Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * postmaster.c
4 : * This program acts as a clearing house for requests to the
5 : * POSTGRES system. Frontend programs connect to the Postmaster,
6 : * and postmaster forks a new backend process to handle the
7 : * connection.
8 : *
9 : * The postmaster also manages system-wide operations such as
10 : * startup and shutdown. The postmaster itself doesn't do those
11 : * operations, mind you --- it just forks off a subprocess to do them
12 : * at the right times. It also takes care of resetting the system
13 : * if a backend crashes.
14 : *
15 : * The postmaster process creates the shared memory and semaphore
16 : * pools during startup, but as a rule does not touch them itself.
17 : * In particular, it is not a member of the PGPROC array of backends
18 : * and so it cannot participate in lock-manager operations. Keeping
19 : * the postmaster away from shared memory operations makes it simpler
20 : * and more reliable. The postmaster is almost always able to recover
21 : * from crashes of individual backends by resetting shared memory;
22 : * if it did much with shared memory then it would be prone to crashing
23 : * along with the backends.
24 : *
25 : * When a request message is received, we now fork() immediately.
26 : * The child process performs authentication of the request, and
27 : * then becomes a backend if successful. This allows the auth code
28 : * to be written in a simple single-threaded style (as opposed to the
29 : * crufty "poor man's multitasking" code that used to be needed).
30 : * More importantly, it ensures that blockages in non-multithreaded
31 : * libraries like SSL or PAM cannot cause denial of service to other
32 : * clients.
33 : *
34 : *
35 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
36 : * Portions Copyright (c) 1994, Regents of the University of California
37 : *
38 : *
39 : * IDENTIFICATION
40 : * src/backend/postmaster/postmaster.c
41 : *
42 : * NOTES
43 : *
44 : * Initialization:
45 : * The Postmaster sets up shared memory data structures
46 : * for the backends.
47 : *
48 : * Synchronization:
49 : * The Postmaster shares memory with the backends but should avoid
50 : * touching shared memory, so as not to become stuck if a crashing
51 : * backend screws up locks or shared memory. Likewise, the Postmaster
52 : * should never block on messages from frontend clients.
53 : *
54 : * Garbage Collection:
55 : * The Postmaster cleans up after backends if they have an emergency
56 : * exit and/or core dump.
57 : *
58 : * Error Reporting:
59 : * Use write_stderr() only for reporting "interactive" errors
60 : * (essentially, bogus arguments on the command line). Once the
61 : * postmaster is launched, use ereport().
62 : *
63 : *-------------------------------------------------------------------------
64 : */
65 :
66 : #include "postgres.h"
67 :
68 : #include <unistd.h>
69 : #include <signal.h>
70 : #include <time.h>
71 : #include <sys/wait.h>
72 : #include <ctype.h>
73 : #include <sys/stat.h>
74 : #include <sys/socket.h>
75 : #include <fcntl.h>
76 : #include <sys/param.h>
77 : #include <netdb.h>
78 : #include <limits.h>
79 :
80 : #ifdef USE_BONJOUR
81 : #include <dns_sd.h>
82 : #endif
83 :
84 : #ifdef USE_SYSTEMD
85 : #include <systemd/sd-daemon.h>
86 : #endif
87 :
88 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
89 : #include <pthread.h>
90 : #endif
91 :
92 : #include "access/xlog.h"
93 : #include "access/xlog_internal.h"
94 : #include "access/xlogrecovery.h"
95 : #include "common/file_perm.h"
96 : #include "common/pg_prng.h"
97 : #include "lib/ilist.h"
98 : #include "libpq/libpq.h"
99 : #include "libpq/pqsignal.h"
100 : #include "pg_getopt.h"
101 : #include "pgstat.h"
102 : #include "port/pg_bswap.h"
103 : #include "postmaster/autovacuum.h"
104 : #include "postmaster/bgworker_internals.h"
105 : #include "postmaster/pgarch.h"
106 : #include "postmaster/postmaster.h"
107 : #include "postmaster/syslogger.h"
108 : #include "postmaster/walsummarizer.h"
109 : #include "replication/logicallauncher.h"
110 : #include "replication/slotsync.h"
111 : #include "replication/walsender.h"
112 : #include "storage/aio_subsys.h"
113 : #include "storage/fd.h"
114 : #include "storage/io_worker.h"
115 : #include "storage/ipc.h"
116 : #include "storage/pmsignal.h"
117 : #include "storage/proc.h"
118 : #include "tcop/backend_startup.h"
119 : #include "tcop/tcopprot.h"
120 : #include "utils/datetime.h"
121 : #include "utils/memutils.h"
122 : #include "utils/pidfile.h"
123 : #include "utils/timestamp.h"
124 : #include "utils/varlena.h"
125 :
126 : #ifdef EXEC_BACKEND
127 : #include "common/file_utils.h"
128 : #include "storage/pg_shmem.h"
129 : #endif
130 :
131 :
132 : /*
133 : * CountChildren and SignalChildren take a bitmask argument to represent
134 : * BackendTypes to count or signal. Define a separate type and functions to
135 : * work with the bitmasks, to avoid accidentally passing a plain BackendType
136 : * in place of a bitmask or vice versa.
137 : */
138 : typedef struct
139 : {
140 : uint32 mask;
141 : } BackendTypeMask;
142 :
143 : StaticAssertDecl(BACKEND_NUM_TYPES < 32, "too many backend types for uint32");
144 :
145 : static const BackendTypeMask BTYPE_MASK_ALL = {(1 << BACKEND_NUM_TYPES) - 1};
146 : static const BackendTypeMask BTYPE_MASK_NONE = {0};
147 :
148 : static inline BackendTypeMask
149 3080 : btmask(BackendType t)
150 : {
151 3080 : BackendTypeMask mask = {.mask = 1 << t};
152 :
153 3080 : return mask;
154 : }
155 :
156 : static inline BackendTypeMask
157 31752 : btmask_add_n(BackendTypeMask mask, int nargs, BackendType *t)
158 : {
159 139712 : for (int i = 0; i < nargs; i++)
160 107960 : mask.mask |= 1 << t[i];
161 31752 : return mask;
162 : }
163 :
164 : #define btmask_add(mask, ...) \
165 : btmask_add_n(mask, \
166 : lengthof(((BackendType[]){__VA_ARGS__})), \
167 : (BackendType[]){__VA_ARGS__} \
168 : )
169 :
170 : static inline BackendTypeMask
171 8612 : btmask_del(BackendTypeMask mask, BackendType t)
172 : {
173 8612 : mask.mask &= ~(1 << t);
174 8612 : return mask;
175 : }
176 :
177 : static inline BackendTypeMask
178 5066 : btmask_all_except_n(int nargs, BackendType *t)
179 : {
180 5066 : BackendTypeMask mask = BTYPE_MASK_ALL;
181 :
182 13678 : for (int i = 0; i < nargs; i++)
183 8612 : mask = btmask_del(mask, t[i]);
184 5066 : return mask;
185 : }
186 :
187 : #define btmask_all_except(...) \
188 : btmask_all_except_n( \
189 : lengthof(((BackendType[]){__VA_ARGS__})), \
190 : (BackendType[]){__VA_ARGS__} \
191 : )
192 :
193 : static inline bool
194 274536 : btmask_contains(BackendTypeMask mask, BackendType t)
195 : {
196 274536 : return (mask.mask & (1 << t)) != 0;
197 : }
198 :
199 :
200 : BackgroundWorker *MyBgworkerEntry = NULL;
201 :
202 : /* The socket number we are listening for connections on */
203 : int PostPortNumber = DEF_PGPORT;
204 :
205 : /* The directory names for Unix socket(s) */
206 : char *Unix_socket_directories;
207 :
208 : /* The TCP listen address(es) */
209 : char *ListenAddresses;
210 :
211 : /*
212 : * SuperuserReservedConnections is the number of backends reserved for
213 : * superuser use, and ReservedConnections is the number of backends reserved
214 : * for use by roles with privileges of the pg_use_reserved_connections
215 : * predefined role. These are taken out of the pool of MaxConnections backend
216 : * slots, so the number of backend slots available for roles that are neither
217 : * superuser nor have privileges of pg_use_reserved_connections is
218 : * (MaxConnections - SuperuserReservedConnections - ReservedConnections).
219 : *
220 : * If the number of remaining slots is less than or equal to
221 : * SuperuserReservedConnections, only superusers can make new connections. If
222 : * the number of remaining slots is greater than SuperuserReservedConnections
223 : * but less than or equal to
224 : * (SuperuserReservedConnections + ReservedConnections), only superusers and
225 : * roles with privileges of pg_use_reserved_connections can make new
226 : * connections. Note that pre-existing superuser and
227 : * pg_use_reserved_connections connections don't count against the limits.
228 : */
229 : int SuperuserReservedConnections;
230 : int ReservedConnections;
231 :
232 : /* The socket(s) we're listening to. */
233 : #define MAXLISTEN 64
234 : static int NumListenSockets = 0;
235 : static pgsocket *ListenSockets = NULL;
236 :
237 : /* still more option variables */
238 : bool EnableSSL = false;
239 :
240 : int PreAuthDelay = 0;
241 : int AuthenticationTimeout = 60;
242 :
243 : bool log_hostname; /* for ps display and logging */
244 :
245 : bool enable_bonjour = false;
246 : char *bonjour_name;
247 : bool restart_after_crash = true;
248 : bool remove_temp_files_after_crash = true;
249 :
250 : /*
251 : * When terminating child processes after fatal errors, like a crash of a
252 : * child process, we normally send SIGQUIT -- and most other comments in this
253 : * file are written on the assumption that we do -- but developers might
254 : * prefer to use SIGABRT to collect per-child core dumps.
255 : */
256 : bool send_abort_for_crash = false;
257 : bool send_abort_for_kill = false;
258 :
259 : /* special child processes; NULL when not running */
260 : static PMChild *StartupPMChild = NULL,
261 : *BgWriterPMChild = NULL,
262 : *CheckpointerPMChild = NULL,
263 : *WalWriterPMChild = NULL,
264 : *WalReceiverPMChild = NULL,
265 : *WalSummarizerPMChild = NULL,
266 : *AutoVacLauncherPMChild = NULL,
267 : *PgArchPMChild = NULL,
268 : *SysLoggerPMChild = NULL,
269 : *SlotSyncWorkerPMChild = NULL;
270 :
271 : /* Startup process's status */
272 : typedef enum
273 : {
274 : STARTUP_NOT_RUNNING,
275 : STARTUP_RUNNING,
276 : STARTUP_SIGNALED, /* we sent it a SIGQUIT or SIGKILL */
277 : STARTUP_CRASHED,
278 : } StartupStatusEnum;
279 :
280 : static StartupStatusEnum StartupStatus = STARTUP_NOT_RUNNING;
281 :
282 : /* Startup/shutdown state */
283 : #define NoShutdown 0
284 : #define SmartShutdown 1
285 : #define FastShutdown 2
286 : #define ImmediateShutdown 3
287 :
288 : static int Shutdown = NoShutdown;
289 :
290 : static bool FatalError = false; /* T if recovering from backend crash */
291 :
292 : /*
293 : * We use a simple state machine to control startup, shutdown, and
294 : * crash recovery (which is rather like shutdown followed by startup).
295 : *
296 : * After doing all the postmaster initialization work, we enter PM_STARTUP
297 : * state and the startup process is launched. The startup process begins by
298 : * reading the control file and other preliminary initialization steps.
299 : * In a normal startup, or after crash recovery, the startup process exits
300 : * with exit code 0 and we switch to PM_RUN state. However, archive recovery
301 : * is handled specially since it takes much longer and we would like to support
302 : * hot standby during archive recovery.
303 : *
304 : * When the startup process is ready to start archive recovery, it signals the
305 : * postmaster, and we switch to PM_RECOVERY state. The background writer and
306 : * checkpointer are launched, while the startup process continues applying WAL.
307 : * If Hot Standby is enabled, then, after reaching a consistent point in WAL
308 : * redo, startup process signals us again, and we switch to PM_HOT_STANDBY
309 : * state and begin accepting connections to perform read-only queries. When
310 : * archive recovery is finished, the startup process exits with exit code 0
311 : * and we switch to PM_RUN state.
312 : *
313 : * Normal child backends can only be launched when we are in PM_RUN or
314 : * PM_HOT_STANDBY state. (connsAllowed can also restrict launching.)
315 : * In other states we handle connection requests by launching "dead-end"
316 : * child processes, which will simply send the client an error message and
317 : * quit. (We track these in the ActiveChildList so that we can know when they
318 : * are all gone; this is important because they're still connected to shared
319 : * memory, and would interfere with an attempt to destroy the shmem segment,
320 : * possibly leading to SHMALL failure when we try to make a new one.)
321 : * In PM_WAIT_DEAD_END state we are waiting for all the dead-end children
322 : * to drain out of the system, and therefore stop accepting connection
323 : * requests at all until the last existing child has quit (which hopefully
324 : * will not be very long).
325 : *
326 : * Notice that this state variable does not distinguish *why* we entered
327 : * states later than PM_RUN --- Shutdown and FatalError must be consulted
328 : * to find that out. FatalError is never true in PM_RECOVERY, PM_HOT_STANDBY,
329 : * or PM_RUN states, nor in PM_WAIT_XLOG_SHUTDOWN states (because we don't
330 : * enter those states when trying to recover from a crash). It can be true in
331 : * PM_STARTUP state, because we don't clear it until we've successfully
332 : * started WAL redo.
333 : */
334 : typedef enum
335 : {
336 : PM_INIT, /* postmaster starting */
337 : PM_STARTUP, /* waiting for startup subprocess */
338 : PM_RECOVERY, /* in archive recovery mode */
339 : PM_HOT_STANDBY, /* in hot standby mode */
340 : PM_RUN, /* normal "database is alive" state */
341 : PM_STOP_BACKENDS, /* need to stop remaining backends */
342 : PM_WAIT_BACKENDS, /* waiting for live backends to exit */
343 : PM_WAIT_XLOG_SHUTDOWN, /* waiting for checkpointer to do shutdown
344 : * ckpt */
345 : PM_WAIT_XLOG_ARCHIVAL, /* waiting for archiver and walsenders to
346 : * finish */
347 : PM_WAIT_IO_WORKERS, /* waiting for io workers to exit */
348 : PM_WAIT_CHECKPOINTER, /* waiting for checkpointer to shut down */
349 : PM_WAIT_DEAD_END, /* waiting for dead-end children to exit */
350 : PM_NO_CHILDREN, /* all important children have exited */
351 : } PMState;
352 :
353 : static PMState pmState = PM_INIT;
354 :
355 : /*
356 : * While performing a "smart shutdown", we restrict new connections but stay
357 : * in PM_RUN or PM_HOT_STANDBY state until all the client backends are gone.
358 : * connsAllowed is a sub-state indicator showing the active restriction.
359 : * It is of no interest unless pmState is PM_RUN or PM_HOT_STANDBY.
360 : */
361 : static bool connsAllowed = true;
362 :
363 : /* Start time of SIGKILL timeout during immediate shutdown or child crash */
364 : /* Zero means timeout is not running */
365 : static time_t AbortStartTime = 0;
366 :
367 : /* Length of said timeout */
368 : #define SIGKILL_CHILDREN_AFTER_SECS 5
369 :
370 : static bool ReachedNormalRunning = false; /* T if we've reached PM_RUN */
371 :
372 : bool ClientAuthInProgress = false; /* T during new-client
373 : * authentication */
374 :
375 : bool redirection_done = false; /* stderr redirected for syslogger? */
376 :
377 : /* received START_AUTOVAC_LAUNCHER signal */
378 : static bool start_autovac_launcher = false;
379 :
380 : /* the launcher needs to be signaled to communicate some condition */
381 : static bool avlauncher_needs_signal = false;
382 :
383 : /* received START_WALRECEIVER signal */
384 : static bool WalReceiverRequested = false;
385 :
386 : /* set when there's a worker that needs to be started up */
387 : static bool StartWorkerNeeded = true;
388 : static bool HaveCrashedWorker = false;
389 :
390 : /* set when signals arrive */
391 : static volatile sig_atomic_t pending_pm_pmsignal;
392 : static volatile sig_atomic_t pending_pm_child_exit;
393 : static volatile sig_atomic_t pending_pm_reload_request;
394 : static volatile sig_atomic_t pending_pm_shutdown_request;
395 : static volatile sig_atomic_t pending_pm_fast_shutdown_request;
396 : static volatile sig_atomic_t pending_pm_immediate_shutdown_request;
397 :
398 : /* event multiplexing object */
399 : static WaitEventSet *pm_wait_set;
400 :
401 : #ifdef USE_SSL
402 : /* Set when and if SSL has been initialized properly */
403 : bool LoadedSSL = false;
404 : #endif
405 :
406 : #ifdef USE_BONJOUR
407 : static DNSServiceRef bonjour_sdref = NULL;
408 : #endif
409 :
410 : /* State for IO worker management. */
411 : static int io_worker_count = 0;
412 : static PMChild *io_worker_children[MAX_IO_WORKERS];
413 :
414 : /*
415 : * postmaster.c - function prototypes
416 : */
417 : static void CloseServerPorts(int status, Datum arg);
418 : static void unlink_external_pid_file(int status, Datum arg);
419 : static void getInstallationPaths(const char *argv0);
420 : static void checkControlFile(void);
421 : static void handle_pm_pmsignal_signal(SIGNAL_ARGS);
422 : static void handle_pm_child_exit_signal(SIGNAL_ARGS);
423 : static void handle_pm_reload_request_signal(SIGNAL_ARGS);
424 : static void handle_pm_shutdown_request_signal(SIGNAL_ARGS);
425 : static void process_pm_pmsignal(void);
426 : static void process_pm_child_exit(void);
427 : static void process_pm_reload_request(void);
428 : static void process_pm_shutdown_request(void);
429 : static void dummy_handler(SIGNAL_ARGS);
430 : static void CleanupBackend(PMChild *bp, int exitstatus);
431 : static void HandleChildCrash(int pid, int exitstatus, const char *procname);
432 : static void LogChildExit(int lev, const char *procname,
433 : int pid, int exitstatus);
434 : static void PostmasterStateMachine(void);
435 : static void UpdatePMState(PMState newState);
436 :
437 : pg_noreturn static void ExitPostmaster(int status);
438 : static int ServerLoop(void);
439 : static int BackendStartup(ClientSocket *client_sock);
440 : static void report_fork_failure_to_client(ClientSocket *client_sock, int errnum);
441 : static CAC_state canAcceptConnections(BackendType backend_type);
442 : static void signal_child(PMChild *pmchild, int signal);
443 : static bool SignalChildren(int signal, BackendTypeMask targetMask);
444 : static void TerminateChildren(int signal);
445 : static int CountChildren(BackendTypeMask targetMask);
446 : static void LaunchMissingBackgroundProcesses(void);
447 : static void maybe_start_bgworkers(void);
448 : static bool maybe_reap_io_worker(int pid);
449 : static void maybe_adjust_io_workers(void);
450 : static bool CreateOptsFile(int argc, char *argv[], char *fullprogname);
451 : static PMChild *StartChildProcess(BackendType type);
452 : static void StartSysLogger(void);
453 : static void StartAutovacuumWorker(void);
454 : static bool StartBackgroundWorker(RegisteredBgWorker *rw);
455 : static void InitPostmasterDeathWatchHandle(void);
456 :
457 : #ifdef WIN32
458 : #define WNOHANG 0 /* ignored, so any integer value will do */
459 :
460 : static pid_t waitpid(pid_t pid, int *exitstatus, int options);
461 : static void WINAPI pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired);
462 :
463 : static HANDLE win32ChildQueue;
464 :
465 : typedef struct
466 : {
467 : HANDLE waitHandle;
468 : HANDLE procHandle;
469 : DWORD procId;
470 : } win32_deadchild_waitinfo;
471 : #endif /* WIN32 */
472 :
473 : /* Macros to check exit status of a child process */
474 : #define EXIT_STATUS_0(st) ((st) == 0)
475 : #define EXIT_STATUS_1(st) (WIFEXITED(st) && WEXITSTATUS(st) == 1)
476 : #define EXIT_STATUS_3(st) (WIFEXITED(st) && WEXITSTATUS(st) == 3)
477 :
478 : #ifndef WIN32
479 : /*
480 : * File descriptors for pipe used to monitor if postmaster is alive.
481 : * First is POSTMASTER_FD_WATCH, second is POSTMASTER_FD_OWN.
482 : */
483 : int postmaster_alive_fds[2] = {-1, -1};
484 : #else
485 : /* Process handle of postmaster used for the same purpose on Windows */
486 : HANDLE PostmasterHandle;
487 : #endif
488 :
489 : /*
490 : * Postmaster main entry point
491 : */
492 : void
493 1792 : PostmasterMain(int argc, char *argv[])
494 : {
495 : int opt;
496 : int status;
497 1792 : char *userDoption = NULL;
498 1792 : bool listen_addr_saved = false;
499 1792 : char *output_config_variable = NULL;
500 :
501 1792 : InitProcessGlobals();
502 :
503 1792 : PostmasterPid = MyProcPid;
504 :
505 1792 : IsPostmasterEnvironment = true;
506 :
507 : /*
508 : * Start our win32 signal implementation
509 : */
510 : #ifdef WIN32
511 : pgwin32_signal_initialize();
512 : #endif
513 :
514 : /*
515 : * We should not be creating any files or directories before we check the
516 : * data directory (see checkDataDir()), but just in case set the umask to
517 : * the most restrictive (owner-only) permissions.
518 : *
519 : * checkDataDir() will reset the umask based on the data directory
520 : * permissions.
521 : */
522 1792 : umask(PG_MODE_MASK_OWNER);
523 :
524 : /*
525 : * By default, palloc() requests in the postmaster will be allocated in
526 : * the PostmasterContext, which is space that can be recycled by backends.
527 : * Allocated data that needs to be available to backends should be
528 : * allocated in TopMemoryContext.
529 : */
530 1792 : PostmasterContext = AllocSetContextCreate(TopMemoryContext,
531 : "Postmaster",
532 : ALLOCSET_DEFAULT_SIZES);
533 1792 : MemoryContextSwitchTo(PostmasterContext);
534 :
535 : /* Initialize paths to installation files */
536 1792 : getInstallationPaths(argv[0]);
537 :
538 : /*
539 : * Set up signal handlers for the postmaster process.
540 : *
541 : * CAUTION: when changing this list, check for side-effects on the signal
542 : * handling setup of child processes. See tcop/postgres.c,
543 : * bootstrap/bootstrap.c, postmaster/bgwriter.c, postmaster/walwriter.c,
544 : * postmaster/autovacuum.c, postmaster/pgarch.c, postmaster/syslogger.c,
545 : * postmaster/bgworker.c and postmaster/checkpointer.c.
