1 /*
2 * Copyright (c) 2020, The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #define LOG_TAG "carwatchdogd"
18 #define DEBUG false // STOPSHIP if true.
19
20 #include "IoOveruseMonitor.h"
21
22 #include "PackageInfoResolver.h"
23
24 #include <WatchdogProperties.sysprop.h>
25 #include <android-base/file.h>
26 #include <android-base/strings.h>
27 #include <android/automotive/watchdog/internal/PackageIdentifier.h>
28 #include <android/automotive/watchdog/internal/UidType.h>
29 #include <binder/IPCThreadState.h>
30 #include <binder/Status.h>
31 #include <log/log.h>
32 #include <processgroup/sched_policy.h>
33
34 #include <pthread.h>
35
36 #include <limits>
37 #include <thread> // NOLINT(build/c++11)
38
39 namespace android {
40 namespace automotive {
41 namespace watchdog {
42
43 namespace {
44
45 using ::android::IPCThreadState;
46 using ::android::sp;
47 using ::android::automotive::watchdog::internal::ComponentType;
48 using ::android::automotive::watchdog::internal::IoOveruseConfiguration;
49 using ::android::automotive::watchdog::internal::IoUsageStats;
50 using ::android::automotive::watchdog::internal::PackageIdentifier;
51 using ::android::automotive::watchdog::internal::PackageInfo;
52 using ::android::automotive::watchdog::internal::PackageIoOveruseStats;
53 using ::android::automotive::watchdog::internal::ResourceOveruseConfiguration;
54 using ::android::automotive::watchdog::internal::UidType;
55 using ::android::automotive::watchdog::internal::UserPackageIoUsageStats;
56 using ::android::base::EndsWith;
57 using ::android::base::Error;
58 using ::android::base::Result;
59 using ::android::base::StringPrintf;
60 using ::android::base::WriteStringToFd;
61 using ::android::binder::Status;
62
63 constexpr int64_t kMaxInt32 = std::numeric_limits<int32_t>::max();
64 constexpr int64_t kMaxInt64 = std::numeric_limits<int64_t>::max();
65 // Minimum written bytes to sync the stats with the Watchdog service.
66 constexpr int64_t kMinSyncWrittenBytes = 100 * 1024;
67 // Minimum percentage of threshold to warn killable applications.
68 constexpr double kDefaultIoOveruseWarnPercentage = 80;
69 // Maximum numer of system-wide stats (from periodic monitoring) to cache.
70 constexpr size_t kMaxPeriodicMonitorBufferSize = 1000;
71 constexpr const char* kHelpText =
72 "\n%s dump options:\n"
73 "%s <package name>, <package name>,...: Reset resource overuse stats for the given package "
74 "names. Value for this flag is a comma-separated value containing package names.\n";
75
uniquePackageIdStr(const std::string & name,userid_t userId)76 std::string uniquePackageIdStr(const std::string& name, userid_t userId) {
77 return StringPrintf("%s:%" PRId32, name.c_str(), userId);
78 }
79
uniquePackageIdStr(const PackageIdentifier & id)80 std::string uniquePackageIdStr(const PackageIdentifier& id) {
81 return uniquePackageIdStr(id.name, multiuser_get_user_id(id.uid));
82 }
83
sum(const PerStateBytes & lhs,const PerStateBytes & rhs)84 PerStateBytes sum(const PerStateBytes& lhs, const PerStateBytes& rhs) {
85 const auto sum = [](const int64_t& l, const int64_t& r) -> int64_t {
86 return (kMaxInt64 - l) > r ? (l + r) : kMaxInt64;
87 };
88 PerStateBytes result;
89 result.foregroundBytes = sum(lhs.foregroundBytes, rhs.foregroundBytes);
90 result.backgroundBytes = sum(lhs.backgroundBytes, rhs.backgroundBytes);
91 result.garageModeBytes = sum(lhs.garageModeBytes, rhs.garageModeBytes);
92 return result;
93 }
94
diff(const PerStateBytes & lhs,const PerStateBytes & rhs)95 PerStateBytes diff(const PerStateBytes& lhs, const PerStateBytes& rhs) {
96 const auto sub = [](const int64_t& l, const int64_t& r) -> int64_t {
97 return l >= r ? (l - r) : 0;
98 };
99 PerStateBytes result;
100 result.foregroundBytes = sub(lhs.foregroundBytes, rhs.foregroundBytes);
101 result.backgroundBytes = sub(lhs.backgroundBytes, rhs.backgroundBytes);
102 result.garageModeBytes = sub(lhs.garageModeBytes, rhs.garageModeBytes);
103 return result;
104 }
105
calculateStartAndDuration(struct tm currentTm)106 std::tuple<int64_t, int64_t> calculateStartAndDuration(struct tm currentTm) {
107 // The stats are stored per-day so the start time is always the beginning of the day.
