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) 1996-2026, 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 3184 : btmask(BackendType t)
150 : {
151 3184 : BackendTypeMask mask = {.mask = 1 << t};
152 :
153 3184 : return mask;
154 : }
155 :
156 : static inline BackendTypeMask
157 32512 : btmask_add_n(BackendTypeMask mask, int nargs, BackendType *t)
158 : {
159 143012 : for (int i = 0; i < nargs; i++)
160 110500 : mask.mask |= 1 << t[i];
161 32512 : 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 8986 : btmask_del(BackendTypeMask mask, BackendType t)
172 : {
173 8986 : mask.mask &= ~(1 << t);
174 8986 : return mask;
175 : }
176 :
177 : static inline BackendTypeMask
178 5242 : btmask_all_except_n(int nargs, BackendType *t)
179 : {
180 5242 : BackendTypeMask mask = BTYPE_MASK_ALL;
181 :
182 14228 : for (int i = 0; i < nargs; i++)
183 8986 : mask = btmask_del(mask, t[i]);
184 5242 : 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 282264 : btmask_contains(BackendTypeMask mask, BackendType t)
195 : {
196 282264 : 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 1852 : PostmasterMain(int argc, char *argv[])
494 : {
495 : int opt;
496 : int status;
497 1852 : char *userDoption = NULL;
498 1852 : bool listen_addr_saved = false;
499 1852 : char *output_config_variable = NULL;
500 :
501 1852 : InitProcessGlobals();
502 :
503 1852 : PostmasterPid = MyProcPid;
504 :
505 1852 : 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 1852 : 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 1852 : PostmasterContext = AllocSetContextCreate(TopMemoryContext,
531 : "Postmaster",
532 : ALLOCSET_DEFAULT_SIZES);
533 1852 : MemoryContextSwitchTo(PostmasterContext);
534 :
535 : /* Initialize paths to installation files */
536 1852 : 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 1852 : pqinitmask();
548 1852 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
549 :
550 1852 : pqsignal(SIGHUP, handle_pm_reload_request_signal);
551 1852 : pqsignal(SIGINT, handle_pm_shutdown_request_signal);
552 1852 : pqsignal(SIGQUIT, handle_pm_shutdown_request_signal);
553 1852 : pqsignal(SIGTERM, handle_pm_shutdown_request_signal);
554 1852 : pqsignal(SIGALRM, SIG_IGN); /* ignored */
555 1852 : pqsignal(SIGPIPE, SIG_IGN); /* ignored */
556 1852 : pqsignal(SIGUSR1, handle_pm_pmsignal_signal);
557 1852 : pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
558 1852 : pqsignal(SIGCHLD, handle_pm_child_exit_signal);
559 :
560 : /* This may configure SIGURG, depending on platform. */
561 1852 : InitializeWaitEventSupport();
562 1852 : 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 1852 : pqsignal(SIGTTIN, SIG_IGN); /* ignored */
573 : #endif
574 : #ifdef SIGTTOU
575 1852 : pqsignal(SIGTTOU, SIG_IGN); /* ignored */
576 : #endif
577 :
578 : /* ignore SIGXFSZ, so that ulimit violations work like disk full */
579 : #ifdef SIGXFSZ
580 1852 : pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
581 : #endif
582 :
583 : /* Begin accepting signals. */
584 1852 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
585 :
586 : /*
587 : * Options setup
588 : */
589 1852 : InitializeGUCOptions();
590 :
591 1852 : 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 6652 : while ((opt = getopt(argc, argv, "B:bC:c:D:d:EeFf:h:ijk:lN:OPp:r:S:sTt:W:-:")) != -1)
599 : {
600 4804 : switch (opt)
601 : {
602 0 : case 'B':
603 0 : SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
604 0 : break;
605 :
606 102 : case 'b':
607 : /* Undocumented flag used for binary upgrades */
608 102 : IsBinaryUpgrade = true;
609 102 : break;
610 :
611 6 : case 'C':
612 6 : output_config_variable = strdup(optarg);
613 6 : break;
614 :
615 1490 : 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 1490 : 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 2330 : ParseLongOption(optarg, &name, &value);
635 2330 : 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 2328 : SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
650 2326 : pfree(name);
651 2326 : pfree(value);
652 2326 : break;
653 : }
654 :
655 1848 : case 'D':
656 1848 : userDoption = strdup(optarg);
657 1848 : break;
658 :
659 0 : case 'd':
660 0 : set_debug_options(atoi(optarg), PGC_POSTMASTER, PGC_S_ARGV);
661 0 : break;
662 :
663 0 : case 'E':
664 0 : SetConfigOption("log_statement", "all", PGC_POSTMASTER, PGC_S_ARGV);
665 0 : break;
666 :
667 0 : case 'e':
668 0 : SetConfigOption("datestyle", "euro", PGC_POSTMASTER, PGC_S_ARGV);
669 0 : break;
670 :
671 196 : case 'F':
672 196 : SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
673 196 : break;
674 :
675 0 : case 'f':
676 0 : if (!set_plan_disabling_options(optarg, PGC_POSTMASTER, PGC_S_ARGV))
677 : {
678 0 : write_stderr("%s: invalid argument for option -f: \"%s\"\n",
679 : progname, optarg);
680 0 : ExitPostmaster(1);
681 : }
682 0 : break;
683 :
684 0 : case 'h':
685 0 : SetConfigOption("listen_addresses", optarg, PGC_POSTMASTER, PGC_S_ARGV);
686 0 : break;
687 :
688 0 : case 'i':
689 0 : SetConfigOption("listen_addresses", "*", PGC_POSTMASTER, PGC_S_ARGV);
690 0 : break;
691 :
692 0 : case 'j':
693 : /* only used by interactive backend */
694 0 : break;
695 :
696 196 : case 'k':
697 196 : SetConfigOption("unix_socket_directories", optarg, PGC_POSTMASTER, PGC_S_ARGV);
698 196 : break;
699 :
700 0 : case 'l':
701 0 : SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
702 0 : break;
703 :
704 0 : case 'N':
705 0 : SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
706 0 : break;
707 :
708 0 : case 'O':
709 0 : SetConfigOption("allow_system_table_mods", "true", PGC_POSTMASTER, PGC_S_ARGV);
710 0 : break;
711 :
712 0 : case 'P':
713 0 : SetConfigOption("ignore_system_indexes", "true", PGC_POSTMASTER, PGC_S_ARGV);
714 0 : break;
715 :
716 126 : case 'p':
717 126 : SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
718 126 : 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 1848 : 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 1848 : if (!SelectConfigFiles(userDoption, progname))
786 0 : ExitPostmaster(2);
787 :
788 1840 : 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 1838 : checkDataDir();
830 :
831 : /* Check that pg_control exists */
832 1838 : checkControlFile();
833 :
834 : /* And switch working directory into it */
835 1838 : ChangeToDataDir();
836 :
837 : /*
838 : * Check for invalid combinations of GUC settings.
839 : */
840 1838 : 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 1838 : 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 1838 : 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 1838 : if (summarize_wal && wal_level == WAL_LEVEL_MINIMAL)
855 0 : ereport(ERROR,
856 : (errmsg("WAL cannot be summarized when \"wal_level\" is \"minimal\"")));
857 1838 : if (sync_replication_slots && wal_level == WAL_LEVEL_MINIMAL)
858 0 : ereport(ERROR,
859 : (errmsg("replication slot synchronization (\"sync_replication_slots\" = on) requires \"wal_level\" to be \"replica\" or \"logical\"")));
860 :
861 : /*
862 : * Other one-time internal sanity checks can go here, if they are fast.
863 : * (Put any slow processing further down, after postmaster.pid creation.)
864 : */
865 1838 : if (!CheckDateTokenTables())
866 : {
867 0 : write_stderr("%s: invalid datetoken tables, please fix\n", progname);
868 0 : ExitPostmaster(1);
869 : }
870 :
871 : /*
872 : * Now that we are done processing the postmaster arguments, reset
873 : * getopt(3) library so that it will work correctly in subprocesses.
874 : */
875 1838 : optind = 1;
876 : #ifdef HAVE_INT_OPTRESET
877 : optreset = 1; /* some systems need this too */
878 : #endif
879 :
880 : /* For debugging: display postmaster environment */
881 1838 : if (message_level_is_interesting(DEBUG3))
882 : {
883 : #if !defined(WIN32)
884 : extern char **environ;
885 : #endif
886 : char **p;
887 : StringInfoData si;
888 :
889 4 : initStringInfo(&si);
890 :
891 4 : appendStringInfoString(&si, "initial environment dump:");
892 172 : for (p = environ; *p; ++p)
893 168 : appendStringInfo(&si, "\n%s", *p);
894 :
895 4 : ereport(DEBUG3, errmsg_internal("%s", si.data));
896 4 : pfree(si.data);
897 : }
898 :
899 : /*
900 : * Create lockfile for data directory.
901 : *
902 : * We want to do this before we try to grab the input sockets, because the
903 : * data directory interlock is more reliable than the socket-file
904 : * interlock (thanks to whoever decided to put socket files in /tmp :-().
905 : * For the same reason, it's best to grab the TCP socket(s) before the
906 : * Unix socket(s).
907 : *
908 : * Also note that this internally sets up the on_proc_exit function that
909 : * is responsible for removing both data directory and socket lockfiles;
910 : * so it must happen before opening sockets so that at exit, the socket
911 : * lockfiles go away after CloseServerPorts runs.
912 : */
913 1838 : CreateDataDirLockFile(true);
914 :
915 : /*
916 : * Read the control file (for error checking and config info).
917 : *
918 : * Since we verify the control file's CRC, this has a useful side effect
919 : * on machines where we need a run-time test for CRC support instructions.
920 : * The postmaster will do the test once at startup, and then its child
921 : * processes will inherit the correct function pointer and not need to
922 : * repeat the test.
923 : */
924 1834 : LocalProcessControlFile(false);
925 :
926 : /*
927 : * Register the apply launcher. It's probably a good idea to call this
928 : * before any modules had a chance to take the background worker slots.
929 : */
930 1834 : ApplyLauncherRegister();
931 :
932 : /*
933 : * process any libraries that should be preloaded at postmaster start
934 : */
935 1834 : process_shared_preload_libraries();
936 :
937 : /*
938 : * Initialize SSL library, if specified.
939 : */
940 : #ifdef USE_SSL
941 1834 : if (EnableSSL)
942 : {
943 64 : (void) secure_initialize(true);
944 54 : LoadedSSL = true;
945 : }
946 : #endif
947 :
948 : /*
949 : * Now that loadable modules have had their chance to alter any GUCs,
950 : * calculate MaxBackends and initialize the machinery to track child
951 : * processes.
952 : */
953 1824 : InitializeMaxBackends();
954 1824 : InitPostmasterChildSlots();
955 :
956 : /*
957 : * Calculate the size of the PGPROC fast-path lock arrays.
958 : */
959 1824 : InitializeFastPathLocks();
960 :
961 : /*
962 : * Give preloaded libraries a chance to request additional shared memory.
963 : */
964 1824 : process_shmem_requests();
965 :
966 : /*
967 : * Now that loadable modules have had their chance to request additional
968 : * shared memory, determine the value of any runtime-computed GUCs that
969 : * depend on the amount of shared memory required.
970 : */
971 1824 : InitializeShmemGUCs();
972 :
973 : /*
974 : * Now that modules have been loaded, we can process any custom resource
975 : * managers specified in the wal_consistency_checking GUC.
976 : */
977 1824 : InitializeWalConsistencyChecking();
978 :
979 : /*
980 : * If -C was specified with a runtime-computed GUC, we held off printing
981 : * the value earlier, as the GUC was not yet initialized. We handle -C
982 : * for most GUCs before we lock the data directory so that the option may
983 : * be used on a running server. However, a handful of GUCs are runtime-
984 : * computed and do not have meaningful values until after locking the data
985 : * directory, and we cannot safely calculate their values earlier on a
986 : * running server. At this point, such GUCs should be properly
987 : * initialized, and we haven't yet set up shared memory, so this is a good
988 : * time to handle the -C option for these special GUCs.
989 : */
990 1824 : if (output_config_variable != NULL)
991 : {
992 2 : const char *config_val = GetConfigOption(output_config_variable,
993 : false, false);
994 :
995 2 : puts(config_val ? config_val : "");
996 2 : ExitPostmaster(0);
997 : }
998 :
999 : /*
1000 : * Set up shared memory and semaphores.
1001 : *
1002 : * Note: if using SysV shmem and/or semas, each postmaster startup will
1003 : * normally choose the same IPC keys. This helps ensure that we will
1004 : * clean up dead IPC objects if the postmaster crashes and is restarted.
1005 : */
1006 1822 : CreateSharedMemoryAndSemaphores();
1007 :
1008 : /*
1009 : * Estimate number of openable files. This must happen after setting up
1010 : * semaphores, because on some platforms semaphores count as open files.
1011 : */
1012 1820 : set_max_safe_fds();
1013 :
1014 : /*
1015 : * Initialize pipe (or process handle on Windows) that allows children to
1016 : * wake up from sleep on postmaster death.
1017 : */
1018 1820 : InitPostmasterDeathWatchHandle();
1019 :
1020 : #ifdef WIN32
1021 :
1022 : /*
1023 : * Initialize I/O completion port used to deliver list of dead children.
1024 : */
1025 : win32ChildQueue = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 1);
1026 : if (win32ChildQueue == NULL)
1027 : ereport(FATAL,
1028 : (errmsg("could not create I/O completion port for child queue")));
1029 : #endif
1030 :
1031 : #ifdef EXEC_BACKEND
1032 : /* Write out nondefault GUC settings for child processes to use */
1033 : write_nondefault_variables(PGC_POSTMASTER);
1034 :
1035 : /*
1036 : * Clean out the temp directory used to transmit parameters to child
1037 : * processes (see internal_forkexec). We must do this before launching
1038 : * any child processes, else we have a race condition: we could remove a
1039 : * parameter file before the child can read it. It should be safe to do
1040 : * so now, because we verified earlier that there are no conflicting
1041 : * Postgres processes in this data directory.
1042 : */
1043 : RemovePgTempFilesInDir(PG_TEMP_FILES_DIR, true, false);
1044 : #endif
1045 :
1046 : /*
1047 : * Forcibly remove the files signaling a standby promotion request.
1048 : * Otherwise, the existence of those files triggers a promotion too early,
1049 : * whether a user wants that or not.
1050 : *
1051 : * This removal of files is usually unnecessary because they can exist
1052 : * only during a few moments during a standby promotion. However there is
1053 : * a race condition: if pg_ctl promote is executed and creates the files
1054 : * during a promotion, the files can stay around even after the server is
1055 : * brought up to be the primary. Then, if a new standby starts by using
1056 : * the backup taken from the new primary, the files can exist at server
1057 : * startup and must be removed in order to avoid an unexpected promotion.
1058 : *
1059 : * Note that promotion signal files need to be removed before the startup
1060 : * process is invoked. Because, after that, they can be used by
1061 : * postmaster's SIGUSR1 signal handler.
1062 : */
1063 1820 : RemovePromoteSignalFiles();
1064 :
1065 : /* Do the same for logrotate signal file */
1066 1820 : RemoveLogrotateSignalFiles();
1067 :
1068 : /* Remove any outdated file holding the current log filenames. */
1069 1820 : if (unlink(LOG_METAINFO_DATAFILE) < 0 && errno != ENOENT)
1070 0 : ereport(LOG,
1071 : (errcode_for_file_access(),
1072 : errmsg("could not remove file \"%s\": %m",
1073 : LOG_METAINFO_DATAFILE)));
1074 :
1075 : /*
1076 : * If enabled, start up syslogger collection subprocess
1077 : */
1078 1820 : if (Logging_collector)
1079 2 : StartSysLogger();
1080 :
1081 : /*
1082 : * Reset whereToSendOutput from DestDebug (its starting state) to
1083 : * DestNone. This stops ereport from sending log messages to stderr unless
1084 : * Log_destination permits. We don't do this until the postmaster is
1085 : * fully launched, since startup failures may as well be reported to
1086 : * stderr.
1087 : *
1088 : * If we are in fact disabling logging to stderr, first emit a log message
1089 : * saying so, to provide a breadcrumb trail for users who may not remember
1090 : * that their logging is configured to go somewhere else.
1091 : */
1092 1820 : if (!(Log_destination & LOG_DESTINATION_STDERR))
1093 0 : ereport(LOG,
1094 : (errmsg("ending log output to stderr"),
1095 : errhint("Future log output will go to log destination \"%s\".",
1096 : Log_destination_string)));
1097 :
1098 1820 : whereToSendOutput = DestNone;
1099 :
1100 : /*
1101 : * Report server startup in log. While we could emit this much earlier,
1102 : * it seems best to do so after starting the log collector, if we intend
1103 : * to use one.
1104 : */
1105 1820 : ereport(LOG,
1106 : (errmsg("starting %s", PG_VERSION_STR)));
1107 :
1108 : /*
1109 : * Establish input sockets.
1110 : *
1111 : * First set up an on_proc_exit function that's charged with closing the
1112 : * sockets again at postmaster shutdown.
1113 : */
1114 1820 : ListenSockets = palloc(MAXLISTEN * sizeof(pgsocket));
1115 1820 : on_proc_exit(CloseServerPorts, 0);
1116 :
1117 1820 : if (ListenAddresses)
1118 : {
1119 : char *rawstring;
1120 : List *elemlist;
1121 : ListCell *l;
1122 1820 : int success = 0;
1123 :
1124 : /* Need a modifiable copy of ListenAddresses */
1125 1820 : rawstring = pstrdup(ListenAddresses);
1126 :
1127 : /* Parse string into list of hostnames */
1128 1820 : if (!SplitGUCList(rawstring, ',', &elemlist))
1129 : {
1130 : /* syntax error in list */
1131 0 : ereport(FATAL,
1132 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1133 : errmsg("invalid list syntax in parameter \"%s\"",
1134 : "listen_addresses")));
1135 : }
1136 :
1137 1880 : foreach(l, elemlist)
1138 : {
1139 60 : char *curhost = (char *) lfirst(l);
1140 :
1141 60 : if (strcmp(curhost, "*") == 0)
1142 0 : status = ListenServerPort(AF_UNSPEC, NULL,
1143 0 : (unsigned short) PostPortNumber,
1144 : NULL,
1145 : ListenSockets,
1146 : &NumListenSockets,
1147 : MAXLISTEN);
1148 : else
1149 60 : status = ListenServerPort(AF_UNSPEC, curhost,
1150 60 : (unsigned short) PostPortNumber,
1151 : NULL,
1152 : ListenSockets,
1153 : &NumListenSockets,
1154 : MAXLISTEN);
1155 :
1156 60 : if (status == STATUS_OK)
1157 : {
1158 60 : success++;
1159 : /* record the first successful host addr in lockfile */
1160 60 : if (!listen_addr_saved)
1161 : {
1162 60 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, curhost);
1163 60 : listen_addr_saved = true;
1164 : }
1165 : }
1166 : else
1167 0 : ereport(WARNING,
1168 : (errmsg("could not create listen socket for \"%s\"",
1169 : curhost)));
1170 : }
1171 :
1172 1820 : if (!success && elemlist != NIL)
1173 0 : ereport(FATAL,
1174 : (errmsg("could not create any TCP/IP sockets")));
1175 :
1176 1820 : list_free(elemlist);
1177 1820 : pfree(rawstring);
1178 : }
1179 :
1180 : #ifdef USE_BONJOUR
1181 : /* Register for Bonjour only if we opened TCP socket(s) */
1182 : if (enable_bonjour && NumListenSockets > 0)
1183 : {
1184 : DNSServiceErrorType err;
1185 :
1186 : /*
1187 : * We pass 0 for interface_index, which will result in registering on
1188 : * all "applicable" interfaces. It's not entirely clear from the
1189 : * DNS-SD docs whether this would be appropriate if we have bound to
1190 : * just a subset of the available network interfaces.
