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 1.362 2004/01/26 22:59:53 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 */
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 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;
186 const bool ExecBackend = false;
190 * Set by the -o option
192 static char ExtraOptions[MAXPGPATH];
195 * These globals control the behavior of the postmaster in case some
196 * backend dumps core. Normally, it kills all peers of the dead backend
197 * and reinitializes shared memory. By specifying -s or -n, we can have
198 * the postmaster stop (rather than kill) peers and not reinitialize
199 * shared data structures.
201 static bool Reinit = true;
202 static int SendStop = false;
204 /* still more option variables */
205 bool NetServer = false; /* listen on TCP/IP */
206 bool EnableSSL = false;
207 bool SilentMode = false; /* silent mode (-S) */
209 int PreAuthDelay = 0;
210 int AuthenticationTimeout = 60;
211 int CheckPointTimeout = 300;
212 int CheckPointWarning = 30;
213 time_t LastSignalledCheckpoint = 0;
215 bool log_hostname; /* for ps display */
217 bool Log_connections = false;
218 bool Db_user_namespace = false;
220 char *rendezvous_name;
222 /* For FNCTL_NONBLOCK */
223 #if defined(WIN32) || defined(__BEOS__)
224 long ioctlsocket_ret=1;
227 /* list of library:init-function to be preloaded */
228 char *preload_libraries_string = NULL;
230 /* Startup/shutdown state */
231 static pid_t StartupPID = 0,
235 static time_t checkpointed = 0;
238 #define SmartShutdown 1
239 #define FastShutdown 2
241 static int Shutdown = NoShutdown;
243 static bool FatalError = false; /* T if recovering from backend crash */
245 bool ClientAuthInProgress = false; /* T during new-client
249 * State for assigning random salts and cancel keys.
250 * Also, the global MyCancelKey passes the cancel key assigned to a given
251 * backend from the postmaster to that backend (via fork).
254 static unsigned int random_seed = 0;
256 static int debug_flag = 0;
262 #ifdef HAVE_INT_OPTRESET
267 * postmaster.c - function prototypes
269 static void pmdaemonize(int argc, char *argv[]);
270 static Port *ConnCreate(int serverFd);
271 static void ConnFree(Port *port);
272 static void reset_shared(unsigned short port);
273 static void SIGHUP_handler(SIGNAL_ARGS);
274 static void pmdie(SIGNAL_ARGS);
275 static void reaper(SIGNAL_ARGS);
276 static void sigusr1_handler(SIGNAL_ARGS);
277 static void dummy_handler(SIGNAL_ARGS);
278 static void CleanupProc(int pid, int exitstatus);
279 static void LogChildExit(int lev, const char *procname,
280 int pid, int exitstatus);
281 static void BackendInit(Port *port);
282 static int BackendRun(Port *port);
283 static void ExitPostmaster(int status);
284 static void usage(const char *);
285 static int ServerLoop(void);
286 static int BackendStartup(Port *port);
287 static int ProcessStartupPacket(Port *port, bool SSLdone);
288 static void processCancelRequest(Port *port, void *pkt);
289 static int initMasks(fd_set *rmask);
290 static void report_fork_failure_to_client(Port *port, int errnum);
291 static enum CAC_state canAcceptConnections(void);
292 static long PostmasterRandom(void);
293 static void RandomSalt(char *cryptSalt, char *md5Salt);
294 static void SignalChildren(int signal);
295 static int CountChildren(void);
296 static bool CreateOptsFile(int argc, char *argv[]);
297 NON_EXEC_STATIC void SSDataBaseInit(int xlop);
298 static pid_t SSDataBase(int xlop);
300 postmaster_error(const char *fmt,...)
301 /* This lets gcc check the format string for consistency. */
302 __attribute__((format(printf, 1, 2)));
306 pid_t win32_forkexec(const char* path, char *argv[]);
308 static void win32_AddChild(pid_t pid, HANDLE handle);
309 static void win32_RemoveChild(pid_t pid);
310 static pid_t win32_waitpid(int *exitstatus);
312 static pid_t *win32_childPIDArray;
313 static HANDLE *win32_childHNDArray;
314 static unsigned long win32_numChildren = 0;
317 static pid_t Backend_forkexec(Port *port);
319 static unsigned long tmpBackendFileNum = 0;
320 void read_backend_variables(unsigned long id, Port *port);
321 static bool write_backend_variables(Port *port);
323 size_t ShmemBackendArraySize(void);
324 void ShmemBackendArrayAllocation(void);
325 static void ShmemBackendArrayAdd(Backend *bn);
326 static void ShmemBackendArrayRemove(pid_t pid);
329 #define StartupDataBase() SSDataBase(BS_XLOG_STARTUP)
330 #define CheckPointDataBase() SSDataBase(BS_XLOG_CHECKPOINT)
331 #define StartBackgroundWriter() SSDataBase(BS_XLOG_BGWRITER)
332 #define ShutdownDataBase() SSDataBase(BS_XLOG_SHUTDOWN)
336 checkDataDir(const char *checkdir)
338 char path[MAXPGPATH];
340 struct stat stat_buf;
342 if (checkdir == NULL)
345 gettext("%s does not know where to find the database system data.\n"
346 "You must specify the directory that contains the database system\n"
347 "either by specifying the -D invocation option or by setting the\n"
348 "PGDATA environment variable.\n"),
353 if (stat(checkdir, &stat_buf) == -1)
357 (errcode_for_file_access(),
358 errmsg("data directory \"%s\" does not exist",
362 (errcode_for_file_access(),
363 errmsg("could not read permissions of directory \"%s\": %m",
368 * Check if the directory has group or world access. If so, reject.
370 * XXX temporarily suppress check when on Windows, because there may not
371 * be proper support for Unix-y file permissions. Need to think of a
372 * reasonable check to apply on Windows.
