• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2012-2014 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 <ctype.h>
18 #include <errno.h>
19 #include <stdio.h>
20 #include <string.h>
21 #include <sys/user.h>
22 #include <time.h>
23 #include <unistd.h>
24 
25 #include <unordered_map>
26 
27 #include <cutils/properties.h>
28 #include <log/logger.h>
29 
30 #include "LogBuffer.h"
31 #include "LogKlog.h"
32 #include "LogReader.h"
33 
34 // Default
35 #define LOG_BUFFER_SIZE (256 * 1024) // Tuned with ro.logd.size per-platform
36 #define log_buffer_size(id) mMaxSize[id]
37 #define LOG_BUFFER_MIN_SIZE (64 * 1024UL)
38 #define LOG_BUFFER_MAX_SIZE (256 * 1024 * 1024UL)
39 
valid_size(unsigned long value)40 static bool valid_size(unsigned long value) {
41     if ((value < LOG_BUFFER_MIN_SIZE) || (LOG_BUFFER_MAX_SIZE < value)) {
42         return false;
43     }
44 
45     long pages = sysconf(_SC_PHYS_PAGES);
46     if (pages < 1) {
47         return true;
48     }
49 
50     long pagesize = sysconf(_SC_PAGESIZE);
51     if (pagesize <= 1) {
52         pagesize = PAGE_SIZE;
53     }
54 
55     // maximum memory impact a somewhat arbitrary ~3%
56     pages = (pages + 31) / 32;
57     unsigned long maximum = pages * pagesize;
58 
59     if ((maximum < LOG_BUFFER_MIN_SIZE) || (LOG_BUFFER_MAX_SIZE < maximum)) {
60         return true;
61     }
62 
63     return value <= maximum;
64 }
65 
property_get_size(const char * key)66 static unsigned long property_get_size(const char *key) {
67     char property[PROPERTY_VALUE_MAX];
68     property_get(key, property, "");
69 
70     char *cp;
71     unsigned long value = strtoul(property, &cp, 10);
72 
73     switch(*cp) {
74     case 'm':
75     case 'M':
76         value *= 1024;
77     /* FALLTHRU */
78     case 'k':
79     case 'K':
80         value *= 1024;
81     /* FALLTHRU */
82     case '\0':
83         break;
84 
85     default:
86         value = 0;
87     }
88 
89     if (!valid_size(value)) {
90         value = 0;
91     }
92 
93     return value;
94 }
95 
init()96 void LogBuffer::init() {
97     static const char global_tuneable[] = "persist.logd.size"; // Settings App
98     static const char global_default[] = "ro.logd.size";       // BoardConfig.mk
99 
100     unsigned long default_size = property_get_size(global_tuneable);
101     if (!default_size) {
102         default_size = property_get_size(global_default);
103         if (!default_size) {
104             default_size = property_get_bool("ro.config.low_ram",
105                                              BOOL_DEFAULT_FALSE)
106                 ? LOG_BUFFER_MIN_SIZE // 64K
107                 : LOG_BUFFER_SIZE;    // 256K
108         }
109     }
110 
111     log_id_for_each(i) {
112         mLastSet[i] = false;
113         mLast[i] = mLogElements.begin();
114 
115         char key[PROP_NAME_MAX];
116 
117         snprintf(key, sizeof(key), "%s.%s",
118                  global_tuneable, android_log_id_to_name(i));
119         unsigned long property_size = property_get_size(key);
120 
121         if (!property_size) {
122             snprintf(key, sizeof(key), "%s.%s",
123                      global_default, android_log_id_to_name(i));
124             property_size = property_get_size(key);
125         }
126 
127         if (!property_size) {
128             property_size = default_size;
129         }
130 
131         if (!property_size) {
132             property_size = LOG_BUFFER_SIZE;
133         }
134 
135         if (setSize(i, property_size)) {
136             setSize(i, LOG_BUFFER_MIN_SIZE);
137         }
138     }
139     bool lastMonotonic = monotonic;
140     monotonic = android_log_clockid() == CLOCK_MONOTONIC;
141     if (lastMonotonic != monotonic) {
142         //
143         // Fixup all timestamps, may not be 100% accurate, but better than
144         // throwing what we have away when we get 'surprised' by a change.
