/src/postgres/src/backend/postmaster/syslogger.c
Line | Count | Source |
1 | | /*------------------------------------------------------------------------- |
2 | | * |
3 | | * syslogger.c |
4 | | * |
5 | | * The system logger (syslogger) appeared in Postgres 8.0. It catches all |
6 | | * stderr output from the postmaster, backends, and other subprocesses |
7 | | * by redirecting to a pipe, and writes it to a set of logfiles. |
8 | | * It's possible to have size and age limits for the logfile configured |
9 | | * in postgresql.conf. If these limits are reached or passed, the |
10 | | * current logfile is closed and a new one is created (rotated). |
11 | | * The logfiles are stored in a subdirectory (configurable in |
12 | | * postgresql.conf), using a user-selectable naming scheme. |
13 | | * |
14 | | * Author: Andreas Pflug <pgadmin@pse-consulting.de> |
15 | | * |
16 | | * Copyright (c) 2004-2026, PostgreSQL Global Development Group |
17 | | * |
18 | | * |
19 | | * IDENTIFICATION |
20 | | * src/backend/postmaster/syslogger.c |
21 | | * |
22 | | *------------------------------------------------------------------------- |
23 | | */ |
24 | | #include "postgres.h" |
25 | | |
26 | | #include <fcntl.h> |
27 | | #include <limits.h> |
28 | | #include <signal.h> |
29 | | #include <time.h> |
30 | | #include <unistd.h> |
31 | | #include <sys/stat.h> |
32 | | #include <sys/time.h> |
33 | | |
34 | | #include "common/file_perm.h" |
35 | | #include "lib/stringinfo.h" |
36 | | #include "libpq/pqsignal.h" |
37 | | #include "miscadmin.h" |
38 | | #include "nodes/pg_list.h" |
39 | | #include "pgstat.h" |
40 | | #include "pgtime.h" |
41 | | #include "port/pg_bitutils.h" |
42 | | #include "postmaster/interrupt.h" |
43 | | #include "postmaster/postmaster.h" |
44 | | #include "postmaster/syslogger.h" |
45 | | #include "storage/dsm.h" |
46 | | #include "storage/fd.h" |
47 | | #include "storage/ipc.h" |
48 | | #include "storage/latch.h" |
49 | | #include "storage/pg_shmem.h" |
50 | | #include "tcop/tcopprot.h" |
51 | | #include "utils/guc.h" |
52 | | #include "utils/memutils.h" |
53 | | #include "utils/ps_status.h" |
54 | | #include "utils/wait_event.h" |
55 | | |
56 | | /* |
57 | | * We read() into a temp buffer twice as big as a chunk, so that any fragment |
58 | | * left after processing can be moved down to the front and we'll still have |
59 | | * room to read a full chunk. |
60 | | */ |
61 | | #define READ_BUF_SIZE (2 * PIPE_CHUNK_SIZE) |
62 | | |
63 | | /* Log rotation signal file path, relative to $PGDATA */ |
64 | 0 | #define LOGROTATE_SIGNAL_FILE "logrotate" |
65 | | |
66 | | |
67 | | /* |
68 | | * GUC parameters. Logging_collector cannot be changed after postmaster |
69 | | * start, but the rest can change at SIGHUP. |
70 | | */ |
71 | | bool Logging_collector = false; |
72 | | int Log_RotationAge = HOURS_PER_DAY * MINS_PER_HOUR; |
73 | | int Log_RotationSize = 10 * 1024; |
74 | | char *Log_directory = NULL; |
75 | | char *Log_filename = NULL; |
76 | | bool Log_truncate_on_rotation = false; |
77 | | int Log_file_mode = S_IRUSR | S_IWUSR; |
78 | | |
79 | | /* |
80 | | * Indicates to be running in the syslogger process, and that the logging |
81 | | * file descriptor(s) have been set up. |
82 | | */ |
83 | | bool syslogger_setup_done = false; |
84 | | |
85 | | /* |
86 | | * Private state |
87 | | */ |
88 | | static pg_time_t next_rotation_time; |
89 | | static bool pipe_eof_seen = false; |
90 | | static bool rotation_disabled = false; |
91 | | static FILE *syslogFile = NULL; |
92 | | static FILE *csvlogFile = NULL; |
93 | | static FILE *jsonlogFile = NULL; |
94 | | NON_EXEC_STATIC pg_time_t first_syslogger_file_time = 0; |
95 | | static char *last_sys_file_name = NULL; |
96 | | static char *last_csv_file_name = NULL; |
97 | | static char *last_json_file_name = NULL; |
98 | | |
99 | | /* |
100 | | * Buffers for saving partial messages from different backends. |
101 | | * |
102 | | * Keep NBUFFER_LISTS lists of these, with the entry for a given source pid |
103 | | * being in the list numbered (pid % NBUFFER_LISTS), so as to cut down on |
104 | | * the number of entries we have to examine for any one incoming message. |
105 | | * There must never be more than one entry for the same source pid. |
106 | | * |
107 | | * An inactive buffer is not removed from its list, just held for re-use. |
108 | | * An inactive buffer has pid == 0 and undefined contents of data. |
109 | | */ |
110 | | typedef struct |
111 | | { |
112 | | int32 pid; /* PID of source process */ |
113 | | StringInfoData data; /* accumulated data, as a StringInfo */ |
114 | | } save_buffer; |
115 | | |
116 | 0 | #define NBUFFER_LISTS 256 |
117 | | static List *buffer_lists[NBUFFER_LISTS]; |
118 | | |
119 | | /* These must be exported for EXEC_BACKEND case ... annoying */ |
120 | | #ifndef WIN32 |
121 | | int syslogPipe[2] = {-1, -1}; |
122 | | #else |
123 | | HANDLE syslogPipe[2] = {0, 0}; |
124 | | #endif |
125 | | |
126 | | #ifdef WIN32 |
127 | | static HANDLE threadHandle = 0; |
128 | | static CRITICAL_SECTION sysloggerSection; |
129 | | #endif |
130 | | |
131 | | /* |
132 | | * Flags set by interrupt handlers for later service in the main loop. |
133 | | */ |
134 | | static volatile sig_atomic_t rotation_requested = false; |
135 | | |
136 | | |
137 | | /* Local subroutines */ |
138 | | #ifdef EXEC_BACKEND |
139 | | static int syslogger_fdget(FILE *file); |
140 | | static FILE *syslogger_fdopen(int fd); |
141 | | #endif |
142 | | static void process_pipe_input(char *logbuffer, int *bytes_in_logbuffer); |
143 | | static void flush_pipe_input(char *logbuffer, int *bytes_in_logbuffer); |
144 | | static FILE *logfile_open(const char *filename, const char *mode, |
145 | | bool allow_errors); |
146 | | |
147 | | #ifdef WIN32 |
148 | | static unsigned int __stdcall pipeThread(void *arg); |
149 | | #endif |
150 | | static void logfile_rotate(bool time_based_rotation, int size_rotation_for); |
151 | | static bool logfile_rotate_dest(bool time_based_rotation, |
152 | | int size_rotation_for, pg_time_t fntime, |
153 | | int target_dest, char **last_file_name, |
154 | | FILE **logFile); |
155 | | static char *logfile_getname(pg_time_t timestamp, const char *suffix); |
156 | | static void set_next_rotation_time(void); |
157 | | static void sigUsr1Handler(SIGNAL_ARGS); |
158 | | static void update_metainfo_datafile(void); |
159 | | |
160 | | typedef struct |
161 | | { |
162 | | int syslogFile; |
163 | | int csvlogFile; |
164 | | int jsonlogFile; |
165 | | } SysloggerStartupData; |
166 | | |
167 | | /* |
168 | | * Main entry point for syslogger process |
169 | | * argc/argv parameters are valid only in EXEC_BACKEND case. |
170 | | */ |
171 | | void |
172 | | SysLoggerMain(const void *startup_data, size_t startup_data_len) |
173 | | { |
174 | | #ifndef WIN32 |
175 | | char logbuffer[READ_BUF_SIZE]; |
176 | | int bytes_in_logbuffer = 0; |
177 | | #endif |
178 | | char *currentLogDir; |
179 | | char *currentLogFilename; |
180 | | int currentLogRotationAge; |
181 | | pg_time_t now; |
182 | | WaitEventSet *wes; |
183 | | |
184 | | #ifndef EXEC_BACKEND |
185 | | |
186 | | /* |
187 | | * In !EXEC_BACKEND, syslogger is immediately ready to take over: the |
188 | | * output files were already opened by postmaster before forking. For the |
189 | | * other case we must wait until the file descriptors have been restored, |