374 #if !defined(__CYGWIN__) && !defined(WIN32)
375 if (stat_buf.st_mode & (S_IRWXG | S_IRWXO))
377 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
378 errmsg("data directory \"%s\" has group or world access",
380 errdetail("Permissions should be u=rwx (0700).")));
383 /* Look for PG_VERSION before looking for pg_control */
384 ValidatePgVersion(checkdir);
386 snprintf(path, sizeof(path), "%s/global/pg_control", checkdir);
388 fp = AllocateFile(path, PG_BINARY_R);
392 gettext("%s: could not find the database system\n"
393 "Expected to find it in the directory \"%s\",\n"
394 "but could not open file \"%s\": %s\n"),
395 progname, checkdir, path, strerror(errno));
402 #ifdef USE_RENDEZVOUS
404 /* reg_reply -- empty callback function for DNSServiceRegistrationCreate() */
406 reg_reply(DNSServiceRegistrationReplyErrorType errorCode, void *context)
413 PostmasterMain(int argc, char *argv[])
417 char original_extraoptions[MAXPGPATH];
418 char *potential_DataDir = NULL;
421 *original_extraoptions = '\0';
425 IsPostmasterEnvironment = true;
428 * Catch standard options before doing much else. This even works on
429 * systems without getopt_long.
433 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
438 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
440 puts("postmaster (PostgreSQL) " PG_VERSION);
446 * for security, no dir or file created can be group or other
449 umask((mode_t) 0077);
451 MyProcPid = PostmasterPid = getpid();
454 * Fire up essential subsystems: memory management
459 * By default, palloc() requests in the postmaster will be allocated
460 * in the PostmasterContext, which is space that can be recycled by
461 * backends. Allocated data that needs to be available to backends
462 * should be allocated in TopMemoryContext.
464 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
466 ALLOCSET_DEFAULT_MINSIZE,
467 ALLOCSET_DEFAULT_INITSIZE,
468 ALLOCSET_DEFAULT_MAXSIZE);
469 MemoryContextSwitchTo(PostmasterContext);
471 IgnoreSystemIndexes(false);
476 InitializeGUCOptions();
478 potential_DataDir = getenv("PGDATA"); /* default value */
482 while ((opt = getopt(argc, argv, "A:a:B:b:c:D:d:Fh:ik:lm:MN:no:p:Ss-:")) != -1)
487 #ifdef USE_ASSERT_CHECKING
488 SetConfigOption("debug_assertions", optarg, PGC_POSTMASTER, PGC_S_ARGV);
490 postmaster_error("assert checking is not compiled in");
494 /* Can no longer set authentication method. */
497 SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
500 /* Can no longer set the backend executable file to use. */
503 potential_DataDir = optarg;
507 /* Turn on debugging for the postmaster. */
508 char *debugstr = palloc(strlen("debug") + strlen(optarg) + 1);
510 sprintf(debugstr, "debug%s", optarg);
511 SetConfigOption("log_min_messages", debugstr,
512 PGC_POSTMASTER, PGC_S_ARGV);
514 debug_flag = atoi(optarg);
518 SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
521 SetConfigOption("virtual_host", optarg, PGC_POSTMASTER, PGC_S_ARGV);
524 SetConfigOption("tcpip_socket", "true", PGC_POSTMASTER, PGC_S_ARGV);
527 SetConfigOption("unix_socket_directory", optarg, PGC_POSTMASTER, PGC_S_ARGV);
531 SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
535 /* Multiplexed backends no longer supported. */
540 * ignore this flag. This may be passed in because the
541 * program was run as 'postgres -M' instead of
546 /* The max number of backends to start. */
547 SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
550 /* Don't reinit shared mem after abnormal exit */
556 * Other options to pass to the backend on the command
557 * line -- useful only for debugging.
559 strcat(ExtraOptions, " ");
560 strcat(ExtraOptions, optarg);
561 strcpy(original_extraoptions, optarg);
564 SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
569 * Start in 'S'ilent mode (disassociate from controlling
570 * tty). You may also think of this as 'S'ysV mode since
571 * it's most badly needed on SysV-derived systems like
574 SetConfigOption("silent_mode", "true", PGC_POSTMASTER, PGC_S_ARGV);
579 * In the event that some backend dumps core, send
580 * SIGSTOP, rather than SIGQUIT, to all its peers. This
581 * lets the wily post_hacker collect core dumps from
592 ParseLongOption(optarg, &name, &value);
597 (errcode(ERRCODE_SYNTAX_ERROR),
598 errmsg("--%s requires a value",
602 (errcode(ERRCODE_SYNTAX_ERROR),
603 errmsg("-c %s requires a value",
607 SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
616 gettext("Try \"%s --help\" for more information.\n"),
623 * Postmaster accepts no non-option switch arguments.
627 postmaster_error("invalid argument: \"%s\"", argv[optind]);
629 gettext("Try \"%s --help\" for more information.\n"),
635 * Now we can set the data directory, and then read postgresql.conf.
637 checkDataDir(potential_DataDir); /* issues error messages */
638 SetDataDir(potential_DataDir);
640 ProcessConfigFile(PGC_POSTMASTER);
642 write_nondefault_variables(PGC_POSTMASTER);
646 * Check for invalid combinations of GUC settings.
648 if (NBuffers < 2 * MaxBackends || NBuffers < 16)
651 * Do not accept -B so small that backends are likely to starve
652 * for lack of buffers. The specific choices here are somewhat
655 postmaster_error("the number of buffers (-B) must be at least twice the number of allowed connections (-N) and at least 16");
659 if (ReservedBackends >= MaxBackends)
661 postmaster_error("superuser_reserved_connections must be less than max_connections");
666 * Other one-time internal sanity checks can go here.
