LOG(INFO) << "shutdown_timeout_timespec.tv_sec: " << shutdown_timeout_timespec.tv_sec;
350
351 int sem_return = 0;
352 while ((sem_return = sem_timedwait_monotonic_np(reboot_semaphore,
353 &shutdown_timeout_timespec)) == -1 &&
354 errno == EINTR) {
355 }
356
357 if (sem_return == -1) {
358 LOG(ERROR) << "Reboot thread timed out";
359
360 if (android::base::GetBoolProperty("ro.debuggable", false) == true) {
361 if (false) {
362 // SEPolicy will block debuggerd from running and this is intentional.
363 // But these lines are left to be enabled during debugging.
364 LOG(INFO) << "Try to dump init process call trace:";
365 const char* vdc_argv[] = {"/system/bin/debuggerd", "-b", "1"};
366 int status;
367 logwrap_fork_execvp(arraysize(vdc_argv), vdc_argv, &status, false, LOG_KLOG,
368 true, nullptr);
369 }
370 LOG(INFO) << "Show stack for all active CPU:";
371 WriteStringToFile("l", PROC_SYSRQ);
372
373 LOG(INFO) << "Show tasks that are in disk sleep(uninterruptable sleep), which are "
374 "like "
375 "blocked in mutex or hardware register access:";
376 WriteStringToFile("w", PROC_SYSRQ);
377 }
378
379 // In shutdown case,notify kernel to sync and umount fs to read-only before shutdown.
380 if (cmd == ANDROID_RB_POWEROFF || cmd == ANDROID_RB_THERMOFF) {
381 WriteStringToFile("s", PROC_SYSRQ);
382
383 WriteStringToFile("u", PROC_SYSRQ);
384
385 RebootSystem(cmd, reboot_target);
386 }
387
388 LOG(ERROR) << "Trigger crash at last!";
389 WriteStringToFile("c", PROC_SYSRQ);
390 } else {
391 timespec current_time_timespec;
392
393 if (clock_gettime(CLOCK_MONOTONIC, ¤t_time_timespec) == -1) {
394 LOG(ERROR) << "clock_gettime() fail! exit RebootMonitorThread()";
395 return;
396 }
397
398 remaining_shutdown_time =
399 shutdown_timeout_timespec.tv_sec - current_time_timespec.tv_sec;
400
401 LOG(INFO) << "remaining_shutdown_time: " << remaining_shutdown_time;
402 }
403 }
404 }
405
406 /* Try umounting all emulated file systems R/W block device cfile systems.
407 * This will just try umount and give it up if it fails.
408 * For fs like ext4, this is ok as file system will be marked as unclean shutdown
409 * and necessary check can be done at the next reboot.
410 * For safer shutdown, caller needs to make sure that
411 * all processes / emulated partition for the target fs are all cleaned-up.
412 *
413 * return true when umount was successful. false when timed out.
414 */
TryUmountAndFsck(unsigned int cmd,bool run_fsck,std::chrono::milliseconds timeout,sem_t * reboot_semaphore)415 static UmountStat TryUmountAndFsck(unsigned int cmd, bool run_fsck,
416 std::chrono::milliseconds timeout, sem_t* reboot_semaphore) {
417 Timer t;
418 std::vector<MountEntry> block_devices;
419 std::vector<MountEntry> emulated_devices;
420
421 if (run_fsck && !FindPartitionsToUmount(&block_devices, &emulated_devices, false)) {
422 return UMOUNT_STAT_ERROR;
423 }
424
425 UmountStat stat = UmountPartitions(timeout - t.duration());
426 if (stat != UMOUNT_STAT_SUCCESS) {
427 LOG(INFO) << "umount timeout, last resort, kill all and try";
428 if (DUMP_ON_UMOUNT_FAILURE) DumpUmountDebuggingInfo();
429 KillAllProcesses();
430 // even if it succeeds, still it is timeout and do not run fsck with all processes killed
431 UmountStat st = UmountPartitions(0ms);
432 if ((st != UMOUNT_STAT_SUCCESS) && DUMP_ON_UMOUNT_FAILURE) DumpUmountDebuggingInfo();
433 }
434
435 if (stat == UMOUNT_STAT_SUCCESS && run_fsck) {
436 LOG(INFO) << "Pause reboot monitor thread before fsck";
437 sem_post(reboot_semaphore);
438
439 // fsck part is excluded from timeout check. It only runs for user initiated shutdown
440 // and should not affect reboot time.
