1 /*-------------------------------------------------------------------------
4 * This program acts as a clearing house for requests to the
5 * POSTGRES system. Frontend programs send a startup message
6 * to the Postmaster and the postmaster uses the info in the
7 * message to setup a backend process.
9 * The postmaster also manages system-wide operations such as
10 * startup, shutdown, and periodic checkpoints. The postmaster
11 * itself doesn't do those operations, mind you --- it just forks
12 * off a subprocess to do them at the right times. It also takes
13 * care of resetting the system if a backend crashes.
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.
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
35 * Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
36 * Portions Copyright (c) 1994, Regents of the University of California
40 * $PostgreSQL: pgsql/src/backend/postmaster/postmaster.c,v 1.389 2004/05/18 20:18:57 momjian Exp $
45 * The Postmaster sets up a few shared memory data structures
46 * for the backends. It should at the very least initialize the
50 * The Postmaster shares memory with the backends but should avoid
51 * touching shared memory, so as not to become stuck if a crashing
52 * backend screws up locks or shared memory. Likewise, the Postmaster
53 * should never block on messages from frontend clients.
56 * The Postmaster cleans up after backends if they have an emergency
57 * exit and/or core dump.
59 *-------------------------------------------------------------------------
70 #include <sys/socket.h>
74 #include <sys/param.h>
75 #include <netinet/in.h>
76 #include <arpa/inet.h>
80 #ifdef HAVE_SYS_SELECT_H
81 #include <sys/select.h>
89 #include <DNSServiceDiscovery/DNSServiceDiscovery.h>
92 #include "catalog/pg_database.h"
93 #include "commands/async.h"
94 #include "lib/dllist.h"
95 #include "libpq/auth.h"
96 #include "libpq/crypt.h"
97 #include "libpq/libpq.h"
98 #include "libpq/pqcomm.h"
99 #include "libpq/pqsignal.h"
100 #include "miscadmin.h"
101 #include "nodes/nodes.h"
102 #include "storage/fd.h"
103 #include "storage/ipc.h"
104 #include "storage/pg_shmem.h"
105 #include "storage/pmsignal.h"
106 #include "storage/proc.h"
107 #include "storage/bufmgr.h"
108 #include "access/xlog.h"
109 #include "tcop/tcopprot.h"
110 #include "utils/guc.h"
111 #include "utils/memutils.h"
112 #include "utils/ps_status.h"
113 #include "bootstrap/bootstrap.h"
117 #define INVALID_SOCK (-1)
119 #ifdef HAVE_SIGPROCMASK
131 * List of active backends (or child processes anyway; we don't actually
132 * know whether a given child has become a backend or is still in the
133 * authorization phase). This is used mainly to keep track of how many
134 * children we have and send them appropriate signals when necessary.
138 pid_t pid; /* process id of backend */
139 long cancel_key; /* cancel key for cancels for this backend */
142 static Dllist *BackendList;
145 #define NUM_BACKENDARRAY_ELEMS (2*MaxBackends)
146 static Backend *ShmemBackendArray;
149 /* The socket number we are listening for connections on */
152 char *ListenAddresses;
155 * MaxBackends is the limit on the number of backends we can start.
156 * Note that a larger MaxBackends value will increase the size of the
157 * shared memory area as well as cause the postmaster to grab more
158 * kernel semaphores, even if you never actually use that many
164 * ReservedBackends is the number of backends reserved for superuser use.
165 * This number is taken out of the pool size given by MaxBackends so
166 * number of backend slots available to non-superusers is
167 * (MaxBackends - ReservedBackends). Note what this really means is
168 * "if there are <= ReservedBackends connections available, only superusers
169 * can make new connections" --- pre-existing superuser connections don't
170 * count against the limit.
172 int ReservedBackends;
175 static const char *progname = NULL;
177 /* The socket(s) we're listening to. */
179 static int ListenSocket[MAXLISTEN];
181 /* Used to reduce macros tests */
183 const bool ExecBackend = true;
185 const bool ExecBackend = false;
189 * Set by the -o option
191 static char ExtraOptions[MAXPGPATH];
194 * These globals control the behavior of the postmaster in case some
195 * backend dumps core. Normally, it kills all peers of the dead backend
196 * and reinitializes shared memory. By specifying -s or -n, we can have
197 * the postmaster stop (rather than kill) peers and not reinitialize
198 * shared data structures.
200 static bool Reinit = true;
201 static int SendStop = false;
203 /* still more option variables */
204 bool EnableSSL = false;
205 bool SilentMode = false; /* silent mode (-S) */
207 int PreAuthDelay = 0;
208 int AuthenticationTimeout = 60;
209 int CheckPointTimeout = 300;
210 int CheckPointWarning = 30;
211 time_t LastSignalledCheckpoint = 0;
213 bool log_hostname; /* for ps display and logging */
214 bool Log_connections = false;
215 bool Db_user_namespace = false;
217 char *rendezvous_name;
219 /* list of library:init-function to be preloaded */
220 char *preload_libraries_string = NULL;
222 /* Startup/shutdown state */
223 static pid_t StartupPID = 0,
227 static time_t checkpointed = 0;
230 #define SmartShutdown 1
231 #define FastShutdown 2
233 static int Shutdown = NoShutdown;
235 static bool FatalError = false; /* T if recovering from backend crash */
237 bool ClientAuthInProgress = false; /* T during new-client
241 * State for assigning random salts and cancel keys.
242 * Also, the global MyCancelKey passes the cancel key assigned to a given
243 * backend from the postmaster to that backend (via fork).
246 static unsigned int random_seed = 0;
248 static int debug_flag = 0;
254 #ifdef HAVE_INT_OPTRESET
259 * postmaster.c - function prototypes
261 static void pmdaemonize(int argc, char *argv[]);
262 static Port *ConnCreate(int serverFd);
263 static void ConnFree(Port *port);
264 static void reset_shared(unsigned short port);
265 static void SIGHUP_handler(SIGNAL_ARGS);
266 static void pmdie(SIGNAL_ARGS);
267 static void reaper(SIGNAL_ARGS);
268 static void sigusr1_handler(SIGNAL_ARGS);
269 static void dummy_handler(SIGNAL_ARGS);
270 static void CleanupProc(int pid, int exitstatus);
271 static void LogChildExit(int lev, const char *procname,
272 int pid, int exitstatus);
273 static void BackendInit(Port *port);
274 static int BackendRun(Port *port);
275 static void ExitPostmaster(int status);
276 static void usage(const char *);
277 static int ServerLoop(void);
278 static int BackendStartup(Port *port);
279 static int ProcessStartupPacket(Port *port, bool SSLdone);
280 static void processCancelRequest(Port *port, void *pkt);
281 static int initMasks(fd_set *rmask);
282 static void report_fork_failure_to_client(Port *port, int errnum);
283 static enum CAC_state canAcceptConnections(void);
284 static long PostmasterRandom(void);
285 static void RandomSalt(char *cryptSalt, char *md5Salt);
286 static void SignalChildren(int signal);
287 static int CountChildren(void);
288 static bool CreateOptsFile(int argc, char *argv[], char *fullprogname);
289 NON_EXEC_STATIC void SSDataBaseInit(int xlop);
290 static pid_t SSDataBase(int xlop);
292 postmaster_error(const char *fmt,...)
293 /* This lets gcc check the format string for consistency. */
294 __attribute__((format(printf, 1, 2)));
299 pid_t win32_forkexec(const char* path, char *argv[]);
301 static void win32_AddChild(pid_t pid, HANDLE handle);
302 static void win32_RemoveChild(pid_t pid);
303 static pid_t win32_waitpid(int *exitstatus);
304 static DWORD WINAPI win32_sigchld_waiter(LPVOID param);
306 static pid_t *win32_childPIDArray;
307 static HANDLE *win32_childHNDArray;
308 static unsigned long win32_numChildren = 0;
311 static pid_t Backend_forkexec(Port *port);
313 static unsigned long tmpBackendFileNum = 0;
314 void read_backend_variables(unsigned long id, Port *port);
315 static bool write_backend_variables(Port *port);
317 static void ShmemBackendArrayAdd(Backend *bn);
318 static void ShmemBackendArrayRemove(pid_t pid);
321 #define StartupDataBase() SSDataBase(BS_XLOG_STARTUP)
322 #define CheckPointDataBase() SSDataBase(BS_XLOG_CHECKPOINT)
323 #define StartBackgroundWriter() SSDataBase(BS_XLOG_BGWRITER)
324 #define ShutdownDataBase() SSDataBase(BS_XLOG_SHUTDOWN)
327 checkDataDir(const char *checkdir)
329 char path[MAXPGPATH];
331 struct stat stat_buf;
333 if (checkdir == NULL)
336 gettext("%s does not know where to find the database system data.\n"
337 "You must specify the directory that contains the database system\n"
338 "either by specifying the -D invocation option or by setting the\n"
339 "PGDATA environment variable.\n"),
344 if (stat(checkdir, &stat_buf) == -1)
348 (errcode_for_file_access(),
349 errmsg("data directory \"%s\" does not exist",
353 (errcode_for_file_access(),
354 errmsg("could not read permissions of directory \"%s\": %m",
359 * Check if the directory has group or world access. If so, reject.
361 * XXX temporarily suppress check when on Windows, because there may not
362 * be proper support for Unix-y file permissions. Need to think of a
363 * reasonable check to apply on Windows.