668 if (!CheckDateTokenTables())
670 postmaster_error("invalid datetoken tables, please fix");
675 * Now that we are done processing the postmaster arguments, reset
676 * getopt(3) library so that it will work correctly in subprocesses.
679 #ifdef HAVE_INT_OPTRESET
680 optreset = 1; /* some systems need this too */
683 /* For debugging: display postmaster environment */
685 extern char **environ;
689 (errmsg_internal("%s: PostmasterMain: initial environ dump:",
692 (errmsg_internal("-----------------------------------------")));
693 for (p = environ; *p; ++p)
695 (errmsg_internal("\t%s", *p)));
697 (errmsg_internal("-----------------------------------------")));
701 * On some systems our dynloader code needs the executable's pathname.
703 if (FindExec(pg_pathname, progname, "postgres") < 0)
705 (errmsg("%s: could not locate postgres executable",
709 * Initialize SSL library, if specified.
712 if (EnableSSL && !NetServer)
714 postmaster_error("TCP/IP connections must be enabled for SSL");
722 * process any libraries that should be preloaded and optionally
725 if (preload_libraries_string)
726 process_preload_libraries(preload_libraries_string);
729 * Fork away from controlling terminal, if -S specified.
731 * Must do this before we grab any interlock files, else the interlocks
732 * will show the wrong PID.
735 pmdaemonize(argc, argv);
738 * Create lockfile for data directory.
740 * We want to do this before we try to grab the input sockets, because
741 * the data directory interlock is more reliable than the socket-file
742 * interlock (thanks to whoever decided to put socket files in /tmp
743 * :-(). For the same reason, it's best to grab the TCP socket before
746 CreateDataDirLockFile(DataDir, true);
749 * Remove old temporary files. At this point there can be no other
750 * Postgres processes running in this directory, so this should be
756 * Establish input sockets.
758 for (i = 0; i < MAXLISTEN; i++)
759 ListenSocket[i] = -1;
763 if (VirtualHost && VirtualHost[0])
769 curhost = VirtualHost;
772 while (*curhost == ' ') /* skip any extra spaces */
774 if (*curhost == '\0')
776 endptr = strchr(curhost, ' ');
782 status = StreamServerPort(AF_UNSPEC, curhost,
783 (unsigned short) PostPortNumber,
785 ListenSocket, MAXLISTEN);
786 if (status != STATUS_OK)
788 (errmsg("could not create listen socket for \"%s\"",
793 curhost = endptr + 1;
801 status = StreamServerPort(AF_UNSPEC, NULL,
802 (unsigned short) PostPortNumber,
804 ListenSocket, MAXLISTEN);
805 if (status != STATUS_OK)
807 (errmsg("could not create TCP/IP listen socket")));
810 #ifdef USE_RENDEZVOUS
811 if (rendezvous_name != NULL)
813 DNSServiceRegistrationCreate(rendezvous_name,
816 htonl(PostPortNumber),
818 (DNSServiceRegistrationReply) reg_reply,
824 #ifdef HAVE_UNIX_SOCKETS
825 status = StreamServerPort(AF_UNIX, NULL,
826 (unsigned short) PostPortNumber,
828 ListenSocket, MAXLISTEN);
829 if (status != STATUS_OK)
831 (errmsg("could not create Unix-domain socket")));
837 * Set up shared memory and semaphores.
839 reset_shared(PostPortNumber);
842 * Initialize the list of active backends.
844 BackendList = DLNewList();
848 * Initialize the child pid/HANDLE arrays
850 /* FIXME: [fork/exec] Ideally, we would resize these arrays with changes
851 * in MaxBackends, but this'll do as a first order solution.
853 win32_childPIDArray = (HANDLE*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(pid_t));
854 win32_childHNDArray = (HANDLE*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(HANDLE));
855 if (!win32_childPIDArray || !win32_childHNDArray)
857 (errcode(ERRCODE_OUT_OF_MEMORY),
858 errmsg("out of memory")));
862 * Record postmaster options. We delay this till now to avoid
863 * recording bogus options (eg, NBuffers too high for available
866 if (!CreateOptsFile(argc, argv))
870 * Set up signal handlers for the postmaster process.
872 * CAUTION: when changing this list, check for side-effects on the signal
873 * handling setup of child processes. See tcop/postgres.c,
874 * bootstrap/bootstrap.c, and postmaster/pgstat.c.
877 PG_SETMASK(&BlockSig);
879 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
880 * children do same */
881 pqsignal(SIGINT, pmdie); /* send SIGTERM and ShutdownDataBase */
882 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
883 pqsignal(SIGTERM, pmdie); /* wait for children and ShutdownDataBase */
884 pqsignal(SIGALRM, SIG_IGN); /* ignored */
885 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
886 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
887 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
888 pqsignal(SIGCHLD, reaper); /* handle child termination */
889 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
890 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
891 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
893 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
897 * Reset whereToSendOutput from Debug (its starting state) to None.
898 * This prevents ereport from sending log messages to stderr unless
899 * the syslog/stderr switch permits. We don't do this until the
900 * postmaster is fully launched, since startup failures may as well be
901 * reported to stderr.
903 whereToSendOutput = None;
906 * On many platforms, the first call of localtime() incurs significant
907 * overhead to load timezone info from the system configuration files.
908 * By doing it once in the postmaster, we avoid having to do it in
909 * every started child process. The savings are not huge, but they
913 time_t now = time(NULL);
915 (void) localtime(&now);
919 * Initialize and try to startup the statistics collector process
925 * Load cached files for client authentication.
933 * We're ready to rock and roll...