190 | | * below. |
191 | | */ |
192 | | syslogger_setup_done = true; |
193 | | #endif |
194 | | |
195 | | /* |
196 | | * Re-open the error output files that were opened by SysLogger_Start(). |
197 | | * |
198 | | * We expect this will always succeed, which is too optimistic, but if it |
199 | | * fails there's not a lot we can do to report the problem anyway. As |
200 | | * coded, we'll just crash on a null pointer dereference after failure... |
201 | | */ |
202 | | #ifdef EXEC_BACKEND |
203 | | { |
204 | | const SysloggerStartupData *slsdata = startup_data; |
205 | | |
206 | | Assert(startup_data_len == sizeof(*slsdata)); |
207 | | syslogFile = syslogger_fdopen(slsdata->syslogFile); |
208 | | csvlogFile = syslogger_fdopen(slsdata->csvlogFile); |
209 | | jsonlogFile = syslogger_fdopen(slsdata->jsonlogFile); |
210 | | |
211 | | syslogger_setup_done = true; |
212 | | } |
213 | | #else |
214 | | Assert(startup_data_len == 0); |
215 | | #endif |
216 | | |
217 | | /* |
218 | | * Now that we're done reading the startup data, release postmaster's |
219 | | * working memory context. |
220 | | */ |
221 | | if (PostmasterContext) |
222 | | { |
223 | | MemoryContextDelete(PostmasterContext); |
224 | | PostmasterContext = NULL; |
225 | | } |
226 | | |
227 | | now = MyStartTime; |
228 | | |
229 | | init_ps_display(NULL); |
230 | | |
231 | | /* |
232 | | * If we restarted, our stderr is already redirected into our own input |
233 | | * pipe. This is of course pretty useless, not to mention that it |
234 | | * interferes with detecting pipe EOF. Point stderr to /dev/null. This |
235 | | * assumes that all interesting messages generated in the syslogger will |
236 | | * come through elog.c and will be sent to write_syslogger_file. |
237 | | */ |
238 | | if (redirection_done) |
239 | | { |
240 | | int fd = open(DEVNULL, O_WRONLY, 0); |
241 | | |
242 | | /* |
243 | | * The closes might look redundant, but they are not: we want to be |
244 | | * darn sure the pipe gets closed even if the open failed. We can |
245 | | * survive running with stderr pointing nowhere, but we can't afford |
246 | | * to have extra pipe input descriptors hanging around. |
247 | | * |
248 | | * As we're just trying to reset these to go to DEVNULL, there's not |
249 | | * much point in checking for failure from the close/dup2 calls here, |
250 | | * if they fail then presumably the file descriptors are closed and |
251 | | * any writes will go into the bitbucket anyway. |
252 | | */ |
253 | | close(STDOUT_FILENO); |
254 | | close(STDERR_FILENO); |
255 | | if (fd != -1) |
256 | | { |
257 | | (void) dup2(fd, STDOUT_FILENO); |
258 | | (void) dup2(fd, STDERR_FILENO); |
259 | | close(fd); |
260 | | } |
261 | | } |
262 | | |
263 | | /* |
264 | | * Syslogger's own stderr can't be the syslogPipe, so set it back to text |
265 | | * mode if we didn't just close it. (It was set to binary in |
266 | | * SubPostmasterMain). |
267 | | */ |
268 | | #ifdef WIN32 |
269 | | else |
270 | | _setmode(STDERR_FILENO, _O_TEXT); |
271 | | #endif |
272 | | |
273 | | /* |
274 | | * Also close our copy of the write end of the pipe. This is needed to |
275 | | * ensure we can detect pipe EOF correctly. (But note that in the restart |
276 | | * case, the postmaster already did this.) |
277 | | */ |
278 | | #ifndef WIN32 |
279 | | if (syslogPipe[1] >= 0) |
280 | | close(syslogPipe[1]); |
281 | | syslogPipe[1] = -1; |
282 | | #else |
283 | | if (syslogPipe[1]) |
284 | | CloseHandle(syslogPipe[1]); |
285 | | syslogPipe[1] = 0; |
286 | | #endif |
287 | | |
288 | | /* |
289 | | * Properly accept or ignore signals the postmaster might send us |
290 | | * |
291 | | * Note: we ignore all termination signals, and instead exit only when all |
292 | | * upstream processes are gone, to ensure we don't miss any dying gasps of |
293 | | * broken backends... |
294 | | */ |
295 | | |
296 | | pqsignal(SIGHUP, SignalHandlerForConfigReload); /* set flag to read config |
297 | | * file */ |
298 | | pqsignal(SIGINT, PG_SIG_IGN); |
299 | | pqsignal(SIGTERM, PG_SIG_IGN); |
300 | | pqsignal(SIGQUIT, PG_SIG_IGN); |
301 | | pqsignal(SIGALRM, PG_SIG_IGN); |
302 | | pqsignal(SIGPIPE, PG_SIG_IGN); |
303 | | pqsignal(SIGUSR1, sigUsr1Handler); /* request log rotation */ |
304 | | pqsignal(SIGUSR2, PG_SIG_IGN); |
305 | | |
306 | | /* |
307 | | * Reset some signals that are accepted by postmaster but not here |
308 | | */ |
309 | | pqsignal(SIGCHLD, PG_SIG_DFL); |
310 | | |
311 | | sigprocmask(SIG_SETMASK, &UnBlockSig, NULL); |
312 | | |
313 | | #ifdef WIN32 |
314 | | /* Fire up separate data transfer thread */ |
315 | | InitializeCriticalSection(&sysloggerSection); |
316 | | EnterCriticalSection(&sysloggerSection); |
317 | | |
318 | | threadHandle = (HANDLE) _beginthreadex(NULL, 0, pipeThread, NULL, 0, NULL); |
319 | | if (threadHandle == 0) |
320 | | elog(FATAL, "could not create syslogger data transfer thread: %m"); |
321 | | #endif /* WIN32 */ |
322 | | |
323 | | /* |
324 | | * Remember active logfiles' name(s). We recompute 'em from the reference |
325 | | * time because passing down just the pg_time_t is a lot cheaper than |
326 | | * passing a whole file path in the EXEC_BACKEND case. |
327 | | */ |
328 | | last_sys_file_name = logfile_getname(first_syslogger_file_time, NULL); |
329 | | if (csvlogFile != NULL) |
330 | | last_csv_file_name = logfile_getname(first_syslogger_file_time, ".csv"); |
331 | | if (jsonlogFile != NULL) |
332 | | last_json_file_name = logfile_getname(first_syslogger_file_time, ".json"); |
333 | | |
334 | | /* remember active logfile parameters */ |
335 | | currentLogDir = pstrdup(Log_directory); |
336 | | currentLogFilename = pstrdup(Log_filename); |
337 | | currentLogRotationAge = Log_RotationAge; |
338 | | /* set next planned rotation time */ |
339 | | set_next_rotation_time(); |
340 | | update_metainfo_datafile(); |
341 | | |
342 | | /* |
343 | | * Reset whereToSendOutput, as the postmaster will do (but hasn't yet, at |
344 | | * the point where we forked). This prevents duplicate output of messages |
345 | | * from syslogger itself. |
346 | | */ |
347 | | whereToSendOutput = DestNone; |
348 | | |
349 | | /* |
350 | | * Set up a reusable WaitEventSet object we'll use to wait for our latch, |
351 | | * and (except on Windows) our socket. |
352 | | * |
353 | | * Unlike all other postmaster child processes, we'll ignore postmaster |
354 | | * death because we want to collect final log output from all backends and |
355 | | * then exit last. We'll do that by running until we see EOF on the |
356 | | * syslog pipe, which implies that all other backends have exited |
357 | | * (including the postmaster). |
358 | | */ |
359 | | wes = CreateWaitEventSet(NULL, 2); |
360 | | AddWaitEventToSet(wes, WL_LATCH_SET, PGINVALID_SOCKET, MyLatch, NULL); |
361 | | #ifndef WIN32 |
362 | | AddWaitEventToSet(wes, WL_SOCKET_READABLE, syslogPipe[0], NULL, NULL); |
363 | | #endif |
364 | | |
365 | | /* main worker loop */ |
366 | | for (;;) |
367 | | { |
368 | | bool time_based_rotation = false; |
369 | | int size_rotation_for = 0; |
370 | | long cur_timeout; |
371 | | WaitEvent event; |
372 | | |
373 | | #ifndef WIN32 |
374 | | int rc; |
375 | | #endif |
376 | | |
377 | | /* Clear any already-pending wakeups */ |
378 | | ResetLatch(MyLatch); |
379 | | |
380 | | /* |
381 | | * Process any requests or signals received recently. |
