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 2828 : btmask(BackendType t)
150 : {
151 2828 : BackendTypeMask mask = {.mask = 1 << t};
152 :
153 2828 : return mask;
154 : }
155 :
156 : static inline BackendTypeMask
157 28844 : btmask_add_n(BackendTypeMask mask, int nargs, BackendType *t)
158 : {
159 127244 : for (int i = 0; i < nargs; i++)
160 98400 : mask.mask |= 1 << t[i];
161 28844 : 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 8096 : btmask_del(BackendTypeMask mask, BackendType t)
172 : {
173 8096 : mask.mask &= ~(1 << t);
174 8096 : return mask;
175 : }
176 :
177 : static inline BackendTypeMask
178 4790 : btmask_all_except_n(int nargs, BackendType *t)
179 : {
180 4790 : BackendTypeMask mask = BTYPE_MASK_ALL;
181 :
182 12886 : for (int i = 0; i < nargs; i++)
183 8096 : mask = btmask_del(mask, t[i]);
184 4790 : 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 229158 : btmask_contains(BackendTypeMask mask, BackendType t)
195 : {
196 229158 : 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 1704 : PostmasterMain(int argc, char *argv[])
494 : {
495 : int opt;
496 : int status;
497 1704 : char *userDoption = NULL;
498 1704 : bool listen_addr_saved = false;
499 1704 : char *output_config_variable = NULL;
500 :
501 1704 : InitProcessGlobals();
502 :
503 1704 : PostmasterPid = MyProcPid;
504 :
505 1704 : 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 1704 : 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 1704 : PostmasterContext = AllocSetContextCreate(TopMemoryContext,
531 : "Postmaster",
532 : ALLOCSET_DEFAULT_SIZES);
533 1704 : MemoryContextSwitchTo(PostmasterContext);
534 :
535 : /* Initialize paths to installation files */
536 1704 : 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 1704 : pqinitmask();
548 1704 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
549 :
550 1704 : pqsignal(SIGHUP, handle_pm_reload_request_signal);
551 1704 : pqsignal(SIGINT, handle_pm_shutdown_request_signal);
552 1704 : pqsignal(SIGQUIT, handle_pm_shutdown_request_signal);
553 1704 : pqsignal(SIGTERM, handle_pm_shutdown_request_signal);
554 1704 : pqsignal(SIGALRM, SIG_IGN); /* ignored */
555 1704 : pqsignal(SIGPIPE, SIG_IGN); /* ignored */
556 1704 : pqsignal(SIGUSR1, handle_pm_pmsignal_signal);
557 1704 : pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
558 1704 : pqsignal(SIGCHLD, handle_pm_child_exit_signal);
559 :
560 : /* This may configure SIGURG, depending on platform. */
561 1704 : InitializeWaitEventSupport();
562 1704 : 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 1704 : pqsignal(SIGTTIN, SIG_IGN); /* ignored */
573 : #endif
574 : #ifdef SIGTTOU
575 1704 : pqsignal(SIGTTOU, SIG_IGN); /* ignored */
576 : #endif
577 :
578 : /* ignore SIGXFSZ, so that ulimit violations work like disk full */
579 : #ifdef SIGXFSZ
580 1704 : pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
581 : #endif
582 :
583 : /* Begin accepting signals. */
584 1704 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
585 :
586 : /*
587 : * Options setup
588 : */
589 1704 : InitializeGUCOptions();
590 :
591 1704 : 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 6280 : while ((opt = getopt(argc, argv, "B:bC:c:D:d:EeFf:h:ijk:lN:OPp:r:S:sTt:W:-:")) != -1)
599 : {
600 4580 : switch (opt)
601 : {
602 0 : case 'B':
603 0 : SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
604 0 : break;
605 :
606 88 : case 'b':
607 : /* Undocumented flag used for binary upgrades */
608 88 : IsBinaryUpgrade = true;
609 88 : break;
610 :
611 6 : case 'C':
612 6 : output_config_variable = strdup(optarg);
613 6 : break;
614 :
615 1380 : 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 1380 : 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 2310 : ParseLongOption(optarg, &name, &value);
635 2310 : 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 2308 : SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
650 2306 : pfree(name);
651 2306 : pfree(value);
652 2306 : break;
653 : }
654 :
655 1700 : case 'D':
656 1700 : userDoption = strdup(optarg);
657 1700 : 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 :
671 182 : case 'F':
672 182 : SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
673 182 : 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 :
696 182 : case 'k':
697 182 : SetConfigOption("unix_socket_directories", optarg, PGC_POSTMASTER, PGC_S_ARGV);
698 182 : 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 :
716 112 : case 'p':
717 112 : SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
718 112 : 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 1700 : 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 1700 : if (!SelectConfigFiles(userDoption, progname))
786 0 : ExitPostmaster(2);
787 :
788 1698 : 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 1696 : checkDataDir();
830 :
831 : /* Check that pg_control exists */
832 1696 : checkControlFile();
833 :
834 : /* And switch working directory into it */
835 1696 : ChangeToDataDir();
836 :
837 : /*
838 : * Check for invalid combinations of GUC settings.
839 : */
840 1696 : 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 1696 : 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 1696 : 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 1696 : if (summarize_wal && wal_level == WAL_LEVEL_MINIMAL)
855 0 : ereport(ERROR,
856 : (errmsg("WAL cannot be summarized when \"wal_level\" is \"minimal\"")));
857 :
858 : /*
859 : * Other one-time internal sanity checks can go here, if they are fast.
860 : * (Put any slow processing further down, after postmaster.pid creation.)
861 : */
862 1696 : if (!CheckDateTokenTables())
863 : {
864 0 : write_stderr("%s: invalid datetoken tables, please fix\n", progname);
865 0 : ExitPostmaster(1);
866 : }
867 :
868 : /*
869 : * Now that we are done processing the postmaster arguments, reset
870 : * getopt(3) library so that it will work correctly in subprocesses.
871 : */
872 1696 : optind = 1;
873 : #ifdef HAVE_INT_OPTRESET
874 : optreset = 1; /* some systems need this too */
875 : #endif
876 :
877 : /* For debugging: display postmaster environment */
878 1696 : if (message_level_is_interesting(DEBUG3))
879 : {
880 : #if !defined(WIN32) || defined(_MSC_VER)
881 : extern char **environ;
882 : #endif
883 : char **p;
884 : StringInfoData si;
885 :
886 4 : initStringInfo(&si);
887 :
888 4 : appendStringInfoString(&si, "initial environment dump:");
889 168 : for (p = environ; *p; ++p)
890 164 : appendStringInfo(&si, "\n%s", *p);
891 :
892 4 : ereport(DEBUG3, errmsg_internal("%s", si.data));
893 4 : pfree(si.data);
894 : }
895 :
896 : /*
897 : * Create lockfile for data directory.
898 : *
899 : * We want to do this before we try to grab the input sockets, because the
900 : * data directory interlock is more reliable than the socket-file
901 : * interlock (thanks to whoever decided to put socket files in /tmp :-().
902 : * For the same reason, it's best to grab the TCP socket(s) before the
903 : * Unix socket(s).
904 : *
905 : * Also note that this internally sets up the on_proc_exit function that
906 : * is responsible for removing both data directory and socket lockfiles;
907 : * so it must happen before opening sockets so that at exit, the socket
908 : * lockfiles go away after CloseServerPorts runs.
909 : */
910 1696 : CreateDataDirLockFile(true);
911 :
912 : /*
913 : * Read the control file (for error checking and config info).
914 : *
915 : * Since we verify the control file's CRC, this has a useful side effect
916 : * on machines where we need a run-time test for CRC support instructions.
917 : * The postmaster will do the test once at startup, and then its child
918 : * processes will inherit the correct function pointer and not need to
919 : * repeat the test.
920 : */
921 1694 : LocalProcessControlFile(false);
922 :
923 : /*
924 : * Register the apply launcher. It's probably a good idea to call this
925 : * before any modules had a chance to take the background worker slots.
926 : */
927 1694 : ApplyLauncherRegister();
928 :
929 : /*
930 : * process any libraries that should be preloaded at postmaster start
931 : */
932 1694 : process_shared_preload_libraries();
933 :
934 : /*
935 : * Initialize SSL library, if specified.
936 : */
937 : #ifdef USE_SSL
938 1694 : if (EnableSSL)
939 : {
940 62 : (void) secure_initialize(true);
941 52 : LoadedSSL = true;
942 : }
943 : #endif
944 :
945 : /*
946 : * Now that loadable modules have had their chance to alter any GUCs,
947 : * calculate MaxBackends and initialize the machinery to track child
948 : * processes.
949 : */
950 1684 : InitializeMaxBackends();
951 1684 : InitPostmasterChildSlots();
952 :
953 : /*
954 : * Calculate the size of the PGPROC fast-path lock arrays.
955 : */
956 1684 : InitializeFastPathLocks();
957 :
958 : /*
959 : * Give preloaded libraries a chance to request additional shared memory.
960 : */
961 1684 : process_shmem_requests();
962 :
963 : /*
964 : * Now that loadable modules have had their chance to request additional
965 : * shared memory, determine the value of any runtime-computed GUCs that
966 : * depend on the amount of shared memory required.
967 : */
968 1684 : InitializeShmemGUCs();
969 :
970 : /*
971 : * Now that modules have been loaded, we can process any custom resource
972 : * managers specified in the wal_consistency_checking GUC.
973 : */
974 1684 : InitializeWalConsistencyChecking();
975 :
976 : /*
977 : * If -C was specified with a runtime-computed GUC, we held off printing
978 : * the value earlier, as the GUC was not yet initialized. We handle -C
979 : * for most GUCs before we lock the data directory so that the option may
980 : * be used on a running server. However, a handful of GUCs are runtime-
981 : * computed and do not have meaningful values until after locking the data
982 : * directory, and we cannot safely calculate their values earlier on a
983 : * running server. At this point, such GUCs should be properly
984 : * initialized, and we haven't yet set up shared memory, so this is a good
985 : * time to handle the -C option for these special GUCs.
986 : */
987 1684 : if (output_config_variable != NULL)
988 : {
989 2 : const char *config_val = GetConfigOption(output_config_variable,
990 : false, false);
991 :
992 2 : puts(config_val ? config_val : "");
993 2 : ExitPostmaster(0);
994 : }
995 :
996 : /*
997 : * Set up shared memory and semaphores.
998 : *
999 : * Note: if using SysV shmem and/or semas, each postmaster startup will
1000 : * normally choose the same IPC keys. This helps ensure that we will
1001 : * clean up dead IPC objects if the postmaster crashes and is restarted.
1002 : */
1003 1682 : CreateSharedMemoryAndSemaphores();
1004 :
1005 : /*
1006 : * Estimate number of openable files. This must happen after setting up
1007 : * semaphores, because on some platforms semaphores count as open files.
1008 : */
1009 1680 : set_max_safe_fds();
1010 :
1011 : /*
1012 : * Initialize pipe (or process handle on Windows) that allows children to
1013 : * wake up from sleep on postmaster death.
1014 : */
1015 1680 : InitPostmasterDeathWatchHandle();
1016 :
1017 : #ifdef WIN32
1018 :
1019 : /*
1020 : * Initialize I/O completion port used to deliver list of dead children.
1021 : */
1022 : win32ChildQueue = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 1);
1023 : if (win32ChildQueue == NULL)
1024 : ereport(FATAL,
1025 : (errmsg("could not create I/O completion port for child queue")));
1026 : #endif
1027 :
1028 : #ifdef EXEC_BACKEND
1029 : /* Write out nondefault GUC settings for child processes to use */
1030 : write_nondefault_variables(PGC_POSTMASTER);
1031 :
1032 : /*
1033 : * Clean out the temp directory used to transmit parameters to child
1034 : * processes (see internal_forkexec). We must do this before launching
1035 : * any child processes, else we have a race condition: we could remove a
1036 : * parameter file before the child can read it. It should be safe to do
1037 : * so now, because we verified earlier that there are no conflicting
1038 : * Postgres processes in this data directory.
1039 : */
1040 : RemovePgTempFilesInDir(PG_TEMP_FILES_DIR, true, false);
1041 : #endif
1042 :
1043 : /*
1044 : * Forcibly remove the files signaling a standby promotion request.
1045 : * Otherwise, the existence of those files triggers a promotion too early,
1046 : * whether a user wants that or not.
1047 : *
1048 : * This removal of files is usually unnecessary because they can exist
1049 : * only during a few moments during a standby promotion. However there is
1050 : * a race condition: if pg_ctl promote is executed and creates the files
1051 : * during a promotion, the files can stay around even after the server is
1052 : * brought up to be the primary. Then, if a new standby starts by using
1053 : * the backup taken from the new primary, the files can exist at server
1054 : * startup and must be removed in order to avoid an unexpected promotion.
1055 : *
1056 : * Note that promotion signal files need to be removed before the startup
1057 : * process is invoked. Because, after that, they can be used by
1058 : * postmaster's SIGUSR1 signal handler.
1059 : */
1060 1680 : RemovePromoteSignalFiles();
1061 :
1062 : /* Do the same for logrotate signal file */
1063 1680 : RemoveLogrotateSignalFiles();
1064 :
1065 : /* Remove any outdated file holding the current log filenames. */
1066 1680 : if (unlink(LOG_METAINFO_DATAFILE) < 0 && errno != ENOENT)
1067 0 : ereport(LOG,
1068 : (errcode_for_file_access(),
1069 : errmsg("could not remove file \"%s\": %m",
1070 : LOG_METAINFO_DATAFILE)));
1071 :
1072 : /*
1073 : * If enabled, start up syslogger collection subprocess
1074 : */
1075 1680 : if (Logging_collector)
1076 2 : StartSysLogger();
1077 :
1078 : /*
1079 : * Reset whereToSendOutput from DestDebug (its starting state) to
1080 : * DestNone. This stops ereport from sending log messages to stderr unless
1081 : * Log_destination permits. We don't do this until the postmaster is
1082 : * fully launched, since startup failures may as well be reported to
1083 : * stderr.
1084 : *
1085 : * If we are in fact disabling logging to stderr, first emit a log message
1086 : * saying so, to provide a breadcrumb trail for users who may not remember
1087 : * that their logging is configured to go somewhere else.
1088 : */
1089 1680 : if (!(Log_destination & LOG_DESTINATION_STDERR))
1090 0 : ereport(LOG,
1091 : (errmsg("ending log output to stderr"),
1092 : errhint("Future log output will go to log destination \"%s\".",
1093 : Log_destination_string)));
1094 :
1095 1680 : whereToSendOutput = DestNone;
1096 :
1097 : /*
1098 : * Report server startup in log. While we could emit this much earlier,
1099 : * it seems best to do so after starting the log collector, if we intend
1100 : * to use one.
1101 : */
1102 1680 : ereport(LOG,
1103 : (errmsg("starting %s", PG_VERSION_STR)));
1104 :
1105 : /*
1106 : * Establish input sockets.
1107 : *
1108 : * First set up an on_proc_exit function that's charged with closing the
1109 : * sockets again at postmaster shutdown.
1110 : */
1111 1680 : ListenSockets = palloc(MAXLISTEN * sizeof(pgsocket));
1112 1680 : on_proc_exit(CloseServerPorts, 0);
1113 :
1114 1680 : if (ListenAddresses)
1115 : {
1116 : char *rawstring;
1117 : List *elemlist;
1118 : ListCell *l;
1119 1680 : int success = 0;
1120 :
1121 : /* Need a modifiable copy of ListenAddresses */
1122 1680 : rawstring = pstrdup(ListenAddresses);
1123 :
1124 : /* Parse string into list of hostnames */
1125 1680 : if (!SplitGUCList(rawstring, ',', &elemlist))
1126 : {
1127 : /* syntax error in list */
1128 0 : ereport(FATAL,
1129 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1130 : errmsg("invalid list syntax in parameter \"%s\"",
1131 : "listen_addresses")));
1132 : }
1133 :
1134 1738 : foreach(l, elemlist)
1135 : {
1136 58 : char *curhost = (char *) lfirst(l);
1137 :
1138 58 : if (strcmp(curhost, "*") == 0)
1139 0 : status = ListenServerPort(AF_UNSPEC, NULL,
1140 0 : (unsigned short) PostPortNumber,
1141 : NULL,
1142 : ListenSockets,
1143 : &NumListenSockets,
1144 : MAXLISTEN);
1145 : else
1146 58 : status = ListenServerPort(AF_UNSPEC, curhost,
1147 58 : (unsigned short) PostPortNumber,
1148 : NULL,
1149 : ListenSockets,
1150 : &NumListenSockets,
1151 : MAXLISTEN);
1152 :
1153 58 : if (status == STATUS_OK)
1154 : {
1155 58 : success++;
1156 : /* record the first successful host addr in lockfile */
1157 58 : if (!listen_addr_saved)
1158 : {
1159 58 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, curhost);
1160 58 : listen_addr_saved = true;
1161 : }
1162 : }
1163 : else
1164 0 : ereport(WARNING,
1165 : (errmsg("could not create listen socket for \"%s\"",
1166 : curhost)));
1167 : }
1168 :
1169 1680 : if (!success && elemlist != NIL)
1170 0 : ereport(FATAL,
1171 : (errmsg("could not create any TCP/IP sockets")));
1172 :
1173 1680 : list_free(elemlist);
1174 1680 : pfree(rawstring);
1175 : }
1176 :
1177 : #ifdef USE_BONJOUR
1178 : /* Register for Bonjour only if we opened TCP socket(s) */
1179 : if (enable_bonjour && NumListenSockets > 0)
1180 : {
1181 : DNSServiceErrorType err;
1182 :
1183 : /*
1184 : * We pass 0 for interface_index, which will result in registering on
1185 : * all "applicable" interfaces. It's not entirely clear from the
1186 : * DNS-SD docs whether this would be appropriate if we have bound to
1187 : * just a subset of the available network interfaces.
1188 : */
1189 : err = DNSServiceRegister(&bonjour_sdref,
1190 : 0,
1191 : 0,
1192 : bonjour_name,
1193 : "_postgresql._tcp.",
1194 : NULL,
1195 : NULL,
1196 : pg_hton16(PostPortNumber),
1197 : 0,
1198 : NULL,
1199 : NULL,
1200 : NULL);
1201 : if (err != kDNSServiceErr_NoError)
1202 : ereport(LOG,
1203 : (errmsg("DNSServiceRegister() failed: error code %ld",
1204 : (long) err)));
1205 :
1206 : /*
1207 : * We don't bother to read the mDNS daemon's reply, and we expect that
1208 : * it will automatically terminate our registration when the socket is
1209 : * closed at postmaster termination. So there's nothing more to be
1210 : * done here. However, the bonjour_sdref is kept around so that
1211 : * forked children can close their copies of the socket.