145         // In-place element fixup so no need to check reader-lock. Entries
146         // should already be in timestamp order, but we could end up with a
147         // few out-of-order entries if new monotonics come in before we
148         // are notified of the reinit change in status. A Typical example would
149         // be:
150         //  --------- beginning of system
151         //      10.494082   184   201 D Cryptfs : Just triggered post_fs_data
152         //  --------- beginning of kernel
153         //       0.000000     0     0 I         : Initializing cgroup subsys
154         // as the act of mounting /data would trigger persist.logd.timestamp to
155         // be corrected. 1/30 corner case YMMV.
156         //
157         pthread_mutex_lock(&mLogElementsLock);
158         LogBufferElementCollection::iterator it = mLogElements.begin();
159         while((it != mLogElements.end())) {
160             LogBufferElement *e = *it;
161             if (monotonic) {
162                 if (!android::isMonotonic(e->mRealTime)) {
163                     LogKlog::convertRealToMonotonic(e->mRealTime);
164                 }
165             } else {
166                 if (android::isMonotonic(e->mRealTime)) {
167                     LogKlog::convertMonotonicToReal(e->mRealTime);
168                 }
169             }
170             ++it;
171         }
172         pthread_mutex_unlock(&mLogElementsLock);
173     }
174 
175     // We may have been triggered by a SIGHUP. Release any sleeping reader
176     // threads to dump their current content.
177     //
178     // NB: this is _not_ performed in the context of a SIGHUP, it is
179     // performed during startup, and in context of reinit administrative thread
180     LogTimeEntry::lock();
181 
182     LastLogTimes::iterator times = mTimes.begin();
183     while(times != mTimes.end()) {
184         LogTimeEntry *entry = (*times);
185         if (entry->owned_Locked()) {
186             entry->triggerReader_Locked();
187         }
188         times++;
189     }
190 
191     LogTimeEntry::unlock();
192 }
193 
LogBuffer(LastLogTimes * times)194 LogBuffer::LogBuffer(LastLogTimes *times):
195         monotonic(android_log_clockid() == CLOCK_MONOTONIC),
196         mTimes(*times) {
197     pthread_mutex_init(&mLogElementsLock, NULL);
198 
199     init();
200 }
201 
log(log_id_t log_id,log_time realtime,uid_t uid,pid_t pid,pid_t tid,const char * msg,unsigned short len)202 int LogBuffer::log(log_id_t log_id, log_time realtime,
203                    uid_t uid, pid_t pid, pid_t tid,
204                    const char *msg, unsigned short len) {
205     if ((log_id >= LOG_ID_MAX) || (log_id < 0)) {
206         return -EINVAL;
207     }
208 
209     LogBufferElement *elem = new LogBufferElement(log_id, realtime,
210                                                   uid, pid, tid, msg, len);
211     if (log_id != LOG_ID_SECURITY) {
212         int prio = ANDROID_LOG_INFO;
213         const char *tag = NULL;
214         if (log_id == LOG_ID_EVENTS) {
215             tag = android::tagToName(elem->getTag());
216         } else {
217             prio = *msg;
218             tag = msg + 1;
219         }
220         if (!__android_log_is_loggable(prio, tag, ANDROID_LOG_VERBOSE)) {
221             // Log traffic received to total
222             pthread_mutex_lock(&mLogElementsLock);
223             stats.add(elem);
224             stats.subtract(elem);
225             pthread_mutex_unlock(&mLogElementsLock);
226             delete elem;
227             return -EACCES;
228         }
229     }
230 
231     pthread_mutex_lock(&mLogElementsLock);
232 
233     // Insert elements in time sorted order if possible
234     //  NB: if end is region locked, place element at end of list
235     LogBufferElementCollection::iterator it = mLogElements.end();
236     LogBufferElementCollection::iterator last = it;
237     while (last != mLogElements.begin()) {
238         --it;
239         if ((*it)->getRealTime() <= realtime) {
240             break;
241         }
242         last = it;
243     }
244 
245     if (last == mLogElements.end()) {
246         mLogElements.push_back(elem);
247     } else {
248         uint64_t end = 1;
249         bool end_set = false;
250         bool end_always = false;
251 
252         LogTimeEntry::lock();
253 
254         LastLogTimes::iterator times = mTimes.begin();
255         while(times != mTimes.end()) {
256             LogTimeEntry *entry = (*times);
257             if (entry->owned_Locked()) {
258                 if (!entry->mNonBlock) {
259                     end_always = true;
260                     break;
261                 }
262                 if (!end_set || (end <= entry->mEnd)) {
263                     end = entry->mEnd;
264                     end_set = true;
265                 }
266             }
267             times++;
268         }
269 
270         if (end_always
271                 || (end_set && (end >= (*last)->getSequence()))) {
272             mLogElements.push_back(elem);
273         } else {
274             mLogElements.insert(last,elem);
275         }
276 
277         LogTimeEntry::unlock();
278     }
279 
280     stats.add(elem);
281     maybePrune(log_id);
282     pthread_mutex_unlock(&mLogElementsLock);
283 
284     return len;
285 }
286 
287 // Prune at most 10% of the log entries or maxPrune, whichever is less.