382 | | */ |
383 | | if (ConfigReloadPending) |
384 | | { |
385 | | ConfigReloadPending = false; |
386 | | ProcessConfigFile(PGC_SIGHUP); |
387 | | |
388 | | /* |
389 | | * Check if the log directory or filename pattern changed in |
390 | | * postgresql.conf. If so, force rotation to make sure we're |
391 | | * writing the logfiles in the right place. |
392 | | */ |
393 | | if (strcmp(Log_directory, currentLogDir) != 0) |
394 | | { |
395 | | pfree(currentLogDir); |
396 | | currentLogDir = pstrdup(Log_directory); |
397 | | rotation_requested = true; |
398 | | |
399 | | /* |
400 | | * Also, create new directory if not present; ignore errors |
401 | | */ |
402 | | (void) MakePGDirectory(Log_directory); |
403 | | } |
404 | | if (strcmp(Log_filename, currentLogFilename) != 0) |
405 | | { |
406 | | pfree(currentLogFilename); |
407 | | currentLogFilename = pstrdup(Log_filename); |
408 | | rotation_requested = true; |
409 | | } |
410 | | |
411 | | /* |
412 | | * Force a rotation if CSVLOG output was just turned on or off and |
413 | | * we need to open or close csvlogFile accordingly. |
414 | | */ |
415 | | if (((Log_destination & LOG_DESTINATION_CSVLOG) != 0) != |
416 | | (csvlogFile != NULL)) |
417 | | rotation_requested = true; |
418 | | |
419 | | /* |
420 | | * Force a rotation if JSONLOG output was just turned on or off |
421 | | * and we need to open or close jsonlogFile accordingly. |
422 | | */ |
423 | | if (((Log_destination & LOG_DESTINATION_JSONLOG) != 0) != |
424 | | (jsonlogFile != NULL)) |
425 | | rotation_requested = true; |
426 | | |
427 | | /* |
428 | | * If rotation time parameter changed, reset next rotation time, |
429 | | * but don't immediately force a rotation. |
430 | | */ |
431 | | if (currentLogRotationAge != Log_RotationAge) |
432 | | { |
433 | | currentLogRotationAge = Log_RotationAge; |
434 | | set_next_rotation_time(); |
435 | | } |
436 | | |
437 | | /* |
438 | | * If we had a rotation-disabling failure, re-enable rotation |
439 | | * attempts after SIGHUP, and force one immediately. |
440 | | */ |
441 | | if (rotation_disabled) |
442 | | { |
443 | | rotation_disabled = false; |
444 | | rotation_requested = true; |
445 | | } |
446 | | |
447 | | /* |
448 | | * Force rewriting last log filename when reloading configuration. |
449 | | * Even if rotation_requested is false, log_destination may have |
450 | | * been changed and we don't want to wait the next file rotation. |
451 | | */ |
452 | | update_metainfo_datafile(); |
453 | | } |
454 | | |
455 | | if (Log_RotationAge > 0 && !rotation_disabled) |
456 | | { |
457 | | /* Do a logfile rotation if it's time */ |
458 | | now = (pg_time_t) time(NULL); |
459 | | if (now >= next_rotation_time) |
460 | | rotation_requested = time_based_rotation = true; |
461 | | } |
462 | | |
463 | | if (!rotation_requested && Log_RotationSize > 0 && !rotation_disabled) |
464 | | { |
465 | | /* Do a rotation if file is too big */ |
466 | | if (ftello(syslogFile) >= Log_RotationSize * (pgoff_t) 1024) |
467 | | { |
468 | | rotation_requested = true; |
469 | | size_rotation_for |= LOG_DESTINATION_STDERR; |
470 | | } |
471 | | if (csvlogFile != NULL && |
472 | | ftello(csvlogFile) >= Log_RotationSize * (pgoff_t) 1024) |
473 | | { |
474 | | rotation_requested = true; |
475 | | size_rotation_for |= LOG_DESTINATION_CSVLOG; |
476 | | } |
477 | | if (jsonlogFile != NULL && |
478 | | ftello(jsonlogFile) >= Log_RotationSize * (pgoff_t) 1024) |
479 | | { |
480 | | rotation_requested = true; |
481 | | size_rotation_for |= LOG_DESTINATION_JSONLOG; |
482 | | } |
483 | | } |
484 | | |
485 | | if (rotation_requested) |
486 | | { |
487 | | /* |
488 | | * Force rotation when both values are zero. It means the request |
489 | | * was sent by pg_rotate_logfile() or "pg_ctl logrotate". |
490 | | */ |
491 | | if (!time_based_rotation && size_rotation_for == 0) |
492 | | size_rotation_for = LOG_DESTINATION_STDERR | |
493 | | LOG_DESTINATION_CSVLOG | |
494 | | LOG_DESTINATION_JSONLOG; |
495 | | logfile_rotate(time_based_rotation, size_rotation_for); |
496 | | } |
497 | | |
498 | | /* |
499 | | * Calculate time till next time-based rotation, so that we don't |
500 | | * sleep longer than that. We assume the value of "now" obtained |
501 | | * above is still close enough. Note we can't make this calculation |
502 | | * until after calling logfile_rotate(), since it will advance |
503 | | * next_rotation_time. |
504 | | * |
505 | | * Also note that we need to beware of overflow in calculation of the |
506 | | * timeout: with large settings of Log_RotationAge, next_rotation_time |
507 | | * could be more than INT_MAX msec in the future. In that case we'll |
508 | | * wait no more than INT_MAX msec, and try again. |
509 | | */ |
510 | | if (Log_RotationAge > 0 && !rotation_disabled) |
511 | | { |
512 | | pg_time_t delay; |
513 | | |
514 | | delay = next_rotation_time - now; |
515 | | if (delay > 0) |
516 | | { |
517 | | if (delay > INT_MAX / 1000) |
518 | | delay = INT_MAX / 1000; |
519 | | cur_timeout = delay * 1000L; /* msec */ |
520 | | } |
521 | | else |
522 | | cur_timeout = 0; |
523 | | } |
524 | | else |
525 | | cur_timeout = -1L; |
526 | | |
527 | | /* |
528 | | * Sleep until there's something to do |
529 | | */ |
530 | | #ifndef WIN32 |
531 | | rc = WaitEventSetWait(wes, cur_timeout, &event, 1, |
532 | | WAIT_EVENT_SYSLOGGER_MAIN); |
533 | | |
534 | | if (rc == 1 && event.events == WL_SOCKET_READABLE) |
535 | | { |
536 | | ssize_t bytesRead; |
537 | | |
538 | | bytesRead = read(syslogPipe[0], |
539 | | logbuffer + bytes_in_logbuffer, |
540 | | sizeof(logbuffer) - bytes_in_logbuffer); |
541 | | if (bytesRead < 0) |
542 | | { |
543 | | if (errno != EINTR) |
544 | | ereport(LOG, |
545 | | (errcode_for_socket_access(), |
546 | | errmsg("could not read from logger pipe: %m"))); |
547 | | } |
548 | | else if (bytesRead > 0) |
549 | | { |
550 | | bytes_in_logbuffer += bytesRead; |
551 | | process_pipe_input(logbuffer, &bytes_in_logbuffer); |
552 | | continue; |
553 | | } |
554 | | else |
555 | | { |
556 | | /* |
557 | | * Zero bytes read when select() is saying read-ready means |
558 | | * EOF on the pipe: that is, there are no longer any processes |
559 | | * with the pipe write end open. Therefore, the postmaster |
560 | | * and all backends are shut down, and we are done. |
561 | | */ |
562 | | pipe_eof_seen = true; |
563 | | |
564 | | /* if there's any data left then force it out now */ |
565 | | flush_pipe_input(logbuffer, &bytes_in_logbuffer); |
566 | | } |
567 | | } |
568 | | #else /* WIN32 */ |
569 | | |
570 | | /* |
571 | | * On Windows we leave it to a separate thread to transfer data and |
572 | | * detect pipe EOF. The main thread just wakes up to handle SIGHUP |
573 | | * and rotation conditions. |
574 | | * |
575 | | * Server code isn't generally thread-safe, so we ensure that only one |
576 | | * of the threads is active at a time by entering the critical section |
577 | | * whenever we're not sleeping. |
578 | | */ |
579 | | LeaveCriticalSection(&sysloggerSection); |
580 | | |
581 | | (void) WaitEventSetWait(wes, cur_timeout, &event, 1, |
582 | | WAIT_EVENT_SYSLOGGER_MAIN); |
583 | | |
584 | | EnterCriticalSection(&sysloggerSection); |
585 | | #endif /* WIN32 */ |
586 | | |
587 | | if (pipe_eof_seen) |
588 | | { |
589 | | /* |