1212 : */
1213 : }
1214 : #endif
1215 :
1216 1680 : if (Unix_socket_directories)
1217 : {
1218 : char *rawstring;
1219 : List *elemlist;
1220 : ListCell *l;
1221 1680 : int success = 0;
1222 :
1223 : /* Need a modifiable copy of Unix_socket_directories */
1224 1680 : rawstring = pstrdup(Unix_socket_directories);
1225 :
1226 : /* Parse string into list of directories */
1227 1680 : if (!SplitDirectoriesString(rawstring, ',', &elemlist))
1228 : {
1229 : /* syntax error in list */
1230 0 : ereport(FATAL,
1231 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1232 : errmsg("invalid list syntax in parameter \"%s\"",
1233 : "unix_socket_directories")));
1234 : }
1235 :
1236 3358 : foreach(l, elemlist)
1237 : {
1238 1678 : char *socketdir = (char *) lfirst(l);
1239 :
1240 1678 : status = ListenServerPort(AF_UNIX, NULL,
1241 1678 : (unsigned short) PostPortNumber,
1242 : socketdir,
1243 : ListenSockets,
1244 : &NumListenSockets,
1245 : MAXLISTEN);
1246 :
1247 1678 : if (status == STATUS_OK)
1248 : {
1249 1678 : success++;
1250 : /* record the first successful Unix socket in lockfile */
1251 1678 : if (success == 1)
1252 1678 : AddToDataDirLockFile(LOCK_FILE_LINE_SOCKET_DIR, socketdir);
1253 : }
1254 : else
1255 0 : ereport(WARNING,
1256 : (errmsg("could not create Unix-domain socket in directory \"%s\"",
1257 : socketdir)));
1258 : }
1259 :
1260 1680 : if (!success && elemlist != NIL)
1261 0 : ereport(FATAL,
1262 : (errmsg("could not create any Unix-domain sockets")));
1263 :
1264 1680 : list_free_deep(elemlist);
1265 1680 : pfree(rawstring);
1266 : }
1267 :
1268 : /*
1269 : * check that we have some socket to listen on
1270 : */
1271 1680 : if (NumListenSockets == 0)
1272 0 : ereport(FATAL,
1273 : (errmsg("no socket created for listening")));
1274 :
1275 : /*
1276 : * If no valid TCP ports, write an empty line for listen address,
1277 : * indicating the Unix socket must be used. Note that this line is not
1278 : * added to the lock file until there is a socket backing it.
1279 : */
1280 1680 : if (!listen_addr_saved)
1281 1622 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, "");
1282 :
1283 : /*
1284 : * Record postmaster options. We delay this till now to avoid recording
1285 : * bogus options (eg, unusable port number).
1286 : */
1287 1680 : if (!CreateOptsFile(argc, argv, my_exec_path))
1288 0 : ExitPostmaster(1);
1289 :
1290 : /*
1291 : * Write the external PID file if requested
1292 : */
1293 1680 : if (external_pid_file)
1294 : {
1295 0 : FILE *fpidfile = fopen(external_pid_file, "w");
1296 :
1297 0 : if (fpidfile)
1298 : {
1299 0 : fprintf(fpidfile, "%d\n", MyProcPid);
1300 0 : fclose(fpidfile);
1301 :
1302 : /* Make PID file world readable */
1303 0 : if (chmod(external_pid_file, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH) != 0)
1304 0 : write_stderr("%s: could not change permissions of external PID file \"%s\": %m\n",
1305 : progname, external_pid_file);
1306 : }
1307 : else
1308 0 : write_stderr("%s: could not write external PID file \"%s\": %m\n",
1309 : progname, external_pid_file);
1310 :
1311 0 : on_proc_exit(unlink_external_pid_file, 0);
1312 : }
1313 :
1314 : /*
1315 : * Remove old temporary files. At this point there can be no other
1316 : * Postgres processes running in this directory, so this should be safe.
1317 : */
1318 1680 : RemovePgTempFiles();
1319 :
1320 : /*
1321 : * Initialize the autovacuum subsystem (again, no process start yet)
1322 : */
1323 1680 : autovac_init();
1324 :
1325 : /*
1326 : * Load configuration files for client authentication.
1327 : */
1328 1680 : if (!load_hba())
1329 : {
1330 : /*
1331 : * It makes no sense to continue if we fail to load the HBA file,
1332 : * since there is no way to connect to the database in this case.
1333 : */
1334 0 : ereport(FATAL,
1335 : /* translator: %s is a configuration file */
1336 : (errmsg("could not load %s", HbaFileName)));
1337 : }
1338 1680 : if (!load_ident())
1339 : {
1340 : /*
1341 : * We can start up without the IDENT file, although it means that you
1342 : * cannot log in using any of the authentication methods that need a
1343 : * user name mapping. load_ident() already logged the details of error
1344 : * to the log.
1345 : */
1346 : }
1347 :
1348 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1349 :
1350 : /*
1351 : * On macOS, libintl replaces setlocale() with a version that calls
1352 : * CFLocaleCopyCurrent() when its second argument is "" and every relevant
1353 : * environment variable is unset or empty. CFLocaleCopyCurrent() makes
1354 : * the process multithreaded. The postmaster calls sigprocmask() and
1355 : * calls fork() without an immediate exec(), both of which have undefined
1356 : * behavior in a multithreaded program. A multithreaded postmaster is the
1357 : * normal case on Windows, which offers neither fork() nor sigprocmask().
1358 : * Currently, macOS is the only platform having pthread_is_threaded_np(),
1359 : * so we need not worry whether this HINT is appropriate elsewhere.
1360 : */
1361 : if (pthread_is_threaded_np() != 0)
1362 : ereport(FATAL,
1363 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1364 : errmsg("postmaster became multithreaded during startup"),
1365 : errhint("Set the LC_ALL environment variable to a valid locale.")));
1366 : #endif
1367 :
1368 : /*
1369 : * Remember postmaster startup time
1370 : */
1371 1680 : PgStartTime = GetCurrentTimestamp();
1372 :
1373 : /*
1374 : * Report postmaster status in the postmaster.pid file, to allow pg_ctl to
1375 : * see what's happening.
1376 : */
1377 1680 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STARTING);
1378 :
1379 1680 : UpdatePMState(PM_STARTUP);
1380 :
1381 : /* Make sure we can perform I/O while starting up. */
1382 1680 : maybe_adjust_io_workers();
1383 :
1384 : /* Start bgwriter and checkpointer so they can help with recovery */
1385 1680 : if (CheckpointerPMChild == NULL)
1386 1680 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
1387 1680 : if (BgWriterPMChild == NULL)
1388 1680 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
1389 :
1390 : /*
1391 : * We're ready to rock and roll...
1392 : */
1393 1680 : StartupPMChild = StartChildProcess(B_STARTUP);
1394 : Assert(StartupPMChild != NULL);
1395 1680 : StartupStatus = STARTUP_RUNNING;
1396 :
1397 : /* Some workers may be scheduled to start now */
1398 1680 : maybe_start_bgworkers();
1399 :
1400 1680 : status = ServerLoop();
1401 :
1402 : /*
1403 : * ServerLoop probably shouldn't ever return, but if it does, close down.
1404 : */
1405 0 : ExitPostmaster(status != STATUS_OK);
1406 :
1407 : abort(); /* not reached */
1408 : }
1409 :
1410 :
1411 : /*
1412 : * on_proc_exit callback to close server's listen sockets
1413 : */
1414 : static void
1415 1674 : CloseServerPorts(int status, Datum arg)
1416 : {
1417 : int i;
1418 :
1419 : /*
1420 : * First, explicitly close all the socket FDs. We used to just let this
1421 : * happen implicitly at postmaster exit, but it's better to close them
1422 : * before we remove the postmaster.pid lockfile; otherwise there's a race
1423 : * condition if a new postmaster wants to re-use the TCP port number.
1424 : */
1425 3406 : for (i = 0; i < NumListenSockets; i++)
1426 : {
1427 1732 : if (closesocket(ListenSockets[i]) != 0)
1428 0 : elog(LOG, "could not close listen socket: %m");
1429 : }
1430 1674 : NumListenSockets = 0;
1431 :
1432 : /*
1433 : * Next, remove any filesystem entries for Unix sockets. To avoid race
1434 : * conditions against incoming postmasters, this must happen after closing
1435 : * the sockets and before removing lock files.
1436 : */
1437 1674 : RemoveSocketFiles();
1438 :
1439 : /*
1440 : * We don't do anything about socket lock files here; those will be
1441 : * removed in a later on_proc_exit callback.
1442 : */
1443 1674 : }
1444 :
1445 : /*
1446 : * on_proc_exit callback to delete external_pid_file
1447 : */
1448 : static void
1449 0 : unlink_external_pid_file(int status, Datum arg)
1450 : {
1451 0 : if (external_pid_file)
1452 0 : unlink(external_pid_file);
1453 0 : }
1454 :
1455 :
1456 : /*
1457 : * Compute and check the directory paths to files that are part of the
1458 : * installation (as deduced from the postgres executable's own location)
1459 : */
1460 : static void
1461 1704 : getInstallationPaths(const char *argv0)
1462 : {
1463 : DIR *pdir;
1464 :
1465 : /* Locate the postgres executable itself */
1466 1704 : if (find_my_exec(argv0, my_exec_path) < 0)
1467 0 : ereport(FATAL,
1468 : (errmsg("%s: could not locate my own executable path", argv0)));
1469 :
1470 : #ifdef EXEC_BACKEND
1471 : /* Locate executable backend before we change working directory */
1472 : if (find_other_exec(argv0, "postgres", PG_BACKEND_VERSIONSTR,
1473 : postgres_exec_path) < 0)
1474 : ereport(FATAL,
1475 : (errmsg("%s: could not locate matching postgres executable",
1476 : argv0)));
1477 : #endif
1478 :
1479 : /*
1480 : * Locate the pkglib directory --- this has to be set early in case we try
1481 : * to load any modules from it in response to postgresql.conf entries.
1482 : */
1483 1704 : get_pkglib_path(my_exec_path, pkglib_path);
1484 :
1485 : /*
1486 : * Verify that there's a readable directory there; otherwise the Postgres
1487 : * installation is incomplete or corrupt. (A typical cause of this
1488 : * failure is that the postgres executable has been moved or hardlinked to
1489 : * some directory that's not a sibling of the installation lib/
1490 : * directory.)
1491 : */
1492 1704 : pdir = AllocateDir(pkglib_path);
1493 1704 : if (pdir == NULL)
1494 0 : ereport(ERROR,
1495 : (errcode_for_file_access(),
1496 : errmsg("could not open directory \"%s\": %m",
1497 : pkglib_path),
1498 : errhint("This may indicate an incomplete PostgreSQL installation, or that the file \"%s\" has been moved away from its proper location.",
1499 : my_exec_path)));
1500 1704 : FreeDir(pdir);
1501 :
1502 : /*
1503 : * It's not worth checking the share/ directory. If the lib/ directory is
1504 : * there, then share/ probably is too.
1505 : */
1506 1704 : }
1507 :
1508 : /*
1509 : * Check that pg_control exists in the correct location in the data directory.
1510 : *
1511 : * No attempt is made to validate the contents of pg_control here. This is
1512 : * just a sanity check to see if we are looking at a real data directory.
1513 : */
1514 : static void
1515 1696 : checkControlFile(void)
1516 : {
1517 : char path[MAXPGPATH];
1518 : FILE *fp;
1519 :
1520 1696 : snprintf(path, sizeof(path), "%s/%s", DataDir, XLOG_CONTROL_FILE);
1521 :
1522 1696 : fp = AllocateFile(path, PG_BINARY_R);
1523 1696 : if (fp == NULL)
1524 : {
1525 0 : write_stderr("%s: could not find the database system\n"
1526 : "Expected to find it in the directory \"%s\",\n"
1527 : "but could not open file \"%s\": %m\n",
1528 : progname, DataDir, path);
1529 0 : ExitPostmaster(2);
1530 : }
1531 1696 : FreeFile(fp);
1532 1696 : }
1533 :
1534 : /*
1535 : * Determine how long should we let ServerLoop sleep, in milliseconds.
1536 : *
1537 : * In normal conditions we wait at most one minute, to ensure that the other
1538 : * background tasks handled by ServerLoop get done even when no requests are
1539 : * arriving. However, if there are background workers waiting to be started,
1540 : * we don't actually sleep so that they are quickly serviced. Other exception
1541 : * cases are as shown in the code.
1542 : */
1543 : static int
1544 246586 : DetermineSleepTime(void)
1545 : {
1546 246586 : TimestampTz next_wakeup = 0;
1547 :
1548 : /*
1549 : * Normal case: either there are no background workers at all, or we're in
1550 : * a shutdown sequence (during which we ignore bgworkers altogether).
1551 : */
1552 246586 : if (Shutdown > NoShutdown ||
1553 234658 : (!StartWorkerNeeded && !HaveCrashedWorker))
1554 : {
1555 246586 : if (AbortStartTime != 0)
1556 : {
1557 : int seconds;
1558 :
1559 : /* time left to abort; clamp to 0 in case it already expired */
1560 2678 : seconds = SIGKILL_CHILDREN_AFTER_SECS -
1561 2678 : (time(NULL) - AbortStartTime);
1562 :
1563 2678 : return Max(seconds * 1000, 0);
1564 : }
1565 : else
1566 243908 : return 60 * 1000;
1567 : }
1568 :
1569 0 : if (StartWorkerNeeded)
1570 0 : return 0;
1571 :
1572 0 : if (HaveCrashedWorker)
1573 : {
1574 : dlist_mutable_iter iter;
1575 :
1576 : /*
1577 : * When there are crashed bgworkers, we sleep just long enough that
1578 : * they are restarted when they request to be. Scan the list to
1579 : * determine the minimum of all wakeup times according to most recent
1580 : * crash time and requested restart interval.
1581 : */
1582 0 : dlist_foreach_modify(iter, &BackgroundWorkerList)
1583 : {
1584 : RegisteredBgWorker *rw;
1585 : TimestampTz this_wakeup;
1586 :
1587 0 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
1588 :
1589 0 : if (rw->rw_crashed_at == 0)
1590 0 : continue;
1591 :
1592 0 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART
1593 0 : || rw->rw_terminate)
1594 : {
1595 0 : ForgetBackgroundWorker(rw);
1596 0 : continue;
1597 : }
1598 :
1599 0 : this_wakeup = TimestampTzPlusMilliseconds(rw->rw_crashed_at,
1600 : 1000L * rw->rw_worker.bgw_restart_time);
1601 0 : if (next_wakeup == 0 || this_wakeup < next_wakeup)
1602 0 : next_wakeup = this_wakeup;
1603 : }
1604 : }
1605 :
1606 0 : if (next_wakeup != 0)
1607 : {
1608 : int ms;
1609 :
1610 : /* result of TimestampDifferenceMilliseconds is in [0, INT_MAX] */
1611 0 : ms = (int) TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
1612 : next_wakeup);
1613 0 : return Min(60 * 1000, ms);
1614 : }
1615 :
1616 0 : return 60 * 1000;
1617 : }
1618 :
1619 : /*
1620 : * Activate or deactivate notifications of server socket events. Since we
1621 : * don't currently have a way to remove events from an existing WaitEventSet,
1622 : * we'll just destroy and recreate the whole thing. This is called during
1623 : * shutdown so we can wait for backends to exit without accepting new
1624 : * connections, and during crash reinitialization when we need to start
1625 : * listening for new connections again. The WaitEventSet will be freed in fork
1626 : * children by ClosePostmasterPorts().
1627 : */
1628 : static void
1629 3372 : ConfigurePostmasterWaitSet(bool accept_connections)
1630 : {
1631 3372 : if (pm_wait_set)
1632 1692 : FreeWaitEventSet(pm_wait_set);
1633 3372 : pm_wait_set = NULL;
1634 :
1635 5062 : pm_wait_set = CreateWaitEventSet(NULL,
1636 1690 : accept_connections ? (1 + NumListenSockets) : 1);
1637 3372 : AddWaitEventToSet(pm_wait_set, WL_LATCH_SET, PGINVALID_SOCKET, MyLatch,
1638 : NULL);
1639 :
1640 3372 : if (accept_connections)
1641 : {
1642 3438 : for (int i = 0; i < NumListenSockets; i++)
1643 1748 : AddWaitEventToSet(pm_wait_set, WL_SOCKET_ACCEPT, ListenSockets[i],
1644 : NULL, NULL);
1645 : }
1646 3372 : }
1647 :
1648 : /*
1649 : * Main idle loop of postmaster
1650 : */
1651 : static int
1652 1680 : ServerLoop(void)
1653 : {
1654 : time_t last_lockfile_recheck_time,
1655 : last_touch_time;
1656 : WaitEvent events[MAXLISTEN];
1657 : int nevents;
1658 :
1659 1680 : ConfigurePostmasterWaitSet(true);
1660 1680 : last_lockfile_recheck_time = last_touch_time = time(NULL);
1661 :
1662 : for (;;)
1663 244906 : {
1664 : time_t now;
1665 :
1666 246586 : nevents = WaitEventSetWait(pm_wait_set,
1667 246586 : DetermineSleepTime(),
1668 : events,
1669 : lengthof(events),
1670 : 0 /* postmaster posts no wait_events */ );
1671 :
1672 : /*
1673 : * Latch set by signal handler, or new connection pending on any of
1674 : * our sockets? If the latter, fork a child process to deal with it.
1675 : */
1676 492374 : for (int i = 0; i < nevents; i++)
1677 : {
1678 247464 : if (events[i].events & WL_LATCH_SET)
1679 219250 : ResetLatch(MyLatch);
1680 :
1681 : /*
1682 : * The following requests are handled unconditionally, even if we
1683 : * didn't see WL_LATCH_SET. This gives high priority to shutdown
1684 : * and reload requests where the latch happens to appear later in
1685 : * events[] or will be reported by a later call to
1686 : * WaitEventSetWait().
1687 : */
1688 247464 : if (pending_pm_shutdown_request)
1689 1666 : process_pm_shutdown_request();
1690 247464 : if (pending_pm_reload_request)
1691 266 : process_pm_reload_request();
1692 247464 : if (pending_pm_child_exit)
1693 44700 : process_pm_child_exit();
1694 245790 : if (pending_pm_pmsignal)
1695 173060 : process_pm_pmsignal();
1696 :
1697 245790 : if (events[i].events & WL_SOCKET_ACCEPT)
1698 : {
1699 : ClientSocket s;
1700 :
1701 28214 : if (AcceptConnection(events[i].fd, &s) == STATUS_OK)
1702 28214 : BackendStartup(&s);
1703 :
1704 : /* We no longer need the open socket in this process */
1705 28212 : if (s.sock != PGINVALID_SOCKET)
1706 : {
1707 28212 : if (closesocket(s.sock) != 0)
1708 0 : elog(LOG, "could not close client socket: %m");
1709 : }
1710 : }
1711 : }
1712 :
1713 : /*
1714 : * If we need to launch any background processes after changing state
1715 : * or because some exited, do so now.