1191 : */
1192 : err = DNSServiceRegister(&bonjour_sdref,
1193 : 0,
1194 : 0,
1195 : bonjour_name,
1196 : "_postgresql._tcp.",
1197 : NULL,
1198 : NULL,
1199 : pg_hton16(PostPortNumber),
1200 : 0,
1201 : NULL,
1202 : NULL,
1203 : NULL);
1204 : if (err != kDNSServiceErr_NoError)
1205 : ereport(LOG,
1206 : (errmsg("DNSServiceRegister() failed: error code %ld",
1207 : (long) err)));
1208 :
1209 : /*
1210 : * We don't bother to read the mDNS daemon's reply, and we expect that
1211 : * it will automatically terminate our registration when the socket is
1212 : * closed at postmaster termination. So there's nothing more to be
1213 : * done here. However, the bonjour_sdref is kept around so that
1214 : * forked children can close their copies of the socket.
1215 : */
1216 : }
1217 : #endif
1218 :
1219 1820 : if (Unix_socket_directories)
1220 : {
1221 : char *rawstring;
1222 : List *elemlist;
1223 : ListCell *l;
1224 1820 : int success = 0;
1225 :
1226 : /* Need a modifiable copy of Unix_socket_directories */
1227 1820 : rawstring = pstrdup(Unix_socket_directories);
1228 :
1229 : /* Parse string into list of directories */
1230 1820 : if (!SplitDirectoriesString(rawstring, ',', &elemlist))
1231 : {
1232 : /* syntax error in list */
1233 0 : ereport(FATAL,
1234 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1235 : errmsg("invalid list syntax in parameter \"%s\"",
1236 : "unix_socket_directories")));
1237 : }
1238 :
1239 3638 : foreach(l, elemlist)
1240 : {
1241 1818 : char *socketdir = (char *) lfirst(l);
1242 :
1243 1818 : status = ListenServerPort(AF_UNIX, NULL,
1244 1818 : (unsigned short) PostPortNumber,
1245 : socketdir,
1246 : ListenSockets,
1247 : &NumListenSockets,
1248 : MAXLISTEN);
1249 :
1250 1818 : if (status == STATUS_OK)
1251 : {
1252 1818 : success++;
1253 : /* record the first successful Unix socket in lockfile */
1254 1818 : if (success == 1)
1255 1818 : AddToDataDirLockFile(LOCK_FILE_LINE_SOCKET_DIR, socketdir);
1256 : }
1257 : else
1258 0 : ereport(WARNING,
1259 : (errmsg("could not create Unix-domain socket in directory \"%s\"",
1260 : socketdir)));
1261 : }
1262 :
1263 1820 : if (!success && elemlist != NIL)
1264 0 : ereport(FATAL,
1265 : (errmsg("could not create any Unix-domain sockets")));
1266 :
1267 1820 : list_free_deep(elemlist);
1268 1820 : pfree(rawstring);
1269 : }
1270 :
1271 : /*
1272 : * check that we have some socket to listen on
1273 : */
1274 1820 : if (NumListenSockets == 0)
1275 0 : ereport(FATAL,
1276 : (errmsg("no socket created for listening")));
1277 :
1278 : /*
1279 : * If no valid TCP ports, write an empty line for listen address,
1280 : * indicating the Unix socket must be used. Note that this line is not
1281 : * added to the lock file until there is a socket backing it.
1282 : */
1283 1820 : if (!listen_addr_saved)
1284 1760 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, "");
1285 :
1286 : /*
1287 : * Record postmaster options. We delay this till now to avoid recording
1288 : * bogus options (eg, unusable port number).
1289 : */
1290 1820 : if (!CreateOptsFile(argc, argv, my_exec_path))
1291 0 : ExitPostmaster(1);
1292 :
1293 : /*
1294 : * Write the external PID file if requested
1295 : */
1296 1820 : if (external_pid_file)
1297 : {
1298 0 : FILE *fpidfile = fopen(external_pid_file, "w");
1299 :
1300 0 : if (fpidfile)
1301 : {
1302 0 : fprintf(fpidfile, "%d\n", MyProcPid);
1303 0 : fclose(fpidfile);
1304 :
1305 : /* Make PID file world readable */
1306 0 : if (chmod(external_pid_file, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH) != 0)
1307 0 : write_stderr("%s: could not change permissions of external PID file \"%s\": %m\n",
1308 : progname, external_pid_file);
1309 : }
1310 : else
1311 0 : write_stderr("%s: could not write external PID file \"%s\": %m\n",
1312 : progname, external_pid_file);
1313 :
1314 0 : on_proc_exit(unlink_external_pid_file, 0);
1315 : }
1316 :
1317 : /*
1318 : * Remove old temporary files. At this point there can be no other
1319 : * Postgres processes running in this directory, so this should be safe.
1320 : */
1321 1820 : RemovePgTempFiles();
1322 :
1323 : /*
1324 : * Initialize the autovacuum subsystem (again, no process start yet)
1325 : */
1326 1820 : autovac_init();
1327 :
1328 : /*
1329 : * Load configuration files for client authentication.
1330 : */
1331 1820 : if (!load_hba())
1332 : {
1333 : /*
1334 : * It makes no sense to continue if we fail to load the HBA file,
1335 : * since there is no way to connect to the database in this case.
1336 : */
1337 0 : ereport(FATAL,
1338 : /* translator: %s is a configuration file */
1339 : (errmsg("could not load %s", HbaFileName)));
1340 : }
1341 1820 : if (!load_ident())
1342 : {
1343 : /*
1344 : * We can start up without the IDENT file, although it means that you
1345 : * cannot log in using any of the authentication methods that need a
1346 : * user name mapping. load_ident() already logged the details of error
1347 : * to the log.
1348 : */
1349 : }
1350 :
1351 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1352 :
1353 : /*
1354 : * On macOS, libintl replaces setlocale() with a version that calls
1355 : * CFLocaleCopyCurrent() when its second argument is "" and every relevant
1356 : * environment variable is unset or empty. CFLocaleCopyCurrent() makes
1357 : * the process multithreaded. The postmaster calls sigprocmask() and
1358 : * calls fork() without an immediate exec(), both of which have undefined
1359 : * behavior in a multithreaded program. A multithreaded postmaster is the
1360 : * normal case on Windows, which offers neither fork() nor sigprocmask().
1361 : * Currently, macOS is the only platform having pthread_is_threaded_np(),
1362 : * so we need not worry whether this HINT is appropriate elsewhere.
1363 : */
1364 : if (pthread_is_threaded_np() != 0)
1365 : ereport(FATAL,
1366 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1367 : errmsg("postmaster became multithreaded during startup"),
1368 : errhint("Set the LC_ALL environment variable to a valid locale.")));
1369 : #endif
1370 :
1371 : /*
1372 : * Remember postmaster startup time
1373 : */
1374 1820 : PgStartTime = GetCurrentTimestamp();
1375 :
1376 : /*
1377 : * Report postmaster status in the postmaster.pid file, to allow pg_ctl to
1378 : * see what's happening.
1379 : */
1380 1820 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STARTING);
1381 :
1382 1820 : UpdatePMState(PM_STARTUP);
1383 :
1384 : /* Make sure we can perform I/O while starting up. */
1385 1820 : maybe_adjust_io_workers();
1386 :
1387 : /* Start bgwriter and checkpointer so they can help with recovery */
1388 1820 : if (CheckpointerPMChild == NULL)
1389 1820 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
1390 1820 : if (BgWriterPMChild == NULL)
1391 1820 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
1392 :
1393 : /*
1394 : * We're ready to rock and roll...
1395 : */
1396 1820 : StartupPMChild = StartChildProcess(B_STARTUP);
1397 : Assert(StartupPMChild != NULL);
1398 1820 : StartupStatus = STARTUP_RUNNING;
1399 :
1400 : /* Some workers may be scheduled to start now */
1401 1820 : maybe_start_bgworkers();
1402 :
1403 1820 : status = ServerLoop();
1404 :
1405 : /*
1406 : * ServerLoop probably shouldn't ever return, but if it does, close down.
1407 : */
1408 0 : ExitPostmaster(status != STATUS_OK);
1409 :
1410 : abort(); /* not reached */
1411 : }
1412 :
1413 :
1414 : /*
1415 : * on_proc_exit callback to close server's listen sockets
1416 : */
1417 : static void
1418 1820 : CloseServerPorts(int status, Datum arg)
1419 : {
1420 : int i;
1421 :
1422 : /*
1423 : * First, explicitly close all the socket FDs. We used to just let this
1424 : * happen implicitly at postmaster exit, but it's better to close them
1425 : * before we remove the postmaster.pid lockfile; otherwise there's a race
1426 : * condition if a new postmaster wants to re-use the TCP port number.
1427 : */
1428 3700 : for (i = 0; i < NumListenSockets; i++)
1429 : {
1430 1880 : if (closesocket(ListenSockets[i]) != 0)
1431 0 : elog(LOG, "could not close listen socket: %m");
1432 : }
1433 1820 : NumListenSockets = 0;
1434 :
1435 : /*
1436 : * Next, remove any filesystem entries for Unix sockets. To avoid race
1437 : * conditions against incoming postmasters, this must happen after closing
1438 : * the sockets and before removing lock files.
1439 : */
1440 1820 : RemoveSocketFiles();
1441 :
1442 : /*
1443 : * We don't do anything about socket lock files here; those will be
1444 : * removed in a later on_proc_exit callback.
1445 : */
1446 1820 : }
1447 :
1448 : /*
1449 : * on_proc_exit callback to delete external_pid_file
1450 : */
1451 : static void
1452 0 : unlink_external_pid_file(int status, Datum arg)
1453 : {
1454 0 : if (external_pid_file)
1455 0 : unlink(external_pid_file);
1456 0 : }
1457 :
1458 :
1459 : /*
1460 : * Compute and check the directory paths to files that are part of the
1461 : * installation (as deduced from the postgres executable's own location)
1462 : */
1463 : static void
1464 1852 : getInstallationPaths(const char *argv0)
1465 : {
1466 : DIR *pdir;
1467 :
1468 : /* Locate the postgres executable itself */
1469 1852 : if (find_my_exec(argv0, my_exec_path) < 0)
1470 0 : ereport(FATAL,
1471 : (errmsg("%s: could not locate my own executable path", argv0)));
1472 :
1473 : #ifdef EXEC_BACKEND
1474 : /* Locate executable backend before we change working directory */
1475 : if (find_other_exec(argv0, "postgres", PG_BACKEND_VERSIONSTR,
1476 : postgres_exec_path) < 0)
1477 : ereport(FATAL,
1478 : (errmsg("%s: could not locate matching postgres executable",
1479 : argv0)));
1480 : #endif
1481 :
1482 : /*
1483 : * Locate the pkglib directory --- this has to be set early in case we try
1484 : * to load any modules from it in response to postgresql.conf entries.
1485 : */
1486 1852 : get_pkglib_path(my_exec_path, pkglib_path);
1487 :
1488 : /*
1489 : * Verify that there's a readable directory there; otherwise the Postgres
1490 : * installation is incomplete or corrupt. (A typical cause of this
1491 : * failure is that the postgres executable has been moved or hardlinked to
1492 : * some directory that's not a sibling of the installation lib/
1493 : * directory.)
1494 : */
1495 1852 : pdir = AllocateDir(pkglib_path);
1496 1852 : if (pdir == NULL)
1497 0 : ereport(ERROR,
1498 : (errcode_for_file_access(),
1499 : errmsg("could not open directory \"%s\": %m",
1500 : pkglib_path),
1501 : errhint("This may indicate an incomplete PostgreSQL installation, or that the file \"%s\" has been moved away from its proper location.",
1502 : my_exec_path)));
1503 1852 : FreeDir(pdir);
1504 :
1505 : /*
1506 : * It's not worth checking the share/ directory. If the lib/ directory is
1507 : * there, then share/ probably is too.
1508 : */
1509 1852 : }
1510 :
1511 : /*
1512 : * Check that pg_control exists in the correct location in the data directory.
1513 : *
1514 : * No attempt is made to validate the contents of pg_control here. This is
1515 : * just a sanity check to see if we are looking at a real data directory.
1516 : */
1517 : static void
1518 1838 : checkControlFile(void)
1519 : {
1520 : char path[MAXPGPATH];
1521 : FILE *fp;
1522 :
1523 1838 : snprintf(path, sizeof(path), "%s/%s", DataDir, XLOG_CONTROL_FILE);
1524 :
1525 1838 : fp = AllocateFile(path, PG_BINARY_R);
1526 1838 : if (fp == NULL)
1527 : {
1528 0 : write_stderr("%s: could not find the database system\n"
1529 : "Expected to find it in the directory \"%s\",\n"
1530 : "but could not open file \"%s\": %m\n",
1531 : progname, DataDir, path);
1532 0 : ExitPostmaster(2);
1533 : }
1534 1838 : FreeFile(fp);
1535 1838 : }
1536 :
1537 : /*
1538 : * Determine how long should we let ServerLoop sleep, in milliseconds.
1539 : *
1540 : * In normal conditions we wait at most one minute, to ensure that the other
1541 : * background tasks handled by ServerLoop get done even when no requests are
1542 : * arriving. However, if there are background workers waiting to be started,
1543 : * we don't actually sleep so that they are quickly serviced. Other exception
1544 : * cases are as shown in the code.
1545 : */
1546 : static int
1547 255860 : DetermineSleepTime(void)
1548 : {
1549 255860 : TimestampTz next_wakeup = 0;
1550 :
1551 : /*
1552 : * Normal case: either there are no background workers at all, or we're in
1553 : * a shutdown sequence (during which we ignore bgworkers altogether).
1554 : */
1555 255860 : if (Shutdown > NoShutdown ||
1556 241866 : (!StartWorkerNeeded && !HaveCrashedWorker))
1557 : {
1558 255860 : if (AbortStartTime != 0)
1559 : {
1560 2502 : time_t curtime = time(NULL);
1561 : int seconds;
1562 :
1563 : /*
1564 : * time left to abort; clamp to 0 if it already expired, or if
1565 : * time goes backwards
1566 : */
1567 2502 : if (curtime < AbortStartTime ||
1568 2502 : curtime - AbortStartTime >= SIGKILL_CHILDREN_AFTER_SECS)
1569 0 : seconds = 0;
1570 : else
1571 2502 : seconds = SIGKILL_CHILDREN_AFTER_SECS -
1572 : (curtime - AbortStartTime);
1573 :
1574 2502 : return seconds * 1000;
1575 : }
1576 : else
1577 253358 : return 60 * 1000;
1578 : }
1579 :
1580 0 : if (StartWorkerNeeded)
1581 0 : return 0;
1582 :
1583 0 : if (HaveCrashedWorker)
1584 : {
1585 : dlist_mutable_iter iter;
1586 :
1587 : /*
1588 : * When there are crashed bgworkers, we sleep just long enough that
1589 : * they are restarted when they request to be. Scan the list to
1590 : * determine the minimum of all wakeup times according to most recent
1591 : * crash time and requested restart interval.
1592 : */
1593 0 : dlist_foreach_modify(iter, &BackgroundWorkerList)
1594 : {
1595 : RegisteredBgWorker *rw;
1596 : TimestampTz this_wakeup;
1597 :
1598 0 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
1599 :
1600 0 : if (rw->rw_crashed_at == 0)
1601 0 : continue;
1602 :
1603 0 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART
1604 0 : || rw->rw_terminate)
1605 : {
1606 0 : ForgetBackgroundWorker(rw);
1607 0 : continue;
1608 : }
1609 :
1610 0 : this_wakeup = TimestampTzPlusMilliseconds(rw->rw_crashed_at,
1611 : 1000L * rw->rw_worker.bgw_restart_time);
1612 0 : if (next_wakeup == 0 || this_wakeup < next_wakeup)
1613 0 : next_wakeup = this_wakeup;
1614 : }
1615 : }
1616 :
1617 0 : if (next_wakeup != 0)
1618 : {
1619 : int ms;
1620 :
1621 : /* result of TimestampDifferenceMilliseconds is in [0, INT_MAX] */
1622 0 : ms = (int) TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
1623 : next_wakeup);
1624 0 : return Min(60 * 1000, ms);
1625 : }
1626 :
1627 0 : return 60 * 1000;
1628 : }
1629 :
1630 : /*
1631 : * Activate or deactivate notifications of server socket events. Since we
1632 : * don't currently have a way to remove events from an existing WaitEventSet,
1633 : * we'll just destroy and recreate the whole thing. This is called during
1634 : * shutdown so we can wait for backends to exit without accepting new
1635 : * connections, and during crash reinitialization when we need to start
1636 : * listening for new connections again. The WaitEventSet will be freed in fork
1637 : * children by ClosePostmasterPorts().
1638 : */
1639 : static void
1640 3654 : ConfigurePostmasterWaitSet(bool accept_connections)
1641 : {
1642 3654 : if (pm_wait_set)
1643 1834 : FreeWaitEventSet(pm_wait_set);
1644 3654 : pm_wait_set = NULL;
1645 :
1646 5484 : pm_wait_set = CreateWaitEventSet(NULL,
1647 1830 : accept_connections ? (1 + NumListenSockets) : 1);
1648 3654 : AddWaitEventToSet(pm_wait_set, WL_LATCH_SET, PGINVALID_SOCKET, MyLatch,
1649 : NULL);
1650 :
1651 3654 : if (accept_connections)
1652 : {
1653 3720 : for (int i = 0; i < NumListenSockets; i++)
1654 1890 : AddWaitEventToSet(pm_wait_set, WL_SOCKET_ACCEPT, ListenSockets[i],
1655 : NULL, NULL);
1656 : }
1657 3654 : }
1658 :
1659 : /*
1660 : * Main idle loop of postmaster
1661 : */
1662 : static int
1663 1820 : ServerLoop(void)
1664 : {
1665 : time_t last_lockfile_recheck_time,
1666 : last_touch_time;
1667 : WaitEvent events[MAXLISTEN];
1668 : int nevents;
1669 :
1670 1820 : ConfigurePostmasterWaitSet(true);
1671 1820 : last_lockfile_recheck_time = last_touch_time = time(NULL);
1672 :
1673 : for (;;)
1674 254040 : {
1675 : time_t now;
1676 :
1677 255860 : nevents = WaitEventSetWait(pm_wait_set,
1678 255860 : DetermineSleepTime(),
1679 : events,
1680 : lengthof(events),
1681 : 0 /* postmaster posts no wait_events */ );
1682 :
1683 : /*
1684 : * Latch set by signal handler, or new connection pending on any of
1685 : * our sockets? If the latter, fork a child process to deal with it.
