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     {   // start of scope for uid found iterator
317         LogBufferIteratorMap::iterator found =
318             mLastWorstUid[id].find(element->getUid());
319         if ((found != mLastWorstUid[id].end())
320                 && (it == found->second)) {
321             mLastWorstUid[id].erase(found);
322         }
323     }
324 
325     if (element->getUid() == AID_SYSTEM) {
326         // start of scope for pid found iterator
327         LogBufferPidIteratorMap::iterator found =
328             mLastWorstPidOfSystem[id].find(element->getPid());
329         if ((found != mLastWorstPidOfSystem[id].end())
330                 && (it == found->second)) {
331             mLastWorstPidOfSystem[id].erase(found);
332         }
333     }
334 
335     bool setLast[LOG_ID_MAX];
336     bool doSetLast = false;
337     log_id_for_each(i) {
338         doSetLast |= setLast[i] = mLastSet[i] && (it == mLast[i]);
339     }
340     it = mLogElements.erase(it);
341     if (doSetLast) {
342         log_id_for_each(i) {
343             if (setLast[i]) {
344                 if (it == mLogElements.end()) { // unlikely
345                     mLastSet[i] = false;
346                 } else {
347                     mLast[i] = it;
348                 }
349             }
350         }
351     }
352     if (coalesce) {
353         stats.erase(element);
354     } else {
355         stats.subtract(element);
356     }
357     delete element;
358 
359     return it;
360 }
361 
362 // Define a temporary mechanism to report the last LogBufferElement pointer
363 // for the specified uid, pid and tid. Used below to help merge-sort when
364 // pruning for worst UID.
365 class LogBufferElementKey {
366     const union {
367         struct {
368             uint16_t uid;
369             uint16_t pid;
370             uint16_t tid;
371             uint16_t padding;
372         } __packed;
373         uint64_t value;
374     } __packed;
375 
376 public:
LogBufferElementKey(uid_t uid,pid_t pid,pid_t tid)377     LogBufferElementKey(uid_t uid, pid_t pid, pid_t tid):
378             uid(uid),
379             pid(pid),
380             tid(tid),
381             padding(0) {
382     }
LogBufferElementKey(uint64_t key)383     LogBufferElementKey(uint64_t key):value(key) { }
384 
getKey()385     uint64_t getKey() { return value; }
386 };
387 
388 class LogBufferElementLast {
389 
390     typedef std::unordered_map<uint64_t, LogBufferElement *> LogBufferElementMap;
391     LogBufferElementMap map;
392 
393 public:
394 
coalesce(LogBufferElement * element,unsigned short dropped)395     bool coalesce(LogBufferElement *element, unsigned short dropped) {
396         LogBufferElementKey key(element->getUid(),
397                                 element->getPid(),
398                                 element->getTid());
399         LogBufferElementMap::iterator it = map.find(key.getKey());
400         if (it != map.end()) {
401             LogBufferElement *found = it->second;
402             unsigned short moreDropped = found->getDropped();
403             if ((dropped + moreDropped) > USHRT_MAX) {
404                 map.erase(it);
405             } else {
406                 found->setDropped(dropped + moreDropped);
407                 return true;
408             }
409         }
410         return false;
411     }
412 
add(LogBufferElement * element)413     void add(LogBufferElement *element) {
414         LogBufferElementKey key(element->getUid(),
415                                 element->getPid(),
416                                 element->getTid());
417         map[key.getKey()] = element;
418     }
419 
clear()420     inline void clear() {
421         map.clear();
422     }
423 
clear(LogBufferElement * element)424     void clear(LogBufferElement *element) {
425         uint64_t current = element->getRealTime().nsec()
426                          - (EXPIRE_RATELIMIT * NS_PER_SEC);
427         for(LogBufferElementMap::iterator it = map.begin(); it != map.end();) {
428             LogBufferElement *mapElement = it->second;
429             if ((mapElement->getDropped() >= EXPIRE_THRESHOLD)
430                     && (current > mapElement->getRealTime().nsec())) {
431                 it = map.erase(it);
432             } else {
433                 ++it;
434             }
435         }
436     }
437 
438 };
439 
440 // prune "pruneRows" of type "id" from the buffer.