1716 : */
1717 244910 : LaunchMissingBackgroundProcesses();
1718 :
1719 : /* If we need to signal the autovacuum launcher, do so now */
1720 244906 : if (avlauncher_needs_signal)
1721 : {
1722 0 : avlauncher_needs_signal = false;
1723 0 : if (AutoVacLauncherPMChild != NULL)
1724 0 : signal_child(AutoVacLauncherPMChild, SIGUSR2);
1725 : }
1726 :
1727 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1728 :
1729 : /*
1730 : * With assertions enabled, check regularly for appearance of
1731 : * additional threads. All builds check at start and exit.
1732 : */
1733 : Assert(pthread_is_threaded_np() == 0);
1734 : #endif
1735 :
1736 : /*
1737 : * Lastly, check to see if it's time to do some things that we don't
1738 : * want to do every single time through the loop, because they're a
1739 : * bit expensive. Note that there's up to a minute of slop in when
1740 : * these tasks will be performed, since DetermineSleepTime() will let
1741 : * us sleep at most that long; except for SIGKILL timeout which has
1742 : * special-case logic there.
1743 : */
1744 244906 : now = time(NULL);
1745 :
1746 : /*
1747 : * If we already sent SIGQUIT to children and they are slow to shut
1748 : * down, it's time to send them SIGKILL (or SIGABRT if requested).
1749 : * This doesn't happen normally, but under certain conditions backends
1750 : * can get stuck while shutting down. This is a last measure to get
1751 : * them unwedged.
1752 : *
1753 : * Note we also do this during recovery from a process crash.
1754 : */
1755 244906 : if ((Shutdown >= ImmediateShutdown || FatalError) &&
1756 2688 : AbortStartTime != 0 &&
1757 2678 : (now - AbortStartTime) >= SIGKILL_CHILDREN_AFTER_SECS)
1758 : {
1759 : /* We were gentle with them before. Not anymore */
1760 0 : ereport(LOG,
1761 : /* translator: %s is SIGKILL or SIGABRT */
1762 : (errmsg("issuing %s to recalcitrant children",
1763 : send_abort_for_kill ? "SIGABRT" : "SIGKILL")));
1764 0 : TerminateChildren(send_abort_for_kill ? SIGABRT : SIGKILL);
1765 : /* reset flag so we don't SIGKILL again */
1766 0 : AbortStartTime = 0;
1767 : }
1768 :
1769 : /*
1770 : * Once a minute, verify that postmaster.pid hasn't been removed or
1771 : * overwritten. If it has, we force a shutdown. This avoids having
1772 : * postmasters and child processes hanging around after their database
1773 : * is gone, and maybe causing problems if a new database cluster is
1774 : * created in the same place. It also provides some protection
1775 : * against a DBA foolishly removing postmaster.pid and manually
1776 : * starting a new postmaster. Data corruption is likely to ensue from
1777 : * that anyway, but we can minimize the damage by aborting ASAP.
1778 : */
1779 244906 : if (now - last_lockfile_recheck_time >= 1 * SECS_PER_MINUTE)
1780 : {
1781 42 : if (!RecheckDataDirLockFile())
1782 : {
1783 0 : ereport(LOG,
1784 : (errmsg("performing immediate shutdown because data directory lock file is invalid")));
1785 0 : kill(MyProcPid, SIGQUIT);
1786 : }
1787 42 : last_lockfile_recheck_time = now;
1788 : }
1789 :
1790 : /*
1791 : * Touch Unix socket and lock files every 58 minutes, to ensure that
1792 : * they are not removed by overzealous /tmp-cleaning tasks. We assume
1793 : * no one runs cleaners with cutoff times of less than an hour ...
1794 : */
1795 244906 : if (now - last_touch_time >= 58 * SECS_PER_MINUTE)
1796 : {
1797 0 : TouchSocketFiles();
1798 0 : TouchSocketLockFiles();
1799 0 : last_touch_time = now;
1800 : }
1801 : }
1802 : }
1803 :
1804 : /*
1805 : * canAcceptConnections --- check to see if database state allows connections
1806 : * of the specified type. backend_type can be B_BACKEND or B_AUTOVAC_WORKER.
1807 : * (Note that we don't yet know whether a normal B_BACKEND connection might
1808 : * turn into a walsender.)
1809 : */
1810 : static CAC_state
1811 30638 : canAcceptConnections(BackendType backend_type)
1812 : {
1813 30638 : CAC_state result = CAC_OK;
1814 :
1815 : Assert(backend_type == B_BACKEND || backend_type == B_AUTOVAC_WORKER);
1816 :
1817 : /*
1818 : * Can't start backends when in startup/shutdown/inconsistent recovery
1819 : * state. We treat autovac workers the same as user backends for this
1820 : * purpose.
1821 : */
1822 30638 : if (pmState != PM_RUN && pmState != PM_HOT_STANDBY)
1823 : {
1824 288 : if (Shutdown > NoShutdown)
1825 50 : return CAC_SHUTDOWN; /* shutdown is pending */
1826 238 : else if (!FatalError && pmState == PM_STARTUP)
1827 228 : return CAC_STARTUP; /* normal startup */
1828 10 : else if (!FatalError && pmState == PM_RECOVERY)
1829 10 : return CAC_NOTHOTSTANDBY; /* not yet ready for hot standby */
1830 : else
1831 0 : return CAC_RECOVERY; /* else must be crash recovery */
1832 : }
1833 :
1834 : /*
1835 : * "Smart shutdown" restrictions are applied only to normal connections,
1836 : * not to autovac workers.
1837 : */
1838 30350 : if (!connsAllowed && backend_type == B_BACKEND)
1839 0 : return CAC_SHUTDOWN; /* shutdown is pending */
1840 :
1841 30350 : return result;
1842 : }
1843 :
1844 : /*
1845 : * ClosePostmasterPorts -- close all the postmaster's open sockets
1846 : *
1847 : * This is called during child process startup to release file descriptors
1848 : * that are not needed by that child process. The postmaster still has
1849 : * them open, of course.
1850 : *
1851 : * Note: we pass am_syslogger as a boolean because we don't want to set
1852 : * the global variable yet when this is called.
1853 : */
1854 : void
1855 43426 : ClosePostmasterPorts(bool am_syslogger)
1856 : {
1857 : /* Release resources held by the postmaster's WaitEventSet. */
1858 43426 : if (pm_wait_set)
1859 : {
1860 36830 : FreeWaitEventSetAfterFork(pm_wait_set);
1861 36830 : pm_wait_set = NULL;
1862 : }
1863 :
1864 : #ifndef WIN32
1865 :
1866 : /*
1867 : * Close the write end of postmaster death watch pipe. It's important to
1868 : * do this as early as possible, so that if postmaster dies, others won't
1869 : * think that it's still running because we're holding the pipe open.
1870 : */
1871 43426 : if (close(postmaster_alive_fds[POSTMASTER_FD_OWN]) != 0)
1872 0 : ereport(FATAL,
1873 : (errcode_for_file_access(),
1874 : errmsg_internal("could not close postmaster death monitoring pipe in child process: %m")));
1875 43426 : postmaster_alive_fds[POSTMASTER_FD_OWN] = -1;
1876 : /* Notify fd.c that we released one pipe FD. */
1877 43426 : ReleaseExternalFD();
1878 : #endif
1879 :
1880 : /*
1881 : * Close the postmaster's listen sockets. These aren't tracked by fd.c,
1882 : * so we don't call ReleaseExternalFD() here.
1883 : *
1884 : * The listen sockets are marked as FD_CLOEXEC, so this isn't needed in
1885 : * EXEC_BACKEND mode.
1886 : */
1887 : #ifndef EXEC_BACKEND
1888 43426 : if (ListenSockets)
1889 : {
1890 87912 : for (int i = 0; i < NumListenSockets; i++)
1891 : {
1892 44488 : if (closesocket(ListenSockets[i]) != 0)
1893 0 : elog(LOG, "could not close listen socket: %m");
1894 : }
1895 43424 : pfree(ListenSockets);
1896 : }
1897 43426 : NumListenSockets = 0;
1898 43426 : ListenSockets = NULL;
1899 : #endif
1900 :
1901 : /*
1902 : * If using syslogger, close the read side of the pipe. We don't bother
1903 : * tracking this in fd.c, either.
1904 : */
1905 43426 : if (!am_syslogger)
1906 : {
1907 : #ifndef WIN32
1908 43424 : if (syslogPipe[0] >= 0)
1909 34 : close(syslogPipe[0]);
1910 43424 : syslogPipe[0] = -1;
1911 : #else
1912 : if (syslogPipe[0])
1913 : CloseHandle(syslogPipe[0]);
1914 : syslogPipe[0] = 0;
1915 : #endif
1916 : }
1917 :
1918 : #ifdef USE_BONJOUR
1919 : /* If using Bonjour, close the connection to the mDNS daemon */
1920 : if (bonjour_sdref)
1921 : close(DNSServiceRefSockFD(bonjour_sdref));
1922 : #endif
1923 43426 : }
1924 :
1925 :
1926 : /*
1927 : * InitProcessGlobals -- set MyStartTime[stamp], random seeds
1928 : *
1929 : * Called early in the postmaster and every backend.
1930 : */
1931 : void
1932 45594 : InitProcessGlobals(void)
1933 : {
1934 45594 : MyStartTimestamp = GetCurrentTimestamp();
1935 45594 : MyStartTime = timestamptz_to_time_t(MyStartTimestamp);
1936 :
1937 : /*
1938 : * Set a different global seed in every process. We want something
1939 : * unpredictable, so if possible, use high-quality random bits for the
1940 : * seed. Otherwise, fall back to a seed based on timestamp and PID.
1941 : */
1942 45594 : if (unlikely(!pg_prng_strong_seed(&pg_global_prng_state)))
1943 : {
1944 : uint64 rseed;
1945 :
1946 : /*
1947 : * Since PIDs and timestamps tend to change more frequently in their
1948 : * least significant bits, shift the timestamp left to allow a larger
1949 : * total number of seeds in a given time period. Since that would
1950 : * leave only 20 bits of the timestamp that cycle every ~1 second,
1951 : * also mix in some higher bits.
1952 : */
1953 0 : rseed = ((uint64) MyProcPid) ^
1954 0 : ((uint64) MyStartTimestamp << 12) ^
1955 0 : ((uint64) MyStartTimestamp >> 20);
1956 :
1957 0 : pg_prng_seed(&pg_global_prng_state, rseed);
1958 : }
1959 :
1960 : /*
1961 : * Also make sure that we've set a good seed for random(3). Use of that
1962 : * is deprecated in core Postgres, but extensions might use it.
1963 : */
1964 : #ifndef WIN32
1965 45594 : srandom(pg_prng_uint32(&pg_global_prng_state));
1966 : #endif
1967 45594 : }
1968 :
1969 : /*
1970 : * Child processes use SIGUSR1 to notify us of 'pmsignals'. pg_ctl uses
1971 : * SIGUSR1 to ask postmaster to check for logrotate and promote files.
1972 : */
1973 : static void
1974 173510 : handle_pm_pmsignal_signal(SIGNAL_ARGS)
1975 : {
1976 173510 : pending_pm_pmsignal = true;
1977 173510 : SetLatch(MyLatch);
1978 173510 : }
1979 :
1980 : /*
1981 : * pg_ctl uses SIGHUP to request a reload of the configuration files.
1982 : */
1983 : static void
1984 266 : handle_pm_reload_request_signal(SIGNAL_ARGS)
1985 : {
1986 266 : pending_pm_reload_request = true;
1987 266 : SetLatch(MyLatch);
1988 266 : }
1989 :
1990 : /*
1991 : * Re-read config files, and tell children to do same.
1992 : */
1993 : static void
1994 266 : process_pm_reload_request(void)
1995 : {
1996 266 : pending_pm_reload_request = false;
1997 :
1998 266 : ereport(DEBUG2,
1999 : (errmsg_internal("postmaster received reload request signal")));
2000 :
2001 266 : if (Shutdown <= SmartShutdown)
2002 : {
2003 266 : ereport(LOG,
2004 : (errmsg("received SIGHUP, reloading configuration files")));
2005 266 : ProcessConfigFile(PGC_SIGHUP);
2006 266 : SignalChildren(SIGHUP, btmask_all_except(B_DEAD_END_BACKEND));
2007 :
2008 : /* Reload authentication config files too */
2009 266 : if (!load_hba())
2010 0 : ereport(LOG,
2011 : /* translator: %s is a configuration file */
2012 : (errmsg("%s was not reloaded", HbaFileName)));
2013 :
2014 266 : if (!load_ident())
2015 0 : ereport(LOG,
2016 : (errmsg("%s was not reloaded", IdentFileName)));
2017 :
2018 : #ifdef USE_SSL
2019 : /* Reload SSL configuration as well */
2020 266 : if (EnableSSL)
2021 : {
2022 2 : if (secure_initialize(false) == 0)
2023 2 : LoadedSSL = true;
2024 : else
2025 0 : ereport(LOG,
2026 : (errmsg("SSL configuration was not reloaded")));
2027 : }
2028 : else
2029 : {
2030 264 : secure_destroy();
2031 264 : LoadedSSL = false;
2032 : }
2033 : #endif
2034 :
2035 : #ifdef EXEC_BACKEND
2036 : /* Update the starting-point file for future children */
2037 : write_nondefault_variables(PGC_SIGHUP);
2038 : #endif
2039 : }
2040 266 : }
2041 :
2042 : /*
2043 : * pg_ctl uses SIGTERM, SIGINT and SIGQUIT to request different types of
2044 : * shutdown.
2045 : */
2046 : static void
2047 1666 : handle_pm_shutdown_request_signal(SIGNAL_ARGS)
2048 : {
2049 1666 : switch (postgres_signal_arg)
2050 : {
2051 76 : case SIGTERM:
2052 : /* smart is implied if the other two flags aren't set */
2053 76 : pending_pm_shutdown_request = true;
2054 76 : break;
2055 922 : case SIGINT:
2056 922 : pending_pm_fast_shutdown_request = true;
2057 922 : pending_pm_shutdown_request = true;
2058 922 : break;
2059 668 : case SIGQUIT:
2060 668 : pending_pm_immediate_shutdown_request = true;
2061 668 : pending_pm_shutdown_request = true;
2062 668 : break;
2063 : }
2064 1666 : SetLatch(MyLatch);
2065 1666 : }
2066 :
2067 : /*
2068 : * Process shutdown request.
2069 : */
2070 : static void
2071 1666 : process_pm_shutdown_request(void)
2072 : {
2073 : int mode;
2074 :
2075 1666 : ereport(DEBUG2,
2076 : (errmsg_internal("postmaster received shutdown request signal")));
2077 :
2078 1666 : pending_pm_shutdown_request = false;
2079 :
2080 : /*
2081 : * If more than one shutdown request signal arrived since the last server
2082 : * loop, take the one that is the most immediate. That matches the
2083 : * priority that would apply if we processed them one by one in any order.
2084 : */
2085 1666 : if (pending_pm_immediate_shutdown_request)
2086 : {
2087 668 : pending_pm_immediate_shutdown_request = false;
2088 668 : pending_pm_fast_shutdown_request = false;
2089 668 : mode = ImmediateShutdown;
2090 : }
2091 998 : else if (pending_pm_fast_shutdown_request)
2092 : {
2093 922 : pending_pm_fast_shutdown_request = false;
2094 922 : mode = FastShutdown;
2095 : }
2096 : else
2097 76 : mode = SmartShutdown;
2098 :
2099 1666 : switch (mode)
2100 : {
2101 76 : case SmartShutdown:
2102 :
2103 : /*
2104 : * Smart Shutdown:
2105 : *
2106 : * Wait for children to end their work, then shut down.
2107 : */
2108 76 : if (Shutdown >= SmartShutdown)
2109 0 : break;
2110 76 : Shutdown = SmartShutdown;
2111 76 : ereport(LOG,
2112 : (errmsg("received smart shutdown request")));
2113 :
2114 : /* Report status */
2115 76 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2116 : #ifdef USE_SYSTEMD
2117 : sd_notify(0, "STOPPING=1");
2118 : #endif
2119 :
2120 : /*
2121 : * If we reached normal running, we go straight to waiting for
2122 : * client backends to exit. If already in PM_STOP_BACKENDS or a
2123 : * later state, do not change it.
2124 : */
2125 76 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2126 76 : connsAllowed = false;
2127 0 : else if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2128 : {
2129 : /* There should be no clients, so proceed to stop children */
2130 0 : UpdatePMState(PM_STOP_BACKENDS);
2131 : }
2132 :
2133 : /*
2134 : * Now wait for online backup mode to end and backends to exit. If
2135 : * that is already the case, PostmasterStateMachine will take the
2136 : * next step.
2137 : */
2138 76 : PostmasterStateMachine();
2139 76 : break;
2140 :
2141 922 : case FastShutdown:
2142 :
2143 : /*
2144 : * Fast Shutdown:
2145 : *
2146 : * Abort all children with SIGTERM (rollback active transactions
2147 : * and exit) and shut down when they are gone.
2148 : */
2149 922 : if (Shutdown >= FastShutdown)
2150 0 : break;
2151 922 : Shutdown = FastShutdown;
2152 922 : ereport(LOG,
2153 : (errmsg("received fast shutdown request")));
2154 :
2155 : /* Report status */
2156 922 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2157 : #ifdef USE_SYSTEMD
2158 : sd_notify(0, "STOPPING=1");
2159 : #endif
2160 :
2161 922 : if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2162 : {
2163 : /* Just shut down background processes silently */
2164 0 : UpdatePMState(PM_STOP_BACKENDS);
2165 : }
2166 922 : else if (pmState == PM_RUN ||
2167 108 : pmState == PM_HOT_STANDBY)
2168 : {
2169 : /* Report that we're about to zap live client sessions */
2170 922 : ereport(LOG,
2171 : (errmsg("aborting any active transactions")));
2172 922 : UpdatePMState(PM_STOP_BACKENDS);
2173 : }
2174 :
2175 : /*
2176 : * PostmasterStateMachine will issue any necessary signals, or
2177 : * take the next step if no child processes need to be killed.
2178 : */
2179 922 : PostmasterStateMachine();
2180 922 : break;
2181 :
2182 668 : case ImmediateShutdown:
2183 :
2184 : /*
2185 : * Immediate Shutdown:
2186 : *
2187 : * abort all children with SIGQUIT, wait for them to exit,
2188 : * terminate remaining ones with SIGKILL, then exit without
2189 : * attempt to properly shut down the data base system.