108 auto startTm = currentTm;
109 startTm.tm_sec = 0;
110 startTm.tm_min = 0;
111 startTm.tm_hour = 0;
112
113 int64_t startTime = static_cast<int64_t>(mktime(&startTm));
114 int64_t currentEpochSeconds = static_cast<int64_t>(mktime(¤tTm));
115 return std::make_tuple(startTime, currentEpochSeconds - startTime);
116 }
117
totalPerStateBytes(PerStateBytes perStateBytes)118 int64_t totalPerStateBytes(PerStateBytes perStateBytes) {
119 const auto sum = [](const int64_t& l, const int64_t& r) -> int64_t {
120 return kMaxInt64 - l > r ? (l + r) : kMaxInt64;
121 };
122 return sum(perStateBytes.foregroundBytes,
123 sum(perStateBytes.backgroundBytes, perStateBytes.garageModeBytes));
124 }
125
calculateOveruseAndForgivenBytes(PerStateBytes writtenBytes,PerStateBytes threshold)126 std::tuple<int32_t, PerStateBytes> calculateOveruseAndForgivenBytes(PerStateBytes writtenBytes,
127 PerStateBytes threshold) {
128 const auto div = [](const int64_t& l, const int64_t& r) -> int32_t {
129 return r > 0 ? (l / r) : 1;
130 };
131 const auto mul = [](const int32_t& l, const int32_t& r) -> int32_t {
132 if (l == 0 || r == 0) {
133 return 0;
134 }
135 return (kMaxInt32 / r) > l ? (l * r) : kMaxInt32;
136 };
137 const auto sum = [](const int32_t& l, const int32_t& r) -> int32_t {
138 return (kMaxInt32 - l) > r ? (l + r) : kMaxInt32;
139 };
140 int32_t foregroundOveruses = div(writtenBytes.foregroundBytes, threshold.foregroundBytes);
141 int32_t backgroundOveruses = div(writtenBytes.backgroundBytes, threshold.backgroundBytes);
142 int32_t garageModeOveruses = div(writtenBytes.garageModeBytes, threshold.garageModeBytes);
143 int32_t totalOveruses = sum(foregroundOveruses, sum(backgroundOveruses, garageModeOveruses));
144
145 PerStateBytes forgivenWriteBytes;
146 forgivenWriteBytes.foregroundBytes = mul(foregroundOveruses, threshold.foregroundBytes);
147 forgivenWriteBytes.backgroundBytes = mul(backgroundOveruses, threshold.backgroundBytes);
148 forgivenWriteBytes.garageModeBytes = mul(garageModeOveruses, threshold.garageModeBytes);
149
150 return std::make_tuple(totalOveruses, forgivenWriteBytes);
151 }
152
153 } // namespace
154
calculateStartAndDuration(const time_t & currentTime)155 std::tuple<int64_t, int64_t> calculateStartAndDuration(const time_t& currentTime) {
156 struct tm currentGmt;
157 gmtime_r(¤tTime, ¤tGmt);
158 return calculateStartAndDuration(currentGmt);
159 }
160
IoOveruseMonitor(const android::sp<IWatchdogServiceHelper> & watchdogServiceHelper)161 IoOveruseMonitor::IoOveruseMonitor(
162 const android::sp<IWatchdogServiceHelper>& watchdogServiceHelper) :
163 mMinSyncWrittenBytes(kMinSyncWrittenBytes),
164 mWatchdogServiceHelper(watchdogServiceHelper),
165 mDidReadTodayPrevBootStats(false),
166 mSystemWideWrittenBytes({}),
167 mPeriodicMonitorBufferSize(0),
168 mLastSystemWideIoMonitorTime(0),
169 mUserPackageDailyIoUsageById({}),
170 mIoOveruseWarnPercentage(0),
171 mLastUserPackageIoMonitorTime(0),
172 mOveruseListenersByUid({}),
173 mBinderDeathRecipient(sp<BinderDeathRecipient>::make(this)) {}
174
init()175 Result<void> IoOveruseMonitor::init() {
176 std::unique_lock writeLock(mRwMutex);
177 if (isInitializedLocked()) {
178 return Error() << "Cannot initialize " << name() << " more than once";
179 }
180 mPeriodicMonitorBufferSize = static_cast<size_t>(