288 //
289 // mLogElementsLock must be held when this function is called.
maybePrune(log_id_t id)290 void LogBuffer::maybePrune(log_id_t id) {
291     size_t sizes = stats.sizes(id);
292     unsigned long maxSize = log_buffer_size(id);
293     if (sizes > maxSize) {
294         size_t sizeOver = sizes - ((maxSize * 9) / 10);
295         size_t elements = stats.realElements(id);
296         size_t minElements = elements / 100;
297         if (minElements < minPrune) {
298             minElements = minPrune;
299         }
300         unsigned long pruneRows = elements * sizeOver / sizes;
301         if (pruneRows < minElements) {
302             pruneRows = minElements;
303         }
304         if (pruneRows > maxPrune) {
305             pruneRows = maxPrune;
306         }
307         prune(id, pruneRows);
308     }
309 }
310 
erase(LogBufferElementCollection::iterator it,bool coalesce)311 LogBufferElementCollection::iterator LogBuffer::erase(
312         LogBufferElementCollection::iterator it, bool coalesce) {
313     LogBufferElement *element = *it;
314     log_id_t id = element->getLogId();
315 
316     // Remove iterator references in the various lists that will become stale
317     // after the element is erased from the main logging list.
318 
319     {   // start of scope for uid found iterator
320         LogBufferIteratorMap::iterator found =
321             mLastWorstUid[id].find(element->getUid());
322         if ((found != mLastWorstUid[id].end())
323                 && (it == found->second)) {
324             mLastWorstUid[id].erase(found);
325         }
326     }
327 
328     {   // start of scope for pid found iterator
329         // element->getUid() may not be AID_SYSTEM for next-best-watermark.
330         LogBufferPidIteratorMap::iterator found =
331             mLastWorstPidOfSystem[id].find(element->getPid());
332         if ((found != mLastWorstPidOfSystem[id].end())
333                 && (it == found->second)) {
334             mLastWorstPidOfSystem[id].erase(found);
335         }
336     }
337 
338     bool setLast[LOG_ID_MAX];
339     bool doSetLast = false;
340     log_id_for_each(i) {
341         doSetLast |= setLast[i] = mLastSet[i] && (it == mLast[i]);
342     }
343     it = mLogElements.erase(it);
344     if (doSetLast) {
345         log_id_for_each(i) {
346             if (setLast[i]) {
347                 if (it == mLogElements.end()) { // unlikely
348                     mLastSet[i] = false;
349                 } else {
350                     mLast[i] = it;
351                 }
352             }
353         }
354     }
355     if (coalesce) {
356         stats.erase(element);
357     } else {
358         stats.subtract(element);
359     }
360     delete element;
361 
362     return it;
363 }
364 
365 // Define a temporary mechanism to report the last LogBufferElement pointer
366 // for the specified uid, pid and tid. Used below to help merge-sort when
367 // pruning for worst UID.