2190 : */
2191 668 : if (Shutdown >= ImmediateShutdown)
2192 0 : break;
2193 668 : Shutdown = ImmediateShutdown;
2194 668 : ereport(LOG,
2195 : (errmsg("received immediate shutdown request")));
2196 :
2197 : /* Report status */
2198 668 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2199 : #ifdef USE_SYSTEMD
2200 : sd_notify(0, "STOPPING=1");
2201 : #endif
2202 :
2203 : /* tell children to shut down ASAP */
2204 : /* (note we don't apply send_abort_for_crash here) */
2205 668 : SetQuitSignalReason(PMQUIT_FOR_STOP);
2206 668 : TerminateChildren(SIGQUIT);
2207 668 : UpdatePMState(PM_WAIT_BACKENDS);
2208 :
2209 : /* set stopwatch for them to die */
2210 668 : AbortStartTime = time(NULL);
2211 :
2212 : /*
2213 : * Now wait for backends to exit. If there are none,
2214 : * PostmasterStateMachine will take the next step.
2215 : */
2216 668 : PostmasterStateMachine();
2217 668 : break;
2218 : }
2219 1666 : }
2220 :
2221 : static void
2222 45098 : handle_pm_child_exit_signal(SIGNAL_ARGS)
2223 : {
2224 45098 : pending_pm_child_exit = true;
2225 45098 : SetLatch(MyLatch);
2226 45098 : }
2227 :
2228 : /*
2229 : * Cleanup after a child process dies.
2230 : */
2231 : static void
2232 44700 : process_pm_child_exit(void)
2233 : {
2234 : int pid; /* process id of dead child process */
2235 : int exitstatus; /* its exit status */
2236 :
2237 44700 : pending_pm_child_exit = false;
2238 :
2239 44700 : ereport(DEBUG4,
2240 : (errmsg_internal("reaping dead processes")));
2241 :
2242 93916 : while ((pid = waitpid(-1, &exitstatus, WNOHANG)) > 0)
2243 : {
2244 : PMChild *pmchild;
2245 :
2246 : /*
2247 : * Check if this child was a startup process.
2248 : */
2249 49218 : if (StartupPMChild && pid == StartupPMChild->pid)
2250 : {
2251 1690 : ReleasePostmasterChildSlot(StartupPMChild);
2252 1690 : StartupPMChild = NULL;
2253 :
2254 : /*
2255 : * Startup process exited in response to a shutdown request (or it
2256 : * completed normally regardless of the shutdown request).
2257 : */
2258 1690 : if (Shutdown > NoShutdown &&
2259 198 : (EXIT_STATUS_0(exitstatus) || EXIT_STATUS_1(exitstatus)))
2260 : {
2261 108 : StartupStatus = STARTUP_NOT_RUNNING;
2262 108 : UpdatePMState(PM_WAIT_BACKENDS);
2263 : /* PostmasterStateMachine logic does the rest */
2264 108 : continue;
2265 : }
2266 :
2267 1582 : if (EXIT_STATUS_3(exitstatus))
2268 : {
2269 0 : ereport(LOG,
2270 : (errmsg("shutdown at recovery target")));
2271 0 : StartupStatus = STARTUP_NOT_RUNNING;
2272 0 : Shutdown = Max(Shutdown, SmartShutdown);
2273 0 : TerminateChildren(SIGTERM);
2274 0 : UpdatePMState(PM_WAIT_BACKENDS);
2275 : /* PostmasterStateMachine logic does the rest */
2276 0 : continue;
2277 : }
2278 :
2279 : /*
2280 : * Unexpected exit of startup process (including FATAL exit)
2281 : * during PM_STARTUP is treated as catastrophic. There are no
2282 : * other processes running yet, so we can just exit.
2283 : */
2284 1582 : if (pmState == PM_STARTUP &&
2285 1198 : StartupStatus != STARTUP_SIGNALED &&
2286 1198 : !EXIT_STATUS_0(exitstatus))
2287 : {
2288 2 : LogChildExit(LOG, _("startup process"),
2289 : pid, exitstatus);
2290 2 : ereport(LOG,
2291 : (errmsg("aborting startup due to startup process failure")));
2292 2 : ExitPostmaster(1);
2293 : }
2294 :
2295 : /*
2296 : * After PM_STARTUP, any unexpected exit (including FATAL exit) of
2297 : * the startup process is catastrophic, so kill other children,
2298 : * and set StartupStatus so we don't try to reinitialize after
2299 : * they're gone. Exception: if StartupStatus is STARTUP_SIGNALED,
2300 : * then we previously sent the startup process a SIGQUIT; so
2301 : * that's probably the reason it died, and we do want to try to
2302 : * restart in that case.
2303 : *
2304 : * This stanza also handles the case where we sent a SIGQUIT
2305 : * during PM_STARTUP due to some dead-end child crashing: in that
2306 : * situation, if the startup process dies on the SIGQUIT, we need
2307 : * to transition to PM_WAIT_BACKENDS state which will allow
2308 : * PostmasterStateMachine to restart the startup process. (On the
2309 : * other hand, the startup process might complete normally, if we
2310 : * were too late with the SIGQUIT. In that case we'll fall
2311 : * through and commence normal operations.)
2312 : */
2313 1580 : if (!EXIT_STATUS_0(exitstatus))
2314 : {
2315 96 : if (StartupStatus == STARTUP_SIGNALED)
2316 : {
2317 90 : StartupStatus = STARTUP_NOT_RUNNING;
2318 90 : if (pmState == PM_STARTUP)
2319 0 : UpdatePMState(PM_WAIT_BACKENDS);
2320 : }
2321 : else
2322 6 : StartupStatus = STARTUP_CRASHED;
2323 96 : HandleChildCrash(pid, exitstatus,
2324 96 : _("startup process"));
2325 96 : continue;
2326 : }
2327 :
2328 : /*
2329 : * Startup succeeded, commence normal operations
2330 : */
2331 1484 : StartupStatus = STARTUP_NOT_RUNNING;
2332 1484 : FatalError = false;
2333 1484 : AbortStartTime = 0;
2334 1484 : ReachedNormalRunning = true;
2335 1484 : UpdatePMState(PM_RUN);
2336 1484 : connsAllowed = true;
2337 :
2338 : /*
2339 : * At the next iteration of the postmaster's main loop, we will
2340 : * crank up the background tasks like the autovacuum launcher and
2341 : * background workers that were not started earlier already.
2342 : */
2343 1484 : StartWorkerNeeded = true;
2344 :
2345 : /* at this point we are really open for business */
2346 1484 : ereport(LOG,
2347 : (errmsg("database system is ready to accept connections")));
2348 :
2349 : /* Report status */
2350 1484 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
2351 : #ifdef USE_SYSTEMD
2352 : sd_notify(0, "READY=1");
2353 : #endif
2354 :
2355 1484 : continue;
2356 : }
2357 :
2358 : /*
2359 : * Was it the bgwriter? Normal exit can be ignored; we'll start a new
2360 : * one at the next iteration of the postmaster's main loop, if
2361 : * necessary. Any other exit condition is treated as a crash.
2362 : */
2363 47528 : if (BgWriterPMChild && pid == BgWriterPMChild->pid)
2364 : {
2365 1682 : ReleasePostmasterChildSlot(BgWriterPMChild);
2366 1682 : BgWriterPMChild = NULL;
2367 1682 : if (!EXIT_STATUS_0(exitstatus))
2368 684 : HandleChildCrash(pid, exitstatus,
2369 684 : _("background writer process"));
2370 1682 : continue;
2371 : }
2372 :
2373 : /*
2374 : * Was it the checkpointer?
2375 : */
2376 45846 : if (CheckpointerPMChild && pid == CheckpointerPMChild->pid)
2377 : {
2378 1682 : ReleasePostmasterChildSlot(CheckpointerPMChild);
2379 1682 : CheckpointerPMChild = NULL;
2380 1682 : if (EXIT_STATUS_0(exitstatus) && pmState == PM_WAIT_CHECKPOINTER)
2381 998 : {
2382 : /*
2383 : * OK, we saw normal exit of the checkpointer after it's been
2384 : * told to shut down. We know checkpointer wrote a shutdown
2385 : * checkpoint, otherwise we'd still be in
2386 : * PM_WAIT_XLOG_SHUTDOWN state.
2387 : *
2388 : * At this point only dead-end children and logger should be
2389 : * left.
2390 : */
2391 998 : UpdatePMState(PM_WAIT_DEAD_END);
2392 998 : ConfigurePostmasterWaitSet(false);
2393 998 : SignalChildren(SIGTERM, btmask_all_except(B_LOGGER));
2394 : }
2395 : else
2396 : {
2397 : /*
2398 : * Any unexpected exit of the checkpointer (including FATAL
2399 : * exit) is treated as a crash.
2400 : */
2401 684 : HandleChildCrash(pid, exitstatus,
2402 684 : _("checkpointer process"));
2403 : }
2404 :
2405 1682 : continue;
2406 : }
2407 :
2408 : /*
2409 : * Was it the wal writer? Normal exit can be ignored; we'll start a
2410 : * new one at the next iteration of the postmaster's main loop, if
2411 : * necessary. Any other exit condition is treated as a crash.
2412 : */
2413 44164 : if (WalWriterPMChild && pid == WalWriterPMChild->pid)
2414 : {
2415 1478 : ReleasePostmasterChildSlot(WalWriterPMChild);
2416 1478 : WalWriterPMChild = NULL;
2417 1478 : if (!EXIT_STATUS_0(exitstatus))
2418 588 : HandleChildCrash(pid, exitstatus,
2419 588 : _("WAL writer process"));
2420 1478 : continue;
2421 : }
2422 :
2423 : /*
2424 : * Was it the wal receiver? If exit status is zero (normal) or one
2425 : * (FATAL exit), we assume everything is all right just like normal
2426 : * backends. (If we need a new wal receiver, we'll start one at the
2427 : * next iteration of the postmaster's main loop.)
2428 : */
2429 42686 : if (WalReceiverPMChild && pid == WalReceiverPMChild->pid)
2430 : {
2431 492 : ReleasePostmasterChildSlot(WalReceiverPMChild);
2432 492 : WalReceiverPMChild = NULL;
2433 492 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2434 34 : HandleChildCrash(pid, exitstatus,
2435 34 : _("WAL receiver process"));
2436 492 : continue;
2437 : }
2438 :
2439 : /*
2440 : * Was it the wal summarizer? Normal exit can be ignored; we'll start
2441 : * a new one at the next iteration of the postmaster's main loop, if
2442 : * necessary. Any other exit condition is treated as a crash.
2443 : */
2444 42194 : if (WalSummarizerPMChild && pid == WalSummarizerPMChild->pid)
2445 : {
2446 38 : ReleasePostmasterChildSlot(WalSummarizerPMChild);
2447 38 : WalSummarizerPMChild = NULL;
2448 38 : if (!EXIT_STATUS_0(exitstatus))
2449 36 : HandleChildCrash(pid, exitstatus,
2450 36 : _("WAL summarizer process"));
2451 38 : continue;
2452 : }
2453 :
2454 : /*
2455 : * Was it the autovacuum launcher? Normal exit can be ignored; we'll
2456 : * start a new one at the next iteration of the postmaster's main
2457 : * loop, if necessary. Any other exit condition is treated as a
2458 : * crash.
2459 : */
2460 42156 : if (AutoVacLauncherPMChild && pid == AutoVacLauncherPMChild->pid)
2461 : {
2462 1242 : ReleasePostmasterChildSlot(AutoVacLauncherPMChild);
2463 1242 : AutoVacLauncherPMChild = NULL;
2464 1242 : if (!EXIT_STATUS_0(exitstatus))
2465 500 : HandleChildCrash(pid, exitstatus,
2466 500 : _("autovacuum launcher process"));
2467 1242 : continue;
2468 : }
2469 :
2470 : /*
2471 : * Was it the archiver? If exit status is zero (normal) or one (FATAL
2472 : * exit), we assume everything is all right just like normal backends
2473 : * and just try to start a new one on the next cycle of the
2474 : * postmaster's main loop, to retry archiving remaining files.
2475 : */
2476 40914 : if (PgArchPMChild && pid == PgArchPMChild->pid)
2477 : {
2478 110 : ReleasePostmasterChildSlot(PgArchPMChild);
2479 110 : PgArchPMChild = NULL;
2480 110 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2481 80 : HandleChildCrash(pid, exitstatus,
2482 80 : _("archiver process"));
2483 110 : continue;
2484 : }
2485 :
2486 : /* Was it the system logger? If so, try to start a new one */
2487 40804 : if (SysLoggerPMChild && pid == SysLoggerPMChild->pid)
2488 : {
2489 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
2490 0 : SysLoggerPMChild = NULL;
2491 :
2492 : /* for safety's sake, launch new logger *first* */
2493 0 : if (Logging_collector)
2494 0 : StartSysLogger();
2495 :
2496 0 : if (!EXIT_STATUS_0(exitstatus))
2497 0 : LogChildExit(LOG, _("system logger process"),
2498 : pid, exitstatus);
2499 0 : continue;
2500 : }
2501 :
2502 : /*
2503 : * Was it the slot sync worker? Normal exit or FATAL exit can be
2504 : * ignored (FATAL can be caused by libpqwalreceiver on receiving
2505 : * shutdown request by the startup process during promotion); we'll
2506 : * start a new one at the next iteration of the postmaster's main
2507 : * loop, if necessary. Any other exit condition is treated as a crash.
2508 : */
2509 40804 : if (SlotSyncWorkerPMChild && pid == SlotSyncWorkerPMChild->pid)
2510 : {
2511 8 : ReleasePostmasterChildSlot(SlotSyncWorkerPMChild);
2512 8 : SlotSyncWorkerPMChild = NULL;
2513 8 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2514 0 : HandleChildCrash(pid, exitstatus,
2515 0 : _("slot sync worker process"));
2516 8 : continue;
2517 : }
2518 :
2519 : /* Was it an IO worker? */
2520 40796 : if (maybe_reap_io_worker(pid))
2521 : {
2522 5134 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2523 2052 : HandleChildCrash(pid, exitstatus, _("io worker"));
2524 :
2525 5134 : maybe_adjust_io_workers();
2526 5134 : continue;
2527 : }
2528 :
2529 : /*
2530 : * Was it a backend or a background worker?
2531 : */
2532 35662 : pmchild = FindPostmasterChildByPid(pid);
2533 35662 : if (pmchild)
2534 : {
2535 35662 : CleanupBackend(pmchild, exitstatus);
2536 : }
2537 :
2538 : /*
2539 : * We don't know anything about this child process. That's highly
2540 : * unexpected, as we do track all the child processes that we fork.
2541 : */
2542 : else
2543 : {
2544 0 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2545 0 : HandleChildCrash(pid, exitstatus, _("untracked child process"));
2546 : else
2547 0 : LogChildExit(LOG, _("untracked child process"), pid, exitstatus);
2548 : }
2549 : } /* loop over pending child-death reports */
2550 :
2551 : /*
2552 : * After cleaning out the SIGCHLD queue, see if we have any state changes
2553 : * or actions to make.
2554 : */
2555 44698 : PostmasterStateMachine();
2556 43026 : }
2557 :
2558 : /*
2559 : * CleanupBackend -- cleanup after terminated backend or background worker.
2560 : *
2561 : * Remove all local state associated with the child process and release its
2562 : * PMChild slot.
2563 : */
2564 : static void
2565 35662 : CleanupBackend(PMChild *bp,
2566 : int exitstatus) /* child's exit status. */
2567 : {
2568 : char namebuf[MAXPGPATH];
2569 : const char *procname;
2570 35662 : bool crashed = false;
2571 35662 : bool logged = false;
2572 : pid_t bp_pid;
2573 : bool bp_bgworker_notify;
2574 : BackendType bp_bkend_type;
2575 : RegisteredBgWorker *rw;
2576 :
2577 : /* Construct a process name for the log message */
2578 35662 : if (bp->bkend_type == B_BG_WORKER)
2579 : {
2580 5030 : snprintf(namebuf, MAXPGPATH, _("background worker \"%s\""),
2581 5030 : bp->rw->rw_worker.bgw_type);
2582 5030 : procname = namebuf;
2583 : }
2584 : else
2585 30632 : procname = _(GetBackendTypeDesc(bp->bkend_type));
2586 :
2587 : /*
2588 : * If a backend dies in an ugly way then we must signal all other backends
2589 : * to quickdie. If exit status is zero (normal) or one (FATAL exit), we
2590 : * assume everything is all right and proceed to remove the backend from
2591 : * the active child list.
2592 : */
2593 35662 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2594 1106 : crashed = true;
2595 :
2596 : #ifdef WIN32
2597 :
2598 : /*
2599 : * On win32, also treat ERROR_WAIT_NO_CHILDREN (128) as nonfatal case,
2600 : * since that sometimes happens under load when the process fails to start
2601 : * properly (long before it starts using shared memory). Microsoft reports
2602 : * it is related to mutex failure:
2603 : * http://archives.postgresql.org/pgsql-hackers/2010-09/msg00790.php
2604 : */
2605 : if (exitstatus == ERROR_WAIT_NO_CHILDREN)
2606 : {
2607 : LogChildExit(LOG, procname, bp->pid, exitstatus);
2608 : logged = true;
2609 : crashed = false;
2610 : }
2611 : #endif
2612 :
2613 : /*
2614 : * Release the PMChild entry.
2615 : *
2616 : * If the process attached to shared memory, this also checks that it
2617 : * detached cleanly.
2618 : */
2619 35662 : bp_pid = bp->pid;
2620 35662 : bp_bgworker_notify = bp->bgworker_notify;
2621 35662 : bp_bkend_type = bp->bkend_type;
2622 35662 : rw = bp->rw;
2623 35662 : if (!ReleasePostmasterChildSlot(bp))
2624 : {
2625 : /*
2626 : * Uh-oh, the child failed to clean itself up. Treat as a crash after
2627 : * all.
2628 : */
2629 728 : crashed = true;
2630 : }
2631 35662 : bp = NULL;
2632 :
2633 35662 : if (crashed)
2634 : {
2635 1106 : HandleChildCrash(bp_pid, exitstatus, procname);
2636 1106 : return;
2637 : }
2638 :
2639 : /*
2640 : * This backend may have been slated to receive SIGUSR1 when some
2641 : * background worker started or stopped. Cancel those notifications, as
2642 : * we don't want to signal PIDs that are not PostgreSQL backends. This
2643 : * gets skipped in the (probably very common) case where the backend has
2644 : * never requested any such notifications.
2645 : */
2646 34556 : if (bp_bgworker_notify)
2647 486 : BackgroundWorkerStopNotifications(bp_pid);
2648 :
2649 : /*
2650 : * If it was a background worker, also update its RegisteredBgWorker
2651 : * entry.