441 for (auto& entry : block_devices) {
442 entry.DoFsck();
443 }
444
445 LOG(INFO) << "Resume reboot monitor thread after fsck";
446 sem_post(reboot_semaphore);
447 }
448 return stat;
449 }
450
451 // zram is able to use backing device on top of a loopback device.
452 // In order to unmount /data successfully, we have to kill the loopback device first
453 #define ZRAM_DEVICE "/dev/block/zram0"
454 #define ZRAM_RESET "/sys/block/zram0/reset"
455 #define ZRAM_BACK_DEV "/sys/block/zram0/backing_dev"
456 #define ZRAM_INITSTATE "/sys/block/zram0/initstate"
KillZramBackingDevice()457 static Result<void> KillZramBackingDevice() {
458 std::string zram_initstate;
459 if (!android::base::ReadFileToString(ZRAM_INITSTATE, &zram_initstate)) {
460 return ErrnoError() << "Failed to read " << ZRAM_INITSTATE;
461 }
462
463 zram_initstate.erase(zram_initstate.length() - 1);
464 if (zram_initstate == "0") {
465 LOG(INFO) << "Zram has not been swapped on";
466 return {};
467 }
468
469 if (access(ZRAM_BACK_DEV, F_OK) != 0 && errno == ENOENT) {
470 LOG(INFO) << "No zram backing device configured";
471 return {};
472 }
473 std::string backing_dev;
474 if (!android::base::ReadFileToString(ZRAM_BACK_DEV, &backing_dev)) {
475 return ErrnoError() << "Failed to read " << ZRAM_BACK_DEV;
476 }
477
478 // cut the last "\n"
479 backing_dev.erase(backing_dev.length() - 1);
480
481 // shutdown zram handle
482 Timer swap_timer;
483 LOG(INFO) << "swapoff() start...";
484 if (swapoff(ZRAM_DEVICE) == -1) {
485 return ErrnoError() << "zram_backing_dev: swapoff (" << backing_dev << ")"
486 << " failed";
487 }
488 LOG(INFO) << "swapoff() took " << swap_timer;;
489
490 if (!WriteStringToFile("1", ZRAM_RESET)) {
491 return Error() << "zram_backing_dev: reset (" << backing_dev << ")"
492 << " failed";
493 }
494
495 if (!android::base::StartsWith(backing_dev, "/dev/block/loop")) {
496 LOG(INFO) << backing_dev << " is not a loop device. Exiting early";
497 return {};
498 }
499
500 // clear loopback device
501 unique_fd loop(TEMP_FAILURE_RETRY(open(backing_dev.c_str(), O_RDWR | O_CLOEXEC)));
502 if (loop.get() < 0) {
503 return ErrnoError() << "zram_backing_dev: open(" << backing_dev << ")"
504 << " failed";
505 }
506
507 if (ioctl(loop.get(), LOOP_CLR_FD, 0) < 0) {
508 return ErrnoError() << "zram_backing_dev: loop_clear (" << backing_dev << ")"
509 << " failed";
510 }
511 LOG(INFO) << "zram_backing_dev: `" << backing_dev << "` is cleared successfully.";
512 return {};
513 }
514
515 // Stops given services, waits for them to be stopped for |timeout| ms.
516 // If terminate is true, then SIGTERM is sent to services, otherwise SIGKILL is sent.
517 // Note that services are stopped in order given by |ServiceList::services_in_shutdown_order|
518 // function.
StopServices(const std::set<std::string> & services,std::chrono::milliseconds timeout,bool terminate)519 static void StopServices(const std::set<std::string>& services, std::chrono::milliseconds timeout,
520 bool terminate) {
521 LOG(INFO) << "Stopping " << services.size() << " services by sending "
522 << (terminate ? "SIGTERM" : "SIGKILL");
523 std::vector<pid_t> pids;
524 pids.reserve(services.size());
525 for (const auto& s : ServiceList::GetInstance().services_in_shutdown_order()) {
526 if (services.count(s->name()) == 0) {
527 continue;
528 }
529 if (s->pid() > 0) {
530 pids.push_back(s->pid());
531 }
532 if (terminate) {
533 s->Terminate();
534 } else {
535 s->Stop();
536 }
537 }
538 if (timeout > 0ms) {
539 WaitToBeReaped(pids, timeout);
540 } else {
541 // Even if we don't to wait for services to stop, we still optimistically reap zombies.