368 class LogBufferElementKey {
369     const union {
370         struct {
371             uint16_t uid;
372             uint16_t pid;
373             uint16_t tid;
374             uint16_t padding;
375         } __packed;
376         uint64_t value;
377     } __packed;
378 
379 public:
LogBufferElementKey(uid_t uid,pid_t pid,pid_t tid)380     LogBufferElementKey(uid_t uid, pid_t pid, pid_t tid):
381             uid(uid),
382             pid(pid),
383             tid(tid),
384             padding(0) {
385     }
LogBufferElementKey(uint64_t key)386     LogBufferElementKey(uint64_t key):value(key) { }
387 
getKey()388     uint64_t getKey() { return value; }
389 };
390 
391 class LogBufferElementLast {
392 
393     typedef std::unordered_map<uint64_t, LogBufferElement *> LogBufferElementMap;
394     LogBufferElementMap map;
395 
396 public:
397 
coalesce(LogBufferElement * element,unsigned short dropped)398     bool coalesce(LogBufferElement *element, unsigned short dropped) {
399         LogBufferElementKey key(element->getUid(),
400                                 element->getPid(),
401                                 element->getTid());
402         LogBufferElementMap::iterator it = map.find(key.getKey());
403         if (it != map.end()) {
404             LogBufferElement *found = it->second;
405             unsigned short moreDropped = found->getDropped();
406             if ((dropped + moreDropped) > USHRT_MAX) {
407                 map.erase(it);
408             } else {
409                 found->setDropped(dropped + moreDropped);
410                 return true;
411             }
412         }
413         return false;
414     }
415 
add(LogBufferElement * element)416     void add(LogBufferElement *element) {
417         LogBufferElementKey key(element->getUid(),
418                                 element->getPid(),
419                                 element->getTid());
420         map[key.getKey()] = element;
421     }
422 
clear()423     inline void clear() {
424         map.clear();
425     }
426 
clear(LogBufferElement * element)427     void clear(LogBufferElement *element) {
428         uint64_t current = element->getRealTime().nsec()
429                          - (EXPIRE_RATELIMIT * NS_PER_SEC);
430         for(LogBufferElementMap::iterator it = map.begin(); it != map.end();) {
431             LogBufferElement *mapElement = it->second;
432             if ((mapElement->getDropped() >= EXPIRE_THRESHOLD)
433                     && (current > mapElement->getRealTime().nsec())) {
434                 it = map.erase(it);
435             } else {
436                 ++it;
437             }
438         }
439     }
440 
441 };
442 
443 // prune "pruneRows" of type "id" from the buffer.
444 //
445 // This garbage collection task is used to expire log entries. It is called to
446 // remove all logs (clear), all UID logs (unprivileged clear), or every
447 // 256 or 10% of the total logs (whichever is less) to prune the logs.
448 //
449 // First there is a prep phase where we discover the reader region lock that
450 // acts as a backstop to any pruning activity to stop there and go no further.
451 //
452 // There are three major pruning loops that follow. All expire from the oldest
453 // entries. Since there are multiple log buffers, the Android logging facility
454 // will appear to drop entries 'in the middle' when looking at multiple log
455 // sources and buffers. This effect is slightly more prominent when we prune
456 // the worst offender by logging source. Thus the logs slowly loose content
457 // and value as you move back in time. This is preferred since chatty sources
458 // invariably move the logs value down faster as less chatty sources would be
459 // expired in the noise.
460 //
461 // The first loop performs blacklisting and worst offender pruning. Falling
462 // through when there are no notable worst offenders and have not hit the
463 // region lock preventing further worst offender pruning. This loop also looks
464 // after managing the chatty log entries and merging to help provide
465 // statistical basis for blame. The chatty entries are not a notification of
466 // how much logs you may have, but instead represent how much logs you would
467 // have had in a virtual log buffer that is extended to cover all the in-memory
468 // logs without loss. They last much longer than the represented pruned logs
469 // since they get multiplied by the gains in the non-chatty log sources.
470 //
471 // The second loop get complicated because an algorithm of watermarks and
472 // history is maintained to reduce the order and keep processing time
473 // down to a minimum at scale. These algorithms can be costly in the face
474 // of larger log buffers, or severly limited processing time granted to a
475 // background task at lowest priority.
476 //
477 // This second loop does straight-up expiration from the end of the logs
478 // (again, remember for the specified log buffer id) but does some whitelist
479 // preservation. Thus whitelist is a Hail Mary low priority, blacklists and
480 // spam filtration all take priority. This second loop also checks if a region
481 // lock is causing us to buffer too much in the logs to help the reader(s),
482 // and will tell the slowest reader thread to skip log entries, and if
483 // persistent and hits a further threshold, kill the reader thread.
484 //
485 // The third thread is optional, and only gets hit if there was a whitelist
486 // and more needs to be pruned against the backstop of the region lock.
487 //
488 // mLogElementsLock must be held when this function is called.
489 //
prune(log_id_t id,unsigned long pruneRows,uid_t caller_uid)490 bool LogBuffer::prune(log_id_t id, unsigned long pruneRows, uid_t caller_uid) {
491     LogTimeEntry *oldest = NULL;
492     bool busy = false;
493     bool clearAll = pruneRows == ULONG_MAX;
494 
495     LogTimeEntry::lock();
496 
497     // Region locked?