1686 : */
1687 509898 : for (int i = 0; i < nevents; i++)
1688 : {
1689 255858 : if (events[i].events & WL_LATCH_SET)
1690 226912 : ResetLatch(MyLatch);
1691 :
1692 : /*
1693 : * The following requests are handled unconditionally, even if we
1694 : * didn't see WL_LATCH_SET. This gives high priority to shutdown
1695 : * and reload requests where the latch happens to appear later in
1696 : * events[] or will be reported by a later call to
1697 : * WaitEventSetWait().
1698 : */
1699 255858 : if (pending_pm_shutdown_request)
1700 1808 : process_pm_shutdown_request();
1701 255858 : if (pending_pm_reload_request)
1702 296 : process_pm_reload_request();
1703 255858 : if (pending_pm_child_exit)
1704 47236 : process_pm_child_exit();
1705 254038 : if (pending_pm_pmsignal)
1706 177628 : process_pm_pmsignal();
1707 :
1708 254038 : if (events[i].events & WL_SOCKET_ACCEPT)
1709 : {
1710 : ClientSocket s;
1711 :
1712 28946 : if (AcceptConnection(events[i].fd, &s) == STATUS_OK)
1713 28946 : BackendStartup(&s);
1714 :
1715 : /* We no longer need the open socket in this process */
1716 28946 : if (s.sock != PGINVALID_SOCKET)
1717 : {
1718 28946 : if (closesocket(s.sock) != 0)
1719 0 : elog(LOG, "could not close client socket: %m");
1720 : }
1721 : }
1722 : }
1723 :
1724 : /*
1725 : * If we need to launch any background processes after changing state
1726 : * or because some exited, do so now.
1727 : */
1728 254040 : LaunchMissingBackgroundProcesses();
1729 :
1730 : /* If we need to signal the autovacuum launcher, do so now */
1731 254040 : if (avlauncher_needs_signal)
1732 : {
1733 0 : avlauncher_needs_signal = false;
1734 0 : if (AutoVacLauncherPMChild != NULL)
1735 0 : signal_child(AutoVacLauncherPMChild, SIGUSR2);
1736 : }
1737 :
1738 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1739 :
1740 : /*
1741 : * With assertions enabled, check regularly for appearance of
1742 : * additional threads. All builds check at start and exit.
1743 : */
1744 : Assert(pthread_is_threaded_np() == 0);
1745 : #endif
1746 :
1747 : /*
1748 : * Lastly, check to see if it's time to do some things that we don't
1749 : * want to do every single time through the loop, because they're a
1750 : * bit expensive. Note that there's up to a minute of slop in when
1751 : * these tasks will be performed, since DetermineSleepTime() will let
1752 : * us sleep at most that long; except for SIGKILL timeout which has
1753 : * special-case logic there.
1754 : */
1755 254040 : now = time(NULL);
1756 :
1757 : /*
1758 : * If we already sent SIGQUIT to children and they are slow to shut
1759 : * down, it's time to send them SIGKILL (or SIGABRT if requested).
1760 : * This doesn't happen normally, but under certain conditions backends
1761 : * can get stuck while shutting down. This is a last measure to get
1762 : * them unwedged.
1763 : *
1764 : * Note we also do this during recovery from a process crash.
1765 : */
1766 254040 : if ((Shutdown >= ImmediateShutdown || FatalError) &&
1767 2522 : AbortStartTime != 0 &&
1768 2502 : (now - AbortStartTime) >= SIGKILL_CHILDREN_AFTER_SECS)
1769 : {
1770 : /* We were gentle with them before. Not anymore */
1771 0 : ereport(LOG,
1772 : /* translator: %s is SIGKILL or SIGABRT */
1773 : (errmsg("issuing %s to recalcitrant children",
1774 : send_abort_for_kill ? "SIGABRT" : "SIGKILL")));
1775 0 : TerminateChildren(send_abort_for_kill ? SIGABRT : SIGKILL);
1776 : /* reset flag so we don't SIGKILL again */
1777 0 : AbortStartTime = 0;
1778 : }
1779 :
1780 : /*
1781 : * Once a minute, verify that postmaster.pid hasn't been removed or
1782 : * overwritten. If it has, we force a shutdown. This avoids having
1783 : * postmasters and child processes hanging around after their database
1784 : * is gone, and maybe causing problems if a new database cluster is
1785 : * created in the same place. It also provides some protection
1786 : * against a DBA foolishly removing postmaster.pid and manually
1787 : * starting a new postmaster. Data corruption is likely to ensue from
1788 : * that anyway, but we can minimize the damage by aborting ASAP.
1789 : */
1790 254040 : if (now - last_lockfile_recheck_time >= 1 * SECS_PER_MINUTE)
1791 : {
1792 74 : if (!RecheckDataDirLockFile())
1793 : {
1794 0 : ereport(LOG,
1795 : (errmsg("performing immediate shutdown because data directory lock file is invalid")));
1796 0 : kill(MyProcPid, SIGQUIT);
1797 : }
1798 74 : last_lockfile_recheck_time = now;
1799 : }
1800 :
1801 : /*
1802 : * Touch Unix socket and lock files every 58 minutes, to ensure that
1803 : * they are not removed by overzealous /tmp-cleaning tasks. We assume
1804 : * no one runs cleaners with cutoff times of less than an hour ...
1805 : */
1806 254040 : if (now - last_touch_time >= 58 * SECS_PER_MINUTE)
1807 : {
1808 0 : TouchSocketFiles();
1809 0 : TouchSocketLockFiles();
1810 0 : last_touch_time = now;
1811 : }
1812 : }
1813 : }
1814 :
1815 : /*
1816 : * canAcceptConnections --- check to see if database state allows connections
1817 : * of the specified type. backend_type can be B_BACKEND or B_AUTOVAC_WORKER.
1818 : * (Note that we don't yet know whether a normal B_BACKEND connection might
1819 : * turn into a walsender.)
1820 : */
1821 : static CAC_state
1822 31240 : canAcceptConnections(BackendType backend_type)
1823 : {
1824 31240 : CAC_state result = CAC_OK;
1825 :
1826 : Assert(backend_type == B_BACKEND || backend_type == B_AUTOVAC_WORKER);
1827 :
1828 : /*
1829 : * Can't start backends when in startup/shutdown/inconsistent recovery
1830 : * state. We treat autovac workers the same as user backends for this
1831 : * purpose.
1832 : */
1833 31240 : if (pmState != PM_RUN && pmState != PM_HOT_STANDBY)
1834 : {
1835 526 : if (Shutdown > NoShutdown)
1836 98 : return CAC_SHUTDOWN; /* shutdown is pending */
1837 428 : else if (!FatalError && pmState == PM_STARTUP)
1838 404 : return CAC_STARTUP; /* normal startup */
1839 24 : else if (!FatalError && pmState == PM_RECOVERY)
1840 14 : return CAC_NOTHOTSTANDBY; /* not yet ready for hot standby */
1841 : else
1842 10 : return CAC_RECOVERY; /* else must be crash recovery */
1843 : }
1844 :
1845 : /*
1846 : * "Smart shutdown" restrictions are applied only to normal connections,
1847 : * not to autovac workers.
1848 : */
1849 30714 : if (!connsAllowed && backend_type == B_BACKEND)
1850 0 : return CAC_SHUTDOWN; /* shutdown is pending */
1851 :
1852 30714 : return result;
1853 : }
1854 :
1855 : /*
1856 : * ClosePostmasterPorts -- close all the postmaster's open sockets
1857 : *
1858 : * This is called during child process startup to release file descriptors
1859 : * that are not needed by that child process. The postmaster still has
1860 : * them open, of course.
1861 : *
1862 : * Note: we pass am_syslogger as a boolean because we don't want to set
1863 : * the global variable yet when this is called.
1864 : */
1865 : void
1866 45388 : ClosePostmasterPorts(bool am_syslogger)
1867 : {
1868 : /* Release resources held by the postmaster's WaitEventSet. */
1869 45388 : if (pm_wait_set)
1870 : {
1871 37974 : FreeWaitEventSetAfterFork(pm_wait_set);
1872 37974 : pm_wait_set = NULL;
1873 : }
1874 :
1875 : #ifndef WIN32
1876 :
1877 : /*
1878 : * Close the write end of postmaster death watch pipe. It's important to
1879 : * do this as early as possible, so that if postmaster dies, others won't
1880 : * think that it's still running because we're holding the pipe open.
1881 : */
1882 45388 : if (close(postmaster_alive_fds[POSTMASTER_FD_OWN]) != 0)
1883 0 : ereport(FATAL,
1884 : (errcode_for_file_access(),
1885 : errmsg_internal("could not close postmaster death monitoring pipe in child process: %m")));
1886 45388 : postmaster_alive_fds[POSTMASTER_FD_OWN] = -1;
1887 : /* Notify fd.c that we released one pipe FD. */
1888 45388 : ReleaseExternalFD();
1889 : #endif
1890 :
1891 : /*
1892 : * Close the postmaster's listen sockets. These aren't tracked by fd.c,
1893 : * so we don't call ReleaseExternalFD() here.
1894 : *
1895 : * The listen sockets are marked as FD_CLOEXEC, so this isn't needed in
1896 : * EXEC_BACKEND mode.
1897 : */
1898 : #ifndef EXEC_BACKEND
1899 45388 : if (ListenSockets)
1900 : {
1901 91862 : for (int i = 0; i < NumListenSockets; i++)
1902 : {
1903 46476 : if (closesocket(ListenSockets[i]) != 0)
1904 0 : elog(LOG, "could not close listen socket: %m");
1905 : }
1906 45386 : pfree(ListenSockets);
1907 : }
1908 45388 : NumListenSockets = 0;
1909 45388 : ListenSockets = NULL;
1910 : #endif
1911 :
1912 : /*
1913 : * If using syslogger, close the read side of the pipe. We don't bother
1914 : * tracking this in fd.c, either.
1915 : */
1916 45388 : if (!am_syslogger)
1917 : {
1918 : #ifndef WIN32
1919 45386 : if (syslogPipe[0] >= 0)
1920 34 : close(syslogPipe[0]);
1921 45386 : syslogPipe[0] = -1;
1922 : #else
1923 : if (syslogPipe[0])
1924 : CloseHandle(syslogPipe[0]);
1925 : syslogPipe[0] = 0;
1926 : #endif
1927 : }
1928 :
1929 : #ifdef USE_BONJOUR
1930 : /* If using Bonjour, close the connection to the mDNS daemon */
1931 : if (bonjour_sdref)
1932 : close(DNSServiceRefSockFD(bonjour_sdref));
1933 : #endif
1934 45388 : }
1935 :
1936 :
1937 : /*
1938 : * InitProcessGlobals -- set MyStartTime[stamp], random seeds
1939 : *
1940 : * Called early in the postmaster and every backend.
1941 : */
1942 : void
1943 47740 : InitProcessGlobals(void)
1944 : {
1945 47740 : MyStartTimestamp = GetCurrentTimestamp();
1946 47740 : MyStartTime = timestamptz_to_time_t(MyStartTimestamp);
1947 :
1948 : /*
1949 : * Set a different global seed in every process. We want something
1950 : * unpredictable, so if possible, use high-quality random bits for the
1951 : * seed. Otherwise, fall back to a seed based on timestamp and PID.
1952 : */
1953 47740 : if (unlikely(!pg_prng_strong_seed(&pg_global_prng_state)))
1954 : {
1955 : uint64 rseed;
1956 :
1957 : /*
1958 : * Since PIDs and timestamps tend to change more frequently in their
1959 : * least significant bits, shift the timestamp left to allow a larger
1960 : * total number of seeds in a given time period. Since that would
1961 : * leave only 20 bits of the timestamp that cycle every ~1 second,
1962 : * also mix in some higher bits.
1963 : */
1964 0 : rseed = ((uint64) MyProcPid) ^
1965 0 : ((uint64) MyStartTimestamp << 12) ^
1966 0 : ((uint64) MyStartTimestamp >> 20);
1967 :
1968 0 : pg_prng_seed(&pg_global_prng_state, rseed);
1969 : }
1970 :
1971 : /*
1972 : * Also make sure that we've set a good seed for random(3). Use of that
1973 : * is deprecated in core Postgres, but extensions might use it.
1974 : */
1975 : #ifndef WIN32
1976 47740 : srandom(pg_prng_uint32(&pg_global_prng_state));
1977 : #endif
1978 47740 : }
1979 :
1980 : /*
1981 : * Child processes use SIGUSR1 to notify us of 'pmsignals'. pg_ctl uses
1982 : * SIGUSR1 to ask postmaster to check for logrotate and promote files.
1983 : */
1984 : static void
1985 178146 : handle_pm_pmsignal_signal(SIGNAL_ARGS)
1986 : {
1987 178146 : pending_pm_pmsignal = true;
1988 178146 : SetLatch(MyLatch);
1989 178146 : }
1990 :
1991 : /*
1992 : * pg_ctl uses SIGHUP to request a reload of the configuration files.
1993 : */
1994 : static void
1995 296 : handle_pm_reload_request_signal(SIGNAL_ARGS)
1996 : {
1997 296 : pending_pm_reload_request = true;
1998 296 : SetLatch(MyLatch);
1999 296 : }
2000 :
2001 : /*
2002 : * Re-read config files, and tell children to do same.
2003 : */
2004 : static void
2005 296 : process_pm_reload_request(void)
2006 : {
2007 296 : pending_pm_reload_request = false;
2008 :
2009 296 : ereport(DEBUG2,
2010 : (errmsg_internal("postmaster received reload request signal")));
2011 :
2012 296 : if (Shutdown <= SmartShutdown)
2013 : {
2014 296 : ereport(LOG,
2015 : (errmsg("received SIGHUP, reloading configuration files")));
2016 296 : ProcessConfigFile(PGC_SIGHUP);
2017 296 : SignalChildren(SIGHUP, btmask_all_except(B_DEAD_END_BACKEND));
2018 :
2019 : /* Reload authentication config files too */
2020 296 : if (!load_hba())
2021 0 : ereport(LOG,
2022 : /* translator: %s is a configuration file */
2023 : (errmsg("%s was not reloaded", HbaFileName)));
2024 :
2025 296 : if (!load_ident())
2026 0 : ereport(LOG,
2027 : (errmsg("%s was not reloaded", IdentFileName)));
2028 :
2029 : #ifdef USE_SSL
2030 : /* Reload SSL configuration as well */
2031 296 : if (EnableSSL)
2032 : {
2033 6 : if (secure_initialize(false) == 0)
2034 4 : LoadedSSL = true;
2035 : else
2036 2 : ereport(LOG,
2037 : (errmsg("SSL configuration was not reloaded")));
2038 : }
2039 : else
2040 : {
2041 290 : secure_destroy();
2042 290 : LoadedSSL = false;
2043 : }
2044 : #endif
2045 :
2046 : #ifdef EXEC_BACKEND
2047 : /* Update the starting-point file for future children */
2048 : write_nondefault_variables(PGC_SIGHUP);
2049 : #endif
2050 : }
2051 296 : }
2052 :
2053 : /*
2054 : * pg_ctl uses SIGTERM, SIGINT and SIGQUIT to request different types of
2055 : * shutdown.
2056 : */
2057 : static void
2058 1808 : handle_pm_shutdown_request_signal(SIGNAL_ARGS)
2059 : {
2060 1808 : switch (postgres_signal_arg)
2061 : {
2062 88 : case SIGTERM:
2063 : /* smart is implied if the other two flags aren't set */
2064 88 : pending_pm_shutdown_request = true;
2065 88 : break;
2066 1040 : case SIGINT:
2067 1040 : pending_pm_fast_shutdown_request = true;
2068 1040 : pending_pm_shutdown_request = true;
2069 1040 : break;
2070 680 : case SIGQUIT:
2071 680 : pending_pm_immediate_shutdown_request = true;
2072 680 : pending_pm_shutdown_request = true;
2073 680 : break;
2074 : }
2075 1808 : SetLatch(MyLatch);
2076 1808 : }
2077 :
2078 : /*
2079 : * Process shutdown request.
2080 : */
2081 : static void
2082 1808 : process_pm_shutdown_request(void)
2083 : {
2084 : int mode;
2085 :
2086 1808 : ereport(DEBUG2,
2087 : (errmsg_internal("postmaster received shutdown request signal")));
2088 :
2089 1808 : pending_pm_shutdown_request = false;
2090 :
2091 : /*
2092 : * If more than one shutdown request signal arrived since the last server
2093 : * loop, take the one that is the most immediate. That matches the
2094 : * priority that would apply if we processed them one by one in any order.
2095 : */
2096 1808 : if (pending_pm_immediate_shutdown_request)
2097 : {
2098 680 : pending_pm_immediate_shutdown_request = false;
2099 680 : pending_pm_fast_shutdown_request = false;
2100 680 : mode = ImmediateShutdown;
2101 : }
2102 1128 : else if (pending_pm_fast_shutdown_request)
2103 : {
2104 1040 : pending_pm_fast_shutdown_request = false;
2105 1040 : mode = FastShutdown;
2106 : }
2107 : else
2108 88 : mode = SmartShutdown;
2109 :
2110 1808 : switch (mode)
2111 : {
2112 88 : case SmartShutdown:
2113 :
2114 : /*
2115 : * Smart Shutdown:
2116 : *
2117 : * Wait for children to end their work, then shut down.
2118 : */
2119 88 : if (Shutdown >= SmartShutdown)
2120 0 : break;
2121 88 : Shutdown = SmartShutdown;
2122 88 : ereport(LOG,
2123 : (errmsg("received smart shutdown request")));
2124 :
2125 : /* Report status */
2126 88 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2127 : #ifdef USE_SYSTEMD
2128 : sd_notify(0, "STOPPING=1");
2129 : #endif
2130 :
2131 : /*
2132 : * If we reached normal running, we go straight to waiting for
2133 : * client backends to exit. If already in PM_STOP_BACKENDS or a
2134 : * later state, do not change it.
2135 : */
2136 88 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2137 88 : connsAllowed = false;
2138 0 : else if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2139 : {
2140 : /* There should be no clients, so proceed to stop children */
2141 0 : UpdatePMState(PM_STOP_BACKENDS);
2142 : }
2143 :
2144 : /*
2145 : * Now wait for online backup mode to end and backends to exit. If
2146 : * that is already the case, PostmasterStateMachine will take the
2147 : * next step.
2148 : */
2149 88 : PostmasterStateMachine();
2150 88 : break;
2151 :
2152 1040 : case FastShutdown:
2153 :
2154 : /*
2155 : * Fast Shutdown:
2156 : *
2157 : * Abort all children with SIGTERM (rollback active transactions
2158 : * and exit) and shut down when they are gone.
2159 : */
2160 1040 : if (Shutdown >= FastShutdown)
2161 0 : break;
2162 1040 : Shutdown = FastShutdown;
2163 1040 : ereport(LOG,
2164 : (errmsg("received fast shutdown request")));
2165 :
2166 : /* Report status */
2167 1040 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2168 : #ifdef USE_SYSTEMD
2169 : sd_notify(0, "STOPPING=1");
2170 : #endif
2171 :
2172 1040 : if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2173 : {
2174 : /* Just shut down background processes silently */
2175 0 : UpdatePMState(PM_STOP_BACKENDS);
2176 : }
2177 1040 : else if (pmState == PM_RUN ||
2178 120 : pmState == PM_HOT_STANDBY)
2179 : {
2180 : /* Report that we're about to zap live client sessions */
2181 1040 : ereport(LOG,
2182 : (errmsg("aborting any active transactions")));
2183 1040 : UpdatePMState(PM_STOP_BACKENDS);
2184 : }
2185 :
2186 : /*
2187 : * PostmasterStateMachine will issue any necessary signals, or
2188 : * take the next step if no child processes need to be killed.