546 : */
547 1792 : pqinitmask();
548 1792 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
549 :
550 1792 : pqsignal(SIGHUP, handle_pm_reload_request_signal);
551 1792 : pqsignal(SIGINT, handle_pm_shutdown_request_signal);
552 1792 : pqsignal(SIGQUIT, handle_pm_shutdown_request_signal);
553 1792 : pqsignal(SIGTERM, handle_pm_shutdown_request_signal);
554 1792 : pqsignal(SIGALRM, SIG_IGN); /* ignored */
555 1792 : pqsignal(SIGPIPE, SIG_IGN); /* ignored */
556 1792 : pqsignal(SIGUSR1, handle_pm_pmsignal_signal);
557 1792 : pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
558 1792 : pqsignal(SIGCHLD, handle_pm_child_exit_signal);
559 :
560 : /* This may configure SIGURG, depending on platform. */
561 1792 : InitializeWaitEventSupport();
562 1792 : InitProcessLocalLatch();
563 :
564 : /*
565 : * No other place in Postgres should touch SIGTTIN/SIGTTOU handling. We
566 : * ignore those signals in a postmaster environment, so that there is no
567 : * risk of a child process freezing up due to writing to stderr. But for
568 : * a standalone backend, their default handling is reasonable. Hence, all
569 : * child processes should just allow the inherited settings to stand.
570 : */
571 : #ifdef SIGTTIN
572 1792 : pqsignal(SIGTTIN, SIG_IGN); /* ignored */
573 : #endif
574 : #ifdef SIGTTOU
575 1792 : pqsignal(SIGTTOU, SIG_IGN); /* ignored */
576 : #endif
577 :
578 : /* ignore SIGXFSZ, so that ulimit violations work like disk full */
579 : #ifdef SIGXFSZ
580 1792 : pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
581 : #endif
582 :
583 : /* Begin accepting signals. */
584 1792 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
585 :
586 : /*
587 : * Options setup
588 : */
589 1792 : InitializeGUCOptions();
590 :
591 1792 : opterr = 1;
592 :
593 : /*
594 : * Parse command-line options. CAUTION: keep this in sync with
595 : * tcop/postgres.c (the option sets should not conflict) and with the
596 : * common help() function in main/main.c.
597 : */
598 6426 : while ((opt = getopt(argc, argv, "B:bC:c:D:d:EeFf:h:ijk:lN:OPp:r:S:sTt:W:-:")) != -1)
599 : {
600 4638 : switch (opt)
601 : {
602 0 : case 'B':
603 0 : SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
604 0 : break;
605 :
606 94 : case 'b':
607 : /* Undocumented flag used for binary upgrades */
608 94 : IsBinaryUpgrade = true;
609 94 : break;
610 :
611 6 : case 'C':
612 6 : output_config_variable = strdup(optarg);
613 6 : break;
614 :
615 1442 : case '-':
616 :
617 : /*
618 : * Error if the user misplaced a special must-be-first option
619 : * for dispatching to a subprogram. parse_dispatch_option()
620 : * returns DISPATCH_POSTMASTER if it doesn't find a match, so
621 : * error for anything else.
622 : */
623 1442 : if (parse_dispatch_option(optarg) != DISPATCH_POSTMASTER)
624 0 : ereport(ERROR,
625 : (errcode(ERRCODE_SYNTAX_ERROR),
626 : errmsg("--%s must be first argument", optarg)));
627 :
628 : /* FALLTHROUGH */
629 : case 'c':
630 : {
631 : char *name,
632 : *value;
633 :
634 2240 : ParseLongOption(optarg, &name, &value);
635 2240 : if (!value)
636 : {
637 2 : if (opt == '-')
638 2 : ereport(ERROR,
639 : (errcode(ERRCODE_SYNTAX_ERROR),
640 : errmsg("--%s requires a value",
641 : optarg)));
642 : else
643 0 : ereport(ERROR,
644 : (errcode(ERRCODE_SYNTAX_ERROR),
645 : errmsg("-c %s requires a value",
646 : optarg)));
647 : }
648 :
649 2238 : SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
650 2236 : pfree(name);
651 2236 : pfree(value);
652 2236 : break;
653 : }
654 :
655 1788 : case 'D':
656 1788 : userDoption = strdup(optarg);
657 1788 : break;
658 :
659 0 : case 'd':
660 0 : set_debug_options(atoi(optarg), PGC_POSTMASTER, PGC_S_ARGV);
661 0 : break;
662 :
663 0 : case 'E':
664 0 : SetConfigOption("log_statement", "all", PGC_POSTMASTER, PGC_S_ARGV);
665 0 : break;
666 :
667 0 : case 'e':
668 0 : SetConfigOption("datestyle", "euro", PGC_POSTMASTER, PGC_S_ARGV);
669 0 : break;
670 :
671 196 : case 'F':
672 196 : SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
673 196 : break;
674 :
675 0 : case 'f':
676 0 : if (!set_plan_disabling_options(optarg, PGC_POSTMASTER, PGC_S_ARGV))
677 : {
678 0 : write_stderr("%s: invalid argument for option -f: \"%s\"\n",
679 : progname, optarg);
680 0 : ExitPostmaster(1);
681 : }
682 0 : break;
683 :
684 0 : case 'h':
685 0 : SetConfigOption("listen_addresses", optarg, PGC_POSTMASTER, PGC_S_ARGV);
686 0 : break;
687 :
688 0 : case 'i':
689 0 : SetConfigOption("listen_addresses", "*", PGC_POSTMASTER, PGC_S_ARGV);
690 0 : break;
691 :
692 0 : case 'j':
693 : /* only used by interactive backend */
694 0 : break;
695 :
696 196 : case 'k':
697 196 : SetConfigOption("unix_socket_directories", optarg, PGC_POSTMASTER, PGC_S_ARGV);
698 196 : break;
699 :
700 0 : case 'l':
701 0 : SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
702 0 : break;
703 :
704 0 : case 'N':
705 0 : SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
706 0 : break;
707 :
708 0 : case 'O':
709 0 : SetConfigOption("allow_system_table_mods", "true", PGC_POSTMASTER, PGC_S_ARGV);
710 0 : break;
711 :
712 0 : case 'P':
713 0 : SetConfigOption("ignore_system_indexes", "true", PGC_POSTMASTER, PGC_S_ARGV);
714 0 : break;
715 :
716 118 : case 'p':
717 118 : SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
718 118 : break;
719 :
720 0 : case 'r':
721 : /* only used by single-user backend */
722 0 : break;
723 :
724 0 : case 'S':
725 0 : SetConfigOption("work_mem", optarg, PGC_POSTMASTER, PGC_S_ARGV);
726 0 : break;
727 :
728 0 : case 's':
729 0 : SetConfigOption("log_statement_stats", "true", PGC_POSTMASTER, PGC_S_ARGV);
730 0 : break;
731 :
732 0 : case 'T':
733 :
734 : /*
735 : * This option used to be defined as sending SIGSTOP after a
736 : * backend crash, but sending SIGABRT seems more useful.
737 : */
738 0 : SetConfigOption("send_abort_for_crash", "true", PGC_POSTMASTER, PGC_S_ARGV);
739 0 : break;
740 :
741 0 : case 't':
742 : {
743 0 : const char *tmp = get_stats_option_name(optarg);
744 :
745 0 : if (tmp)
746 : {
747 0 : SetConfigOption(tmp, "true", PGC_POSTMASTER, PGC_S_ARGV);
748 : }
749 : else
750 : {
751 0 : write_stderr("%s: invalid argument for option -t: \"%s\"\n",
752 : progname, optarg);
753 0 : ExitPostmaster(1);
754 : }
755 0 : break;
756 : }
757 :
758 0 : case 'W':
759 0 : SetConfigOption("post_auth_delay", optarg, PGC_POSTMASTER, PGC_S_ARGV);
760 0 : break;
761 :
762 0 : default:
763 0 : write_stderr("Try \"%s --help\" for more information.\n",
764 : progname);
765 0 : ExitPostmaster(1);
766 : }
767 : }
768 :
769 : /*
770 : * Postmaster accepts no non-option switch arguments.
771 : */
772 1788 : if (optind < argc)
773 : {
774 0 : write_stderr("%s: invalid argument: \"%s\"\n",
775 0 : progname, argv[optind]);
776 0 : write_stderr("Try \"%s --help\" for more information.\n",
777 : progname);
778 0 : ExitPostmaster(1);
779 : }
780 :
781 : /*
782 : * Locate the proper configuration files and data directory, and read
783 : * postgresql.conf for the first time.
784 : */
785 1788 : if (!SelectConfigFiles(userDoption, progname))
786 0 : ExitPostmaster(2);
787 :
788 1780 : if (output_config_variable != NULL)
789 : {
790 : /*
791 : * If this is a runtime-computed GUC, it hasn't yet been initialized,
792 : * and the present value is not useful. However, this is a convenient
793 : * place to print the value for most GUCs because it is safe to run
794 : * postmaster startup to this point even if the server is already
795 : * running. For the handful of runtime-computed GUCs that we cannot
796 : * provide meaningful values for yet, we wait until later in
797 : * postmaster startup to print the value. We won't be able to use -C
798 : * on running servers for those GUCs, but using this option now would
799 : * lead to incorrect results for them.
800 : */
801 4 : int flags = GetConfigOptionFlags(output_config_variable, true);
802 :
803 4 : if ((flags & GUC_RUNTIME_COMPUTED) == 0)
804 : {
805 : /*
806 : * "-C guc" was specified, so print GUC's value and exit. No
807 : * extra permission check is needed because the user is reading
808 : * inside the data dir.
809 : */
810 2 : const char *config_val = GetConfigOption(output_config_variable,
811 : false, false);
812 :
813 2 : puts(config_val ? config_val : "");
814 2 : ExitPostmaster(0);
815 : }
816 :
817 : /*
818 : * A runtime-computed GUC will be printed later on. As we initialize
819 : * a server startup sequence, silence any log messages that may show
820 : * up in the output generated. FATAL and more severe messages are
821 : * useful to show, even if one would only expect at least PANIC. LOG
822 : * entries are hidden.
823 : */
824 2 : SetConfigOption("log_min_messages", "FATAL", PGC_SUSET,
825 : PGC_S_OVERRIDE);
826 : }
827 :
828 : /* Verify that DataDir looks reasonable */
829 1778 : checkDataDir();
830 :
831 : /* Check that pg_control exists */
832 1778 : checkControlFile();
833 :
834 : /* And switch working directory into it */
835 1778 : ChangeToDataDir();
836 :
837 : /*
838 : * Check for invalid combinations of GUC settings.
839 : */
840 1778 : if (SuperuserReservedConnections + ReservedConnections >= MaxConnections)
841 : {
842 0 : write_stderr("%s: \"superuser_reserved_connections\" (%d) plus \"reserved_connections\" (%d) must be less than \"max_connections\" (%d)\n",
843 : progname,
844 : SuperuserReservedConnections, ReservedConnections,
845 : MaxConnections);
846 0 : ExitPostmaster(1);
847 : }
848 1778 : if (XLogArchiveMode > ARCHIVE_MODE_OFF && wal_level == WAL_LEVEL_MINIMAL)
849 0 : ereport(ERROR,
850 : (errmsg("WAL archival cannot be enabled when \"wal_level\" is \"minimal\"")));
851 1778 : if (max_wal_senders > 0 && wal_level == WAL_LEVEL_MINIMAL)
852 0 : ereport(ERROR,
853 : (errmsg("WAL streaming (\"max_wal_senders\" > 0) requires \"wal_level\" to be \"replica\" or \"logical\"")));
854 1778 : if (summarize_wal && wal_level == WAL_LEVEL_MINIMAL)
855 0 : ereport(ERROR,
856 : (errmsg("WAL cannot be summarized when \"wal_level\" is \"minimal\"")));
857 1778 : if (sync_replication_slots && wal_level < WAL_LEVEL_LOGICAL)
858 0 : ereport(ERROR,
859 : (errmsg("replication slot synchronization (\"sync_replication_slots\" = on) requires \"wal_level\" >= \"logical\"")));
860 :
861 : /*
862 : * Other one-time internal sanity checks can go here, if they are fast.
863 : * (Put any slow processing further down, after postmaster.pid creation.)
864 : */
865 1778 : if (!CheckDateTokenTables())
866 : {
867 0 : write_stderr("%s: invalid datetoken tables, please fix\n", progname);
868 0 : ExitPostmaster(1);
869 : }
870 :
871 : /*
872 : * Now that we are done processing the postmaster arguments, reset
873 : * getopt(3) library so that it will work correctly in subprocesses.
874 : */
875 1778 : optind = 1;
876 : #ifdef HAVE_INT_OPTRESET
877 : optreset = 1; /* some systems need this too */
878 : #endif
879 :
880 : /* For debugging: display postmaster environment */
881 1778 : if (message_level_is_interesting(DEBUG3))
882 : {
883 : #if !defined(WIN32)
884 : extern char **environ;
885 : #endif
886 : char **p;
887 : StringInfoData si;
888 :
889 4 : initStringInfo(&si);
890 :
891 4 : appendStringInfoString(&si, "initial environment dump:");
892 172 : for (p = environ; *p; ++p)
893 168 : appendStringInfo(&si, "\n%s", *p);
894 :
895 4 : ereport(DEBUG3, errmsg_internal("%s", si.data));
896 4 : pfree(si.data);
897 : }
898 :
899 : /*
900 : * Create lockfile for data directory.
901 : *
902 : * We want to do this before we try to grab the input sockets, because the
903 : * data directory interlock is more reliable than the socket-file
904 : * interlock (thanks to whoever decided to put socket files in /tmp :-().
905 : * For the same reason, it's best to grab the TCP socket(s) before the
906 : * Unix socket(s).
907 : *
908 : * Also note that this internally sets up the on_proc_exit function that
909 : * is responsible for removing both data directory and socket lockfiles;
910 : * so it must happen before opening sockets so that at exit, the socket
911 : * lockfiles go away after CloseServerPorts runs.
912 : */
913 1778 : CreateDataDirLockFile(true);
914 :
915 : /*
916 : * Read the control file (for error checking and config info).
917 : *
918 : * Since we verify the control file's CRC, this has a useful side effect
919 : * on machines where we need a run-time test for CRC support instructions.
920 : * The postmaster will do the test once at startup, and then its child
921 : * processes will inherit the correct function pointer and not need to
922 : * repeat the test.
923 : */
924 1772 : LocalProcessControlFile(false);
925 :
926 : /*
927 : * Register the apply launcher. It's probably a good idea to call this
928 : * before any modules had a chance to take the background worker slots.
929 : */
930 1772 : ApplyLauncherRegister();
931 :
932 : /*
933 : * process any libraries that should be preloaded at postmaster start
934 : */
935 1772 : process_shared_preload_libraries();
936 :
937 : /*
938 : * Initialize SSL library, if specified.
939 : */
940 : #ifdef USE_SSL
941 1772 : if (EnableSSL)
942 : {
943 64 : (void) secure_initialize(true);
944 54 : LoadedSSL = true;
945 : }
946 : #endif
947 :
948 : /*
949 : * Now that loadable modules have had their chance to alter any GUCs,
950 : * calculate MaxBackends and initialize the machinery to track child
951 : * processes.
952 : */
953 1762 : InitializeMaxBackends();
954 1762 : InitPostmasterChildSlots();
955 :
956 : /*
957 : * Calculate the size of the PGPROC fast-path lock arrays.
958 : */
959 1762 : InitializeFastPathLocks();
960 :
961 : /*
962 : * Give preloaded libraries a chance to request additional shared memory.
963 : */
964 1762 : process_shmem_requests();
965 :
966 : /*
967 : * Now that loadable modules have had their chance to request additional
968 : * shared memory, determine the value of any runtime-computed GUCs that
969 : * depend on the amount of shared memory required.
970 : */
971 1762 : InitializeShmemGUCs();
972 :
973 : /*
974 : * Now that modules have been loaded, we can process any custom resource
975 : * managers specified in the wal_consistency_checking GUC.
976 : */
977 1762 : InitializeWalConsistencyChecking();
978 :
979 : /*
980 : * If -C was specified with a runtime-computed GUC, we held off printing
981 : * the value earlier, as the GUC was not yet initialized. We handle -C
982 : * for most GUCs before we lock the data directory so that the option may
983 : * be used on a running server. However, a handful of GUCs are runtime-
984 : * computed and do not have meaningful values until after locking the data
985 : * directory, and we cannot safely calculate their values earlier on a
986 : * running server. At this point, such GUCs should be properly
987 : * initialized, and we haven't yet set up shared memory, so this is a good
988 : * time to handle the -C option for these special GUCs.
989 : */
990 1762 : if (output_config_variable != NULL)
991 : {
992 2 : const char *config_val = GetConfigOption(output_config_variable,
993 : false, false);
994 :
995 2 : puts(config_val ? config_val : "");
996 2 : ExitPostmaster(0);
997 : }
998 :
999 : /*
1000 : * Set up shared memory and semaphores.
1001 : *
1002 : * Note: if using SysV shmem and/or semas, each postmaster startup will
1003 : * normally choose the same IPC keys. This helps ensure that we will
1004 : * clean up dead IPC objects if the postmaster crashes and is restarted.
1005 : */
1006 1760 : CreateSharedMemoryAndSemaphores();
1007 :
1008 : /*
1009 : * Estimate number of openable files. This must happen after setting up
1010 : * semaphores, because on some platforms semaphores count as open files.
1011 : */
1012 1758 : set_max_safe_fds();
1013 :
1014 : /*
1015 : * Initialize pipe (or process handle on Windows) that allows children to
1016 : * wake up from sleep on postmaster death.
1017 : */
1018 1758 : InitPostmasterDeathWatchHandle();
1019 :
1020 : #ifdef WIN32
1021 :
1022 : /*
1023 : * Initialize I/O completion port used to deliver list of dead children.
1024 : */
1025 : win32ChildQueue = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 1);
1026 : if (win32ChildQueue == NULL)
1027 : ereport(FATAL,
1028 : (errmsg("could not create I/O completion port for child queue")));
1029 : #endif
1030 :
1031 : #ifdef EXEC_BACKEND
1032 : /* Write out nondefault GUC settings for child processes to use */
1033 : write_nondefault_variables(PGC_POSTMASTER);
1034 :
1035 : /*
1036 : * Clean out the temp directory used to transmit parameters to child
1037 : * processes (see internal_forkexec). We must do this before launching
1038 : * any child processes, else we have a race condition: we could remove a
1039 : * parameter file before the child can read it. It should be safe to do
1040 : * so now, because we verified earlier that there are no conflicting
1041 : * Postgres processes in this data directory.
1042 : */
1043 : RemovePgTempFilesInDir(PG_TEMP_FILES_DIR, true, false);
1044 : #endif
1045 :
1046 : /*
1047 : * Forcibly remove the files signaling a standby promotion request.
1048 : * Otherwise, the existence of those files triggers a promotion too early,
1049 : * whether a user wants that or not.
1050 : *
1051 : * This removal of files is usually unnecessary because they can exist
1052 : * only during a few moments during a standby promotion. However there is
1053 : * a race condition: if pg_ctl promote is executed and creates the files
1054 : * during a promotion, the files can stay around even after the server is
1055 : * brought up to be the primary. Then, if a new standby starts by using
1056 : * the backup taken from the new primary, the files can exist at server
1057 : * startup and must be removed in order to avoid an unexpected promotion.
1058 : *
1059 : * Note that promotion signal files need to be removed before the startup
1060 : * process is invoked. Because, after that, they can be used by
1061 : * postmaster's SIGUSR1 signal handler.
1062 : */
1063 1758 : RemovePromoteSignalFiles();
1064 :
1065 : /* Do the same for logrotate signal file */
1066 1758 : RemoveLogrotateSignalFiles();
1067 :
1068 : /* Remove any outdated file holding the current log filenames. */
1069 1758 : if (unlink(LOG_METAINFO_DATAFILE) < 0 && errno != ENOENT)
1070 0 : ereport(LOG,
1071 : (errcode_for_file_access(),
1072 : errmsg("could not remove file \"%s\": %m",
1073 : LOG_METAINFO_DATAFILE)));
1074 :
1075 : /*
1076 : * If enabled, start up syslogger collection subprocess
1077 : */
1078 1758 : if (Logging_collector)
1079 2 : StartSysLogger();
1080 :
1081 : /*
1082 : * Reset whereToSendOutput from DestDebug (its starting state) to
1083 : * DestNone. This stops ereport from sending log messages to stderr unless
1084 : * Log_destination permits. We don't do this until the postmaster is
1085 : * fully launched, since startup failures may as well be reported to
1086 : * stderr.
1087 : *
1088 : * If we are in fact disabling logging to stderr, first emit a log message
1089 : * saying so, to provide a breadcrumb trail for users who may not remember
1090 : * that their logging is configured to go somewhere else.
1091 : */
1092 1758 : if (!(Log_destination & LOG_DESTINATION_STDERR))
1093 0 : ereport(LOG,
1094 : (errmsg("ending log output to stderr"),
1095 : errhint("Future log output will go to log destination \"%s\".",
1096 : Log_destination_string)));
1097 :
1098 1758 : whereToSendOutput = DestNone;
1099 :
1100 : /*
1101 : * Report server startup in log. While we could emit this much earlier,
1102 : * it seems best to do so after starting the log collector, if we intend
1103 : * to use one.
1104 : */
1105 1758 : ereport(LOG,
1106 : (errmsg("starting %s", PG_VERSION_STR)));
1107 :
1108 : /*
1109 : * Establish input sockets.
1110 : *
1111 : * First set up an on_proc_exit function that's charged with closing the
1112 : * sockets again at postmaster shutdown.