181 sysprop::periodicMonitorBufferSize().value_or(kDefaultPeriodicMonitorBufferSize));
182 if (mPeriodicMonitorBufferSize == 0 ||
183 mPeriodicMonitorBufferSize > kMaxPeriodicMonitorBufferSize) {
184 return Error() << "Periodic monitor buffer size cannot be zero or above "
185 << kDefaultPeriodicMonitorBufferSize << ". Received "
186 << mPeriodicMonitorBufferSize;
187 }
188 mIoOveruseWarnPercentage = static_cast<double>(
189 sysprop::ioOveruseWarnPercentage().value_or(kDefaultIoOveruseWarnPercentage));
190 mIoOveruseConfigs = sp<IoOveruseConfigs>::make();
191 mPackageInfoResolver = PackageInfoResolver::getInstance();
192 mPackageInfoResolver->setPackageConfigurations(mIoOveruseConfigs->vendorPackagePrefixes(),
193 mIoOveruseConfigs->packagesToAppCategories());
194 if (DEBUG) {
195 ALOGD("Initialized %s data processor", name().c_str());
196 }
197 return {};
198 }
199
terminate()200 void IoOveruseMonitor::terminate() {
201 std::unique_lock writeLock(mRwMutex);
202
203 ALOGW("Terminating %s", name().c_str());
204 mWatchdogServiceHelper.clear();
205 mIoOveruseConfigs.clear();
206 mSystemWideWrittenBytes.clear();
207 mUserPackageDailyIoUsageById.clear();
208 for (const auto& [uid, listener] : mOveruseListenersByUid) {
209 BnResourceOveruseListener::asBinder(listener)->unlinkToDeath(mBinderDeathRecipient);
210 }
211 mBinderDeathRecipient.clear();
212 mOveruseListenersByUid.clear();
213 if (DEBUG) {
214 ALOGD("Terminated %s data processor", name().c_str());
215 }
216 return;
217 }
218
219 Result<void> IoOveruseMonitor::onPeriodicCollection(
220 time_t time, SystemState systemState,
221 const android::wp<UidStatsCollectorInterface>& uidStatsCollector,
222 [[maybe_unused]] const android::wp<ProcStat>& procStat) {
223 android::sp<UidStatsCollectorInterface> uidStatsCollectorSp = uidStatsCollector.promote();
224 if (uidStatsCollectorSp == nullptr) {
225 return Error() << "Per-UID I/O stats collector must not be null";
226 }
227
228 std::unique_lock writeLock(mRwMutex);
229 if (!mDidReadTodayPrevBootStats) {
230 syncTodayIoUsageStatsLocked();
231 }
232 struct tm prevGmt, curGmt;
233 gmtime_r(&mLastUserPackageIoMonitorTime, &prevGmt);
234 gmtime_r(&time, &curGmt);
235 if (prevGmt.tm_yday != curGmt.tm_yday || prevGmt.tm_year != curGmt.tm_year) {
236 /*
237 * Date changed so reset the daily I/O usage cache. CarWatchdogService automatically handles
238 * date change on |CarWatchdogService.latestIoOveruseStats| call.
239 */
240 mUserPackageDailyIoUsageById.clear();
241 }
242 mLastUserPackageIoMonitorTime = time;
243 const auto [startTime, durationInSeconds] = calculateStartAndDuration(curGmt);
244
245 auto uidStats = uidStatsCollectorSp->deltaStats();
246 if (uidStats.empty()) {
247 return {};
248 }
249 std::unordered_map<uid_t, IoOveruseStats> overusingNativeStats;
250 bool isGarageModeActive = systemState == SystemState::GARAGE_MODE;
251 for (const auto& curUidStats : uidStats) {
252 if (curUidStats.ioStats.sumWriteBytes() == 0 || !curUidStats.hasPackageInfo()) {
253 /* 1. Ignore UIDs with zero written bytes since the last collection because they are
254 * either already accounted for or no writes made since system start.
255 *
256 * 2. UID stats without package info is not useful because the stats isn't attributed to
257 * any package/service.