2189 : */
2190 1040 : PostmasterStateMachine();
2191 1040 : break;
2192 :
2193 680 : case ImmediateShutdown:
2194 :
2195 : /*
2196 : * Immediate Shutdown:
2197 : *
2198 : * abort all children with SIGQUIT, wait for them to exit,
2199 : * terminate remaining ones with SIGKILL, then exit without
2200 : * attempt to properly shut down the data base system.
2201 : */
2202 680 : if (Shutdown >= ImmediateShutdown)
2203 0 : break;
2204 680 : Shutdown = ImmediateShutdown;
2205 680 : ereport(LOG,
2206 : (errmsg("received immediate shutdown request")));
2207 :
2208 : /* Report status */
2209 680 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2210 : #ifdef USE_SYSTEMD
2211 : sd_notify(0, "STOPPING=1");
2212 : #endif
2213 :
2214 : /* tell children to shut down ASAP */
2215 : /* (note we don't apply send_abort_for_crash here) */
2216 680 : SetQuitSignalReason(PMQUIT_FOR_STOP);
2217 680 : TerminateChildren(SIGQUIT);
2218 680 : UpdatePMState(PM_WAIT_BACKENDS);
2219 :
2220 : /* set stopwatch for them to die */
2221 680 : AbortStartTime = time(NULL);
2222 :
2223 : /*
2224 : * Now wait for backends to exit. If there are none,
2225 : * PostmasterStateMachine will take the next step.
2226 : */
2227 680 : PostmasterStateMachine();
2228 680 : break;
2229 : }
2230 1808 : }
2231 :
2232 : static void
2233 47324 : handle_pm_child_exit_signal(SIGNAL_ARGS)
2234 : {
2235 47324 : pending_pm_child_exit = true;
2236 47324 : SetLatch(MyLatch);
2237 47324 : }
2238 :
2239 : /*
2240 : * Cleanup after a child process dies.
2241 : */
2242 : static void
2243 47236 : process_pm_child_exit(void)
2244 : {
2245 : int pid; /* process id of dead child process */
2246 : int exitstatus; /* its exit status */
2247 :
2248 47236 : pending_pm_child_exit = false;
2249 :
2250 47236 : ereport(DEBUG4,
2251 : (errmsg_internal("reaping dead processes")));
2252 :
2253 98656 : while ((pid = waitpid(-1, &exitstatus, WNOHANG)) > 0)
2254 : {
2255 : PMChild *pmchild;
2256 :
2257 : /*
2258 : * Check if this child was a startup process.
2259 : */
2260 51426 : if (StartupPMChild && pid == StartupPMChild->pid)
2261 : {
2262 1830 : ReleasePostmasterChildSlot(StartupPMChild);
2263 1830 : StartupPMChild = NULL;
2264 :
2265 : /*
2266 : * Startup process exited in response to a shutdown request (or it
2267 : * completed normally regardless of the shutdown request).
2268 : */
2269 1830 : if (Shutdown > NoShutdown &&
2270 210 : (EXIT_STATUS_0(exitstatus) || EXIT_STATUS_1(exitstatus)))
2271 : {
2272 122 : StartupStatus = STARTUP_NOT_RUNNING;
2273 122 : UpdatePMState(PM_WAIT_BACKENDS);
2274 : /* PostmasterStateMachine logic does the rest */
2275 122 : continue;
2276 : }
2277 :
2278 1708 : if (EXIT_STATUS_3(exitstatus))
2279 : {
2280 0 : ereport(LOG,
2281 : (errmsg("shutdown at recovery target")));
2282 0 : StartupStatus = STARTUP_NOT_RUNNING;
2283 0 : Shutdown = Max(Shutdown, SmartShutdown);
2284 0 : TerminateChildren(SIGTERM);
2285 0 : UpdatePMState(PM_WAIT_BACKENDS);
2286 : /* PostmasterStateMachine logic does the rest */
2287 0 : continue;
2288 : }
2289 :
2290 : /*
2291 : * Unexpected exit of startup process (including FATAL exit)
2292 : * during PM_STARTUP is treated as catastrophic. There are no
2293 : * other processes running yet, so we can just exit.
2294 : */
2295 1708 : if (pmState == PM_STARTUP &&
2296 1312 : StartupStatus != STARTUP_SIGNALED &&
2297 1312 : !EXIT_STATUS_0(exitstatus))
2298 : {
2299 6 : LogChildExit(LOG, _("startup process"),
2300 : pid, exitstatus);
2301 6 : ereport(LOG,
2302 : (errmsg("aborting startup due to startup process failure")));
2303 6 : ExitPostmaster(1);
2304 : }
2305 :
2306 : /*
2307 : * After PM_STARTUP, any unexpected exit (including FATAL exit) of
2308 : * the startup process is catastrophic, so kill other children,
2309 : * and set StartupStatus so we don't try to reinitialize after
2310 : * they're gone. Exception: if StartupStatus is STARTUP_SIGNALED,
2311 : * then we previously sent the startup process a SIGQUIT; so
2312 : * that's probably the reason it died, and we do want to try to
2313 : * restart in that case.
2314 : *
2315 : * This stanza also handles the case where we sent a SIGQUIT
2316 : * during PM_STARTUP due to some dead-end child crashing: in that
2317 : * situation, if the startup process dies on the SIGQUIT, we need
2318 : * to transition to PM_WAIT_BACKENDS state which will allow
2319 : * PostmasterStateMachine to restart the startup process. (On the
2320 : * other hand, the startup process might complete normally, if we
2321 : * were too late with the SIGQUIT. In that case we'll fall
2322 : * through and commence normal operations.)
2323 : */
2324 1702 : if (!EXIT_STATUS_0(exitstatus))
2325 : {
2326 94 : if (StartupStatus == STARTUP_SIGNALED)
2327 : {
2328 88 : StartupStatus = STARTUP_NOT_RUNNING;
2329 88 : if (pmState == PM_STARTUP)
2330 0 : UpdatePMState(PM_WAIT_BACKENDS);
2331 : }
2332 : else
2333 6 : StartupStatus = STARTUP_CRASHED;
2334 94 : HandleChildCrash(pid, exitstatus,
2335 94 : _("startup process"));
2336 94 : continue;
2337 : }
2338 :
2339 : /*
2340 : * Startup succeeded, commence normal operations
2341 : */
2342 1608 : StartupStatus = STARTUP_NOT_RUNNING;
2343 1608 : FatalError = false;
2344 1608 : AbortStartTime = 0;
2345 1608 : ReachedNormalRunning = true;
2346 1608 : UpdatePMState(PM_RUN);
2347 1608 : connsAllowed = true;
2348 :
2349 : /*
2350 : * At the next iteration of the postmaster's main loop, we will
2351 : * crank up the background tasks like the autovacuum launcher and
2352 : * background workers that were not started earlier already.
2353 : */
2354 1608 : StartWorkerNeeded = true;
2355 :
2356 : /* at this point we are really open for business */
2357 1608 : ereport(LOG,
2358 : (errmsg("database system is ready to accept connections")));
2359 :
2360 : /* Report status */
2361 1608 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
2362 : #ifdef USE_SYSTEMD
2363 : sd_notify(0, "READY=1");
2364 : #endif
2365 :
2366 1608 : continue;
2367 : }
2368 :
2369 : /*
2370 : * Was it the bgwriter? Normal exit can be ignored; we'll start a new
2371 : * one at the next iteration of the postmaster's main loop, if
2372 : * necessary. Any other exit condition is treated as a crash.
2373 : */
2374 49596 : if (BgWriterPMChild && pid == BgWriterPMChild->pid)
2375 : {
2376 1824 : ReleasePostmasterChildSlot(BgWriterPMChild);
2377 1824 : BgWriterPMChild = NULL;
2378 1824 : if (!EXIT_STATUS_0(exitstatus))
2379 696 : HandleChildCrash(pid, exitstatus,
2380 696 : _("background writer process"));
2381 1824 : continue;
2382 : }
2383 :
2384 : /*
2385 : * Was it the checkpointer?
2386 : */
2387 47772 : if (CheckpointerPMChild && pid == CheckpointerPMChild->pid)
2388 : {
2389 1824 : ReleasePostmasterChildSlot(CheckpointerPMChild);
2390 1824 : CheckpointerPMChild = NULL;
2391 1824 : if (EXIT_STATUS_0(exitstatus) && pmState == PM_WAIT_CHECKPOINTER)
2392 1128 : {
2393 : /*
2394 : * OK, we saw normal exit of the checkpointer after it's been
2395 : * told to shut down. We know checkpointer wrote a shutdown
2396 : * checkpoint, otherwise we'd still be in
2397 : * PM_WAIT_XLOG_SHUTDOWN state.
2398 : *
2399 : * At this point only dead-end children and logger should be
2400 : * left.
2401 : */
2402 1128 : UpdatePMState(PM_WAIT_DEAD_END);
2403 1128 : ConfigurePostmasterWaitSet(false);
2404 1128 : SignalChildren(SIGTERM, btmask_all_except(B_LOGGER));
2405 : }
2406 : else
2407 : {
2408 : /*
2409 : * Any unexpected exit of the checkpointer (including FATAL
2410 : * exit) is treated as a crash.
2411 : */
2412 696 : HandleChildCrash(pid, exitstatus,
2413 696 : _("checkpointer process"));
2414 : }
2415 :
2416 1824 : continue;
2417 : }
2418 :
2419 : /*
2420 : * Was it the wal writer? Normal exit can be ignored; we'll start a
2421 : * new one at the next iteration of the postmaster's main loop, if
2422 : * necessary. Any other exit condition is treated as a crash.
2423 : */
2424 45948 : if (WalWriterPMChild && pid == WalWriterPMChild->pid)
2425 : {
2426 1608 : ReleasePostmasterChildSlot(WalWriterPMChild);
2427 1608 : WalWriterPMChild = NULL;
2428 1608 : if (!EXIT_STATUS_0(exitstatus))
2429 600 : HandleChildCrash(pid, exitstatus,
2430 600 : _("WAL writer process"));
2431 1608 : continue;
2432 : }
2433 :
2434 : /*
2435 : * Was it the wal receiver? If exit status is zero (normal) or one
2436 : * (FATAL exit), we assume everything is all right just like normal
2437 : * backends. (If we need a new wal receiver, we'll start one at the
2438 : * next iteration of the postmaster's main loop.)
2439 : */
2440 44340 : if (WalReceiverPMChild && pid == WalReceiverPMChild->pid)
2441 : {
2442 496 : ReleasePostmasterChildSlot(WalReceiverPMChild);
2443 496 : WalReceiverPMChild = NULL;
2444 496 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2445 42 : HandleChildCrash(pid, exitstatus,
2446 42 : _("WAL receiver process"));
2447 496 : continue;
2448 : }
2449 :
2450 : /*
2451 : * Was it the wal summarizer? Normal exit can be ignored; we'll start
2452 : * a new one at the next iteration of the postmaster's main loop, if
2453 : * necessary. Any other exit condition is treated as a crash.
2454 : */
2455 43844 : if (WalSummarizerPMChild && pid == WalSummarizerPMChild->pid)
2456 : {
2457 42 : ReleasePostmasterChildSlot(WalSummarizerPMChild);
2458 42 : WalSummarizerPMChild = NULL;
2459 42 : if (!EXIT_STATUS_0(exitstatus))
2460 36 : HandleChildCrash(pid, exitstatus,
2461 36 : _("WAL summarizer process"));
2462 42 : continue;
2463 : }
2464 :
2465 : /*
2466 : * Was it the autovacuum launcher? Normal exit can be ignored; we'll
2467 : * start a new one at the next iteration of the postmaster's main
2468 : * loop, if necessary. Any other exit condition is treated as a
2469 : * crash.
2470 : */
2471 43802 : if (AutoVacLauncherPMChild && pid == AutoVacLauncherPMChild->pid)
2472 : {
2473 1344 : ReleasePostmasterChildSlot(AutoVacLauncherPMChild);
2474 1344 : AutoVacLauncherPMChild = NULL;
2475 1344 : if (!EXIT_STATUS_0(exitstatus))
2476 510 : HandleChildCrash(pid, exitstatus,
2477 510 : _("autovacuum launcher process"));
2478 1344 : continue;
2479 : }
2480 :
2481 : /*
2482 : * Was it the archiver? If exit status is zero (normal) or one (FATAL
2483 : * exit), we assume everything is all right just like normal backends
2484 : * and just try to start a new one on the next cycle of the
2485 : * postmaster's main loop, to retry archiving remaining files.
2486 : */
2487 42458 : if (PgArchPMChild && pid == PgArchPMChild->pid)
2488 : {
2489 114 : ReleasePostmasterChildSlot(PgArchPMChild);
2490 114 : PgArchPMChild = NULL;
2491 114 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2492 78 : HandleChildCrash(pid, exitstatus,
2493 78 : _("archiver process"));
2494 114 : continue;
2495 : }
2496 :
2497 : /* Was it the system logger? If so, try to start a new one */
2498 42344 : if (SysLoggerPMChild && pid == SysLoggerPMChild->pid)
2499 : {
2500 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
2501 0 : SysLoggerPMChild = NULL;
2502 :
2503 : /* for safety's sake, launch new logger *first* */
2504 0 : if (Logging_collector)
2505 0 : StartSysLogger();
2506 :
2507 0 : if (!EXIT_STATUS_0(exitstatus))
2508 0 : LogChildExit(LOG, _("system logger process"),
2509 : pid, exitstatus);
2510 0 : continue;
2511 : }
2512 :
2513 : /*
2514 : * Was it the slot sync worker? Normal exit or FATAL exit can be
2515 : * ignored (FATAL can be caused by libpqwalreceiver on receiving
2516 : * shutdown request by the startup process during promotion); we'll
2517 : * start a new one at the next iteration of the postmaster's main
2518 : * loop, if necessary. Any other exit condition is treated as a crash.
2519 : */
2520 42344 : if (SlotSyncWorkerPMChild && pid == SlotSyncWorkerPMChild->pid)
2521 : {
2522 8 : ReleasePostmasterChildSlot(SlotSyncWorkerPMChild);
2523 8 : SlotSyncWorkerPMChild = NULL;
2524 8 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2525 0 : HandleChildCrash(pid, exitstatus,
2526 0 : _("slot sync worker process"));
2527 8 : continue;
2528 : }
2529 :
2530 : /* Was it an IO worker? */
2531 42336 : if (maybe_reap_io_worker(pid))
2532 : {
2533 5538 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2534 2088 : HandleChildCrash(pid, exitstatus, _("io worker"));
2535 :
2536 5538 : maybe_adjust_io_workers();
2537 5538 : continue;
2538 : }
2539 :
2540 : /*
2541 : * Was it a backend or a background worker?
2542 : */
2543 36798 : pmchild = FindPostmasterChildByPid(pid);
2544 36798 : if (pmchild)
2545 : {
2546 36798 : CleanupBackend(pmchild, exitstatus);
2547 : }
2548 :
2549 : /*
2550 : * We don't know anything about this child process. That's highly
2551 : * unexpected, as we do track all the child processes that we fork.
2552 : */
2553 : else
2554 : {
2555 0 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2556 0 : HandleChildCrash(pid, exitstatus, _("untracked child process"));
2557 : else
2558 0 : LogChildExit(LOG, _("untracked child process"), pid, exitstatus);
2559 : }
2560 : } /* loop over pending child-death reports */
2561 :
2562 : /*
2563 : * After cleaning out the SIGCHLD queue, see if we have any state changes
2564 : * or actions to make.
2565 : */
2566 47230 : PostmasterStateMachine();
2567 45416 : }
2568 :
2569 : /*
2570 : * CleanupBackend -- cleanup after terminated backend or background worker.
2571 : *
2572 : * Remove all local state associated with the child process and release its
2573 : * PMChild slot.
2574 : */
2575 : static void
2576 36798 : CleanupBackend(PMChild *bp,
2577 : int exitstatus) /* child's exit status. */
2578 : {
2579 : char namebuf[MAXPGPATH];
2580 : const char *procname;
2581 36798 : bool crashed = false;
2582 36798 : bool logged = false;
2583 : pid_t bp_pid;
2584 : bool bp_bgworker_notify;
2585 : BackendType bp_bkend_type;
2586 : RegisteredBgWorker *rw;
2587 :
2588 : /* Construct a process name for the log message */
2589 36798 : if (bp->bkend_type == B_BG_WORKER)
2590 : {
2591 5558 : snprintf(namebuf, MAXPGPATH, _("background worker \"%s\""),
2592 5558 : bp->rw->rw_worker.bgw_type);
2593 5558 : procname = namebuf;
2594 : }
2595 : else
2596 31240 : procname = _(GetBackendTypeDesc(bp->bkend_type));
2597 :
2598 : /*
2599 : * If a backend dies in an ugly way then we must signal all other backends
2600 : * to quickdie. If exit status is zero (normal) or one (FATAL exit), we
2601 : * assume everything is all right and proceed to remove the backend from
2602 : * the active child list.
2603 : */
2604 36798 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
2605 1646 : crashed = true;
2606 :
2607 : #ifdef WIN32
2608 :
2609 : /*
2610 : * On win32, also treat ERROR_WAIT_NO_CHILDREN (128) as nonfatal case,
2611 : * since that sometimes happens under load when the process fails to start
2612 : * properly (long before it starts using shared memory). Microsoft reports
2613 : * it is related to mutex failure:
2614 : * http://archives.postgresql.org/pgsql-hackers/2010-09/msg00790.php
2615 : */
2616 : if (exitstatus == ERROR_WAIT_NO_CHILDREN)
2617 : {
2618 : LogChildExit(LOG, procname, bp->pid, exitstatus);
2619 : logged = true;
2620 : crashed = false;
2621 : }
2622 : #endif
2623 :
2624 : /*
2625 : * Release the PMChild entry.
2626 : *
2627 : * If the process attached to shared memory, this also checks that it
2628 : * detached cleanly.
2629 : */
2630 36798 : bp_pid = bp->pid;
2631 36798 : bp_bgworker_notify = bp->bgworker_notify;
2632 36798 : bp_bkend_type = bp->bkend_type;
2633 36798 : rw = bp->rw;
2634 36798 : if (!ReleasePostmasterChildSlot(bp))
2635 : {
2636 : /*
2637 : * Uh-oh, the child failed to clean itself up. Treat as a crash after
2638 : * all.
2639 : */
2640 768 : crashed = true;
2641 : }
2642 36798 : bp = NULL;
2643 :
2644 : /*
2645 : * In a crash case, exit immediately without resetting background worker
2646 : * state. However, if restart_after_crash is enabled, the background
2647 : * worker state (e.g., rw_pid) still needs be reset so the worker can
2648 : * restart after crash recovery. This reset is handled in
2649 : * ResetBackgroundWorkerCrashTimes(), not here.
2650 : */
2651 36798 : if (crashed)
2652 : {
2653 1646 : HandleChildCrash(bp_pid, exitstatus, procname);
2654 1646 : return;
2655 : }
2656 :
2657 : /*
2658 : * This backend may have been slated to receive SIGUSR1 when some
2659 : * background worker started or stopped. Cancel those notifications, as
2660 : * we don't want to signal PIDs that are not PostgreSQL backends. This
2661 : * gets skipped in the (probably very common) case where the backend has
2662 : * never requested any such notifications.