2652 : */
2653 34556 : if (bp_bkend_type == B_BG_WORKER)
2654 : {
2655 4410 : if (!EXIT_STATUS_0(exitstatus))
2656 : {
2657 : /* Record timestamp, so we know when to restart the worker. */
2658 1208 : rw->rw_crashed_at = GetCurrentTimestamp();
2659 : }
2660 : else
2661 : {
2662 : /* Zero exit status means terminate */
2663 3202 : rw->rw_crashed_at = 0;
2664 3202 : rw->rw_terminate = true;
2665 : }
2666 :
2667 4410 : rw->rw_pid = 0;
2668 4410 : ReportBackgroundWorkerExit(rw); /* report child death */
2669 :
2670 4410 : if (!logged)
2671 : {
2672 4410 : LogChildExit(EXIT_STATUS_0(exitstatus) ? DEBUG1 : LOG,
2673 : procname, bp_pid, exitstatus);
2674 4410 : logged = true;
2675 : }
2676 :
2677 : /* have it be restarted */
2678 4410 : HaveCrashedWorker = true;
2679 : }
2680 :
2681 34556 : if (!logged)
2682 30146 : LogChildExit(DEBUG2, procname, bp_pid, exitstatus);
2683 : }
2684 :
2685 : /*
2686 : * Transition into FatalError state, in response to something bad having
2687 : * happened. Commonly the caller will have logged the reason for entering
2688 : * FatalError state.
2689 : *
2690 : * This should only be called when not already in FatalError or
2691 : * ImmediateShutdown state.
2692 : */
2693 : static void
2694 16 : HandleFatalError(QuitSignalReason reason, bool consider_sigabrt)
2695 : {
2696 : int sigtosend;
2697 :
2698 : Assert(!FatalError);
2699 : Assert(Shutdown != ImmediateShutdown);
2700 :
2701 16 : SetQuitSignalReason(reason);
2702 :
2703 16 : if (consider_sigabrt && send_abort_for_crash)
2704 0 : sigtosend = SIGABRT;
2705 : else
2706 16 : sigtosend = SIGQUIT;
2707 :
2708 : /*
2709 : * Signal all other child processes to exit.
2710 : *
2711 : * We could exclude dead-end children here, but at least when sending
2712 : * SIGABRT it seems better to include them.
2713 : */
2714 16 : TerminateChildren(sigtosend);
2715 :
2716 16 : FatalError = true;
2717 :
2718 : /*
2719 : * Choose the appropriate new state to react to the fatal error. Unless we
2720 : * were already in the process of shutting down, we go through
2721 : * PM_WAIT_BACKENDS. For errors during the shutdown sequence, we directly
2722 : * switch to PM_WAIT_DEAD_END.
2723 : */
2724 16 : switch (pmState)
2725 : {
2726 0 : case PM_INIT:
2727 : /* shouldn't have any children */
2728 : Assert(false);
2729 0 : break;
2730 0 : case PM_STARTUP:
2731 : /* should have been handled in process_pm_child_exit */
2732 : Assert(false);
2733 0 : break;
2734 :
2735 : /* wait for children to die */
2736 16 : case PM_RECOVERY:
2737 : case PM_HOT_STANDBY:
2738 : case PM_RUN:
2739 : case PM_STOP_BACKENDS:
2740 16 : UpdatePMState(PM_WAIT_BACKENDS);
2741 16 : break;
2742 :
2743 0 : case PM_WAIT_BACKENDS:
2744 : /* there might be more backends to wait for */
2745 0 : break;
2746 :
2747 0 : case PM_WAIT_XLOG_SHUTDOWN:
2748 : case PM_WAIT_XLOG_ARCHIVAL:
2749 : case PM_WAIT_CHECKPOINTER:
2750 : case PM_WAIT_IO_WORKERS:
2751 :
2752 : /*
2753 : * NB: Similar code exists in PostmasterStateMachine()'s handling
2754 : * of FatalError in PM_STOP_BACKENDS/PM_WAIT_BACKENDS states.
2755 : */
2756 0 : ConfigurePostmasterWaitSet(false);
2757 0 : UpdatePMState(PM_WAIT_DEAD_END);
2758 0 : break;
2759 :
2760 0 : case PM_WAIT_DEAD_END:
2761 : case PM_NO_CHILDREN:
2762 0 : break;
2763 : }
2764 :
2765 : /*
2766 : * .. and if this doesn't happen quickly enough, now the clock is ticking
2767 : * for us to kill them without mercy.
2768 : */
2769 16 : if (AbortStartTime == 0)
2770 16 : AbortStartTime = time(NULL);
2771 16 : }
2772 :
2773 : /*
2774 : * HandleChildCrash -- cleanup after failed backend, bgwriter, checkpointer,
2775 : * walwriter, autovacuum, archiver, slot sync worker, or background worker.
2776 : *
2777 : * The objectives here are to clean up our local state about the child
2778 : * process, and to signal all other remaining children to quickdie.
2779 : *
2780 : * The caller has already released its PMChild slot.
2781 : */
2782 : static void
2783 5860 : HandleChildCrash(int pid, int exitstatus, const char *procname)
2784 : {
2785 : /*
2786 : * We only log messages and send signals if this is the first process
2787 : * crash and we're not doing an immediate shutdown; otherwise, we're only
2788 : * here to update postmaster's idea of live processes. If we have already
2789 : * signaled children, nonzero exit status is to be expected, so don't
2790 : * clutter log.
2791 : */
2792 5860 : if (FatalError || Shutdown == ImmediateShutdown)
2793 5844 : return;
2794 :
2795 16 : LogChildExit(LOG, procname, pid, exitstatus);
2796 16 : ereport(LOG,
2797 : (errmsg("terminating any other active server processes")));
2798 :
2799 : /*
2800 : * Switch into error state. The crashed process has already been removed
2801 : * from ActiveChildList.
2802 : */
2803 16 : HandleFatalError(PMQUIT_FOR_CRASH, true);
2804 : }
2805 :
2806 : /*
2807 : * Log the death of a child process.
2808 : */
2809 : static void
2810 34574 : LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2811 : {
2812 : /*
2813 : * size of activity_buffer is arbitrary, but set equal to default
2814 : * track_activity_query_size
2815 : */
2816 : char activity_buffer[1024];
2817 34574 : const char *activity = NULL;
2818 :
2819 34574 : if (!EXIT_STATUS_0(exitstatus))
2820 1808 : activity = pgstat_get_crashed_backend_activity(pid,
2821 : activity_buffer,
2822 : sizeof(activity_buffer));
2823 :
2824 34574 : if (WIFEXITED(exitstatus))
2825 34566 : ereport(lev,
2826 :
2827 : /*------
2828 : translator: %s is a noun phrase describing a child process, such as
2829 : "server process" */
2830 : (errmsg("%s (PID %d) exited with exit code %d",
2831 : procname, pid, WEXITSTATUS(exitstatus)),
2832 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2833 8 : else if (WIFSIGNALED(exitstatus))
2834 : {
2835 : #if defined(WIN32)
2836 : ereport(lev,
2837 :
2838 : /*------
2839 : translator: %s is a noun phrase describing a child process, such as
2840 : "server process" */
2841 : (errmsg("%s (PID %d) was terminated by exception 0x%X",
2842 : procname, pid, WTERMSIG(exitstatus)),
2843 : errhint("See C include file \"ntstatus.h\" for a description of the hexadecimal value."),
2844 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2845 : #else
2846 8 : ereport(lev,
2847 :
2848 : /*------
2849 : translator: %s is a noun phrase describing a child process, such as
2850 : "server process" */
2851 : (errmsg("%s (PID %d) was terminated by signal %d: %s",
2852 : procname, pid, WTERMSIG(exitstatus),
2853 : pg_strsignal(WTERMSIG(exitstatus))),
2854 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2855 : #endif
2856 : }
2857 : else
2858 0 : ereport(lev,
2859 :
2860 : /*------
2861 : translator: %s is a noun phrase describing a child process, such as
2862 : "server process" */
2863 : (errmsg("%s (PID %d) exited with unrecognized status %d",
2864 : procname, pid, exitstatus),
2865 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2866 34574 : }
2867 :
2868 : /*
2869 : * Advance the postmaster's state machine and take actions as appropriate
2870 : *
2871 : * This is common code for process_pm_shutdown_request(),
2872 : * process_pm_child_exit() and process_pm_pmsignal(), which process the signals
2873 : * that might mean we need to change state.
2874 : */
2875 : static void
2876 49676 : PostmasterStateMachine(void)
2877 : {
2878 : /* If we're doing a smart shutdown, try to advance that state. */
2879 49676 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2880 : {
2881 36148 : if (!connsAllowed)
2882 : {
2883 : /*
2884 : * This state ends when we have no normal client backends running.
2885 : * Then we're ready to stop other children.
2886 : */
2887 148 : if (CountChildren(btmask(B_BACKEND)) == 0)
2888 76 : UpdatePMState(PM_STOP_BACKENDS);
2889 : }
2890 : }
2891 :
2892 : /*
2893 : * In the PM_WAIT_BACKENDS state, wait for all the regular backends and
2894 : * processes like autovacuum and background workers that are comparable to
2895 : * backends to exit.
2896 : *
2897 : * PM_STOP_BACKENDS is a transient state that means the same as
2898 : * PM_WAIT_BACKENDS, but we signal the processes first, before waiting for
2899 : * them. Treating it as a distinct pmState allows us to share this code
2900 : * across multiple shutdown code paths.
2901 : */
2902 49676 : if (pmState == PM_STOP_BACKENDS || pmState == PM_WAIT_BACKENDS)
2903 : {
2904 8488 : BackendTypeMask targetMask = BTYPE_MASK_NONE;
2905 :
2906 : /*
2907 : * PM_WAIT_BACKENDS state ends when we have no regular backends, no
2908 : * autovac launcher or workers, and no bgworkers (including
2909 : * unconnected ones).
2910 : */
2911 8488 : targetMask = btmask_add(targetMask,
2912 : B_BACKEND,
2913 : B_AUTOVAC_LAUNCHER,
2914 : B_AUTOVAC_WORKER,
2915 : B_BG_WORKER);
2916 :
2917 : /*
2918 : * No walwriter, bgwriter, slot sync worker, or WAL summarizer either.
2919 : */
2920 8488 : targetMask = btmask_add(targetMask,
2921 : B_WAL_WRITER,
2922 : B_BG_WRITER,
2923 : B_SLOTSYNC_WORKER,
2924 : B_WAL_SUMMARIZER);
2925 :
2926 : /* If we're in recovery, also stop startup and walreceiver procs */
2927 8488 : targetMask = btmask_add(targetMask,
2928 : B_STARTUP,
2929 : B_WAL_RECEIVER);
2930 :
2931 : /*
2932 : * If we are doing crash recovery or an immediate shutdown then we
2933 : * expect archiver, checkpointer, io workers and walsender to exit as
2934 : * well, otherwise not.
2935 : */
2936 8488 : if (FatalError || Shutdown >= ImmediateShutdown)
2937 3380 : targetMask = btmask_add(targetMask,
2938 : B_CHECKPOINTER,
2939 : B_ARCHIVER,
2940 : B_IO_WORKER,
2941 : B_WAL_SENDER);
2942 :
2943 : /*
2944 : * Normally archiver, checkpointer, IO workers and walsenders will
2945 : * continue running; they will be terminated later after writing the
2946 : * checkpoint record. We also let dead-end children to keep running
2947 : * for now. The syslogger process exits last.
2948 : *
2949 : * This assertion checks that we have covered all backend types,
2950 : * either by including them in targetMask, or by noting here that they
2951 : * are allowed to continue running.
2952 : */
2953 : #ifdef USE_ASSERT_CHECKING
2954 : {
2955 : BackendTypeMask remainMask = BTYPE_MASK_NONE;
2956 :
2957 : remainMask = btmask_add(remainMask,
2958 : B_DEAD_END_BACKEND,
2959 : B_LOGGER);
2960 :
2961 : /*
2962 : * Archiver, checkpointer, IO workers, and walsender may or may
2963 : * not be in targetMask already.
2964 : */
2965 : remainMask = btmask_add(remainMask,
2966 : B_ARCHIVER,
2967 : B_CHECKPOINTER,
2968 : B_IO_WORKER,
2969 : B_WAL_SENDER);
2970 :
2971 : /* these are not real postmaster children */
2972 : remainMask = btmask_add(remainMask,
2973 : B_INVALID,
2974 : B_STANDALONE_BACKEND);
2975 :
2976 : /* All types should be included in targetMask or remainMask */
2977 : Assert((remainMask.mask | targetMask.mask) == BTYPE_MASK_ALL.mask);
2978 : }
2979 : #endif
2980 :
2981 : /* If we had not yet signaled the processes to exit, do so now */
2982 8488 : if (pmState == PM_STOP_BACKENDS)
2983 : {
2984 : /*
2985 : * Forget any pending requests for background workers, since we're
2986 : * no longer willing to launch any new workers. (If additional
2987 : * requests arrive, BackgroundWorkerStateChange will reject them.)
2988 : */
2989 998 : ForgetUnstartedBackgroundWorkers();
2990 :
2991 998 : SignalChildren(SIGTERM, targetMask);
2992 :
2993 998 : UpdatePMState(PM_WAIT_BACKENDS);
2994 : }
2995 :
2996 : /* Are any of the target processes still running? */
2997 8488 : if (CountChildren(targetMask) == 0)
2998 : {
2999 1682 : if (Shutdown >= ImmediateShutdown || FatalError)
3000 : {
3001 : /*
3002 : * Stop any dead-end children and stop creating new ones.
3003 : *
3004 : * NB: Similar code exists in HandleFatalError(), when the
3005 : * error happens in pmState > PM_WAIT_BACKENDS.
3006 : */
3007 684 : UpdatePMState(PM_WAIT_DEAD_END);
3008 684 : ConfigurePostmasterWaitSet(false);
3009 684 : SignalChildren(SIGQUIT, btmask(B_DEAD_END_BACKEND));
3010 :
3011 : /*
3012 : * We already SIGQUIT'd auxiliary processes (other than
3013 : * logger), if any, when we started immediate shutdown or
3014 : * entered FatalError state.
3015 : */
3016 : }
3017 : else
3018 : {
3019 : /*
3020 : * If we get here, we are proceeding with normal shutdown. All
3021 : * the regular children are gone, and it's time to tell the
3022 : * checkpointer to do a shutdown checkpoint.
3023 : */
3024 : Assert(Shutdown > NoShutdown);
3025 : /* Start the checkpointer if not running */
3026 998 : if (CheckpointerPMChild == NULL)
3027 0 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3028 : /* And tell it to write the shutdown checkpoint */
3029 998 : if (CheckpointerPMChild != NULL)
3030 : {
3031 998 : signal_child(CheckpointerPMChild, SIGINT);
3032 998 : UpdatePMState(PM_WAIT_XLOG_SHUTDOWN);
3033 : }
3034 : else
3035 : {
3036 : /*
3037 : * If we failed to fork a checkpointer, just shut down.
3038 : * Any required cleanup will happen at next restart. We
3039 : * set FatalError so that an "abnormal shutdown" message
3040 : * gets logged when we exit.
3041 : *
3042 : * We don't consult send_abort_for_crash here, as it's
3043 : * unlikely that dumping cores would illuminate the reason
3044 : * for checkpointer fork failure.
3045 : *
3046 : * XXX: It may be worth to introduce a different PMQUIT
3047 : * value that signals that the cluster is in a bad state,
3048 : * without a process having crashed. But right now this
3049 : * path is very unlikely to be reached, so it isn't
3050 : * obviously worthwhile adding a distinct error message in
3051 : * quickdie().
3052 : */
3053 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3054 : }
3055 : }
3056 : }
3057 : }
3058 :
3059 : /*
3060 : * The state transition from PM_WAIT_XLOG_SHUTDOWN to
3061 : * PM_WAIT_XLOG_ARCHIVAL is in process_pm_pmsignal(), in response to
3062 : * PMSIGNAL_XLOG_IS_SHUTDOWN.
3063 : */
3064 :
3065 49676 : if (pmState == PM_WAIT_XLOG_ARCHIVAL)
3066 : {
3067 : /*
3068 : * PM_WAIT_XLOG_ARCHIVAL state ends when there are no children other
3069 : * than checkpointer, io workers and dead-end children left. There
3070 : * shouldn't be any regular backends left by now anyway; what we're
3071 : * really waiting for is for walsenders and archiver to exit.
3072 : */
3073 1102 : if (CountChildren(btmask_all_except(B_CHECKPOINTER, B_IO_WORKER,
3074 : B_LOGGER, B_DEAD_END_BACKEND)) == 0)
3075 : {
3076 998 : UpdatePMState(PM_WAIT_IO_WORKERS);
3077 998 : SignalChildren(SIGUSR2, btmask(B_IO_WORKER));
3078 : }
3079 : }
3080 :
3081 49676 : if (pmState == PM_WAIT_IO_WORKERS)
3082 : {
3083 : /*
3084 : * PM_WAIT_IO_WORKERS state ends when there's only checkpointer and
3085 : * dead-end children left.
3086 : */
3087 3898 : if (io_worker_count == 0)
3088 : {
3089 998 : UpdatePMState(PM_WAIT_CHECKPOINTER);
3090 :
3091 : /*
3092 : * Now that the processes mentioned above are gone, tell
3093 : * checkpointer to shut down too. That allows checkpointer to
3094 : * perform some last bits of cleanup without other processes
3095 : * interfering.
3096 : */
3097 998 : if (CheckpointerPMChild != NULL)
3098 998 : signal_child(CheckpointerPMChild, SIGUSR2);
3099 : }
3100 : }
3101 :
3102 : /*
3103 : * The state transition from PM_WAIT_CHECKPOINTER to PM_WAIT_DEAD_END is
3104 : * in process_pm_child_exit().
3105 : */
3106 :
3107 49676 : if (pmState == PM_WAIT_DEAD_END)
3108 : {
3109 : /*
3110 : * PM_WAIT_DEAD_END state ends when all other children are gone except
3111 : * for the logger. During normal shutdown, all that remains are
3112 : * dead-end backends, but in FatalError processing we jump straight
3113 : * here with more processes remaining. Note that they have already
3114 : * been sent appropriate shutdown signals, either during a normal
3115 : * state transition leading up to PM_WAIT_DEAD_END, or during
3116 : * FatalError processing.
3117 : *
3118 : * The reason we wait is to protect against a new postmaster starting
3119 : * conflicting subprocesses; this isn't an ironclad protection, but it
3120 : * at least helps in the shutdown-and-immediately-restart scenario.
3121 : */
3122 1740 : if (CountChildren(btmask_all_except(B_LOGGER)) == 0)
3123 : {
3124 : /* These other guys should be dead already */
3125 : Assert(StartupPMChild == NULL);
3126 : Assert(WalReceiverPMChild == NULL);
3127 : Assert(WalSummarizerPMChild == NULL);
3128 : Assert(BgWriterPMChild == NULL);
3129 : Assert(CheckpointerPMChild == NULL);
3130 : Assert(WalWriterPMChild == NULL);
3131 : Assert(AutoVacLauncherPMChild == NULL);
3132 : Assert(SlotSyncWorkerPMChild == NULL);
3133 : /* syslogger is not considered here */
3134 1682 : UpdatePMState(PM_NO_CHILDREN);
3135 : }
3136 : }
3137 :
3138 : /*
3139 : * If we've been told to shut down, we exit as soon as there are no
3140 : * remaining children. If there was a crash, cleanup will occur at the
3141 : * next startup. (Before PostgreSQL 8.3, we tried to recover from the
3142 : * crash before exiting, but that seems unwise if we are quitting because
3143 : * we got SIGTERM from init --- there may well not be time for recovery
3144 : * before init decides to SIGKILL us.)