258 */
259 continue;
260 }
261 UserPackageIoUsage curUsage(curUidStats.packageInfo, curUidStats.ioStats,
262 isGarageModeActive);
263 UserPackageIoUsage* dailyIoUsage;
264 if (auto cachedUsage = mUserPackageDailyIoUsageById.find(curUsage.id());
265 cachedUsage != mUserPackageDailyIoUsageById.end()) {
266 cachedUsage->second += curUsage;
267 dailyIoUsage = &cachedUsage->second;
268 } else {
269 if (auto prevBootStats = mPrevBootIoUsageStatsById.find(curUsage.id());
270 prevBootStats != mPrevBootIoUsageStatsById.end()) {
271 curUsage += prevBootStats->second;
272 mPrevBootIoUsageStatsById.erase(prevBootStats);
273 }
274 const auto& [it, wasInserted] = mUserPackageDailyIoUsageById.insert(
275 std::pair(curUsage.id(), std::move(curUsage)));
276 dailyIoUsage = &it->second;
277 }
278
279 const auto threshold = mIoOveruseConfigs->fetchThreshold(dailyIoUsage->packageInfo);
280
281 const auto deltaWrittenBytes =
282 diff(dailyIoUsage->writtenBytes, dailyIoUsage->forgivenWriteBytes);
283 const auto [currentOveruses, forgivenWriteBytes] =
284 calculateOveruseAndForgivenBytes(deltaWrittenBytes, threshold);
285 dailyIoUsage->totalOveruses += currentOveruses;
286 dailyIoUsage->forgivenWriteBytes =
287 sum(dailyIoUsage->forgivenWriteBytes, forgivenWriteBytes);
288
289 PackageIoOveruseStats stats;
290 stats.uid = curUidStats.packageInfo.packageIdentifier.uid;
291 stats.shouldNotify = false;
292 stats.forgivenWriteBytes = dailyIoUsage->forgivenWriteBytes;
293 stats.ioOveruseStats.startTime = startTime;
294 stats.ioOveruseStats.durationInSeconds = durationInSeconds;
295 stats.ioOveruseStats.writtenBytes = dailyIoUsage->writtenBytes;
296 stats.ioOveruseStats.totalOveruses = dailyIoUsage->totalOveruses;
297 stats.ioOveruseStats.remainingWriteBytes = diff(threshold, deltaWrittenBytes);
298 stats.ioOveruseStats.killableOnOveruse =
299 mIoOveruseConfigs->isSafeToKill(dailyIoUsage->packageInfo);
300
301 const auto& remainingWriteBytes = stats.ioOveruseStats.remainingWriteBytes;
__anon39c7dd3a0802(double remaining, double threshold) 302 const auto exceedsWarnThreshold = [&](double remaining, double threshold) {
303 if (threshold == 0) {
304 return true;
305 }
306 double usedPercent = (100 - (remaining / threshold) * 100);
307 return usedPercent > mIoOveruseWarnPercentage;
308 };
309 bool shouldSyncWatchdogService =
310 (totalPerStateBytes(dailyIoUsage->writtenBytes) -
311 dailyIoUsage->lastSyncedWrittenBytes) >= mMinSyncWrittenBytes;
312 if (currentOveruses > 0) {
313 dailyIoUsage->isPackageWarned = false;
314 /*
315 * Send notifications for native service I/O overuses as well because system listeners
316 * need to be notified of all I/O overuses.
317 */
318 stats.shouldNotify = true;
319 if (dailyIoUsage->packageInfo.uidType == UidType::NATIVE) {
320 overusingNativeStats[stats.uid] = stats.ioOveruseStats;
321 }
322 shouldSyncWatchdogService = true;
323 } else if (dailyIoUsage->packageInfo.uidType != UidType::NATIVE &&
324 stats.ioOveruseStats.killableOnOveruse && !dailyIoUsage->isPackageWarned &&
325 (exceedsWarnThreshold(remainingWriteBytes.foregroundBytes,
326 threshold.foregroundBytes) ||
327 exceedsWarnThreshold(remainingWriteBytes.backgroundBytes,
328 threshold.backgroundBytes) ||
329 exceedsWarnThreshold(remainingWriteBytes.garageModeBytes,
330 threshold.garageModeBytes))) {
331 /*
332 * No need to warn native services or applications that won't be killed on I/O overuse
333 * as they will be sent a notification when they exceed their daily threshold.
334 */
335 stats.shouldNotify = true;
336 // Avoid duplicate warning before the daily threshold exceeded notification is sent.
337 dailyIoUsage->isPackageWarned = true;
338 shouldSyncWatchdogService = true;
339 }
340 if (shouldSyncWatchdogService) {
341 dailyIoUsage->lastSyncedWrittenBytes = totalPerStateBytes(dailyIoUsage->writtenBytes);
342 mLatestIoOveruseStats.emplace_back(std::move(stats));
343 }
344 }
345 if (!overusingNativeStats.empty()) {
346 notifyNativePackagesLocked(overusingNativeStats);
347 }
348 if (mLatestIoOveruseStats.empty()) {
349 return {};
350 }
351 if (const auto status = mWatchdogServiceHelper->latestIoOveruseStats(mLatestIoOveruseStats);
352 !status.isOk()) {
353 // Don't clear the cache as it can be pushed again on the next collection.