935 StartupPID = StartupDataBase();
937 status = ServerLoop();
940 * ServerLoop probably shouldn't ever return, but if it does, close
943 ExitPostmaster(status != STATUS_OK);
945 return 0; /* not reached */
949 pmdaemonize(int argc, char *argv[])
955 struct itimerval prof_itimer;
959 /* see comments in BackendRun */
960 getitimer(ITIMER_PROF, &prof_itimer);
964 /* FIXME: [fork/exec] to be implemented? */
969 if (pid == (pid_t) -1)
971 postmaster_error("could not fork background process: %s",
977 /* Parent should just exit, without doing any atexit cleanup */
982 setitimer(ITIMER_PROF, &prof_itimer, NULL);
985 MyProcPid = getpid(); /* reset MyProcPid to child */
987 /* GH: If there's no setsid(), we hopefully don't need silent mode.
988 * Until there's a better solution.
993 postmaster_error("could not dissociate from controlling TTY: %s",
998 i = open(NULL_DEV, O_RDWR | PG_BINARY);
1008 * Print out help message
1011 usage(const char *progname)
1013 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
1014 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
1015 printf(gettext("Options:\n"));
1016 #ifdef USE_ASSERT_CHECKING
1017 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
1019 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
1020 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
1021 printf(gettext(" -d 1-5 debugging level\n"));
1022 printf(gettext(" -D DATADIR database directory\n"));
1023 printf(gettext(" -F turn fsync off\n"));
1024 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
1025 printf(gettext(" -i enable TCP/IP connections\n"));
1026 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
1028 printf(gettext(" -l enable SSL connections\n"));
1030 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
1031 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
1032 printf(gettext(" -p PORT port number to listen on\n"));
1033 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1034 printf(gettext(" --help show this help, then exit\n"));
1035 printf(gettext(" --version output version information, then exit\n"));
1037 printf(gettext("\nDeveloper options:\n"));
1038 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1039 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1041 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1042 "configuration settings and how to set them on the command line or in\n"
1043 "the configuration file.\n\n"
1044 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1057 gettimeofday(&now, &tz);
1059 nSockets = initMasks(&readmask);
1065 struct timeval timeout;
1068 * The timeout for the select() below is normally set on the basis
1069 * of the time to the next checkpoint. However, if for some
1070 * reason we don't have a next-checkpoint time, time out after 60
1071 * seconds. This keeps checkpoint scheduling from locking up when
1072 * we get new connection requests infrequently (since we are
1073 * likely to detect checkpoint completion just after enabling
1074 * signals below, after we've already made the decision about how
1075 * long to wait this time).
1077 timeout.tv_sec = 60;
1078 timeout.tv_usec = 0;
1080 if (CheckPointPID == 0 && checkpointed &&
1081 Shutdown == NoShutdown && !FatalError && random_seed != 0)
1083 time_t now = time(NULL);
1085 if (CheckPointTimeout + checkpointed > now)
1088 * Not time for checkpoint yet, so set select timeout
1090 timeout.tv_sec = CheckPointTimeout + checkpointed - now;
1094 /* Time to make the checkpoint... */
1095 CheckPointPID = CheckPointDataBase();
1098 * if fork failed, schedule another try at 0.1 normal
1101 if (CheckPointPID == 0)
1103 timeout.tv_sec = CheckPointTimeout / 10;
1104 checkpointed = now + timeout.tv_sec - CheckPointTimeout;
1110 * If no background writer process is running and we should
1111 * do background writing, start one. It doesn't matter if
1112 * this fails, we'll just try again later.
1114 if (BgWriterPID == 0 && BgWriterPercent > 0 &&
1115 Shutdown == NoShutdown && !FatalError && random_seed != 0)
1117 BgWriterPID = StartBackgroundWriter();
1121 * Wait for something to happen.
1123 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1125 PG_SETMASK(&UnBlockSig);
1127 if (select(nSockets, &rmask, NULL, NULL, &timeout) < 0)
1129 PG_SETMASK(&BlockSig);
1130 if (errno == EINTR || errno == EWOULDBLOCK)
1133 (errcode_for_socket_access(),
1134 errmsg("select() failed in postmaster: %m")));
1135 return STATUS_ERROR;
1139 * Block all signals until we wait again. (This makes it safe for
1140 * our signal handlers to do nontrivial work.)
1142 PG_SETMASK(&BlockSig);
1145 * Select a random seed at the time of first receiving a request.
1147 while (random_seed == 0)
1149 gettimeofday(&later, &tz);
1152 * We are not sure how much precision is in tv_usec, so we
1153 * swap the nibbles of 'later' and XOR them with 'now'. On the
1154 * off chance that the result is 0, we loop until it isn't.
1156 random_seed = now.tv_usec ^
1157 ((later.tv_usec << 16) |
1158 ((later.tv_usec >> 16) & 0xffff));
1162 * New connection pending on any of our sockets? If so, fork a
1163 * child process to deal with it.
1165 for (i = 0; i < MAXLISTEN; i++)
1167 if (ListenSocket[i] == -1)
1169 if (FD_ISSET(ListenSocket[i], &rmask))
1171 port = ConnCreate(ListenSocket[i]);
1174 BackendStartup(port);
1177 * We no longer need the open socket or port structure
1180 StreamClose(port->sock);
1186 /* If we have lost the stats collector, try to start a new one */
1187 if (!pgstat_is_running)
1194 * Initialise the masks for select() for the ports
1195 * we are listening on. Return the number of sockets to listen on.
1199 initMasks(fd_set *rmask)
1206 for (i = 0; i < MAXLISTEN; i++)
1208 int fd = ListenSocket[i];
1222 * Read the startup packet and do something according to it.
1224 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1225 * not return at all.
1227 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1228 * if that's what you want. Return STATUS_ERROR if you don't want to
1229 * send anything to the client, which would typically be appropriate
1230 * if we detect a communications failure.)
1233 ProcessStartupPacket(Port *port, bool SSLdone)
1237 ProtocolVersion proto;
1238 MemoryContext oldcontext;
1240 if (pq_getbytes((char *) &len, 4) == EOF)
1243 * EOF after SSLdone probably means the client didn't like our
1244 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1245 * so don't clutter the log with a complaint.