3145 : *
3146 : * Note that the syslogger continues to run. It will exit when it sees
3147 : * EOF on its input pipe, which happens when there are no more upstream
3148 : * processes.
3149 : */
3150 49676 : if (Shutdown > NoShutdown && pmState == PM_NO_CHILDREN)
3151 : {
3152 1666 : if (FatalError)
3153 : {
3154 0 : ereport(LOG, (errmsg("abnormal database system shutdown")));
3155 0 : ExitPostmaster(1);
3156 : }
3157 : else
3158 : {
3159 : /*
3160 : * Normal exit from the postmaster is here. We don't need to log
3161 : * anything here, since the UnlinkLockFiles proc_exit callback
3162 : * will do so, and that should be the last user-visible action.
3163 : */
3164 1666 : ExitPostmaster(0);
3165 : }
3166 : }
3167 :
3168 : /*
3169 : * If the startup process failed, or the user does not want an automatic
3170 : * restart after backend crashes, wait for all non-syslogger children to
3171 : * exit, and then exit postmaster. We don't try to reinitialize when the
3172 : * startup process fails, because more than likely it will just fail again
3173 : * and we will keep trying forever.
3174 : */
3175 48010 : if (pmState == PM_NO_CHILDREN)
3176 : {
3177 16 : if (StartupStatus == STARTUP_CRASHED)
3178 : {
3179 6 : ereport(LOG,
3180 : (errmsg("shutting down due to startup process failure")));
3181 6 : ExitPostmaster(1);
3182 : }
3183 10 : if (!restart_after_crash)
3184 : {
3185 0 : ereport(LOG,
3186 : (errmsg("shutting down because \"restart_after_crash\" is off")));
3187 0 : ExitPostmaster(1);
3188 : }
3189 : }
3190 :
3191 : /*
3192 : * If we need to recover from a crash, wait for all non-syslogger children
3193 : * to exit, then reset shmem and start the startup process.
3194 : */
3195 48004 : if (FatalError && pmState == PM_NO_CHILDREN)
3196 : {
3197 10 : ereport(LOG,
3198 : (errmsg("all server processes terminated; reinitializing")));
3199 :
3200 : /* remove leftover temporary files after a crash */
3201 10 : if (remove_temp_files_after_crash)
3202 8 : RemovePgTempFiles();
3203 :
3204 : /* allow background workers to immediately restart */
3205 10 : ResetBackgroundWorkerCrashTimes();
3206 :
3207 10 : shmem_exit(1);
3208 :
3209 : /* re-read control file into local memory */
3210 10 : LocalProcessControlFile(true);
3211 :
3212 : /* re-create shared memory and semaphores */
3213 10 : CreateSharedMemoryAndSemaphores();
3214 :
3215 10 : UpdatePMState(PM_STARTUP);
3216 :
3217 : /* Make sure we can perform I/O while starting up. */
3218 10 : maybe_adjust_io_workers();
3219 :
3220 10 : StartupPMChild = StartChildProcess(B_STARTUP);
3221 : Assert(StartupPMChild != NULL);
3222 10 : StartupStatus = STARTUP_RUNNING;
3223 : /* crash recovery started, reset SIGKILL flag */
3224 10 : AbortStartTime = 0;
3225 :
3226 : /* start accepting server socket connection events again */
3227 10 : ConfigurePostmasterWaitSet(true);
3228 : }
3229 48004 : }
3230 :
3231 : static const char *
3232 1940 : pmstate_name(PMState state)
3233 : {
3234 : #define PM_TOSTR_CASE(sym) case sym: return #sym
3235 1940 : switch (state)
3236 : {
3237 108 : PM_TOSTR_CASE(PM_INIT);
3238 216 : PM_TOSTR_CASE(PM_STARTUP);
3239 32 : PM_TOSTR_CASE(PM_RECOVERY);
3240 24 : PM_TOSTR_CASE(PM_HOT_STANDBY);
3241 202 : PM_TOSTR_CASE(PM_RUN);
3242 160 : PM_TOSTR_CASE(PM_STOP_BACKENDS);
3243 228 : PM_TOSTR_CASE(PM_WAIT_BACKENDS);
3244 160 : PM_TOSTR_CASE(PM_WAIT_XLOG_SHUTDOWN);
3245 160 : PM_TOSTR_CASE(PM_WAIT_XLOG_ARCHIVAL);
3246 160 : PM_TOSTR_CASE(PM_WAIT_IO_WORKERS);
3247 220 : PM_TOSTR_CASE(PM_WAIT_DEAD_END);
3248 160 : PM_TOSTR_CASE(PM_WAIT_CHECKPOINTER);
3249 110 : PM_TOSTR_CASE(PM_NO_CHILDREN);
3250 : }
3251 : #undef PM_TOSTR_CASE
3252 :
3253 0 : pg_unreachable();
3254 : return ""; /* silence compiler */
3255 : }
3256 :
3257 : /*
3258 : * Simple wrapper for updating pmState. The main reason to have this wrapper
3259 : * is that it makes it easy to log all state transitions.
3260 : */
3261 : static void
3262 14108 : UpdatePMState(PMState newState)
3263 : {
3264 14108 : elog(DEBUG1, "updating PMState from %s to %s",
3265 : pmstate_name(pmState), pmstate_name(newState));
3266 14108 : pmState = newState;
3267 14108 : }
3268 :
3269 : /*
3270 : * Launch background processes after state change, or relaunch after an
3271 : * existing process has exited.
3272 : *
3273 : * Check the current pmState and the status of any background processes. If
3274 : * there are any background processes missing that should be running in the
3275 : * current state, but are not, launch them.
3276 : */
3277 : static void
3278 244910 : LaunchMissingBackgroundProcesses(void)
3279 : {
3280 : /* Syslogger is active in all states */
3281 244910 : if (SysLoggerPMChild == NULL && Logging_collector)
3282 0 : StartSysLogger();
3283 :
3284 : /*
3285 : * The number of configured workers might have changed, or a prior start
3286 : * of a worker might have failed. Check if we need to start/stop any
3287 : * workers.
3288 : *
3289 : * A config file change will always lead to this function being called, so
3290 : * we always will process the config change in a timely manner.
3291 : */
3292 244910 : maybe_adjust_io_workers();
3293 :
3294 : /*
3295 : * The checkpointer and the background writer are active from the start,
3296 : * until shutdown is initiated.
3297 : *
3298 : * (If the checkpointer is not running when we enter the
3299 : * PM_WAIT_XLOG_SHUTDOWN state, it is launched one more time to perform
3300 : * the shutdown checkpoint. That's done in PostmasterStateMachine(), not
3301 : * here.)
3302 : */
3303 244910 : if (pmState == PM_RUN || pmState == PM_RECOVERY ||
3304 15598 : pmState == PM_HOT_STANDBY || pmState == PM_STARTUP)
3305 : {
3306 232992 : if (CheckpointerPMChild == NULL)
3307 10 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3308 232992 : if (BgWriterPMChild == NULL)
3309 10 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
3310 : }
3311 :
3312 : /*
3313 : * WAL writer is needed only in normal operation (else we cannot be
3314 : * writing any new WAL).
3315 : */
3316 244910 : if (WalWriterPMChild == NULL && pmState == PM_RUN)
3317 1484 : WalWriterPMChild = StartChildProcess(B_WAL_WRITER);
3318 :
3319 : /*
3320 : * We don't want autovacuum to run in binary upgrade mode because
3321 : * autovacuum might update relfrozenxid for empty tables before the
3322 : * physical files are put in place.
3323 : */
3324 263268 : if (!IsBinaryUpgrade && AutoVacLauncherPMChild == NULL &&
3325 25672 : (AutoVacuumingActive() || start_autovac_launcher) &&
3326 11044 : pmState == PM_RUN)
3327 : {
3328 1248 : AutoVacLauncherPMChild = StartChildProcess(B_AUTOVAC_LAUNCHER);
3329 1248 : if (AutoVacLauncherPMChild != NULL)
3330 1248 : start_autovac_launcher = false; /* signal processed */
3331 : }
3332 :
3333 : /*
3334 : * If WAL archiving is enabled always, we are allowed to start archiver
3335 : * even during recovery.
3336 : */
3337 244910 : if (PgArchPMChild == NULL &&
3338 242792 : ((XLogArchivingActive() && pmState == PM_RUN) ||
3339 242792 : (XLogArchivingAlways() && (pmState == PM_RECOVERY || pmState == PM_HOT_STANDBY))) &&
3340 104 : PgArchCanRestart())
3341 104 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3342 :
3343 : /*
3344 : * If we need to start a slot sync worker, try to do that now
3345 : *
3346 : * We allow to start the slot sync worker when we are on a hot standby,
3347 : * fast or immediate shutdown is not in progress, slot sync parameters are
3348 : * configured correctly, and it is the first time of worker's launch, or
3349 : * enough time has passed since the worker was launched last.
3350 : */
3351 244910 : if (SlotSyncWorkerPMChild == NULL && pmState == PM_HOT_STANDBY &&
3352 3414 : Shutdown <= SmartShutdown && sync_replication_slots &&
3353 26 : ValidateSlotSyncParams(LOG) && SlotSyncWorkerCanRestart())
3354 8 : SlotSyncWorkerPMChild = StartChildProcess(B_SLOTSYNC_WORKER);
3355 :
3356 : /*
3357 : * If we need to start a WAL receiver, try to do that now
3358 : *
3359 : * Note: if a walreceiver process is already running, it might seem that
3360 : * we should clear WalReceiverRequested. However, there's a race
3361 : * condition if the walreceiver terminates and the startup process
3362 : * immediately requests a new one: it's quite possible to get the signal
3363 : * for the request before reaping the dead walreceiver process. Better to
3364 : * risk launching an extra walreceiver than to miss launching one we need.
3365 : * (The walreceiver code has logic to recognize that it should go away if
3366 : * not needed.)
3367 : */
3368 244910 : if (WalReceiverRequested)
3369 : {
3370 686 : if (WalReceiverPMChild == NULL &&
3371 498 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
3372 496 : pmState == PM_HOT_STANDBY) &&
3373 492 : Shutdown <= SmartShutdown)
3374 : {
3375 492 : WalReceiverPMChild = StartChildProcess(B_WAL_RECEIVER);
3376 492 : if (WalReceiverPMChild != 0)
3377 492 : WalReceiverRequested = false;
3378 : /* else leave the flag set, so we'll try again later */
3379 : }
3380 : }
3381 :
3382 : /* If we need to start a WAL summarizer, try to do that now */
3383 244910 : if (summarize_wal && WalSummarizerPMChild == NULL &&
3384 128 : (pmState == PM_RUN || pmState == PM_HOT_STANDBY) &&
3385 38 : Shutdown <= SmartShutdown)
3386 38 : WalSummarizerPMChild = StartChildProcess(B_WAL_SUMMARIZER);
3387 :
3388 : /* Get other worker processes running, if needed */
3389 244910 : if (StartWorkerNeeded || HaveCrashedWorker)
3390 12718 : maybe_start_bgworkers();
3391 244906 : }
3392 :
3393 : /*
3394 : * Return string representation of signal.
3395 : *
3396 : * Because this is only implemented for signals we already rely on in this
3397 : * file we don't need to deal with unimplemented or same-numeric-value signals
3398 : * (as we'd e.g. have to for EWOULDBLOCK / EAGAIN).
3399 : */
3400 : static const char *
3401 34 : pm_signame(int signal)
3402 : {
3403 : #define PM_TOSTR_CASE(sym) case sym: return #sym
3404 34 : switch (signal)
3405 : {
3406 0 : PM_TOSTR_CASE(SIGABRT);
3407 0 : PM_TOSTR_CASE(SIGCHLD);
3408 0 : PM_TOSTR_CASE(SIGHUP);
3409 4 : PM_TOSTR_CASE(SIGINT);
3410 0 : PM_TOSTR_CASE(SIGKILL);
3411 0 : PM_TOSTR_CASE(SIGQUIT);
3412 20 : PM_TOSTR_CASE(SIGTERM);
3413 0 : PM_TOSTR_CASE(SIGUSR1);
3414 10 : PM_TOSTR_CASE(SIGUSR2);
3415 0 : default:
3416 : /* all signals sent by postmaster should be listed here */
3417 : Assert(false);
3418 0 : return "(unknown)";
3419 : }
3420 : #undef PM_TOSTR_CASE
3421 :
3422 : return ""; /* silence compiler */
3423 : }
3424 :
3425 : /*
3426 : * Send a signal to a postmaster child process
3427 : *
3428 : * On systems that have setsid(), each child process sets itself up as a
3429 : * process group leader. For signals that are generally interpreted in the
3430 : * appropriate fashion, we signal the entire process group not just the
3431 : * direct child process. This allows us to, for example, SIGQUIT a blocked
3432 : * archive_recovery script, or SIGINT a script being run by a backend via
3433 : * system().
3434 : *
3435 : * There is a race condition for recently-forked children: they might not
3436 : * have executed setsid() yet. So we signal the child directly as well as
3437 : * the group. We assume such a child will handle the signal before trying
3438 : * to spawn any grandchild processes. We also assume that signaling the
3439 : * child twice will not cause any problems.
3440 : */
3441 : static void
3442 17888 : signal_child(PMChild *pmchild, int signal)
3443 : {
3444 17888 : pid_t pid = pmchild->pid;
3445 :
3446 17888 : ereport(DEBUG3,
3447 : (errmsg_internal("sending signal %d/%s to %s process with pid %d",
3448 : signal, pm_signame(signal),
3449 : GetBackendTypeDesc(pmchild->bkend_type),
3450 : (int) pmchild->pid)));
3451 :
3452 17888 : if (kill(pid, signal) < 0)
3453 0 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) pid, signal);
3454 : #ifdef HAVE_SETSID
3455 17888 : switch (signal)
3456 : {
3457 11178 : case SIGINT:
3458 : case SIGTERM:
3459 : case SIGQUIT:
3460 : case SIGKILL:
3461 : case SIGABRT:
3462 11178 : if (kill(-pid, signal) < 0)
3463 70 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) (-pid), signal);
3464 11178 : break;
3465 6710 : default:
3466 6710 : break;
3467 : }
3468 : #endif
3469 17888 : }
3470 :
3471 : /*
3472 : * Send a signal to the targeted children.
3473 : */
3474 : static bool
3475 5626 : SignalChildren(int signal, BackendTypeMask targetMask)
3476 : {
3477 : dlist_iter iter;
3478 5626 : bool signaled = false;
3479 :
3480 30710 : dlist_foreach(iter, &ActiveChildList)
3481 : {
3482 25084 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3483 :
3484 : /*
3485 : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3486 : * if any B_BACKEND backends have recently announced that they are
3487 : * actually WAL senders.
3488 : */
3489 25084 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3490 12616 : bp->bkend_type == B_BACKEND)
3491 : {
3492 654 : if (IsPostmasterChildWalSender(bp->child_slot))
3493 76 : bp->bkend_type = B_WAL_SENDER;
3494 : }
3495 :
3496 25084 : if (!btmask_contains(targetMask, bp->bkend_type))
3497 9308 : continue;
3498 :
3499 15776 : signal_child(bp, signal);
3500 15776 : signaled = true;
3501 : }
3502 5626 : return signaled;
3503 : }
3504 :
3505 : /*
3506 : * Send a termination signal to children. This considers all of our children
3507 : * processes, except syslogger.
3508 : */
3509 : static void
3510 684 : TerminateChildren(int signal)
3511 : {
3512 684 : SignalChildren(signal, btmask_all_except(B_LOGGER));
3513 684 : if (StartupPMChild != NULL)
3514 : {
3515 90 : if (signal == SIGQUIT || signal == SIGKILL || signal == SIGABRT)
3516 90 : StartupStatus = STARTUP_SIGNALED;
3517 : }
3518 684 : }
3519 :
3520 : /*
3521 : * BackendStartup -- start backend process
3522 : *
3523 : * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
3524 : *
3525 : * Note: if you change this code, also consider StartAutovacuumWorker and
3526 : * StartBackgroundWorker.
3527 : */
3528 : static int
3529 28214 : BackendStartup(ClientSocket *client_sock)
3530 : {
3531 28214 : PMChild *bn = NULL;
3532 : pid_t pid;
3533 : BackendStartupData startup_data;
3534 : CAC_state cac;
3535 :
3536 : /*
3537 : * Capture time that Postmaster got a socket from accept (for logging
3538 : * connection establishment and setup total duration).
3539 : */
3540 28214 : startup_data.socket_created = GetCurrentTimestamp();
3541 :
3542 : /*
3543 : * Allocate and assign the child slot. Note we must do this before
3544 : * forking, so that we can handle failures (out of memory or child-process
3545 : * slots) cleanly.
3546 : */
3547 28214 : cac = canAcceptConnections(B_BACKEND);
3548 28214 : if (cac == CAC_OK)
3549 : {
3550 : /* Can change later to B_WAL_SENDER */
3551 27926 : bn = AssignPostmasterChildSlot(B_BACKEND);
3552 27926 : if (!bn)
3553 : {
3554 : /*
3555 : * Too many regular child processes; launch a dead-end child
3556 : * process instead.
3557 : */
3558 56 : cac = CAC_TOOMANY;
3559 : }
3560 : }
3561 28214 : if (!bn)
3562 : {
3563 344 : bn = AllocDeadEndChild();
3564 344 : if (!bn)
3565 : {
3566 0 : ereport(LOG,
3567 : (errcode(ERRCODE_OUT_OF_MEMORY),
3568 : errmsg("out of memory")));
3569 0 : return STATUS_ERROR;
3570 : }
3571 : }
3572 :
3573 : /* Pass down canAcceptConnections state */
3574 28214 : startup_data.canAcceptConnections = cac;
3575 28214 : bn->rw = NULL;
3576 :
3577 : /* Hasn't asked to be notified about any bgworkers yet */
3578 28214 : bn->bgworker_notify = false;
3579 :
3580 28214 : pid = postmaster_child_launch(bn->bkend_type, bn->child_slot,
3581 : &startup_data, sizeof(startup_data),
3582 : client_sock);
3583 28212 : if (pid < 0)
3584 : {
3585 : /* in parent, fork failed */
3586 0 : int save_errno = errno;
3587 :
3588 0 : (void) ReleasePostmasterChildSlot(bn);
3589 0 : errno = save_errno;
3590 0 : ereport(LOG,
3591 : (errmsg("could not fork new process for connection: %m")));
3592 0 : report_fork_failure_to_client(client_sock, save_errno);
3593 0 : return STATUS_ERROR;
3594 : }
3595 :
3596 : /* in parent, successful fork */
3597 28212 : ereport(DEBUG2,
3598 : (errmsg_internal("forked new %s, pid=%d socket=%d",
3599 : GetBackendTypeDesc(bn->bkend_type),
3600 : (int) pid, (int) client_sock->sock)));
3601 :
3602 : /*
3603 : * Everything's been successful, it's safe to add this backend to our list
3604 : * of backends.