354 ALOGW("Failed to push the latest I/O overuse stats to watchdog service: %s",
355 status.toString8().c_str());
356 } else {
357 mLatestIoOveruseStats.clear();
358 if (DEBUG) {
359 ALOGD("Pushed latest I/O overuse stats to watchdog service");
360 }
361 }
362 return {};
363 }
364
365 Result<void> IoOveruseMonitor::onCustomCollection(
366 time_t time, SystemState systemState,
367 [[maybe_unused]] const std::unordered_set<std::string>& filterPackages,
368 const android::wp<UidStatsCollectorInterface>& uidStatsCollector,
369 const android::wp<ProcStat>& procStat) {
370 // Nothing special for custom collection.
371 return onPeriodicCollection(time, systemState, uidStatsCollector, procStat);
372 }
373
onPeriodicMonitor(time_t time,const android::wp<IProcDiskStatsInterface> & procDiskStats,const std::function<void ()> & alertHandler)374 Result<void> IoOveruseMonitor::onPeriodicMonitor(
375 time_t time, const android::wp<IProcDiskStatsInterface>& procDiskStats,
376 const std::function<void()>& alertHandler) {
377 if (procDiskStats == nullptr) {
378 return Error() << "Proc disk stats collector must not be null";
379 }
380
381 std::unique_lock writeLock(mRwMutex);
382 if (mLastSystemWideIoMonitorTime == 0) {
383 /*
384 * Do not record the first disk stats as it reflects the aggregated disks stats since the
385 * system boot up and is not in sync with the polling period. This will lead to spurious
386 * I/O overuse alerting.
387 */
388 mLastSystemWideIoMonitorTime = time;
389 return {};
390 }
391 const auto diskStats = procDiskStats.promote()->deltaSystemWideDiskStats();
392 mSystemWideWrittenBytes.push_back(
393 {.pollDurationInSecs = difftime(time, mLastSystemWideIoMonitorTime),
394 .bytesInKib = diskStats.numKibWritten});
395 for (const auto& threshold : mIoOveruseConfigs->systemWideAlertThresholds()) {
396 int64_t accountedWrittenKib = 0;
397 double accountedDurationInSecs = 0;
398 size_t accountedPolls = 0;
399 for (auto rit = mSystemWideWrittenBytes.rbegin(); rit != mSystemWideWrittenBytes.rend();
400 ++rit) {
401 accountedWrittenKib += rit->bytesInKib;
402 accountedDurationInSecs += rit->pollDurationInSecs;
403 ++accountedPolls;
404 if (accountedDurationInSecs >= threshold.durationInSeconds) {
405 break;
406 }
407 }
408 // Heuristic to handle spurious alerting when the buffer is partially filled.
409 if (const size_t bufferSize = mSystemWideWrittenBytes.size();
410 accountedPolls == bufferSize && bufferSize < mPeriodicMonitorBufferSize + 1 &&
411 threshold.durationInSeconds > accountedDurationInSecs) {
412 continue;
413 }
414 const double thresholdKbps = threshold.writtenBytesPerSecond / 1024.0;
415 if (const auto kbps = accountedWrittenKib / accountedDurationInSecs;
416 kbps >= thresholdKbps) {
417 alertHandler();
418 break;
419 }
420 }
421 if (mSystemWideWrittenBytes.size() > mPeriodicMonitorBufferSize) {
422 mSystemWideWrittenBytes.erase(mSystemWideWrittenBytes.begin()); // Erase the oldest entry.
423 }
424 mLastSystemWideIoMonitorTime = time;
425 return {};
426 }
427
428 Result<void> IoOveruseMonitor::onDump([[maybe_unused]] int fd) const {
429 // TODO(b/183436216): Dump the list of killed/disabled packages. Dump the list of packages that
430 // exceed xx% of their threshold.