1113 : */
1114 1758 : ListenSockets = palloc(MAXLISTEN * sizeof(pgsocket));
1115 1758 : on_proc_exit(CloseServerPorts, 0);
1116 :
1117 1758 : if (ListenAddresses)
1118 : {
1119 : char *rawstring;
1120 : List *elemlist;
1121 : ListCell *l;
1122 1758 : int success = 0;
1123 :
1124 : /* Need a modifiable copy of ListenAddresses */
1125 1758 : rawstring = pstrdup(ListenAddresses);
1126 :
1127 : /* Parse string into list of hostnames */
1128 1758 : if (!SplitGUCList(rawstring, ',', &elemlist))
1129 : {
1130 : /* syntax error in list */
1131 0 : ereport(FATAL,
1132 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1133 : errmsg("invalid list syntax in parameter \"%s\"",
1134 : "listen_addresses")));
1135 : }
1136 :
1137 1818 : foreach(l, elemlist)
1138 : {
1139 60 : char *curhost = (char *) lfirst(l);
1140 :
1141 60 : if (strcmp(curhost, "*") == 0)
1142 0 : status = ListenServerPort(AF_UNSPEC, NULL,
1143 0 : (unsigned short) PostPortNumber,
1144 : NULL,
1145 : ListenSockets,
1146 : &NumListenSockets,
1147 : MAXLISTEN);
1148 : else
1149 60 : status = ListenServerPort(AF_UNSPEC, curhost,
1150 60 : (unsigned short) PostPortNumber,
1151 : NULL,
1152 : ListenSockets,
1153 : &NumListenSockets,
1154 : MAXLISTEN);
1155 :
1156 60 : if (status == STATUS_OK)
1157 : {
1158 60 : success++;
1159 : /* record the first successful host addr in lockfile */
1160 60 : if (!listen_addr_saved)
1161 : {
1162 60 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, curhost);
1163 60 : listen_addr_saved = true;
1164 : }
1165 : }
1166 : else
1167 0 : ereport(WARNING,
1168 : (errmsg("could not create listen socket for \"%s\"",
1169 : curhost)));
1170 : }
1171 :
1172 1758 : if (!success && elemlist != NIL)
1173 0 : ereport(FATAL,
1174 : (errmsg("could not create any TCP/IP sockets")));
1175 :
1176 1758 : list_free(elemlist);
1177 1758 : pfree(rawstring);
1178 : }
1179 :
1180 : #ifdef USE_BONJOUR
1181 : /* Register for Bonjour only if we opened TCP socket(s) */
1182 : if (enable_bonjour && NumListenSockets > 0)
1183 : {
1184 : DNSServiceErrorType err;
1185 :
1186 : /*
1187 : * We pass 0 for interface_index, which will result in registering on
1188 : * all "applicable" interfaces. It's not entirely clear from the
1189 : * DNS-SD docs whether this would be appropriate if we have bound to
1190 : * just a subset of the available network interfaces.
1191 : */
1192 : err = DNSServiceRegister(&bonjour_sdref,
1193 : 0,
1194 : 0,
1195 : bonjour_name,
1196 : "_postgresql._tcp.",
1197 : NULL,
1198 : NULL,
1199 : pg_hton16(PostPortNumber),
1200 : 0,
1201 : NULL,
1202 : NULL,
1203 : NULL);
1204 : if (err != kDNSServiceErr_NoError)
1205 : ereport(LOG,
1206 : (errmsg("DNSServiceRegister() failed: error code %ld",
1207 : (long) err)));
1208 :
1209 : /*
1210 : * We don't bother to read the mDNS daemon's reply, and we expect that
1211 : * it will automatically terminate our registration when the socket is
1212 : * closed at postmaster termination. So there's nothing more to be
1213 : * done here. However, the bonjour_sdref is kept around so that
1214 : * forked children can close their copies of the socket.
1215 : */
1216 : }
1217 : #endif
1218 :
1219 1758 : if (Unix_socket_directories)
1220 : {
1221 : char *rawstring;
1222 : List *elemlist;
1223 : ListCell *l;
1224 1758 : int success = 0;
1225 :
1226 : /* Need a modifiable copy of Unix_socket_directories */
1227 1758 : rawstring = pstrdup(Unix_socket_directories);
1228 :
1229 : /* Parse string into list of directories */
1230 1758 : if (!SplitDirectoriesString(rawstring, ',', &elemlist))
1231 : {
1232 : /* syntax error in list */
1233 0 : ereport(FATAL,
1234 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1235 : errmsg("invalid list syntax in parameter \"%s\"",
1236 : "unix_socket_directories")));
1237 : }
1238 :
1239 3514 : foreach(l, elemlist)
1240 : {
1241 1756 : char *socketdir = (char *) lfirst(l);
1242 :
1243 1756 : status = ListenServerPort(AF_UNIX, NULL,
1244 1756 : (unsigned short) PostPortNumber,
1245 : socketdir,
1246 : ListenSockets,
1247 : &NumListenSockets,
1248 : MAXLISTEN);
1249 :
1250 1756 : if (status == STATUS_OK)
1251 : {
1252 1756 : success++;
1253 : /* record the first successful Unix socket in lockfile */
1254 1756 : if (success == 1)
1255 1756 : AddToDataDirLockFile(LOCK_FILE_LINE_SOCKET_DIR, socketdir);
1256 : }
1257 : else
1258 0 : ereport(WARNING,
1259 : (errmsg("could not create Unix-domain socket in directory \"%s\"",
1260 : socketdir)));
1261 : }
1262 :
1263 1758 : if (!success && elemlist != NIL)
1264 0 : ereport(FATAL,
1265 : (errmsg("could not create any Unix-domain sockets")));
1266 :
1267 1758 : list_free_deep(elemlist);
1268 1758 : pfree(rawstring);
1269 : }
1270 :
1271 : /*
1272 : * check that we have some socket to listen on
1273 : */
1274 1758 : if (NumListenSockets == 0)
1275 0 : ereport(FATAL,
1276 : (errmsg("no socket created for listening")));
1277 :
1278 : /*
1279 : * If no valid TCP ports, write an empty line for listen address,
1280 : * indicating the Unix socket must be used. Note that this line is not
1281 : * added to the lock file until there is a socket backing it.
1282 : */
1283 1758 : if (!listen_addr_saved)
1284 1698 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, "");
1285 :
1286 : /*
1287 : * Record postmaster options. We delay this till now to avoid recording
1288 : * bogus options (eg, unusable port number).
1289 : */
1290 1758 : if (!CreateOptsFile(argc, argv, my_exec_path))
1291 0 : ExitPostmaster(1);
1292 :
1293 : /*
1294 : * Write the external PID file if requested
1295 : */
1296 1758 : if (external_pid_file)
1297 : {
1298 0 : FILE *fpidfile = fopen(external_pid_file, "w");
1299 :
1300 0 : if (fpidfile)
1301 : {
1302 0 : fprintf(fpidfile, "%d\n", MyProcPid);
1303 0 : fclose(fpidfile);
1304 :
1305 : /* Make PID file world readable */
1306 0 : if (chmod(external_pid_file, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH) != 0)
1307 0 : write_stderr("%s: could not change permissions of external PID file \"%s\": %m\n",
1308 : progname, external_pid_file);
1309 : }
1310 : else
1311 0 : write_stderr("%s: could not write external PID file \"%s\": %m\n",
1312 : progname, external_pid_file);
1313 :
1314 0 : on_proc_exit(unlink_external_pid_file, 0);
1315 : }
1316 :
1317 : /*
1318 : * Remove old temporary files. At this point there can be no other
1319 : * Postgres processes running in this directory, so this should be safe.
1320 : */
1321 1758 : RemovePgTempFiles();
1322 :
1323 : /*
1324 : * Initialize the autovacuum subsystem (again, no process start yet)
1325 : */
1326 1758 : autovac_init();
1327 :
1328 : /*
1329 : * Load configuration files for client authentication.
1330 : */
1331 1758 : if (!load_hba())
1332 : {
1333 : /*
1334 : * It makes no sense to continue if we fail to load the HBA file,
1335 : * since there is no way to connect to the database in this case.
1336 : */
1337 0 : ereport(FATAL,
1338 : /* translator: %s is a configuration file */
1339 : (errmsg("could not load %s", HbaFileName)));
1340 : }
1341 1758 : if (!load_ident())
1342 : {
1343 : /*
1344 : * We can start up without the IDENT file, although it means that you
1345 : * cannot log in using any of the authentication methods that need a
1346 : * user name mapping. load_ident() already logged the details of error
1347 : * to the log.
1348 : */
1349 : }
1350 :
1351 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1352 :
1353 : /*
1354 : * On macOS, libintl replaces setlocale() with a version that calls
1355 : * CFLocaleCopyCurrent() when its second argument is "" and every relevant
1356 : * environment variable is unset or empty. CFLocaleCopyCurrent() makes
1357 : * the process multithreaded. The postmaster calls sigprocmask() and
1358 : * calls fork() without an immediate exec(), both of which have undefined
1359 : * behavior in a multithreaded program. A multithreaded postmaster is the
1360 : * normal case on Windows, which offers neither fork() nor sigprocmask().
1361 : * Currently, macOS is the only platform having pthread_is_threaded_np(),
1362 : * so we need not worry whether this HINT is appropriate elsewhere.
1363 : */
1364 : if (pthread_is_threaded_np() != 0)
1365 : ereport(FATAL,
1366 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1367 : errmsg("postmaster became multithreaded during startup"),
1368 : errhint("Set the LC_ALL environment variable to a valid locale.")));
1369 : #endif
1370 :
1371 : /*
1372 : * Remember postmaster startup time
1373 : */
1374 1758 : PgStartTime = GetCurrentTimestamp();
1375 :
1376 : /*
1377 : * Report postmaster status in the postmaster.pid file, to allow pg_ctl to
1378 : * see what's happening.
1379 : */
1380 1758 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STARTING);
1381 :
1382 1758 : UpdatePMState(PM_STARTUP);
1383 :
1384 : /* Make sure we can perform I/O while starting up. */
1385 1758 : maybe_adjust_io_workers();
1386 :
1387 : /* Start bgwriter and checkpointer so they can help with recovery */
1388 1758 : if (CheckpointerPMChild == NULL)
1389 1758 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
1390 1758 : if (BgWriterPMChild == NULL)
1391 1758 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
1392 :
1393 : /*
1394 : * We're ready to rock and roll...
1395 : */
1396 1758 : StartupPMChild = StartChildProcess(B_STARTUP);
1397 : Assert(StartupPMChild != NULL);
1398 1758 : StartupStatus = STARTUP_RUNNING;
1399 :
1400 : /* Some workers may be scheduled to start now */
1401 1758 : maybe_start_bgworkers();
1402 :
1403 1758 : status = ServerLoop();
1404 :
1405 : /*
1406 : * ServerLoop probably shouldn't ever return, but if it does, close down.
1407 : */
1408 0 : ExitPostmaster(status != STATUS_OK);
1409 :
1410 : abort(); /* not reached */
1411 : }
1412 :
1413 :
1414 : /*
1415 : * on_proc_exit callback to close server's listen sockets
1416 : */
1417 : static void
1418 1758 : CloseServerPorts(int status, Datum arg)
1419 : {
1420 : int i;
1421 :
1422 : /*
1423 : * First, explicitly close all the socket FDs. We used to just let this
1424 : * happen implicitly at postmaster exit, but it's better to close them
1425 : * before we remove the postmaster.pid lockfile; otherwise there's a race
1426 : * condition if a new postmaster wants to re-use the TCP port number.
1427 : */
1428 3576 : for (i = 0; i < NumListenSockets; i++)
1429 : {
1430 1818 : if (closesocket(ListenSockets[i]) != 0)
1431 0 : elog(LOG, "could not close listen socket: %m");
1432 : }
1433 1758 : NumListenSockets = 0;
1434 :
1435 : /*
1436 : * Next, remove any filesystem entries for Unix sockets. To avoid race
1437 : * conditions against incoming postmasters, this must happen after closing
1438 : * the sockets and before removing lock files.
1439 : */
1440 1758 : RemoveSocketFiles();
1441 :
1442 : /*
1443 : * We don't do anything about socket lock files here; those will be
1444 : * removed in a later on_proc_exit callback.
1445 : */
1446 1758 : }
1447 :
1448 : /*
1449 : * on_proc_exit callback to delete external_pid_file
1450 : */
1451 : static void
1452 0 : unlink_external_pid_file(int status, Datum arg)
1453 : {
1454 0 : if (external_pid_file)
1455 0 : unlink(external_pid_file);
1456 0 : }
1457 :
1458 :
1459 : /*
1460 : * Compute and check the directory paths to files that are part of the
1461 : * installation (as deduced from the postgres executable's own location)
1462 : */
1463 : static void
1464 1792 : getInstallationPaths(const char *argv0)
1465 : {
1466 : DIR *pdir;
1467 :
1468 : /* Locate the postgres executable itself */
1469 1792 : if (find_my_exec(argv0, my_exec_path) < 0)
1470 0 : ereport(FATAL,
1471 : (errmsg("%s: could not locate my own executable path", argv0)));
1472 :
1473 : #ifdef EXEC_BACKEND
1474 : /* Locate executable backend before we change working directory */
1475 : if (find_other_exec(argv0, "postgres", PG_BACKEND_VERSIONSTR,
1476 : postgres_exec_path) < 0)
1477 : ereport(FATAL,
1478 : (errmsg("%s: could not locate matching postgres executable",
1479 : argv0)));
1480 : #endif
1481 :
1482 : /*
1483 : * Locate the pkglib directory --- this has to be set early in case we try
1484 : * to load any modules from it in response to postgresql.conf entries.
1485 : */
1486 1792 : get_pkglib_path(my_exec_path, pkglib_path);
1487 :
1488 : /*
1489 : * Verify that there's a readable directory there; otherwise the Postgres
1490 : * installation is incomplete or corrupt. (A typical cause of this
1491 : * failure is that the postgres executable has been moved or hardlinked to
1492 : * some directory that's not a sibling of the installation lib/
1493 : * directory.)
1494 : */
1495 1792 : pdir = AllocateDir(pkglib_path);
1496 1792 : if (pdir == NULL)
1497 0 : ereport(ERROR,
1498 : (errcode_for_file_access(),
1499 : errmsg("could not open directory \"%s\": %m",
1500 : pkglib_path),
1501 : errhint("This may indicate an incomplete PostgreSQL installation, or that the file \"%s\" has been moved away from its proper location.",
1502 : my_exec_path)));
1503 1792 : FreeDir(pdir);
1504 :
1505 : /*
1506 : * It's not worth checking the share/ directory. If the lib/ directory is
1507 : * there, then share/ probably is too.
1508 : */
1509 1792 : }
1510 :
1511 : /*
1512 : * Check that pg_control exists in the correct location in the data directory.
1513 : *
1514 : * No attempt is made to validate the contents of pg_control here. This is
1515 : * just a sanity check to see if we are looking at a real data directory.
1516 : */
1517 : static void
1518 1778 : checkControlFile(void)
1519 : {
1520 : char path[MAXPGPATH];
1521 : FILE *fp;
1522 :
1523 1778 : snprintf(path, sizeof(path), "%s/%s", DataDir, XLOG_CONTROL_FILE);
1524 :
1525 1778 : fp = AllocateFile(path, PG_BINARY_R);
1526 1778 : if (fp == NULL)
1527 : {
1528 0 : write_stderr("%s: could not find the database system\n"
1529 : "Expected to find it in the directory \"%s\",\n"
1530 : "but could not open file \"%s\": %m\n",
1531 : progname, DataDir, path);
1532 0 : ExitPostmaster(2);
1533 : }
1534 1778 : FreeFile(fp);
1535 1778 : }
1536 :
1537 : /*
1538 : * Determine how long should we let ServerLoop sleep, in milliseconds.
1539 : *
1540 : * In normal conditions we wait at most one minute, to ensure that the other
1541 : * background tasks handled by ServerLoop get done even when no requests are
1542 : * arriving. However, if there are background workers waiting to be started,
1543 : * we don't actually sleep so that they are quickly serviced. Other exception
1544 : * cases are as shown in the code.
1545 : */
1546 : static int
1547 242476 : DetermineSleepTime(void)
1548 : {
1549 242476 : TimestampTz next_wakeup = 0;
1550 :
1551 : /*
1552 : * Normal case: either there are no background workers at all, or we're in
1553 : * a shutdown sequence (during which we ignore bgworkers altogether).
1554 : */
1555 242476 : if (Shutdown > NoShutdown ||
1556 228930 : (!StartWorkerNeeded && !HaveCrashedWorker))
1557 : {
1558 242476 : if (AbortStartTime != 0)
1559 : {
1560 2506 : time_t curtime = time(NULL);
1561 : int seconds;
1562 :
1563 : /*
1564 : * time left to abort; clamp to 0 if it already expired, or if
1565 : * time goes backwards
1566 : */
1567 2506 : if (curtime < AbortStartTime ||
1568 2506 : curtime - AbortStartTime >= SIGKILL_CHILDREN_AFTER_SECS)
1569 0 : seconds = 0;
1570 : else
1571 2506 : seconds = SIGKILL_CHILDREN_AFTER_SECS -
1572 : (curtime - AbortStartTime);
1573 :
1574 2506 : return seconds * 1000;
1575 : }
1576 : else
1577 239970 : return 60 * 1000;
1578 : }
1579 :
1580 0 : if (StartWorkerNeeded)
1581 0 : return 0;
1582 :
1583 0 : if (HaveCrashedWorker)
1584 : {
1585 : dlist_mutable_iter iter;
1586 :
1587 : /*
1588 : * When there are crashed bgworkers, we sleep just long enough that
1589 : * they are restarted when they request to be. Scan the list to
1590 : * determine the minimum of all wakeup times according to most recent
1591 : * crash time and requested restart interval.
1592 : */
1593 0 : dlist_foreach_modify(iter, &BackgroundWorkerList)
1594 : {
1595 : RegisteredBgWorker *rw;
1596 : TimestampTz this_wakeup;
1597 :
1598 0 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
1599 :
1600 0 : if (rw->rw_crashed_at == 0)
1601 0 : continue;
1602 :
1603 0 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART
1604 0 : || rw->rw_terminate)
1605 : {
1606 0 : ForgetBackgroundWorker(rw);
1607 0 : continue;
1608 : }
1609 :
1610 0 : this_wakeup = TimestampTzPlusMilliseconds(rw->rw_crashed_at,
1611 : 1000L * rw->rw_worker.bgw_restart_time);
1612 0 : if (next_wakeup == 0 || this_wakeup < next_wakeup)
1613 0 : next_wakeup = this_wakeup;
1614 : }
1615 : }
1616 :
1617 0 : if (next_wakeup != 0)
1618 : {
1619 : int ms;
1620 :
1621 : /* result of TimestampDifferenceMilliseconds is in [0, INT_MAX] */
1622 0 : ms = (int) TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
1623 : next_wakeup);
1624 0 : return Min(60 * 1000, ms);
1625 : }
1626 :
1627 0 : return 60 * 1000;
1628 : }
1629 :
1630 : /*
1631 : * Activate or deactivate notifications of server socket events. Since we
1632 : * don't currently have a way to remove events from an existing WaitEventSet,
1633 : * we'll just destroy and recreate the whole thing. This is called during
1634 : * shutdown so we can wait for backends to exit without accepting new
1635 : * connections, and during crash reinitialization when we need to start
1636 : * listening for new connections again. The WaitEventSet will be freed in fork
1637 : * children by ClosePostmasterPorts().
1638 : */
1639 : static void
1640 3534 : ConfigurePostmasterWaitSet(bool accept_connections)
1641 : {
1642 3534 : if (pm_wait_set)
1643 1776 : FreeWaitEventSet(pm_wait_set);
1644 3534 : pm_wait_set = NULL;
1645 :
1646 5302 : pm_wait_set = CreateWaitEventSet(NULL,
1647 1768 : accept_connections ? (1 + NumListenSockets) : 1);
1648 3534 : AddWaitEventToSet(pm_wait_set, WL_LATCH_SET, PGINVALID_SOCKET, MyLatch,
1649 : NULL);
1650 :
1651 3534 : if (accept_connections)
1652 : {
1653 3596 : for (int i = 0; i < NumListenSockets; i++)
1654 1828 : AddWaitEventToSet(pm_wait_set, WL_SOCKET_ACCEPT, ListenSockets[i],
1655 : NULL, NULL);
1656 : }
1657 3534 : }
1658 :
1659 : /*
1660 : * Main idle loop of postmaster
1661 : */
1662 : static int
1663 1758 : ServerLoop(void)
1664 : {
1665 : time_t last_lockfile_recheck_time,
1666 : last_touch_time;
1667 : WaitEvent events[MAXLISTEN];
1668 : int nevents;
1669 :
1670 1758 : ConfigurePostmasterWaitSet(true);
1671 1758 : last_lockfile_recheck_time = last_touch_time = time(NULL);
1672 :
1673 : for (;;)
1674 240718 : {
1675 : time_t now;
1676 :
1677 242476 : nevents = WaitEventSetWait(pm_wait_set,
1678 242476 : DetermineSleepTime(),
1679 : events,
1680 : lengthof(events),
1681 : 0 /* postmaster posts no wait_events */ );
1682 :
1683 : /*
1684 : * Latch set by signal handler, or new connection pending on any of
1685 : * our sockets? If the latter, fork a child process to deal with it.
1686 : */
1687 483194 : for (int i = 0; i < nevents; i++)
1688 : {
1689 242476 : if (events[i].events & WL_LATCH_SET)
1690 214544 : ResetLatch(MyLatch);
1691 :
1692 : /*
1693 : * The following requests are handled unconditionally, even if we
1694 : * didn't see WL_LATCH_SET. This gives high priority to shutdown
1695 : * and reload requests where the latch happens to appear later in
1696 : * events[] or will be reported by a later call to
1697 : * WaitEventSetWait().
1698 : */
1699 242476 : if (pending_pm_shutdown_request)
1700 1750 : process_pm_shutdown_request();
1701 242476 : if (pending_pm_reload_request)
1702 280 : process_pm_reload_request();
1703 242476 : if (pending_pm_child_exit)
1704 47676 : process_pm_child_exit();
1705 240718 : if (pending_pm_pmsignal)
1706 164894 : process_pm_pmsignal();
1707 :
1708 240718 : if (events[i].events & WL_SOCKET_ACCEPT)
1709 : {
1710 : ClientSocket s;
1711 :
1712 27932 : if (AcceptConnection(events[i].fd, &s) == STATUS_OK)
1713 27932 : BackendStartup(&s);
1714 :
1715 : /* We no longer need the open socket in this process */
1716 27932 : if (s.sock != PGINVALID_SOCKET)
1717 : {
1718 27932 : if (closesocket(s.sock) != 0)
1719 0 : elog(LOG, "could not close client socket: %m");
1720 : }
1721 : }
1722 : }
1723 :
1724 : /*
1725 : * If we need to launch any background processes after changing state
1726 : * or because some exited, do so now.
1727 : */
1728 240718 : LaunchMissingBackgroundProcesses();
1729 :
1730 : /* If we need to signal the autovacuum launcher, do so now */
1731 240718 : if (avlauncher_needs_signal)
1732 : {
1733 0 : avlauncher_needs_signal = false;
1734 0 : if (AutoVacLauncherPMChild != NULL)
1735 0 : signal_child(AutoVacLauncherPMChild, SIGUSR2);
1736 : }
1737 :
1738 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1739 :
1740 : /*
1741 : * With assertions enabled, check regularly for appearance of
1742 : * additional threads. All builds check at start and exit.