498     LastLogTimes::iterator times = mTimes.begin();
499     while(times != mTimes.end()) {
500         LogTimeEntry *entry = (*times);
501         if (entry->owned_Locked() && entry->isWatching(id)
502                 && (!oldest ||
503                     (oldest->mStart > entry->mStart) ||
504                     ((oldest->mStart == entry->mStart) &&
505                      (entry->mTimeout.tv_sec || entry->mTimeout.tv_nsec)))) {
506             oldest = entry;
507         }
508         times++;
509     }
510 
511     LogBufferElementCollection::iterator it;
512 
513     if (caller_uid != AID_ROOT) {
514         // Only here if clearAll condition (pruneRows == ULONG_MAX)
515         it = mLastSet[id] ? mLast[id] : mLogElements.begin();
516         while (it != mLogElements.end()) {
517             LogBufferElement *element = *it;
518 
519             if ((element->getLogId() != id) || (element->getUid() != caller_uid)) {
520                 ++it;
521                 continue;
522             }
523 
524             if (!mLastSet[id] || ((*mLast[id])->getLogId() != id)) {
525                 mLast[id] = it;
526                 mLastSet[id] = true;
527             }
528 
529             if (oldest && (oldest->mStart <= element->getSequence())) {
530                 busy = true;
531                 if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
532                     oldest->triggerReader_Locked();
533                 } else {
534                     oldest->triggerSkip_Locked(id, pruneRows);
535                 }
536                 break;
537             }
538 
539             it = erase(it);
540             pruneRows--;
541         }
542         LogTimeEntry::unlock();
543         return busy;
544     }
545 
546     // prune by worst offenders; by blacklist, UID, and by PID of system UID
547     bool hasBlacklist = (id != LOG_ID_SECURITY) && mPrune.naughty();
548     while (!clearAll && (pruneRows > 0)) {
549         // recalculate the worst offender on every batched pass
550         uid_t worst = (uid_t) -1;
551         size_t worst_sizes = 0;
552         size_t second_worst_sizes = 0;
553         pid_t worstPid = 0; // POSIX guarantees PID != 0
554 
555         if (worstUidEnabledForLogid(id) && mPrune.worstUidEnabled()) {
556             {   // begin scope for UID sorted list
557                 std::unique_ptr<const UidEntry *[]> sorted = stats.sort(
558                     AID_ROOT, (pid_t)0, 2, id);
559 
560                 if (sorted.get() && sorted[0] && sorted[1]) {
561                     worst_sizes = sorted[0]->getSizes();
562                     // Calculate threshold as 12.5% of available storage
563                     size_t threshold = log_buffer_size(id) / 8;
564                     if ((worst_sizes > threshold)
565                         // Allow time horizon to extend roughly tenfold, assume
566                         // average entry length is 100 characters.
567                             && (worst_sizes > (10 * sorted[0]->getDropped()))) {
568                         worst = sorted[0]->getKey();
569                         second_worst_sizes = sorted[1]->getSizes();
570                         if (second_worst_sizes < threshold) {
571                             second_worst_sizes = threshold;
572                         }
573                     }
574                 }
575             }
576 
577             if ((worst == AID_SYSTEM) && mPrune.worstPidOfSystemEnabled()) {
578                 // begin scope of PID sorted list
579                 std::unique_ptr<const PidEntry *[]> sorted = stats.sort(
580                     worst, (pid_t)0, 2, id, worst);
581                 if (sorted.get() && sorted[0] && sorted[1]) {
582                     worstPid = sorted[0]->getKey();
583                     second_worst_sizes = worst_sizes
584                                        - sorted[0]->getSizes()
585                                        + sorted[1]->getSizes();
586                 }
587             }
588         }
589 
590         // skip if we have neither worst nor naughty filters
591         if ((worst == (uid_t) -1) && !hasBlacklist) {
592             break;
593         }
594 
595         bool kick = false;
596         bool leading = true;
597         it = mLastSet[id] ? mLast[id] : mLogElements.begin();
598         // Perform at least one mandatory garbage collection cycle in following
599         // - clear leading chatty tags
600         // - coalesce chatty tags
601         // - check age-out of preserved logs
602         bool gc = pruneRows <= 1;
603         if (!gc && (worst != (uid_t) -1)) {
604             {   // begin scope for uid worst found iterator
605                 LogBufferIteratorMap::iterator found = mLastWorstUid[id].find(worst);
606                 if ((found != mLastWorstUid[id].end())
607                         && (found->second != mLogElements.end())) {
608                     leading = false;
609                     it = found->second;
610                 }
611             }
612             if (worstPid) {
613                 // begin scope for pid worst found iterator