2663 : */
2664 35152 : if (bp_bgworker_notify)
2665 518 : BackgroundWorkerStopNotifications(bp_pid);
2666 :
2667 : /*
2668 : * If it was an autovacuum worker, wake up the launcher so that it can
2669 : * immediately launch a new worker or rebalance to cost limit setting of
2670 : * the remaining workers.
2671 : */
2672 35152 : if (bp_bkend_type == B_AUTOVAC_WORKER && AutoVacLauncherPMChild != NULL)
2673 2288 : signal_child(AutoVacLauncherPMChild, SIGUSR2);
2674 :
2675 : /*
2676 : * If it was a background worker, also update its RegisteredBgWorker
2677 : * entry.
2678 : */
2679 35152 : if (bp_bkend_type == B_BG_WORKER)
2680 : {
2681 4882 : if (!EXIT_STATUS_0(exitstatus))
2682 : {
2683 : /* Record timestamp, so we know when to restart the worker. */
2684 1394 : rw->rw_crashed_at = GetCurrentTimestamp();
2685 : }
2686 : else
2687 : {
2688 : /* Zero exit status means terminate */
2689 3488 : rw->rw_crashed_at = 0;
2690 3488 : rw->rw_terminate = true;
2691 : }
2692 :
2693 4882 : rw->rw_pid = 0;
2694 4882 : ReportBackgroundWorkerExit(rw); /* report child death */
2695 :
2696 4882 : if (!logged)
2697 : {
2698 4882 : LogChildExit(EXIT_STATUS_0(exitstatus) ? DEBUG1 : LOG,
2699 : procname, bp_pid, exitstatus);
2700 4882 : logged = true;
2701 : }
2702 :
2703 : /* have it be restarted */
2704 4882 : HaveCrashedWorker = true;
2705 : }
2706 :
2707 35152 : if (!logged)
2708 30270 : LogChildExit(DEBUG2, procname, bp_pid, exitstatus);
2709 : }
2710 :
2711 : /*
2712 : * Transition into FatalError state, in response to something bad having
2713 : * happened. Commonly the caller will have logged the reason for entering
2714 : * FatalError state.
2715 : *
2716 : * This should only be called when not already in FatalError or
2717 : * ImmediateShutdown state.
2718 : */
2719 : static void
2720 16 : HandleFatalError(QuitSignalReason reason, bool consider_sigabrt)
2721 : {
2722 : int sigtosend;
2723 :
2724 : Assert(!FatalError);
2725 : Assert(Shutdown != ImmediateShutdown);
2726 :
2727 16 : SetQuitSignalReason(reason);
2728 :
2729 16 : if (consider_sigabrt && send_abort_for_crash)
2730 0 : sigtosend = SIGABRT;
2731 : else
2732 16 : sigtosend = SIGQUIT;
2733 :
2734 : /*
2735 : * Signal all other child processes to exit.
2736 : *
2737 : * We could exclude dead-end children here, but at least when sending
2738 : * SIGABRT it seems better to include them.
2739 : */
2740 16 : TerminateChildren(sigtosend);
2741 :
2742 16 : FatalError = true;
2743 :
2744 : /*
2745 : * Choose the appropriate new state to react to the fatal error. Unless we
2746 : * were already in the process of shutting down, we go through
2747 : * PM_WAIT_BACKENDS. For errors during the shutdown sequence, we directly
2748 : * switch to PM_WAIT_DEAD_END.
2749 : */
2750 16 : switch (pmState)
2751 : {
2752 0 : case PM_INIT:
2753 : /* shouldn't have any children */
2754 : Assert(false);
2755 0 : break;
2756 0 : case PM_STARTUP:
2757 : /* should have been handled in process_pm_child_exit */
2758 : Assert(false);
2759 0 : break;
2760 :
2761 : /* wait for children to die */
2762 16 : case PM_RECOVERY:
2763 : case PM_HOT_STANDBY:
2764 : case PM_RUN:
2765 : case PM_STOP_BACKENDS:
2766 16 : UpdatePMState(PM_WAIT_BACKENDS);
2767 16 : break;
2768 :
2769 0 : case PM_WAIT_BACKENDS:
2770 : /* there might be more backends to wait for */
2771 0 : break;
2772 :
2773 0 : case PM_WAIT_XLOG_SHUTDOWN:
2774 : case PM_WAIT_XLOG_ARCHIVAL:
2775 : case PM_WAIT_CHECKPOINTER:
2776 : case PM_WAIT_IO_WORKERS:
2777 :
2778 : /*
2779 : * NB: Similar code exists in PostmasterStateMachine()'s handling
2780 : * of FatalError in PM_STOP_BACKENDS/PM_WAIT_BACKENDS states.
2781 : */
2782 0 : ConfigurePostmasterWaitSet(false);
2783 0 : UpdatePMState(PM_WAIT_DEAD_END);
2784 0 : break;
2785 :
2786 0 : case PM_WAIT_DEAD_END:
2787 : case PM_NO_CHILDREN:
2788 0 : break;
2789 : }
2790 :
2791 : /*
2792 : * .. and if this doesn't happen quickly enough, now the clock is ticking
2793 : * for us to kill them without mercy.
2794 : */
2795 16 : if (AbortStartTime == 0)
2796 16 : AbortStartTime = time(NULL);
2797 16 : }
2798 :
2799 : /*
2800 : * HandleChildCrash -- cleanup after failed backend, bgwriter, checkpointer,
2801 : * walwriter, autovacuum, archiver, slot sync worker, or background worker.
2802 : *
2803 : * The objectives here are to clean up our local state about the child
2804 : * process, and to signal all other remaining children to quickdie.
2805 : *
2806 : * The caller has already released its PMChild slot.
2807 : */
2808 : static void
2809 6486 : HandleChildCrash(int pid, int exitstatus, const char *procname)
2810 : {
2811 : /*
2812 : * We only log messages and send signals if this is the first process
2813 : * crash and we're not doing an immediate shutdown; otherwise, we're only
2814 : * here to update postmaster's idea of live processes. If we have already
2815 : * signaled children, nonzero exit status is to be expected, so don't
2816 : * clutter log.
2817 : */
2818 6486 : if (FatalError || Shutdown == ImmediateShutdown)
2819 6470 : return;
2820 :
2821 16 : LogChildExit(LOG, procname, pid, exitstatus);
2822 16 : ereport(LOG,
2823 : (errmsg("terminating any other active server processes")));
2824 :
2825 : /*
2826 : * Switch into error state. The crashed process has already been removed
2827 : * from ActiveChildList.
2828 : */
2829 16 : HandleFatalError(PMQUIT_FOR_CRASH, true);
2830 : }
2831 :
2832 : /*
2833 : * Log the death of a child process.
2834 : */
2835 : static void
2836 35174 : LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2837 : {
2838 : /*
2839 : * size of activity_buffer is arbitrary, but set equal to default
2840 : * track_activity_query_size
2841 : */
2842 : char activity_buffer[1024];
2843 35174 : const char *activity = NULL;
2844 :
2845 35174 : if (!EXIT_STATUS_0(exitstatus))
2846 2238 : activity = pgstat_get_crashed_backend_activity(pid,
2847 : activity_buffer,
2848 : sizeof(activity_buffer));
2849 :
2850 35174 : if (WIFEXITED(exitstatus))
2851 35166 : ereport(lev,
2852 :
2853 : /*------
2854 : translator: %s is a noun phrase describing a child process, such as
2855 : "server process" */
2856 : (errmsg("%s (PID %d) exited with exit code %d",
2857 : procname, pid, WEXITSTATUS(exitstatus)),
2858 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2859 8 : else if (WIFSIGNALED(exitstatus))
2860 : {
2861 : #if defined(WIN32)
2862 : ereport(lev,
2863 :
2864 : /*------
2865 : translator: %s is a noun phrase describing a child process, such as
2866 : "server process" */
2867 : (errmsg("%s (PID %d) was terminated by exception 0x%X",
2868 : procname, pid, WTERMSIG(exitstatus)),
2869 : errhint("See C include file \"ntstatus.h\" for a description of the hexadecimal value."),
2870 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2871 : #else
2872 8 : ereport(lev,
2873 :
2874 : /*------
2875 : translator: %s is a noun phrase describing a child process, such as
2876 : "server process" */
2877 : (errmsg("%s (PID %d) was terminated by signal %d: %s",
2878 : procname, pid, WTERMSIG(exitstatus),
2879 : pg_strsignal(WTERMSIG(exitstatus))),
2880 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2881 : #endif
2882 : }
2883 : else
2884 0 : ereport(lev,
2885 :
2886 : /*------
2887 : translator: %s is a noun phrase describing a child process, such as
2888 : "server process" */
2889 : (errmsg("%s (PID %d) exited with unrecognized status %d",
2890 : procname, pid, exitstatus),
2891 : activity ? errdetail("Failed process was running: %s", activity) : 0));
2892 35174 : }
2893 :
2894 : /*
2895 : * Advance the postmaster's state machine and take actions as appropriate
2896 : *
2897 : * This is common code for process_pm_shutdown_request(),
2898 : * process_pm_child_exit() and process_pm_pmsignal(), which process the signals
2899 : * that might mean we need to change state.
2900 : */
2901 : static void
2902 52692 : PostmasterStateMachine(void)
2903 : {
2904 : /* If we're doing a smart shutdown, try to advance that state. */
2905 52692 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2906 : {
2907 37036 : if (!connsAllowed)
2908 : {
2909 : /*
2910 : * This state ends when we have no normal client backends running.
2911 : * Then we're ready to stop other children.
2912 : */
2913 232 : if (CountChildren(btmask(B_BACKEND)) == 0)
2914 88 : UpdatePMState(PM_STOP_BACKENDS);
2915 : }
2916 : }
2917 :
2918 : /*
2919 : * In the PM_WAIT_BACKENDS state, wait for all the regular backends and
2920 : * processes like autovacuum and background workers that are comparable to
2921 : * backends to exit.
2922 : *
2923 : * PM_STOP_BACKENDS is a transient state that means the same as
2924 : * PM_WAIT_BACKENDS, but we signal the processes first, before waiting for
2925 : * them. Treating it as a distinct pmState allows us to share this code
2926 : * across multiple shutdown code paths.
2927 : */
2928 52692 : if (pmState == PM_STOP_BACKENDS || pmState == PM_WAIT_BACKENDS)
2929 : {
2930 9774 : BackendTypeMask targetMask = BTYPE_MASK_NONE;
2931 :
2932 : /*
2933 : * PM_WAIT_BACKENDS state ends when we have no regular backends, no
2934 : * autovac launcher or workers, and no bgworkers (including
2935 : * unconnected ones).
2936 : */
2937 9774 : targetMask = btmask_add(targetMask,
2938 : B_BACKEND,
2939 : B_AUTOVAC_LAUNCHER,
2940 : B_AUTOVAC_WORKER,
2941 : B_BG_WORKER);
2942 :
2943 : /*
2944 : * No walwriter, bgwriter, slot sync worker, or WAL summarizer either.
2945 : */
2946 9774 : targetMask = btmask_add(targetMask,
2947 : B_WAL_WRITER,
2948 : B_BG_WRITER,
2949 : B_SLOTSYNC_WORKER,
2950 : B_WAL_SUMMARIZER);
2951 :
2952 : /* If we're in recovery, also stop startup and walreceiver procs */
2953 9774 : targetMask = btmask_add(targetMask,
2954 : B_STARTUP,
2955 : B_WAL_RECEIVER);
2956 :
2957 : /*
2958 : * If we are doing crash recovery or an immediate shutdown then we
2959 : * expect archiver, checkpointer, io workers and walsender to exit as
2960 : * well, otherwise not.
2961 : */
2962 9774 : if (FatalError || Shutdown >= ImmediateShutdown)
2963 3190 : targetMask = btmask_add(targetMask,
2964 : B_CHECKPOINTER,
2965 : B_ARCHIVER,
2966 : B_IO_WORKER,
2967 : B_WAL_SENDER);
2968 :
2969 : /*
2970 : * Normally archiver, checkpointer, IO workers and walsenders will
2971 : * continue running; they will be terminated later after writing the
2972 : * checkpoint record. We also let dead-end children to keep running
2973 : * for now. The syslogger process exits last.
2974 : *
2975 : * This assertion checks that we have covered all backend types,
2976 : * either by including them in targetMask, or by noting here that they
2977 : * are allowed to continue running.
2978 : */
2979 : #ifdef USE_ASSERT_CHECKING
2980 : {
2981 : BackendTypeMask remainMask = BTYPE_MASK_NONE;
2982 :
2983 : remainMask = btmask_add(remainMask,
2984 : B_DEAD_END_BACKEND,
2985 : B_LOGGER);
2986 :
2987 : /*
2988 : * Archiver, checkpointer, IO workers, and walsender may or may
2989 : * not be in targetMask already.
2990 : */
2991 : remainMask = btmask_add(remainMask,
2992 : B_ARCHIVER,
2993 : B_CHECKPOINTER,
2994 : B_IO_WORKER,
2995 : B_WAL_SENDER);
2996 :
2997 : /* these are not real postmaster children */
2998 : remainMask = btmask_add(remainMask,
2999 : B_INVALID,
3000 : B_STANDALONE_BACKEND);
3001 :
3002 : /* All types should be included in targetMask or remainMask */
3003 : Assert((remainMask.mask | targetMask.mask) == BTYPE_MASK_ALL.mask);
3004 : }
3005 : #endif
3006 :
3007 : /* If we had not yet signaled the processes to exit, do so now */
3008 9774 : if (pmState == PM_STOP_BACKENDS)
3009 : {
3010 : /*
3011 : * Forget any pending requests for background workers, since we're
3012 : * no longer willing to launch any new workers. (If additional
3013 : * requests arrive, BackgroundWorkerStateChange will reject them.)
3014 : */
3015 1128 : ForgetUnstartedBackgroundWorkers();
3016 :
3017 1128 : SignalChildren(SIGTERM, targetMask);
3018 :
3019 1128 : UpdatePMState(PM_WAIT_BACKENDS);
3020 : }
3021 :
3022 : /* Are any of the target processes still running? */
3023 9774 : if (CountChildren(targetMask) == 0)
3024 : {
3025 1824 : if (Shutdown >= ImmediateShutdown || FatalError)
3026 : {
3027 : /*
3028 : * Stop any dead-end children and stop creating new ones.
3029 : *
3030 : * NB: Similar code exists in HandleFatalError(), when the
3031 : * error happens in pmState > PM_WAIT_BACKENDS.
3032 : */
3033 696 : UpdatePMState(PM_WAIT_DEAD_END);
3034 696 : ConfigurePostmasterWaitSet(false);
3035 696 : SignalChildren(SIGQUIT, btmask(B_DEAD_END_BACKEND));
3036 :
3037 : /*
3038 : * We already SIGQUIT'd auxiliary processes (other than
3039 : * logger), if any, when we started immediate shutdown or
3040 : * entered FatalError state.
3041 : */
3042 : }
3043 : else
3044 : {
3045 : /*
3046 : * If we get here, we are proceeding with normal shutdown. All
3047 : * the regular children are gone, and it's time to tell the
3048 : * checkpointer to do a shutdown checkpoint.
3049 : */
3050 : Assert(Shutdown > NoShutdown);
3051 : /* Start the checkpointer if not running */
3052 1128 : if (CheckpointerPMChild == NULL)
3053 0 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3054 : /* And tell it to write the shutdown checkpoint */
3055 1128 : if (CheckpointerPMChild != NULL)
3056 : {
3057 1128 : signal_child(CheckpointerPMChild, SIGINT);
3058 1128 : UpdatePMState(PM_WAIT_XLOG_SHUTDOWN);
3059 : }
3060 : else
3061 : {
3062 : /*
3063 : * If we failed to fork a checkpointer, just shut down.
3064 : * Any required cleanup will happen at next restart. We
3065 : * set FatalError so that an "abnormal shutdown" message
3066 : * gets logged when we exit.
3067 : *
3068 : * We don't consult send_abort_for_crash here, as it's
3069 : * unlikely that dumping cores would illuminate the reason
3070 : * for checkpointer fork failure.
3071 : *
3072 : * XXX: It may be worth to introduce a different PMQUIT
3073 : * value that signals that the cluster is in a bad state,
3074 : * without a process having crashed. But right now this
3075 : * path is very unlikely to be reached, so it isn't
3076 : * obviously worthwhile adding a distinct error message in
3077 : * quickdie().
3078 : */
3079 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3080 : }
3081 : }
3082 : }
3083 : }
3084 :
3085 : /*
3086 : * The state transition from PM_WAIT_XLOG_SHUTDOWN to
3087 : * PM_WAIT_XLOG_ARCHIVAL is in process_pm_pmsignal(), in response to
3088 : * PMSIGNAL_XLOG_IS_SHUTDOWN.
3089 : */
3090 :
3091 52692 : if (pmState == PM_WAIT_XLOG_ARCHIVAL)
3092 : {
3093 : /*
3094 : * PM_WAIT_XLOG_ARCHIVAL state ends when there are no children other
3095 : * than checkpointer, io workers and dead-end children left. There
3096 : * shouldn't be any regular backends left by now anyway; what we're
3097 : * really waiting for is for walsenders and archiver to exit.
3098 : */
3099 1248 : if (CountChildren(btmask_all_except(B_CHECKPOINTER, B_IO_WORKER,
3100 : B_LOGGER, B_DEAD_END_BACKEND)) == 0)
3101 : {
3102 1128 : UpdatePMState(PM_WAIT_IO_WORKERS);
3103 1128 : SignalChildren(SIGUSR2, btmask(B_IO_WORKER));
3104 : }
3105 : }
3106 :
3107 52692 : if (pmState == PM_WAIT_IO_WORKERS)
3108 : {
3109 : /*
3110 : * PM_WAIT_IO_WORKERS state ends when there's only checkpointer and
3111 : * dead-end children left.
3112 : */
3113 4540 : if (io_worker_count == 0)
3114 : {
3115 1128 : UpdatePMState(PM_WAIT_CHECKPOINTER);
3116 :
3117 : /*
3118 : * Now that the processes mentioned above are gone, tell
3119 : * checkpointer to shut down too. That allows checkpointer to
3120 : * perform some last bits of cleanup without other processes
3121 : * interfering.
3122 : */
3123 1128 : if (CheckpointerPMChild != NULL)
3124 1128 : signal_child(CheckpointerPMChild, SIGUSR2);
3125 : }
3126 : }
3127 :
3128 : /*
3129 : * The state transition from PM_WAIT_CHECKPOINTER to PM_WAIT_DEAD_END is
3130 : * in process_pm_child_exit().
3131 : */
3132 :
3133 52692 : if (pmState == PM_WAIT_DEAD_END)
3134 : {
3135 : /*
3136 : * PM_WAIT_DEAD_END state ends when all other children are gone except
3137 : * for the logger. During normal shutdown, all that remains are
3138 : * dead-end backends, but in FatalError processing we jump straight
3139 : * here with more processes remaining. Note that they have already
3140 : * been sent appropriate shutdown signals, either during a normal
3141 : * state transition leading up to PM_WAIT_DEAD_END, or during
3142 : * FatalError processing.