1249 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1250 errmsg("incomplete startup packet")));
1251 return STATUS_ERROR;
1257 if (len < (int32) sizeof(ProtocolVersion) ||
1258 len > MAX_STARTUP_PACKET_LENGTH)
1261 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1262 errmsg("invalid length of startup packet")));
1263 return STATUS_ERROR;
1267 * Allocate at least the size of an old-style startup packet, plus one
1268 * extra byte, and make sure all are zeroes. This ensures we will
1269 * have null termination of all strings, in both fixed- and
1270 * variable-length packet layouts.
1272 if (len <= (int32) sizeof(StartupPacket))
1273 buf = palloc0(sizeof(StartupPacket) + 1);
1275 buf = palloc0(len + 1);
1277 if (pq_getbytes(buf, len) == EOF)
1280 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1281 errmsg("incomplete startup packet")));
1282 return STATUS_ERROR;
1286 * The first field is either a protocol version number or a special
1289 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1291 if (proto == CANCEL_REQUEST_CODE)
1293 processCancelRequest(port, buf);
1294 return 127; /* XXX */
1297 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1302 /* No SSL when disabled or on Unix sockets */
1303 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1306 SSLok = 'S'; /* Support for SSL */
1308 SSLok = 'N'; /* No support for SSL */
1310 if (send(port->sock, &SSLok, 1, 0) != 1)
1313 (errcode_for_socket_access(),
1314 errmsg("failed to send SSL negotiation response: %m")));
1315 return STATUS_ERROR; /* close the connection */
1319 if (SSLok == 'S' && secure_open_server(port) == -1)
1320 return STATUS_ERROR;
1322 /* regular startup packet, cancel, etc packet should follow... */
1323 /* but not another SSL negotiation request */
1324 return ProcessStartupPacket(port, true);
1327 /* Could add additional special packet types here */
1330 * Set FrontendProtocol now so that ereport() knows what format to
1331 * send if we fail during startup.
1333 FrontendProtocol = proto;
1335 /* Check we can handle the protocol the frontend is using. */
1337 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1338 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1339 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1340 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1342 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1343 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1344 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1345 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1346 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1347 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1350 * Now fetch parameters out of startup packet and save them into the
1351 * Port structure. All data structures attached to the Port struct
1352 * must be allocated in TopMemoryContext so that they won't disappear
1353 * when we pass them to PostgresMain (see BackendRun). We need not
1354 * worry about leaking this storage on failure, since we aren't in the
1355 * postmaster process anymore.
1357 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1359 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1361 int32 offset = sizeof(ProtocolVersion);
1364 * Scan packet body for name/option pairs. We can assume any
1365 * string beginning within the packet body is null-terminated,
1366 * thanks to zeroing extra byte above.
1368 port->guc_options = NIL;
1370 while (offset < len)
1372 char *nameptr = ((char *) buf) + offset;
1376 if (*nameptr == '\0')
1377 break; /* found packet terminator */
1378 valoffset = offset + strlen(nameptr) + 1;
1379 if (valoffset >= len)
1380 break; /* missing value, will complain below */
1381 valptr = ((char *) buf) + valoffset;
1383 if (strcmp(nameptr, "database") == 0)
1384 port->database_name = pstrdup(valptr);
1385 else if (strcmp(nameptr, "user") == 0)
1386 port->user_name = pstrdup(valptr);
1387 else if (strcmp(nameptr, "options") == 0)
1388 port->cmdline_options = pstrdup(valptr);
1391 /* Assume it's a generic GUC option */
1392 port->guc_options = lappend(port->guc_options,
1394 port->guc_options = lappend(port->guc_options,
1397 offset = valoffset + strlen(valptr) + 1;
1401 * If we didn't find a packet terminator exactly at the end of the
1402 * given packet length, complain.
1404 if (offset != len - 1)
1406 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1407 errmsg("invalid startup packet layout: expected terminator as last byte")));
1412 * Get the parameters from the old-style, fixed-width-fields
1413 * startup packet as C strings. The packet destination was
1414 * cleared first so a short packet has zeros silently added. We
1415 * have to be prepared to truncate the pstrdup result for oversize
1418 StartupPacket *packet = (StartupPacket *) buf;
1420 port->database_name = pstrdup(packet->database);
1421 if (strlen(port->database_name) > sizeof(packet->database))
1422 port->database_name[sizeof(packet->database)] = '\0';
1423 port->user_name = pstrdup(packet->user);
1424 if (strlen(port->user_name) > sizeof(packet->user))
1425 port->user_name[sizeof(packet->user)] = '\0';
1426 port->cmdline_options = pstrdup(packet->options);
1427 if (strlen(port->cmdline_options) > sizeof(packet->options))
1428 port->cmdline_options[sizeof(packet->options)] = '\0';
1429 port->guc_options = NIL;
1432 /* Check a user name was given. */
1433 if (port->user_name == NULL || port->user_name[0] == '\0')
1435 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1436 errmsg("no PostgreSQL user name specified in startup packet")));
1438 /* The database defaults to the user name. */
1439 if (port->database_name == NULL || port->database_name[0] == '\0')
1440 port->database_name = pstrdup(port->user_name);
1442 if (Db_user_namespace)
1445 * If user@, it is a global user, remove '@'. We only want to do
1446 * this if there is an '@' at the end and no earlier in the user
1447 * string or they may fake as a local user of another database
1448 * attaching to this database.
1450 if (strchr(port->user_name, '@') ==
1451 port->user_name + strlen(port->user_name) - 1)
1452 *strchr(port->user_name, '@') = '\0';
1455 /* Append '@' and dbname */
1458 db_user = palloc(strlen(port->user_name) +
1459 strlen(port->database_name) + 2);
1460 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1461 port->user_name = db_user;
1466 * Truncate given database and user names to length of a Postgres
1467 * name. This avoids lookup failures when overlength names are given.