441 //
442 // This garbage collection task is used to expire log entries. It is called to
443 // remove all logs (clear), all UID logs (unprivileged clear), or every
444 // 256 or 10% of the total logs (whichever is less) to prune the logs.
445 //
446 // First there is a prep phase where we discover the reader region lock that
447 // acts as a backstop to any pruning activity to stop there and go no further.
448 //
449 // There are three major pruning loops that follow. All expire from the oldest
450 // entries. Since there are multiple log buffers, the Android logging facility
451 // will appear to drop entries 'in the middle' when looking at multiple log
452 // sources and buffers. This effect is slightly more prominent when we prune
453 // the worst offender by logging source. Thus the logs slowly loose content
454 // and value as you move back in time. This is preferred since chatty sources
455 // invariably move the logs value down faster as less chatty sources would be
456 // expired in the noise.
457 //
458 // The first loop performs blacklisting and worst offender pruning. Falling
459 // through when there are no notable worst offenders and have not hit the
460 // region lock preventing further worst offender pruning. This loop also looks
461 // after managing the chatty log entries and merging to help provide
462 // statistical basis for blame. The chatty entries are not a notification of
463 // how much logs you may have, but instead represent how much logs you would
464 // have had in a virtual log buffer that is extended to cover all the in-memory
465 // logs without loss. They last much longer than the represented pruned logs
466 // since they get multiplied by the gains in the non-chatty log sources.
467 //
468 // The second loop get complicated because an algorithm of watermarks and
469 // history is maintained to reduce the order and keep processing time
470 // down to a minimum at scale. These algorithms can be costly in the face
471 // of larger log buffers, or severly limited processing time granted to a
472 // background task at lowest priority.
473 //
474 // This second loop does straight-up expiration from the end of the logs
475 // (again, remember for the specified log buffer id) but does some whitelist
476 // preservation. Thus whitelist is a Hail Mary low priority, blacklists and
477 // spam filtration all take priority. This second loop also checks if a region
478 // lock is causing us to buffer too much in the logs to help the reader(s),
479 // and will tell the slowest reader thread to skip log entries, and if
480 // persistent and hits a further threshold, kill the reader thread.
481 //
482 // The third thread is optional, and only gets hit if there was a whitelist
483 // and more needs to be pruned against the backstop of the region lock.
484 //
485 // mLogElementsLock must be held when this function is called.
486 //
prune(log_id_t id,unsigned long pruneRows,uid_t caller_uid)487 bool LogBuffer::prune(log_id_t id, unsigned long pruneRows, uid_t caller_uid) {
488     LogTimeEntry *oldest = NULL;
489     bool busy = false;
490     bool clearAll = pruneRows == ULONG_MAX;
491 
492     LogTimeEntry::lock();
493 
494     // Region locked?