365 #if !defined(__CYGWIN__) && !defined(WIN32)
366 if (stat_buf.st_mode & (S_IRWXG | S_IRWXO))
368 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
369 errmsg("data directory \"%s\" has group or world access",
371 errdetail("Permissions should be u=rwx (0700).")));
374 /* Look for PG_VERSION before looking for pg_control */
375 ValidatePgVersion(checkdir);
377 snprintf(path, sizeof(path), "%s/global/pg_control", checkdir);
379 fp = AllocateFile(path, PG_BINARY_R);
383 gettext("%s: could not find the database system\n"
384 "Expected to find it in the directory \"%s\",\n"
385 "but could not open file \"%s\": %s\n"),
386 progname, checkdir, path, strerror(errno));
393 #ifdef USE_RENDEZVOUS
395 /* reg_reply -- empty callback function for DNSServiceRegistrationCreate() */
397 reg_reply(DNSServiceRegistrationReplyErrorType errorCode, void *context)
404 PostmasterMain(int argc, char *argv[])
408 char original_extraoptions[MAXPGPATH];
409 char *potential_DataDir = NULL;
412 *original_extraoptions = '\0';
414 progname = get_progname(argv[0]);
416 IsPostmasterEnvironment = true;
419 * Catch standard options before doing much else. This even works on
420 * systems without getopt_long.
424 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
429 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
431 puts("postmaster (PostgreSQL) " PG_VERSION);
437 * for security, no dir or file created can be group or other
440 umask((mode_t) 0077);
442 MyProcPid = PostmasterPid = getpid();
445 * Fire up essential subsystems: memory management
450 * By default, palloc() requests in the postmaster will be allocated
451 * in the PostmasterContext, which is space that can be recycled by
452 * backends. Allocated data that needs to be available to backends
453 * should be allocated in TopMemoryContext.
455 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
457 ALLOCSET_DEFAULT_MINSIZE,
458 ALLOCSET_DEFAULT_INITSIZE,
459 ALLOCSET_DEFAULT_MAXSIZE);
460 MemoryContextSwitchTo(PostmasterContext);
462 IgnoreSystemIndexes(false);
467 InitializeGUCOptions();
469 potential_DataDir = getenv("PGDATA"); /* default value */
473 while ((opt = getopt(argc, argv, "A:a:B:b:c:D:d:Fh:ik:lm:MN:no:p:Ss-:")) != -1)
478 #ifdef USE_ASSERT_CHECKING
479 SetConfigOption("debug_assertions", optarg, PGC_POSTMASTER, PGC_S_ARGV);
481 postmaster_error("assert checking is not compiled in");
485 /* Can no longer set authentication method. */
488 SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
491 /* Can no longer set the backend executable file to use. */
494 potential_DataDir = optarg;
498 /* Turn on debugging for the postmaster. */
499 char *debugstr = palloc(strlen("debug") + strlen(optarg) + 1);
501 sprintf(debugstr, "debug%s", optarg);
502 SetConfigOption("log_min_messages", debugstr,
503 PGC_POSTMASTER, PGC_S_ARGV);
505 debug_flag = atoi(optarg);
509 SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
512 SetConfigOption("listen_addresses", optarg, PGC_POSTMASTER, PGC_S_ARGV);
515 SetConfigOption("listen_addresses", "*", PGC_POSTMASTER, PGC_S_ARGV);
518 SetConfigOption("unix_socket_directory", optarg, PGC_POSTMASTER, PGC_S_ARGV);
522 SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
526 /* Multiplexed backends no longer supported. */
531 * ignore this flag. This may be passed in because the
532 * program was run as 'postgres -M' instead of
537 /* The max number of backends to start. */
538 SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
541 /* Don't reinit shared mem after abnormal exit */
547 * Other options to pass to the backend on the command
548 * line -- useful only for debugging.
550 strcat(ExtraOptions, " ");
551 strcat(ExtraOptions, optarg);
552 strcpy(original_extraoptions, optarg);
555 SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
560 * Start in 'S'ilent mode (disassociate from controlling
561 * tty). You may also think of this as 'S'ysV mode since
562 * it's most badly needed on SysV-derived systems like
565 SetConfigOption("silent_mode", "true", PGC_POSTMASTER, PGC_S_ARGV);
570 * In the event that some backend dumps core, send
571 * SIGSTOP, rather than SIGQUIT, to all its peers. This
572 * lets the wily post_hacker collect core dumps from
583 ParseLongOption(optarg, &name, &value);
588 (errcode(ERRCODE_SYNTAX_ERROR),
589 errmsg("--%s requires a value",
593 (errcode(ERRCODE_SYNTAX_ERROR),
594 errmsg("-c %s requires a value",
598 SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
607 gettext("Try \"%s --help\" for more information.\n"),
614 * Postmaster accepts no non-option switch arguments.
618 postmaster_error("invalid argument: \"%s\"", argv[optind]);
620 gettext("Try \"%s --help\" for more information.\n"),
626 * Now we can set the data directory, and then read postgresql.conf.
628 checkDataDir(potential_DataDir); /* issues error messages */
629 SetDataDir(potential_DataDir);
631 ProcessConfigFile(PGC_POSTMASTER);
633 write_nondefault_variables(PGC_POSTMASTER);
637 * Check for invalid combinations of GUC settings.
639 if (NBuffers < 2 * MaxBackends || NBuffers < 16)
642 * Do not accept -B so small that backends are likely to starve
643 * for lack of buffers. The specific choices here are somewhat
646 postmaster_error("the number of buffers (-B) must be at least twice the number of allowed connections (-N) and at least 16");
650 if (ReservedBackends >= MaxBackends)
652 postmaster_error("superuser_reserved_connections must be less than max_connections");
657 * Other one-time internal sanity checks can go here.
659 if (!CheckDateTokenTables())
661 postmaster_error("invalid datetoken tables, please fix");
666 * Now that we are done processing the postmaster arguments, reset
667 * getopt(3) library so that it will work correctly in subprocesses.
670 #ifdef HAVE_INT_OPTRESET
671 optreset = 1; /* some systems need this too */
674 /* For debugging: display postmaster environment */
676 extern char **environ;
680 (errmsg_internal("%s: PostmasterMain: initial environ dump:",
683 (errmsg_internal("-----------------------------------------")));
684 for (p = environ; *p; ++p)
686 (errmsg_internal("\t%s", *p)));
688 (errmsg_internal("-----------------------------------------")));
691 if (strlen(pkglib_path) == 0)
692 get_pkglib_path(my_exec_path, pkglib_path);
695 if (find_other_exec(argv[0], "postgres", PG_VERSIONSTR, postgres_exec_path) < 0)
697 (errmsg("%s: could not locate postgres executable or non-matching version",
702 * Initialize SSL library, if specified.
710 * process any libraries that should be preloaded and optionally
713 if (preload_libraries_string)
714 process_preload_libraries(preload_libraries_string);
717 * Fork away from controlling terminal, if -S specified.
719 * Must do this before we grab any interlock files, else the interlocks
720 * will show the wrong PID.
723 pmdaemonize(argc, argv);
726 * Create lockfile for data directory.
728 * We want to do this before we try to grab the input sockets, because
729 * the data directory interlock is more reliable than the socket-file
730 * interlock (thanks to whoever decided to put socket files in /tmp
731 * :-(). For the same reason, it's best to grab the TCP socket before
734 CreateDataDirLockFile(DataDir, true);
737 * Remove old temporary files. At this point there can be no other
738 * Postgres processes running in this directory, so this should be
744 * Establish input sockets.
746 for (i = 0; i < MAXLISTEN; i++)
747 ListenSocket[i] = -1;
755 curhost = ListenAddresses;
758 /* ignore whitespace */
759 while (isspace((unsigned char) *curhost))
761 if (*curhost == '\0')
764 while (*endptr != '\0' && !isspace((unsigned char) *endptr))
768 if (strcmp(curhost,"*") == 0)
769 status = StreamServerPort(AF_UNSPEC, NULL,
770 (unsigned short) PostPortNumber,
772 ListenSocket, MAXLISTEN);
774 status = StreamServerPort(AF_UNSPEC, curhost,
775 (unsigned short) PostPortNumber,
777 ListenSocket, MAXLISTEN);
778 if (status != STATUS_OK)
780 (errmsg("could not create listen socket for \"%s\"",
790 #ifdef USE_RENDEZVOUS
791 /* Register for Rendezvous only if we opened TCP socket(s) */
792 if (ListenSocket[0] != -1 && rendezvous_name != NULL)
794 DNSServiceRegistrationCreate(rendezvous_name,
797 htonl(PostPortNumber),
799 (DNSServiceRegistrationReply) reg_reply,
804 #ifdef HAVE_UNIX_SOCKETS
805 status = StreamServerPort(AF_UNIX, NULL,
806 (unsigned short) PostPortNumber,
808 ListenSocket, MAXLISTEN);
809 if (status != STATUS_OK)
811 (errmsg("could not create Unix-domain socket")));
815 * check that we have some socket to listen on
817 if (ListenSocket[0] == -1)
819 (errmsg("no socket created for listening")));
824 * Set up shared memory and semaphores.
826 reset_shared(PostPortNumber);
829 * Estimate number of openable files. This must happen after setting up
830 * semaphores, because on some platforms semaphores count as open files.
835 * Initialize the list of active backends.
837 BackendList = DLNewList();
841 * Initialize the child pid/HANDLE arrays
843 win32_childPIDArray = (pid_t*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(pid_t));
844 win32_childHNDArray = (HANDLE*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(HANDLE));
845 if (!win32_childPIDArray || !win32_childHNDArray)
847 (errcode(ERRCODE_OUT_OF_MEMORY),
848 errmsg("out of memory")));
852 * Record postmaster options. We delay this till now to avoid
853 * recording bogus options (eg, NBuffers too high for available
856 if (!CreateOptsFile(argc, argv, my_exec_path))
860 * Set up signal handlers for the postmaster process.
862 * CAUTION: when changing this list, check for side-effects on the signal
863 * handling setup of child processes. See tcop/postgres.c,
864 * bootstrap/bootstrap.c, and postmaster/pgstat.c.
867 PG_SETMASK(&BlockSig);
869 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
870 * children do same */
871 pqsignal(SIGINT, pmdie); /* send SIGTERM and ShutdownDataBase */
872 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
873 pqsignal(SIGTERM, pmdie); /* wait for children and ShutdownDataBase */
874 pqsignal(SIGALRM, SIG_IGN); /* ignored */
875 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
876 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
877 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
878 pqsignal(SIGCHLD, reaper); /* handle child termination */
879 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
880 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
881 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
883 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
887 * Reset whereToSendOutput from Debug (its starting state) to None.