3143 : *
3144 : * The reason we wait is to protect against a new postmaster starting
3145 : * conflicting subprocesses; this isn't an ironclad protection, but it
3146 : * at least helps in the shutdown-and-immediately-restart scenario.
3147 : */
3148 1874 : if (CountChildren(btmask_all_except(B_LOGGER)) == 0)
3149 : {
3150 : /* These other guys should be dead already */
3151 : Assert(StartupPMChild == NULL);
3152 : Assert(WalReceiverPMChild == NULL);
3153 : Assert(WalSummarizerPMChild == NULL);
3154 : Assert(BgWriterPMChild == NULL);
3155 : Assert(CheckpointerPMChild == NULL);
3156 : Assert(WalWriterPMChild == NULL);
3157 : Assert(AutoVacLauncherPMChild == NULL);
3158 : Assert(SlotSyncWorkerPMChild == NULL);
3159 : /* syslogger is not considered here */
3160 1824 : UpdatePMState(PM_NO_CHILDREN);
3161 : }
3162 : }
3163 :
3164 : /*
3165 : * If we've been told to shut down, we exit as soon as there are no
3166 : * remaining children. If there was a crash, cleanup will occur at the
3167 : * next startup. (Before PostgreSQL 8.3, we tried to recover from the
3168 : * crash before exiting, but that seems unwise if we are quitting because
3169 : * we got SIGTERM from init --- there may well not be time for recovery
3170 : * before init decides to SIGKILL us.)
3171 : *
3172 : * Note that the syslogger continues to run. It will exit when it sees
3173 : * EOF on its input pipe, which happens when there are no more upstream
3174 : * processes.
3175 : */
3176 52692 : if (Shutdown > NoShutdown && pmState == PM_NO_CHILDREN)
3177 : {
3178 1808 : if (FatalError)
3179 : {
3180 0 : ereport(LOG, (errmsg("abnormal database system shutdown")));
3181 0 : ExitPostmaster(1);
3182 : }
3183 : else
3184 : {
3185 : /*
3186 : * Normal exit from the postmaster is here. We don't need to log
3187 : * anything here, since the UnlinkLockFiles proc_exit callback
3188 : * will do so, and that should be the last user-visible action.
3189 : */
3190 1808 : ExitPostmaster(0);
3191 : }
3192 : }
3193 :
3194 : /*
3195 : * If the startup process failed, or the user does not want an automatic
3196 : * restart after backend crashes, wait for all non-syslogger children to
3197 : * exit, and then exit postmaster. We don't try to reinitialize when the
3198 : * startup process fails, because more than likely it will just fail again
3199 : * and we will keep trying forever.
3200 : */
3201 50884 : if (pmState == PM_NO_CHILDREN)
3202 : {
3203 16 : if (StartupStatus == STARTUP_CRASHED)
3204 : {
3205 6 : ereport(LOG,
3206 : (errmsg("shutting down due to startup process failure")));
3207 6 : ExitPostmaster(1);
3208 : }
3209 10 : if (!restart_after_crash)
3210 : {
3211 0 : ereport(LOG,
3212 : (errmsg("shutting down because \"restart_after_crash\" is off")));
3213 0 : ExitPostmaster(1);
3214 : }
3215 : }
3216 :
3217 : /*
3218 : * If we need to recover from a crash, wait for all non-syslogger children
3219 : * to exit, then reset shmem and start the startup process.
3220 : */
3221 50878 : if (FatalError && pmState == PM_NO_CHILDREN)
3222 : {
3223 10 : ereport(LOG,
3224 : (errmsg("all server processes terminated; reinitializing")));
3225 :
3226 : /* remove leftover temporary files after a crash */
3227 10 : if (remove_temp_files_after_crash)
3228 8 : RemovePgTempFiles();
3229 :
3230 : /* allow background workers to immediately restart */
3231 10 : ResetBackgroundWorkerCrashTimes();
3232 :
3233 10 : shmem_exit(1);
3234 :
3235 : /* re-read control file into local memory */
3236 10 : LocalProcessControlFile(true);
3237 :
3238 : /* re-create shared memory and semaphores */
3239 10 : CreateSharedMemoryAndSemaphores();
3240 :
3241 10 : UpdatePMState(PM_STARTUP);
3242 :
3243 : /* Make sure we can perform I/O while starting up. */
3244 10 : maybe_adjust_io_workers();
3245 :
3246 10 : StartupPMChild = StartChildProcess(B_STARTUP);
3247 : Assert(StartupPMChild != NULL);
3248 10 : StartupStatus = STARTUP_RUNNING;
3249 : /* crash recovery started, reset SIGKILL flag */
3250 10 : AbortStartTime = 0;
3251 :
3252 : /* start accepting server socket connection events again */
3253 10 : ConfigurePostmasterWaitSet(true);
3254 : }
3255 50878 : }
3256 :
3257 : static const char *
3258 2624 : pmstate_name(PMState state)
3259 : {
3260 : #define PM_TOSTR_CASE(sym) case sym: return #sym
3261 2624 : switch (state)
3262 : {
3263 142 : PM_TOSTR_CASE(PM_INIT);
3264 282 : PM_TOSTR_CASE(PM_STARTUP);
3265 56 : PM_TOSTR_CASE(PM_RECOVERY);
3266 50 : PM_TOSTR_CASE(PM_HOT_STANDBY);
3267 258 : PM_TOSTR_CASE(PM_RUN);
3268 220 : PM_TOSTR_CASE(PM_STOP_BACKENDS);
3269 304 : PM_TOSTR_CASE(PM_WAIT_BACKENDS);
3270 220 : PM_TOSTR_CASE(PM_WAIT_XLOG_SHUTDOWN);
3271 220 : PM_TOSTR_CASE(PM_WAIT_XLOG_ARCHIVAL);
3272 220 : PM_TOSTR_CASE(PM_WAIT_IO_WORKERS);
3273 288 : PM_TOSTR_CASE(PM_WAIT_DEAD_END);
3274 220 : PM_TOSTR_CASE(PM_WAIT_CHECKPOINTER);
3275 144 : PM_TOSTR_CASE(PM_NO_CHILDREN);
3276 : }
3277 : #undef PM_TOSTR_CASE
3278 :
3279 0 : pg_unreachable();
3280 : return ""; /* silence compiler */
3281 : }
3282 :
3283 : /*
3284 : * Simple wrapper for updating pmState. The main reason to have this wrapper
3285 : * is that it makes it easy to log all state transitions.
3286 : */
3287 : static void
3288 15508 : UpdatePMState(PMState newState)
3289 : {
3290 15508 : elog(DEBUG1, "updating PMState from %s to %s",
3291 : pmstate_name(pmState), pmstate_name(newState));
3292 15508 : pmState = newState;
3293 15508 : }
3294 :
3295 : /*
3296 : * Launch background processes after state change, or relaunch after an
3297 : * existing process has exited.
3298 : *
3299 : * Check the current pmState and the status of any background processes. If
3300 : * there are any background processes missing that should be running in the
3301 : * current state, but are not, launch them.
3302 : */
3303 : static void
3304 254040 : LaunchMissingBackgroundProcesses(void)
3305 : {
3306 : /* Syslogger is active in all states */
3307 254040 : if (SysLoggerPMChild == NULL && Logging_collector)
3308 0 : StartSysLogger();
3309 :
3310 : /*
3311 : * The number of configured workers might have changed, or a prior start
3312 : * of a worker might have failed. Check if we need to start/stop any
3313 : * workers.
3314 : *
3315 : * A config file change will always lead to this function being called, so
3316 : * we always will process the config change in a timely manner.
3317 : */
3318 254040 : maybe_adjust_io_workers();
3319 :
3320 : /*
3321 : * The checkpointer and the background writer are active from the start,
3322 : * until shutdown is initiated.
3323 : *
3324 : * (If the checkpointer is not running when we enter the
3325 : * PM_WAIT_XLOG_SHUTDOWN state, it is launched one more time to perform
3326 : * the shutdown checkpoint. That's done in PostmasterStateMachine(), not
3327 : * here.)
3328 : */
3329 254040 : if (pmState == PM_RUN || pmState == PM_RECOVERY ||
3330 18762 : pmState == PM_HOT_STANDBY || pmState == PM_STARTUP)
3331 : {
3332 240116 : if (CheckpointerPMChild == NULL)
3333 10 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3334 240116 : if (BgWriterPMChild == NULL)
3335 10 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
3336 : }
3337 :
3338 : /*
3339 : * WAL writer is needed only in normal operation (else we cannot be
3340 : * writing any new WAL).
3341 : */
3342 254040 : if (WalWriterPMChild == NULL && pmState == PM_RUN)
3343 1608 : WalWriterPMChild = StartChildProcess(B_WAL_WRITER);
3344 :
3345 : /*
3346 : * We don't want autovacuum to run in binary upgrade mode because
3347 : * autovacuum might update relfrozenxid for empty tables before the
3348 : * physical files are put in place.
3349 : */
3350 275334 : if (!IsBinaryUpgrade && AutoVacLauncherPMChild == NULL &&
3351 29726 : (AutoVacuumingActive() || start_autovac_launcher) &&
3352 12862 : pmState == PM_RUN)
3353 : {
3354 1344 : AutoVacLauncherPMChild = StartChildProcess(B_AUTOVAC_LAUNCHER);
3355 1344 : if (AutoVacLauncherPMChild != NULL)
3356 1344 : start_autovac_launcher = false; /* signal processed */
3357 : }
3358 :
3359 : /*
3360 : * If WAL archiving is enabled always, we are allowed to start archiver
3361 : * even during recovery.
3362 : */
3363 254040 : if (PgArchPMChild == NULL &&
3364 251310 : ((XLogArchivingActive() && pmState == PM_RUN) ||
3365 251310 : (XLogArchivingAlways() && (pmState == PM_RECOVERY || pmState == PM_HOT_STANDBY))) &&
3366 108 : PgArchCanRestart())
3367 108 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3368 :
3369 : /*
3370 : * If we need to start a slot sync worker, try to do that now
3371 : *
3372 : * We allow to start the slot sync worker when we are on a hot standby,
3373 : * fast or immediate shutdown is not in progress, slot sync parameters are
3374 : * configured correctly, and it is the first time of worker's launch, or
3375 : * enough time has passed since the worker was launched last.
3376 : */
3377 254040 : if (SlotSyncWorkerPMChild == NULL && pmState == PM_HOT_STANDBY &&
3378 4326 : Shutdown <= SmartShutdown && sync_replication_slots &&
3379 42 : ValidateSlotSyncParams(LOG) && SlotSyncWorkerCanRestart())
3380 8 : SlotSyncWorkerPMChild = StartChildProcess(B_SLOTSYNC_WORKER);
3381 :
3382 : /*
3383 : * If we need to start a WAL receiver, try to do that now
3384 : *
3385 : * Note: if a walreceiver process is already running, it might seem that
3386 : * we should clear WalReceiverRequested. However, there's a race
3387 : * condition if the walreceiver terminates and the startup process
3388 : * immediately requests a new one: it's quite possible to get the signal
3389 : * for the request before reaping the dead walreceiver process. Better to
3390 : * risk launching an extra walreceiver than to miss launching one we need.
3391 : * (The walreceiver code has logic to recognize that it should go away if
3392 : * not needed.)
3393 : */
3394 254040 : if (WalReceiverRequested)
3395 : {
3396 778 : if (WalReceiverPMChild == NULL &&
3397 536 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
3398 534 : pmState == PM_HOT_STANDBY) &&
3399 496 : Shutdown <= SmartShutdown)
3400 : {
3401 496 : WalReceiverPMChild = StartChildProcess(B_WAL_RECEIVER);
3402 496 : if (WalReceiverPMChild != NULL)
3403 496 : WalReceiverRequested = false;
3404 : /* else leave the flag set, so we'll try again later */
3405 : }
3406 : }
3407 :
3408 : /* If we need to start a WAL summarizer, try to do that now */
3409 254040 : if (summarize_wal && WalSummarizerPMChild == NULL &&
3410 182 : (pmState == PM_RUN || pmState == PM_HOT_STANDBY) &&
3411 42 : Shutdown <= SmartShutdown)
3412 42 : WalSummarizerPMChild = StartChildProcess(B_WAL_SUMMARIZER);
3413 :
3414 : /* Get other worker processes running, if needed */
3415 254040 : if (StartWorkerNeeded || HaveCrashedWorker)
3416 14978 : maybe_start_bgworkers();
3417 254040 : }
3418 :
3419 : /*
3420 : * Return string representation of signal.
3421 : *
3422 : * Because this is only implemented for signals we already rely on in this
3423 : * file we don't need to deal with unimplemented or same-numeric-value signals
3424 : * (as we'd e.g. have to for EWOULDBLOCK / EAGAIN).
3425 : */
3426 : static const char *
3427 36 : pm_signame(int signal)
3428 : {
3429 : #define PM_TOSTR_CASE(sym) case sym: return #sym
3430 36 : switch (signal)
3431 : {
3432 0 : PM_TOSTR_CASE(SIGABRT);
3433 0 : PM_TOSTR_CASE(SIGCHLD);
3434 0 : PM_TOSTR_CASE(SIGHUP);
3435 4 : PM_TOSTR_CASE(SIGINT);
3436 0 : PM_TOSTR_CASE(SIGKILL);
3437 0 : PM_TOSTR_CASE(SIGQUIT);
3438 22 : PM_TOSTR_CASE(SIGTERM);
3439 0 : PM_TOSTR_CASE(SIGUSR1);
3440 10 : PM_TOSTR_CASE(SIGUSR2);
3441 0 : default:
3442 : /* all signals sent by postmaster should be listed here */
3443 : Assert(false);
3444 0 : return "(unknown)";
3445 : }
3446 : #undef PM_TOSTR_CASE
3447 :
3448 : return ""; /* silence compiler */
3449 : }
3450 :
3451 : /*
3452 : * Send a signal to a postmaster child process
3453 : *
3454 : * On systems that have setsid(), each child process sets itself up as a
3455 : * process group leader. For signals that are generally interpreted in the
3456 : * appropriate fashion, we signal the entire process group not just the
3457 : * direct child process. This allows us to, for example, SIGQUIT a blocked
3458 : * archive_recovery script, or SIGINT a script being run by a backend via
3459 : * system().
3460 : *
3461 : * There is a race condition for recently-forked children: they might not
3462 : * have executed setsid() yet. So we signal the child directly as well as
3463 : * the group. We assume such a child will handle the signal before trying
3464 : * to spawn any grandchild processes. We also assume that signaling the
3465 : * child twice will not cause any problems.
3466 : */
3467 : static void
3468 23270 : signal_child(PMChild *pmchild, int signal)
3469 : {
3470 23270 : pid_t pid = pmchild->pid;
3471 :
3472 23270 : ereport(DEBUG3,
3473 : (errmsg_internal("sending signal %d/%s to %s process with pid %d",
3474 : signal, pm_signame(signal),
3475 : GetBackendTypeDesc(pmchild->bkend_type),
3476 : (int) pmchild->pid)));
3477 :
3478 23270 : if (kill(pid, signal) < 0)
3479 0 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) pid, signal);
3480 : #ifdef HAVE_SETSID
3481 23270 : switch (signal)
3482 : {
3483 13210 : case SIGINT:
3484 : case SIGTERM:
3485 : case SIGQUIT:
3486 : case SIGKILL:
3487 : case SIGABRT:
3488 13210 : if (kill(-pid, signal) < 0)
3489 8 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) (-pid), signal);
3490 13210 : break;
3491 10060 : default:
3492 10060 : break;
3493 : }
3494 : #endif
3495 23270 : }
3496 :
3497 : /*
3498 : * Send a signal to the targeted children.
3499 : */
3500 : static bool
3501 6200 : SignalChildren(int signal, BackendTypeMask targetMask)
3502 : {
3503 : dlist_iter iter;
3504 6200 : bool signaled = false;
3505 :
3506 35250 : dlist_foreach(iter, &ActiveChildList)
3507 : {
3508 29050 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3509 :
3510 : /*
3511 : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3512 : * if any B_BACKEND backends have recently announced that they are
3513 : * actually WAL senders.
3514 : */
3515 29050 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3516 14914 : bp->bkend_type == B_BACKEND)
3517 : {
3518 1418 : if (IsPostmasterChildWalSender(bp->child_slot))
3519 82 : bp->bkend_type = B_WAL_SENDER;
3520 : }
3521 :
3522 29050 : if (!btmask_contains(targetMask, bp->bkend_type))
3523 10456 : continue;
3524 :
3525 18594 : signal_child(bp, signal);
3526 18594 : signaled = true;
3527 : }
3528 6200 : return signaled;
3529 : }
3530 :
3531 : /*
3532 : * Send a termination signal to children. This considers all of our children
3533 : * processes, except syslogger.
3534 : */
3535 : static void
3536 696 : TerminateChildren(int signal)
3537 : {
3538 696 : SignalChildren(signal, btmask_all_except(B_LOGGER));
3539 696 : if (StartupPMChild != NULL)
3540 : {
3541 90 : if (signal == SIGQUIT || signal == SIGKILL || signal == SIGABRT)
3542 90 : StartupStatus = STARTUP_SIGNALED;
3543 : }
3544 696 : }
3545 :
3546 : /*
3547 : * BackendStartup -- start backend process
3548 : *
3549 : * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
3550 : *
3551 : * Note: if you change this code, also consider StartAutovacuumWorker and
3552 : * StartBackgroundWorker.
3553 : */
3554 : static int
3555 28946 : BackendStartup(ClientSocket *client_sock)
3556 : {
3557 28946 : PMChild *bn = NULL;
3558 : pid_t pid;
3559 : BackendStartupData startup_data;
3560 : CAC_state cac;
3561 :
3562 : /*
3563 : * Capture time that Postmaster got a socket from accept (for logging
3564 : * connection establishment and setup total duration).
3565 : */
3566 28946 : startup_data.socket_created = GetCurrentTimestamp();
3567 :
3568 : /*
3569 : * Allocate and assign the child slot. Note we must do this before
3570 : * forking, so that we can handle failures (out of memory or child-process
3571 : * slots) cleanly.
3572 : */
3573 28946 : cac = canAcceptConnections(B_BACKEND);
3574 28946 : if (cac == CAC_OK)
3575 : {
3576 : /* Can change later to B_WAL_SENDER */
3577 28420 : bn = AssignPostmasterChildSlot(B_BACKEND);
3578 28420 : if (!bn)
3579 : {
3580 : /*
3581 : * Too many regular child processes; launch a dead-end child
3582 : * process instead.