1469 if (strlen(port->database_name) >= NAMEDATALEN)
1470 port->database_name[NAMEDATALEN - 1] = '\0';
1471 if (strlen(port->user_name) >= NAMEDATALEN)
1472 port->user_name[NAMEDATALEN - 1] = '\0';
1475 * Done putting stuff in TopMemoryContext.
1477 MemoryContextSwitchTo(oldcontext);
1480 * If we're going to reject the connection due to database state, say
1481 * so now instead of wasting cycles on an authentication exchange.
1482 * (This also allows a pg_ping utility to be written.)
1484 switch (port->canAcceptConnections)
1488 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1489 errmsg("the database system is starting up")));
1493 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1494 errmsg("the database system is shutting down")));
1498 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1499 errmsg("the database system is in recovery mode")));
1503 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1504 errmsg("sorry, too many clients already")));
1516 * The client has sent a cancel request packet, not a normal
1517 * start-a-new-connection packet. Perform the necessary processing.
1518 * Nothing is sent back to the client.
1521 processCancelRequest(Port *port, void *pkt)
1523 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1525 long cancelAuthCode;
1527 #ifndef EXEC_BACKEND
1533 backendPID = (int) ntohl(canc->backendPID);
1534 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1536 if (backendPID == CheckPointPID)
1539 (errmsg_internal("ignoring cancel request for checkpoint process %d",
1543 else if (backendPID == BgWriterPID)
1546 (errmsg_internal("ignoring cancel request for bgwriter process %d",
1551 /* See if we have a matching backend */
1552 #ifndef EXEC_BACKEND
1553 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1555 bp = (Backend *) DLE_VAL(curr);
1557 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1559 bp = (Backend*) &ShmemBackendArray[i];
1561 if (bp->pid == backendPID)
1563 if (bp->cancel_key == cancelAuthCode)
1565 /* Found a match; signal that backend to cancel current op */
1567 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1569 kill(bp->pid, SIGINT);
1572 /* Right PID, wrong key: no way, Jose */
1574 (errmsg_internal("bad key in cancel request for process %d",
1580 /* No matching backend */
1582 (errmsg_internal("bad pid in cancel request for process %d",
1587 * canAcceptConnections --- check to see if database state allows connections.
1589 static enum CAC_state
1590 canAcceptConnections(void)
1592 /* Can't start backends when in startup/shutdown/recovery state. */
1593 if (Shutdown > NoShutdown)
1594 return CAC_SHUTDOWN;
1598 return CAC_RECOVERY;
1601 * Don't start too many children.
1603 * We allow more connections than we can have backends here because some
1604 * might still be authenticating; they might fail auth, or some
1605 * existing backend might exit before the auth cycle is completed. The
1606 * exact MaxBackends limit is enforced when a new backend tries to
1607 * join the shared-inval backend array.
1609 if (CountChildren() >= 2 * MaxBackends)
1617 * ConnCreate -- create a local connection data structure
1620 ConnCreate(int serverFd)
1624 if (!(port = (Port *) calloc(1, sizeof(Port))))
1627 (errcode(ERRCODE_OUT_OF_MEMORY),
1628 errmsg("out of memory")));
1632 if (StreamConnection(serverFd, port) != STATUS_OK)
1634 StreamClose(port->sock);
1641 * Precompute password salt values to use for this connection.
1642 * It's slightly annoying to do this long in advance of knowing
1643 * whether we'll need 'em or not, but we must do the random()
1644 * calls before we fork, not after. Else the postmaster's random
1645 * sequence won't get advanced, and all backends would end up
1646 * using the same salt...
1648 RandomSalt(port->cryptSalt, port->md5Salt);
1656 * ConnFree -- free a local connection data structure
1659 ConnFree(Port *conn)
1669 * ClosePostmasterPorts -- close all the postmaster's open sockets
1671 * This is called during child process startup to release file descriptors
1672 * that are not needed by that child process. The postmaster still has
1673 * them open, of course.
1676 ClosePostmasterPorts(bool pgstat_too)
1680 /* Close the listen sockets */
1681 for (i = 0; i < MAXLISTEN; i++)
1683 if (ListenSocket[i] != -1)
1685 StreamClose(ListenSocket[i]);
1686 ListenSocket[i] = -1;
1690 /* Close pgstat control sockets, unless we're starting pgstat itself */
1692 pgstat_close_sockets();
1697 * reset_shared -- reset shared memory and semaphores
1700 reset_shared(unsigned short port)
1703 * Create or re-create shared memory and semaphores.
1705 * Note: in each "cycle of life" we will normally assign the same IPC
1706 * keys (if using SysV shmem and/or semas), since the port number is
1707 * used to determine IPC keys. This helps ensure that we will clean
1708 * up dead IPC objects if the postmaster crashes and is restarted.
1710 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1715 * SIGHUP -- reread config files, and tell children to do same
1718 SIGHUP_handler(SIGNAL_ARGS)
1720 int save_errno = errno;
1722 PG_SETMASK(&BlockSig);
1724 if (Shutdown <= SmartShutdown)
1727 (errmsg("received SIGHUP, reloading configuration files")));
1728 ProcessConfigFile(PGC_SIGHUP);
1730 write_nondefault_variables(PGC_SIGHUP);
1732 SignalChildren(SIGHUP);
1737 * Tell the background writer to terminate so that we
1738 * will start a new one with a possibly changed config
1740 if (BgWriterPID != 0)
1741 kill(BgWriterPID, SIGTERM);
1744 PG_SETMASK(&UnBlockSig);
1752 * pmdie -- signal handler for processing various postmaster signals.
1757 int save_errno = errno;
1759 PG_SETMASK(&BlockSig);
1762 (errmsg_internal("postmaster received signal %d",
1763 postgres_signal_arg)));
1765 switch (postgres_signal_arg)
1772 * Wait for children to end their work and ShutdownDataBase.