888 * This prevents ereport from sending log messages to stderr unless
889 * the syslog/stderr switch permits. We don't do this until the
890 * postmaster is fully launched, since startup failures may as well be
891 * reported to stderr.
893 whereToSendOutput = None;
896 * On many platforms, the first call of localtime() incurs significant
897 * overhead to load timezone info from the system configuration files.
898 * By doing it once in the postmaster, we avoid having to do it in
899 * every started child process. The savings are not huge, but they
903 time_t now = time(NULL);
905 (void) localtime(&now);
909 * Initialize and try to startup the statistics collector process
915 * Load cached files for client authentication.
923 * We're ready to rock and roll...
925 StartupPID = StartupDataBase();
927 status = ServerLoop();
930 * ServerLoop probably shouldn't ever return, but if it does, close
933 ExitPostmaster(status != STATUS_OK);
935 return 0; /* not reached */
939 pmdaemonize(int argc, char *argv[])
943 elog(FATAL,"SilentMode not supported under WIN32");
949 struct itimerval prof_itimer;
953 /* see comments in BackendRun */
954 getitimer(ITIMER_PROF, &prof_itimer);
958 if (pid == (pid_t) -1)
960 postmaster_error("could not fork background process: %s",
966 /* Parent should just exit, without doing any atexit cleanup */
971 setitimer(ITIMER_PROF, &prof_itimer, NULL);
974 MyProcPid = getpid(); /* reset MyProcPid to child */
976 /* GH: If there's no setsid(), we hopefully don't need silent mode.
977 * Until there's a better solution.
982 postmaster_error("could not dissociate from controlling TTY: %s",
987 i = open(NULL_DEV, O_RDWR | PG_BINARY);
998 * Print out help message
1001 usage(const char *progname)
1003 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
1004 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
1005 printf(gettext("Options:\n"));
1006 #ifdef USE_ASSERT_CHECKING
1007 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
1009 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
1010 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
1011 printf(gettext(" -d 1-5 debugging level\n"));
1012 printf(gettext(" -D DATADIR database directory\n"));
1013 printf(gettext(" -F turn fsync off\n"));
1014 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
1015 printf(gettext(" -i enable TCP/IP connections\n"));
1016 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
1018 printf(gettext(" -l enable SSL connections\n"));
1020 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
1021 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
1022 printf(gettext(" -p PORT port number to listen on\n"));
1023 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1024 printf(gettext(" --help show this help, then exit\n"));
1025 printf(gettext(" --version output version information, then exit\n"));
1027 printf(gettext("\nDeveloper options:\n"));
1028 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1029 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1031 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1032 "configuration settings and how to set them on the command line or in\n"
1033 "the configuration file.\n\n"
1034 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1047 gettimeofday(&now, &tz);
1049 nSockets = initMasks(&readmask);
1055 struct timeval timeout;
1058 * The timeout for the select() below is normally set on the basis
1059 * of the time to the next checkpoint. However, if for some
1060 * reason we don't have a next-checkpoint time, time out after 60
1061 * seconds. This keeps checkpoint scheduling from locking up when
1062 * we get new connection requests infrequently (since we are
1063 * likely to detect checkpoint completion just after enabling
1064 * signals below, after we've already made the decision about how
1065 * long to wait this time).
1067 timeout.tv_sec = 60;
1068 timeout.tv_usec = 0;
1070 if (CheckPointPID == 0 && checkpointed &&
1071 StartupPID == 0 && Shutdown == NoShutdown &&
1072 !FatalError && random_seed != 0)
1074 time_t now = time(NULL);
1076 if (CheckPointTimeout + checkpointed > now)
1079 * Not time for checkpoint yet, so set select timeout
1081 timeout.tv_sec = CheckPointTimeout + checkpointed - now;
1085 /* Time to make the checkpoint... */
1086 CheckPointPID = CheckPointDataBase();
1089 * if fork failed, schedule another try at 0.1 normal
1092 if (CheckPointPID == 0)
1094 timeout.tv_sec = CheckPointTimeout / 10;
1095 checkpointed = now + timeout.tv_sec - CheckPointTimeout;
1101 * If no background writer process is running and we should
1102 * do background writing, start one. It doesn't matter if
1103 * this fails, we'll just try again later.
1105 if (BgWriterPID == 0 && BgWriterPercent > 0 &&
1106 StartupPID == 0 && Shutdown == NoShutdown &&
1107 !FatalError && random_seed != 0)
1109 BgWriterPID = StartBackgroundWriter();
1113 * Wait for something to happen.
1115 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1117 PG_SETMASK(&UnBlockSig);
1119 if (select(nSockets, &rmask, NULL, NULL, &timeout) < 0)
1121 PG_SETMASK(&BlockSig);
1122 if (errno == EINTR || errno == EWOULDBLOCK)
1125 (errcode_for_socket_access(),
1126 errmsg("select() failed in postmaster: %m")));
1127 return STATUS_ERROR;
1131 * Block all signals until we wait again. (This makes it safe for
1132 * our signal handlers to do nontrivial work.)
1134 PG_SETMASK(&BlockSig);
1137 * Select a random seed at the time of first receiving a request.
1139 while (random_seed == 0)
1141 gettimeofday(&later, &tz);
1144 * We are not sure how much precision is in tv_usec, so we
1145 * swap the nibbles of 'later' and XOR them with 'now'. On the
1146 * off chance that the result is 0, we loop until it isn't.
1148 random_seed = now.tv_usec ^
1149 ((later.tv_usec << 16) |
1150 ((later.tv_usec >> 16) & 0xffff));
1154 * New connection pending on any of our sockets? If so, fork a
1155 * child process to deal with it.
1157 for (i = 0; i < MAXLISTEN; i++)
1159 if (ListenSocket[i] == -1)
1161 if (FD_ISSET(ListenSocket[i], &rmask))
1163 port = ConnCreate(ListenSocket[i]);
1166 BackendStartup(port);
1169 * We no longer need the open socket or port structure
1172 StreamClose(port->sock);
1178 /* If we have lost the stats collector, try to start a new one */
1179 if (!pgstat_is_running)
1186 * Initialise the masks for select() for the ports
1187 * we are listening on. Return the number of sockets to listen on.
1191 initMasks(fd_set *rmask)
1198 for (i = 0; i < MAXLISTEN; i++)
1200 int fd = ListenSocket[i];
1214 * Read the startup packet and do something according to it.
1216 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1217 * not return at all.
1219 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1220 * if that's what you want. Return STATUS_ERROR if you don't want to
1221 * send anything to the client, which would typically be appropriate
1222 * if we detect a communications failure.)
1225 ProcessStartupPacket(Port *port, bool SSLdone)
1229 ProtocolVersion proto;
1230 MemoryContext oldcontext;
1232 if (pq_getbytes((char *) &len, 4) == EOF)
1235 * EOF after SSLdone probably means the client didn't like our
1236 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1237 * so don't clutter the log with a complaint.
1241 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1242 errmsg("incomplete startup packet")));
1243 return STATUS_ERROR;
1249 if (len < (int32) sizeof(ProtocolVersion) ||
1250 len > MAX_STARTUP_PACKET_LENGTH)
1253 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1254 errmsg("invalid length of startup packet")));
1255 return STATUS_ERROR;
1259 * Allocate at least the size of an old-style startup packet, plus one
1260 * extra byte, and make sure all are zeroes. This ensures we will
1261 * have null termination of all strings, in both fixed- and
1262 * variable-length packet layouts.
1264 if (len <= (int32) sizeof(StartupPacket))
1265 buf = palloc0(sizeof(StartupPacket) + 1);
1267 buf = palloc0(len + 1);
1269 if (pq_getbytes(buf, len) == EOF)
1272 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1273 errmsg("incomplete startup packet")));
1274 return STATUS_ERROR;
1278 * The first field is either a protocol version number or a special
1281 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1283 if (proto == CANCEL_REQUEST_CODE)
1285 processCancelRequest(port, buf);
1286 return 127; /* XXX */
1289 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1294 /* No SSL when disabled or on Unix sockets */
1295 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1298 SSLok = 'S'; /* Support for SSL */
1300 SSLok = 'N'; /* No support for SSL */
1302 if (send(port->sock, &SSLok, 1, 0) != 1)
1305 (errcode_for_socket_access(),
1306 errmsg("failed to send SSL negotiation response: %m")));
1307 return STATUS_ERROR; /* close the connection */
1311 if (SSLok == 'S' && secure_open_server(port) == -1)
1312 return STATUS_ERROR;
1314 /* regular startup packet, cancel, etc packet should follow... */
1315 /* but not another SSL negotiation request */
1316 return ProcessStartupPacket(port, true);
1319 /* Could add additional special packet types here */
1322 * Set FrontendProtocol now so that ereport() knows what format to
1323 * send if we fail during startup.
1325 FrontendProtocol = proto;
1327 /* Check we can handle the protocol the frontend is using. */
1329 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1330 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1331 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1332 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1334 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1335 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1336 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1337 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1338 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1339 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1342 * Now fetch parameters out of startup packet and save them into the
1343 * Port structure. All data structures attached to the Port struct
1344 * must be allocated in TopMemoryContext so that they won't disappear
1345 * when we pass them to PostgresMain (see BackendRun). We need not
1346 * worry about leaking this storage on failure, since we aren't in the
1347 * postmaster process anymore.
1349 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1351 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1353 int32 offset = sizeof(ProtocolVersion);
1356 * Scan packet body for name/option pairs. We can assume any
1357 * string beginning within the packet body is null-terminated,
1358 * thanks to zeroing extra byte above.