3583 : */
3584 56 : cac = CAC_TOOMANY;
3585 : }
3586 : }
3587 28946 : if (!bn)
3588 : {
3589 582 : bn = AllocDeadEndChild();
3590 582 : if (!bn)
3591 : {
3592 0 : ereport(LOG,
3593 : (errcode(ERRCODE_OUT_OF_MEMORY),
3594 : errmsg("out of memory")));
3595 0 : return STATUS_ERROR;
3596 : }
3597 : }
3598 :
3599 : /* Pass down canAcceptConnections state */
3600 28946 : startup_data.canAcceptConnections = cac;
3601 28946 : bn->rw = NULL;
3602 :
3603 : /* Hasn't asked to be notified about any bgworkers yet */
3604 28946 : bn->bgworker_notify = false;
3605 :
3606 28946 : pid = postmaster_child_launch(bn->bkend_type, bn->child_slot,
3607 : &startup_data, sizeof(startup_data),
3608 : client_sock);
3609 28946 : if (pid < 0)
3610 : {
3611 : /* in parent, fork failed */
3612 0 : int save_errno = errno;
3613 :
3614 0 : (void) ReleasePostmasterChildSlot(bn);
3615 0 : errno = save_errno;
3616 0 : ereport(LOG,
3617 : (errmsg("could not fork new process for connection: %m")));
3618 0 : report_fork_failure_to_client(client_sock, save_errno);
3619 0 : return STATUS_ERROR;
3620 : }
3621 :
3622 : /* in parent, successful fork */
3623 28946 : ereport(DEBUG2,
3624 : (errmsg_internal("forked new %s, pid=%d socket=%d",
3625 : GetBackendTypeDesc(bn->bkend_type),
3626 : (int) pid, (int) client_sock->sock)));
3627 :
3628 : /*
3629 : * Everything's been successful, it's safe to add this backend to our list
3630 : * of backends.
3631 : */
3632 28946 : bn->pid = pid;
3633 28946 : return STATUS_OK;
3634 : }
3635 :
3636 : /*
3637 : * Try to report backend fork() failure to client before we close the
3638 : * connection. Since we do not care to risk blocking the postmaster on
3639 : * this connection, we set the connection to non-blocking and try only once.
3640 : *
3641 : * This is grungy special-purpose code; we cannot use backend libpq since
3642 : * it's not up and running.
3643 : */
3644 : static void
3645 0 : report_fork_failure_to_client(ClientSocket *client_sock, int errnum)
3646 : {
3647 : char buffer[1000];
3648 : int rc;
3649 :
3650 : /* Format the error message packet (always V2 protocol) */
3651 0 : snprintf(buffer, sizeof(buffer), "E%s%s\n",
3652 : _("could not fork new process for connection: "),
3653 : strerror(errnum));
3654 :
3655 : /* Set port to non-blocking. Don't do send() if this fails */
3656 0 : if (!pg_set_noblock(client_sock->sock))
3657 0 : return;
3658 :
3659 : /* We'll retry after EINTR, but ignore all other failures */
3660 : do
3661 : {
3662 0 : rc = send(client_sock->sock, buffer, strlen(buffer) + 1, 0);
3663 0 : } while (rc < 0 && errno == EINTR);
3664 : }
3665 :
3666 : /*
3667 : * ExitPostmaster -- cleanup
3668 : *
3669 : * Do NOT call exit() directly --- always go through here!
3670 : */
3671 : static void
3672 1824 : ExitPostmaster(int status)
3673 : {
3674 : #ifdef HAVE_PTHREAD_IS_THREADED_NP
3675 :
3676 : /*
3677 : * There is no known cause for a postmaster to become multithreaded after
3678 : * startup. However, we might reach here via an error exit before
3679 : * reaching the test in PostmasterMain, so provide the same hint as there.
3680 : * This message uses LOG level, because an unclean shutdown at this point
3681 : * would usually not look much different from a clean shutdown.
3682 : */
3683 : if (pthread_is_threaded_np() != 0)
3684 : ereport(LOG,
3685 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3686 : errmsg("postmaster became multithreaded"),
3687 : errhint("Set the LC_ALL environment variable to a valid locale.")));
3688 : #endif
3689 :
3690 : /* should cleanup shared memory and kill all backends */
3691 :
3692 : /*
3693 : * Not sure of the semantics here. When the Postmaster dies, should the
3694 : * backends all be killed? probably not.
3695 : *
3696 : * MUST -- vadim 05-10-1999
3697 : */
3698 :
3699 1824 : proc_exit(status);
3700 : }
3701 :
3702 : /*
3703 : * Handle pmsignal conditions representing requests from backends,
3704 : * and check for promote and logrotate requests from pg_ctl.
3705 : */
3706 : static void
3707 177628 : process_pm_pmsignal(void)
3708 : {
3709 177628 : bool request_state_update = false;
3710 :
3711 177628 : pending_pm_pmsignal = false;
3712 :
3713 177628 : ereport(DEBUG2,
3714 : (errmsg_internal("postmaster received pmsignal signal")));
3715 :
3716 : /*
3717 : * RECOVERY_STARTED and BEGIN_HOT_STANDBY signals are ignored in
3718 : * unexpected states. If the startup process quickly starts up, completes
3719 : * recovery, exits, we might process the death of the startup process
3720 : * first. We don't want to go back to recovery in that case.
3721 : */
3722 177628 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_STARTED) &&
3723 518 : pmState == PM_STARTUP && Shutdown == NoShutdown)
3724 : {
3725 : /* WAL redo has started. We're out of reinitialization. */
3726 518 : FatalError = false;
3727 518 : AbortStartTime = 0;
3728 518 : reachedConsistency = false;
3729 :
3730 : /*
3731 : * Start the archiver if we're responsible for (re-)archiving received
3732 : * files.
3733 : */
3734 : Assert(PgArchPMChild == NULL);
3735 518 : if (XLogArchivingAlways())
3736 6 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3737 :
3738 : /*
3739 : * If we aren't planning to enter hot standby mode later, treat
3740 : * RECOVERY_STARTED as meaning we're out of startup, and report status
3741 : * accordingly.
3742 : */
3743 518 : if (!EnableHotStandby)
3744 : {
3745 4 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STANDBY);
3746 : #ifdef USE_SYSTEMD
3747 : sd_notify(0, "READY=1");
3748 : #endif
3749 : }
3750 :
3751 518 : UpdatePMState(PM_RECOVERY);
3752 : }
3753 :
3754 177628 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_CONSISTENT) &&
3755 336 : pmState == PM_RECOVERY && Shutdown == NoShutdown)
3756 : {
3757 336 : reachedConsistency = true;
3758 : }
3759 :
3760 177628 : if (CheckPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY) &&
3761 318 : (pmState == PM_RECOVERY && Shutdown == NoShutdown))
3762 : {
3763 318 : ereport(LOG,
3764 : (errmsg("database system is ready to accept read-only connections")));
3765 :
3766 : /* Report status */
3767 318 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
3768 : #ifdef USE_SYSTEMD
3769 : sd_notify(0, "READY=1");
3770 : #endif
3771 :
3772 318 : UpdatePMState(PM_HOT_STANDBY);
3773 318 : connsAllowed = true;
3774 :
3775 : /* Some workers may be scheduled to start now */
3776 318 : StartWorkerNeeded = true;
3777 : }
3778 :
3779 : /* Process background worker state changes. */
3780 177628 : if (CheckPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE))
3781 : {
3782 : /* Accept new worker requests only if not stopping. */
3783 2704 : BackgroundWorkerStateChange(pmState < PM_STOP_BACKENDS);
3784 2704 : StartWorkerNeeded = true;
3785 : }
3786 :
3787 : /* Tell syslogger to rotate logfile if requested */
3788 177628 : if (SysLoggerPMChild != NULL)
3789 : {
3790 4 : if (CheckLogrotateSignal())
3791 : {
3792 2 : signal_child(SysLoggerPMChild, SIGUSR1);
3793 2 : RemoveLogrotateSignalFiles();
3794 : }
3795 2 : else if (CheckPostmasterSignal(PMSIGNAL_ROTATE_LOGFILE))
3796 : {
3797 0 : signal_child(SysLoggerPMChild, SIGUSR1);
3798 : }
3799 : }
3800 :
3801 177628 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER) &&
3802 167926 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3803 : {
3804 : /*
3805 : * Start one iteration of the autovacuum daemon, even if autovacuuming
3806 : * is nominally not enabled. This is so we can have an active defense
3807 : * against transaction ID wraparound. We set a flag for the main loop
3808 : * to do it rather than trying to do it here --- this is because the
3809 : * autovac process itself may send the signal, and we want to handle
3810 : * that by launching another iteration as soon as the current one
3811 : * completes.
3812 : */
3813 167926 : start_autovac_launcher = true;
3814 : }
3815 :
3816 177628 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER) &&
3817 2294 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3818 : {
3819 : /* The autovacuum launcher wants us to start a worker process. */
3820 2294 : StartAutovacuumWorker();
3821 : }
3822 :
3823 177628 : if (CheckPostmasterSignal(PMSIGNAL_START_WALRECEIVER))
3824 : {
3825 : /* Startup Process wants us to start the walreceiver process. */
3826 516 : WalReceiverRequested = true;
3827 : }
3828 :
3829 177628 : if (CheckPostmasterSignal(PMSIGNAL_XLOG_IS_SHUTDOWN))
3830 : {
3831 : /* Checkpointer completed the shutdown checkpoint */
3832 1128 : if (pmState == PM_WAIT_XLOG_SHUTDOWN)
3833 : {
3834 : /*
3835 : * If we have an archiver subprocess, tell it to do a last archive
3836 : * cycle and quit. Likewise, if we have walsender processes, tell
3837 : * them to send any remaining WAL and quit.
3838 : */
3839 : Assert(Shutdown > NoShutdown);
3840 :
3841 : /* Waken archiver for the last time */
3842 1128 : if (PgArchPMChild != NULL)
3843 34 : signal_child(PgArchPMChild, SIGUSR2);
3844 :
3845 : /*
3846 : * Waken walsenders for the last time. No regular backends should
3847 : * be around anymore.
3848 : */
3849 1128 : SignalChildren(SIGUSR2, btmask(B_WAL_SENDER));
3850 :
3851 1128 : UpdatePMState(PM_WAIT_XLOG_ARCHIVAL);
3852 : }
3853 0 : else if (!FatalError && Shutdown != ImmediateShutdown)
3854 : {
3855 : /*
3856 : * Checkpointer only ought to perform the shutdown checkpoint
3857 : * during shutdown. If somehow checkpointer did so in another
3858 : * situation, we have no choice but to crash-restart.
3859 : *
3860 : * It's possible however that we get PMSIGNAL_XLOG_IS_SHUTDOWN
3861 : * outside of PM_WAIT_XLOG_SHUTDOWN if an orderly shutdown was
3862 : * "interrupted" by a crash or an immediate shutdown.
3863 : */
3864 0 : ereport(LOG,
3865 : (errmsg("WAL was shut down unexpectedly")));
3866 :
3867 : /*
3868 : * Doesn't seem likely to help to take send_abort_for_crash into
3869 : * account here.
3870 : */
3871 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3872 : }
3873 :
3874 : /*
3875 : * Need to run PostmasterStateMachine() to check if we already can go
3876 : * to the next state.
3877 : */
3878 1128 : request_state_update = true;
3879 : }
3880 :
3881 : /*
3882 : * Try to advance postmaster's state machine, if a child requests it.
3883 : */
3884 177628 : if (CheckPostmasterSignal(PMSIGNAL_ADVANCE_STATE_MACHINE))
3885 : {
3886 2526 : request_state_update = true;
3887 : }
3888 :
3889 : /*
3890 : * Be careful about the order of this action relative to this function's
3891 : * other actions. Generally, this should be after other actions, in case
3892 : * they have effects PostmasterStateMachine would need to know about.
3893 : * However, we should do it before the CheckPromoteSignal step, which
3894 : * cannot have any (immediate) effect on the state machine, but does
3895 : * depend on what state we're in now.
3896 : */
3897 177628 : if (request_state_update)
3898 : {
3899 3654 : PostmasterStateMachine();
3900 : }
3901 :
3902 177628 : if (StartupPMChild != NULL &&
3903 1666 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
3904 2708 : pmState == PM_HOT_STANDBY) &&
3905 1666 : CheckPromoteSignal())
3906 : {
3907 : /*
3908 : * Tell startup process to finish recovery.
3909 : *
3910 : * Leave the promote signal file in place and let the Startup process
3911 : * do the unlink.
3912 : */
3913 96 : signal_child(StartupPMChild, SIGUSR2);
3914 : }
3915 177628 : }
3916 :
3917 : /*
3918 : * Dummy signal handler
3919 : *
3920 : * We use this for signals that we don't actually use in the postmaster,
3921 : * but we do use in backends. If we were to SIG_IGN such signals in the
3922 : * postmaster, then a newly started backend might drop a signal that arrives
3923 : * before it's able to reconfigure its signal processing. (See notes in
3924 : * tcop/postgres.c.)
3925 : */
3926 : static void
3927 0 : dummy_handler(SIGNAL_ARGS)
3928 : {
3929 0 : }
3930 :
3931 : /*
3932 : * Count up number of child processes of specified types.
3933 : */
3934 : static int
3935 13128 : CountChildren(BackendTypeMask targetMask)
3936 : {
3937 : dlist_iter iter;
3938 13128 : int cnt = 0;
3939 :
3940 78166 : dlist_foreach(iter, &ActiveChildList)
3941 : {
3942 65038 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3943 :
3944 : /*
3945 : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3946 : * if any B_BACKEND backends have recently announced that they are
3947 : * actually WAL senders.
3948 : */
3949 65038 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3950 46742 : bp->bkend_type == B_BACKEND)
3951 : {
3952 3074 : if (IsPostmasterChildWalSender(bp->child_slot))
3953 0 : bp->bkend_type = B_WAL_SENDER;
3954 : }
3955 :
3956 65038 : if (!btmask_contains(targetMask, bp->bkend_type))
3957 34024 : continue;
3958 :
3959 31014 : ereport(DEBUG4,
3960 : (errmsg_internal("%s process %d is still running",
3961 : GetBackendTypeDesc(bp->bkend_type), (int) bp->pid)));
3962 :
3963 31014 : cnt++;
3964 : }
3965 13128 : return cnt;
3966 : }
3967 :
3968 :
3969 : /*
3970 : * StartChildProcess -- start an auxiliary process for the postmaster
3971 : *
3972 : * "type" determines what kind of child will be started. All child types
3973 : * initially go to AuxiliaryProcessMain, which will handle common setup.
3974 : *
3975 : * Return value of StartChildProcess is subprocess' PMChild entry, or NULL on
3976 : * failure.
3977 : */
3978 : static PMChild *
3979 16952 : StartChildProcess(BackendType type)
3980 : {
3981 : PMChild *pmchild;
3982 : pid_t pid;
3983 :
3984 16952 : pmchild = AssignPostmasterChildSlot(type);
3985 16952 : if (!pmchild)
3986 : {
3987 0 : if (type == B_AUTOVAC_WORKER)
3988 0 : ereport(LOG,
3989 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
3990 : errmsg("no slot available for new autovacuum worker process")));
3991 : else
3992 : {
3993 : /* shouldn't happen because we allocate enough slots */
3994 0 : elog(LOG, "no postmaster child slot available for aux process");
3995 : }
3996 0 : return NULL;
3997 : }
3998 :
3999 16952 : pid = postmaster_child_launch(type, pmchild->child_slot, NULL, 0, NULL);
4000 16952 : if (pid < 0)
4001 : {
4002 : /* in parent, fork failed */
4003 0 : ReleasePostmasterChildSlot(pmchild);
4004 0 : ereport(LOG,
4005 : (errmsg("could not fork \"%s\" process: %m", PostmasterChildName(type))));
4006 :
4007 : /*
4008 : * fork failure is fatal during startup, but there's no need to choke
4009 : * immediately if starting other child types fails.
4010 : */
4011 0 : if (type == B_STARTUP)
4012 0 : ExitPostmaster(1);
4013 0 : return NULL;
4014 : }
4015 :
4016 : /* in parent, successful fork */
4017 16952 : pmchild->pid = pid;
4018 16952 : return pmchild;
4019 : }
4020 :
4021 : /*
4022 : * StartSysLogger -- start the syslogger process
4023 : */
4024 : void
4025 2 : StartSysLogger(void)
4026 : {
4027 : Assert(SysLoggerPMChild == NULL);
4028 :
4029 2 : SysLoggerPMChild = AssignPostmasterChildSlot(B_LOGGER);
4030 2 : if (!SysLoggerPMChild)
4031 0 : elog(PANIC, "no postmaster child slot available for syslogger");
4032 2 : SysLoggerPMChild->pid = SysLogger_Start(SysLoggerPMChild->child_slot);
4033 2 : if (SysLoggerPMChild->pid == 0)
4034 : {
4035 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
4036 0 : SysLoggerPMChild = NULL;
4037 : }
4038 2 : }
4039 :
4040 : /*
4041 : * StartAutovacuumWorker
4042 : * Start an autovac worker process.
4043 : *
4044 : * This function is here because it enters the resulting PID into the
4045 : * postmaster's private backends list.
4046 : *
4047 : * NB -- this code very roughly matches BackendStartup.
4048 : */
4049 : static void
4050 2294 : StartAutovacuumWorker(void)
4051 : {
4052 : PMChild *bn;
4053 :
4054 : /*
4055 : * If not in condition to run a process, don't try, but handle it like a
4056 : * fork failure. This does not normally happen, since the signal is only
4057 : * supposed to be sent by autovacuum launcher when it's OK to do it, but
4058 : * we have to check to avoid race-condition problems during DB state
4059 : * changes.
4060 : */
4061 2294 : if (canAcceptConnections(B_AUTOVAC_WORKER) == CAC_OK)
4062 : {
4063 2294 : bn = StartChildProcess(B_AUTOVAC_WORKER);
4064 2294 : if (bn)
4065 : {
4066 2294 : bn->bgworker_notify = false;
4067 2294 : bn->rw = NULL;
4068 2294 : return;
4069 : }
4070 : else
4071 : {
4072 : /*
4073 : * fork failed, fall through to report -- actual error message was
4074 : * logged by StartChildProcess
4075 : */
4076 : }
4077 : }
4078 :
4079 : /*
4080 : * Report the failure to the launcher, if it's running. (If it's not, we
4081 : * might not even be connected to shared memory, so don't try to call
4082 : * AutoVacWorkerFailed.) Note that we also need to signal it so that it
4083 : * responds to the condition, but we don't do that here, instead waiting
4084 : * for ServerLoop to do it. This way we avoid a ping-pong signaling in
4085 : * quick succession between the autovac launcher and postmaster in case
4086 : * things get ugly.
4087 : */
4088 0 : if (AutoVacLauncherPMChild != NULL)
4089 : {
4090 0 : AutoVacWorkerFailed();
4091 0 : avlauncher_needs_signal = true;
4092 : }
4093 : }
4094 :
4095 :
4096 : /*
4097 : * Create the opts file
4098 : */
4099 : static bool
4100 1820 : CreateOptsFile(int argc, char *argv[], char *fullprogname)
4101 : {
4102 : FILE *fp;
4103 : int i;
4104 :
4105 : #define OPTS_FILE "postmaster.opts"
4106 :
4107 1820 : if ((fp = fopen(OPTS_FILE, "w")) == NULL)
4108 : {
4109 0 : ereport(LOG,
4110 : (errcode_for_file_access(),
4111 : errmsg("could not create file \"%s\": %m", OPTS_FILE)));
4112 0 : return false;
4113 : }
4114 :
4115 1820 : fprintf(fp, "%s", fullprogname);
4116 9546 : for (i = 1; i < argc; i++)
4117 7726 : fprintf(fp, " \"%s\"", argv[i]);
4118 1820 : fputs("\n", fp);
4119 :
4120 1820 : if (fclose(fp))
4121 : {
4122 0 : ereport(LOG,
4123 : (errcode_for_file_access(),
4124 : errmsg("could not write file \"%s\": %m", OPTS_FILE)));
4125 0 : return false;
4126 : }
4127 :
4128 1820 : return true;
4129 : }
4130 :
4131 :
4132 : /*
4133 : * Start a new bgworker.