590 | | * seeing this message on the real stderr is annoying - so we make |
591 | | * it DEBUG1 to suppress in normal use. |
592 | | */ |
593 | | ereport(DEBUG1, |
594 | | (errmsg_internal("logger shutting down"))); |
595 | | |
596 | | /* |
597 | | * Normal exit from the syslogger is here. Note that we |
598 | | * deliberately do not close syslogFile before exiting; this is to |
599 | | * allow for the possibility of elog messages being generated |
600 | | * inside proc_exit. Regular exit() will take care of flushing |
601 | | * and closing stdio channels. |
602 | | */ |
603 | | proc_exit(0); |
604 | | } |
605 | | } |
606 | | } |
607 | | |
608 | | /* |
609 | | * Postmaster subroutine to start a syslogger subprocess. |
610 | | */ |
611 | | int |
612 | | SysLogger_Start(int child_slot) |
613 | 0 | { |
614 | 0 | pid_t sysloggerPid; |
615 | 0 | char *filename; |
616 | | #ifdef EXEC_BACKEND |
617 | | SysloggerStartupData startup_data; |
618 | | #endif /* EXEC_BACKEND */ |
619 | |
|
620 | 0 | Assert(Logging_collector); |
621 | | |
622 | | /* |
623 | | * If first time through, create the pipe which will receive stderr |
624 | | * output. |
625 | | * |
626 | | * If the syslogger crashes and needs to be restarted, we continue to use |
627 | | * the same pipe (indeed must do so, since extant backends will be writing |
628 | | * into that pipe). |
629 | | * |
630 | | * This means the postmaster must continue to hold the read end of the |
631 | | * pipe open, so we can pass it down to the reincarnated syslogger. This |
632 | | * is a bit klugy but we have little choice. |
633 | | * |
634 | | * Also note that we don't bother counting the pipe FDs by calling |
635 | | * Reserve/ReleaseExternalFD. There's no real need to account for them |
636 | | * accurately in the postmaster or syslogger process, and both ends of the |
637 | | * pipe will wind up closed in all other postmaster children. |
638 | | */ |
639 | 0 | #ifndef WIN32 |
640 | 0 | if (syslogPipe[0] < 0) |
641 | 0 | { |
642 | 0 | if (pipe(syslogPipe) < 0) |
643 | 0 | ereport(FATAL, |
644 | 0 | (errcode_for_socket_access(), |
645 | 0 | errmsg("could not create pipe for syslog: %m"))); |
646 | 0 | } |
647 | | #else |
648 | | if (!syslogPipe[0]) |
649 | | { |
650 | | SECURITY_ATTRIBUTES sa; |
651 | | |
652 | | memset(&sa, 0, sizeof(SECURITY_ATTRIBUTES)); |
653 | | sa.nLength = sizeof(SECURITY_ATTRIBUTES); |
654 | | sa.bInheritHandle = TRUE; |
655 | | |
656 | | if (!CreatePipe(&syslogPipe[0], &syslogPipe[1], &sa, 32768)) |
657 | | ereport(FATAL, |
658 | | (errcode_for_file_access(), |
659 | | errmsg("could not create pipe for syslog: %m"))); |
660 | | } |
661 | | #endif |
662 | | |
663 | | /* |
664 | | * Create log directory if not present; ignore errors |
665 | | */ |
666 | 0 | (void) MakePGDirectory(Log_directory); |
667 | | |
668 | | /* |
669 | | * The initial logfile is created right in the postmaster, to verify that |
670 | | * the Log_directory is writable. We save the reference time so that the |
671 | | * syslogger child process can recompute this file name. |
672 | | * |
673 | | * It might look a bit strange to re-do this during a syslogger restart, |
674 | | * but we must do so since the postmaster closed syslogFile after the |
675 | | * previous fork (and remembering that old file wouldn't be right anyway). |
676 | | * Note we always append here, we won't overwrite any existing file. This |
677 | | * is consistent with the normal rules, because by definition this is not |
678 | | * a time-based rotation. |
679 | | */ |
680 | 0 | first_syslogger_file_time = time(NULL); |
681 | |
|
682 | 0 | filename = logfile_getname(first_syslogger_file_time, NULL); |
683 | |
|
684 | 0 | syslogFile = logfile_open(filename, "a", false); |
685 | |
|
686 | 0 | pfree(filename); |
687 | | |
688 | | /* |
689 | | * Likewise for the initial CSV log file, if that's enabled. (Note that |
690 | | * we open syslogFile even when only CSV output is nominally enabled, |
691 | | * since some code paths will write to syslogFile anyway.) |
692 | | */ |
693 | 0 | if (Log_destination & LOG_DESTINATION_CSVLOG) |
694 | 0 | { |
695 | 0 | filename = logfile_getname(first_syslogger_file_time, ".csv"); |
696 | |
|
697 | 0 | csvlogFile = logfile_open(filename, "a", false); |
698 | |
|
699 | 0 | pfree(filename); |
700 | 0 | } |
701 | | |
702 | | /* |
703 | | * Likewise for the initial JSON log file, if that's enabled. (Note that |
704 | | * we open syslogFile even when only JSON output is nominally enabled, |
705 | | * since some code paths will write to syslogFile anyway.) |
706 | | */ |
707 | 0 | if (Log_destination & LOG_DESTINATION_JSONLOG) |
708 | 0 | { |
709 | 0 | filename = logfile_getname(first_syslogger_file_time, ".json"); |
710 | |
|
711 | 0 | jsonlogFile = logfile_open(filename, "a", false); |
712 | |
|
713 | 0 | pfree(filename); |
714 | 0 | } |
715 | |
|
716 | | #ifdef EXEC_BACKEND |
717 | | startup_data.syslogFile = syslogger_fdget(syslogFile); |
718 | | startup_data.csvlogFile = syslogger_fdget(csvlogFile); |
719 | | startup_data.jsonlogFile = syslogger_fdget(jsonlogFile); |
720 | | sysloggerPid = postmaster_child_launch(B_LOGGER, child_slot, |
721 | | &startup_data, sizeof(startup_data), NULL); |
722 | | #else |
723 | 0 | sysloggerPid = postmaster_child_launch(B_LOGGER, child_slot, |
724 | 0 | NULL, 0, NULL); |
725 | 0 | #endif /* EXEC_BACKEND */ |
726 | |
|
727 | 0 | if (sysloggerPid == -1) |
728 | 0 | { |
729 | 0 | ereport(LOG, |
730 | 0 | (errmsg("could not fork system logger: %m"))); |
731 | 0 | return 0; |
732 | 0 | } |
733 | | |
734 | | /* success, in postmaster */ |
735 | | |
736 | | /* now we redirect stderr, if not done already */ |
737 | 0 | if (!redirection_done) |
738 | 0 | { |
739 | | #ifdef WIN32 |
740 | | int fd; |
741 | | #endif |
742 | | |
743 | | /* |
744 | | * Leave a breadcrumb trail when redirecting, in case the user forgets |
745 | | * that redirection is active and looks only at the original stderr |
746 | | * target file. |
747 | | */ |
748 | 0 | ereport(LOG, |
749 | 0 | (errmsg("redirecting log output to logging collector process"), |
750 | 0 | errhint("Future log output will appear in directory \"%s\".", |
751 | 0 | Log_directory))); |
752 | | |
753 | 0 | #ifndef WIN32 |
754 | 0 | fflush(stdout); |
755 | 0 | if (dup2(syslogPipe[1], STDOUT_FILENO) < 0) |
756 | 0 | ereport(FATAL, |
757 | 0 | (errcode_for_file_access(), |
758 | 0 | errmsg("could not redirect stdout: %m"))); |
759 | 0 | fflush(stderr); |
760 | 0 | if (dup2(syslogPipe[1], STDERR_FILENO) < 0) |
761 | 0 | ereport(FATAL, |
762 | 0 | (errcode_for_file_access(), |
763 | 0 | errmsg("could not redirect stderr: %m"))); |
764 | | /* Now we are done with the write end of the pipe. */ |
765 | 0 | close(syslogPipe[1]); |
766 | 0 | syslogPipe[1] = -1; |
767 | | #else |
768 | | |
769 | | /* |
770 | | * open the pipe in binary mode and make sure stderr is binary after |
771 | | * it's been dup'ed into, to avoid disturbing the pipe chunking |
772 | | * protocol. |
773 | | */ |
774 | | fflush(stderr); |
775 | | fd = _open_osfhandle((intptr_t) syslogPipe[1], |
776 | | _O_APPEND | _O_BINARY); |
777 | | if (dup2(fd, STDERR_FILENO) < 0) |
778 | | ereport(FATAL, |
779 | | (errcode_for_file_access(), |
780 | | errmsg("could not redirect stderr: %m"))); |
781 | | close(fd); |
782 | | _setmode(STDERR_FILENO, _O_BINARY); |
783 | | |
784 | | /* |