1743 : */
1744 : Assert(pthread_is_threaded_np() == 0);
1745 : #endif
1746 :
1747 : /*
1748 : * Lastly, check to see if it's time to do some things that we don't
1749 : * want to do every single time through the loop, because they're a
1750 : * bit expensive. Note that there's up to a minute of slop in when
1751 : * these tasks will be performed, since DetermineSleepTime() will let
1752 : * us sleep at most that long; except for SIGKILL timeout which has
1753 : * special-case logic there.
1754 : */
1755 240718 : now = time(NULL);
1756 :
1757 : /*
1758 : * If we already sent SIGQUIT to children and they are slow to shut
1759 : * down, it's time to send them SIGKILL (or SIGABRT if requested).
1760 : * This doesn't happen normally, but under certain conditions backends
1761 : * can get stuck while shutting down. This is a last measure to get
1762 : * them unwedged.
1763 : *
1764 : * Note we also do this during recovery from a process crash.
1765 : */
1766 240718 : if ((Shutdown >= ImmediateShutdown || FatalError) &&
1767 2522 : AbortStartTime != 0 &&
1768 2506 : (now - AbortStartTime) >= SIGKILL_CHILDREN_AFTER_SECS)
1769 : {
1770 : /* We were gentle with them before. Not anymore */
1771 0 : ereport(LOG,
1772 : /* translator: %s is SIGKILL or SIGABRT */
1773 : (errmsg("issuing %s to recalcitrant children",
1774 : send_abort_for_kill ? "SIGABRT" : "SIGKILL")));
1775 0 : TerminateChildren(send_abort_for_kill ? SIGABRT : SIGKILL);
1776 : /* reset flag so we don't SIGKILL again */
1777 0 : AbortStartTime = 0;
1778 : }
1779 :
1780 : /*
1781 : * Once a minute, verify that postmaster.pid hasn't been removed or
1782 : * overwritten. If it has, we force a shutdown. This avoids having
1783 : * postmasters and child processes hanging around after their database
1784 : * is gone, and maybe causing problems if a new database cluster is
1785 : * created in the same place. It also provides some protection
1786 : * against a DBA foolishly removing postmaster.pid and manually
1787 : * starting a new postmaster. Data corruption is likely to ensue from
1788 : * that anyway, but we can minimize the damage by aborting ASAP.
1789 : */
1790 240718 : if (now - last_lockfile_recheck_time >= 1 * SECS_PER_MINUTE)
1791 : {
1792 66 : if (!RecheckDataDirLockFile())
1793 : {
1794 0 : ereport(LOG,
1795 : (errmsg("performing immediate shutdown because data directory lock file is invalid")));
1796 0 : kill(MyProcPid, SIGQUIT);
1797 : }
1798 66 : last_lockfile_recheck_time = now;
1799 : }
1800 :
1801 : /*
1802 : * Touch Unix socket and lock files every 58 minutes, to ensure that
1803 : * they are not removed by overzealous /tmp-cleaning tasks. We assume
1804 : * no one runs cleaners with cutoff times of less than an hour ...
1805 : */
1806 240718 : if (now - last_touch_time >= 58 * SECS_PER_MINUTE)
1807 : {
1808 0 : TouchSocketFiles();
1809 0 : TouchSocketLockFiles();
1810 0 : last_touch_time = now;
1811 : }
1812 : }
1813 : }
1814 :
1815 : /*
1816 : * canAcceptConnections --- check to see if database state allows connections
1817 : * of the specified type. backend_type can be B_BACKEND or B_AUTOVAC_WORKER.
1818 : * (Note that we don't yet know whether a normal B_BACKEND connection might
1819 : * turn into a walsender.)
1820 : */
1821 : static CAC_state
1822 32192 : canAcceptConnections(BackendType backend_type)
1823 : {
1824 32192 : CAC_state result = CAC_OK;
1825 :
1826 : Assert(backend_type == B_BACKEND || backend_type == B_AUTOVAC_WORKER);
1827 :
1828 : /*
1829 : * Can't start backends when in startup/shutdown/inconsistent recovery
1830 : * state. We treat autovac workers the same as user backends for this
1831 : * purpose.
1832 : */
1833 32192 : if (pmState != PM_RUN && pmState != PM_HOT_STANDBY)
1834 : {
1835 678 : if (Shutdown > NoShutdown)
1836 96 : return CAC_SHUTDOWN; /* shutdown is pending */
1837 582 : else if (!FatalError && pmState == PM_STARTUP)
1838 548 : return CAC_STARTUP; /* normal startup */
1839 34 : else if (!FatalError && pmState == PM_RECOVERY)
1840 28 : return CAC_NOTHOTSTANDBY; /* not yet ready for hot standby */
1841 : else
1842 6 : return CAC_RECOVERY; /* else must be crash recovery */
1843 : }
1844 :
1845 : /*
1846 : * "Smart shutdown" restrictions are applied only to normal connections,
1847 : * not to autovac workers.
1848 : */
1849 31514 : if (!connsAllowed && backend_type == B_BACKEND)
1850 0 : return CAC_SHUTDOWN; /* shutdown is pending */
1851 :
1852 31514 : return result;
1853 : }
1854 :
1855 : /*
1856 : * ClosePostmasterPorts -- close all the postmaster's open sockets
1857 : *
1858 : * This is called during child process startup to release file descriptors
1859 : * that are not needed by that child process. The postmaster still has
1860 : * them open, of course.
1861 : *
1862 : * Note: we pass am_syslogger as a boolean because we don't want to set
1863 : * the global variable yet when this is called.
1864 : */
1865 : void
1866 45866 : ClosePostmasterPorts(bool am_syslogger)
1867 : {
1868 : /* Release resources held by the postmaster's WaitEventSet. */
1869 45866 : if (pm_wait_set)
1870 : {
1871 38794 : FreeWaitEventSetAfterFork(pm_wait_set);
1872 38794 : pm_wait_set = NULL;
1873 : }
1874 :
1875 : #ifndef WIN32
1876 :
1877 : /*
1878 : * Close the write end of postmaster death watch pipe. It's important to
1879 : * do this as early as possible, so that if postmaster dies, others won't
1880 : * think that it's still running because we're holding the pipe open.
1881 : */
1882 45866 : if (close(postmaster_alive_fds[POSTMASTER_FD_OWN]) != 0)
1883 0 : ereport(FATAL,
1884 : (errcode_for_file_access(),
1885 : errmsg_internal("could not close postmaster death monitoring pipe in child process: %m")));
1886 45866 : postmaster_alive_fds[POSTMASTER_FD_OWN] = -1;
1887 : /* Notify fd.c that we released one pipe FD. */
1888 45866 : ReleaseExternalFD();
1889 : #endif
1890 :
1891 : /*
1892 : * Close the postmaster's listen sockets. These aren't tracked by fd.c,
1893 : * so we don't call ReleaseExternalFD() here.
1894 : *
1895 : * The listen sockets are marked as FD_CLOEXEC, so this isn't needed in
1896 : * EXEC_BACKEND mode.
1897 : */
1898 : #ifndef EXEC_BACKEND
1899 45866 : if (ListenSockets)
1900 : {
1901 92816 : for (int i = 0; i < NumListenSockets; i++)
1902 : {
1903 46952 : if (closesocket(ListenSockets[i]) != 0)
1904 0 : elog(LOG, "could not close listen socket: %m");
1905 : }
1906 45864 : pfree(ListenSockets);
1907 : }
1908 45866 : NumListenSockets = 0;
1909 45866 : ListenSockets = NULL;
1910 : #endif
1911 :
1912 : /*
1913 : * If using syslogger, close the read side of the pipe. We don't bother
1914 : * tracking this in fd.c, either.
1915 : */
1916 45866 : if (!am_syslogger)
1917 : {
1918 : #ifndef WIN32
1919 45864 : if (syslogPipe[0] >= 0)
1920 34 : close(syslogPipe[0]);
1921 45864 : syslogPipe[0] = -1;
1922 : #else
1923 : if (syslogPipe[0])
1924 : CloseHandle(syslogPipe[0]);
1925 : syslogPipe[0] = 0;
1926 : #endif
1927 : }
1928 :
1929 : #ifdef USE_BONJOUR
1930 : /* If using Bonjour, close the connection to the mDNS daemon */
1931 : if (bonjour_sdref)
1932 : close(DNSServiceRefSockFD(bonjour_sdref));
1933 : #endif
1934 45866 : }
1935 :
1936 :
1937 : /*
1938 : * InitProcessGlobals -- set MyStartTime[stamp], random seeds
1939 : *
1940 : * Called early in the postmaster and every backend.
1941 : */
1942 : void
1943 48150 : InitProcessGlobals(void)
1944 : {
1945 48150 : MyStartTimestamp = GetCurrentTimestamp();
1946 48150 : MyStartTime = timestamptz_to_time_t(MyStartTimestamp);
1947 :
1948 : /*
1949 : * Set a different global seed in every process. We want something
1950 : * unpredictable, so if possible, use high-quality random bits for the
1951 : * seed. Otherwise, fall back to a seed based on timestamp and PID.
1952 : */
1953 48150 : if (unlikely(!pg_prng_strong_seed(&pg_global_prng_state)))
1954 : {
1955 : uint64 rseed;
1956 :
1957 : /*
1958 : * Since PIDs and timestamps tend to change more frequently in their
1959 : * least significant bits, shift the timestamp left to allow a larger
1960 : * total number of seeds in a given time period. Since that would
1961 : * leave only 20 bits of the timestamp that cycle every ~1 second,
1962 : * also mix in some higher bits.
1963 : */
1964 0 : rseed = ((uint64) MyProcPid) ^
1965 0 : ((uint64) MyStartTimestamp << 12) ^
1966 0 : ((uint64) MyStartTimestamp >> 20);
1967 :
1968 0 : pg_prng_seed(&pg_global_prng_state, rseed);
1969 : }
1970 :
1971 : /*
1972 : * Also make sure that we've set a good seed for random(3). Use of that
1973 : * is deprecated in core Postgres, but extensions might use it.
1974 : */
1975 : #ifndef WIN32
1976 48150 : srandom(pg_prng_uint32(&pg_global_prng_state));
1977 : #endif
1978 48150 : }
1979 :
1980 : /*
1981 : * Child processes use SIGUSR1 to notify us of 'pmsignals'. pg_ctl uses
1982 : * SIGUSR1 to ask postmaster to check for logrotate and promote files.
1983 : */
1984 : static void
1985 165544 : handle_pm_pmsignal_signal(SIGNAL_ARGS)
1986 : {
1987 165544 : pending_pm_pmsignal = true;
1988 165544 : SetLatch(MyLatch);
1989 165544 : }
1990 :
1991 : /*
1992 : * pg_ctl uses SIGHUP to request a reload of the configuration files.
1993 : */
1994 : static void
1995 280 : handle_pm_reload_request_signal(SIGNAL_ARGS)
1996 : {
1997 280 : pending_pm_reload_request = true;
1998 280 : SetLatch(MyLatch);
1999 280 : }
2000 :
2001 : /*
2002 : * Re-read config files, and tell children to do same.
2003 : */
2004 : static void
2005 280 : process_pm_reload_request(void)
2006 : {
2007 280 : pending_pm_reload_request = false;
2008 :
2009 280 : ereport(DEBUG2,
2010 : (errmsg_internal("postmaster received reload request signal")));
2011 :
2012 280 : if (Shutdown <= SmartShutdown)
2013 : {
2014 280 : ereport(LOG,
2015 : (errmsg("received SIGHUP, reloading configuration files")));
2016 280 : ProcessConfigFile(PGC_SIGHUP);
2017 280 : SignalChildren(SIGHUP, btmask_all_except(B_DEAD_END_BACKEND));
2018 :
2019 : /* Reload authentication config files too */
2020 280 : if (!load_hba())
2021 0 : ereport(LOG,
2022 : /* translator: %s is a configuration file */
2023 : (errmsg("%s was not reloaded", HbaFileName)));
2024 :
2025 280 : if (!load_ident())
2026 0 : ereport(LOG,
2027 : (errmsg("%s was not reloaded", IdentFileName)));
2028 :
2029 : #ifdef USE_SSL
2030 : /* Reload SSL configuration as well */
2031 280 : if (EnableSSL)
2032 : {
2033 6 : if (secure_initialize(false) == 0)
2034 4 : LoadedSSL = true;
2035 : else
2036 2 : ereport(LOG,
2037 : (errmsg("SSL configuration was not reloaded")));
2038 : }
2039 : else
2040 : {
2041 274 : secure_destroy();
2042 274 : LoadedSSL = false;
2043 : }
2044 : #endif
2045 :
2046 : #ifdef EXEC_BACKEND
2047 : /* Update the starting-point file for future children */
2048 : write_nondefault_variables(PGC_SIGHUP);
2049 : #endif
2050 : }
2051 280 : }
2052 :
2053 : /*
2054 : * pg_ctl uses SIGTERM, SIGINT and SIGQUIT to request different types of
2055 : * shutdown.
2056 : */
2057 : static void
2058 1750 : handle_pm_shutdown_request_signal(SIGNAL_ARGS)
2059 : {
2060 1750 : switch (postgres_signal_arg)
2061 : {
2062 80 : case SIGTERM:
2063 : /* smart is implied if the other two flags aren't set */
2064 80 : pending_pm_shutdown_request = true;
2065 80 : break;
2066 996 : case SIGINT:
2067 996 : pending_pm_fast_shutdown_request = true;
2068 996 : pending_pm_shutdown_request = true;
2069 996 : break;
2070 674 : case SIGQUIT:
2071 674 : pending_pm_immediate_shutdown_request = true;
2072 674 : pending_pm_shutdown_request = true;
2073 674 : break;
2074 : }
2075 1750 : SetLatch(MyLatch);
2076 1750 : }
2077 :
2078 : /*
2079 : * Process shutdown request.
2080 : */
2081 : static void
2082 1750 : process_pm_shutdown_request(void)
2083 : {
2084 : int mode;
2085 :
2086 1750 : ereport(DEBUG2,
2087 : (errmsg_internal("postmaster received shutdown request signal")));
2088 :
2089 1750 : pending_pm_shutdown_request = false;
2090 :
2091 : /*
2092 : * If more than one shutdown request signal arrived since the last server
2093 : * loop, take the one that is the most immediate. That matches the
2094 : * priority that would apply if we processed them one by one in any order.
2095 : */
2096 1750 : if (pending_pm_immediate_shutdown_request)
2097 : {
2098 674 : pending_pm_immediate_shutdown_request = false;
2099 674 : pending_pm_fast_shutdown_request = false;
2100 674 : mode = ImmediateShutdown;
2101 : }
2102 1076 : else if (pending_pm_fast_shutdown_request)
2103 : {
2104 996 : pending_pm_fast_shutdown_request = false;
2105 996 : mode = FastShutdown;
2106 : }
2107 : else
2108 80 : mode = SmartShutdown;
2109 :
2110 1750 : switch (mode)
2111 : {
2112 80 : case SmartShutdown:
2113 :
2114 : /*
2115 : * Smart Shutdown:
2116 : *
2117 : * Wait for children to end their work, then shut down.
2118 : */
2119 80 : if (Shutdown >= SmartShutdown)
2120 0 : break;
2121 80 : Shutdown = SmartShutdown;
2122 80 : ereport(LOG,
2123 : (errmsg("received smart shutdown request")));
2124 :
2125 : /* Report status */
2126 80 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2127 : #ifdef USE_SYSTEMD
2128 : sd_notify(0, "STOPPING=1");
2129 : #endif
2130 :
2131 : /*
2132 : * If we reached normal running, we go straight to waiting for
2133 : * client backends to exit. If already in PM_STOP_BACKENDS or a
2134 : * later state, do not change it.
2135 : */
2136 80 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2137 80 : connsAllowed = false;
2138 0 : else if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2139 : {
2140 : /* There should be no clients, so proceed to stop children */
2141 0 : UpdatePMState(PM_STOP_BACKENDS);
2142 : }
2143 :
2144 : /*
2145 : * Now wait for online backup mode to end and backends to exit. If
2146 : * that is already the case, PostmasterStateMachine will take the
2147 : * next step.
2148 : */
2149 80 : PostmasterStateMachine();
2150 80 : break;
2151 :
2152 996 : case FastShutdown:
2153 :
2154 : /*
2155 : * Fast Shutdown:
2156 : *
2157 : * Abort all children with SIGTERM (rollback active transactions
2158 : * and exit) and shut down when they are gone.
2159 : */
2160 996 : if (Shutdown >= FastShutdown)
2161 0 : break;
2162 996 : Shutdown = FastShutdown;
2163 996 : ereport(LOG,
2164 : (errmsg("received fast shutdown request")));
2165 :
2166 : /* Report status */
2167 996 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2168 : #ifdef USE_SYSTEMD
2169 : sd_notify(0, "STOPPING=1");
2170 : #endif
2171 :
2172 996 : if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2173 : {
2174 : /* Just shut down background processes silently */
2175 0 : UpdatePMState(PM_STOP_BACKENDS);
2176 : }
2177 996 : else if (pmState == PM_RUN ||
2178 116 : pmState == PM_HOT_STANDBY)
2179 : {
2180 : /* Report that we're about to zap live client sessions */
2181 996 : ereport(LOG,
2182 : (errmsg("aborting any active transactions")));
2183 996 : UpdatePMState(PM_STOP_BACKENDS);
2184 : }
2185 :
2186 : /*
2187 : * PostmasterStateMachine will issue any necessary signals, or
2188 : * take the next step if no child processes need to be killed.
2189 : */
2190 996 : PostmasterStateMachine();
2191 996 : break;
2192 :
2193 674 : case ImmediateShutdown:
2194 :
2195 : /*
2196 : * Immediate Shutdown:
2197 : *
2198 : * abort all children with SIGQUIT, wait for them to exit,
2199 : * terminate remaining ones with SIGKILL, then exit without
2200 : * attempt to properly shut down the data base system.
2201 : */
2202 674 : if (Shutdown >= ImmediateShutdown)
2203 0 : break;
2204 674 : Shutdown = ImmediateShutdown;
2205 674 : ereport(LOG,
2206 : (errmsg("received immediate shutdown request")));
2207 :
2208 : /* Report status */
2209 674 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2210 : #ifdef USE_SYSTEMD
2211 : sd_notify(0, "STOPPING=1");
2212 : #endif
2213 :
2214 : /* tell children to shut down ASAP */
2215 : /* (note we don't apply send_abort_for_crash here) */
2216 674 : SetQuitSignalReason(PMQUIT_FOR_STOP);
2217 674 : TerminateChildren(SIGQUIT);
2218 674 : UpdatePMState(PM_WAIT_BACKENDS);
2219 :
2220 : /* set stopwatch for them to die */
2221 674 : AbortStartTime = time(NULL);
2222 :
2223 : /*
2224 : * Now wait for backends to exit. If there are none,
2225 : * PostmasterStateMachine will take the next step.
2226 : */
2227 674 : PostmasterStateMachine();
2228 674 : break;
2229 : }
2230 1750 : }
2231 :
2232 : static void
2233 47768 : handle_pm_child_exit_signal(SIGNAL_ARGS)
2234 : {
2235 47768 : pending_pm_child_exit = true;
2236 47768 : SetLatch(MyLatch);
2237 47768 : }
2238 :
2239 : /*
2240 : * Cleanup after a child process dies.
2241 : */
2242 : static void
2243 47676 : process_pm_child_exit(void)
2244 : {
2245 : int pid; /* process id of dead child process */
2246 : int exitstatus; /* its exit status */
2247 :
2248 47676 : pending_pm_child_exit = false;
2249 :
2250 47676 : ereport(DEBUG4,
2251 : (errmsg_internal("reaping dead processes")));
2252 :
2253 99544 : while ((pid = waitpid(-1, &exitstatus, WNOHANG)) > 0)
2254 : {
2255 : PMChild *pmchild;
2256 :
2257 : /*
2258 : * Check if this child was a startup process.
2259 : */
2260 51870 : if (StartupPMChild && pid == StartupPMChild->pid)
2261 : {
2262 1768 : ReleasePostmasterChildSlot(StartupPMChild);
2263 1768 : StartupPMChild = NULL;
2264 :
2265 : /*
2266 : * Startup process exited in response to a shutdown request (or it
2267 : * completed normally regardless of the shutdown request).
2268 : */
2269 1768 : if (Shutdown > NoShutdown &&
2270 208 : (EXIT_STATUS_0(exitstatus) || EXIT_STATUS_1(exitstatus)))
2271 : {
2272 120 : StartupStatus = STARTUP_NOT_RUNNING;
2273 120 : UpdatePMState(PM_WAIT_BACKENDS);
2274 : /* PostmasterStateMachine logic does the rest */
2275 120 : continue;
2276 : }
2277 :
2278 1648 : if (EXIT_STATUS_3(exitstatus))
2279 : {
2280 0 : ereport(LOG,
2281 : (errmsg("shutdown at recovery target")));
2282 0 : StartupStatus = STARTUP_NOT_RUNNING;
2283 0 : Shutdown = Max(Shutdown, SmartShutdown);
2284 0 : TerminateChildren(SIGTERM);
2285 0 : UpdatePMState(PM_WAIT_BACKENDS);
2286 : /* PostmasterStateMachine logic does the rest */
2287 0 : continue;
2288 : }
2289 :
2290 : /*
2291 : * Unexpected exit of startup process (including FATAL exit)
2292 : * during PM_STARTUP is treated as catastrophic. There are no
2293 : * other processes running yet, so we can just exit.
2294 : */
2295 1648 : if (pmState == PM_STARTUP &&
2296 1266 : StartupStatus != STARTUP_SIGNALED &&
2297 1266 : !EXIT_STATUS_0(exitstatus))
2298 : {
2299 2 : LogChildExit(LOG, _("startup process"),
2300 : pid, exitstatus);
2301 2 : ereport(LOG,
2302 : (errmsg("aborting startup due to startup process failure")));
2303 2 : ExitPostmaster(1);
2304 : }
2305 :
2306 : /*
2307 : * After PM_STARTUP, any unexpected exit (including FATAL exit) of
2308 : * the startup process is catastrophic, so kill other children,
2309 : * and set StartupStatus so we don't try to reinitialize after
2310 : * they're gone. Exception: if StartupStatus is STARTUP_SIGNALED,
2311 : * then we previously sent the startup process a SIGQUIT; so
2312 : * that's probably the reason it died, and we do want to try to
2313 : * restart in that case.