542 ReapAnyOutstandingChildren();
543 }
544 }
545
546 // Like StopServices, but also logs all the services that failed to stop after the provided timeout.
547 // Returns number of violators.
StopServicesAndLogViolations(const std::set<std::string> & services,std::chrono::milliseconds timeout,bool terminate)548 int StopServicesAndLogViolations(const std::set<std::string>& services,
549 std::chrono::milliseconds timeout, bool terminate) {
550 StopServices(services, timeout, terminate);
551 int still_running = 0;
552 for (const auto& s : ServiceList::GetInstance()) {
553 if (s->IsRunning() && services.count(s->name())) {
554 LOG(ERROR) << "[service-misbehaving] : service '" << s->name() << "' is still running "
555 << timeout.count() << "ms after receiving "
556 << (terminate ? "SIGTERM" : "SIGKILL");
557 still_running++;
558 }
559 }
560 return still_running;
561 }
562
UnmountAllApexes()563 static Result<void> UnmountAllApexes() {
564 const char* args[] = {"/system/bin/apexd", "--unmount-all"};
565 int status;
566 if (logwrap_fork_execvp(arraysize(args), args, &status, false, LOG_KLOG, true, nullptr) != 0) {
567 return ErrnoError() << "Failed to call '/system/bin/apexd --unmount-all'";
568 }
569 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
570 return {};
571 }
572 return Error() << "'/system/bin/apexd --unmount-all' failed : " << status;
573 }
574
575 //* Reboot / shutdown the system.
576 // cmd ANDROID_RB_* as defined in android_reboot.h
577 // reason Reason string like "reboot", "shutdown,userrequested"
578 // reboot_target Reboot target string like "bootloader". Otherwise, it should be an empty string.
579 // run_fsck Whether to run fsck after umount is done.
580 //
DoReboot(unsigned int cmd,const std::string & reason,const std::string & reboot_target,bool run_fsck)581 static void DoReboot(unsigned int cmd, const std::string& reason, const std::string& reboot_target,
582 bool run_fsck) {
583 Timer t;
584 LOG(INFO) << "Reboot start, reason: " << reason << ", reboot_target: " << reboot_target;
585
586 bool is_thermal_shutdown = cmd == ANDROID_RB_THERMOFF;
587
588 auto shutdown_timeout = 0ms;
589 if (!SHUTDOWN_ZERO_TIMEOUT) {
590 constexpr unsigned int shutdown_timeout_default = 6;
591 constexpr unsigned int max_thermal_shutdown_timeout = 3;
592 auto shutdown_timeout_final = android::base::GetUintProperty("ro.build.shutdown_timeout",
593 shutdown_timeout_default);
594 if (is_thermal_shutdown && shutdown_timeout_final > max_thermal_shutdown_timeout) {
595 shutdown_timeout_final = max_thermal_shutdown_timeout;
596 }
597 shutdown_timeout = std::chrono::seconds(shutdown_timeout_final);
598 }
599 LOG(INFO) << "Shutdown timeout: " << shutdown_timeout.count() << " ms";
600
601 sem_t reboot_semaphore;
602 if (sem_init(&reboot_semaphore, false, 0) == -1) {
603 // These should never fail, but if they do, skip the graceful reboot and reboot immediately.
604 LOG(ERROR) << "sem_init() fail and RebootSystem() return!";
605 RebootSystem(cmd, reboot_target);
606 }
607
608 // Start a thread to monitor init shutdown process
609 LOG(INFO) << "Create reboot monitor thread.";
610 bool reboot_monitor_run = true;
611 std::thread reboot_monitor_thread(&RebootMonitorThread, cmd, reboot_target, &reboot_semaphore,
612 shutdown_timeout, &reboot_monitor_run);
613 reboot_monitor_thread.detach();
614
615 // Start reboot monitor thread
616 sem_post(&reboot_semaphore);
617
618 // Ensure last reboot reason is reduced to canonical
619 // alias reported in bootloader or system boot reason.