1360 port->guc_options = NIL;
1362 while (offset < len)
1364 char *nameptr = ((char *) buf) + offset;
1368 if (*nameptr == '\0')
1369 break; /* found packet terminator */
1370 valoffset = offset + strlen(nameptr) + 1;
1371 if (valoffset >= len)
1372 break; /* missing value, will complain below */
1373 valptr = ((char *) buf) + valoffset;
1375 if (strcmp(nameptr, "database") == 0)
1376 port->database_name = pstrdup(valptr);
1377 else if (strcmp(nameptr, "user") == 0)
1378 port->user_name = pstrdup(valptr);
1379 else if (strcmp(nameptr, "options") == 0)
1380 port->cmdline_options = pstrdup(valptr);
1383 /* Assume it's a generic GUC option */
1384 port->guc_options = lappend(port->guc_options,
1386 port->guc_options = lappend(port->guc_options,
1389 offset = valoffset + strlen(valptr) + 1;
1393 * If we didn't find a packet terminator exactly at the end of the
1394 * given packet length, complain.
1396 if (offset != len - 1)
1398 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1399 errmsg("invalid startup packet layout: expected terminator as last byte")));
1404 * Get the parameters from the old-style, fixed-width-fields
1405 * startup packet as C strings. The packet destination was
1406 * cleared first so a short packet has zeros silently added. We
1407 * have to be prepared to truncate the pstrdup result for oversize
1410 StartupPacket *packet = (StartupPacket *) buf;
1412 port->database_name = pstrdup(packet->database);
1413 if (strlen(port->database_name) > sizeof(packet->database))
1414 port->database_name[sizeof(packet->database)] = '\0';
1415 port->user_name = pstrdup(packet->user);
1416 if (strlen(port->user_name) > sizeof(packet->user))
1417 port->user_name[sizeof(packet->user)] = '\0';
1418 port->cmdline_options = pstrdup(packet->options);
1419 if (strlen(port->cmdline_options) > sizeof(packet->options))
1420 port->cmdline_options[sizeof(packet->options)] = '\0';
1421 port->guc_options = NIL;
1424 /* Check a user name was given. */
1425 if (port->user_name == NULL || port->user_name[0] == '\0')
1427 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1428 errmsg("no PostgreSQL user name specified in startup packet")));
1430 /* The database defaults to the user name. */
1431 if (port->database_name == NULL || port->database_name[0] == '\0')
1432 port->database_name = pstrdup(port->user_name);
1434 if (Db_user_namespace)
1437 * If user@, it is a global user, remove '@'. We only want to do
1438 * this if there is an '@' at the end and no earlier in the user
1439 * string or they may fake as a local user of another database
1440 * attaching to this database.
1442 if (strchr(port->user_name, '@') ==
1443 port->user_name + strlen(port->user_name) - 1)
1444 *strchr(port->user_name, '@') = '\0';
1447 /* Append '@' and dbname */
1450 db_user = palloc(strlen(port->user_name) +
1451 strlen(port->database_name) + 2);
1452 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1453 port->user_name = db_user;
1458 * Truncate given database and user names to length of a Postgres
1459 * name. This avoids lookup failures when overlength names are given.
1461 if (strlen(port->database_name) >= NAMEDATALEN)
1462 port->database_name[NAMEDATALEN - 1] = '\0';
1463 if (strlen(port->user_name) >= NAMEDATALEN)
1464 port->user_name[NAMEDATALEN - 1] = '\0';
1467 * Done putting stuff in TopMemoryContext.
1469 MemoryContextSwitchTo(oldcontext);
1472 * If we're going to reject the connection due to database state, say
1473 * so now instead of wasting cycles on an authentication exchange.
1474 * (This also allows a pg_ping utility to be written.)
1476 switch (port->canAcceptConnections)
1480 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1481 errmsg("the database system is starting up")));
1485 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1486 errmsg("the database system is shutting down")));
1490 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1491 errmsg("the database system is in recovery mode")));
1495 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1496 errmsg("sorry, too many clients already")));
1508 * The client has sent a cancel request packet, not a normal
1509 * start-a-new-connection packet. Perform the necessary processing.
1510 * Nothing is sent back to the client.
1513 processCancelRequest(Port *port, void *pkt)
1515 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1517 long cancelAuthCode;
1519 #ifndef EXEC_BACKEND
1525 backendPID = (int) ntohl(canc->backendPID);
1526 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1528 if (backendPID == CheckPointPID)
1531 (errmsg_internal("ignoring cancel request for checkpoint process %d",
1535 else if (backendPID == BgWriterPID)
1538 (errmsg_internal("ignoring cancel request for bgwriter process %d",
1543 /* See if we have a matching backend */
1544 #ifndef EXEC_BACKEND
1545 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1547 bp = (Backend *) DLE_VAL(curr);
1549 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1551 bp = (Backend*) &ShmemBackendArray[i];
1553 if (bp->pid == backendPID)
1555 if (bp->cancel_key == cancelAuthCode)
1557 /* Found a match; signal that backend to cancel current op */
1559 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1561 kill(bp->pid, SIGINT);
1564 /* Right PID, wrong key: no way, Jose */
1566 (errmsg_internal("bad key in cancel request for process %d",
1572 /* No matching backend */
1574 (errmsg_internal("bad pid in cancel request for process %d",
1579 * canAcceptConnections --- check to see if database state allows connections.
1581 static enum CAC_state
1582 canAcceptConnections(void)
1584 /* Can't start backends when in startup/shutdown/recovery state. */
1585 if (Shutdown > NoShutdown)
1586 return CAC_SHUTDOWN;
1590 return CAC_RECOVERY;
1593 * Don't start too many children.
1595 * We allow more connections than we can have backends here because some
1596 * might still be authenticating; they might fail auth, or some
1597 * existing backend might exit before the auth cycle is completed. The
1598 * exact MaxBackends limit is enforced when a new backend tries to
1599 * join the shared-inval backend array.
1601 if (CountChildren() >= 2 * MaxBackends)
1609 * ConnCreate -- create a local connection data structure
1612 ConnCreate(int serverFd)
1616 if (!(port = (Port *) calloc(1, sizeof(Port))))
1619 (errcode(ERRCODE_OUT_OF_MEMORY),
1620 errmsg("out of memory")));
1624 if (StreamConnection(serverFd, port) != STATUS_OK)
1626 StreamClose(port->sock);
1633 * Precompute password salt values to use for this connection.
1634 * It's slightly annoying to do this long in advance of knowing
1635 * whether we'll need 'em or not, but we must do the random()
1636 * calls before we fork, not after. Else the postmaster's random
1637 * sequence won't get advanced, and all backends would end up
1638 * using the same salt...
1640 RandomSalt(port->cryptSalt, port->md5Salt);
1648 * ConnFree -- free a local connection data structure
1651 ConnFree(Port *conn)
1661 * ClosePostmasterPorts -- close all the postmaster's open sockets
1663 * This is called during child process startup to release file descriptors
1664 * that are not needed by that child process. The postmaster still has
1665 * them open, of course.
1668 ClosePostmasterPorts(bool pgstat_too)
1672 /* Close the listen sockets */
1673 for (i = 0; i < MAXLISTEN; i++)
1675 if (ListenSocket[i] != -1)
1677 StreamClose(ListenSocket[i]);
1678 ListenSocket[i] = -1;
1682 /* Close pgstat control sockets, unless we're starting pgstat itself */
1684 pgstat_close_sockets();
1689 * reset_shared -- reset shared memory and semaphores
1692 reset_shared(unsigned short port)
1695 * Create or re-create shared memory and semaphores.
1697 * Note: in each "cycle of life" we will normally assign the same IPC
1698 * keys (if using SysV shmem and/or semas), since the port number is
1699 * used to determine IPC keys. This helps ensure that we will clean
1700 * up dead IPC objects if the postmaster crashes and is restarted.
1702 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1707 * SIGHUP -- reread config files, and tell children to do same
1710 SIGHUP_handler(SIGNAL_ARGS)
1712 int save_errno = errno;
1714 PG_SETMASK(&BlockSig);
1716 if (Shutdown <= SmartShutdown)
1719 (errmsg("received SIGHUP, reloading configuration files")));
1720 ProcessConfigFile(PGC_SIGHUP);
1722 write_nondefault_variables(PGC_SIGHUP);
1724 SignalChildren(SIGHUP);
1729 * Tell the background writer to terminate so that we
1730 * will start a new one with a possibly changed config
1732 if (BgWriterPID != 0)
1733 kill(BgWriterPID, SIGTERM);
1736 PG_SETMASK(&UnBlockSig);
1744 * pmdie -- signal handler for processing various postmaster signals.
1749 int save_errno = errno;
1751 PG_SETMASK(&BlockSig);
1754 (errmsg_internal("postmaster received signal %d",
1755 postgres_signal_arg)));
1757 switch (postgres_signal_arg)
1764 * Wait for children to end their work and ShutdownDataBase.
1766 if (Shutdown >= SmartShutdown)
1768 Shutdown = SmartShutdown;
1770 (errmsg("received smart shutdown request")));
1772 /* Must tell bgwriter to quit, or it never will... */
1773 if (BgWriterPID != 0)
1774 kill(BgWriterPID, SIGTERM);
1776 if (DLGetHead(BackendList)) /* let reaper() handle this */
1780 * No children left. Shutdown data base system.
1782 if (StartupPID > 0 || FatalError) /* let reaper() handle
1785 if (ShutdownPID > 0)
1787 elog(PANIC, "shutdown process %d already running",
1792 ShutdownPID = ShutdownDataBase();
1800 * Abort all children with SIGTERM (rollback active transactions
1801 * and exit) and ShutdownDataBase when they are gone.
1803 if (Shutdown >= FastShutdown)
1805 Shutdown = FastShutdown;
1807 (errmsg("received fast shutdown request")));
1809 if (DLGetHead(BackendList))
1814 (errmsg("aborting any active transactions")));
1815 SignalChildren(SIGTERM);
1816 /* reaper() does the rest */
1822 * No children left. Shutdown data base system.
1824 * Unlike the previous case, it is not an error for the shutdown
1825 * process to be running already (we could get SIGTERM followed
1826 * shortly later by SIGINT).