2314 : *
2315 : * This stanza also handles the case where we sent a SIGQUIT
2316 : * during PM_STARTUP due to some dead-end child crashing: in that
2317 : * situation, if the startup process dies on the SIGQUIT, we need
2318 : * to transition to PM_WAIT_BACKENDS state which will allow
2319 : * PostmasterStateMachine to restart the startup process. (On the
2320 : * other hand, the startup process might complete normally, if we
2321 : * were too late with the SIGQUIT. In that case we'll fall
2322 : * through and commence normal operations.)
2323 : */
2324 1646 : if (!EXIT_STATUS_0(exitstatus))
2325 : {
2326 94 : if (StartupStatus == STARTUP_SIGNALED)
2327 : {
2328 88 : StartupStatus = STARTUP_NOT_RUNNING;
2329 88 : if (pmState == PM_STARTUP)
2330 0 : UpdatePMState(PM_WAIT_BACKENDS);
2331 : }
2332 : else
2333 6 : StartupStatus = STARTUP_CRASHED;
2334 94 : HandleChildCrash(pid, exitstatus,
2335 94 : _("startup process"));
2336 94 : continue;
2337 : }
2338 :
2339 : /*
2340 : * Startup succeeded, commence normal operations
2341 : */
2342 1552 : StartupStatus = STARTUP_NOT_RUNNING;
2343 1552 : FatalError = false;
2344 1552 : AbortStartTime = 0;
2345 1552 : ReachedNormalRunning = true;
2346 1552 : UpdatePMState(PM_RUN);
2347 1552 : connsAllowed = true;
2348 :
2349 : /*
2350 : * At the next iteration of the postmaster's main loop, we will
2351 : * crank up the background tasks like the autovacuum launcher and
2352 : * background workers that were not started earlier already.
2353 : */
2354 1552 : StartWorkerNeeded = true;
2355 :
2356 : /* at this point we are really open for business */
2357 1552 : ereport(LOG,
2358 : (errmsg("database system is ready to accept connections")));
2359 :
2360 : /* Report status */
2361 1552 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
2362 : #ifdef USE_SYSTEMD
2363 : sd_notify(0, "READY=1");
2364 : #endif
2365 :
2366 1552 : continue;
2367 : }
2368 :
2369 : /*
2370 : * Was it the bgwriter? Normal exit can be ignored; we'll start a new
2371 : * one at the next iteration of the postmaster's main loop, if
2372 : * necessary. Any other exit condition is treated as a crash.
2373 : */
2374 50102 : if (BgWriterPMChild && pid == BgWriterPMChild->pid)
2375 : {
2376 1766 : ReleasePostmasterChildSlot(BgWriterPMChild);
2377 1766 : BgWriterPMChild = NULL;
2378 1766 : if (!EXIT_STATUS_0(exitstatus))
2379 690 : HandleChildCrash(pid, exitstatus,
2380 690 : _("background writer process"));
2381 1766 : continue;
2382 : }
2383 :
2384 : /*
2385 : * Was it the checkpointer?
2386 : */
2387 48336 : if (CheckpointerPMChild && pid == CheckpointerPMChild->pid)
2388 : {
2389 1766 : ReleasePostmasterChildSlot(CheckpointerPMChild);
2390 1766 : CheckpointerPMChild = NULL;
2391 1766 : if (EXIT_STATUS_0(exitstatus) && pmState == PM_WAIT_CHECKPOINTER)
2392 1076 : {
2393 : /*
2394 : * OK, we saw normal exit of the checkpointer after it's been
2395 : * told to shut down. We know checkpointer wrote a shutdown
2396 : * checkpoint, otherwise we'd still be in
2397 : * PM_WAIT_XLOG_SHUTDOWN state.
2398 : *
2399 : * At this point only dead-end children and logger should be
2400 : * left.
2401 : */
2402 1076 : UpdatePMState(PM_WAIT_DEAD_END);
2403 1076 : ConfigurePostmasterWaitSet(false);
2404 1076 : SignalChildren(SIGTERM, btmask_all_except(B_LOGGER));
2405 : }
2406 : else
2407 : {
2408 : /*
2409 : * Any unexpected exit of the checkpointer (including FATAL
2410 : * exit) is treated as a crash.
2411 : */
2412 690 : HandleChildCrash(pid, exitstatus,
2413 690 : _("checkpointer process"));
2414 : }
2415 :
2416 1766 : continue;
2417 : }
2418 :
2419 : /*
2420 : * Was it the wal writer? Normal exit can be ignored; we'll start a
2421 : * new one at the next iteration of the postmaster's main loop, if
2422 : * necessary. Any other exit condition is treated as a crash.
2423 : */
2424 46570 : if (WalWriterPMChild && pid == WalWriterPMChild->pid)
2425 : {
2426 1552 : ReleasePostmasterChildSlot(WalWriterPMChild);
2427 1552 : WalWriterPMChild = NULL;
2428 1552 : if (!EXIT_STATUS_0(exitstatus))
2429 592 : HandleChildCrash(pid, exitstatus,
2430 592 : _("WAL writer process"));
2431 1552 : continue;
2432 : }
2433 :
2434 : /*
2435 : * Was it the wal receiver? If exit status is zero (normal) or one
2436 : * (FATAL exit), we assume everything is all right just like normal
2437 : * backends. (If we need a new wal receiver, we'll start one at the
2438 : * next iteration of the postmaster's main loop.)
2439 : */
2440 45018 : if (WalReceiverPMChild && pid == WalReceiverPMChild->pid)
2441 : {
2442 506 : ReleasePostmasterChildSlot(WalReceiverPMChild);
2443 506 : WalReceiverPMChild = NULL;
2444 506 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2445 46 : HandleChildCrash(pid, exitstatus,
2446 46 : _("WAL receiver process"));
2447 506 : continue;
2448 : }
2449 :
2450 : /*
2451 : * Was it the wal summarizer? Normal exit can be ignored; we'll start
2452 : * a new one at the next iteration of the postmaster's main loop, if
2453 : * necessary. Any other exit condition is treated as a crash.
2454 : */
2455 44512 : if (WalSummarizerPMChild && pid == WalSummarizerPMChild->pid)
2456 : {
2457 38 : ReleasePostmasterChildSlot(WalSummarizerPMChild);
2458 38 : WalSummarizerPMChild = NULL;
2459 38 : if (!EXIT_STATUS_0(exitstatus))
2460 36 : HandleChildCrash(pid, exitstatus,
2461 36 : _("WAL summarizer process"));
2462 38 : continue;
2463 : }
2464 :
2465 : /*
2466 : * Was it the autovacuum launcher? Normal exit can be ignored; we'll
2467 : * start a new one at the next iteration of the postmaster's main
2468 : * loop, if necessary. Any other exit condition is treated as a
2469 : * crash.
2470 : */
2471 44474 : if (AutoVacLauncherPMChild && pid == AutoVacLauncherPMChild->pid)
2472 : {
2473 1298 : ReleasePostmasterChildSlot(AutoVacLauncherPMChild);
2474 1298 : AutoVacLauncherPMChild = NULL;
2475 1298 : if (!EXIT_STATUS_0(exitstatus))
2476 502 : HandleChildCrash(pid, exitstatus,
2477 502 : _("autovacuum launcher process"));
2478 1298 : continue;
2479 : }
2480 :
2481 : /*
2482 : * Was it the archiver? If exit status is zero (normal) or one (FATAL
2483 : * exit), we assume everything is all right just like normal backends
2484 : * and just try to start a new one on the next cycle of the
2485 : * postmaster's main loop, to retry archiving remaining files.
2486 : */
2487 43176 : if (PgArchPMChild && pid == PgArchPMChild->pid)
2488 : {
2489 110 : ReleasePostmasterChildSlot(PgArchPMChild);
2490 110 : PgArchPMChild = NULL;
2491 110 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2492 80 : HandleChildCrash(pid, exitstatus,
2493 80 : _("archiver process"));
2494 110 : continue;
2495 : }
2496 :
2497 : /* Was it the system logger? If so, try to start a new one */
2498 43066 : if (SysLoggerPMChild && pid == SysLoggerPMChild->pid)
2499 : {
2500 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
2501 0 : SysLoggerPMChild = NULL;
2502 :
2503 : /* for safety's sake, launch new logger *first* */
2504 0 : if (Logging_collector)
2505 0 : StartSysLogger();
2506 :
2507 0 : if (!EXIT_STATUS_0(exitstatus))
2508 0 : LogChildExit(LOG, _("system logger process"),
2509 : pid, exitstatus);
2510 0 : continue;
2511 : }
2512 :
2513 : /*
2514 : * Was it the slot sync worker? Normal exit or FATAL exit can be
2515 : * ignored (FATAL can be caused by libpqwalreceiver on receiving
2516 : * shutdown request by the startup process during promotion); we'll
2517 : * start a new one at the next iteration of the postmaster's main
2518 : * loop, if necessary. Any other exit condition is treated as a crash.
2519 : */
2520 43066 : if (SlotSyncWorkerPMChild && pid == SlotSyncWorkerPMChild->pid)
2521 : {
2522 10 : ReleasePostmasterChildSlot(SlotSyncWorkerPMChild);
2523 10 : SlotSyncWorkerPMChild = NULL;
2524 10 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2525 0 : HandleChildCrash(pid, exitstatus,
2526 0 : _("slot sync worker process"));
2527 10 : continue;
2528 : }
2529 :
2530 : /* Was it an IO worker? */
2531 43056 : if (maybe_reap_io_worker(pid))
2532 : {
2533 5364 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2534 2070 : HandleChildCrash(pid, exitstatus, _("io worker"));
2535 :
2536 5364 : maybe_adjust_io_workers();
2537 5364 : continue;
2538 : }
2539 :
2540 : /*
2541 : * Was it a backend or a background worker?
2542 : */
2543 37692 : pmchild = FindPostmasterChildByPid(pid);
2544 37692 : if (pmchild)
2545 : {
2546 37692 : CleanupBackend(pmchild, exitstatus);
2547 : }
2548 :
2549 : /*
2550 : * We don't know anything about this child process. That's highly
2551 : * unexpected, as we do track all the child processes that we fork.
2552 : */
2553 : else
2554 : {
2555 0 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2556 0 : HandleChildCrash(pid, exitstatus, _("untracked child process"));
2557 : else
2558 0 : LogChildExit(LOG, _("untracked child process"), pid, exitstatus);
2559 : }
2560 : } /* loop over pending child-death reports */
2561 :
2562 : /*
2563 : * After cleaning out the SIGCHLD queue, see if we have any state changes
2564 : * or actions to make.
2565 : */
2566 47674 : PostmasterStateMachine();
2567 45918 : }
2568 :
2569 : /*
2570 : * CleanupBackend -- cleanup after terminated backend or background worker.
2571 : *
2572 : * Remove all local state associated with the child process and release its
2573 : * PMChild slot.
2574 : */
2575 : static void
2576 37692 : CleanupBackend(PMChild *bp,
2577 : int exitstatus) /* child's exit status. */
2578 : {
2579 : char namebuf[MAXPGPATH];
2580 : const char *procname;
2581 37692 : bool crashed = false;
2582 37692 : bool logged = false;
2583 : pid_t bp_pid;
2584 : bool bp_bgworker_notify;
2585 : BackendType bp_bkend_type;
2586 : RegisteredBgWorker *rw;
2587 :
2588 : /* Construct a process name for the log message */
2589 37692 : if (bp->bkend_type == B_BG_WORKER)
2590 : {
2591 5500 : snprintf(namebuf, MAXPGPATH, _("background worker \"%s\""),
2592 5500 : bp->rw->rw_worker.bgw_type);
2593 5500 : procname = namebuf;
2594 : }
2595 : else
2596 32192 : procname = _(GetBackendTypeDesc(bp->bkend_type));
2597 :
2598 : /*
2599 : * If a backend dies in an ugly way then we must signal all other backends
2600 : * to quickdie. If exit status is zero (normal) or one (FATAL exit), we
2601 : * assume everything is all right and proceed to remove the backend from
2602 : * the active child list.
2603 : */
2604 37692 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2605 1648 : crashed = true;
2606 :
2607 : #ifdef WIN32
2608 :
2609 : /*
2610 : * On win32, also treat ERROR_WAIT_NO_CHILDREN (128) as nonfatal case,
2611 : * since that sometimes happens under load when the process fails to start
2612 : * properly (long before it starts using shared memory). Microsoft reports
2613 : * it is related to mutex failure:
2614 : * http://archives.postgresql.org/pgsql-hackers/2010-09/msg00790.php
2615 : */
2616 : if (exitstatus == ERROR_WAIT_NO_CHILDREN)
2617 : {
2618 : LogChildExit(LOG, procname, bp->pid, exitstatus);
2619 : logged = true;
2620 : crashed = false;
2621 : }
2622 : #endif
2623 :
2624 : /*
2625 : * Release the PMChild entry.
2626 : *
2627 : * If the process attached to shared memory, this also checks that it
2628 : * detached cleanly.
2629 : */
2630 37692 : bp_pid = bp->pid;
2631 37692 : bp_bgworker_notify = bp->bgworker_notify;
2632 37692 : bp_bkend_type = bp->bkend_type;
2633 37692 : rw = bp->rw;
2634 37692 : if (!ReleasePostmasterChildSlot(bp))
2635 : {
2636 : /*
2637 : * Uh-oh, the child failed to clean itself up. Treat as a crash after
2638 : * all.
2639 : */
2640 758 : crashed = true;
2641 : }
2642 37692 : bp = NULL;
2643 :
2644 : /*
2645 : * In a crash case, exit immediately without resetting background worker
2646 : * state. However, if restart_after_crash is enabled, the background
2647 : * worker state (e.g., rw_pid) still needs be reset so the worker can
2648 : * restart after crash recovery. This reset is handled in
2649 : * ResetBackgroundWorkerCrashTimes(), not here.
2650 : */
2651 37692 : if (crashed)
2652 : {
2653 1648 : HandleChildCrash(bp_pid, exitstatus, procname);
2654 1648 : return;
2655 : }
2656 :
2657 : /*
2658 : * This backend may have been slated to receive SIGUSR1 when some
2659 : * background worker started or stopped. Cancel those notifications, as
2660 : * we don't want to signal PIDs that are not PostgreSQL backends. This
2661 : * gets skipped in the (probably very common) case where the backend has
2662 : * never requested any such notifications.
2663 : */
2664 36044 : if (bp_bgworker_notify)
2665 500 : BackgroundWorkerStopNotifications(bp_pid);
2666 :
2667 : /*
2668 : * If it was a background worker, also update its RegisteredBgWorker
2669 : * entry.
2670 : */
2671 36044 : if (bp_bkend_type == B_BG_WORKER)
2672 : {
2673 4842 : if (!EXIT_STATUS_0(exitstatus))
2674 : {
2675 : /* Record timestamp, so we know when to restart the worker. */
2676 1352 : rw->rw_crashed_at = GetCurrentTimestamp();
2677 : }
2678 : else
2679 : {
2680 : /* Zero exit status means terminate */
2681 3490 : rw->rw_crashed_at = 0;
2682 3490 : rw->rw_terminate = true;
2683 : }
2684 :
2685 4842 : rw->rw_pid = 0;
2686 4842 : ReportBackgroundWorkerExit(rw); /* report child death */
2687 :
2688 4842 : if (!logged)
2689 : {
2690 4842 : LogChildExit(EXIT_STATUS_0(exitstatus) ? DEBUG1 : LOG,
2691 : procname, bp_pid, exitstatus);
2692 4842 : logged = true;
2693 : }
2694 :
2695 : /* have it be restarted */
2696 4842 : HaveCrashedWorker = true;
2697 : }
2698 :
2699 36044 : if (!logged)
2700 31202 : LogChildExit(DEBUG2, procname, bp_pid, exitstatus);
2701 : }
2702 :
2703 : /*
2704 : * Transition into FatalError state, in response to something bad having
2705 : * happened. Commonly the caller will have logged the reason for entering
2706 : * FatalError state.
2707 : *
2708 : * This should only be called when not already in FatalError or
2709 : * ImmediateShutdown state.
2710 : */
2711 : static void
2712 16 : HandleFatalError(QuitSignalReason reason, bool consider_sigabrt)
2713 : {
2714 : int sigtosend;
2715 :
2716 : Assert(!FatalError);
2717 : Assert(Shutdown != ImmediateShutdown);
2718 :
2719 16 : SetQuitSignalReason(reason);
2720 :
2721 16 : if (consider_sigabrt && send_abort_for_crash)
2722 0 : sigtosend = SIGABRT;
2723 : else
2724 16 : sigtosend = SIGQUIT;
2725 :
2726 : /*
2727 : * Signal all other child processes to exit.
2728 : *
2729 : * We could exclude dead-end children here, but at least when sending
2730 : * SIGABRT it seems better to include them.
2731 : */
2732 16 : TerminateChildren(sigtosend);
2733 :
2734 16 : FatalError = true;
2735 :
2736 : /*
2737 : * Choose the appropriate new state to react to the fatal error. Unless we
2738 : * were already in the process of shutting down, we go through
2739 : * PM_WAIT_BACKENDS. For errors during the shutdown sequence, we directly
2740 : * switch to PM_WAIT_DEAD_END.
2741 : */
2742 16 : switch (pmState)
2743 : {
2744 0 : case PM_INIT:
2745 : /* shouldn't have any children */
2746 : Assert(false);
2747 0 : break;
2748 0 : case PM_STARTUP:
2749 : /* should have been handled in process_pm_child_exit */
2750 : Assert(false);
2751 0 : break;
2752 :
2753 : /* wait for children to die */
2754 16 : case PM_RECOVERY:
2755 : case PM_HOT_STANDBY:
2756 : case PM_RUN:
2757 : case PM_STOP_BACKENDS:
2758 16 : UpdatePMState(PM_WAIT_BACKENDS);
2759 16 : break;
2760 :
2761 0 : case PM_WAIT_BACKENDS:
2762 : /* there might be more backends to wait for */
2763 0 : break;
2764 :
2765 0 : case PM_WAIT_XLOG_SHUTDOWN:
2766 : case PM_WAIT_XLOG_ARCHIVAL:
2767 : case PM_WAIT_CHECKPOINTER:
2768 : case PM_WAIT_IO_WORKERS:
2769 :
2770 : /*
2771 : * NB: Similar code exists in PostmasterStateMachine()'s handling
2772 : * of FatalError in PM_STOP_BACKENDS/PM_WAIT_BACKENDS states.
2773 : */
2774 0 : ConfigurePostmasterWaitSet(false);
2775 0 : UpdatePMState(PM_WAIT_DEAD_END);
2776 0 : break;
2777 :
2778 0 : case PM_WAIT_DEAD_END:
2779 : case PM_NO_CHILDREN:
2780 0 : break;
2781 : }
2782 :
2783 : /*
2784 : * .. and if this doesn't happen quickly enough, now the clock is ticking
2785 : * for us to kill them without mercy.
2786 : */
2787 16 : if (AbortStartTime == 0)
2788 16 : AbortStartTime = time(NULL);
2789 16 : }
2790 :
2791 : /*
2792 : * HandleChildCrash -- cleanup after failed backend, bgwriter, checkpointer,
2793 : * walwriter, autovacuum, archiver, slot sync worker, or background worker.
2794 : *
2795 : * The objectives here are to clean up our local state about the child
2796 : * process, and to signal all other remaining children to quickdie.
2797 : *
2798 : * The caller has already released its PMChild slot.
2799 : */
2800 : static void
2801 6448 : HandleChildCrash(int pid, int exitstatus, const char *procname)
2802 : {
2803 : /*
2804 : * We only log messages and send signals if this is the first process
2805 : * crash and we're not doing an immediate shutdown; otherwise, we're only
2806 : * here to update postmaster's idea of live processes. If we have already
2807 : * signaled children, nonzero exit status is to be expected, so don't
2808 : * clutter log.
2809 : */
2810 6448 : if (FatalError || Shutdown == ImmediateShutdown)
2811 6432 : return;
2812 :
2813 16 : LogChildExit(LOG, procname, pid, exitstatus);
2814 16 : ereport(LOG,
2815 : (errmsg("terminating any other active server processes")));
2816 :
2817 : /*
2818 : * Switch into error state. The crashed process has already been removed
2819 : * from ActiveChildList.
2820 : */
2821 16 : HandleFatalError(PMQUIT_FOR_CRASH, true);
2822 : }
2823 :
2824 : /*
2825 : * Log the death of a child process.
2826 : */
2827 : static void
2828 36062 : LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2829 : {
2830 : /*
2831 : * size of activity_buffer is arbitrary, but set equal to default
2832 : * track_activity_query_size
2833 : */
2834 : char activity_buffer[1024];
2835 36062 : const char *activity = NULL;
2836 :
2837 36062 : if (!EXIT_STATUS_0(exitstatus))
2838 2334 : activity = pgstat_get_crashed_backend_activity(pid,
2839 : activity_buffer,
2840 : sizeof(activity_buffer));
2841 :
2842 36062 : if (WIFEXITED(exitstatus))
2843 36054 : ereport(lev,
2844 :
2845 : /*------
2846 : translator: %s is a noun phrase describing a child process, such as
2847 : "server process" */
2848 : (errmsg("%s (PID %d) exited with exit code %d",
2849 : procname, pid, WEXITSTATUS(exitstatus)),
2850 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2851 8 : else if (WIFSIGNALED(exitstatus))
2852 : {
2853 : #if defined(WIN32)
2854 : ereport(lev,
2855 :
2856 : /*------
2857 : translator: %s is a noun phrase describing a child process, such as
2858 : "server process" */
2859 : (errmsg("%s (PID %d) was terminated by exception 0x%X",
2860 : procname, pid, WTERMSIG(exitstatus)),
2861 : errhint("See C include file \"ntstatus.h\" for a description of the hexadecimal value."),
2862 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2863 : #else
2864 8 : ereport(lev,
2865 :
2866 : /*------
2867 : translator: %s is a noun phrase describing a child process, such as
2868 : "server process" */
2869 : (errmsg("%s (PID %d) was terminated by signal %d: %s",
2870 : procname, pid, WTERMSIG(exitstatus),
2871 : pg_strsignal(WTERMSIG(exitstatus))),
2872 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2873 : #endif
2874 : }
2875 : else
2876 0 : ereport(lev,
2877 :
2878 : /*------
2879 : translator: %s is a noun phrase describing a child process, such as
2880 : "server process" */
2881 : (errmsg("%s (PID %d) exited with unrecognized status %d",
2882 : procname, pid, exitstatus),
2883 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2884 36062 : }
2885 :
2886 : /*
2887 : * Advance the postmaster's state machine and take actions as appropriate
2888 : *
2889 : * This is common code for process_pm_shutdown_request(),
2890 : * process_pm_child_exit() and process_pm_pmsignal(), which process the signals
2891 : * that might mean we need to change state.