431 return {};
432 }
433
dumpHelpText(int fd) const434 bool IoOveruseMonitor::dumpHelpText(int fd) const {
435 return WriteStringToFd(StringPrintf(kHelpText, name().c_str(), kResetResourceOveruseStatsFlag),
436 fd);
437 }
438
syncTodayIoUsageStatsLocked()439 void IoOveruseMonitor::syncTodayIoUsageStatsLocked() {
440 std::vector<UserPackageIoUsageStats> userPackageIoUsageStats;
441 if (const auto status = mWatchdogServiceHelper->getTodayIoUsageStats(&userPackageIoUsageStats);
442 !status.isOk()) {
443 ALOGE("Failed to fetch today I/O usage stats collected during previous boot: %s",
444 status.exceptionMessage().c_str());
445 return;
446 }
447 for (const auto& statsEntry : userPackageIoUsageStats) {
448 std::string uniqueId = uniquePackageIdStr(statsEntry.packageName,
449 static_cast<userid_t>(statsEntry.userId));
450 if (auto it = mUserPackageDailyIoUsageById.find(uniqueId);
451 it != mUserPackageDailyIoUsageById.end()) {
452 it->second += statsEntry.ioUsageStats;
453 continue;
454 }
455 mPrevBootIoUsageStatsById.insert(std::pair(uniqueId, statsEntry.ioUsageStats));
456 }
457 mDidReadTodayPrevBootStats = true;
458 }
459
notifyNativePackagesLocked(const std::unordered_map<uid_t,IoOveruseStats> & statsByUid)460 void IoOveruseMonitor::notifyNativePackagesLocked(
461 const std::unordered_map<uid_t, IoOveruseStats>& statsByUid) {
462 for (const auto& [uid, ioOveruseStats] : statsByUid) {
463 IResourceOveruseListener* listener;
464 if (const auto it = mOveruseListenersByUid.find(uid); it == mOveruseListenersByUid.end()) {
465 continue;
466 } else {
467 listener = it->second.get();
468 }
469 ResourceOveruseStats stats;
470 stats.set<ResourceOveruseStats::ioOveruseStats>(ioOveruseStats);
471 listener->onOveruse(stats);
472 }
473 if (DEBUG) {
474 ALOGD("Notified native packages on I/O overuse");
475 }
476 }
477
updateResourceOveruseConfigurations(const std::vector<ResourceOveruseConfiguration> & configs)478 Result<void> IoOveruseMonitor::updateResourceOveruseConfigurations(
479 const std::vector<ResourceOveruseConfiguration>& configs) {
480 std::unique_lock writeLock(mRwMutex);
481 if (!isInitializedLocked()) {
482 return Error(Status::EX_ILLEGAL_STATE) << name() << " is not initialized";
483 }
484 if (const auto result = mIoOveruseConfigs->update(configs); !result.ok()) {
485 return result;
486 }
487 std::thread writeToDiskThread([&]() {
488 if (set_sched_policy(0, SP_BACKGROUND) != 0) {
489 ALOGW("Failed to set background scheduling priority for writing resource overuse "
490 "configs to disk");
491 }
492 if (int result = pthread_setname_np(pthread_self(), "ResOveruseCfgWr"); result != 0) {
493 ALOGE("Failed to set thread name to 'ResOveruseCfgWr'");
494 }
495 std::unique_lock writeLock(mRwMutex);
496 if (const auto result = mIoOveruseConfigs->writeToDisk(); !result.ok()) {
497 ALOGE("Failed to write resource overuse configs to disk: %s",
498 result.error().message().c_str());
499 }
500 });
501 writeToDiskThread.detach();
502 return {};
503 }
504
getResourceOveruseConfigurations(std::vector<ResourceOveruseConfiguration> * configs) const505 Result<void> IoOveruseMonitor::getResourceOveruseConfigurations(
506 std::vector<ResourceOveruseConfiguration>* configs) const {
507 std::shared_lock readLock(mRwMutex);
508 if (!isInitializedLocked()) {
509 return Error(Status::EX_ILLEGAL_STATE) << name() << " is not initialized";
510 }
511 mIoOveruseConfigs->get(configs);
512 return {};
513 }
514
addIoOveruseListener(const sp<IResourceOveruseListener> & listener)515 Result<void> IoOveruseMonitor::addIoOveruseListener(const sp<IResourceOveruseListener>& listener) {
516 pid_t callingPid = IPCThreadState::self()->getCallingPid();
517 uid_t callingUid = IPCThreadState::self()->getCallingUid();
518 std::unique_lock writeLock(mRwMutex);
519 auto binder = BnResourceOveruseListener::asBinder(listener);
520 if (findListenerAndProcessLocked(binder, nullptr)) {
521 ALOGW("Failed to register the I/O overuse listener (pid: %d, uid: %d) as it is already "
522 "registered",
523 callingPid, callingUid);
524 return {};
525 }
526 if (const auto status = binder->linkToDeath(mBinderDeathRecipient); status != OK) {
527 return Error(Status::EX_ILLEGAL_STATE)
528 << "(pid " << callingPid << ", uid: " << callingUid << ") is dead";
529 }
530 mOveruseListenersByUid[callingUid] = listener;
531 if (DEBUG) {
532 ALOGD("Added I/O overuse listener for uid: %d", callingUid);