495     LastLogTimes::iterator times = mTimes.begin();
496     while(times != mTimes.end()) {
497         LogTimeEntry *entry = (*times);
498         if (entry->owned_Locked() && entry->isWatching(id)
499                 && (!oldest ||
500                     (oldest->mStart > entry->mStart) ||
501                     ((oldest->mStart == entry->mStart) &&
502                      (entry->mTimeout.tv_sec || entry->mTimeout.tv_nsec)))) {
503             oldest = entry;
504         }
505         times++;
506     }
507 
508     LogBufferElementCollection::iterator it;
509 
510     if (caller_uid != AID_ROOT) {
511         // Only here if clearAll condition (pruneRows == ULONG_MAX)
512         it = mLastSet[id] ? mLast[id] : mLogElements.begin();
513         while (it != mLogElements.end()) {
514             LogBufferElement *element = *it;
515 
516             if ((element->getLogId() != id) || (element->getUid() != caller_uid)) {
517                 ++it;
518                 continue;
519             }
520 
521             if (!mLastSet[id] || ((*mLast[id])->getLogId() != id)) {
522                 mLast[id] = it;
523                 mLastSet[id] = true;
524             }
525 
526             if (oldest && (oldest->mStart <= element->getSequence())) {
527                 busy = true;
528                 if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
529                     oldest->triggerReader_Locked();
530                 } else {
531                     oldest->triggerSkip_Locked(id, pruneRows);
532                 }
533                 break;
534             }
535 
536             it = erase(it);
537             pruneRows--;
538         }
539         LogTimeEntry::unlock();
540         return busy;
541     }
542 
543     // prune by worst offenders; by blacklist, UID, and by PID of system UID
544     bool hasBlacklist = (id != LOG_ID_SECURITY) && mPrune.naughty();
545     while (!clearAll && (pruneRows > 0)) {
546         // recalculate the worst offender on every batched pass
547         uid_t worst = (uid_t) -1;
548         size_t worst_sizes = 0;
549         size_t second_worst_sizes = 0;
550         pid_t worstPid = 0; // POSIX guarantees PID != 0
551 
552         if (worstUidEnabledForLogid(id) && mPrune.worstUidEnabled()) {
553             {   // begin scope for UID sorted list
554                 std::unique_ptr<const UidEntry *[]> sorted = stats.sort(
555                     AID_ROOT, (pid_t)0, 2, id);
556 
557                 if (sorted.get() && sorted[0] && sorted[1]) {
558                     worst_sizes = sorted[0]->getSizes();
559                     // Calculate threshold as 12.5% of available storage
560                     size_t threshold = log_buffer_size(id) / 8;
561                     if ((worst_sizes > threshold)
562                         // Allow time horizon to extend roughly tenfold, assume
563                         // average entry length is 100 characters.
564                             && (worst_sizes > (10 * sorted[0]->getDropped()))) {
565                         worst = sorted[0]->getKey();
566                         second_worst_sizes = sorted[1]->getSizes();
567                         if (second_worst_sizes < threshold) {
568                             second_worst_sizes = threshold;
569                         }
570                     }
571                 }
572             }
573 
574             if ((worst == AID_SYSTEM) && mPrune.worstPidOfSystemEnabled()) {
575                 // begin scope of PID sorted list
576                 std::unique_ptr<const PidEntry *[]> sorted = stats.sort(
577                     worst, (pid_t)0, 2, id, worst);
578                 if (sorted.get() && sorted[0] && sorted[1]) {
579                     worstPid = sorted[0]->getKey();
580                     second_worst_sizes = worst_sizes
581                                        - sorted[0]->getSizes()
582                                        + sorted[1]->getSizes();
583                 }
584             }
585         }
586 
587         // skip if we have neither worst nor naughty filters
588         if ((worst == (uid_t) -1) && !hasBlacklist) {
589             break;
590         }
591 
592         bool kick = false;
593         bool leading = true;
594         it = mLastSet[id] ? mLast[id] : mLogElements.begin();
595         // Perform at least one mandatory garbage collection cycle in following
596         // - clear leading chatty tags
597         // - coalesce chatty tags
598         // - check age-out of preserved logs
599         bool gc = pruneRows <= 1;
600         if (!gc && (worst != (uid_t) -1)) {
601             {   // begin scope for uid worst found iterator
602                 LogBufferIteratorMap::iterator found = mLastWorstUid[id].find(worst);
603                 if ((found != mLastWorstUid[id].end())