2892 : */
2893 : static void
2894 52958 : PostmasterStateMachine(void)
2895 : {
2896 : /* If we're doing a smart shutdown, try to advance that state. */
2897 52958 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2898 : {
2899 37816 : if (!connsAllowed)
2900 : {
2901 : /*
2902 : * This state ends when we have no normal client backends running.
2903 : * Then we're ready to stop other children.
2904 : */
2905 238 : if (CountChildren(btmask(B_BACKEND)) == 0)
2906 80 : UpdatePMState(PM_STOP_BACKENDS);
2907 : }
2908 : }
2909 :
2910 : /*
2911 : * In the PM_WAIT_BACKENDS state, wait for all the regular backends and
2912 : * processes like autovacuum and background workers that are comparable to
2913 : * backends to exit.
2914 : *
2915 : * PM_STOP_BACKENDS is a transient state that means the same as
2916 : * PM_WAIT_BACKENDS, but we signal the processes first, before waiting for
2917 : * them. Treating it as a distinct pmState allows us to share this code
2918 : * across multiple shutdown code paths.
2919 : */
2920 52958 : if (pmState == PM_STOP_BACKENDS || pmState == PM_WAIT_BACKENDS)
2921 : {
2922 9524 : BackendTypeMask targetMask = BTYPE_MASK_NONE;
2923 :
2924 : /*
2925 : * PM_WAIT_BACKENDS state ends when we have no regular backends, no
2926 : * autovac launcher or workers, and no bgworkers (including
2927 : * unconnected ones).
2928 : */
2929 9524 : targetMask = btmask_add(targetMask,
2930 : B_BACKEND,
2931 : B_AUTOVAC_LAUNCHER,
2932 : B_AUTOVAC_WORKER,
2933 : B_BG_WORKER);
2934 :
2935 : /*
2936 : * No walwriter, bgwriter, slot sync worker, or WAL summarizer either.
2937 : */
2938 9524 : targetMask = btmask_add(targetMask,
2939 : B_WAL_WRITER,
2940 : B_BG_WRITER,
2941 : B_SLOTSYNC_WORKER,
2942 : B_WAL_SUMMARIZER);
2943 :
2944 : /* If we're in recovery, also stop startup and walreceiver procs */
2945 9524 : targetMask = btmask_add(targetMask,
2946 : B_STARTUP,
2947 : B_WAL_RECEIVER);
2948 :
2949 : /*
2950 : * If we are doing crash recovery or an immediate shutdown then we
2951 : * expect archiver, checkpointer, io workers and walsender to exit as
2952 : * well, otherwise not.
2953 : */
2954 9524 : if (FatalError || Shutdown >= ImmediateShutdown)
2955 3180 : targetMask = btmask_add(targetMask,
2956 : B_CHECKPOINTER,
2957 : B_ARCHIVER,
2958 : B_IO_WORKER,
2959 : B_WAL_SENDER);
2960 :
2961 : /*
2962 : * Normally archiver, checkpointer, IO workers and walsenders will
2963 : * continue running; they will be terminated later after writing the
2964 : * checkpoint record. We also let dead-end children to keep running
2965 : * for now. The syslogger process exits last.
2966 : *
2967 : * This assertion checks that we have covered all backend types,
2968 : * either by including them in targetMask, or by noting here that they
2969 : * are allowed to continue running.
2970 : */
2971 : #ifdef USE_ASSERT_CHECKING
2972 : {
2973 : BackendTypeMask remainMask = BTYPE_MASK_NONE;
2974 :
2975 : remainMask = btmask_add(remainMask,
2976 : B_DEAD_END_BACKEND,
2977 : B_LOGGER);
2978 :
2979 : /*
2980 : * Archiver, checkpointer, IO workers, and walsender may or may
2981 : * not be in targetMask already.
2982 : */
2983 : remainMask = btmask_add(remainMask,
2984 : B_ARCHIVER,
2985 : B_CHECKPOINTER,
2986 : B_IO_WORKER,
2987 : B_WAL_SENDER);
2988 :
2989 : /* these are not real postmaster children */
2990 : remainMask = btmask_add(remainMask,
2991 : B_INVALID,
2992 : B_STANDALONE_BACKEND);
2993 :
2994 : /* All types should be included in targetMask or remainMask */
2995 : Assert((remainMask.mask | targetMask.mask) == BTYPE_MASK_ALL.mask);
2996 : }
2997 : #endif
2998 :
2999 : /* If we had not yet signaled the processes to exit, do so now */
3000 9524 : if (pmState == PM_STOP_BACKENDS)
3001 : {
3002 : /*
3003 : * Forget any pending requests for background workers, since we're
3004 : * no longer willing to launch any new workers. (If additional
3005 : * requests arrive, BackgroundWorkerStateChange will reject them.)
3006 : */
3007 1076 : ForgetUnstartedBackgroundWorkers();
3008 :
3009 1076 : SignalChildren(SIGTERM, targetMask);
3010 :
3011 1076 : UpdatePMState(PM_WAIT_BACKENDS);
3012 : }
3013 :
3014 : /* Are any of the target processes still running? */
3015 9524 : if (CountChildren(targetMask) == 0)
3016 : {
3017 1766 : if (Shutdown >= ImmediateShutdown || FatalError)
3018 : {
3019 : /*
3020 : * Stop any dead-end children and stop creating new ones.
3021 : *
3022 : * NB: Similar code exists in HandleFatalError(), when the
3023 : * error happens in pmState > PM_WAIT_BACKENDS.
3024 : */
3025 690 : UpdatePMState(PM_WAIT_DEAD_END);
3026 690 : ConfigurePostmasterWaitSet(false);
3027 690 : SignalChildren(SIGQUIT, btmask(B_DEAD_END_BACKEND));
3028 :
3029 : /*
3030 : * We already SIGQUIT'd auxiliary processes (other than
3031 : * logger), if any, when we started immediate shutdown or
3032 : * entered FatalError state.
3033 : */
3034 : }
3035 : else
3036 : {
3037 : /*
3038 : * If we get here, we are proceeding with normal shutdown. All
3039 : * the regular children are gone, and it's time to tell the
3040 : * checkpointer to do a shutdown checkpoint.
3041 : */
3042 : Assert(Shutdown > NoShutdown);
3043 : /* Start the checkpointer if not running */
3044 1076 : if (CheckpointerPMChild == NULL)
3045 0 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3046 : /* And tell it to write the shutdown checkpoint */
3047 1076 : if (CheckpointerPMChild != NULL)
3048 : {
3049 1076 : signal_child(CheckpointerPMChild, SIGINT);
3050 1076 : UpdatePMState(PM_WAIT_XLOG_SHUTDOWN);
3051 : }
3052 : else
3053 : {
3054 : /*
3055 : * If we failed to fork a checkpointer, just shut down.
3056 : * Any required cleanup will happen at next restart. We
3057 : * set FatalError so that an "abnormal shutdown" message
3058 : * gets logged when we exit.
3059 : *
3060 : * We don't consult send_abort_for_crash here, as it's
3061 : * unlikely that dumping cores would illuminate the reason
3062 : * for checkpointer fork failure.
3063 : *
3064 : * XXX: It may be worth to introduce a different PMQUIT
3065 : * value that signals that the cluster is in a bad state,
3066 : * without a process having crashed. But right now this
3067 : * path is very unlikely to be reached, so it isn't
3068 : * obviously worthwhile adding a distinct error message in
3069 : * quickdie().
3070 : */
3071 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3072 : }
3073 : }
3074 : }
3075 : }
3076 :
3077 : /*
3078 : * The state transition from PM_WAIT_XLOG_SHUTDOWN to
3079 : * PM_WAIT_XLOG_ARCHIVAL is in process_pm_pmsignal(), in response to
3080 : * PMSIGNAL_XLOG_IS_SHUTDOWN.
3081 : */
3082 :
3083 52958 : if (pmState == PM_WAIT_XLOG_ARCHIVAL)
3084 : {
3085 : /*
3086 : * PM_WAIT_XLOG_ARCHIVAL state ends when there are no children other
3087 : * than checkpointer, io workers and dead-end children left. There
3088 : * shouldn't be any regular backends left by now anyway; what we're
3089 : * really waiting for is for walsenders and archiver to exit.
3090 : */
3091 1182 : if (CountChildren(btmask_all_except(B_CHECKPOINTER, B_IO_WORKER,
3092 : B_LOGGER, B_DEAD_END_BACKEND)) == 0)
3093 : {
3094 1076 : UpdatePMState(PM_WAIT_IO_WORKERS);
3095 1076 : SignalChildren(SIGUSR2, btmask(B_IO_WORKER));
3096 : }
3097 : }
3098 :
3099 52958 : if (pmState == PM_WAIT_IO_WORKERS)
3100 : {
3101 : /*
3102 : * PM_WAIT_IO_WORKERS state ends when there's only checkpointer and
3103 : * dead-end children left.
3104 : */
3105 4302 : if (io_worker_count == 0)
3106 : {
3107 1076 : UpdatePMState(PM_WAIT_CHECKPOINTER);
3108 :
3109 : /*
3110 : * Now that the processes mentioned above are gone, tell
3111 : * checkpointer to shut down too. That allows checkpointer to
3112 : * perform some last bits of cleanup without other processes
3113 : * interfering.
3114 : */
3115 1076 : if (CheckpointerPMChild != NULL)
3116 1076 : signal_child(CheckpointerPMChild, SIGUSR2);
3117 : }
3118 : }
3119 :
3120 : /*
3121 : * The state transition from PM_WAIT_CHECKPOINTER to PM_WAIT_DEAD_END is
3122 : * in process_pm_child_exit().
3123 : */
3124 :
3125 52958 : if (pmState == PM_WAIT_DEAD_END)
3126 : {
3127 : /*
3128 : * PM_WAIT_DEAD_END state ends when all other children are gone except
3129 : * for the logger. During normal shutdown, all that remains are
3130 : * dead-end backends, but in FatalError processing we jump straight
3131 : * here with more processes remaining. Note that they have already
3132 : * been sent appropriate shutdown signals, either during a normal
3133 : * state transition leading up to PM_WAIT_DEAD_END, or during
3134 : * FatalError processing.
3135 : *
3136 : * The reason we wait is to protect against a new postmaster starting
3137 : * conflicting subprocesses; this isn't an ironclad protection, but it
3138 : * at least helps in the shutdown-and-immediately-restart scenario.
3139 : */
3140 1838 : if (CountChildren(btmask_all_except(B_LOGGER)) == 0)
3141 : {
3142 : /* These other guys should be dead already */
3143 : Assert(StartupPMChild == NULL);
3144 : Assert(WalReceiverPMChild == NULL);
3145 : Assert(WalSummarizerPMChild == NULL);
3146 : Assert(BgWriterPMChild == NULL);
3147 : Assert(CheckpointerPMChild == NULL);
3148 : Assert(WalWriterPMChild == NULL);
3149 : Assert(AutoVacLauncherPMChild == NULL);
3150 : Assert(SlotSyncWorkerPMChild == NULL);
3151 : /* syslogger is not considered here */
3152 1766 : UpdatePMState(PM_NO_CHILDREN);
3153 : }
3154 : }
3155 :
3156 : /*
3157 : * If we've been told to shut down, we exit as soon as there are no
3158 : * remaining children. If there was a crash, cleanup will occur at the
3159 : * next startup. (Before PostgreSQL 8.3, we tried to recover from the
3160 : * crash before exiting, but that seems unwise if we are quitting because
3161 : * we got SIGTERM from init --- there may well not be time for recovery
3162 : * before init decides to SIGKILL us.)
3163 : *
3164 : * Note that the syslogger continues to run. It will exit when it sees
3165 : * EOF on its input pipe, which happens when there are no more upstream
3166 : * processes.
3167 : */
3168 52958 : if (Shutdown > NoShutdown && pmState == PM_NO_CHILDREN)
3169 : {
3170 1750 : if (FatalError)
3171 : {
3172 0 : ereport(LOG, (errmsg("abnormal database system shutdown")));
3173 0 : ExitPostmaster(1);
3174 : }
3175 : else
3176 : {
3177 : /*
3178 : * Normal exit from the postmaster is here. We don't need to log
3179 : * anything here, since the UnlinkLockFiles proc_exit callback
3180 : * will do so, and that should be the last user-visible action.
3181 : */
3182 1750 : ExitPostmaster(0);
3183 : }
3184 : }
3185 :
3186 : /*
3187 : * If the startup process failed, or the user does not want an automatic
3188 : * restart after backend crashes, wait for all non-syslogger children to
3189 : * exit, and then exit postmaster. We don't try to reinitialize when the
3190 : * startup process fails, because more than likely it will just fail again
3191 : * and we will keep trying forever.
3192 : */
3193 51208 : if (pmState == PM_NO_CHILDREN)
3194 : {
3195 16 : if (StartupStatus == STARTUP_CRASHED)
3196 : {
3197 6 : ereport(LOG,
3198 : (errmsg("shutting down due to startup process failure")));
3199 6 : ExitPostmaster(1);
3200 : }
3201 10 : if (!restart_after_crash)
3202 : {
3203 0 : ereport(LOG,
3204 : (errmsg("shutting down because \"restart_after_crash\" is off")));
3205 0 : ExitPostmaster(1);
3206 : }
3207 : }
3208 :
3209 : /*
3210 : * If we need to recover from a crash, wait for all non-syslogger children
3211 : * to exit, then reset shmem and start the startup process.
3212 : */
3213 51202 : if (FatalError && pmState == PM_NO_CHILDREN)
3214 : {
3215 10 : ereport(LOG,
3216 : (errmsg("all server processes terminated; reinitializing")));
3217 :
3218 : /* remove leftover temporary files after a crash */
3219 10 : if (remove_temp_files_after_crash)
3220 8 : RemovePgTempFiles();
3221 :
3222 : /* allow background workers to immediately restart */
3223 10 : ResetBackgroundWorkerCrashTimes();
3224 :
3225 10 : shmem_exit(1);
3226 :
3227 : /* re-read control file into local memory */
3228 10 : LocalProcessControlFile(true);
3229 :
3230 : /* re-create shared memory and semaphores */
3231 10 : CreateSharedMemoryAndSemaphores();
3232 :
3233 10 : UpdatePMState(PM_STARTUP);
3234 :
3235 : /* Make sure we can perform I/O while starting up. */
3236 10 : maybe_adjust_io_workers();
3237 :
3238 10 : StartupPMChild = StartChildProcess(B_STARTUP);
3239 : Assert(StartupPMChild != NULL);
3240 10 : StartupStatus = STARTUP_RUNNING;
3241 : /* crash recovery started, reset SIGKILL flag */
3242 10 : AbortStartTime = 0;
3243 :
3244 : /* start accepting server socket connection events again */
3245 10 : ConfigurePostmasterWaitSet(true);
3246 : }
3247 51202 : }
3248 :
3249 : static const char *
3250 2012 : pmstate_name(PMState state)
3251 : {
3252 : #define PM_TOSTR_CASE(sym) case sym: return #sym
3253 2012 : switch (state)
3254 : {
3255 112 : PM_TOSTR_CASE(PM_INIT);
3256 224 : PM_TOSTR_CASE(PM_STARTUP);
3257 32 : PM_TOSTR_CASE(PM_RECOVERY);
3258 26 : PM_TOSTR_CASE(PM_HOT_STANDBY);
3259 210 : PM_TOSTR_CASE(PM_RUN);
3260 164 : PM_TOSTR_CASE(PM_STOP_BACKENDS);
3261 240 : PM_TOSTR_CASE(PM_WAIT_BACKENDS);
3262 164 : PM_TOSTR_CASE(PM_WAIT_XLOG_SHUTDOWN);
3263 164 : PM_TOSTR_CASE(PM_WAIT_XLOG_ARCHIVAL);
3264 164 : PM_TOSTR_CASE(PM_WAIT_IO_WORKERS);
3265 232 : PM_TOSTR_CASE(PM_WAIT_DEAD_END);
3266 164 : PM_TOSTR_CASE(PM_WAIT_CHECKPOINTER);
3267 116 : PM_TOSTR_CASE(PM_NO_CHILDREN);
3268 : }
3269 : #undef PM_TOSTR_CASE
3270 :
3271 0 : pg_unreachable();
3272 : return ""; /* silence compiler */
3273 : }
3274 :
3275 : /*
3276 : * Simple wrapper for updating pmState. The main reason to have this wrapper
3277 : * is that it makes it easy to log all state transitions.
3278 : */
3279 : static void
3280 14924 : UpdatePMState(PMState newState)
3281 : {
3282 14924 : elog(DEBUG1, "updating PMState from %s to %s",
3283 : pmstate_name(pmState), pmstate_name(newState));
3284 14924 : pmState = newState;
3285 14924 : }
3286 :
3287 : /*
3288 : * Launch background processes after state change, or relaunch after an
3289 : * existing process has exited.
3290 : *
3291 : * Check the current pmState and the status of any background processes. If
3292 : * there are any background processes missing that should be running in the
3293 : * current state, but are not, launch them.
3294 : */
3295 : static void
3296 240718 : LaunchMissingBackgroundProcesses(void)
3297 : {
3298 : /* Syslogger is active in all states */
3299 240718 : if (SysLoggerPMChild == NULL && Logging_collector)
3300 0 : StartSysLogger();
3301 :
3302 : /*
3303 : * The number of configured workers might have changed, or a prior start
3304 : * of a worker might have failed. Check if we need to start/stop any
3305 : * workers.
3306 : *
3307 : * A config file change will always lead to this function being called, so
3308 : * we always will process the config change in a timely manner.
3309 : */
3310 240718 : maybe_adjust_io_workers();
3311 :
3312 : /*
3313 : * The checkpointer and the background writer are active from the start,
3314 : * until shutdown is initiated.
3315 : *
3316 : * (If the checkpointer is not running when we enter the
3317 : * PM_WAIT_XLOG_SHUTDOWN state, it is launched one more time to perform
3318 : * the shutdown checkpoint. That's done in PostmasterStateMachine(), not
3319 : * here.)
3320 : */
3321 240718 : if (pmState == PM_RUN || pmState == PM_RECOVERY ||
3322 17924 : pmState == PM_HOT_STANDBY || pmState == PM_STARTUP)
3323 : {
3324 227274 : if (CheckpointerPMChild == NULL)
3325 10 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3326 227274 : if (BgWriterPMChild == NULL)
3327 10 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
3328 : }
3329 :
3330 : /*
3331 : * WAL writer is needed only in normal operation (else we cannot be
3332 : * writing any new WAL).
3333 : */
3334 240718 : if (WalWriterPMChild == NULL && pmState == PM_RUN)
3335 1552 : WalWriterPMChild = StartChildProcess(B_WAL_WRITER);
3336 :
3337 : /*
3338 : * We don't want autovacuum to run in binary upgrade mode because
3339 : * autovacuum might update relfrozenxid for empty tables before the
3340 : * physical files are put in place.
3341 : */
3342 261336 : if (!IsBinaryUpgrade && AutoVacLauncherPMChild == NULL &&
3343 28912 : (AutoVacuumingActive() || start_autovac_launcher) &&
3344 12324 : pmState == PM_RUN)
3345 : {
3346 1298 : AutoVacLauncherPMChild = StartChildProcess(B_AUTOVAC_LAUNCHER);
3347 1298 : if (AutoVacLauncherPMChild != NULL)
3348 1298 : start_autovac_launcher = false; /* signal processed */
3349 : }
3350 :
3351 : /*
3352 : * If WAL archiving is enabled always, we are allowed to start archiver
3353 : * even during recovery.
3354 : */
3355 240718 : if (PgArchPMChild == NULL &&
3356 238516 : ((XLogArchivingActive() && pmState == PM_RUN) ||
3357 238516 : (XLogArchivingAlways() && (pmState == PM_RECOVERY || pmState == PM_HOT_STANDBY))) &&
3358 104 : PgArchCanRestart())
3359 104 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3360 :
3361 : /*
3362 : * If we need to start a slot sync worker, try to do that now
3363 : *
3364 : * We allow to start the slot sync worker when we are on a hot standby,
3365 : * fast or immediate shutdown is not in progress, slot sync parameters are
3366 : * configured correctly, and it is the first time of worker's launch, or
3367 : * enough time has passed since the worker was launched last.
3368 : */
3369 240718 : if (SlotSyncWorkerPMChild == NULL && pmState == PM_HOT_STANDBY &&
3370 3786 : Shutdown <= SmartShutdown && sync_replication_slots &&
3371 30 : ValidateSlotSyncParams(LOG) && SlotSyncWorkerCanRestart())
3372 10 : SlotSyncWorkerPMChild = StartChildProcess(B_SLOTSYNC_WORKER);
3373 :
3374 : /*
3375 : * If we need to start a WAL receiver, try to do that now
3376 : *
3377 : * Note: if a walreceiver process is already running, it might seem that
3378 : * we should clear WalReceiverRequested. However, there's a race
3379 : * condition if the walreceiver terminates and the startup process
3380 : * immediately requests a new one: it's quite possible to get the signal
3381 : * for the request before reaping the dead walreceiver process. Better to
3382 : * risk launching an extra walreceiver than to miss launching one we need.
3383 : * (The walreceiver code has logic to recognize that it should go away if
3384 : * not needed.)
3385 : */
3386 240718 : if (WalReceiverRequested)
3387 : {
3388 822 : if (WalReceiverPMChild == NULL &&
3389 554 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
3390 552 : pmState == PM_HOT_STANDBY) &&
3391 506 : Shutdown <= SmartShutdown)
3392 : {
3393 506 : WalReceiverPMChild = StartChildProcess(B_WAL_RECEIVER);
3394 506 : if (WalReceiverPMChild != NULL)
3395 506 : WalReceiverRequested = false;
3396 : /* else leave the flag set, so we'll try again later */
3397 : }
3398 : }
3399 :
3400 : /* If we need to start a WAL summarizer, try to do that now */
3401 240718 : if (summarize_wal && WalSummarizerPMChild == NULL &&
3402 96 : (pmState == PM_RUN || pmState == PM_HOT_STANDBY) &&
3403 38 : Shutdown <= SmartShutdown)
3404 38 : WalSummarizerPMChild = StartChildProcess(B_WAL_SUMMARIZER);
3405 :
3406 : /* Get other worker processes running, if needed */
3407 240718 : if (StartWorkerNeeded || HaveCrashedWorker)
3408 14414 : maybe_start_bgworkers();
3409 240718 : }
3410 :
3411 : /*
3412 : * Return string representation of signal.