1828 if (StartupPID > 0 || FatalError) /* let reaper() handle
1831 if (ShutdownPID == 0)
1832 ShutdownPID = ShutdownDataBase();
1838 * Immediate Shutdown:
1840 * abort all children with SIGQUIT and exit without attempt to
1841 * properly shutdown data base system.
1844 (errmsg("received immediate shutdown request")));
1845 if (ShutdownPID > 0)
1846 kill(ShutdownPID, SIGQUIT);
1848 kill(StartupPID, SIGQUIT);
1849 if (DLGetHead(BackendList))
1850 SignalChildren(SIGQUIT);
1855 PG_SETMASK(&UnBlockSig);
1861 * Reaper -- signal handler to cleanup after a backend (child) dies.
1866 int save_errno = errno;
1869 int status; /* backend exit status */
1872 union wait status; /* backend exit status */
1876 int pid; /* process id of dead backend */
1878 PG_SETMASK(&BlockSig);
1881 (errmsg_internal("reaping dead processes")));
1883 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1885 exitstatus = status;
1888 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1890 exitstatus = status.w_status;
1892 while ((pid = win32_waitpid(&exitstatus)) > 0)
1895 * We need to do this here, and not in CleanupProc, since this
1896 * is to be called on all children when we are done with them.
1897 * Could move to LogChildExit, but that seems like asking for
1900 win32_RemoveChild(pid);
1905 * Check if this child was the statistics collector. If so, try to
1906 * start a new one. (If fail, we'll try again in future cycles of
1909 if (pgstat_ispgstat(pid))
1911 LogChildExit(LOG, gettext("statistics collector process"),
1918 * Check if this child was a shutdown or startup process.
1920 if (ShutdownPID > 0 && pid == ShutdownPID)
1922 if (exitstatus != 0)
1924 LogChildExit(LOG, gettext("shutdown process"),
1928 /* Normal postmaster exit is here */
1932 if (StartupPID > 0 && pid == StartupPID)
1934 if (exitstatus != 0)
1936 LogChildExit(LOG, gettext("startup process"),
1939 (errmsg("aborting startup due to startup process failure")));
1945 * Startup succeeded - remember its ID and RedoRecPtr.
1947 * NB: this MUST happen before we fork a checkpoint or shutdown
1948 * subprocess, else they will have wrong local ThisStartUpId.
1952 FatalError = false; /* done with recovery */
1955 * Arrange for first checkpoint to occur after standard delay.
1958 checkpointed = time(NULL);
1961 * Go to shutdown mode if a shutdown request was pending.
1963 if (Shutdown > NoShutdown)
1965 if (ShutdownPID > 0)
1967 elog(PANIC, "startup process %d died while shutdown process %d already running",
1968 pid, (int) ShutdownPID);
1971 ShutdownPID = ShutdownDataBase();
1978 * Else do standard child cleanup.
1980 CleanupProc(pid, exitstatus);
1982 } /* loop over pending child-death reports */
1987 * Wait for all children exit, then reset shmem and
1990 if (DLGetHead(BackendList) || StartupPID > 0 || ShutdownPID > 0)
1993 (errmsg("all server processes terminated; reinitializing")));
1996 reset_shared(PostPortNumber);
1998 StartupPID = StartupDataBase();
2003 if (Shutdown > NoShutdown)
2005 if (DLGetHead(BackendList))
2007 if (StartupPID > 0 || ShutdownPID > 0)
2009 ShutdownPID = ShutdownDataBase();
2013 PG_SETMASK(&UnBlockSig);
2020 * CleanupProc -- cleanup after terminated backend.
2022 * Remove all local state associated with backend.
2025 CleanupProc(int pid,
2026 int exitstatus) /* child's exit status. */
2032 LogChildExit(DEBUG2, gettext("child process"), pid, exitstatus);
2035 * If a backend dies in an ugly way (i.e. exit status not 0) then we
2036 * must signal all other backends to quickdie. If exit status is zero
2037 * we assume everything is hunky dory and simply remove the backend
2038 * from the active backend list.
2040 if (exitstatus == 0)
2042 curr = DLGetHead(BackendList);
2045 bp = (Backend *) DLE_VAL(curr);
2049 ShmemBackendArrayRemove(bp->pid);
2056 curr = DLGetSucc(curr);
2059 if (pid == CheckPointPID)
2064 checkpointed = time(NULL);
2065 /* Update RedoRecPtr for future child backends */
2066 GetSavedRedoRecPtr();
2069 else if (pid == BgWriterPID)
2077 /* below here we're dealing with a non-normal exit */
2079 /* Make log entry unless we did so already */
2083 (pid == CheckPointPID) ? gettext("checkpoint process") :
2084 (pid == BgWriterPID) ? gettext("bgwriter process") :
2085 gettext("server process"),
2088 (errmsg("terminating any other active server processes")));
2091 curr = DLGetHead(BackendList);
2094 next = DLGetSucc(curr);
2095 bp = (Backend *) DLE_VAL(curr);
2099 * This backend is still alive. Unless we did so already,
2100 * tell it to commit hara-kiri.
2102 * SIGQUIT is the special signal that says exit without proc_exit
2103 * and let the user know what's going on. But if SendStop is
2104 * set (-s on command line), then we send SIGSTOP instead, so
2105 * that we can get core dumps from all backends by hand.
2110 (errmsg_internal("sending %s to process %d",
2111 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2113 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2119 * Found entry for freshly-dead backend, so remove it.
2122 ShmemBackendArrayRemove(bp->pid);
2131 if (pid == CheckPointPID)
2136 else if (pid == BgWriterPID)
2143 * Tell the collector about backend termination
2152 * Log the death of a child process.
2155 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2157 if (WIFEXITED(exitstatus))
2161 * translator: %s is a noun phrase describing a child process,
2162 * such as "server process"
2164 (errmsg("%s (PID %d) exited with exit code %d",
2165 procname, pid, WEXITSTATUS(exitstatus))));
2166 else if (WIFSIGNALED(exitstatus))
2170 * translator: %s is a noun phrase describing a child process,
2171 * such as "server process"
2173 (errmsg("%s (PID %d) was terminated by signal %d",
2174 procname, pid, WTERMSIG(exitstatus))));
2179 * translator: %s is a noun phrase describing a child process,
2180 * such as "server process"
2182 (errmsg("%s (PID %d) exited with unexpected status %d",
2183 procname, pid, exitstatus)));
2187 * Send a signal to all backend children.
2190 SignalChildren(int signal)
2196 curr = DLGetHead(BackendList);
2199 next = DLGetSucc(curr);
2200 bp = (Backend *) DLE_VAL(curr);
2202 if (bp->pid != MyProcPid)
2205 (errmsg_internal("sending signal %d to process %d",
2208 kill(bp->pid, signal);
2216 * BackendStartup -- start backend process
2218 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2221 BackendStartup(Port *port)
2223 Backend *bn; /* for backend cleanup */
2226 #ifdef LINUX_PROFILE
2227 struct itimerval prof_itimer;
2231 * Compute the cancel key that will be assigned to this backend. The
2232 * backend will have its own copy in the forked-off process' value of
2233 * MyCancelKey, so that it can transmit the key to the frontend.
2235 MyCancelKey = PostmasterRandom();
2238 * Make room for backend data structure. Better before the fork() so
2239 * we can handle failure cleanly.
2241 bn = (Backend *) malloc(sizeof(Backend));
2245 (errcode(ERRCODE_OUT_OF_MEMORY),
2246 errmsg("out of memory")));
2247 return STATUS_ERROR;
2251 * Flush stdio channels just before fork, to avoid double-output
2252 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2253 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2254 * coredump if we do. Presently stdout and stderr are the only stdio
2255 * output channels used by the postmaster, so fflush'ing them should
2261 #ifdef LINUX_PROFILE
2264 * Linux's fork() resets the profiling timer in the child process. If
2265 * we want to profile child processes then we need to save and restore
2266 * the timer setting. This is a waste of time if not profiling,
2267 * however, so only do it if commanded by specific -DLINUX_PROFILE
2270 getitimer(ITIMER_PROF, &prof_itimer);
2274 /* Specific beos actions before backend startup */
2275 beos_before_backend_startup();
2278 port->canAcceptConnections = canAcceptConnections();
2280 pid = Backend_forkexec(port);
2284 if (pid == 0) /* child */
2286 #ifdef LINUX_PROFILE
2287 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2291 /* Specific beos backend startup actions */
2292 beos_backend_startup();
2296 proc_exit(BackendRun(port));
2300 /* in parent, error */
2303 int save_errno = errno;
2306 /* Specific beos backend startup actions */
2307 beos_backend_startup_failed();
2312 (errmsg("could not fork new process for connection: %m")));
2313 report_fork_failure_to_client(port, save_errno);
2314 return STATUS_ERROR;
2317 /* in parent, normal */
2319 (errmsg_internal("forked new backend, pid=%d socket=%d",
2320 (int) pid, port->sock)));
2323 * Everything's been successful, it's safe to add this backend to our
2327 bn->cancel_key = MyCancelKey;
2329 ShmemBackendArrayAdd(bn);
2331 DLAddHead(BackendList, DLNewElem(bn));
2337 * Try to report backend fork() failure to client before we close the
2338 * connection. Since we do not care to risk blocking the postmaster on
2339 * this connection, we set the connection to non-blocking and try only once.
2341 * This is grungy special-purpose code; we cannot use backend libpq since
2342 * it's not up and running.
2345 report_fork_failure_to_client(Port *port, int errnum)
2349 /* Format the error message packet (always V2 protocol) */
2350 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2351 gettext("could not fork new process for connection: "),
2354 /* Set port to non-blocking. Don't do send() if this fails */
2355 if (!set_noblock(port->sock))
2358 send(port->sock, buffer, strlen(buffer) + 1, 0);
2363 * split_opts -- split a string of options and append it to an argv array
2365 * NB: the string is destructively modified!