620 size_t skip = 0;
621 std::vector<std::string> reasons = Split(reason, ",");
622 if (reasons.size() >= 2 && reasons[0] == "reboot" &&
623 (reasons[1] == "recovery" || reasons[1] == "bootloader" || reasons[1] == "cold" ||
624 reasons[1] == "hard" || reasons[1] == "warm")) {
625 skip = strlen("reboot,");
626 }
627 PersistRebootReason(reason.c_str() + skip, true);
628
629 // If /data isn't mounted then we can skip the extra reboot steps below, since we don't need to
630 // worry about unmounting it.
631 if (!IsDataMounted()) {
632 sync();
633 RebootSystem(cmd, reboot_target);
634 abort();
635 }
636
637 // watchdogd is a vendor specific component but should be alive to complete shutdown safely.
638 const std::set<std::string> to_starts{"watchdogd"};
639 std::set<std::string> stop_first;
640 for (const auto& s : ServiceList::GetInstance()) {
641 if (kDebuggingServices.count(s->name())) {
642 // keep debugging tools until non critical ones are all gone.
643 s->SetShutdownCritical();
644 } else if (to_starts.count(s->name())) {
645 if (auto result = s->Start(); !result.ok()) {
646 LOG(ERROR) << "Could not start shutdown 'to_start' service '" << s->name()
647 << "': " << result.error();
648 }
649 s->SetShutdownCritical();
650 } else if (s->IsShutdownCritical()) {
651 // Start shutdown critical service if not started.
652 if (auto result = s->Start(); !result.ok()) {
653 LOG(ERROR) << "Could not start shutdown critical service '" << s->name()
654 << "': " << result.error();
655 }
656 } else {
657 stop_first.insert(s->name());
658 }
659 }
660
661 // remaining operations (specifically fsck) may take a substantial duration
662 if (cmd == ANDROID_RB_POWEROFF || is_thermal_shutdown) {
663 TurnOffBacklight();
664 }
665
666 Service* boot_anim = ServiceList::GetInstance().FindService("bootanim");
667 Service* surface_flinger = ServiceList::GetInstance().FindService("surfaceflinger");
668 if (boot_anim != nullptr && surface_flinger != nullptr && surface_flinger->IsRunning()) {
669 bool do_shutdown_animation = GetBoolProperty("ro.init.shutdown_animation", false);
670
671 if (do_shutdown_animation) {
672 SetProperty("service.bootanim.exit", "0");
673 SetProperty("service.bootanim.progress", "0");
674 // Could be in the middle of animation. Stop and start so that it can pick
675 // up the right mode.
676 boot_anim->Stop();
677 }
678
679 for (const auto& service : ServiceList::GetInstance()) {
680 if (service->classnames().count("animation") == 0) {
681 continue;
682 }
683
684 // start all animation classes if stopped.
685 if (do_shutdown_animation) {
686 service->Start();
687 }
688 service->SetShutdownCritical(); // will not check animation class separately
689 }
690
691 if (do_shutdown_animation) {
692 boot_anim->Start();
693 surface_flinger->SetShutdownCritical();
694 boot_anim->SetShutdownCritical();
695 }
696 }
697
698 // optional shutdown step
699 // 1. terminate all services except shutdown critical ones. wait for delay to finish
700 if (shutdown_timeout > 0ms) {
701 StopServicesAndLogViolations(stop_first, shutdown_timeout / 2, true /* SIGTERM */);
702 }
703 // Send SIGKILL to ones that didn't terminate cleanly.
704 StopServicesAndLogViolations(stop_first, 0ms, false /* SIGKILL */);
705 SubcontextTerminate();
706 // Reap subcontext pids.
707 ReapAnyOutstandingChildren();
708
709 // 3. send volume abort_fuse and volume shutdown to vold
710 Service* vold_service = ServiceList::GetInstance().FindService("vold");
711 if (vold_service != nullptr && vold_service->IsRunning()) {
712 // Manually abort FUSE connections, since the FUSE daemon is already dead
713 // at this point, and unmounting it might hang.