614                 LogBufferPidIteratorMap::iterator found
615                     = mLastWorstPidOfSystem[id].find(worstPid);
616                 if ((found != mLastWorstPidOfSystem[id].end())
617                         && (found->second != mLogElements.end())) {
618                     leading = false;
619                     it = found->second;
620                 }
621             }
622         }
623         static const timespec too_old = {
624             EXPIRE_HOUR_THRESHOLD * 60 * 60, 0
625         };
626         LogBufferElementCollection::iterator lastt;
627         lastt = mLogElements.end();
628         --lastt;
629         LogBufferElementLast last;
630         while (it != mLogElements.end()) {
631             LogBufferElement *element = *it;
632 
633             if (oldest && (oldest->mStart <= element->getSequence())) {
634                 busy = true;
635                 if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
636                     oldest->triggerReader_Locked();
637                 }
638                 break;
639             }
640 
641             if (element->getLogId() != id) {
642                 ++it;
643                 continue;
644             }
645             // below this point element->getLogId() == id
646 
647             if (leading && (!mLastSet[id] || ((*mLast[id])->getLogId() != id))) {
648                 mLast[id] = it;
649                 mLastSet[id] = true;
650             }
651 
652             unsigned short dropped = element->getDropped();
653 
654             // remove any leading drops
655             if (leading && dropped) {
656                 it = erase(it);
657                 continue;
658             }
659 
660             if (dropped && last.coalesce(element, dropped)) {
661                 it = erase(it, true);
662                 continue;
663             }
664 
665             if (hasBlacklist && mPrune.naughty(element)) {
666                 last.clear(element);
667                 it = erase(it);
668                 if (dropped) {
669                     continue;
670                 }
671 
672                 pruneRows--;
673                 if (pruneRows == 0) {
674                     break;
675                 }
676 
677                 if (element->getUid() == worst) {
678                     kick = true;
679                     if (worst_sizes < second_worst_sizes) {
680                         break;
681                     }
682                     worst_sizes -= element->getMsgLen();
683                 }
684                 continue;
685             }
686 
687             if ((element->getRealTime() < ((*lastt)->getRealTime() - too_old))
688                     || (element->getRealTime() > (*lastt)->getRealTime())) {
689                 break;
690             }
691 
692             if (dropped) {
693                 last.add(element);
694                 if (worstPid
695                         && ((!gc && (element->getPid() == worstPid))
696                             || (mLastWorstPidOfSystem[id].find(element->getPid())
697                                 == mLastWorstPidOfSystem[id].end()))) {
698                     // element->getUid() may not be AID_SYSTEM, next best
699                     // watermark if current one empty.
700                     mLastWorstPidOfSystem[id][element->getPid()] = it;
701                 }
702                 if ((!gc && !worstPid && (element->getUid() == worst))
703                         || (mLastWorstUid[id].find(element->getUid())
704                             == mLastWorstUid[id].end())) {
705                     mLastWorstUid[id][element->getUid()] = it;
706                 }
707                 ++it;
708                 continue;
709             }
710 
711             if ((element->getUid() != worst)
712                     || (worstPid && (element->getPid() != worstPid))) {
713                 leading = false;
714                 last.clear(element);
715                 ++it;
716                 continue;
717             }
718             // key == worst below here
719             // If worstPid set, then element->getPid() == worstPid below here
720 
721             pruneRows--;
722             if (pruneRows == 0) {
723                 break;
724             }
725 
726             kick = true;
727 
728             unsigned short len = element->getMsgLen();
729 
730             // do not create any leading drops
731             if (leading) {
732                 it = erase(it);
733             } else {
734                 stats.drop(element);
735                 element->setDropped(1);
736                 if (last.coalesce(element, 1)) {
737                     it = erase(it, true);
738                 } else {
739                     last.add(element);
740                     if (worstPid && (!gc
741                                 || (mLastWorstPidOfSystem[id].find(worstPid)
742                                     == mLastWorstPidOfSystem[id].end()))) {
743                         // element->getUid() may not be AID_SYSTEM, next best
744                         // watermark if current one empty.