2367 * Since no current POSTGRES arguments require any quoting characters,
2368 * we can use the simple-minded tactic of assuming each set of space-
2369 * delimited characters is a separate argv element.
2371 * If you don't like that, well, we *used* to pass the whole option string
2372 * as ONE argument to execl(), which was even less intelligent...
2375 split_opts(char **argv, int *argcp, char *s)
2379 while (isspace((unsigned char) *s))
2383 argv[(*argcp)++] = s;
2384 while (*s && !isspace((unsigned char) *s))
2393 * BackendInit/Run -- perform authentication [BackendInit], and if successful,
2394 * set up the backend's argument list [BackendRun] and invoke
2398 * Shouldn't return at all.
2399 * If PostgresMain() fails, return status.
2402 BackendInit(Port *port)
2407 char remote_host[NI_MAXHOST];
2408 char remote_port[NI_MAXSERV];
2409 char remote_ps_data[NI_MAXHOST];
2411 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2413 ClientAuthInProgress = true; /* limit visibility of log messages */
2415 /* We don't want the postmaster's proc_exit() handlers */
2419 * Signal handlers setting is moved to tcop/postgres...
2422 /* save start time for end of session reporting */
2423 gettimeofday(&(port->session_start),NULL);
2425 /* set these to empty in case they are needed before we set them up */
2426 port->remote_host = "";
2427 port->remote_port = "";
2428 port->commandTag = "";
2430 /* Save port etc. for ps status */
2433 /* Reset MyProcPid to new backend's pid */
2434 MyProcPid = getpid();
2437 * Initialize libpq and enable reporting of ereport errors to the
2438 * client. Must do this now because authentication uses libpq to send
2441 pq_init(); /* initialize libpq to talk to client */
2442 whereToSendOutput = Remote; /* now safe to ereport to client */
2445 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2446 * during any client authentication related communication. Otherwise
2447 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2448 * if a buggy client blocks a backend during authentication.
2450 pqsignal(SIGTERM, authdie);
2451 pqsignal(SIGQUIT, authdie);
2452 pqsignal(SIGALRM, authdie);
2453 PG_SETMASK(&AuthBlockSig);
2456 * Get the remote host name and port for logging and status display.
2458 remote_host[0] = '\0';
2459 remote_port[0] = '\0';
2460 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2461 remote_host, sizeof(remote_host),
2462 remote_port, sizeof(remote_port),
2463 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2465 getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2466 remote_host, sizeof(remote_host),
2467 remote_port, sizeof(remote_port),
2468 NI_NUMERICHOST | NI_NUMERICSERV);
2470 snprintf(remote_ps_data, sizeof(remote_ps_data),
2471 remote_port[0] == '\0' ? "%s" : "%s(%s)",
2472 remote_host, remote_port);
2474 if (Log_connections)
2476 (errmsg("connection received: host=%s port=%s",
2477 remote_host, remote_port)));
2480 * save remote_host and remote_port in port stucture
2482 port->remote_host = strdup(remote_host);
2483 port->remote_port = strdup(remote_port);
2486 * Ready to begin client interaction. We will give up and exit(0)
2487 * after a time delay, so that a broken client can't hog a connection
2488 * indefinitely. PreAuthDelay doesn't count against the time limit.
2490 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2491 elog(FATAL, "could not set timer for authorization timeout");
2494 * Receive the startup packet (which might turn out to be a cancel
2497 status = ProcessStartupPacket(port, false);
2499 if (status != STATUS_OK)
2503 * Now that we have the user and database name, we can set the process
2504 * title for ps. It's good to do this as early as possible in
2507 init_ps_display(port->user_name, port->database_name, remote_ps_data);
2508 set_ps_display("authentication");
2511 * Now perform authentication exchange.
2513 ClientAuthentication(port); /* might not return, if failure */
2516 * Done with authentication. Disable timeout, and prevent
2517 * SIGTERM/SIGQUIT again until backend startup is complete.
2519 if (!disable_sig_alarm(false))
2520 elog(FATAL, "could not disable timer for authorization timeout");
2521 PG_SETMASK(&BlockSig);
2523 if (Log_connections)
2525 (errmsg("connection authorized: user=%s database=%s",
2526 port->user_name, port->database_name)));
2529 * Don't want backend to be able to see the postmaster random number
2530 * generator state. We have to clobber the static random_seed *and*
2531 * start a new random sequence in the random() library function.
2534 gettimeofday(&now, &tz);
2535 srandom((unsigned int) now.tv_usec);
2540 BackendRun(Port *port)
2550 * Let's clean up ourselves as the postmaster child, and
2551 * close the postmaster's other sockets
2553 ClosePostmasterPorts(true);
2556 * PreAuthDelay is a debugging aid for investigating problems in the
2557 * authentication cycle: it can be set in postgresql.conf to allow
2558 * time to attach to the newly-forked backend with a debugger. (See
2559 * also the -W backend switch, which we allow clients to pass through
2560 * PGOPTIONS, but it is not honored until after authentication.)
2562 if (PreAuthDelay > 0)
2563 pg_usleep(PreAuthDelay*1000000L);
2565 /* Will exit on failure */
2570 * Now, build the argv vector that will be given to PostgresMain.
2572 * The layout of the command line is
2573 * postgres [secure switches] -p databasename [insecure switches]
2574 * where the switches after -p come from the client request.
2576 * The maximum possible number of commandline arguments that could come
2577 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2581 maxac = 10; /* for fixed args supplied below */
2582 maxac += (strlen(ExtraOptions) + 1) / 2;
2583 if (port->cmdline_options)
2584 maxac += (strlen(port->cmdline_options) + 1) / 2;
2586 av = (char **) MemoryContextAlloc(TopMemoryContext,
2587 maxac * sizeof(char *));
2590 av[ac++] = "postgres";
2593 * Pass the requested debugging level along to the backend.
2597 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2598 av[ac++] = debugbuf;
2602 * Pass any backend switches specified with -o in the postmaster's own
2603 * command line. We assume these are secure.
2605 split_opts(av, &ac, ExtraOptions);
2607 /* Tell the backend what protocol the frontend is using. */
2608 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2609 av[ac++] = protobuf;
2612 /* pass data dir before end of secure switches (-p) */
2618 * Tell the backend it is being called from the postmaster, and which
2619 * database to use. -p marks the end of secure switches.
2622 av[ac++] = port->database_name;
2625 * Pass the (insecure) option switches from the connection request.
2626 * (It's OK to mangle port->cmdline_options now.)
2628 if (port->cmdline_options)
2629 split_opts(av, &ac, port->cmdline_options);
2636 * Release postmaster's working memory context so that backend can
2637 * recycle the space. Note this does not trash *MyProcPort, because
2638 * ConnCreate() allocated that space with malloc() ... else we'd need
2639 * to copy the Port data here. Also, subsidiary data such as the
2640 * username isn't lost either; see ProcessStartupPacket().
2642 MemoryContextSwitchTo(TopMemoryContext);
2643 #ifndef EXEC_BACKEND
2644 MemoryContextDelete(PostmasterContext);
2646 PostmasterContext = NULL;
2649 * Debug: print arguments being passed to backend
2652 (errmsg_internal("%s child[%d]: starting with (",
2653 progname, getpid())));
2654 for (i = 0; i < ac; ++i)
2656 (errmsg_internal("\t%s", av[i])));
2658 (errmsg_internal(")")));
2660 ClientAuthInProgress = false; /* client_min_messages is active
2663 return (PostgresMain(ac, av, port->user_name));
2671 * SubPostmasterMain -- prepare the fork/exec'd process to be in an equivalent
2672 * state (for calling BackendRun) as a forked process.
2675 * Shouldn't return at all.
2678 SubPostmasterMain(int argc, char* argv[])
2680 unsigned long backendID;
2683 memset((void*)&port, 0, sizeof(Port));
2686 /* Do this sooner rather than later... */
2687 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2689 /* In EXEC case we will not have inherited these settings */
2690 IsPostmasterEnvironment = true;
2691 whereToSendOutput = None;
2693 /* Setup global context */
2694 MemoryContextInit();
2695 InitializeGUCOptions();
2697 /* Parse passed-in context */
2699 backendID = (unsigned long)atol(argv[argc++]);
2700 DataDir = strdup(argv[argc++]);
2702 /* Read in file-based context */
2703 read_nondefault_variables();
2704 read_backend_variables(backendID,&port);
2706 /* Remaining initialization */
2707 pgstat_init_forkexec_backend();
2709 /* FIXME: [fork/exec] Ugh */
2715 /* Attach process to shared segments */
2716 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2719 proc_exit(BackendRun(&port));
2724 * Backend_forkexec -- fork/exec off a backend process
2727 * the pid of the fork/exec'd process
2730 Backend_forkexec(Port *port)
2734 int ac = 0, bufc = 0, i;
2735 char buf[2][MAXPGPATH];
2737 if (!write_backend_variables(port))
2738 return -1; /* log made by write_backend_variables */
2740 av[ac++] = "postgres";
2741 av[ac++] = "-forkexec";
2743 /* Format up context to pass to exec'd process */
2744 snprintf(buf[bufc++],MAXPGPATH,"%lu",tmpBackendFileNum);
2745 snprintf(buf[bufc++],MAXPGPATH,"\"%s\"",DataDir);
2747 /* Add to the arg list */
2748 Assert(bufc <= lengthof(buf));
2749 for (i = 0; i < bufc; i++)
2752 /* FIXME: [fork/exec] ExtraOptions? */
2755 Assert(ac <= lengthof(av));
2758 pid = win32_forkexec(postgres_exec_path, av); /* logs on error */
2760 /* Fire off execv in child */
2761 if ((pid = fork()) == 0 && (execv(postgres_exec_path, av) == -1))
2763 * FIXME: [fork/exec] suggestions for what to do here?
2764 * Probably OK to issue error (unlike pgstat case)
2768 return pid; /* Parent returns pid */
2775 * ExitPostmaster -- cleanup
2777 * Do NOT call exit() directly --- always go through here!