3413 : *
3414 : * Because this is only implemented for signals we already rely on in this
3415 : * file we don't need to deal with unimplemented or same-numeric-value signals
3416 : * (as we'd e.g. have to for EWOULDBLOCK / EAGAIN).
3417 : */
3418 : static const char *
3419 34 : pm_signame(int signal)
3420 : {
3421 : #define PM_TOSTR_CASE(sym) case sym: return #sym
3422 34 : switch (signal)
3423 : {
3424 0 : PM_TOSTR_CASE(SIGABRT);
3425 0 : PM_TOSTR_CASE(SIGCHLD);
3426 0 : PM_TOSTR_CASE(SIGHUP);
3427 4 : PM_TOSTR_CASE(SIGINT);
3428 0 : PM_TOSTR_CASE(SIGKILL);
3429 0 : PM_TOSTR_CASE(SIGQUIT);
3430 20 : PM_TOSTR_CASE(SIGTERM);
3431 0 : PM_TOSTR_CASE(SIGUSR1);
3432 10 : PM_TOSTR_CASE(SIGUSR2);
3433 0 : default:
3434 : /* all signals sent by postmaster should be listed here */
3435 : Assert(false);
3436 0 : return "(unknown)";
3437 : }
3438 : #undef PM_TOSTR_CASE
3439 :
3440 : return ""; /* silence compiler */
3441 : }
3442 :
3443 : /*
3444 : * Send a signal to a postmaster child process
3445 : *
3446 : * On systems that have setsid(), each child process sets itself up as a
3447 : * process group leader. For signals that are generally interpreted in the
3448 : * appropriate fashion, we signal the entire process group not just the
3449 : * direct child process. This allows us to, for example, SIGQUIT a blocked
3450 : * archive_recovery script, or SIGINT a script being run by a backend via
3451 : * system().
3452 : *
3453 : * There is a race condition for recently-forked children: they might not
3454 : * have executed setsid() yet. So we signal the child directly as well as
3455 : * the group. We assume such a child will handle the signal before trying
3456 : * to spawn any grandchild processes. We also assume that signaling the
3457 : * child twice will not cause any problems.
3458 : */
3459 : static void
3460 20336 : signal_child(PMChild *pmchild, int signal)
3461 : {
3462 20336 : pid_t pid = pmchild->pid;
3463 :
3464 20336 : ereport(DEBUG3,
3465 : (errmsg_internal("sending signal %d/%s to %s process with pid %d",
3466 : signal, pm_signame(signal),
3467 : GetBackendTypeDesc(pmchild->bkend_type),
3468 : (int) pmchild->pid)));
3469 :
3470 20336 : if (kill(pid, signal) < 0)
3471 0 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) pid, signal);
3472 : #ifdef HAVE_SETSID
3473 20336 : switch (signal)
3474 : {
3475 12964 : case SIGINT:
3476 : case SIGTERM:
3477 : case SIGQUIT:
3478 : case SIGKILL:
3479 : case SIGABRT:
3480 12964 : if (kill(-pid, signal) < 0)
3481 16 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) (-pid), signal);
3482 12964 : break;
3483 7372 : default:
3484 7372 : break;
3485 : }
3486 : #endif
3487 20336 : }
3488 :
3489 : /*
3490 : * Send a signal to the targeted children.
3491 : */
3492 : static bool
3493 5964 : SignalChildren(int signal, BackendTypeMask targetMask)
3494 : {
3495 : dlist_iter iter;
3496 5964 : bool signaled = false;
3497 :
3498 34004 : dlist_foreach(iter, &ActiveChildList)
3499 : {
3500 28040 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3501 :
3502 : /*
3503 : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3504 : * if any B_BACKEND backends have recently announced that they are
3505 : * actually WAL senders.
3506 : */
3507 28040 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3508 14298 : bp->bkend_type == B_BACKEND)
3509 : {
3510 1408 : if (IsPostmasterChildWalSender(bp->child_slot))
3511 74 : bp->bkend_type = B_WAL_SENDER;
3512 : }
3513 :
3514 28040 : if (!btmask_contains(targetMask, bp->bkend_type))
3515 9976 : continue;
3516 :
3517 18064 : signal_child(bp, signal);
3518 18064 : signaled = true;
3519 : }
3520 5964 : return signaled;
3521 : }
3522 :
3523 : /*
3524 : * Send a termination signal to children. This considers all of our children
3525 : * processes, except syslogger.
3526 : */
3527 : static void
3528 690 : TerminateChildren(int signal)
3529 : {
3530 690 : SignalChildren(signal, btmask_all_except(B_LOGGER));
3531 690 : if (StartupPMChild != NULL)
3532 : {
3533 92 : if (signal == SIGQUIT || signal == SIGKILL || signal == SIGABRT)
3534 92 : StartupStatus = STARTUP_SIGNALED;
3535 : }
3536 690 : }
3537 :
3538 : /*
3539 : * BackendStartup -- start backend process
3540 : *
3541 : * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
3542 : *
3543 : * Note: if you change this code, also consider StartAutovacuumWorker and
3544 : * StartBackgroundWorker.
3545 : */
3546 : static int
3547 27932 : BackendStartup(ClientSocket *client_sock)
3548 : {
3549 27932 : PMChild *bn = NULL;
3550 : pid_t pid;
3551 : BackendStartupData startup_data;
3552 : CAC_state cac;
3553 :
3554 : /*
3555 : * Capture time that Postmaster got a socket from accept (for logging
3556 : * connection establishment and setup total duration).
3557 : */
3558 27932 : startup_data.socket_created = GetCurrentTimestamp();
3559 :
3560 : /*
3561 : * Allocate and assign the child slot. Note we must do this before
3562 : * forking, so that we can handle failures (out of memory or child-process
3563 : * slots) cleanly.
3564 : */
3565 27932 : cac = canAcceptConnections(B_BACKEND);
3566 27932 : if (cac == CAC_OK)
3567 : {
3568 : /* Can change later to B_WAL_SENDER */
3569 27254 : bn = AssignPostmasterChildSlot(B_BACKEND);
3570 27254 : if (!bn)
3571 : {
3572 : /*
3573 : * Too many regular child processes; launch a dead-end child
3574 : * process instead.
3575 : */
3576 56 : cac = CAC_TOOMANY;
3577 : }
3578 : }
3579 27932 : if (!bn)
3580 : {
3581 734 : bn = AllocDeadEndChild();
3582 734 : if (!bn)
3583 : {
3584 0 : ereport(LOG,
3585 : (errcode(ERRCODE_OUT_OF_MEMORY),
3586 : errmsg("out of memory")));
3587 0 : return STATUS_ERROR;
3588 : }
3589 : }
3590 :
3591 : /* Pass down canAcceptConnections state */
3592 27932 : startup_data.canAcceptConnections = cac;
3593 27932 : bn->rw = NULL;
3594 :
3595 : /* Hasn't asked to be notified about any bgworkers yet */
3596 27932 : bn->bgworker_notify = false;
3597 :
3598 27932 : pid = postmaster_child_launch(bn->bkend_type, bn->child_slot,
3599 : &startup_data, sizeof(startup_data),
3600 : client_sock);
3601 27932 : if (pid < 0)
3602 : {
3603 : /* in parent, fork failed */
3604 0 : int save_errno = errno;
3605 :
3606 0 : (void) ReleasePostmasterChildSlot(bn);
3607 0 : errno = save_errno;
3608 0 : ereport(LOG,
3609 : (errmsg("could not fork new process for connection: %m")));
3610 0 : report_fork_failure_to_client(client_sock, save_errno);
3611 0 : return STATUS_ERROR;
3612 : }
3613 :
3614 : /* in parent, successful fork */
3615 27932 : ereport(DEBUG2,
3616 : (errmsg_internal("forked new %s, pid=%d socket=%d",
3617 : GetBackendTypeDesc(bn->bkend_type),
3618 : (int) pid, (int) client_sock->sock)));
3619 :
3620 : /*
3621 : * Everything's been successful, it's safe to add this backend to our list
3622 : * of backends.
3623 : */
3624 27932 : bn->pid = pid;
3625 27932 : return STATUS_OK;
3626 : }
3627 :
3628 : /*
3629 : * Try to report backend fork() failure to client before we close the
3630 : * connection. Since we do not care to risk blocking the postmaster on
3631 : * this connection, we set the connection to non-blocking and try only once.
3632 : *
3633 : * This is grungy special-purpose code; we cannot use backend libpq since
3634 : * it's not up and running.
3635 : */
3636 : static void
3637 0 : report_fork_failure_to_client(ClientSocket *client_sock, int errnum)
3638 : {
3639 : char buffer[1000];
3640 : int rc;
3641 :
3642 : /* Format the error message packet (always V2 protocol) */
3643 0 : snprintf(buffer, sizeof(buffer), "E%s%s\n",
3644 : _("could not fork new process for connection: "),
3645 : strerror(errnum));
3646 :
3647 : /* Set port to non-blocking. Don't do send() if this fails */
3648 0 : if (!pg_set_noblock(client_sock->sock))
3649 0 : return;
3650 :
3651 : /* We'll retry after EINTR, but ignore all other failures */
3652 : do
3653 : {
3654 0 : rc = send(client_sock->sock, buffer, strlen(buffer) + 1, 0);
3655 0 : } while (rc < 0 && errno == EINTR);
3656 : }
3657 :
3658 : /*
3659 : * ExitPostmaster -- cleanup
3660 : *
3661 : * Do NOT call exit() directly --- always go through here!
3662 : */
3663 : static void
3664 1762 : ExitPostmaster(int status)
3665 : {
3666 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
3667 :
3668 : /*
3669 : * There is no known cause for a postmaster to become multithreaded after
3670 : * startup. However, we might reach here via an error exit before
3671 : * reaching the test in PostmasterMain, so provide the same hint as there.
3672 : * This message uses LOG level, because an unclean shutdown at this point
3673 : * would usually not look much different from a clean shutdown.
3674 : */
3675 : if (pthread_is_threaded_np() != 0)
3676 : ereport(LOG,
3677 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3678 : errmsg("postmaster became multithreaded"),
3679 : errhint("Set the LC_ALL environment variable to a valid locale.")));
3680 : #endif
3681 :
3682 : /* should cleanup shared memory and kill all backends */
3683 :
3684 : /*
3685 : * Not sure of the semantics here. When the Postmaster dies, should the
3686 : * backends all be killed? probably not.
3687 : *
3688 : * MUST -- vadim 05-10-1999
3689 : */
3690 :
3691 1762 : proc_exit(status);
3692 : }
3693 :
3694 : /*
3695 : * Handle pmsignal conditions representing requests from backends,
3696 : * and check for promote and logrotate requests from pg_ctl.
3697 : */
3698 : static void
3699 164894 : process_pm_pmsignal(void)
3700 : {
3701 164894 : bool request_state_update = false;
3702 :
3703 164894 : pending_pm_pmsignal = false;
3704 :
3705 164894 : ereport(DEBUG2,
3706 : (errmsg_internal("postmaster received pmsignal signal")));
3707 :
3708 : /*
3709 : * RECOVERY_STARTED and BEGIN_HOT_STANDBY signals are ignored in
3710 : * unexpected states. If the startup process quickly starts up, completes
3711 : * recovery, exits, we might process the death of the startup process
3712 : * first. We don't want to go back to recovery in that case.
3713 : */
3714 164894 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_STARTED) &&
3715 502 : pmState == PM_STARTUP && Shutdown == NoShutdown)
3716 : {
3717 : /* WAL redo has started. We're out of reinitialization. */
3718 502 : FatalError = false;
3719 502 : AbortStartTime = 0;
3720 502 : reachedConsistency = false;
3721 :
3722 : /*
3723 : * Start the archiver if we're responsible for (re-)archiving received
3724 : * files.
3725 : */
3726 : Assert(PgArchPMChild == NULL);
3727 502 : if (XLogArchivingAlways())
3728 6 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3729 :
3730 : /*
3731 : * If we aren't planning to enter hot standby mode later, treat
3732 : * RECOVERY_STARTED as meaning we're out of startup, and report status
3733 : * accordingly.
3734 : */
3735 502 : if (!EnableHotStandby)
3736 : {
3737 4 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STANDBY);
3738 : #ifdef USE_SYSTEMD
3739 : sd_notify(0, "READY=1");
3740 : #endif
3741 : }
3742 :
3743 502 : UpdatePMState(PM_RECOVERY);
3744 : }
3745 :
3746 164894 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_CONSISTENT) &&
3747 320 : pmState == PM_RECOVERY && Shutdown == NoShutdown)
3748 : {
3749 320 : reachedConsistency = true;
3750 : }
3751 :
3752 164894 : if (CheckPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY) &&
3753 304 : (pmState == PM_RECOVERY && Shutdown == NoShutdown))
3754 : {
3755 304 : ereport(LOG,
3756 : (errmsg("database system is ready to accept read-only connections")));
3757 :
3758 : /* Report status */
3759 304 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
3760 : #ifdef USE_SYSTEMD
3761 : sd_notify(0, "READY=1");
3762 : #endif
3763 :
3764 304 : UpdatePMState(PM_HOT_STANDBY);
3765 304 : connsAllowed = true;
3766 :
3767 : /* Some workers may be scheduled to start now */
3768 304 : StartWorkerNeeded = true;
3769 : }
3770 :
3771 : /* Process background worker state changes. */
3772 164894 : if (CheckPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE))
3773 : {
3774 : /* Accept new worker requests only if not stopping. */
3775 2570 : BackgroundWorkerStateChange(pmState < PM_STOP_BACKENDS);
3776 2570 : StartWorkerNeeded = true;
3777 : }
3778 :
3779 : /* Tell syslogger to rotate logfile if requested */
3780 164894 : if (SysLoggerPMChild != NULL)
3781 : {
3782 4 : if (CheckLogrotateSignal())
3783 : {
3784 2 : signal_child(SysLoggerPMChild, SIGUSR1);
3785 2 : RemoveLogrotateSignalFiles();
3786 : }
3787 2 : else if (CheckPostmasterSignal(PMSIGNAL_ROTATE_LOGFILE))
3788 : {
3789 0 : signal_child(SysLoggerPMChild, SIGUSR1);
3790 : }
3791 : }
3792 :
3793 164894 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER) &&
3794 153416 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3795 : {
3796 : /*
3797 : * Start one iteration of the autovacuum daemon, even if autovacuuming
3798 : * is nominally not enabled. This is so we can have an active defense
3799 : * against transaction ID wraparound. We set a flag for the main loop
3800 : * to do it rather than trying to do it here --- this is because the
3801 : * autovac process itself may send the signal, and we want to handle
3802 : * that by launching another iteration as soon as the current one
3803 : * completes.
3804 : */
3805 153416 : start_autovac_launcher = true;
3806 : }
3807 :
3808 164894 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER) &&
3809 4264 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3810 : {
3811 : /* The autovacuum launcher wants us to start a worker process. */
3812 4260 : StartAutovacuumWorker();
3813 : }
3814 :
3815 164894 : if (CheckPostmasterSignal(PMSIGNAL_START_WALRECEIVER))
3816 : {
3817 : /* Startup Process wants us to start the walreceiver process. */
3818 522 : WalReceiverRequested = true;
3819 : }
3820 :
3821 164894 : if (CheckPostmasterSignal(PMSIGNAL_XLOG_IS_SHUTDOWN))
3822 : {
3823 : /* Checkpointer completed the shutdown checkpoint */
3824 1076 : if (pmState == PM_WAIT_XLOG_SHUTDOWN)
3825 : {
3826 : /*
3827 : * If we have an archiver subprocess, tell it to do a last archive
3828 : * cycle and quit. Likewise, if we have walsender processes, tell
3829 : * them to send any remaining WAL and quit.
3830 : */
3831 : Assert(Shutdown > NoShutdown);
3832 :
3833 : /* Waken archiver for the last time */
3834 1076 : if (PgArchPMChild != NULL)
3835 30 : signal_child(PgArchPMChild, SIGUSR2);
3836 :
3837 : /*
3838 : * Waken walsenders for the last time. No regular backends should
3839 : * be around anymore.
3840 : */
3841 1076 : SignalChildren(SIGUSR2, btmask(B_WAL_SENDER));
3842 :
3843 1076 : UpdatePMState(PM_WAIT_XLOG_ARCHIVAL);
3844 : }
3845 0 : else if (!FatalError && Shutdown != ImmediateShutdown)
3846 : {
3847 : /*
3848 : * Checkpointer only ought to perform the shutdown checkpoint
3849 : * during shutdown. If somehow checkpointer did so in another
3850 : * situation, we have no choice but to crash-restart.
3851 : *
3852 : * It's possible however that we get PMSIGNAL_XLOG_IS_SHUTDOWN
3853 : * outside of PM_WAIT_XLOG_SHUTDOWN if an orderly shutdown was
3854 : * "interrupted" by a crash or an immediate shutdown.
3855 : */
3856 0 : ereport(LOG,
3857 : (errmsg("WAL was shut down unexpectedly")));
3858 :
3859 : /*
3860 : * Doesn't seem likely to help to take send_abort_for_crash into
3861 : * account here.
3862 : */
3863 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3864 : }
3865 :
3866 : /*
3867 : * Need to run PostmasterStateMachine() to check if we already can go
3868 : * to the next state.
3869 : */
3870 1076 : request_state_update = true;
3871 : }
3872 :
3873 : /*
3874 : * Try to advance postmaster's state machine, if a child requests it.
3875 : */
3876 164894 : if (CheckPostmasterSignal(PMSIGNAL_ADVANCE_STATE_MACHINE))
3877 : {
3878 2458 : request_state_update = true;
3879 : }
3880 :
3881 : /*
3882 : * Be careful about the order of this action relative to this function's
3883 : * other actions. Generally, this should be after other actions, in case
3884 : * they have effects PostmasterStateMachine would need to know about.
3885 : * However, we should do it before the CheckPromoteSignal step, which
3886 : * cannot have any (immediate) effect on the state machine, but does
3887 : * depend on what state we're in now.
3888 : */
3889 164894 : if (request_state_update)
3890 : {
3891 3534 : PostmasterStateMachine();
3892 : }
3893 :
3894 164894 : if (StartupPMChild != NULL &&
3895 1586 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
3896 2600 : pmState == PM_HOT_STANDBY) &&
3897 1586 : CheckPromoteSignal())
3898 : {
3899 : /*
3900 : * Tell startup process to finish recovery.
3901 : *
3902 : * Leave the promote signal file in place and let the Startup process
3903 : * do the unlink.
3904 : */
3905 88 : signal_child(StartupPMChild, SIGUSR2);
3906 : }
3907 164894 : }
3908 :
3909 : /*
3910 : * Dummy signal handler
3911 : *
3912 : * We use this for signals that we don't actually use in the postmaster,
3913 : * but we do use in backends. If we were to SIG_IGN such signals in the
3914 : * postmaster, then a newly started backend might drop a signal that arrives
3915 : * before it's able to reconfigure its signal processing. (See notes in
3916 : * tcop/postgres.c.)
3917 : */
3918 : static void
3919 0 : dummy_handler(SIGNAL_ARGS)
3920 : {
3921 0 : }
3922 :
3923 : /*
3924 : * Count up number of child processes of specified types.
3925 : */
3926 : static int
3927 12782 : CountChildren(BackendTypeMask targetMask)
3928 : {
3929 : dlist_iter iter;
3930 12782 : int cnt = 0;
3931 :
3932 76254 : dlist_foreach(iter, &ActiveChildList)
3933 : {
3934 63472 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3935 :
3936 : /*
3937 : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3938 : * if any B_BACKEND backends have recently announced that they are
3939 : * actually WAL senders.
3940 : */
3941 63472 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3942 45274 : bp->bkend_type == B_BACKEND)
3943 : {
3944 3072 : if (IsPostmasterChildWalSender(bp->child_slot))
3945 0 : bp->bkend_type = B_WAL_SENDER;
3946 : }
3947 :
3948 63472 : if (!btmask_contains(targetMask, bp->bkend_type))
3949 32848 : continue;
3950 :
3951 30624 : ereport(DEBUG4,
3952 : (errmsg_internal("%s process %d is still running",
3953 : GetBackendTypeDesc(bp->bkend_type), (int) bp->pid)));
3954 :
3955 30624 : cnt++;
3956 : }
3957 12782 : return cnt;
3958 : }
3959 :
3960 :
3961 : /*
3962 : * StartChildProcess -- start an auxiliary process for the postmaster
3963 : *
3964 : * "type" determines what kind of child will be started. All child types
3965 : * initially go to AuxiliaryProcessMain, which will handle common setup.
3966 : *
3967 : * Return value of StartChildProcess is subprocess' PMChild entry, or NULL on
3968 : * failure.
3969 : */
3970 : static PMChild *
3971 18448 : StartChildProcess(BackendType type)
3972 : {
3973 : PMChild *pmchild;
3974 : pid_t pid;
3975 :
3976 18448 : pmchild = AssignPostmasterChildSlot(type);
3977 18448 : if (!pmchild)
3978 : {
3979 0 : if (type == B_AUTOVAC_WORKER)
3980 0 : ereport(LOG,
3981 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
3982 : errmsg("no slot available for new autovacuum worker process")));
3983 : else
3984 : {
3985 : /* shouldn't happen because we allocate enough slots */
3986 0 : elog(LOG, "no postmaster child slot available for aux process");
3987 : }
3988 0 : return NULL;
3989 : }
3990 :
3991 18448 : pid = postmaster_child_launch(type, pmchild->child_slot, NULL, 0, NULL);
3992 18448 : if (pid < 0)
3993 : {
3994 : /* in parent, fork failed */
3995 0 : ReleasePostmasterChildSlot(pmchild);
3996 0 : ereport(LOG,
3997 : (errmsg("could not fork \"%s\" process: %m", PostmasterChildName(type))));
3998 :
3999 : /*
4000 : * fork failure is fatal during startup, but there's no need to choke
4001 : * immediately if starting other child types fails.
4002 : */
4003 0 : if (type == B_STARTUP)
4004 0 : ExitPostmaster(1);
4005 0 : return NULL;
4006 : }
4007 :
4008 : /* in parent, successful fork */
4009 18448 : pmchild->pid = pid;
4010 18448 : return pmchild;
4011 : }
4012 :
4013 : /*
4014 : * StartSysLogger -- start the syslogger process
4015 : */
4016 : void
4017 2 : StartSysLogger(void)
4018 : {
4019 : Assert(SysLoggerPMChild == NULL);
4020 :
4021 2 : SysLoggerPMChild = AssignPostmasterChildSlot(B_LOGGER);
4022 2 : if (!SysLoggerPMChild)
4023 0 : elog(PANIC, "no postmaster child slot available for syslogger");
4024 2 : SysLoggerPMChild->pid = SysLogger_Start(SysLoggerPMChild->child_slot);
4025 2 : if (SysLoggerPMChild->pid == 0)
4026 : {
4027 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
4028 0 : SysLoggerPMChild = NULL;
4029 : }
4030 2 : }
4031 :
4032 : /*
4033 : * StartAutovacuumWorker
4034 : * Start an autovac worker process.