785 | | * Now we are done with the write end of the pipe. CloseHandle() must |
786 | | * not be called because the preceding close() closes the underlying |
787 | | * handle. |
788 | | */ |
789 | | syslogPipe[1] = 0; |
790 | | #endif |
791 | 0 | redirection_done = true; |
792 | 0 | } |
793 | | |
794 | | /* postmaster will never write the file(s); close 'em */ |
795 | 0 | fclose(syslogFile); |
796 | 0 | syslogFile = NULL; |
797 | 0 | if (csvlogFile != NULL) |
798 | 0 | { |
799 | 0 | fclose(csvlogFile); |
800 | 0 | csvlogFile = NULL; |
801 | 0 | } |
802 | 0 | if (jsonlogFile != NULL) |
803 | 0 | { |
804 | 0 | fclose(jsonlogFile); |
805 | 0 | jsonlogFile = NULL; |
806 | 0 | } |
807 | 0 | return (int) sysloggerPid; |
808 | 0 | } |
809 | | |
810 | | |
811 | | #ifdef EXEC_BACKEND |
812 | | |
813 | | /* |
814 | | * syslogger_fdget() - |
815 | | * |
816 | | * Utility wrapper to grab the file descriptor of an opened error output |
817 | | * file. Used when building the command to fork the logging collector. |
818 | | */ |
819 | | static int |
820 | | syslogger_fdget(FILE *file) |
821 | | { |
822 | | #ifndef WIN32 |
823 | | if (file != NULL) |
824 | | return fileno(file); |
825 | | else |
826 | | return -1; |
827 | | #else |
828 | | if (file != NULL) |
829 | | return (int) _get_osfhandle(_fileno(file)); |
830 | | else |
831 | | return 0; |
832 | | #endif /* WIN32 */ |
833 | | } |
834 | | |
835 | | /* |
836 | | * syslogger_fdopen() - |
837 | | * |
838 | | * Utility wrapper to re-open an error output file, using the given file |
839 | | * descriptor. Used when parsing arguments in a forked logging collector. |
840 | | */ |
841 | | static FILE * |
842 | | syslogger_fdopen(int fd) |
843 | | { |
844 | | FILE *file = NULL; |
845 | | |
846 | | #ifndef WIN32 |
847 | | if (fd != -1) |
848 | | { |
849 | | file = fdopen(fd, "a"); |
850 | | setvbuf(file, NULL, PG_IOLBF, 0); |
851 | | } |
852 | | #else /* WIN32 */ |
853 | | if (fd != 0) |
854 | | { |
855 | | fd = _open_osfhandle(fd, _O_APPEND | _O_TEXT); |
856 | | if (fd > 0) |
857 | | { |
858 | | file = fdopen(fd, "a"); |
859 | | setvbuf(file, NULL, PG_IOLBF, 0); |
860 | | } |
861 | | } |
862 | | #endif /* WIN32 */ |
863 | | |
864 | | return file; |
865 | | } |
866 | | #endif /* EXEC_BACKEND */ |
867 | | |
868 | | |
869 | | /* -------------------------------- |
870 | | * pipe protocol handling |
871 | | * -------------------------------- |
872 | | */ |
873 | | |
874 | | /* |
875 | | * Process data received through the syslogger pipe. |
876 | | * |
877 | | * This routine interprets the log pipe protocol which sends log messages as |
878 | | * (hopefully atomic) chunks - such chunks are detected and reassembled here. |
879 | | * |
880 | | * The protocol has a header that starts with two nul bytes, then has a 16 bit |
881 | | * length, the pid of the sending process, and a flag to indicate if it is |
882 | | * the last chunk in a message. Incomplete chunks are saved until we read some |
883 | | * more, and non-final chunks are accumulated until we get the final chunk. |
884 | | * |
885 | | * All of this is to avoid 2 problems: |
886 | | * . partial messages being written to logfiles (messes rotation), and |
887 | | * . messages from different backends being interleaved (messages garbled). |
888 | | * |
889 | | * Any non-protocol messages are written out directly. These should only come |
890 | | * from non-PostgreSQL sources, however (e.g. third party libraries writing to |
891 | | * stderr). |
892 | | * |
893 | | * logbuffer is the data input buffer, and *bytes_in_logbuffer is the number |
894 | | * of bytes present. On exit, any not-yet-eaten data is left-justified in |
895 | | * logbuffer, and *bytes_in_logbuffer is updated. |
896 | | */ |
897 | | static void |
898 | | process_pipe_input(char *logbuffer, int *bytes_in_logbuffer) |
899 | 0 | { |
900 | 0 | char *cursor = logbuffer; |
901 | 0 | int count = *bytes_in_logbuffer; |
902 | 0 | int dest = LOG_DESTINATION_STDERR; |
903 | | |
904 | | /* While we have enough for a header, process data... */ |
905 | 0 | while (count >= (int) (offsetof(PipeProtoHeader, data) + 1)) |
906 | 0 | { |
907 | 0 | PipeProtoHeader p; |
908 | 0 | int chunklen; |
909 | 0 | uint8 dest_flags; |
910 | | |
911 | | /* Do we have a valid header? */ |
912 | 0 | memcpy(&p, cursor, offsetof(PipeProtoHeader, data)); |
913 | 0 | dest_flags = p.flags & (PIPE_PROTO_DEST_STDERR | |
914 | 0 | PIPE_PROTO_DEST_CSVLOG | |
915 | 0 | PIPE_PROTO_DEST_JSONLOG); |
916 | 0 | if (p.nuls[0] == '\0' && p.nuls[1] == '\0' && |
917 | 0 | p.len > 0 && p.len <= PIPE_MAX_PAYLOAD && |
918 | 0 | p.pid != 0 && |
919 | 0 | pg_number_of_ones[dest_flags] == 1) |
920 | 0 | { |
921 | 0 | List *buffer_list; |
922 | 0 | ListCell *cell; |
923 | 0 | save_buffer *existing_slot = NULL, |
924 | 0 | *free_slot = NULL; |
925 | 0 | StringInfo str; |
926 | |
|
927 | 0 | chunklen = PIPE_HEADER_SIZE + p.len; |
928 | | |
929 | | /* Fall out of loop if we don't have the whole chunk yet */ |
930 | 0 | if (count < chunklen) |
931 | 0 | break; |
932 | | |
933 | 0 | if ((p.flags & PIPE_PROTO_DEST_STDERR) != 0) |
934 | 0 | dest = LOG_DESTINATION_STDERR; |
935 | 0 | else if ((p.flags & PIPE_PROTO_DEST_CSVLOG) != 0) |
936 | 0 | dest = LOG_DESTINATION_CSVLOG; |
937 | 0 | else if ((p.flags & PIPE_PROTO_DEST_JSONLOG) != 0) |
938 | 0 | dest = LOG_DESTINATION_JSONLOG; |
939 | 0 | else |
940 | 0 | { |
941 | | /* this should never happen as of the header validation */ |
942 | 0 | Assert(false); |
943 | 0 | } |
944 | | |
945 | | /* Locate any existing buffer for this source pid */ |
946 | 0 | buffer_list = buffer_lists[p.pid % NBUFFER_LISTS]; |
947 | 0 | foreach(cell, buffer_list) |
948 | 0 | { |
949 | 0 | save_buffer *buf = (save_buffer *) lfirst(cell); |
950 | |
|
951 | 0 | if (buf->pid == p.pid) |
952 | 0 | { |
953 | 0 | existing_slot = buf; |
954 | 0 | break; |
955 | 0 | } |
956 | 0 | if (buf->pid == 0 && free_slot == NULL) |
957 | 0 | free_slot = buf; |
958 | 0 | } |
959 | |
|
960 | 0 | if ((p.flags & PIPE_PROTO_IS_LAST) == 0) |
961 | 0 | { |
962 | | /* |
963 | | * Save a complete non-final chunk in a per-pid buffer |
964 | | */ |
965 | 0 | if (existing_slot != NULL) |
966 | 0 | { |
967 | | /* Add chunk to data from preceding chunks */ |
968 | 0 | str = &(existing_slot->data); |
969 | 0 | appendBinaryStringInfo(str, |
970 | 0 | cursor + PIPE_HEADER_SIZE, |
971 | 0 | p.len); |
972 | 0 | } |
973 | 0 | else |
974 | 0 | { |
975 | | /* First chunk of message, save in a new buffer */ |
976 | 0 | if (free_slot == NULL) |
977 | 0 | { |
978 | | /* |
979 | | * Need a free slot, but there isn't one in the list, |
980 | | * so create a new one and extend the list with it. |
981 | | */ |
982 | 0 | free_slot = palloc_object(save_buffer); |
983 | 0 | buffer_list = lappend(buffer_list, free_slot); |
984 | 0 | buffer_lists[p.pid % NBUFFER_LISTS] = buffer_list; |
985 | 0 | } |
986 | 0 | free_slot->pid = p.pid; |
987 | 0 | str = &(free_slot->data); |
988 | 0 | initStringInfo(str); |
989 | 0 | appendBinaryStringInfo(str, |
990 | 0 | cursor + PIPE_HEADER_SIZE, |
991 | 0 | p.len); |
992 | 0 | } |
993 | 0 | } |
994 | 0 | else |
995 | 0 | { |
996 | | /* |
997 | | * Final chunk --- add it to anything saved for that pid, and |
998 | | * either way write the whole thing out. |
999 | | */ |
1000 | 0 | if (existing_slot != NULL) |
1001 | 0 | { |
1002 | 0 | str = &(existing_slot->data); |