533 }
534 return {};
535 }
536
removeIoOveruseListener(const sp<IResourceOveruseListener> & listener)537 Result<void> IoOveruseMonitor::removeIoOveruseListener(
538 const sp<IResourceOveruseListener>& listener) {
539 std::unique_lock writeLock(mRwMutex);
540 const auto processor = [&](ListenersByUidMap& listeners, ListenersByUidMap::const_iterator it) {
541 auto binder = BnResourceOveruseListener::asBinder(it->second);
542 binder->unlinkToDeath(mBinderDeathRecipient);
543 listeners.erase(it);
544 };
545 if (const auto binder = BnResourceOveruseListener::asBinder(listener);
546 !findListenerAndProcessLocked(binder, processor)) {
547 return Error(Status::EX_ILLEGAL_ARGUMENT) << "Listener is not previously registered";
548 }
549 if (DEBUG) {
550 ALOGD("Removed I/O overuse listener for uid: %d", IPCThreadState::self()->getCallingUid());
551 }
552 return {};
553 }
554
getIoOveruseStats(IoOveruseStats * ioOveruseStats) const555 Result<void> IoOveruseMonitor::getIoOveruseStats(IoOveruseStats* ioOveruseStats) const {
556 if (!isInitialized()) {
557 return Error(Status::EX_ILLEGAL_STATE) << "I/O overuse monitor is not initialized";
558 }
559 uid_t callingUid = IPCThreadState::self()->getCallingUid();
560 const auto packageInfosByUid = mPackageInfoResolver->getPackageInfosForUids({callingUid});
561 const PackageInfo* packageInfo;
562 if (const auto it = packageInfosByUid.find(callingUid); it == packageInfosByUid.end()) {
563 return Error(Status::EX_ILLEGAL_ARGUMENT)
564 << "Package information not available for calling UID(" << callingUid << ")";
565 } else {
566 packageInfo = &it->second;
567 }
568 std::shared_lock readLock(mRwMutex);
569 const UserPackageIoUsage* dailyIoUsage;
570 if (const auto it = mUserPackageDailyIoUsageById.find(
571 uniquePackageIdStr(packageInfo->packageIdentifier));
572 it == mUserPackageDailyIoUsageById.end()) {
573 return Error(Status::EX_ILLEGAL_ARGUMENT)
574 << "Calling UID " << callingUid << " doesn't have I/O overuse stats";
575 } else {
576 dailyIoUsage = &it->second;
577 }
578 ioOveruseStats->killableOnOveruse = mIoOveruseConfigs->isSafeToKill(*packageInfo);
579 const auto thresholdBytes = mIoOveruseConfigs->fetchThreshold(*packageInfo);
580 ioOveruseStats->remainingWriteBytes =
581 diff(thresholdBytes,
582 diff(dailyIoUsage->writtenBytes, dailyIoUsage->forgivenWriteBytes));
583 ioOveruseStats->totalOveruses = dailyIoUsage->totalOveruses;
584 ioOveruseStats->writtenBytes = dailyIoUsage->writtenBytes;
585 const auto [startTime, durationInSeconds] =
586 calculateStartAndDuration(mLastUserPackageIoMonitorTime);
587 ioOveruseStats->startTime = startTime;
588 ioOveruseStats->durationInSeconds = durationInSeconds;
589 if (DEBUG) {
590 ALOGD("Returning I/O overuse stats for uid: %d", callingUid);
591 }
592 return {};
593 }
594
resetIoOveruseStats(const std::vector<std::string> & packageNames)595 Result<void> IoOveruseMonitor::resetIoOveruseStats(const std::vector<std::string>& packageNames) {
596 if (const auto status = mWatchdogServiceHelper->resetResourceOveruseStats(packageNames);
597 !status.isOk()) {
598 return Error() << "Failed to reset stats in watchdog service: " << status.toString8();
599 }
600 std::unordered_set<std::string> uniquePackageNames;
601 std::copy(packageNames.begin(), packageNames.end(),
602 std::inserter(uniquePackageNames, uniquePackageNames.end()));
603 std::unique_lock writeLock(mRwMutex);
604 for (auto& [key, usage] : mUserPackageDailyIoUsageById) {
605 if (uniquePackageNames.find(usage.packageInfo.packageIdentifier.name) !=
606 uniquePackageNames.end()) {
607 usage.resetStats();
608 }
609 }
610 return {};
611 }
612
removeStatsForUser(userid_t userId)613 void IoOveruseMonitor::removeStatsForUser(userid_t userId) {
614 std::unique_lock writeLock(mRwMutex);
615 for (auto it = mUserPackageDailyIoUsageById.begin();
616 it != mUserPackageDailyIoUsageById.end();) {
617 if (multiuser_get_user_id(it->second.packageInfo.packageIdentifier.uid) == userId) {
618 it = mUserPackageDailyIoUsageById.erase(it);
619 } else {
620 ++it;
621 }
622 }
623 // |mPrevBootIoUsageStatsById| keys are constructed using |uniquePackageIdStr| method. Thus, the
624 // key suffix would contain the userId. The value in this map is |IoUsageStats|, which doesn't
625 // contain the userId, so this is the only way to delete cached previous boot stats for
626 // the removed user.