745                         mLastWorstPidOfSystem[id][worstPid] = it;
746                     }
747                     if ((!gc && !worstPid) || (mLastWorstUid[id].find(worst)
748                                 == mLastWorstUid[id].end())) {
749                         mLastWorstUid[id][worst] = it;
750                     }
751                     ++it;
752                 }
753             }
754             if (worst_sizes < second_worst_sizes) {
755                 break;
756             }
757             worst_sizes -= len;
758         }
759         last.clear();
760 
761         if (!kick || !mPrune.worstUidEnabled()) {
762             break; // the following loop will ask bad clients to skip/drop
763         }
764     }
765 
766     bool whitelist = false;
767     bool hasWhitelist = (id != LOG_ID_SECURITY) && mPrune.nice() && !clearAll;
768     it = mLastSet[id] ? mLast[id] : mLogElements.begin();
769     while((pruneRows > 0) && (it != mLogElements.end())) {
770         LogBufferElement *element = *it;
771 
772         if (element->getLogId() != id) {
773             it++;
774             continue;
775         }
776 
777         if (!mLastSet[id] || ((*mLast[id])->getLogId() != id)) {
778             mLast[id] = it;
779             mLastSet[id] = true;
780         }
781 
782         if (oldest && (oldest->mStart <= element->getSequence())) {
783             busy = true;
784             if (whitelist) {
785                 break;
786             }
787 
788             if (stats.sizes(id) > (2 * log_buffer_size(id))) {
789                 // kick a misbehaving log reader client off the island
790                 oldest->release_Locked();
791             } else if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
792                 oldest->triggerReader_Locked();
793             } else {
794                 oldest->triggerSkip_Locked(id, pruneRows);
795             }
796             break;
797         }
798 
799         if (hasWhitelist && !element->getDropped() && mPrune.nice(element)) {
800             // WhiteListed
801             whitelist = true;
802             it++;
803             continue;
804         }
805 
806         it = erase(it);
807         pruneRows--;
808     }
809 
810     // Do not save the whitelist if we are reader range limited
811     if (whitelist && (pruneRows > 0)) {
812         it = mLastSet[id] ? mLast[id] : mLogElements.begin();
813         while((it != mLogElements.end()) && (pruneRows > 0)) {
814             LogBufferElement *element = *it;
815 
816             if (element->getLogId() != id) {
817                 ++it;
818                 continue;
819             }
820 
821             if (!mLastSet[id] || ((*mLast[id])->getLogId() != id)) {
822                 mLast[id] = it;
823                 mLastSet[id] = true;
824             }
825 
826             if (oldest && (oldest->mStart <= element->getSequence())) {
827                 busy = true;
828                 if (stats.sizes(id) > (2 * log_buffer_size(id))) {
829                     // kick a misbehaving log reader client off the island
830                     oldest->release_Locked();
831                 } else if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
832                     oldest->triggerReader_Locked();
833                 } else {
834                     oldest->triggerSkip_Locked(id, pruneRows);
835                 }
836                 break;
837             }
838 
839             it = erase(it);
840             pruneRows--;
841         }
842     }
843 
844     LogTimeEntry::unlock();
845 
846     return (pruneRows > 0) && busy;
847 }
848 
849 // clear all rows of type "id" from the buffer.
clear(log_id_t id,uid_t uid)850 bool LogBuffer::clear(log_id_t id, uid_t uid) {
851     bool busy = true;
852     // If it takes more than 4 tries (seconds) to clear, then kill reader(s)
853     for (int retry = 4;;) {
854         if (retry == 1) { // last pass
855             // Check if it is still busy after the sleep, we say prune
856             // one entry, not another clear run, so we are looking for
857             // the quick side effect of the return value to tell us if
858             // we have a _blocked_ reader.
859             pthread_mutex_lock(&mLogElementsLock);
860             busy = prune(id, 1, uid);
861             pthread_mutex_unlock(&mLogElementsLock);
862             // It is still busy, blocked reader(s), lets kill them all!
863             // otherwise, lets be a good citizen and preserve the slow
864             // readers and let the clear run (below) deal with determining
865             // if we are still blocked and return an error code to caller.