604                         && (found->second != mLogElements.end())) {
605                     leading = false;
606                     it = found->second;
607                 }
608             }
609             if (worstPid) {
610                 // begin scope for pid worst found iterator
611                 LogBufferPidIteratorMap::iterator found
612                     = mLastWorstPidOfSystem[id].find(worstPid);
613                 if ((found != mLastWorstPidOfSystem[id].end())
614                         && (found->second != mLogElements.end())) {
615                     leading = false;
616                     it = found->second;
617                 }
618             }
619         }
620         static const timespec too_old = {
621             EXPIRE_HOUR_THRESHOLD * 60 * 60, 0
622         };
623         LogBufferElementCollection::iterator lastt;
624         lastt = mLogElements.end();
625         --lastt;
626         LogBufferElementLast last;
627         while (it != mLogElements.end()) {
628             LogBufferElement *element = *it;
629 
630             if (oldest && (oldest->mStart <= element->getSequence())) {
631                 busy = true;
632                 if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
633                     oldest->triggerReader_Locked();
634                 }
635                 break;
636             }
637 
638             if (element->getLogId() != id) {
639                 ++it;
640                 continue;
641             }
642 
643             if (leading && (!mLastSet[id] || ((*mLast[id])->getLogId() != id))) {
644                 mLast[id] = it;
645                 mLastSet[id] = true;
646             }
647 
648             unsigned short dropped = element->getDropped();
649 
650             // remove any leading drops
651             if (leading && dropped) {
652                 it = erase(it);
653                 continue;
654             }
655 
656             if (dropped && last.coalesce(element, dropped)) {
657                 it = erase(it, true);
658                 continue;
659             }
660 
661             if (hasBlacklist && mPrune.naughty(element)) {
662                 last.clear(element);
663                 it = erase(it);
664                 if (dropped) {
665                     continue;
666                 }
667 
668                 pruneRows--;
669                 if (pruneRows == 0) {
670                     break;
671                 }
672 
673                 if (element->getUid() == worst) {
674                     kick = true;
675                     if (worst_sizes < second_worst_sizes) {
676                         break;
677                     }
678                     worst_sizes -= element->getMsgLen();
679                 }
680                 continue;
681             }
682 
683             if ((element->getRealTime() < ((*lastt)->getRealTime() - too_old))
684                     || (element->getRealTime() > (*lastt)->getRealTime())) {
685                 break;
686             }
687 
688             if (dropped) {
689                 last.add(element);
690                 if (worstPid
691                         && ((!gc && (element->getPid() == worstPid))
692                             || (mLastWorstPidOfSystem[id].find(element->getPid())
693                                 == mLastWorstPidOfSystem[id].end()))) {
694                     mLastWorstPidOfSystem[id][element->getUid()] = it;
695                 }
696                 if ((!gc && !worstPid && (element->getUid() == worst))
697                         || (mLastWorstUid[id].find(element->getUid())
698                             == mLastWorstUid[id].end())) {
699                     mLastWorstUid[id][element->getUid()] = it;
700                 }
701                 ++it;
702                 continue;
703             }
704 
705             if ((element->getUid() != worst)
706                     || (worstPid && (element->getPid() != worstPid))) {
707                 leading = false;
708                 last.clear(element);
709                 ++it;
710                 continue;
711             }
712 
713             pruneRows--;
714             if (pruneRows == 0) {
715                 break;
716             }
717 
718             kick = true;
719 
720             unsigned short len = element->getMsgLen();
721 
722             // do not create any leading drops
723             if (leading) {
724                 it = erase(it);
725             } else {
726                 stats.drop(element);
727                 element->setDropped(1);
728                 if (last.coalesce(element, 1)) {
729                     it = erase(it, true);
730                 } else {
731                     last.add(element);
732                     if (worstPid && (!gc