2780 ExitPostmaster(int status)
2782 /* should cleanup shared memory and kill all backends */
2785 * Not sure of the semantics here. When the Postmaster dies, should
2786 * the backends all be killed? probably not.
2788 * MUST -- vadim 05-10-1999
2790 /* Should I use true instead? */
2791 ClosePostmasterPorts(false);
2797 * sigusr1_handler - handle signal conditions from child processes
2800 sigusr1_handler(SIGNAL_ARGS)
2802 int save_errno = errno;
2804 PG_SETMASK(&BlockSig);
2806 if (CheckPostmasterSignal(PMSIGNAL_DO_CHECKPOINT))
2808 if (CheckPointWarning != 0)
2811 * This only times checkpoints forced by running out of
2812 * segment files. Other checkpoints could reduce the
2813 * frequency of forced checkpoints.
2815 time_t now = time(NULL);
2817 if (LastSignalledCheckpoint != 0)
2819 int elapsed_secs = now - LastSignalledCheckpoint;
2821 if (elapsed_secs < CheckPointWarning)
2823 (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
2825 errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
2827 LastSignalledCheckpoint = now;
2831 * Request to schedule a checkpoint
2833 * Ignore request if checkpoint is already running or checkpointing
2834 * is currently disabled
2836 if (CheckPointPID == 0 && checkpointed &&
2837 StartupPID == 0 && Shutdown == NoShutdown &&
2838 !FatalError && random_seed != 0)
2840 CheckPointPID = CheckPointDataBase();
2841 /* note: if fork fails, CheckPointPID stays 0; nothing happens */
2845 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2848 * Password or group file has changed.
2854 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2857 * Send SIGUSR2 to all children (triggers AsyncNotifyHandler). See
2858 * storage/ipc/sinvaladt.c for the use of this.
2860 if (Shutdown == NoShutdown)
2861 SignalChildren(SIGUSR2);
2864 PG_SETMASK(&UnBlockSig);
2871 * Dummy signal handler
2873 * We use this for signals that we don't actually use in the postmaster,
2874 * but we do use in backends. If we SIG_IGN such signals in the postmaster,
2875 * then a newly started backend might drop a signal that arrives before it's
2876 * able to reconfigure its signal processing. (See notes in postgres.c.)
2879 dummy_handler(SIGNAL_ARGS)
2885 * CharRemap: given an int in range 0..61, produce textual encoding of it
2886 * per crypt(3) conventions.
2910 RandomSalt(char *cryptSalt, char *md5Salt)
2912 long rand = PostmasterRandom();
2914 cryptSalt[0] = CharRemap(rand % 62);
2915 cryptSalt[1] = CharRemap(rand / 62);
2918 * It's okay to reuse the first random value for one of the MD5 salt
2919 * bytes, since only one of the two salts will be sent to the client.
2920 * After that we need to compute more random bits.
2922 * We use % 255, sacrificing one possible byte value, so as to ensure
2923 * that all bits of the random() value participate in the result.
2924 * While at it, add one to avoid generating any null bytes.
2926 md5Salt[0] = (rand % 255) + 1;
2927 rand = PostmasterRandom();
2928 md5Salt[1] = (rand % 255) + 1;
2929 rand = PostmasterRandom();
2930 md5Salt[2] = (rand % 255) + 1;
2931 rand = PostmasterRandom();
2932 md5Salt[3] = (rand % 255) + 1;
2939 PostmasterRandom(void)
2941 static bool initialized = false;
2945 Assert(random_seed != 0);
2946 srandom(random_seed);
2954 * Count up number of child processes.
2963 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2965 bp = (Backend *) DLE_VAL(curr);
2966 if (bp->pid != MyProcPid)
2969 /* Checkpoint and bgwriter will be in the list, discount them */
2970 if (CheckPointPID != 0)
2972 if (BgWriterPID != 0)
2978 * Fire off a subprocess for startup/shutdown/checkpoint/bgwriter.
2980 * Return value of SSDataBase is subprocess' PID, or 0 if failed to start subprocess
2981 * (0 is returned only for checkpoint/bgwriter cases).
2983 * note: in the EXEC_BACKEND case, we delay the fork until argument list has been
2986 NON_EXEC_STATIC void
2987 SSDataBaseInit(int xlop)
2989 const char *statmsg;
2991 IsUnderPostmaster = true; /* we are a postmaster subprocess
2995 /* In EXEC case we will not have inherited these settings */
2996 IsPostmasterEnvironment = true;
2997 whereToSendOutput = None;
3000 MyProcPid = getpid(); /* reset MyProcPid */
3002 /* Lose the postmaster's on-exit routines and port connections */
3006 * Identify myself via ps
3010 case BS_XLOG_STARTUP:
3011 statmsg = "startup subprocess";
3013 case BS_XLOG_CHECKPOINT:
3014 statmsg = "checkpoint subprocess";
3016 case BS_XLOG_BGWRITER:
3017 statmsg = "bgwriter subprocess";
3019 case BS_XLOG_SHUTDOWN:
3020 statmsg = "shutdown subprocess";
3023 statmsg = "??? subprocess";
3026 init_ps_display(statmsg, "", "");
3032 SSDataBase(int xlop)
3036 #ifndef EXEC_BACKEND
3037 #ifdef LINUX_PROFILE
3038 struct itimerval prof_itimer;
3042 char ddirbuf[MAXPGPATH];
3048 #ifndef EXEC_BACKEND
3049 #ifdef LINUX_PROFILE
3050 /* see comments in BackendRun */
3051 getitimer(ITIMER_PROF, &prof_itimer);
3055 /* Specific beos actions before backend startup */
3056 beos_before_backend_startup();
3059 /* Non EXEC_BACKEND case; fork here */
3060 if ((pid = fork()) == 0) /* child */
3068 #ifndef EXEC_BACKEND
3069 #ifdef LINUX_PROFILE
3070 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3074 /* Specific beos actions after backend startup */
3075 beos_backend_startup();
3078 /* Close the postmaster's sockets */
3079 ClosePostmasterPorts(true);
3081 SSDataBaseInit(xlop);
3083 if (!write_backend_variables(NULL))
3084 return -1; /* log issued by write_backend_variables */
3087 /* Set up command-line arguments for subprocess */
3088 av[ac++] = "postgres";
3093 snprintf(nbbuf, sizeof(nbbuf), "-B%d", NBuffers);
3096 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
3100 /* pass data dir before end of secure switches (-p) */
3101 snprintf(ddirbuf, MAXPGPATH, "\"%s\"", DataDir);
3105 /* and the backend identifier + dbname */
3106 snprintf(idbuf, sizeof(idbuf), "-p%lu,template1", tmpBackendFileNum);
3110 av[ac++] = "template1";
3115 Assert(ac < lengthof(av));
3118 /* EXEC_BACKEND case; fork/exec here */
3120 pid = win32_forkexec(postgres_exec_path, av); /* logs on error */
3122 if ((pid = fork()) == 0 && (execv(postgres_exec_path, av) == -1))
3125 elog(ERROR, "unable to execv in SSDataBase: %m");
3130 BootstrapMain(ac, av);
3138 #ifndef EXEC_BACKEND
3140 /* Specific beos actions before backend startup */
3141 beos_backend_startup_failed();
3146 case BS_XLOG_STARTUP:
3148 (errmsg("could not fork startup process: %m")));
3150 case BS_XLOG_CHECKPOINT:
3152 (errmsg("could not fork checkpoint process: %m")));
3154 case BS_XLOG_BGWRITER:
3156 (errmsg("could not fork bgwriter process: %m")));
3158 case BS_XLOG_SHUTDOWN:
3160 (errmsg("could not fork shutdown process: %m")));
3164 (errmsg("could not fork process: %m")));
3169 * fork failure is fatal during startup/shutdown, but there's no
3170 * need to choke if a routine checkpoint or starting a background
3173 if (xlop == BS_XLOG_CHECKPOINT)
3175 if (xlop == BS_XLOG_BGWRITER)
3181 * The startup and shutdown processes are not considered normal
3182 * backends, but the checkpoint and bgwriter processes are.
3183 * They must be added to the list of backends.
3185 if (xlop == BS_XLOG_CHECKPOINT || xlop == BS_XLOG_BGWRITER)
3187 if (!(bn = (Backend *) malloc(sizeof(Backend))))
3190 (errcode(ERRCODE_OUT_OF_MEMORY),
3191 errmsg("out of memory")));
3196 bn->cancel_key = PostmasterRandom();
3198 ShmemBackendArrayAdd(bn);
3200 DLAddHead(BackendList, DLNewElem(bn));
3203 * Since this code is executed periodically, it's a fine place to
3204 * do other actions that should happen every now and then on no
3205 * particular schedule. Such as...
3208 TouchSocketLockFile();
3216 * Create the opts file
3219 CreateOptsFile(int argc, char *argv[], char *fullprogname)
3221 char filename[MAXPGPATH];
3225 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3227 if ((fp = fopen(filename, "w")) == NULL)
3229 elog(LOG, "could not create file \"%s\": %m", filename);
3233 fprintf(fp, "%s", fullprogname);
3234 for (i = 1; i < argc; i++)
3235 fprintf(fp, " '%s'", argv[i]);
3240 elog(LOG, "could not write file \"%s\": %m", filename);
3248 * This should be used only for reporting "interactive" errors (essentially,
3249 * bogus arguments on the command line). Once the postmaster is launched,
3250 * use ereport. In particular, don't use this for anything that occurs
3251 * after pmdaemonize.
3254 postmaster_error(const char *fmt,...)