714 CallVdc("volume", "abort_fuse");
715 CallVdc("volume", "shutdown");
716 vold_service->Stop();
717 } else {
718 LOG(INFO) << "vold not running, skipping vold shutdown";
719 }
720 // logcat stopped here
721 StopServices(kDebuggingServices, 0ms, false /* SIGKILL */);
722 // 4. sync, try umount, and optionally run fsck for user shutdown
723 {
724 Timer sync_timer;
725 LOG(INFO) << "sync() before umount...";
726 sync();
727 LOG(INFO) << "sync() before umount took" << sync_timer;
728 }
729 // 5. drop caches and disable zram backing device, if exist
730 KillZramBackingDevice();
731
732 LOG(INFO) << "Ready to unmount apexes. So far shutdown sequence took " << t;
733 // 6. unmount active apexes, otherwise they might prevent clean unmount of /data.
734 if (auto ret = UnmountAllApexes(); !ret.ok()) {
735 LOG(ERROR) << ret.error();
736 }
737 UmountStat stat =
738 TryUmountAndFsck(cmd, run_fsck, shutdown_timeout - t.duration(), &reboot_semaphore);
739 // Follow what linux shutdown is doing: one more sync with little bit delay
740 {
741 Timer sync_timer;
742 LOG(INFO) << "sync() after umount...";
743 sync();
744 LOG(INFO) << "sync() after umount took" << sync_timer;
745 }
746 if (!is_thermal_shutdown) std::this_thread::sleep_for(100ms);
747 LogShutdownTime(stat, &t);
748
749 // Send signal to terminate reboot monitor thread.
750 reboot_monitor_run = false;
751 sem_post(&reboot_semaphore);
752
753 // Reboot regardless of umount status. If umount fails, fsck after reboot will fix it.
754 RebootSystem(cmd, reboot_target);
755 abort();
756 }
757
EnterShutdown()758 static void EnterShutdown() {
759 LOG(INFO) << "Entering shutdown mode";
760 shutting_down = true;
761 // Skip wait for prop if it is in progress
762 ResetWaitForProp();
763 // Clear EXEC flag if there is one pending
764 for (const auto& s : ServiceList::GetInstance()) {
765 s->UnSetExec();
766 }
767 }
768
LeaveShutdown()769 static void LeaveShutdown() {
770 LOG(INFO) << "Leaving shutdown mode";
771 shutting_down = false;
772 StartSendingMessages();
773 }
774
GetMillisProperty(const std::string & name,std::chrono::milliseconds default_value)775 static std::chrono::milliseconds GetMillisProperty(const std::string& name,
776 std::chrono::milliseconds default_value) {
777 auto value = GetUintProperty(name, static_cast<uint64_t>(default_value.count()));
778 return std::chrono::milliseconds(std::move(value));
779 }
780
DoUserspaceReboot()781 static Result<void> DoUserspaceReboot() {
782 LOG(INFO) << "Userspace reboot initiated";
783 // An ugly way to pass a more precise reason on why fallback to hard reboot was triggered.
784 std::string sub_reason = "";
785 auto guard = android::base::make_scope_guard([&sub_reason] {
786 // Leave shutdown so that we can handle a full reboot.
787 LeaveShutdown();
788 trigger_shutdown("reboot,userspace_failed,shutdown_aborted," + sub_reason);
789 });
790 // Triggering userspace-reboot-requested will result in a bunch of setprop
791 // actions. We should make sure, that all of them are propagated before
792 // proceeding with userspace reboot. Synchronously setting sys.init.userspace_reboot.in_progress
793 // property is not perfect, but it should do the trick.
794 if (!android::sysprop::InitProperties::userspace_reboot_in_progress(true)) {
795 sub_reason = "setprop";
796 return Error() << "Failed to set sys.init.userspace_reboot.in_progress property";
797 }
798 EnterShutdown();
799 if (!SetProperty("sys.powerctl", "")) {
800 sub_reason = "resetprop";
801 return Error() << "Failed to reset sys.powerctl property";
802 }
803 std::set<std::string> stop_first;
804 // Remember the services that were enabled. We will need to manually enable them again otherwise
805 // triggers like class_start won't restart them.
806 std::set<std::string> were_enabled;
807 for (const auto& s : ServiceList::GetInstance().services_in_shutdown_order()) {
808 if (s->is_post_data() && !kDebuggingServices.count(s->name())) {
809 stop_first.insert(s->name());
810 }
811 // TODO(ioffe): we should also filter out temporary services here.