627 std::string keySuffix = StringPrintf(":%" PRId32, userId);
628 for (auto it = mPrevBootIoUsageStatsById.begin(); it != mPrevBootIoUsageStatsById.end();) {
629 if (EndsWith(it->first, keySuffix)) {
630 it = mPrevBootIoUsageStatsById.erase(it);
631 } else {
632 ++it;
633 }
634 }
635 for (auto it = mLatestIoOveruseStats.begin(); it != mLatestIoOveruseStats.end();) {
636 if (multiuser_get_user_id(it->uid) == userId) {
637 it = mLatestIoOveruseStats.erase(it);
638 } else {
639 ++it;
640 }
641 }
642 }
643
handleBinderDeath(const wp<IBinder> & who)644 void IoOveruseMonitor::handleBinderDeath(const wp<IBinder>& who) {
645 std::unique_lock writeLock(mRwMutex);
646 IBinder* binder = who.unsafe_get();
647 findListenerAndProcessLocked(binder,
648 [&](ListenersByUidMap& listeners,
649 ListenersByUidMap::const_iterator it) {
650 ALOGW("Resource overuse notification handler died for uid(%d)",
651 it->first);
652 listeners.erase(it);
653 });
654 }
655
findListenerAndProcessLocked(const sp<IBinder> & binder,const Processor & processor)656 bool IoOveruseMonitor::findListenerAndProcessLocked(const sp<IBinder>& binder,
657 const Processor& processor) {
658 for (auto it = mOveruseListenersByUid.begin(); it != mOveruseListenersByUid.end(); ++it) {
659 if (BnResourceOveruseListener::asBinder(it->second) != binder) {
660 continue;
661 }
662 if (processor != nullptr) {
663 processor(mOveruseListenersByUid, it);
664 }
665 return true;
666 }
667 return false;
668 }
669
UserPackageIoUsage(const PackageInfo & pkgInfo,const UidIoStats & uidIoStats,const bool isGarageModeActive)670 IoOveruseMonitor::UserPackageIoUsage::UserPackageIoUsage(const PackageInfo& pkgInfo,
671 const UidIoStats& uidIoStats,
672 const bool isGarageModeActive) {
673 packageInfo = pkgInfo;
674 if (isGarageModeActive) {
675 writtenBytes.garageModeBytes = uidIoStats.sumWriteBytes();
676 } else {
677 writtenBytes.foregroundBytes = uidIoStats.metrics[WRITE_BYTES][FOREGROUND];
678 writtenBytes.backgroundBytes = uidIoStats.metrics[WRITE_BYTES][BACKGROUND];
679 }
680 }
681
operator +=(const UserPackageIoUsage & r)682 IoOveruseMonitor::UserPackageIoUsage& IoOveruseMonitor::UserPackageIoUsage::operator+=(
683 const UserPackageIoUsage& r) {
684 if (id() == r.id()) {
685 packageInfo = r.packageInfo;
686 }
687 writtenBytes = sum(writtenBytes, r.writtenBytes);
688 return *this;
689 }
690
operator +=(const IoUsageStats & ioUsageStats)691 IoOveruseMonitor::UserPackageIoUsage& IoOveruseMonitor::UserPackageIoUsage::operator+=(
692 const IoUsageStats& ioUsageStats) {
693 writtenBytes = sum(writtenBytes, ioUsageStats.writtenBytes);
694 forgivenWriteBytes = sum(forgivenWriteBytes, ioUsageStats.forgivenWriteBytes);
695 totalOveruses += ioUsageStats.totalOveruses;
696 return *this;
697 }
698
id() const699 const std::string IoOveruseMonitor::UserPackageIoUsage::id() const {
700 return uniquePackageIdStr(packageInfo.packageIdentifier);
701 }
702
resetStats()703 void IoOveruseMonitor::UserPackageIoUsage::resetStats() {
704 writtenBytes = {};
705 forgivenWriteBytes = {};
706 totalOveruses = 0;
707 isPackageWarned = false;
708 lastSyncedWrittenBytes = 0;
709 }
710
711 } // namespace watchdog
712 } // namespace automotive
713 } // namespace android
714