3258 fprintf(stderr, "%s: ", progname);
3260 vfprintf(stderr, gettext(fmt), ap);
3262 fprintf(stderr, "\n");
3269 * The following need to be available to the read/write_backend_variables
3272 #include "storage/spin.h"
3273 extern XLogRecPtr RedoRecPtr;
3274 extern XLogwrtResult LogwrtResult;
3275 extern slock_t *ShmemLock;
3276 extern slock_t *ShmemIndexLock;
3277 extern void *ShmemIndexAlloc;
3278 typedef struct LWLock LWLock;
3279 extern LWLock *LWLockArray;
3280 extern slock_t *ProcStructLock;
3281 extern int pgStatSock;
3283 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3284 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3285 #define get_tmp_backend_file_name(buf,id) \
3289 "%s/%s/%s.backend_var.%lu", \
3291 PG_TEMP_FILES_DIR, \
3292 PG_TEMP_FILE_PREFIX, \
3297 write_backend_variables(Port *port)
3299 char filename[MAXPGPATH];
3301 get_tmp_backend_file_name(filename,++tmpBackendFileNum);
3304 fp = AllocateFile(filename, PG_BINARY_W);
3307 /* As per OpenTemporaryFile... */
3308 char dirname[MAXPGPATH];
3309 sprintf(dirname,"%s/%s",DataDir,PG_TEMP_FILES_DIR);
3310 mkdir(dirname, S_IRWXU);
3312 fp = AllocateFile(filename, PG_BINARY_W);
3316 (errcode_for_file_access(),
3317 errmsg("could not write to file \"%s\": %m", filename)));
3325 write_var(port->sock,fp);
3326 write_var(port->proto,fp);
3327 write_var(port->laddr,fp);
3328 write_var(port->raddr,fp);
3329 write_var(port->canAcceptConnections,fp);
3330 write_var(port->cryptSalt,fp);
3331 write_var(port->md5Salt,fp);
3333 write_var(MyCancelKey,fp);
3335 write_var(RedoRecPtr,fp);
3336 write_var(LogwrtResult,fp);
3338 write_var(UsedShmemSegID,fp);
3339 write_var(UsedShmemSegAddr,fp);
3341 write_var(ShmemLock,fp);
3342 write_var(ShmemIndexLock,fp);
3343 write_var(ShmemVariableCache,fp);
3344 write_var(ShmemIndexAlloc,fp);
3345 write_var(ShmemBackendArray,fp);
3347 write_var(LWLockArray,fp);
3348 write_var(ProcStructLock,fp);
3349 write_var(pgStatSock,fp);
3351 write_var(PreAuthDelay,fp);
3352 write_var(debug_flag,fp);
3353 write_var(PostmasterPid,fp);
3359 (errcode_for_file_access(),
3360 errmsg("could not write to file \"%s\": %m", filename)));
3368 read_backend_variables(unsigned long id, Port *port)
3370 char filename[MAXPGPATH];
3372 get_tmp_backend_file_name(filename,id);
3375 fp = AllocateFile(filename, PG_BINARY_R);
3379 (errcode_for_file_access(),
3380 errmsg("could not read from backend_variables file \"%s\": %m", filename)));
3387 read_var(port->sock,fp);
3388 read_var(port->proto,fp);
3389 read_var(port->laddr,fp);
3390 read_var(port->raddr,fp);
3391 read_var(port->canAcceptConnections,fp);
3392 read_var(port->cryptSalt,fp);
3393 read_var(port->md5Salt,fp);
3395 read_var(MyCancelKey,fp);
3397 read_var(RedoRecPtr,fp);
3398 read_var(LogwrtResult,fp);
3400 read_var(UsedShmemSegID,fp);
3401 read_var(UsedShmemSegAddr,fp);
3403 read_var(ShmemLock,fp);
3404 read_var(ShmemIndexLock,fp);
3405 read_var(ShmemVariableCache,fp);
3406 read_var(ShmemIndexAlloc,fp);
3407 read_var(ShmemBackendArray,fp);
3409 read_var(LWLockArray,fp);
3410 read_var(ProcStructLock,fp);
3411 read_var(pgStatSock,fp);
3413 read_var(PreAuthDelay,fp);
3414 read_var(debug_flag,fp);
3415 read_var(PostmasterPid,fp);
3419 if (unlink(filename) != 0)
3421 (errcode_for_file_access(),
3422 errmsg("could not remove file \"%s\": %m", filename)));
3426 size_t ShmemBackendArraySize(void)
3428 return (NUM_BACKENDARRAY_ELEMS*sizeof(Backend));
3431 void ShmemBackendArrayAllocation(void)
3433 size_t size = ShmemBackendArraySize();
3434 ShmemBackendArray = (Backend*)ShmemAlloc(size);
3435 memset(ShmemBackendArray, 0, size);
3438 static void ShmemBackendArrayAdd(Backend *bn)
3441 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3443 /* Find an empty slot */
3444 if (ShmemBackendArray[i].pid == 0)
3446 ShmemBackendArray[i] = *bn;
3452 (errmsg_internal("unable to add backend entry")));
3455 static void ShmemBackendArrayRemove(pid_t pid)
3458 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3460 if (ShmemBackendArray[i].pid == pid)
3462 /* Mark the slot as empty */
3463 ShmemBackendArray[i].pid = 0;
3469 (errmsg_internal("unable to find backend entry with pid %d",
3477 pid_t win32_forkexec(const char* path, char *argv[])
3480 PROCESS_INFORMATION pi;
3483 char cmdLine[MAXPGPATH];
3484 HANDLE childHandleCopy;
3485 HANDLE waiterThread;
3487 /* Format the cmd line */
3488 snprintf(cmdLine,sizeof(cmdLine),"\"%s\"",path);
3490 while (argv[++i] != NULL)
3492 /* FIXME: [fork/exec] some strlen checks might be prudent here */
3493 strcat(cmdLine," ");
3494 strcat(cmdLine,argv[i]);
3498 * The following snippet can disappear when we consistently
3499 * use forward slashes.
3502 while (*(p++) != '\0')
3503 if (*p == '/') *p = '\\';
3505 memset(&pi,0,sizeof(pi));
3506 memset(&si,0,sizeof(si));
3508 if (!CreateProcess(NULL,cmdLine,NULL,NULL,TRUE,0,NULL,NULL,&si,&pi))
3510 elog(ERROR,"CreateProcess call failed (%i): %m",(int)GetLastError());
3514 if (!IsUnderPostmaster)
3515 /* We are the Postmaster creating a child... */
3516 win32_AddChild(pi.dwProcessId,pi.hProcess);
3518 if (!DuplicateHandle(GetCurrentProcess(),
3520 GetCurrentProcess(),
3524 DUPLICATE_SAME_ACCESS))
3526 (errmsg_internal("failed to duplicate child handle: %i",(int)GetLastError())));
3527 waiterThread = CreateThread(NULL, 64*1024, win32_sigchld_waiter, (LPVOID)childHandleCopy, 0, NULL);
3530 (errmsg_internal("failed to create sigchld waiter thread: %i",(int)GetLastError())));
3531 CloseHandle(waiterThread);
3533 if (IsUnderPostmaster)
3534 CloseHandle(pi.hProcess);
3535 CloseHandle(pi.hThread);
3537 return pi.dwProcessId;
3541 * Note: The following three functions must not be interrupted (eg. by signals).
3542 * As the Postgres Win32 signalling architecture (currently) requires polling,
3543 * or APC checking functions which aren't used here, this is not an issue.
3545 * We keep two separate arrays, instead of a single array of pid/HANDLE structs,
3546 * to avoid having to re-create a handle array for WaitForMultipleObjects on
3547 * each call to win32_waitpid.
3550 static void win32_AddChild(pid_t pid, HANDLE handle)
3552 Assert(win32_childPIDArray && win32_childHNDArray);
3553 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3555 win32_childPIDArray[win32_numChildren] = pid;
3556 win32_childHNDArray[win32_numChildren] = handle;
3557 ++win32_numChildren;
3561 (errmsg_internal("unable to add child entry with pid %lu",
3565 static void win32_RemoveChild(pid_t pid)
3568 Assert(win32_childPIDArray && win32_childHNDArray);
3570 for (i = 0; i < win32_numChildren; i++)
3572 if (win32_childPIDArray[i] == pid)
3574 CloseHandle(win32_childHNDArray[i]);
3576 /* Swap last entry into the "removed" one */
3577 --win32_numChildren;
3578 win32_childPIDArray[i] = win32_childPIDArray[win32_numChildren];
3579 win32_childHNDArray[i] = win32_childHNDArray[win32_numChildren];
3585 (errmsg_internal("unable to find child entry with pid %lu",
3589 static pid_t win32_waitpid(int *exitstatus)
3591 Assert(win32_childPIDArray && win32_childHNDArray);
3592 elog(DEBUG3,"waiting on %lu children",win32_numChildren);
3594 if (win32_numChildren > 0)
3597 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't
3598 * want to run queued APCs here.
3602 DWORD ret = WaitForMultipleObjects(win32_numChildren,win32_childHNDArray,FALSE,0);
3608 (errmsg_internal("failed to wait on %lu children: %i",
3609 win32_numChildren,(int)GetLastError())));
3610 /* Fall through to WAIT_TIMEOUTs return */
3613 /* No children have finished */
3617 /* Get the exit code, and return the PID of, the respective process */
3618 index = ret-WAIT_OBJECT_0;
3619 Assert(index >= 0 && index < win32_numChildren);
3620 if (!GetExitCodeProcess(win32_childHNDArray[index],&exitCode))
3622 * If we get this far, this should never happen, but, then again...
3623 * No choice other than to assume a catastrophic failure.
3626 (errmsg_internal("failed to get exit code for child %lu",
3627 win32_childPIDArray[index])));
3628 *exitstatus = (int)exitCode;
3629 return win32_childPIDArray[index];
3637 /* Note! Code belows executes on separate threads, one for
3638 each child process created */
3639 static DWORD WINAPI win32_sigchld_waiter(LPVOID param) {
3640 HANDLE procHandle = (HANDLE)param;
3642 DWORD r = WaitForSingleObject(procHandle, INFINITE);
3643 if (r == WAIT_OBJECT_0)
3644 pg_queue_signal(SIGCHLD);
3646 fprintf(stderr,"ERROR: Failed to wait on child process handle: %i\n",(int)GetLastError());
3647 CloseHandle(procHandle);