812 if (s->is_post_data() && s->IsEnabled()) {
813 were_enabled.insert(s->name());
814 }
815 }
816 {
817 Timer sync_timer;
818 LOG(INFO) << "sync() before terminating services...";
819 sync();
820 LOG(INFO) << "sync() took " << sync_timer;
821 }
822 auto sigterm_timeout = GetMillisProperty("init.userspace_reboot.sigterm.timeoutmillis", 5s);
823 auto sigkill_timeout = GetMillisProperty("init.userspace_reboot.sigkill.timeoutmillis", 10s);
824 LOG(INFO) << "Timeout to terminate services: " << sigterm_timeout.count() << "ms "
825 << "Timeout to kill services: " << sigkill_timeout.count() << "ms";
826 std::string services_file_name = "/metadata/userspacereboot/services.txt";
827 const int flags = O_RDWR | O_CREAT | O_SYNC | O_APPEND | O_CLOEXEC;
828 StopServicesAndLogViolations(stop_first, sigterm_timeout, true /* SIGTERM */);
829 if (int r = StopServicesAndLogViolations(stop_first, sigkill_timeout, false /* SIGKILL */);
830 r > 0) {
831 auto fd = unique_fd(TEMP_FAILURE_RETRY(open(services_file_name.c_str(), flags, 0666)));
832 android::base::WriteStringToFd("Post-data services still running: \n", fd);
833 for (const auto& s : ServiceList::GetInstance()) {
834 if (s->IsRunning() && stop_first.count(s->name())) {
835 android::base::WriteStringToFd(s->name() + "\n", fd);
836 }
837 }
838 sub_reason = "sigkill";
839 return Error() << r << " post-data services are still running";
840 }
841 if (auto result = KillZramBackingDevice(); !result.ok()) {
842 sub_reason = "zram";
843 return result;
844 }
845 if (auto result = CallVdc("volume", "reset"); !result.ok()) {
846 sub_reason = "vold_reset";
847 return result;
848 }
849 const auto& debugging_services = GetPostDataDebuggingServices();
850 if (int r = StopServicesAndLogViolations(debugging_services, sigkill_timeout,
851 false /* SIGKILL */);
852 r > 0) {
853 auto fd = unique_fd(TEMP_FAILURE_RETRY(open(services_file_name.c_str(), flags, 0666)));
854 android::base::WriteStringToFd("Debugging services still running: \n", fd);
855 for (const auto& s : ServiceList::GetInstance()) {
856 if (s->IsRunning() && debugging_services.count(s->name())) {
857 android::base::WriteStringToFd(s->name() + "\n", fd);
858 }
859 }
860 sub_reason = "sigkill_debug";
861 return Error() << r << " debugging services are still running";
862 }
863 {
864 Timer sync_timer;
865 LOG(INFO) << "sync() after stopping services...";
866 sync();
867 LOG(INFO) << "sync() took " << sync_timer;
868 }
869 if (auto result = UnmountAllApexes(); !result.ok()) {
870 sub_reason = "apex";
871 return result;
872 }
873 if (!SwitchToMountNamespaceIfNeeded(NS_BOOTSTRAP).ok()) {
874 sub_reason = "ns_switch";
875 return Error() << "Failed to switch to bootstrap namespace";
876 }
877 // Remove services that were defined in an APEX.
878 ServiceList::GetInstance().RemoveServiceIf([](const std::unique_ptr<Service>& s) -> bool {
879 if (s->is_from_apex()) {
880 LOG(INFO) << "Removing service '" << s->name() << "' because it's defined in an APEX";
881 return true;
882 }
883 return false;
884 });
885 // Re-enable services
886 for (const auto& s : ServiceList::GetInstance()) {
887 if (were_enabled.count(s->name())) {
888 LOG(INFO) << "Re-enabling service '" << s->name() << "'";
889 s->Enable();
890 }
891 }
892 ServiceList::GetInstance().ResetState();
893 LeaveShutdown();
894 ActionManager::GetInstance().QueueEventTrigger("userspace-reboot-resume");
895 guard.Disable(); // Go on with userspace reboot.