866             if (busy) {
867                 LogTimeEntry::lock();
868                 LastLogTimes::iterator times = mTimes.begin();
869                 while (times != mTimes.end()) {
870                     LogTimeEntry *entry = (*times);
871                     // Killer punch
872                     if (entry->owned_Locked() && entry->isWatching(id)) {
873                         entry->release_Locked();
874                     }
875                     times++;
876                 }
877                 LogTimeEntry::unlock();
878             }
879         }
880         pthread_mutex_lock(&mLogElementsLock);
881         busy = prune(id, ULONG_MAX, uid);
882         pthread_mutex_unlock(&mLogElementsLock);
883         if (!busy || !--retry) {
884             break;
885         }
886         sleep (1); // Let reader(s) catch up after notification
887     }
888     return busy;
889 }
890 
891 // get the used space associated with "id".
getSizeUsed(log_id_t id)892 unsigned long LogBuffer::getSizeUsed(log_id_t id) {
893     pthread_mutex_lock(&mLogElementsLock);
894     size_t retval = stats.sizes(id);
895     pthread_mutex_unlock(&mLogElementsLock);
896     return retval;
897 }
898 
899 // set the total space allocated to "id"
setSize(log_id_t id,unsigned long size)900 int LogBuffer::setSize(log_id_t id, unsigned long size) {
901     // Reasonable limits ...
902     if (!valid_size(size)) {
903         return -1;
904     }
905     pthread_mutex_lock(&mLogElementsLock);
906     log_buffer_size(id) = size;
907     pthread_mutex_unlock(&mLogElementsLock);
908     return 0;
909 }
910 
911 // get the total space allocated to "id"
getSize(log_id_t id)912 unsigned long LogBuffer::getSize(log_id_t id) {
913     pthread_mutex_lock(&mLogElementsLock);
914     size_t retval = log_buffer_size(id);
915     pthread_mutex_unlock(&mLogElementsLock);
916     return retval;
917 }
918 
flushTo(SocketClient * reader,const uint64_t start,bool privileged,bool security,int (* filter)(const LogBufferElement * element,void * arg),void * arg)919 uint64_t LogBuffer::flushTo(
920         SocketClient *reader, const uint64_t start,
921         bool privileged, bool security,
922         int (*filter)(const LogBufferElement *element, void *arg), void *arg) {
923     LogBufferElementCollection::iterator it;
924     uint64_t max = start;
925     uid_t uid = reader->getUid();
926 
927     pthread_mutex_lock(&mLogElementsLock);
928 
929     if (start <= 1) {
930         // client wants to start from the beginning
931         it = mLogElements.begin();
932     } else {
933         // Client wants to start from some specified time. Chances are
934         // we are better off starting from the end of the time sorted list.
935         for (it = mLogElements.end(); it != mLogElements.begin(); /* do nothing */) {
936             --it;
937             LogBufferElement *element = *it;
938             if (element->getSequence() <= start) {
939                 it++;
940                 break;
941             }
942         }
943     }
944 
945     for (; it != mLogElements.end(); ++it) {
946         LogBufferElement *element = *it;
947 
948         if (!privileged && (element->getUid() != uid)) {
949             continue;
950         }
951 
952         if (!security && (element->getLogId() == LOG_ID_SECURITY)) {
953             continue;
954         }
955 
956         if (element->getSequence() <= start) {
957             continue;
958         }
959 
960         // NB: calling out to another object with mLogElementsLock held (safe)
961         if (filter) {
962             int ret = (*filter)(element, arg);
963             if (ret == false) {
964                 continue;
965             }
966             if (ret != true) {
967                 break;
968             }
969         }
970 
971         pthread_mutex_unlock(&mLogElementsLock);
972 
973         // range locking in LastLogTimes looks after us
974         max = element->flushTo(reader, this, privileged);
975 
976         if (max == element->FLUSH_ERROR) {
977             return max;
978         }
979 
980         pthread_mutex_lock(&mLogElementsLock);
981     }
982     pthread_mutex_unlock(&mLogElementsLock);
983 
984     return max;
985 }
986 
formatStatistics(uid_t uid,pid_t pid,unsigned int logMask)987 std::string LogBuffer::formatStatistics(uid_t uid, pid_t pid,
988                                         unsigned int logMask) {
989     pthread_mutex_lock(&mLogElementsLock);
990 
991     std::string ret = stats.format(uid, pid, logMask);
992 
993     pthread_mutex_unlock(&mLogElementsLock);
994 
995     return ret;
996 }
997