1003 | 0 | appendBinaryStringInfo(str, |
1004 | 0 | cursor + PIPE_HEADER_SIZE, |
1005 | 0 | p.len); |
1006 | 0 | write_syslogger_file(str->data, str->len, dest); |
1007 | | /* Mark the buffer unused, and reclaim string storage */ |
1008 | 0 | existing_slot->pid = 0; |
1009 | 0 | pfree(str->data); |
1010 | 0 | } |
1011 | 0 | else |
1012 | 0 | { |
1013 | | /* The whole message was one chunk, evidently. */ |
1014 | 0 | write_syslogger_file(cursor + PIPE_HEADER_SIZE, p.len, |
1015 | 0 | dest); |
1016 | 0 | } |
1017 | 0 | } |
1018 | | |
1019 | | /* Finished processing this chunk */ |
1020 | 0 | cursor += chunklen; |
1021 | 0 | count -= chunklen; |
1022 | 0 | } |
1023 | 0 | else |
1024 | 0 | { |
1025 | | /* Process non-protocol data */ |
1026 | | |
1027 | | /* |
1028 | | * Look for the start of a protocol header. If found, dump data |
1029 | | * up to there and repeat the loop. Otherwise, dump it all and |
1030 | | * fall out of the loop. (Note: we want to dump it all if at all |
1031 | | * possible, so as to avoid dividing non-protocol messages across |
1032 | | * logfiles. We expect that in many scenarios, a non-protocol |
1033 | | * message will arrive all in one read(), and we want to respect |
1034 | | * the read() boundary if possible.) |
1035 | | */ |
1036 | 0 | for (chunklen = 1; chunklen < count; chunklen++) |
1037 | 0 | { |
1038 | 0 | if (cursor[chunklen] == '\0') |
1039 | 0 | break; |
1040 | 0 | } |
1041 | | /* fall back on the stderr log as the destination */ |
1042 | 0 | write_syslogger_file(cursor, chunklen, LOG_DESTINATION_STDERR); |
1043 | 0 | cursor += chunklen; |
1044 | 0 | count -= chunklen; |
1045 | 0 | } |
1046 | 0 | } |
1047 | | |
1048 | | /* We don't have a full chunk, so left-align what remains in the buffer */ |
1049 | 0 | if (count > 0 && cursor != logbuffer) |
1050 | 0 | memmove(logbuffer, cursor, count); |
1051 | 0 | *bytes_in_logbuffer = count; |
1052 | 0 | } |
1053 | | |
1054 | | /* |
1055 | | * Force out any buffered data |
1056 | | * |
1057 | | * This is currently used only at syslogger shutdown, but could perhaps be |
1058 | | * useful at other times, so it is careful to leave things in a clean state. |
1059 | | */ |
1060 | | static void |
1061 | | flush_pipe_input(char *logbuffer, int *bytes_in_logbuffer) |
1062 | 0 | { |
1063 | 0 | int i; |
1064 | | |
1065 | | /* Dump any incomplete protocol messages */ |
1066 | 0 | for (i = 0; i < NBUFFER_LISTS; i++) |
1067 | 0 | { |
1068 | 0 | List *list = buffer_lists[i]; |
1069 | 0 | ListCell *cell; |
1070 | |
|
1071 | 0 | foreach(cell, list) |
1072 | 0 | { |
1073 | 0 | save_buffer *buf = (save_buffer *) lfirst(cell); |
1074 | |
|
1075 | 0 | if (buf->pid != 0) |
1076 | 0 | { |
1077 | 0 | StringInfo str = &(buf->data); |
1078 | |
|
1079 | 0 | write_syslogger_file(str->data, str->len, |
1080 | 0 | LOG_DESTINATION_STDERR); |
1081 | | /* Mark the buffer unused, and reclaim string storage */ |
1082 | 0 | buf->pid = 0; |
1083 | 0 | pfree(str->data); |
1084 | 0 | } |
1085 | 0 | } |
1086 | 0 | } |
1087 | | |
1088 | | /* |
1089 | | * Force out any remaining pipe data as-is; we don't bother trying to |
1090 | | * remove any protocol headers that may exist in it. |
1091 | | */ |
1092 | 0 | if (*bytes_in_logbuffer > 0) |
1093 | 0 | write_syslogger_file(logbuffer, *bytes_in_logbuffer, |
1094 | 0 | LOG_DESTINATION_STDERR); |
1095 | 0 | *bytes_in_logbuffer = 0; |
1096 | 0 | } |
1097 | | |
1098 | | |
1099 | | /* -------------------------------- |
1100 | | * logfile routines |
1101 | | * -------------------------------- |
1102 | | */ |
1103 | | |
1104 | | /* |
1105 | | * Write text to the currently open logfile |
1106 | | * |
1107 | | * This is exported so that elog.c can call it when MyBackendType is B_LOGGER. |
1108 | | * This allows the syslogger process to record elog messages of its own, |
1109 | | * even though its stderr does not point at the syslog pipe. |
1110 | | */ |
1111 | | void |
1112 | | write_syslogger_file(const char *buffer, int count, int destination) |
1113 | 0 | { |
1114 | 0 | int rc; |
1115 | 0 | FILE *logfile; |
1116 | | |
1117 | | /* |
1118 | | * If we're told to write to a structured log file, but it's not open, |
1119 | | * dump the data to syslogFile (which is always open) instead. This can |
1120 | | * happen if structured output is enabled after postmaster start and we've |
1121 | | * been unable to open logFile. There are also race conditions during a |
1122 | | * parameter change whereby backends might send us structured output |
1123 | | * before we open the logFile or after we close it. Writing formatted |
1124 | | * output to the regular log file isn't great, but it beats dropping log |
1125 | | * output on the floor. |
1126 | | * |
1127 | | * Think not to improve this by trying to open logFile on-the-fly. Any |
1128 | | * failure in that would lead to recursion. |
1129 | | */ |
1130 | 0 | if ((destination & LOG_DESTINATION_CSVLOG) && csvlogFile != NULL) |
1131 | 0 | logfile = csvlogFile; |
1132 | 0 | else if ((destination & LOG_DESTINATION_JSONLOG) && jsonlogFile != NULL) |
1133 | 0 | logfile = jsonlogFile; |
1134 | 0 | else |
1135 | 0 | logfile = syslogFile; |
1136 | |
|
1137 | 0 | rc = fwrite(buffer, 1, count, logfile); |
1138 | | |
1139 | | /* |
1140 | | * Try to report any failure. We mustn't use ereport because it would |
1141 | | * just recurse right back here, but write_stderr is OK: it will write |
1142 | | * either to the postmaster's original stderr, or to /dev/null, but never |
1143 | | * to our input pipe which would result in a different sort of looping. |
1144 | | */ |
1145 | 0 | if (rc != count) |
1146 | 0 | write_stderr("could not write to log file: %m\n"); |
1147 | 0 | } |
1148 | | |
1149 | | #ifdef WIN32 |
1150 | | |
1151 | | /* |
1152 | | * Worker thread to transfer data from the pipe to the current logfile. |
1153 | | * |
1154 | | * We need this because on Windows, WaitForMultipleObjects does not work on |
1155 | | * unnamed pipes: it always reports "signaled", so the blocking ReadFile won't |
1156 | | * allow for SIGHUP; and select is for sockets only. |
1157 | | */ |
1158 | | static unsigned int __stdcall |
1159 | | pipeThread(void *arg) |
1160 | | { |
1161 | | char logbuffer[READ_BUF_SIZE]; |
1162 | | int bytes_in_logbuffer = 0; |
1163 | | |
1164 | | for (;;) |
1165 | | { |
1166 | | DWORD bytesRead; |
1167 | | BOOL result; |
1168 | | |
1169 | | result = ReadFile(syslogPipe[0], |
1170 | | logbuffer + bytes_in_logbuffer, |
1171 | | sizeof(logbuffer) - bytes_in_logbuffer, |
1172 | | &bytesRead, 0); |
1173 | | |
1174 | | /* |
1175 | | * Enter critical section before doing anything that might touch |
1176 | | * global state shared by the main thread. Anything that uses |
1177 | | * palloc()/pfree() in particular are not safe outside the critical |
1178 | | * section. |
1179 | | */ |
1180 | | EnterCriticalSection(&sysloggerSection); |
1181 | | if (!result) |
1182 | | { |
1183 | | DWORD error = GetLastError(); |
1184 | | |
1185 | | if (error == ERROR_HANDLE_EOF || |
1186 | | error == ERROR_BROKEN_PIPE) |
1187 | | break; |
1188 | | _dosmaperr(error); |
1189 | | ereport(LOG, |
1190 | | (errcode_for_file_access(), |
1191 | | errmsg("could not read from logger pipe: %m"))); |
1192 | | } |
1193 | | else if (bytesRead > 0) |
1194 | | { |
1195 | | bytes_in_logbuffer += bytesRead; |
1196 | | process_pipe_input(logbuffer, &bytes_in_logbuffer); |