896 return {};
897 }
898
UserspaceRebootWatchdogThread()899 static void UserspaceRebootWatchdogThread() {
900 auto started_timeout = GetMillisProperty("init.userspace_reboot.started.timeoutmillis", 10s);
901 if (!WaitForProperty("sys.init.userspace_reboot.in_progress", "1", started_timeout)) {
902 LOG(ERROR) << "Userspace reboot didn't start in " << started_timeout.count()
903 << "ms. Switching to full reboot";
904 // Init might be wedged, don't try to write reboot reason into a persistent property and do
905 // a dirty reboot.
906 PersistRebootReason("userspace_failed,watchdog_triggered,failed_to_start", false);
907 RebootSystem(ANDROID_RB_RESTART2, "userspace_failed,watchdog_triggered,failed_to_start");
908 }
909 LOG(INFO) << "Starting userspace reboot watchdog";
910 auto watchdog_timeout = GetMillisProperty("init.userspace_reboot.watchdog.timeoutmillis", 5min);
911 LOG(INFO) << "UserspaceRebootWatchdog timeout: " << watchdog_timeout.count() << "ms";
912 if (!WaitForProperty("sys.boot_completed", "1", watchdog_timeout)) {
913 LOG(ERROR) << "Failed to boot in " << watchdog_timeout.count()
914 << "ms. Switching to full reboot";
915 // In this case device is in a boot loop. Only way to recover is to do dirty reboot.
916 // Since init might be wedged, don't try to write reboot reason into a persistent property.
917 PersistRebootReason("userspace_failed,watchdog_triggered,failed_to_boot", false);
918 RebootSystem(ANDROID_RB_RESTART2, "userspace_failed,watchdog_triggered,failed_to_boot");
919 }
920 LOG(INFO) << "Device booted, stopping userspace reboot watchdog";
921 }
922
HandleUserspaceReboot()923 static void HandleUserspaceReboot() {
924 if (!android::sysprop::InitProperties::is_userspace_reboot_supported().value_or(false)) {
925 LOG(ERROR) << "Attempted a userspace reboot on a device that doesn't support it";
926 return;
927 }
928 // Spinnig up a separate thread will fail the setns call later in the boot sequence.
929 // Fork a new process to monitor userspace reboot while we are investigating a better solution.
930 pid_t pid = fork();
931 if (pid < 0) {
932 PLOG(ERROR) << "Failed to fork process for userspace reboot watchdog. Switching to full "
933 << "reboot";
934 trigger_shutdown("reboot,userspace_failed,watchdog_fork");
935 return;
936 }
937 if (pid == 0) {
938 // Child
939 UserspaceRebootWatchdogThread();
940 _exit(EXIT_SUCCESS);
941 }
942 LOG(INFO) << "Clearing queue and starting userspace-reboot-requested trigger";
943 auto& am = ActionManager::GetInstance();
944 am.ClearQueue();
945 am.QueueEventTrigger("userspace-reboot-requested");
946 auto handler = [](const BuiltinArguments&) { return DoUserspaceReboot(); };
947 am.QueueBuiltinAction(handler, "userspace-reboot");
948 }
949
950 /**
951 * Check if "command" field is set in bootloader message.
952 *
953 * If "command" field is broken (contains non-printable characters prior to
954 * terminating zero), it will be zeroed.
955 *
956 * @param[in,out] boot Bootloader message (BCB) structure
957 * @return true if "command" field is already set, and false if it's empty
958 */
CommandIsPresent(bootloader_message * boot)959 static bool CommandIsPresent(bootloader_message* boot) {
960 if (boot->command[0] == '\0')
961 return false;
962
963 for (size_t i = 0; i < arraysize(boot->command); ++i) {
964 if (boot->command[i] == '\0')
965 return true;
966 if (!isprint(boot->command[i]))
967 break;
968 }
969
970 memset(boot->command, 0, sizeof(boot->command));
971 return false;
972 }
973
HandlePowerctlMessage(const std::string & command)974 void HandlePowerctlMessage(const std::string& command) {
975 unsigned int cmd = 0;
976 std::vector<std::string> cmd_params = Split(command, ",");
977 std::string reboot_target = "";
978 bool run_fsck = false;
979 bool command_invalid = false;
980 bool userspace_reboot = false;
981
982 if (cmd_params[0] == "shutdown") {
983 cmd = ANDROID_RB_POWEROFF;
984 if (cmd_params.size() >= 2) {
985 if (cmd_params[1] == "userrequested") {
986 // The shutdown reason is PowerManager.SHUTDOWN_USER_REQUESTED.