1774 if (BgWriterPID != 0)
1775 kill(BgWriterPID, SIGTERM);
1776 if (Shutdown >= SmartShutdown)
1778 Shutdown = SmartShutdown;
1780 (errmsg("received smart shutdown request")));
1781 if (DLGetHead(BackendList)) /* let reaper() handle this */
1785 * No children left. Shutdown data base system.
1787 if (StartupPID > 0 || FatalError) /* let reaper() handle
1790 if (ShutdownPID > 0)
1792 elog(PANIC, "shutdown process %d already running",
1797 ShutdownPID = ShutdownDataBase();
1805 * abort all children with SIGTERM (rollback active transactions
1806 * and exit) and ShutdownDataBase when they are gone.
1808 if (BgWriterPID != 0)
1809 kill(BgWriterPID, SIGTERM);
1810 if (Shutdown >= FastShutdown)
1813 (errmsg("received fast shutdown request")));
1814 if (DLGetHead(BackendList)) /* let reaper() handle this */
1816 Shutdown = FastShutdown;
1820 (errmsg("aborting any active transactions")));
1821 SignalChildren(SIGTERM);
1825 if (Shutdown > NoShutdown)
1827 Shutdown = FastShutdown;
1830 Shutdown = FastShutdown;
1833 * No children left. Shutdown data base system.
1835 if (StartupPID > 0 || FatalError) /* let reaper() handle
1838 if (ShutdownPID > 0)
1840 elog(PANIC, "shutdown process %d already running",
1845 ShutdownPID = ShutdownDataBase();
1851 * Immediate Shutdown:
1853 * abort all children with SIGQUIT and exit without attempt to
1854 * properly shutdown data base system.
1856 if (BgWriterPID != 0)
1857 kill(BgWriterPID, SIGQUIT);
1859 (errmsg("received immediate shutdown request")));
1860 if (ShutdownPID > 0)
1861 kill(ShutdownPID, SIGQUIT);
1863 kill(StartupPID, SIGQUIT);
1864 if (DLGetHead(BackendList))
1865 SignalChildren(SIGQUIT);
1870 PG_SETMASK(&UnBlockSig);
1876 * Reaper -- signal handler to cleanup after a backend (child) dies.
1881 int save_errno = errno;
1884 int status; /* backend exit status */
1887 union wait status; /* backend exit status */
1891 int pid; /* process id of dead backend */
1893 PG_SETMASK(&BlockSig);
1896 (errmsg_internal("reaping dead processes")));
1898 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1900 exitstatus = status;
1903 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1905 exitstatus = status.w_status;
1907 while ((pid = win32_waitpid(&exitstatus)) > 0)
1910 * We need to do this here, and not in CleanupProc, since this
1911 * is to be called on all children when we are done with them.
1912 * Could move to LogChildExit, but that seems like asking for
1915 win32_RemoveChild(pid);
1920 * Check if this child was the statistics collector. If so, try to
1921 * start a new one. (If fail, we'll try again in future cycles of
1924 if (pgstat_ispgstat(pid))
1926 LogChildExit(LOG, gettext("statistics collector process"),
1933 * Check if this child was a shutdown or startup process.
1935 if (ShutdownPID > 0 && pid == ShutdownPID)
1937 if (exitstatus != 0)
1939 LogChildExit(LOG, gettext("shutdown process"),
1943 /* Normal postmaster exit is here */
1947 if (StartupPID > 0 && pid == StartupPID)
1949 if (exitstatus != 0)
1951 LogChildExit(LOG, gettext("startup process"),
1954 (errmsg("aborting startup due to startup process failure")));
1960 * Startup succeeded - remember its ID and RedoRecPtr.
1962 * NB: this MUST happen before we fork a checkpoint or shutdown
1963 * subprocess, else they will have wrong local ThisStartUpId.
1967 FatalError = false; /* done with recovery */
1970 * Arrange for first checkpoint to occur after standard delay.
1973 checkpointed = time(NULL);
1975 if (BgWriterPID == 0 && BgWriterPercent > 0 &&
1976 Shutdown == NoShutdown && !FatalError && random_seed != 0)
1978 BgWriterPID = StartBackgroundWriter();
1982 * Go to shutdown mode if a shutdown request was pending.
1984 if (Shutdown > NoShutdown)
1986 if (ShutdownPID > 0)
1988 elog(PANIC, "startup process %d died while shutdown process %d already running",
1989 pid, (int) ShutdownPID);
1992 ShutdownPID = ShutdownDataBase();
1999 * Else do standard child cleanup.
2001 CleanupProc(pid, exitstatus);
2003 } /* loop over pending child-death reports */
2008 * Wait for all children exit, then reset shmem and
2011 if (DLGetHead(BackendList) || StartupPID > 0 || ShutdownPID > 0)
2014 (errmsg("all server processes terminated; reinitializing")));
2017 reset_shared(PostPortNumber);
2019 StartupPID = StartupDataBase();
2024 if (Shutdown > NoShutdown)
2026 if (DLGetHead(BackendList))
2028 if (StartupPID > 0 || ShutdownPID > 0)
2030 ShutdownPID = ShutdownDataBase();
2034 PG_SETMASK(&UnBlockSig);
2040 * CleanupProc -- cleanup after terminated backend.
2042 * Remove all local state associated with backend.
2045 CleanupProc(int pid,
2046 int exitstatus) /* child's exit status. */
2052 LogChildExit(DEBUG2, gettext("child process"), pid, exitstatus);
2055 * If a backend dies in an ugly way (i.e. exit status not 0) then we
2056 * must signal all other backends to quickdie. If exit status is zero
2057 * we assume everything is hunky dory and simply remove the backend
2058 * from the active backend list.