4035 : *
4036 : * This function is here because it enters the resulting PID into the
4037 : * postmaster's private backends list.
4038 : *
4039 : * NB -- this code very roughly matches BackendStartup.
4040 : */
4041 : static void
4042 4260 : StartAutovacuumWorker(void)
4043 : {
4044 : PMChild *bn;
4045 :
4046 : /*
4047 : * If not in condition to run a process, don't try, but handle it like a
4048 : * fork failure. This does not normally happen, since the signal is only
4049 : * supposed to be sent by autovacuum launcher when it's OK to do it, but
4050 : * we have to check to avoid race-condition problems during DB state
4051 : * changes.
4052 : */
4053 4260 : if (canAcceptConnections(B_AUTOVAC_WORKER) == CAC_OK)
4054 : {
4055 4260 : bn = StartChildProcess(B_AUTOVAC_WORKER);
4056 4260 : if (bn)
4057 : {
4058 4260 : bn->bgworker_notify = false;
4059 4260 : bn->rw = NULL;
4060 4260 : return;
4061 : }
4062 : else
4063 : {
4064 : /*
4065 : * fork failed, fall through to report -- actual error message was
4066 : * logged by StartChildProcess
4067 : */
4068 : }
4069 : }
4070 :
4071 : /*
4072 : * Report the failure to the launcher, if it's running. (If it's not, we
4073 : * might not even be connected to shared memory, so don't try to call
4074 : * AutoVacWorkerFailed.) Note that we also need to signal it so that it
4075 : * responds to the condition, but we don't do that here, instead waiting
4076 : * for ServerLoop to do it. This way we avoid a ping-pong signaling in
4077 : * quick succession between the autovac launcher and postmaster in case
4078 : * things get ugly.
4079 : */
4080 0 : if (AutoVacLauncherPMChild != NULL)
4081 : {
4082 0 : AutoVacWorkerFailed();
4083 0 : avlauncher_needs_signal = true;
4084 : }
4085 : }
4086 :
4087 :
4088 : /*
4089 : * Create the opts file
4090 : */
4091 : static bool
4092 1758 : CreateOptsFile(int argc, char *argv[], char *fullprogname)
4093 : {
4094 : FILE *fp;
4095 : int i;
4096 :
4097 : #define OPTS_FILE "postmaster.opts"
4098 :
4099 1758 : if ((fp = fopen(OPTS_FILE, "w")) == NULL)
4100 : {
4101 0 : ereport(LOG,
4102 : (errcode_for_file_access(),
4103 : errmsg("could not create file \"%s\": %m", OPTS_FILE)));
4104 0 : return false;
4105 : }
4106 :
4107 1758 : fprintf(fp, "%s", fullprogname);
4108 9202 : for (i = 1; i < argc; i++)
4109 7444 : fprintf(fp, " \"%s\"", argv[i]);
4110 1758 : fputs("\n", fp);
4111 :
4112 1758 : if (fclose(fp))
4113 : {
4114 0 : ereport(LOG,
4115 : (errcode_for_file_access(),
4116 : errmsg("could not write file \"%s\": %m", OPTS_FILE)));
4117 0 : return false;
4118 : }
4119 :
4120 1758 : return true;
4121 : }
4122 :
4123 :
4124 : /*
4125 : * Start a new bgworker.
4126 : * Starting time conditions must have been checked already.
4127 : *
4128 : * Returns true on success, false on failure.
4129 : * In either case, update the RegisteredBgWorker's state appropriately.
4130 : *
4131 : * NB -- this code very roughly matches BackendStartup.
4132 : */
4133 : static bool
4134 5500 : StartBackgroundWorker(RegisteredBgWorker *rw)
4135 : {
4136 : PMChild *bn;
4137 : pid_t worker_pid;
4138 :
4139 : Assert(rw->rw_pid == 0);
4140 :
4141 : /*
4142 : * Allocate and assign the child slot. Note we must do this before
4143 : * forking, so that we can handle failures (out of memory or child-process
4144 : * slots) cleanly.
4145 : *
4146 : * Treat failure as though the worker had crashed. That way, the
4147 : * postmaster will wait a bit before attempting to start it again; if we
4148 : * tried again right away, most likely we'd find ourselves hitting the
4149 : * same resource-exhaustion condition.
4150 : */
4151 5500 : bn = AssignPostmasterChildSlot(B_BG_WORKER);
4152 5500 : if (bn == NULL)
4153 : {
4154 0 : ereport(LOG,
4155 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
4156 : errmsg("no slot available for new background worker process")));
4157 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4158 0 : return false;
4159 : }
4160 5500 : bn->rw = rw;
4161 5500 : bn->bkend_type = B_BG_WORKER;
4162 5500 : bn->bgworker_notify = false;
4163 :
4164 5500 : ereport(DEBUG1,
4165 : (errmsg_internal("starting background worker process \"%s\"",
4166 : rw->rw_worker.bgw_name)));
4167 :
4168 5500 : worker_pid = postmaster_child_launch(B_BG_WORKER, bn->child_slot,
4169 5500 : &rw->rw_worker, sizeof(BackgroundWorker), NULL);
4170 5500 : if (worker_pid == -1)
4171 : {
4172 : /* in postmaster, fork failed ... */
4173 0 : ereport(LOG,
4174 : (errmsg("could not fork background worker process: %m")));
4175 : /* undo what AssignPostmasterChildSlot did */
4176 0 : ReleasePostmasterChildSlot(bn);
4177 :
4178 : /* mark entry as crashed, so we'll try again later */
4179 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4180 0 : return false;
4181 : }
4182 :
4183 : /* in postmaster, fork successful ... */
4184 5500 : rw->rw_pid = worker_pid;
4185 5500 : bn->pid = rw->rw_pid;
4186 5500 : ReportBackgroundWorkerPID(rw);
4187 5500 : return true;
4188 : }
4189 :
4190 : /*
4191 : * Does the current postmaster state require starting a worker with the
4192 : * specified start_time?
4193 : */
4194 : static bool
4195 7460 : bgworker_should_start_now(BgWorkerStartTime start_time)
4196 : {
4197 7460 : switch (pmState)
4198 : {
4199 2 : case PM_NO_CHILDREN:
4200 : case PM_WAIT_CHECKPOINTER:
4201 : case PM_WAIT_DEAD_END:
4202 : case PM_WAIT_XLOG_ARCHIVAL:
4203 : case PM_WAIT_XLOG_SHUTDOWN:
4204 : case PM_WAIT_IO_WORKERS:
4205 : case PM_WAIT_BACKENDS:
4206 : case PM_STOP_BACKENDS:
4207 2 : break;
4208 :
4209 5500 : case PM_RUN:
4210 5500 : if (start_time == BgWorkerStart_RecoveryFinished)
4211 2604 : return true;
4212 : /* fall through */
4213 :
4214 : case PM_HOT_STANDBY:
4215 3192 : if (start_time == BgWorkerStart_ConsistentState)
4216 2896 : return true;
4217 : /* fall through */
4218 :
4219 : case PM_RECOVERY:
4220 : case PM_STARTUP:
4221 : case PM_INIT:
4222 1958 : if (start_time == BgWorkerStart_PostmasterStart)
4223 0 : return true;
4224 : /* fall through */
4225 : }
4226 :
4227 1960 : return false;
4228 : }
4229 :
4230 : /*
4231 : * If the time is right, start background worker(s).
4232 : *
4233 : * As a side effect, the bgworker control variables are set or reset
4234 : * depending on whether more workers may need to be started.
4235 : *
4236 : * We limit the number of workers started per call, to avoid consuming the
4237 : * postmaster's attention for too long when many such requests are pending.
4238 : * As long as StartWorkerNeeded is true, ServerLoop will not block and will
4239 : * call this function again after dealing with any other issues.
4240 : */
4241 : static void
4242 16172 : maybe_start_bgworkers(void)
4243 : {
4244 : #define MAX_BGWORKERS_TO_LAUNCH 100
4245 16172 : int num_launched = 0;
4246 16172 : TimestampTz now = 0;
4247 : dlist_mutable_iter iter;
4248 :
4249 : /*
4250 : * During crash recovery, we have no need to be called until the state
4251 : * transition out of recovery.
4252 : */
4253 16172 : if (FatalError)
4254 : {
4255 0 : StartWorkerNeeded = false;
4256 0 : HaveCrashedWorker = false;
4257 0 : return;
4258 : }
4259 :
4260 : /* Don't need to be called again unless we find a reason for it below */
4261 16172 : StartWorkerNeeded = false;
4262 16172 : HaveCrashedWorker = false;
4263 :
4264 43318 : dlist_foreach_modify(iter, &BackgroundWorkerList)
4265 : {
4266 : RegisteredBgWorker *rw;
4267 :
4268 27146 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
4269 :
4270 : /* ignore if already running */
4271 27146 : if (rw->rw_pid != 0)
4272 13620 : continue;
4273 :
4274 : /* if marked for death, clean up and remove from list */
4275 13526 : if (rw->rw_terminate)
4276 : {
4277 0 : ForgetBackgroundWorker(rw);
4278 0 : continue;
4279 : }
4280 :
4281 : /*
4282 : * If this worker has crashed previously, maybe it needs to be
4283 : * restarted (unless on registration it specified it doesn't want to
4284 : * be restarted at all). Check how long ago did a crash last happen.
4285 : * If the last crash is too recent, don't start it right away; let it
4286 : * be restarted once enough time has passed.
4287 : */
4288 13526 : if (rw->rw_crashed_at != 0)
4289 : {
4290 6068 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
4291 0 : {
4292 : int notify_pid;
4293 :
4294 0 : notify_pid = rw->rw_worker.bgw_notify_pid;
4295 :
4296 0 : ForgetBackgroundWorker(rw);
4297 :
4298 : /* Report worker is gone now. */
4299 0 : if (notify_pid != 0)
4300 0 : kill(notify_pid, SIGUSR1);
4301 :
4302 0 : continue;
4303 : }
4304 :
4305 : /* read system time only when needed */
4306 6068 : if (now == 0)
4307 6068 : now = GetCurrentTimestamp();
4308 :
4309 6068 : if (!TimestampDifferenceExceeds(rw->rw_crashed_at, now,
4310 6068 : rw->rw_worker.bgw_restart_time * 1000))
4311 : {
4312 : /* Set flag to remember that we have workers to start later */
4313 6066 : HaveCrashedWorker = true;
4314 6066 : continue;
4315 : }
4316 : }
4317 :
4318 7460 : if (bgworker_should_start_now(rw->rw_worker.bgw_start_time))
4319 : {
4320 : /* reset crash time before trying to start worker */
4321 5500 : rw->rw_crashed_at = 0;
4322 :
4323 : /*
4324 : * Try to start the worker.
4325 : *
4326 : * On failure, give up processing workers for now, but set
4327 : * StartWorkerNeeded so we'll come back here on the next iteration
4328 : * of ServerLoop to try again. (We don't want to wait, because
4329 : * there might be additional ready-to-run workers.) We could set
4330 : * HaveCrashedWorker as well, since this worker is now marked
4331 : * crashed, but there's no need because the next run of this
4332 : * function will do that.
4333 : */
4334 5500 : if (!StartBackgroundWorker(rw))
4335 : {
4336 0 : StartWorkerNeeded = true;
4337 0 : return;
4338 : }
4339 :
4340 : /*
4341 : * If we've launched as many workers as allowed, quit, but have
4342 : * ServerLoop call us again to look for additional ready-to-run
4343 : * workers. There might not be any, but we'll find out the next
4344 : * time we run.
4345 : */
4346 5500 : if (++num_launched >= MAX_BGWORKERS_TO_LAUNCH)
4347 : {
4348 0 : StartWorkerNeeded = true;
4349 0 : return;
4350 : }
4351 : }
4352 : }
4353 : }
4354 :
4355 : static bool
4356 43056 : maybe_reap_io_worker(int pid)
4357 : {
4358 1255546 : for (int i = 0; i < MAX_IO_WORKERS; ++i)
4359 : {
4360 1217854 : if (io_worker_children[i] &&
4361 118598 : io_worker_children[i]->pid == pid)
4362 : {
4363 5364 : ReleasePostmasterChildSlot(io_worker_children[i]);
4364 :
4365 5364 : --io_worker_count;
4366 5364 : io_worker_children[i] = NULL;
4367 5364 : return true;
4368 : }
4369 : }
4370 37692 : return false;
4371 : }
4372 :
4373 : /*
4374 : * Start or stop IO workers, to close the gap between the number of running
4375 : * workers and the number of configured workers. Used to respond to change of
4376 : * the io_workers GUC (by increasing and decreasing the number of workers), as
4377 : * well as workers terminating in response to errors (by starting
4378 : * "replacement" workers).
4379 : */
4380 : static void
4381 247850 : maybe_adjust_io_workers(void)
4382 : {
4383 247850 : if (!pgaio_workers_enabled())
4384 112 : return;
4385 :
4386 : /*
4387 : * If we're in final shutting down state, then we're just waiting for all
4388 : * processes to exit.
4389 : */
4390 247738 : if (pmState >= PM_WAIT_IO_WORKERS)
4391 7752 : return;
4392 :
4393 : /* Don't start new workers during an immediate shutdown either. */
4394 239986 : if (Shutdown >= ImmediateShutdown)
4395 4458 : return;
4396 :
4397 : /*
4398 : * Don't start new workers if we're in the shutdown phase of a crash
4399 : * restart. But we *do* need to start if we're already starting up again.
4400 : */
4401 235528 : if (FatalError && pmState >= PM_STOP_BACKENDS)
4402 104 : return;
4403 :
4404 : Assert(pmState < PM_WAIT_IO_WORKERS);
4405 :
4406 : /* Not enough running? */
4407 240794 : while (io_worker_count < io_workers)
4408 : {
4409 : PMChild *child;
4410 : int i;
4411 :
4412 : /* find unused entry in io_worker_children array */
4413 11722 : for (i = 0; i < MAX_IO_WORKERS; ++i)
4414 : {
4415 11722 : if (io_worker_children[i] == NULL)
4416 5370 : break;
4417 : }
4418 5370 : if (i == MAX_IO_WORKERS)
4419 0 : elog(ERROR, "could not find a free IO worker slot");
4420 :
4421 : /* Try to launch one. */
4422 5370 : child = StartChildProcess(B_IO_WORKER);
4423 5370 : if (child != NULL)
4424 : {
4425 5370 : io_worker_children[i] = child;
4426 5370 : ++io_worker_count;
4427 : }
4428 : else
4429 0 : break; /* try again next time */
4430 : }
4431 :
4432 : /* Too many running? */
4433 235424 : if (io_worker_count > io_workers)
4434 : {
4435 : /* ask the IO worker in the highest slot to exit */
4436 976 : for (int i = MAX_IO_WORKERS - 1; i >= 0; --i)
4437 : {
4438 976 : if (io_worker_children[i] != NULL)
4439 : {
4440 100 : kill(io_worker_children[i]->pid, SIGUSR2);
4441 100 : break;
4442 : }
4443 : }
4444 : }
4445 : }
4446 :
4447 :
4448 : /*
4449 : * When a backend asks to be notified about worker state changes, we
4450 : * set a flag in its backend entry. The background worker machinery needs
4451 : * to know when such backends exit.
4452 : */
4453 : bool
4454 4038 : PostmasterMarkPIDForWorkerNotify(int pid)
4455 : {
4456 : dlist_iter iter;
4457 : PMChild *bp;
4458 :
4459 9858 : dlist_foreach(iter, &ActiveChildList)
4460 : {
4461 9858 : bp = dlist_container(PMChild, elem, iter.cur);
4462 9858 : if (bp->pid == pid)
4463 : {
4464 4038 : bp->bgworker_notify = true;
4465 4038 : return true;
4466 : }
4467 : }
4468 0 : return false;
4469 : }
4470 :
4471 : #ifdef WIN32
4472 :
4473 : /*
4474 : * Subset implementation of waitpid() for Windows. We assume pid is -1
4475 : * (that is, check all child processes) and options is WNOHANG (don't wait).
4476 : */
4477 : static pid_t
4478 : waitpid(pid_t pid, int *exitstatus, int options)
4479 : {
4480 : win32_deadchild_waitinfo *childinfo;
4481 : DWORD exitcode;
4482 : DWORD dwd;
4483 : ULONG_PTR key;
4484 : OVERLAPPED *ovl;
4485 :
4486 : /* Try to consume one win32_deadchild_waitinfo from the queue. */
4487 : if (!GetQueuedCompletionStatus(win32ChildQueue, &dwd, &key, &ovl, 0))
4488 : {
4489 : errno = EAGAIN;
4490 : return -1;
4491 : }
4492 :
4493 : childinfo = (win32_deadchild_waitinfo *) key;
4494 : pid = childinfo->procId;
4495 :
4496 : /*
4497 : * Remove handle from wait - required even though it's set to wait only
4498 : * once
4499 : */
4500 : UnregisterWaitEx(childinfo->waitHandle, NULL);
4501 :
4502 : if (!GetExitCodeProcess(childinfo->procHandle, &exitcode))
4503 : {
4504 : /*
4505 : * Should never happen. Inform user and set a fixed exitcode.
4506 : */
4507 : write_stderr("could not read exit code for process\n");
4508 : exitcode = 255;
4509 : }
4510 : *exitstatus = exitcode;
4511 :
4512 : /*
4513 : * Close the process handle. Only after this point can the PID can be
4514 : * recycled by the kernel.
4515 : */
4516 : CloseHandle(childinfo->procHandle);
4517 :
4518 : /*
4519 : * Free struct that was allocated before the call to
4520 : * RegisterWaitForSingleObject()
4521 : */
4522 : pfree(childinfo);
4523 :
4524 : return pid;
4525 : }
4526 :
4527 : /*
4528 : * Note! Code below executes on a thread pool! All operations must
4529 : * be thread safe! Note that elog() and friends must *not* be used.
4530 : */
4531 : static void WINAPI
4532 : pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
4533 : {
4534 : /* Should never happen, since we use INFINITE as timeout value. */
4535 : if (TimerOrWaitFired)
4536 : return;
4537 :
4538 : /*
4539 : * Post the win32_deadchild_waitinfo object for waitpid() to deal with. If
4540 : * that fails, we leak the object, but we also leak a whole process and
4541 : * get into an unrecoverable state, so there's not much point in worrying
4542 : * about that. We'd like to panic, but we can't use that infrastructure
4543 : * from this thread.
4544 : */
4545 : if (!PostQueuedCompletionStatus(win32ChildQueue,
4546 : 0,
4547 : (ULONG_PTR) lpParameter,
4548 : NULL))
4549 : write_stderr("could not post child completion status\n");
4550 :
4551 : /* Queue SIGCHLD signal. */
4552 : pg_queue_signal(SIGCHLD);
4553 : }
4554 :
4555 : /*
4556 : * Queue a waiter to signal when this child dies. The wait will be handled
4557 : * automatically by an operating system thread pool. The memory and the
4558 : * process handle will be freed by a later call to waitpid().
4559 : */
4560 : void
4561 : pgwin32_register_deadchild_callback(HANDLE procHandle, DWORD procId)
4562 : {
4563 : win32_deadchild_waitinfo *childinfo;
4564 :
4565 : childinfo = palloc(sizeof(win32_deadchild_waitinfo));
4566 : childinfo->procHandle = procHandle;
4567 : childinfo->procId = procId;
4568 :
4569 : if (!RegisterWaitForSingleObject(&childinfo->waitHandle,
4570 : procHandle,
4571 : pgwin32_deadchild_callback,
4572 : childinfo,
4573 : INFINITE,
4574 : WT_EXECUTEONLYONCE | WT_EXECUTEINWAITTHREAD))
4575 : ereport(FATAL,
4576 : (errmsg_internal("could not register process for wait: error code %lu",
4577 : GetLastError())));
4578 : }
4579 :
4580 : #endif /* WIN32 */
4581 :
4582 : /*
4583 : * Initialize one and only handle for monitoring postmaster death.
4584 : *
4585 : * Called once in the postmaster, so that child processes can subsequently
4586 : * monitor if their parent is dead.
4587 : */
4588 : static void
4589 1758 : InitPostmasterDeathWatchHandle(void)
4590 : {
4591 : #ifndef WIN32
4592 :
4593 : /*
4594 : * Create a pipe. Postmaster holds the write end of the pipe open
4595 : * (POSTMASTER_FD_OWN), and children hold the read end. Children can pass
4596 : * the read file descriptor to select() to wake up in case postmaster
4597 : * dies, or check for postmaster death with a (read() == 0). Children must
4598 : * close the write end as soon as possible after forking, because EOF
4599 : * won't be signaled in the read end until all processes have closed the
4600 : * write fd. That is taken care of in ClosePostmasterPorts().
4601 : */
4602 : Assert(MyProcPid == PostmasterPid);
4603 1758 : if (pipe(postmaster_alive_fds) < 0)
4604 0 : ereport(FATAL,
4605 : (errcode_for_file_access(),
4606 : errmsg_internal("could not create pipe to monitor postmaster death: %m")));
4607 :
4608 : /* Notify fd.c that we've eaten two FDs for the pipe. */
4609 1758 : ReserveExternalFD();
4610 1758 : ReserveExternalFD();
4611 :
4612 : /*
4613 : * Set O_NONBLOCK to allow testing for the fd's presence with a read()
4614 : * call.
4615 : */
4616 1758 : if (fcntl(postmaster_alive_fds[POSTMASTER_FD_WATCH], F_SETFL, O_NONBLOCK) == -1)
4617 0 : ereport(FATAL,
4618 : (errcode_for_socket_access(),
4619 : errmsg_internal("could not set postmaster death monitoring pipe to nonblocking mode: %m")));
4620 : #else
4621 :
4622 : /*
4623 : * On Windows, we use a process handle for the same purpose.
4624 : */
4625 : if (DuplicateHandle(GetCurrentProcess(),
4626 : GetCurrentProcess(),
4627 : GetCurrentProcess(),
4628 : &PostmasterHandle,
4629 : 0,
4630 : TRUE,
4631 : DUPLICATE_SAME_ACCESS) == 0)
4632 : ereport(FATAL,
4633 : (errmsg_internal("could not duplicate postmaster handle: error code %lu",
4634 : GetLastError())));
4635 : #endif /* WIN32 */
4636 1758 : }
|