1197 | | } |
1198 | | |
1199 | | /* |
1200 | | * If we've filled the current logfile, nudge the main thread to do a |
1201 | | * log rotation. |
1202 | | */ |
1203 | | if (Log_RotationSize > 0) |
1204 | | { |
1205 | | if (ftello(syslogFile) >= Log_RotationSize * (pgoff_t) 1024 || |
1206 | | (csvlogFile != NULL && |
1207 | | ftello(csvlogFile) >= Log_RotationSize * (pgoff_t) 1024) || |
1208 | | (jsonlogFile != NULL && |
1209 | | ftello(jsonlogFile) >= Log_RotationSize * (pgoff_t) 1024)) |
1210 | | SetLatch(MyLatch); |
1211 | | } |
1212 | | LeaveCriticalSection(&sysloggerSection); |
1213 | | } |
1214 | | |
1215 | | /* We exit the above loop only upon detecting pipe EOF */ |
1216 | | pipe_eof_seen = true; |
1217 | | |
1218 | | /* if there's any data left then force it out now */ |
1219 | | flush_pipe_input(logbuffer, &bytes_in_logbuffer); |
1220 | | |
1221 | | /* set the latch to waken the main thread, which will quit */ |
1222 | | SetLatch(MyLatch); |
1223 | | |
1224 | | LeaveCriticalSection(&sysloggerSection); |
1225 | | _endthread(); |
1226 | | return 0; |
1227 | | } |
1228 | | #endif /* WIN32 */ |
1229 | | |
1230 | | /* |
1231 | | * Open a new logfile with proper permissions and buffering options. |
1232 | | * |
1233 | | * If allow_errors is true, we just log any open failure and return NULL |
1234 | | * (with errno still correct for the fopen failure). |
1235 | | * Otherwise, errors are treated as fatal. |
1236 | | */ |
1237 | | static FILE * |
1238 | | logfile_open(const char *filename, const char *mode, bool allow_errors) |
1239 | 0 | { |
1240 | 0 | FILE *fh; |
1241 | 0 | mode_t oumask; |
1242 | | |
1243 | | /* |
1244 | | * Note we do not let Log_file_mode disable IWUSR, since we certainly want |
1245 | | * to be able to write the files ourselves. |
1246 | | */ |
1247 | 0 | oumask = umask((mode_t) ((~(Log_file_mode | S_IWUSR)) & (S_IRWXU | S_IRWXG | S_IRWXO))); |
1248 | 0 | fh = fopen(filename, mode); |
1249 | 0 | umask(oumask); |
1250 | |
|
1251 | 0 | if (fh) |
1252 | 0 | { |
1253 | 0 | setvbuf(fh, NULL, PG_IOLBF, 0); |
1254 | |
|
1255 | | #ifdef WIN32 |
1256 | | /* use CRLF line endings on Windows */ |
1257 | | _setmode(_fileno(fh), _O_TEXT); |
1258 | | #endif |
1259 | 0 | } |
1260 | 0 | else |
1261 | 0 | { |
1262 | 0 | int save_errno = errno; |
1263 | |
|
1264 | 0 | ereport(allow_errors ? LOG : FATAL, |
1265 | 0 | (errcode_for_file_access(), |
1266 | 0 | errmsg("could not open log file \"%s\": %m", |
1267 | 0 | filename))); |
1268 | 0 | errno = save_errno; |
1269 | 0 | } |
1270 | | |
1271 | 0 | return fh; |
1272 | 0 | } |
1273 | | |
1274 | | /* |
1275 | | * Do logfile rotation for a single destination, as specified by target_dest. |
1276 | | * The information stored in *last_file_name and *logFile is updated on a |
1277 | | * successful file rotation. |
1278 | | * |
1279 | | * Returns false if the rotation has been stopped, or true to move on to |
1280 | | * the processing of other formats. |
1281 | | */ |
1282 | | static bool |
1283 | | logfile_rotate_dest(bool time_based_rotation, int size_rotation_for, |
1284 | | pg_time_t fntime, int target_dest, |
1285 | | char **last_file_name, FILE **logFile) |
1286 | 0 | { |
1287 | 0 | char *logFileExt = NULL; |
1288 | 0 | char *filename; |
1289 | 0 | FILE *fh; |
1290 | | |
1291 | | /* |
1292 | | * If the target destination was just turned off, close the previous file |
1293 | | * and unregister its data. This cannot happen for stderr as syslogFile |
1294 | | * is assumed to be always opened even if stderr is disabled in |
1295 | | * log_destination. |
1296 | | */ |
1297 | 0 | if ((Log_destination & target_dest) == 0 && |
1298 | 0 | target_dest != LOG_DESTINATION_STDERR) |
1299 | 0 | { |
1300 | 0 | if (*logFile != NULL) |
1301 | 0 | fclose(*logFile); |
1302 | 0 | *logFile = NULL; |
1303 | 0 | if (*last_file_name != NULL) |
1304 | 0 | pfree(*last_file_name); |
1305 | 0 | *last_file_name = NULL; |
1306 | 0 | return true; |
1307 | 0 | } |
1308 | | |
1309 | | /* |
1310 | | * Leave if it is not time for a rotation or if the target destination has |
1311 | | * no need to do a rotation based on the size of its file. |
1312 | | */ |
1313 | 0 | if (!time_based_rotation && (size_rotation_for & target_dest) == 0) |
1314 | 0 | return true; |
1315 | | |
1316 | | /* file extension depends on the destination type */ |
1317 | 0 | if (target_dest == LOG_DESTINATION_STDERR) |
1318 | 0 | logFileExt = NULL; |
1319 | 0 | else if (target_dest == LOG_DESTINATION_CSVLOG) |
1320 | 0 | logFileExt = ".csv"; |
1321 | 0 | else if (target_dest == LOG_DESTINATION_JSONLOG) |
1322 | 0 | logFileExt = ".json"; |
1323 | 0 | else |
1324 | 0 | { |
1325 | | /* cannot happen */ |
1326 | 0 | Assert(false); |
1327 | 0 | } |
1328 | | |
1329 | | /* build the new file name */ |
1330 | 0 | filename = logfile_getname(fntime, logFileExt); |
1331 | | |
1332 | | /* |
1333 | | * Decide whether to overwrite or append. We can overwrite if (a) |
1334 | | * Log_truncate_on_rotation is set, (b) the rotation was triggered by |
1335 | | * elapsed time and not something else, and (c) the computed file name is |
1336 | | * different from what we were previously logging into. |
1337 | | */ |
1338 | 0 | if (Log_truncate_on_rotation && time_based_rotation && |
1339 | 0 | *last_file_name != NULL && |
1340 | 0 | strcmp(filename, *last_file_name) != 0) |
1341 | 0 | fh = logfile_open(filename, "w", true); |
1342 | 0 | else |
1343 | 0 | fh = logfile_open(filename, "a", true); |
1344 | |
|
1345 | 0 | if (!fh) |
1346 | 0 | { |
1347 | | /* |
1348 | | * ENFILE/EMFILE are not too surprising on a busy system; just keep |
1349 | | * using the old file till we manage to get a new one. Otherwise, |
1350 | | * assume something's wrong with Log_directory and stop trying to |
1351 | | * create files. |
1352 | | */ |
1353 | 0 | if (errno != ENFILE && errno != EMFILE) |
1354 | 0 | { |
1355 | 0 | ereport(LOG, |
1356 | 0 | (errmsg("disabling automatic rotation (use SIGHUP to re-enable)"))); |
1357 | 0 | rotation_disabled = true; |
1358 | 0 | } |
1359 | | |
1360 | 0 | if (filename) |
1361 | 0 | pfree(filename); |
1362 | 0 | return false; |
1363 | 0 | } |
1364 | | |
1365 | | /* fill in the new information */ |
1366 | 0 | if (*logFile != NULL) |
1367 | 0 | fclose(*logFile); |
1368 | 0 | *logFile = fh; |
1369 | | |
1370 | | /* instead of pfree'ing filename, remember it for next time */ |
1371 | 0 | if (*last_file_name != NULL) |
1372 | 0 | pfree(*last_file_name); |
1373 | 0 | *last_file_name = filename; |
1374 | |
|
1375 | 0 | return true; |
1376 | 0 | } |
1377 | | |
1378 | | /* |
1379 | | * perform logfile rotation |
1380 | | */ |
1381 | | static void |
1382 | | logfile_rotate(bool time_based_rotation, int size_rotation_for) |
1383 | 0 | { |
1384 | 0 | pg_time_t fntime; |
1385 | |
|
1386 | 0 | rotation_requested = false; |
1387 | | |
1388 | | /* |
1389 | | * When doing a time-based rotation, invent the new logfile name based on |
1390 | | * the planned rotation time, not current time, to avoid "slippage" in the |
1391 | | * file name when we don't do the rotation immediately. |
1392 | | */ |
1393 | 0 | if (time_based_rotation) |
1394 | 0 | fntime = next_rotation_time; |
1395 | 0 | else |
1396 | 0 | fntime = time(NULL); |
1397 | | |
1398 | | /* file rotation for stderr */ |
1399 | 0 | if (!logfile_rotate_dest(time_based_rotation, size_rotation_for, fntime, |