4134 : * Starting time conditions must have been checked already.
4135 : *
4136 : * Returns true on success, false on failure.
4137 : * In either case, update the RegisteredBgWorker's state appropriately.
4138 : *
4139 : * NB -- this code very roughly matches BackendStartup.
4140 : */
4141 : static bool
4142 5558 : StartBackgroundWorker(RegisteredBgWorker *rw)
4143 : {
4144 : PMChild *bn;
4145 : pid_t worker_pid;
4146 :
4147 : Assert(rw->rw_pid == 0);
4148 :
4149 : /*
4150 : * Allocate and assign the child slot. Note we must do this before
4151 : * forking, so that we can handle failures (out of memory or child-process
4152 : * slots) cleanly.
4153 : *
4154 : * Treat failure as though the worker had crashed. That way, the
4155 : * postmaster will wait a bit before attempting to start it again; if we
4156 : * tried again right away, most likely we'd find ourselves hitting the
4157 : * same resource-exhaustion condition.
4158 : */
4159 5558 : bn = AssignPostmasterChildSlot(B_BG_WORKER);
4160 5558 : if (bn == NULL)
4161 : {
4162 0 : ereport(LOG,
4163 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
4164 : errmsg("no slot available for new background worker process")));
4165 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4166 0 : return false;
4167 : }
4168 5558 : bn->rw = rw;
4169 5558 : bn->bkend_type = B_BG_WORKER;
4170 5558 : bn->bgworker_notify = false;
4171 :
4172 5558 : ereport(DEBUG1,
4173 : (errmsg_internal("starting background worker process \"%s\"",
4174 : rw->rw_worker.bgw_name)));
4175 :
4176 5558 : worker_pid = postmaster_child_launch(B_BG_WORKER, bn->child_slot,
4177 5558 : &rw->rw_worker, sizeof(BackgroundWorker), NULL);
4178 5558 : if (worker_pid == -1)
4179 : {
4180 : /* in postmaster, fork failed ... */
4181 0 : ereport(LOG,
4182 : (errmsg("could not fork background worker process: %m")));
4183 : /* undo what AssignPostmasterChildSlot did */
4184 0 : ReleasePostmasterChildSlot(bn);
4185 :
4186 : /* mark entry as crashed, so we'll try again later */
4187 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4188 0 : return false;
4189 : }
4190 :
4191 : /* in postmaster, fork successful ... */
4192 5558 : rw->rw_pid = worker_pid;
4193 5558 : bn->pid = rw->rw_pid;
4194 5558 : ReportBackgroundWorkerPID(rw);
4195 5558 : return true;
4196 : }
4197 :
4198 : /*
4199 : * Does the current postmaster state require starting a worker with the
4200 : * specified start_time?
4201 : */
4202 : static bool
4203 7584 : bgworker_should_start_now(BgWorkerStartTime start_time)
4204 : {
4205 7584 : switch (pmState)
4206 : {
4207 0 : case PM_NO_CHILDREN:
4208 : case PM_WAIT_CHECKPOINTER:
4209 : case PM_WAIT_DEAD_END:
4210 : case PM_WAIT_XLOG_ARCHIVAL:
4211 : case PM_WAIT_XLOG_SHUTDOWN:
4212 : case PM_WAIT_IO_WORKERS:
4213 : case PM_WAIT_BACKENDS:
4214 : case PM_STOP_BACKENDS:
4215 0 : break;
4216 :
4217 5558 : case PM_RUN:
4218 5558 : if (start_time == BgWorkerStart_RecoveryFinished)
4219 2664 : return true;
4220 : /* fall through */
4221 :
4222 : case PM_HOT_STANDBY:
4223 3204 : if (start_time == BgWorkerStart_ConsistentState)
4224 2894 : return true;
4225 : /* fall through */
4226 :
4227 : case PM_RECOVERY:
4228 : case PM_STARTUP:
4229 : case PM_INIT:
4230 2026 : if (start_time == BgWorkerStart_PostmasterStart)
4231 0 : return true;
4232 : /* fall through */
4233 : }
4234 :
4235 2026 : return false;
4236 : }
4237 :
4238 : /*
4239 : * If the time is right, start background worker(s).
4240 : *
4241 : * As a side effect, the bgworker control variables are set or reset
4242 : * depending on whether more workers may need to be started.
4243 : *
4244 : * We limit the number of workers started per call, to avoid consuming the
4245 : * postmaster's attention for too long when many such requests are pending.
4246 : * As long as StartWorkerNeeded is true, ServerLoop will not block and will
4247 : * call this function again after dealing with any other issues.
4248 : */
4249 : static void
4250 16798 : maybe_start_bgworkers(void)
4251 : {
4252 : #define MAX_BGWORKERS_TO_LAUNCH 100
4253 16798 : int num_launched = 0;
4254 16798 : TimestampTz now = 0;
4255 : dlist_mutable_iter iter;
4256 :
4257 : /*
4258 : * During crash recovery, we have no need to be called until the state
4259 : * transition out of recovery.
4260 : */
4261 16798 : if (FatalError)
4262 : {
4263 0 : StartWorkerNeeded = false;
4264 0 : HaveCrashedWorker = false;
4265 0 : return;
4266 : }
4267 :
4268 : /* Don't need to be called again unless we find a reason for it below */
4269 16798 : StartWorkerNeeded = false;
4270 16798 : HaveCrashedWorker = false;
4271 :
4272 44846 : dlist_foreach_modify(iter, &BackgroundWorkerList)
4273 : {
4274 : RegisteredBgWorker *rw;
4275 :
4276 28048 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
4277 :
4278 : /* ignore if already running */
4279 28048 : if (rw->rw_pid != 0)
4280 14046 : continue;
4281 :
4282 : /* if marked for death, clean up and remove from list */
4283 14002 : if (rw->rw_terminate)
4284 : {
4285 0 : ForgetBackgroundWorker(rw);
4286 0 : continue;
4287 : }
4288 :
4289 : /*
4290 : * If this worker has crashed previously, maybe it needs to be
4291 : * restarted (unless on registration it specified it doesn't want to
4292 : * be restarted at all). Check how long ago did a crash last happen.
4293 : * If the last crash is too recent, don't start it right away; let it
4294 : * be restarted once enough time has passed.
4295 : */
4296 14002 : if (rw->rw_crashed_at != 0)
4297 : {
4298 6418 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
4299 0 : {
4300 : int notify_pid;
4301 :
4302 0 : notify_pid = rw->rw_worker.bgw_notify_pid;
4303 :
4304 0 : ForgetBackgroundWorker(rw);
4305 :
4306 : /* Report worker is gone now. */
4307 0 : if (notify_pid != 0)
4308 0 : kill(notify_pid, SIGUSR1);
4309 :
4310 0 : continue;
4311 : }
4312 :
4313 : /* read system time only when needed */
4314 6418 : if (now == 0)
4315 6418 : now = GetCurrentTimestamp();
4316 :
4317 6418 : if (!TimestampDifferenceExceeds(rw->rw_crashed_at, now,
4318 6418 : rw->rw_worker.bgw_restart_time * 1000))
4319 : {
4320 : /* Set flag to remember that we have workers to start later */
4321 6418 : HaveCrashedWorker = true;
4322 6418 : continue;
4323 : }
4324 : }
4325 :
4326 7584 : if (bgworker_should_start_now(rw->rw_worker.bgw_start_time))
4327 : {
4328 : /* reset crash time before trying to start worker */
4329 5558 : rw->rw_crashed_at = 0;
4330 :
4331 : /*
4332 : * Try to start the worker.
4333 : *
4334 : * On failure, give up processing workers for now, but set
4335 : * StartWorkerNeeded so we'll come back here on the next iteration
4336 : * of ServerLoop to try again. (We don't want to wait, because
4337 : * there might be additional ready-to-run workers.) We could set
4338 : * HaveCrashedWorker as well, since this worker is now marked
4339 : * crashed, but there's no need because the next run of this
4340 : * function will do that.
4341 : */
4342 5558 : if (!StartBackgroundWorker(rw))
4343 : {
4344 0 : StartWorkerNeeded = true;
4345 0 : return;
4346 : }
4347 :
4348 : /*
4349 : * If we've launched as many workers as allowed, quit, but have
4350 : * ServerLoop call us again to look for additional ready-to-run
4351 : * workers. There might not be any, but we'll find out the next
4352 : * time we run.
4353 : */
4354 5558 : if (++num_launched >= MAX_BGWORKERS_TO_LAUNCH)
4355 : {
4356 0 : StartWorkerNeeded = true;
4357 0 : return;
4358 : }
4359 : }
4360 : }
4361 : }
4362 :
4363 : static bool
4364 42336 : maybe_reap_io_worker(int pid)
4365 : {
4366 1226436 : for (int i = 0; i < MAX_IO_WORKERS; ++i)
4367 : {
4368 1189638 : if (io_worker_children[i] &&
4369 116224 : io_worker_children[i]->pid == pid)
4370 : {
4371 5538 : ReleasePostmasterChildSlot(io_worker_children[i]);
4372 :
4373 5538 : --io_worker_count;
4374 5538 : io_worker_children[i] = NULL;
4375 5538 : return true;
4376 : }
4377 : }
4378 36798 : return false;
4379 : }
4380 :
4381 : /*
4382 : * Start or stop IO workers, to close the gap between the number of running
4383 : * workers and the number of configured workers. Used to respond to change of
4384 : * the io_workers GUC (by increasing and decreasing the number of workers), as
4385 : * well as workers terminating in response to errors (by starting
4386 : * "replacement" workers).
4387 : */
4388 : static void
4389 261408 : maybe_adjust_io_workers(void)
4390 : {
4391 261408 : if (!pgaio_workers_enabled())
4392 112 : return;
4393 :
4394 : /*
4395 : * If we're in final shutting down state, then we're just waiting for all
4396 : * processes to exit.
4397 : */
4398 261296 : if (pmState >= PM_WAIT_IO_WORKERS)
4399 8126 : return;
4400 :
4401 : /* Don't start new workers during an immediate shutdown either. */
4402 253170 : if (Shutdown >= ImmediateShutdown)
4403 4464 : return;
4404 :
4405 : /*
4406 : * Don't start new workers if we're in the shutdown phase of a crash
4407 : * restart. But we *do* need to start if we're already starting up again.
4408 : */
4409 248706 : if (FatalError && pmState >= PM_STOP_BACKENDS)
4410 122 : return;
4411 :
4412 : Assert(pmState < PM_WAIT_IO_WORKERS);
4413 :
4414 : /* Not enough running? */
4415 254140 : while (io_worker_count < io_workers)
4416 : {
4417 : PMChild *child;
4418 : int i;
4419 :
4420 : /* find unused entry in io_worker_children array */
4421 12138 : for (i = 0; i < MAX_IO_WORKERS; ++i)
4422 : {
4423 12138 : if (io_worker_children[i] == NULL)
4424 5556 : break;
4425 : }
4426 5556 : if (i == MAX_IO_WORKERS)
4427 0 : elog(ERROR, "could not find a free IO worker slot");
4428 :
4429 : /* Try to launch one. */
4430 5556 : child = StartChildProcess(B_IO_WORKER);
4431 5556 : if (child != NULL)
4432 : {
4433 5556 : io_worker_children[i] = child;
4434 5556 : ++io_worker_count;
4435 : }
4436 : else
4437 0 : break; /* try again next time */
4438 : }
4439 :
4440 : /* Too many running? */
4441 248584 : if (io_worker_count > io_workers)
4442 : {
4443 : /* ask the IO worker in the highest slot to exit */
4444 978 : for (int i = MAX_IO_WORKERS - 1; i >= 0; --i)
4445 : {
4446 978 : if (io_worker_children[i] != NULL)
4447 : {
4448 96 : kill(io_worker_children[i]->pid, SIGUSR2);
4449 96 : break;
4450 : }
4451 : }
4452 : }
4453 : }
4454 :
4455 :
4456 : /*
4457 : * When a backend asks to be notified about worker state changes, we
4458 : * set a flag in its backend entry. The background worker machinery needs
4459 : * to know when such backends exit.
4460 : */
4461 : bool
4462 4048 : PostmasterMarkPIDForWorkerNotify(int pid)
4463 : {
4464 : dlist_iter iter;
4465 : PMChild *bp;
4466 :
4467 11998 : dlist_foreach(iter, &ActiveChildList)
4468 : {
4469 11998 : bp = dlist_container(PMChild, elem, iter.cur);
4470 11998 : if (bp->pid == pid)
4471 : {
4472 4048 : bp->bgworker_notify = true;
4473 4048 : return true;
4474 : }
4475 : }
4476 0 : return false;
4477 : }
4478 :
4479 : #ifdef WIN32
4480 :
4481 : /*
4482 : * Subset implementation of waitpid() for Windows. We assume pid is -1
4483 : * (that is, check all child processes) and options is WNOHANG (don't wait).
4484 : */
4485 : static pid_t
4486 : waitpid(pid_t pid, int *exitstatus, int options)
4487 : {
4488 : win32_deadchild_waitinfo *childinfo;
4489 : DWORD exitcode;
4490 : DWORD dwd;
4491 : ULONG_PTR key;
4492 : OVERLAPPED *ovl;
4493 :
4494 : /* Try to consume one win32_deadchild_waitinfo from the queue. */
4495 : if (!GetQueuedCompletionStatus(win32ChildQueue, &dwd, &key, &ovl, 0))
4496 : {
4497 : errno = EAGAIN;
4498 : return -1;
4499 : }
4500 :
4501 : childinfo = (win32_deadchild_waitinfo *) key;
4502 : pid = childinfo->procId;
4503 :
4504 : /*
4505 : * Remove handle from wait - required even though it's set to wait only
4506 : * once
4507 : */
4508 : UnregisterWaitEx(childinfo->waitHandle, NULL);
4509 :
4510 : if (!GetExitCodeProcess(childinfo->procHandle, &exitcode))
4511 : {
4512 : /*
4513 : * Should never happen. Inform user and set a fixed exitcode.
4514 : */
4515 : write_stderr("could not read exit code for process\n");
4516 : exitcode = 255;
4517 : }
4518 : *exitstatus = exitcode;
4519 :
4520 : /*
4521 : * Close the process handle. Only after this point can the PID can be
4522 : * recycled by the kernel.
4523 : */
4524 : CloseHandle(childinfo->procHandle);
4525 :
4526 : /*
4527 : * Free struct that was allocated before the call to
4528 : * RegisterWaitForSingleObject()
4529 : */
4530 : pfree(childinfo);
4531 :
4532 : return pid;
4533 : }
4534 :
4535 : /*
4536 : * Note! Code below executes on a thread pool! All operations must
4537 : * be thread safe! Note that elog() and friends must *not* be used.
4538 : */
4539 : static void WINAPI
4540 : pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
4541 : {
4542 : /* Should never happen, since we use INFINITE as timeout value. */
4543 : if (TimerOrWaitFired)
4544 : return;
4545 :
4546 : /*
4547 : * Post the win32_deadchild_waitinfo object for waitpid() to deal with. If
4548 : * that fails, we leak the object, but we also leak a whole process and
4549 : * get into an unrecoverable state, so there's not much point in worrying
4550 : * about that. We'd like to panic, but we can't use that infrastructure
4551 : * from this thread.
4552 : */
4553 : if (!PostQueuedCompletionStatus(win32ChildQueue,
4554 : 0,
4555 : (ULONG_PTR) lpParameter,
4556 : NULL))
4557 : write_stderr("could not post child completion status\n");
4558 :
4559 : /* Queue SIGCHLD signal. */
4560 : pg_queue_signal(SIGCHLD);
4561 : }
4562 :
4563 : /*
4564 : * Queue a waiter to signal when this child dies. The wait will be handled
4565 : * automatically by an operating system thread pool. The memory and the
4566 : * process handle will be freed by a later call to waitpid().
4567 : */
4568 : void
4569 : pgwin32_register_deadchild_callback(HANDLE procHandle, DWORD procId)
4570 : {
4571 : win32_deadchild_waitinfo *childinfo;
4572 :
4573 : childinfo = palloc_object(win32_deadchild_waitinfo);
4574 : childinfo->procHandle = procHandle;
4575 : childinfo->procId = procId;
4576 :
4577 : if (!RegisterWaitForSingleObject(&childinfo->waitHandle,
4578 : procHandle,
4579 : pgwin32_deadchild_callback,
4580 : childinfo,
4581 : INFINITE,
4582 : WT_EXECUTEONLYONCE | WT_EXECUTEINWAITTHREAD))
4583 : ereport(FATAL,
4584 : (errmsg_internal("could not register process for wait: error code %lu",
4585 : GetLastError())));
4586 : }
4587 :
4588 : #endif /* WIN32 */
4589 :
4590 : /*
4591 : * Initialize one and only handle for monitoring postmaster death.
4592 : *
4593 : * Called once in the postmaster, so that child processes can subsequently
4594 : * monitor if their parent is dead.
4595 : */
4596 : static void
4597 1820 : InitPostmasterDeathWatchHandle(void)
4598 : {
4599 : #ifndef WIN32
4600 :
4601 : /*
4602 : * Create a pipe. Postmaster holds the write end of the pipe open
4603 : * (POSTMASTER_FD_OWN), and children hold the read end. Children can pass
4604 : * the read file descriptor to select() to wake up in case postmaster
4605 : * dies, or check for postmaster death with a (read() == 0). Children must
4606 : * close the write end as soon as possible after forking, because EOF
4607 : * won't be signaled in the read end until all processes have closed the
4608 : * write fd. That is taken care of in ClosePostmasterPorts().
4609 : */
4610 : Assert(MyProcPid == PostmasterPid);
4611 1820 : if (pipe(postmaster_alive_fds) < 0)
4612 0 : ereport(FATAL,
4613 : (errcode_for_file_access(),
4614 : errmsg_internal("could not create pipe to monitor postmaster death: %m")));
4615 :
4616 : /* Notify fd.c that we've eaten two FDs for the pipe. */
4617 1820 : ReserveExternalFD();
4618 1820 : ReserveExternalFD();
4619 :
4620 : /*
4621 : * Set O_NONBLOCK to allow testing for the fd's presence with a read()
4622 : * call.
4623 : */
4624 1820 : if (fcntl(postmaster_alive_fds[POSTMASTER_FD_WATCH], F_SETFL, O_NONBLOCK) == -1)
4625 0 : ereport(FATAL,
4626 : (errcode_for_socket_access(),
4627 : errmsg_internal("could not set postmaster death monitoring pipe to nonblocking mode: %m")));
4628 : #else
4629 :
4630 : /*
4631 : * On Windows, we use a process handle for the same purpose.
4632 : */
4633 : if (DuplicateHandle(GetCurrentProcess(),
4634 : GetCurrentProcess(),
4635 : GetCurrentProcess(),
4636 : &PostmasterHandle,
4637 : 0,
4638 : TRUE,
4639 : DUPLICATE_SAME_ACCESS) == 0)
4640 : ereport(FATAL,
4641 : (errmsg_internal("could not duplicate postmaster handle: error code %lu",
4642 : GetLastError())));
4643 : #endif /* WIN32 */
4644 1820 : }
|