987 // Run fsck once the file system is remounted in read-only mode.
988 run_fsck = true;
989 } else if (cmd_params[1] == "thermal") {
990 // Turn off sources of heat immediately.
991 TurnOffBacklight();
992 // run_fsck is false to avoid delay
993 cmd = ANDROID_RB_THERMOFF;
994 }
995 }
996 } else if (cmd_params[0] == "reboot") {
997 cmd = ANDROID_RB_RESTART2;
998 if (cmd_params.size() >= 2) {
999 reboot_target = cmd_params[1];
1000 if (reboot_target == "userspace") {
1001 LOG(INFO) << "Userspace reboot requested";
1002 userspace_reboot = true;
1003 }
1004 // adb reboot fastboot should boot into bootloader for devices not
1005 // supporting logical partitions.
1006 if (reboot_target == "fastboot" &&
1007 !android::base::GetBoolProperty("ro.boot.dynamic_partitions", false)) {
1008 reboot_target = "bootloader";
1009 }
1010 // When rebooting to the bootloader notify the bootloader writing
1011 // also the BCB.
1012 if (reboot_target == "bootloader") {
1013 std::string err;
1014 if (!write_reboot_bootloader(&err)) {
1015 LOG(ERROR) << "reboot-bootloader: Error writing "
1016 "bootloader_message: "
1017 << err;
1018 }
1019 } else if (reboot_target == "recovery") {
1020 bootloader_message boot = {};
1021 if (std::string err; !read_bootloader_message(&boot, &err)) {
1022 LOG(ERROR) << "Failed to read bootloader message: " << err;
1023 }
1024 // Update the boot command field if it's empty, and preserve
1025 // the other arguments in the bootloader message.
1026 if (!CommandIsPresent(&boot)) {
1027 strlcpy(boot.command, "boot-recovery", sizeof(boot.command));
1028 if (std::string err; !write_bootloader_message(boot, &err)) {
1029 LOG(ERROR) << "Failed to set bootloader message: " << err;
1030 return;
1031 }
1032 }
1033 } else if (reboot_target == "sideload" || reboot_target == "sideload-auto-reboot" ||
1034 reboot_target == "fastboot") {
1035 std::string arg = reboot_target == "sideload-auto-reboot" ? "sideload_auto_reboot"
1036 : reboot_target;
1037 const std::vector<std::string> options = {
1038 "--" + arg,
1039 };
1040 std::string err;
1041 if (!write_bootloader_message(options, &err)) {
1042 LOG(ERROR) << "Failed to set bootloader message: " << err;
1043 return;
1044 }
1045 reboot_target = "recovery";
1046 }
1047
1048 // If there are additional parameter, pass them along
1049 for (size_t i = 2; (cmd_params.size() > i) && cmd_params[i].size(); ++i) {
1050 reboot_target += "," + cmd_params[i];
1051 }
1052 }
1053 } else {
1054 command_invalid = true;
1055 }
1056 if (command_invalid) {
1057 LOG(ERROR) << "powerctl: unrecognized command '" << command << "'";
1058 return;
1059 }
1060
1061 // We do not want to process any messages (queue'ing triggers, shutdown messages, control
1062 // messages, etc) from properties during reboot.
1063 StopSendingMessages();
1064
1065 if (userspace_reboot) {
1066 HandleUserspaceReboot();
1067 return;
1068 }
1069
1070 LOG(INFO) << "Clear action queue and start shutdown trigger";
1071 ActionManager::GetInstance().ClearQueue();
1072 // Queue shutdown trigger first
1073 ActionManager::GetInstance().QueueEventTrigger("shutdown");
1074 // Queue built-in shutdown_done
1075 auto shutdown_handler = [cmd, command, reboot_target, run_fsck](const BuiltinArguments&) {
1076 DoReboot(cmd, command, reboot_target, run_fsck);
1077 return Result<void>{};
1078 };
1079 ActionManager::GetInstance().QueueBuiltinAction(shutdown_handler, "shutdown_done");
1080
1081 EnterShutdown();
1082 }
1083
IsShuttingDown()1084 bool IsShuttingDown() {
1085 return shutting_down;
1086 }
1087
1088 } // namespace init
1089 } // namespace android
1090