1400 | 0 | LOG_DESTINATION_STDERR, &last_sys_file_name, |
1401 | 0 | &syslogFile)) |
1402 | 0 | return; |
1403 | | |
1404 | | /* file rotation for csvlog */ |
1405 | 0 | if (!logfile_rotate_dest(time_based_rotation, size_rotation_for, fntime, |
1406 | 0 | LOG_DESTINATION_CSVLOG, &last_csv_file_name, |
1407 | 0 | &csvlogFile)) |
1408 | 0 | return; |
1409 | | |
1410 | | /* file rotation for jsonlog */ |
1411 | 0 | if (!logfile_rotate_dest(time_based_rotation, size_rotation_for, fntime, |
1412 | 0 | LOG_DESTINATION_JSONLOG, &last_json_file_name, |
1413 | 0 | &jsonlogFile)) |
1414 | 0 | return; |
1415 | | |
1416 | 0 | update_metainfo_datafile(); |
1417 | |
|
1418 | 0 | set_next_rotation_time(); |
1419 | 0 | } |
1420 | | |
1421 | | |
1422 | | /* |
1423 | | * construct logfile name using timestamp information |
1424 | | * |
1425 | | * If suffix isn't NULL, append it to the name, replacing any ".log" |
1426 | | * that may be in the pattern. |
1427 | | * |
1428 | | * Result is palloc'd. |
1429 | | */ |
1430 | | static char * |
1431 | | logfile_getname(pg_time_t timestamp, const char *suffix) |
1432 | 0 | { |
1433 | 0 | char *filename; |
1434 | 0 | int len; |
1435 | |
|
1436 | 0 | filename = palloc(MAXPGPATH); |
1437 | |
|
1438 | 0 | snprintf(filename, MAXPGPATH, "%s/", Log_directory); |
1439 | |
|
1440 | 0 | len = strlen(filename); |
1441 | | |
1442 | | /* treat Log_filename as a strftime pattern */ |
1443 | 0 | pg_strftime(filename + len, MAXPGPATH - len, Log_filename, |
1444 | 0 | pg_localtime(×tamp, log_timezone)); |
1445 | |
|
1446 | 0 | if (suffix != NULL) |
1447 | 0 | { |
1448 | 0 | len = strlen(filename); |
1449 | 0 | if (len > 4 && (strcmp(filename + (len - 4), ".log") == 0)) |
1450 | 0 | len -= 4; |
1451 | 0 | strlcpy(filename + len, suffix, MAXPGPATH - len); |
1452 | 0 | } |
1453 | |
|
1454 | 0 | return filename; |
1455 | 0 | } |
1456 | | |
1457 | | /* |
1458 | | * Determine the next planned rotation time, and store in next_rotation_time. |
1459 | | */ |
1460 | | static void |
1461 | | set_next_rotation_time(void) |
1462 | 0 | { |
1463 | 0 | pg_time_t now; |
1464 | 0 | struct pg_tm *tm; |
1465 | 0 | int rotinterval; |
1466 | | |
1467 | | /* nothing to do if time-based rotation is disabled */ |
1468 | 0 | if (Log_RotationAge <= 0) |
1469 | 0 | return; |
1470 | | |
1471 | | /* |
1472 | | * The requirements here are to choose the next time > now that is a |
1473 | | * "multiple" of the log rotation interval. "Multiple" can be interpreted |
1474 | | * fairly loosely. In this version we align to log_timezone rather than |
1475 | | * GMT. |
1476 | | */ |
1477 | 0 | rotinterval = Log_RotationAge * SECS_PER_MINUTE; /* convert to seconds */ |
1478 | 0 | now = (pg_time_t) time(NULL); |
1479 | 0 | tm = pg_localtime(&now, log_timezone); |
1480 | 0 | now += tm->tm_gmtoff; |
1481 | 0 | now -= now % rotinterval; |
1482 | 0 | now += rotinterval; |
1483 | 0 | now -= tm->tm_gmtoff; |
1484 | 0 | next_rotation_time = now; |
1485 | 0 | } |
1486 | | |
1487 | | /* |
1488 | | * Store the name of the file(s) where the log collector, when enabled, writes |
1489 | | * log messages. Useful for finding the name(s) of the current log file(s) |
1490 | | * when there is time-based logfile rotation. Filenames are stored in a |
1491 | | * temporary file and which is renamed into the final destination for |
1492 | | * atomicity. The file is opened with the same permissions as what gets |
1493 | | * created in the data directory and has proper buffering options. |
1494 | | */ |
1495 | | static void |
1496 | | update_metainfo_datafile(void) |
1497 | | { |
1498 | | FILE *fh; |
1499 | | mode_t oumask; |
1500 | | |
1501 | | if (!(Log_destination & LOG_DESTINATION_STDERR) && |
1502 | | !(Log_destination & LOG_DESTINATION_CSVLOG) && |
1503 | | !(Log_destination & LOG_DESTINATION_JSONLOG)) |
1504 | | { |
1505 | | if (unlink(LOG_METAINFO_DATAFILE) < 0 && errno != ENOENT) |
1506 | | ereport(LOG, |
1507 | | (errcode_for_file_access(), |
1508 | | errmsg("could not remove file \"%s\": %m", |
1509 | | LOG_METAINFO_DATAFILE))); |
1510 | | return; |
1511 | | } |
1512 | | |
1513 | | /* use the same permissions as the data directory for the new file */ |
1514 | | oumask = umask(pg_mode_mask); |
1515 | | fh = fopen(LOG_METAINFO_DATAFILE_TMP, "w"); |
1516 | | umask(oumask); |
1517 | | |
1518 | | if (fh) |
1519 | | { |
1520 | | setvbuf(fh, NULL, PG_IOLBF, 0); |
1521 | | |
1522 | | #ifdef WIN32 |
1523 | | /* use CRLF line endings on Windows */ |
1524 | | _setmode(_fileno(fh), _O_TEXT); |
1525 | | #endif |
1526 | | } |
1527 | | else |
1528 | | { |
1529 | | ereport(LOG, |
1530 | | (errcode_for_file_access(), |
1531 | | errmsg("could not open file \"%s\": %m", |
1532 | | LOG_METAINFO_DATAFILE_TMP))); |
1533 | | return; |
1534 | | } |
1535 | | |
1536 | | if (last_sys_file_name && (Log_destination & LOG_DESTINATION_STDERR)) |
1537 | | { |
1538 | | if (fprintf(fh, "stderr %s\n", last_sys_file_name) < 0) |
1539 | | { |
1540 | | ereport(LOG, |
1541 | | (errcode_for_file_access(), |
1542 | | errmsg("could not write file \"%s\": %m", |
1543 | | LOG_METAINFO_DATAFILE_TMP))); |
1544 | | fclose(fh); |
1545 | | return; |
1546 | | } |
1547 | | } |
1548 | | |
1549 | | if (last_csv_file_name && (Log_destination & LOG_DESTINATION_CSVLOG)) |
1550 | | { |
1551 | | if (fprintf(fh, "csvlog %s\n", last_csv_file_name) < 0) |
1552 | | { |
1553 | | ereport(LOG, |
1554 | | (errcode_for_file_access(), |
1555 | | errmsg("could not write file \"%s\": %m", |
1556 | | LOG_METAINFO_DATAFILE_TMP))); |
1557 | | fclose(fh); |
1558 | | return; |
1559 | | } |
1560 | | } |
1561 | | |
1562 | | if (last_json_file_name && (Log_destination & LOG_DESTINATION_JSONLOG)) |
1563 | | { |
1564 | | if (fprintf(fh, "jsonlog %s\n", last_json_file_name) < 0) |
1565 | | { |
1566 | | ereport(LOG, |
1567 | | (errcode_for_file_access(), |
1568 | | errmsg("could not write file \"%s\": %m", |
1569 | | LOG_METAINFO_DATAFILE_TMP))); |
1570 | | fclose(fh); |
1571 | | return; |
1572 | | } |
1573 | | } |
1574 | | fclose(fh); |
1575 | | |
1576 | | if (rename(LOG_METAINFO_DATAFILE_TMP, LOG_METAINFO_DATAFILE) != 0) |
1577 | | ereport(LOG, |
1578 | | (errcode_for_file_access(), |
1579 | | errmsg("could not rename file \"%s\" to \"%s\": %m", |
1580 | | LOG_METAINFO_DATAFILE_TMP, LOG_METAINFO_DATAFILE))); |
1581 | | } |
1582 | | |
1583 | | /* -------------------------------- |
1584 | | * signal handler routines |
1585 | | * -------------------------------- |
1586 | | */ |
1587 | | |
1588 | | /* |
1589 | | * Check to see if a log rotation request has arrived. Should be |
1590 | | * called by postmaster after receiving SIGUSR1. |
1591 | | */ |
1592 | | bool |
1593 | | CheckLogrotateSignal(void) |
1594 | 0 | { |
1595 | 0 | struct stat stat_buf; |
1596 | |
|
1597 | 0 | if (stat(LOGROTATE_SIGNAL_FILE, &stat_buf) == 0) |
1598 | 0 | return true; |
1599 | | |
1600 | 0 | return false; |
1601 | 0 | } |
1602 | | |
1603 | | /* |
1604 | | * Remove the file signaling a log rotation request. |
1605 | | */ |
1606 | | void |
1607 | | RemoveLogrotateSignalFiles(void) |
1608 | 0 | { |
1609 | 0 | unlink(LOGROTATE_SIGNAL_FILE); |
1610 | 0 | } |
1611 | | |
1612 | | /* SIGUSR1: set flag to rotate logfile */ |
1613 | | static void |
1614 | | sigUsr1Handler(SIGNAL_ARGS) |
1615 | 0 | { |
1616 | | rotation_requested = true; |
1617 | 0 | SetLatch(MyLatch); |
1618 | 0 | } |