3605 : */
3606 28212 : bn->pid = pid;
3607 28212 : return STATUS_OK;
3608 : }
3609 :
3610 : /*
3611 : * Try to report backend fork() failure to client before we close the
3612 : * connection. Since we do not care to risk blocking the postmaster on
3613 : * this connection, we set the connection to non-blocking and try only once.
3614 : *
3615 : * This is grungy special-purpose code; we cannot use backend libpq since
3616 : * it's not up and running.
3617 : */
3618 : static void
3619 0 : report_fork_failure_to_client(ClientSocket *client_sock, int errnum)
3620 : {
3621 : char buffer[1000];
3622 : int rc;
3623 :
3624 : /* Format the error message packet (always V2 protocol) */
3625 0 : snprintf(buffer, sizeof(buffer), "E%s%s\n",
3626 : _("could not fork new process for connection: "),
3627 : strerror(errnum));
3628 :
3629 : /* Set port to non-blocking. Don't do send() if this fails */
3630 0 : if (!pg_set_noblock(client_sock->sock))
3631 0 : return;
3632 :
3633 : /* We'll retry after EINTR, but ignore all other failures */
3634 : do
3635 : {
3636 0 : rc = send(client_sock->sock, buffer, strlen(buffer) + 1, 0);
3637 0 : } while (rc < 0 && errno == EINTR);
3638 : }
3639 :
3640 : /*
3641 : * ExitPostmaster -- cleanup
3642 : *
3643 : * Do NOT call exit() directly --- always go through here!
3644 : */
3645 : static void
3646 1678 : ExitPostmaster(int status)
3647 : {
3648 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
3649 :
3650 : /*
3651 : * There is no known cause for a postmaster to become multithreaded after
3652 : * startup. However, we might reach here via an error exit before
3653 : * reaching the test in PostmasterMain, so provide the same hint as there.
3654 : * This message uses LOG level, because an unclean shutdown at this point
3655 : * would usually not look much different from a clean shutdown.
3656 : */
3657 : if (pthread_is_threaded_np() != 0)
3658 : ereport(LOG,
3659 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3660 : errmsg("postmaster became multithreaded"),
3661 : errhint("Set the LC_ALL environment variable to a valid locale.")));
3662 : #endif
3663 :
3664 : /* should cleanup shared memory and kill all backends */
3665 :
3666 : /*
3667 : * Not sure of the semantics here. When the Postmaster dies, should the
3668 : * backends all be killed? probably not.
3669 : *
3670 : * MUST -- vadim 05-10-1999
3671 : */
3672 :
3673 1678 : proc_exit(status);
3674 : }
3675 :
3676 : /*
3677 : * Handle pmsignal conditions representing requests from backends,
3678 : * and check for promote and logrotate requests from pg_ctl.
3679 : */
3680 : static void
3681 173060 : process_pm_pmsignal(void)
3682 : {
3683 173060 : bool request_state_update = false;
3684 :
3685 173060 : pending_pm_pmsignal = false;
3686 :
3687 173060 : ereport(DEBUG2,
3688 : (errmsg_internal("postmaster received pmsignal signal")));
3689 :
3690 : /*
3691 : * RECOVERY_STARTED and BEGIN_HOT_STANDBY signals are ignored in
3692 : * unexpected states. If the startup process quickly starts up, completes
3693 : * recovery, exits, we might process the death of the startup process
3694 : * first. We don't want to go back to recovery in that case.
3695 : */
3696 173060 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_STARTED) &&
3697 492 : pmState == PM_STARTUP && Shutdown == NoShutdown)
3698 : {
3699 : /* WAL redo has started. We're out of reinitialization. */
3700 492 : FatalError = false;
3701 492 : AbortStartTime = 0;
3702 492 : reachedConsistency = false;
3703 :
3704 : /*
3705 : * Start the archiver if we're responsible for (re-)archiving received
3706 : * files.
3707 : */
3708 : Assert(PgArchPMChild == NULL);
3709 492 : if (XLogArchivingAlways())
3710 6 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3711 :
3712 : /*
3713 : * If we aren't planning to enter hot standby mode later, treat
3714 : * RECOVERY_STARTED as meaning we're out of startup, and report status
3715 : * accordingly.
3716 : */
3717 492 : if (!EnableHotStandby)
3718 : {
3719 4 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STANDBY);
3720 : #ifdef USE_SYSTEMD
3721 : sd_notify(0, "READY=1");
3722 : #endif
3723 : }
3724 :
3725 492 : UpdatePMState(PM_RECOVERY);
3726 : }
3727 :
3728 173060 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_CONSISTENT) &&
3729 314 : pmState == PM_RECOVERY && Shutdown == NoShutdown)
3730 : {
3731 314 : reachedConsistency = true;
3732 : }
3733 :
3734 173060 : if (CheckPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY) &&
3735 298 : (pmState == PM_RECOVERY && Shutdown == NoShutdown))
3736 : {
3737 298 : ereport(LOG,
3738 : (errmsg("database system is ready to accept read-only connections")));
3739 :
3740 : /* Report status */
3741 298 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
3742 : #ifdef USE_SYSTEMD
3743 : sd_notify(0, "READY=1");
3744 : #endif
3745 :
3746 298 : UpdatePMState(PM_HOT_STANDBY);
3747 298 : connsAllowed = true;
3748 :
3749 : /* Some workers may be scheduled to start now */
3750 298 : StartWorkerNeeded = true;
3751 : }
3752 :
3753 : /* Process background worker state changes. */
3754 173060 : if (CheckPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE))
3755 : {
3756 : /* Accept new worker requests only if not stopping. */
3757 2186 : BackgroundWorkerStateChange(pmState < PM_STOP_BACKENDS);
3758 2186 : StartWorkerNeeded = true;
3759 : }
3760 :
3761 : /* Tell syslogger to rotate logfile if requested */
3762 173060 : if (SysLoggerPMChild != NULL)
3763 : {
3764 4 : if (CheckLogrotateSignal())
3765 : {
3766 2 : signal_child(SysLoggerPMChild, SIGUSR1);
3767 2 : RemoveLogrotateSignalFiles();
3768 : }
3769 2 : else if (CheckPostmasterSignal(PMSIGNAL_ROTATE_LOGFILE))
3770 : {
3771 0 : signal_child(SysLoggerPMChild, SIGUSR1);
3772 : }
3773 : }
3774 :
3775 173060 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER) &&
3776 164046 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3777 : {
3778 : /*
3779 : * Start one iteration of the autovacuum daemon, even if autovacuuming
3780 : * is nominally not enabled. This is so we can have an active defense
3781 : * against transaction ID wraparound. We set a flag for the main loop
3782 : * to do it rather than trying to do it here --- this is because the
3783 : * autovac process itself may send the signal, and we want to handle
3784 : * that by launching another iteration as soon as the current one
3785 : * completes.
3786 : */
3787 164046 : start_autovac_launcher = true;
3788 : }
3789 :
3790 173060 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER) &&
3791 2424 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3792 : {
3793 : /* The autovacuum launcher wants us to start a worker process. */
3794 2424 : StartAutovacuumWorker();
3795 : }
3796 :
3797 173060 : if (CheckPostmasterSignal(PMSIGNAL_START_WALRECEIVER))
3798 : {
3799 : /* Startup Process wants us to start the walreceiver process. */
3800 496 : WalReceiverRequested = true;
3801 : }
3802 :
3803 173060 : if (CheckPostmasterSignal(PMSIGNAL_XLOG_IS_SHUTDOWN))
3804 : {
3805 : /* Checkpointer completed the shutdown checkpoint */
3806 998 : if (pmState == PM_WAIT_XLOG_SHUTDOWN)
3807 : {
3808 : /*
3809 : * If we have an archiver subprocess, tell it to do a last archive
3810 : * cycle and quit. Likewise, if we have walsender processes, tell
3811 : * them to send any remaining WAL and quit.
3812 : */
3813 : Assert(Shutdown > NoShutdown);
3814 :
3815 : /* Waken archiver for the last time */
3816 998 : if (PgArchPMChild != NULL)
3817 30 : signal_child(PgArchPMChild, SIGUSR2);
3818 :
3819 : /*
3820 : * Waken walsenders for the last time. No regular backends should
3821 : * be around anymore.
3822 : */
3823 998 : SignalChildren(SIGUSR2, btmask(B_WAL_SENDER));
3824 :
3825 998 : UpdatePMState(PM_WAIT_XLOG_ARCHIVAL);
3826 : }
3827 0 : else if (!FatalError && Shutdown != ImmediateShutdown)
3828 : {
3829 : /*
3830 : * Checkpointer only ought to perform the shutdown checkpoint
3831 : * during shutdown. If somehow checkpointer did so in another
3832 : * situation, we have no choice but to crash-restart.
3833 : *
3834 : * It's possible however that we get PMSIGNAL_XLOG_IS_SHUTDOWN
3835 : * outside of PM_WAIT_XLOG_SHUTDOWN if an orderly shutdown was
3836 : * "interrupted" by a crash or an immediate shutdown.
3837 : */
3838 0 : ereport(LOG,
3839 : (errmsg("WAL was shut down unexpectedly")));
3840 :
3841 : /*
3842 : * Doesn't seem likely to help to take send_abort_for_crash into
3843 : * account here.
3844 : */
3845 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3846 : }
3847 :
3848 : /*
3849 : * Need to run PostmasterStateMachine() to check if we already can go
3850 : * to the next state.
3851 : */
3852 998 : request_state_update = true;
3853 : }
3854 :
3855 : /*
3856 : * Try to advance postmaster's state machine, if a child requests it.
3857 : */
3858 173060 : if (CheckPostmasterSignal(PMSIGNAL_ADVANCE_STATE_MACHINE))
3859 : {
3860 2314 : request_state_update = true;
3861 : }
3862 :
3863 : /*
3864 : * Be careful about the order of this action relative to this function's
3865 : * other actions. Generally, this should be after other actions, in case
3866 : * they have effects PostmasterStateMachine would need to know about.
3867 : * However, we should do it before the CheckPromoteSignal step, which
3868 : * cannot have any (immediate) effect on the state machine, but does
3869 : * depend on what state we're in now.
3870 : */
3871 173060 : if (request_state_update)
3872 : {
3873 3312 : PostmasterStateMachine();
3874 : }
3875 :
3876 173060 : if (StartupPMChild != NULL &&
3877 1536 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
3878 2516 : pmState == PM_HOT_STANDBY) &&
3879 1536 : CheckPromoteSignal())
3880 : {
3881 : /*
3882 : * Tell startup process to finish recovery.
3883 : *
3884 : * Leave the promote signal file in place and let the Startup process
3885 : * do the unlink.
3886 : */
3887 84 : signal_child(StartupPMChild, SIGUSR2);
3888 : }
3889 173060 : }
3890 :
3891 : /*
3892 : * Dummy signal handler
3893 : *
3894 : * We use this for signals that we don't actually use in the postmaster,
3895 : * but we do use in backends. If we were to SIG_IGN such signals in the
3896 : * postmaster, then a newly started backend might drop a signal that arrives
3897 : * before it's able to reconfigure its signal processing. (See notes in
3898 : * tcop/postgres.c.)
3899 : */
3900 : static void
3901 0 : dummy_handler(SIGNAL_ARGS)
3902 : {
3903 0 : }
3904 :
3905 : /*
3906 : * Count up number of child processes of specified types.
3907 : */
3908 : static int
3909 11478 : CountChildren(BackendTypeMask targetMask)
3910 : {
3911 : dlist_iter iter;
3912 11478 : int cnt = 0;
3913 :
3914 62780 : dlist_foreach(iter, &ActiveChildList)
3915 : {
3916 51302 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3917 :
3918 : /*
3919 : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3920 : * if any B_BACKEND backends have recently announced that they are
3921 : * actually WAL senders.
3922 : */
3923 51302 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3924 33908 : bp->bkend_type == B_BACKEND)
3925 : {
3926 1010 : if (IsPostmasterChildWalSender(bp->child_slot))
3927 0 : bp->bkend_type = B_WAL_SENDER;
3928 : }
3929 :
3930 51302 : if (!btmask_contains(targetMask, bp->bkend_type))
3931 26922 : continue;
3932 :
3933 24380 : ereport(DEBUG4,
3934 : (errmsg_internal("%s process %d is still running",
3935 : GetBackendTypeDesc(bp->bkend_type), (int) bp->pid)));
3936 :
3937 24380 : cnt++;
3938 : }
3939 11478 : return cnt;
3940 : }
3941 :
3942 :
3943 : /*
3944 : * StartChildProcess -- start an auxiliary process for the postmaster
3945 : *
3946 : * "type" determines what kind of child will be started. All child types
3947 : * initially go to AuxiliaryProcessMain, which will handle common setup.
3948 : *
3949 : * Return value of StartChildProcess is subprocess' PMChild entry, or NULL on
3950 : * failure.
3951 : */
3952 : static PMChild *
3953 16032 : StartChildProcess(BackendType type)
3954 : {
3955 : PMChild *pmchild;
3956 : pid_t pid;
3957 :
3958 16032 : pmchild = AssignPostmasterChildSlot(type);
3959 16032 : if (!pmchild)
3960 : {
3961 0 : if (type == B_AUTOVAC_WORKER)
3962 0 : ereport(LOG,
3963 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
3964 : errmsg("no slot available for new autovacuum worker process")));
3965 : else
3966 : {
3967 : /* shouldn't happen because we allocate enough slots */
3968 0 : elog(LOG, "no postmaster child slot available for aux process");
3969 : }
3970 0 : return NULL;
3971 : }
3972 :
3973 16032 : pid = postmaster_child_launch(type, pmchild->child_slot, NULL, 0, NULL);
3974 16032 : if (pid < 0)
3975 : {
3976 : /* in parent, fork failed */
3977 0 : ReleasePostmasterChildSlot(pmchild);
3978 0 : ereport(LOG,
3979 : (errmsg("could not fork \"%s\" process: %m", PostmasterChildName(type))));
3980 :
3981 : /*
3982 : * fork failure is fatal during startup, but there's no need to choke
3983 : * immediately if starting other child types fails.
3984 : */
3985 0 : if (type == B_STARTUP)
3986 0 : ExitPostmaster(1);
3987 0 : return NULL;
3988 : }
3989 :
3990 : /* in parent, successful fork */
3991 16032 : pmchild->pid = pid;
3992 16032 : return pmchild;
3993 : }
3994 :
3995 : /*
3996 : * StartSysLogger -- start the syslogger process
3997 : */
3998 : void
3999 2 : StartSysLogger(void)
4000 : {
4001 : Assert(SysLoggerPMChild == NULL);
4002 :
4003 2 : SysLoggerPMChild = AssignPostmasterChildSlot(B_LOGGER);
4004 2 : if (!SysLoggerPMChild)
4005 0 : elog(PANIC, "no postmaster child slot available for syslogger");
4006 2 : SysLoggerPMChild->pid = SysLogger_Start(SysLoggerPMChild->child_slot);
4007 2 : if (SysLoggerPMChild->pid == 0)
4008 : {
4009 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
4010 0 : SysLoggerPMChild = NULL;
4011 : }
4012 2 : }
4013 :
4014 : /*
4015 : * StartAutovacuumWorker
4016 : * Start an autovac worker process.
4017 : *
4018 : * This function is here because it enters the resulting PID into the
4019 : * postmaster's private backends list.
4020 : *
4021 : * NB -- this code very roughly matches BackendStartup.
4022 : */
4023 : static void
4024 2424 : StartAutovacuumWorker(void)
4025 : {
4026 : PMChild *bn;
4027 :
4028 : /*
4029 : * If not in condition to run a process, don't try, but handle it like a
4030 : * fork failure. This does not normally happen, since the signal is only
4031 : * supposed to be sent by autovacuum launcher when it's OK to do it, but
4032 : * we have to check to avoid race-condition problems during DB state
4033 : * changes.
4034 : */
4035 2424 : if (canAcceptConnections(B_AUTOVAC_WORKER) == CAC_OK)
4036 : {
4037 2424 : bn = StartChildProcess(B_AUTOVAC_WORKER);
4038 2424 : if (bn)
4039 : {
4040 2424 : bn->bgworker_notify = false;
4041 2424 : bn->rw = NULL;
4042 2424 : return;
4043 : }
4044 : else
4045 : {
4046 : /*
4047 : * fork failed, fall through to report -- actual error message was
4048 : * logged by StartChildProcess
4049 : */
4050 : }
4051 : }
4052 :
4053 : /*
4054 : * Report the failure to the launcher, if it's running. (If it's not, we
4055 : * might not even be connected to shared memory, so don't try to call
4056 : * AutoVacWorkerFailed.) Note that we also need to signal it so that it
4057 : * responds to the condition, but we don't do that here, instead waiting
4058 : * for ServerLoop to do it. This way we avoid a ping-pong signaling in
4059 : * quick succession between the autovac launcher and postmaster in case
4060 : * things get ugly.
4061 : */
4062 0 : if (AutoVacLauncherPMChild != NULL)
4063 : {
4064 0 : AutoVacWorkerFailed();
4065 0 : avlauncher_needs_signal = true;
4066 : }
4067 : }
4068 :
4069 :
4070 : /*
4071 : * Create the opts file
4072 : */
4073 : static bool
4074 1680 : CreateOptsFile(int argc, char *argv[], char *fullprogname)
4075 : {
4076 : FILE *fp;
4077 : int i;
4078 :
4079 : #define OPTS_FILE "postmaster.opts"
4080 :
4081 1680 : if ((fp = fopen(OPTS_FILE, "w")) == NULL)
4082 : {
4083 0 : ereport(LOG,
4084 : (errcode_for_file_access(),
4085 : errmsg("could not create file \"%s\": %m", OPTS_FILE)));
4086 0 : return false;
4087 : }
4088 :
4089 1680 : fprintf(fp, "%s", fullprogname);
4090 9114 : for (i = 1; i < argc; i++)
4091 7434 : fprintf(fp, " \"%s\"", argv[i]);
4092 1680 : fputs("\n", fp);
4093 :
4094 1680 : if (fclose(fp))
4095 : {
4096 0 : ereport(LOG,
4097 : (errcode_for_file_access(),
4098 : errmsg("could not write file \"%s\": %m", OPTS_FILE)));
4099 0 : return false;
4100 : }
4101 :
4102 1680 : return true;
4103 : }
4104 :
4105 :
4106 : /*
4107 : * Start a new bgworker.
4108 : * Starting time conditions must have been checked already.
4109 : *
4110 : * Returns true on success, false on failure.
4111 : * In either case, update the RegisteredBgWorker's state appropriately.