733                                 || (mLastWorstPidOfSystem[id].find(worstPid)
734                                     == mLastWorstPidOfSystem[id].end()))) {
735                         mLastWorstPidOfSystem[id][worstPid] = it;
736                     }
737                     if ((!gc && !worstPid) || (mLastWorstUid[id].find(worst)
738                                 == mLastWorstUid[id].end())) {
739                         mLastWorstUid[id][worst] = it;
740                     }
741                     ++it;
742                 }
743             }
744             if (worst_sizes < second_worst_sizes) {
745                 break;
746             }
747             worst_sizes -= len;
748         }
749         last.clear();
750 
751         if (!kick || !mPrune.worstUidEnabled()) {
752             break; // the following loop will ask bad clients to skip/drop
753         }
754     }
755 
756     bool whitelist = false;
757     bool hasWhitelist = (id != LOG_ID_SECURITY) && mPrune.nice() && !clearAll;
758     it = mLastSet[id] ? mLast[id] : mLogElements.begin();
759     while((pruneRows > 0) && (it != mLogElements.end())) {
760         LogBufferElement *element = *it;
761 
762         if (element->getLogId() != id) {
763             it++;
764             continue;
765         }
766 
767         if (!mLastSet[id] || ((*mLast[id])->getLogId() != id)) {
768             mLast[id] = it;
769             mLastSet[id] = true;
770         }
771 
772         if (oldest && (oldest->mStart <= element->getSequence())) {
773             busy = true;
774             if (whitelist) {
775                 break;
776             }
777 
778             if (stats.sizes(id) > (2 * log_buffer_size(id))) {
779                 // kick a misbehaving log reader client off the island
780                 oldest->release_Locked();
781             } else if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
782                 oldest->triggerReader_Locked();
783             } else {
784                 oldest->triggerSkip_Locked(id, pruneRows);
785             }
786             break;
787         }
788 
789         if (hasWhitelist && !element->getDropped() && mPrune.nice(element)) {
790             // WhiteListed
791             whitelist = true;
792             it++;
793             continue;
794         }
795 
796         it = erase(it);
797         pruneRows--;
798     }
799 
800     // Do not save the whitelist if we are reader range limited
801     if (whitelist && (pruneRows > 0)) {
802         it = mLastSet[id] ? mLast[id] : mLogElements.begin();
803         while((it != mLogElements.end()) && (pruneRows > 0)) {
804             LogBufferElement *element = *it;
805 
806             if (element->getLogId() != id) {
807                 ++it;
808                 continue;
809             }
810 
811             if (!mLastSet[id] || ((*mLast[id])->getLogId() != id)) {
812                 mLast[id] = it;
813                 mLastSet[id] = true;
814             }
815 
816             if (oldest && (oldest->mStart <= element->getSequence())) {
817                 busy = true;
818                 if (stats.sizes(id) > (2 * log_buffer_size(id))) {
819                     // kick a misbehaving log reader client off the island
820                     oldest->release_Locked();
821                 } else if (oldest->mTimeout.tv_sec || oldest->mTimeout.tv_nsec) {
822                     oldest->triggerReader_Locked();
823                 } else {
824                     oldest->triggerSkip_Locked(id, pruneRows);
825                 }
826                 break;
827             }
828 
829             it = erase(it);
830             pruneRows--;
831         }
832     }
833 
834     LogTimeEntry::unlock();
835 
836     return (pruneRows > 0) && busy;
837 }
838 
839 // clear all rows of type "id" from the buffer.
clear(log_id_t id,uid_t uid)840 bool LogBuffer::clear(log_id_t id, uid_t uid) {
841     bool busy = true;
842     // If it takes more than 4 tries (seconds) to clear, then kill reader(s)
843     for (int retry = 4;;) {
844         if (retry == 1) { // last pass
845             // Check if it is still busy after the sleep, we say prune
846             // one entry, not another clear run, so we are looking for
847             // the quick side effect of the return value to tell us if
848             // we have a _blocked_ reader.
849             pthread_mutex_lock(&mLogElementsLock);
850             busy = prune(id, 1, uid);
851             pthread_mutex_unlock(&mLogElementsLock);
852             // It is still busy, blocked reader(s), lets kill them all!
853             // otherwise, lets be a good citizen and preserve the slow
854             // readers and let the clear run (below) deal with determining
855             // if we are still blocked and return an error code to caller.