4112 : *
4113 : * NB -- this code very roughly matches BackendStartup.
4114 : */
4115 : static bool
4116 5036 : StartBackgroundWorker(RegisteredBgWorker *rw)
4117 : {
4118 : PMChild *bn;
4119 : pid_t worker_pid;
4120 :
4121 : Assert(rw->rw_pid == 0);
4122 :
4123 : /*
4124 : * Allocate and assign the child slot. Note we must do this before
4125 : * forking, so that we can handle failures (out of memory or child-process
4126 : * slots) cleanly.
4127 : *
4128 : * Treat failure as though the worker had crashed. That way, the
4129 : * postmaster will wait a bit before attempting to start it again; if we
4130 : * tried again right away, most likely we'd find ourselves hitting the
4131 : * same resource-exhaustion condition.
4132 : */
4133 5036 : bn = AssignPostmasterChildSlot(B_BG_WORKER);
4134 5036 : if (bn == NULL)
4135 : {
4136 0 : ereport(LOG,
4137 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
4138 : errmsg("no slot available for new background worker process")));
4139 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4140 0 : return false;
4141 : }
4142 5036 : bn->rw = rw;
4143 5036 : bn->bkend_type = B_BG_WORKER;
4144 5036 : bn->bgworker_notify = false;
4145 :
4146 5036 : ereport(DEBUG1,
4147 : (errmsg_internal("starting background worker process \"%s\"",
4148 : rw->rw_worker.bgw_name)));
4149 :
4150 5036 : worker_pid = postmaster_child_launch(B_BG_WORKER, bn->child_slot,
4151 5036 : &rw->rw_worker, sizeof(BackgroundWorker), NULL);
4152 5032 : if (worker_pid == -1)
4153 : {
4154 : /* in postmaster, fork failed ... */
4155 0 : ereport(LOG,
4156 : (errmsg("could not fork background worker process: %m")));
4157 : /* undo what AssignPostmasterChildSlot did */
4158 0 : ReleasePostmasterChildSlot(bn);
4159 :
4160 : /* mark entry as crashed, so we'll try again later */
4161 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4162 0 : return false;
4163 : }
4164 :
4165 : /* in postmaster, fork successful ... */
4166 5032 : rw->rw_pid = worker_pid;
4167 5032 : bn->pid = rw->rw_pid;
4168 5032 : ReportBackgroundWorkerPID(rw);
4169 5032 : return true;
4170 : }
4171 :
4172 : /*
4173 : * Does the current postmaster state require starting a worker with the
4174 : * specified start_time?
4175 : */
4176 : static bool
4177 6918 : bgworker_should_start_now(BgWorkerStartTime start_time)
4178 : {
4179 6918 : switch (pmState)
4180 : {
4181 2 : case PM_NO_CHILDREN:
4182 : case PM_WAIT_CHECKPOINTER:
4183 : case PM_WAIT_DEAD_END:
4184 : case PM_WAIT_XLOG_ARCHIVAL:
4185 : case PM_WAIT_XLOG_SHUTDOWN:
4186 : case PM_WAIT_IO_WORKERS:
4187 : case PM_WAIT_BACKENDS:
4188 : case PM_STOP_BACKENDS:
4189 2 : break;
4190 :
4191 5036 : case PM_RUN:
4192 5036 : if (start_time == BgWorkerStart_RecoveryFinished)
4193 2278 : return true;
4194 : /* fall through */
4195 :
4196 : case PM_HOT_STANDBY:
4197 3048 : if (start_time == BgWorkerStart_ConsistentState)
4198 2758 : return true;
4199 : /* fall through */
4200 :
4201 : case PM_RECOVERY:
4202 : case PM_STARTUP:
4203 : case PM_INIT:
4204 1880 : if (start_time == BgWorkerStart_PostmasterStart)
4205 0 : return true;
4206 : /* fall through */
4207 : }
4208 :
4209 1882 : return false;
4210 : }
4211 :
4212 : /*
4213 : * If the time is right, start background worker(s).
4214 : *
4215 : * As a side effect, the bgworker control variables are set or reset
4216 : * depending on whether more workers may need to be started.
4217 : *
4218 : * We limit the number of workers started per call, to avoid consuming the
4219 : * postmaster's attention for too long when many such requests are pending.
4220 : * As long as StartWorkerNeeded is true, ServerLoop will not block and will
4221 : * call this function again after dealing with any other issues.
4222 : */
4223 : static void
4224 14398 : maybe_start_bgworkers(void)
4225 : {
4226 : #define MAX_BGWORKERS_TO_LAUNCH 100
4227 14398 : int num_launched = 0;
4228 14398 : TimestampTz now = 0;
4229 : dlist_mutable_iter iter;
4230 :
4231 : /*
4232 : * During crash recovery, we have no need to be called until the state
4233 : * transition out of recovery.
4234 : */
4235 14398 : if (FatalError)
4236 : {
4237 0 : StartWorkerNeeded = false;
4238 0 : HaveCrashedWorker = false;
4239 0 : return;
4240 : }
4241 :
4242 : /* Don't need to be called again unless we find a reason for it below */
4243 14398 : StartWorkerNeeded = false;
4244 14398 : HaveCrashedWorker = false;
4245 :
4246 37024 : dlist_foreach_modify(iter, &BackgroundWorkerList)
4247 : {
4248 : RegisteredBgWorker *rw;
4249 :
4250 22630 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
4251 :
4252 : /* ignore if already running */
4253 22630 : if (rw->rw_pid != 0)
4254 10314 : continue;
4255 :
4256 : /* if marked for death, clean up and remove from list */
4257 12316 : if (rw->rw_terminate)
4258 : {
4259 0 : ForgetBackgroundWorker(rw);
4260 0 : continue;
4261 : }
4262 :
4263 : /*
4264 : * If this worker has crashed previously, maybe it needs to be
4265 : * restarted (unless on registration it specified it doesn't want to
4266 : * be restarted at all). Check how long ago did a crash last happen.
4267 : * If the last crash is too recent, don't start it right away; let it
4268 : * be restarted once enough time has passed.
4269 : */
4270 12316 : if (rw->rw_crashed_at != 0)
4271 : {
4272 5400 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
4273 : {
4274 : int notify_pid;
4275 :
4276 0 : notify_pid = rw->rw_worker.bgw_notify_pid;
4277 :
4278 0 : ForgetBackgroundWorker(rw);
4279 :
4280 : /* Report worker is gone now. */
4281 0 : if (notify_pid != 0)
4282 0 : kill(notify_pid, SIGUSR1);
4283 :
4284 0 : continue;
4285 : }
4286 :
4287 : /* read system time only when needed */
4288 5400 : if (now == 0)
4289 5400 : now = GetCurrentTimestamp();
4290 :
4291 5400 : if (!TimestampDifferenceExceeds(rw->rw_crashed_at, now,
4292 5400 : rw->rw_worker.bgw_restart_time * 1000))
4293 : {
4294 : /* Set flag to remember that we have workers to start later */
4295 5398 : HaveCrashedWorker = true;
4296 5398 : continue;
4297 : }
4298 : }
4299 :
4300 6918 : if (bgworker_should_start_now(rw->rw_worker.bgw_start_time))
4301 : {
4302 : /* reset crash time before trying to start worker */
4303 5036 : rw->rw_crashed_at = 0;
4304 :
4305 : /*
4306 : * Try to start the worker.
4307 : *
4308 : * On failure, give up processing workers for now, but set
4309 : * StartWorkerNeeded so we'll come back here on the next iteration
4310 : * of ServerLoop to try again. (We don't want to wait, because
4311 : * there might be additional ready-to-run workers.) We could set
4312 : * HaveCrashedWorker as well, since this worker is now marked
4313 : * crashed, but there's no need because the next run of this
4314 : * function will do that.
4315 : */
4316 5036 : if (!StartBackgroundWorker(rw))
4317 : {
4318 0 : StartWorkerNeeded = true;
4319 0 : return;
4320 : }
4321 :
4322 : /*
4323 : * If we've launched as many workers as allowed, quit, but have
4324 : * ServerLoop call us again to look for additional ready-to-run
4325 : * workers. There might not be any, but we'll find out the next
4326 : * time we run.
4327 : */
4328 5032 : if (++num_launched >= MAX_BGWORKERS_TO_LAUNCH)
4329 : {
4330 0 : StartWorkerNeeded = true;
4331 0 : return;
4332 : }
4333 : }
4334 : }
4335 : }
4336 :
4337 : static bool
4338 40796 : maybe_reap_io_worker(int pid)
4339 : {
4340 1188680 : for (int id = 0; id < MAX_IO_WORKERS; ++id)
4341 : {
4342 1153018 : if (io_worker_children[id] &&
4343 114282 : io_worker_children[id]->pid == pid)
4344 : {
4345 5134 : ReleasePostmasterChildSlot(io_worker_children[id]);
4346 :
4347 5134 : --io_worker_count;
4348 5134 : io_worker_children[id] = NULL;
4349 5134 : return true;
4350 : }
4351 : }
4352 35662 : return false;
4353 : }
4354 :
4355 : /*
4356 : * Start or stop IO workers, to close the gap between the number of running
4357 : * workers and the number of configured workers. Used to respond to change of
4358 : * the io_workers GUC (by increasing and decreasing the number of workers), as
4359 : * well as workers terminating in response to errors (by starting
4360 : * "replacement" workers).
4361 : */
4362 : static void
4363 251734 : maybe_adjust_io_workers(void)
4364 : {
4365 251734 : if (!pgaio_workers_enabled())
4366 106 : return;
4367 :
4368 : /*
4369 : * If we're in final shutting down state, then we're just waiting for all
4370 : * processes to exit.
4371 : */
4372 251628 : if (pmState >= PM_WAIT_IO_WORKERS)
4373 7056 : return;
4374 :
4375 : /* Don't start new workers during an immediate shutdown either. */
4376 244572 : if (Shutdown >= ImmediateShutdown)
4377 4620 : return;
4378 :
4379 : /*
4380 : * Don't start new workers if we're in the shutdown phase of a crash
4381 : * restart. But we *do* need to start if we're already starting up again.
4382 : */
4383 239952 : if (FatalError && pmState >= PM_STOP_BACKENDS)
4384 110 : return;
4385 :
4386 : Assert(pmState < PM_WAIT_IO_WORKERS);
4387 :
4388 : /* Not enough running? */
4389 245000 : while (io_worker_count < io_workers)
4390 : {
4391 : PMChild *child;
4392 : int id;
4393 :
4394 : /* find unused entry in io_worker_children array */
4395 11882 : for (id = 0; id < MAX_IO_WORKERS; ++id)
4396 : {
4397 11882 : if (io_worker_children[id] == NULL)
4398 5158 : break;
4399 : }
4400 5158 : if (id == MAX_IO_WORKERS)
4401 0 : elog(ERROR, "could not find a free IO worker ID");
4402 :
4403 : /* Try to launch one. */
4404 5158 : child = StartChildProcess(B_IO_WORKER);
4405 5158 : if (child != NULL)
4406 : {
4407 5158 : io_worker_children[id] = child;
4408 5158 : ++io_worker_count;
4409 : }
4410 : else
4411 0 : break; /* try again next time */
4412 : }
4413 :
4414 : /* Too many running? */
4415 239842 : if (io_worker_count > io_workers)
4416 : {
4417 : /* ask the IO worker in the highest slot to exit */
4418 554 : for (int id = MAX_IO_WORKERS - 1; id >= 0; --id)
4419 : {
4420 554 : if (io_worker_children[id] != NULL)
4421 : {
4422 60 : kill(io_worker_children[id]->pid, SIGUSR2);
4423 60 : break;
4424 : }
4425 : }
4426 : }
4427 : }
4428 :
4429 :
4430 : /*
4431 : * When a backend asks to be notified about worker state changes, we
4432 : * set a flag in its backend entry. The background worker machinery needs
4433 : * to know when such backends exit.
4434 : */
4435 : bool
4436 3646 : PostmasterMarkPIDForWorkerNotify(int pid)
4437 : {
4438 : dlist_iter iter;
4439 : PMChild *bp;
4440 :
4441 7618 : dlist_foreach(iter, &ActiveChildList)
4442 : {
4443 7618 : bp = dlist_container(PMChild, elem, iter.cur);
4444 7618 : if (bp->pid == pid)
4445 : {
4446 3646 : bp->bgworker_notify = true;
4447 3646 : return true;
4448 : }
4449 : }
4450 0 : return false;
4451 : }
4452 :
4453 : #ifdef WIN32
4454 :
4455 : /*
4456 : * Subset implementation of waitpid() for Windows. We assume pid is -1
4457 : * (that is, check all child processes) and options is WNOHANG (don't wait).
4458 : */
4459 : static pid_t
4460 : waitpid(pid_t pid, int *exitstatus, int options)
4461 : {
4462 : win32_deadchild_waitinfo *childinfo;
4463 : DWORD exitcode;
4464 : DWORD dwd;
4465 : ULONG_PTR key;
4466 : OVERLAPPED *ovl;
4467 :
4468 : /* Try to consume one win32_deadchild_waitinfo from the queue. */
4469 : if (!GetQueuedCompletionStatus(win32ChildQueue, &dwd, &key, &ovl, 0))
4470 : {
4471 : errno = EAGAIN;
4472 : return -1;
4473 : }
4474 :
4475 : childinfo = (win32_deadchild_waitinfo *) key;
4476 : pid = childinfo->procId;
4477 :
4478 : /*
4479 : * Remove handle from wait - required even though it's set to wait only
4480 : * once
4481 : */
4482 : UnregisterWaitEx(childinfo->waitHandle, NULL);
4483 :
4484 : if (!GetExitCodeProcess(childinfo->procHandle, &exitcode))
4485 : {
4486 : /*
4487 : * Should never happen. Inform user and set a fixed exitcode.
4488 : */
4489 : write_stderr("could not read exit code for process\n");
4490 : exitcode = 255;
4491 : }
4492 : *exitstatus = exitcode;
4493 :
4494 : /*
4495 : * Close the process handle. Only after this point can the PID can be
4496 : * recycled by the kernel.
4497 : */
4498 : CloseHandle(childinfo->procHandle);
4499 :
4500 : /*
4501 : * Free struct that was allocated before the call to
4502 : * RegisterWaitForSingleObject()
4503 : */
4504 : pfree(childinfo);
4505 :
4506 : return pid;
4507 : }
4508 :
4509 : /*
4510 : * Note! Code below executes on a thread pool! All operations must
4511 : * be thread safe! Note that elog() and friends must *not* be used.
4512 : */
4513 : static void WINAPI
4514 : pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
4515 : {
4516 : /* Should never happen, since we use INFINITE as timeout value. */
4517 : if (TimerOrWaitFired)
4518 : return;
4519 :
4520 : /*
4521 : * Post the win32_deadchild_waitinfo object for waitpid() to deal with. If
4522 : * that fails, we leak the object, but we also leak a whole process and
4523 : * get into an unrecoverable state, so there's not much point in worrying
4524 : * about that. We'd like to panic, but we can't use that infrastructure
4525 : * from this thread.
4526 : */
4527 : if (!PostQueuedCompletionStatus(win32ChildQueue,
4528 : 0,
4529 : (ULONG_PTR) lpParameter,
4530 : NULL))
4531 : write_stderr("could not post child completion status\n");
4532 :
4533 : /* Queue SIGCHLD signal. */
4534 : pg_queue_signal(SIGCHLD);
4535 : }
4536 :
4537 : /*
4538 : * Queue a waiter to signal when this child dies. The wait will be handled
4539 : * automatically by an operating system thread pool. The memory and the
4540 : * process handle will be freed by a later call to waitpid().
4541 : */
4542 : void
4543 : pgwin32_register_deadchild_callback(HANDLE procHandle, DWORD procId)
4544 : {
4545 : win32_deadchild_waitinfo *childinfo;
4546 :
4547 : childinfo = palloc(sizeof(win32_deadchild_waitinfo));
4548 : childinfo->procHandle = procHandle;
4549 : childinfo->procId = procId;
4550 :
4551 : if (!RegisterWaitForSingleObject(&childinfo->waitHandle,
4552 : procHandle,
4553 : pgwin32_deadchild_callback,
4554 : childinfo,
4555 : INFINITE,
4556 : WT_EXECUTEONLYONCE | WT_EXECUTEINWAITTHREAD))
4557 : ereport(FATAL,
4558 : (errmsg_internal("could not register process for wait: error code %lu",
4559 : GetLastError())));
4560 : }
4561 :
4562 : #endif /* WIN32 */
4563 :
4564 : /*
4565 : * Initialize one and only handle for monitoring postmaster death.
4566 : *
4567 : * Called once in the postmaster, so that child processes can subsequently
4568 : * monitor if their parent is dead.
4569 : */
4570 : static void
4571 1680 : InitPostmasterDeathWatchHandle(void)
4572 : {
4573 : #ifndef WIN32
4574 :
4575 : /*
4576 : * Create a pipe. Postmaster holds the write end of the pipe open
4577 : * (POSTMASTER_FD_OWN), and children hold the read end. Children can pass
4578 : * the read file descriptor to select() to wake up in case postmaster
4579 : * dies, or check for postmaster death with a (read() == 0). Children must
4580 : * close the write end as soon as possible after forking, because EOF
4581 : * won't be signaled in the read end until all processes have closed the
4582 : * write fd. That is taken care of in ClosePostmasterPorts().
4583 : */
4584 : Assert(MyProcPid == PostmasterPid);
4585 1680 : if (pipe(postmaster_alive_fds) < 0)
4586 0 : ereport(FATAL,
4587 : (errcode_for_file_access(),
4588 : errmsg_internal("could not create pipe to monitor postmaster death: %m")));
4589 :
4590 : /* Notify fd.c that we've eaten two FDs for the pipe. */
4591 1680 : ReserveExternalFD();
4592 1680 : ReserveExternalFD();
4593 :
4594 : /*
4595 : * Set O_NONBLOCK to allow testing for the fd's presence with a read()
4596 : * call.
4597 : */
4598 1680 : if (fcntl(postmaster_alive_fds[POSTMASTER_FD_WATCH], F_SETFL, O_NONBLOCK) == -1)
4599 0 : ereport(FATAL,
4600 : (errcode_for_socket_access(),
4601 : errmsg_internal("could not set postmaster death monitoring pipe to nonblocking mode: %m")));
4602 : #else
4603 :
4604 : /*
4605 : * On Windows, we use a process handle for the same purpose.
4606 : */
4607 : if (DuplicateHandle(GetCurrentProcess(),
4608 : GetCurrentProcess(),
4609 : GetCurrentProcess(),
4610 : &PostmasterHandle,
4611 : 0,
4612 : TRUE,
4613 : DUPLICATE_SAME_ACCESS) == 0)
4614 : ereport(FATAL,
4615 : (errmsg_internal("could not duplicate postmaster handle: error code %lu",
4616 : GetLastError())));
4617 : #endif /* WIN32 */
4618 1680 : }
|