856             if (busy) {
857                 LogTimeEntry::lock();
858                 LastLogTimes::iterator times = mTimes.begin();
859                 while (times != mTimes.end()) {
860                     LogTimeEntry *entry = (*times);
861                     // Killer punch
862                     if (entry->owned_Locked() && entry->isWatching(id)) {
863                         entry->release_Locked();
864                     }
865                     times++;
866                 }
867                 LogTimeEntry::unlock();
868             }
869         }
870         pthread_mutex_lock(&mLogElementsLock);
871         busy = prune(id, ULONG_MAX, uid);
872         pthread_mutex_unlock(&mLogElementsLock);
873         if (!busy || !--retry) {
874             break;
875         }
876         sleep (1); // Let reader(s) catch up after notification
877     }
878     return busy;
879 }
880 
881 // get the used space associated with "id".
getSizeUsed(log_id_t id)882 unsigned long LogBuffer::getSizeUsed(log_id_t id) {
883     pthread_mutex_lock(&mLogElementsLock);
884     size_t retval = stats.sizes(id);
885     pthread_mutex_unlock(&mLogElementsLock);
886     return retval;
887 }
888 
889 // set the total space allocated to "id"
setSize(log_id_t id,unsigned long size)890 int LogBuffer::setSize(log_id_t id, unsigned long size) {
891     // Reasonable limits ...
892     if (!valid_size(size)) {
893         return -1;
894     }
895     pthread_mutex_lock(&mLogElementsLock);
896     log_buffer_size(id) = size;
897     pthread_mutex_unlock(&mLogElementsLock);
898     return 0;
899 }
900 
901 // get the total space allocated to "id"
getSize(log_id_t id)902 unsigned long LogBuffer::getSize(log_id_t id) {
903     pthread_mutex_lock(&mLogElementsLock);
904     size_t retval = log_buffer_size(id);
905     pthread_mutex_unlock(&mLogElementsLock);
906     return retval;
907 }
908 
flushTo(SocketClient * reader,const uint64_t start,bool privileged,bool security,int (* filter)(const LogBufferElement * element,void * arg),void * arg)909 uint64_t LogBuffer::flushTo(
910         SocketClient *reader, const uint64_t start,
911         bool privileged, bool security,
912         int (*filter)(const LogBufferElement *element, void *arg), void *arg) {
913     LogBufferElementCollection::iterator it;
914     uint64_t max = start;
915     uid_t uid = reader->getUid();
916 
917     pthread_mutex_lock(&mLogElementsLock);
918 
919     if (start <= 1) {
920         // client wants to start from the beginning
921         it = mLogElements.begin();
922     } else {
923         // Client wants to start from some specified time. Chances are
924         // we are better off starting from the end of the time sorted list.
925         for (it = mLogElements.end(); it != mLogElements.begin(); /* do nothing */) {
926             --it;
927             LogBufferElement *element = *it;
928             if (element->getSequence() <= start) {
929                 it++;
930                 break;
931             }
932         }
933     }
934 
935     for (; it != mLogElements.end(); ++it) {
936         LogBufferElement *element = *it;
937 
938         if (!privileged && (element->getUid() != uid)) {
939             continue;
940         }
941 
942         if (!security && (element->getLogId() == LOG_ID_SECURITY)) {
943             continue;
944         }
945 
946         if (element->getSequence() <= start) {
947             continue;
948         }
949 
950         // NB: calling out to another object with mLogElementsLock held (safe)
951         if (filter) {
952             int ret = (*filter)(element, arg);
953             if (ret == false) {
954                 continue;
955             }
956             if (ret != true) {
957                 break;
958             }
959         }
960 
961         pthread_mutex_unlock(&mLogElementsLock);
962 
963         // range locking in LastLogTimes looks after us
964         max = element->flushTo(reader, this, privileged);
965 
966         if (max == element->FLUSH_ERROR) {
967             return max;
968         }
969 
970         pthread_mutex_lock(&mLogElementsLock);
971     }
972     pthread_mutex_unlock(&mLogElementsLock);
973 
974     return max;
975 }
976 
formatStatistics(uid_t uid,pid_t pid,unsigned int logMask)977 std::string LogBuffer::formatStatistics(uid_t uid, pid_t pid,
978                                         unsigned int logMask) {
979     pthread_mutex_lock(&mLogElementsLock);
980 
981     std::string ret = stats.format(uid, pid, logMask);
982 
983     pthread_mutex_unlock(&mLogElementsLock);
984 
985     return ret;
986 }
987