2060 if (exitstatus == 0)
2062 curr = DLGetHead(BackendList);
2065 bp = (Backend *) DLE_VAL(curr);
2069 ShmemBackendArrayRemove(bp->pid);
2076 curr = DLGetSucc(curr);
2079 if (pid == CheckPointPID)
2084 checkpointed = time(NULL);
2085 /* Update RedoRecPtr for future child backends */
2086 GetSavedRedoRecPtr();
2089 else if (pid == BgWriterPID)
2097 /* below here we're dealing with a non-normal exit */
2099 /* Make log entry unless we did so already */
2103 (pid == CheckPointPID) ? gettext("checkpoint process") :
2104 (pid == BgWriterPID) ? gettext("bgwriter process") :
2105 gettext("server process"),
2108 (errmsg("terminating any other active server processes")));
2111 curr = DLGetHead(BackendList);
2114 next = DLGetSucc(curr);
2115 bp = (Backend *) DLE_VAL(curr);
2119 * This backend is still alive. Unless we did so already,
2120 * tell it to commit hara-kiri.
2122 * SIGQUIT is the special signal that says exit without proc_exit
2123 * and let the user know what's going on. But if SendStop is
2124 * set (-s on command line), then we send SIGSTOP instead, so
2125 * that we can get core dumps from all backends by hand.
2130 (errmsg_internal("sending %s to process %d",
2131 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2133 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2139 * Found entry for freshly-dead backend, so remove it.
2142 ShmemBackendArrayRemove(bp->pid);
2151 if (pid == CheckPointPID)
2156 else if (pid == BgWriterPID)
2163 * Tell the collector about backend termination
2172 * Log the death of a child process.
2175 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2177 if (WIFEXITED(exitstatus))
2181 * translator: %s is a noun phrase describing a child process,
2182 * such as "server process"
2184 (errmsg("%s (PID %d) exited with exit code %d",
2185 procname, pid, WEXITSTATUS(exitstatus))));
2186 else if (WIFSIGNALED(exitstatus))
2190 * translator: %s is a noun phrase describing a child process,
2191 * such as "server process"
2193 (errmsg("%s (PID %d) was terminated by signal %d",
2194 procname, pid, WTERMSIG(exitstatus))));
2199 * translator: %s is a noun phrase describing a child process,
2200 * such as "server process"
2202 (errmsg("%s (PID %d) exited with unexpected status %d",
2203 procname, pid, exitstatus)));
2207 * Send a signal to all backend children.
2210 SignalChildren(int signal)
2216 curr = DLGetHead(BackendList);
2219 next = DLGetSucc(curr);
2220 bp = (Backend *) DLE_VAL(curr);
2222 if (bp->pid != MyProcPid)
2225 (errmsg_internal("sending signal %d to process %d",
2228 kill(bp->pid, signal);
2236 * BackendStartup -- start backend process
2238 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2241 BackendStartup(Port *port)
2243 Backend *bn; /* for backend cleanup */
2246 #ifdef LINUX_PROFILE
2247 struct itimerval prof_itimer;
2251 * Compute the cancel key that will be assigned to this backend. The
2252 * backend will have its own copy in the forked-off process' value of
2253 * MyCancelKey, so that it can transmit the key to the frontend.
2255 MyCancelKey = PostmasterRandom();
2258 * Make room for backend data structure. Better before the fork() so
2259 * we can handle failure cleanly.
2261 bn = (Backend *) malloc(sizeof(Backend));
2265 (errcode(ERRCODE_OUT_OF_MEMORY),
2266 errmsg("out of memory")));
2267 return STATUS_ERROR;
2271 * Flush stdio channels just before fork, to avoid double-output
2272 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2273 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2274 * coredump if we do. Presently stdout and stderr are the only stdio
2275 * output channels used by the postmaster, so fflush'ing them should
2281 #ifdef LINUX_PROFILE
2284 * Linux's fork() resets the profiling timer in the child process. If
2285 * we want to profile child processes then we need to save and restore
2286 * the timer setting. This is a waste of time if not profiling,
2287 * however, so only do it if commanded by specific -DLINUX_PROFILE
2290 getitimer(ITIMER_PROF, &prof_itimer);
2294 /* Specific beos actions before backend startup */
2295 beos_before_backend_startup();
2298 port->canAcceptConnections = canAcceptConnections();
2300 pid = Backend_forkexec(port);
2304 if (pid == 0) /* child */
2306 #ifdef LINUX_PROFILE
2307 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2311 /* Specific beos backend startup actions */
2312 beos_backend_startup();
2316 proc_exit(BackendRun(port));
2320 /* in parent, error */
2323 int save_errno = errno;
2326 /* Specific beos backend startup actions */
2327 beos_backend_startup_failed();
2332 (errmsg("could not fork new process for connection: %m")));
2333 report_fork_failure_to_client(port, save_errno);
2334 return STATUS_ERROR;
2337 /* in parent, normal */
2339 (errmsg_internal("forked new backend, pid=%d socket=%d",
2340 (int) pid, port->sock)));
2343 * Everything's been successful, it's safe to add this backend to our
2347 bn->cancel_key = MyCancelKey;
2349 ShmemBackendArrayAdd(bn);
2351 DLAddHead(BackendList, DLNewElem(bn));
2357 * Try to report backend fork() failure to client before we close the
2358 * connection. Since we do not care to risk blocking the postmaster on
2359 * this connection, we set the connection to non-blocking and try only once.
2361 * This is grungy special-purpose code; we cannot use backend libpq since
2362 * it's not up and running.
2365 report_fork_failure_to_client(Port *port, int errnum)
2369 /* Format the error message packet (always V2 protocol) */
2370 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2371 gettext("could not fork new process for connection: "),
2374 /* Set port to non-blocking. Don't do send() if this fails */
2375 if (FCNTL_NONBLOCK(port->sock) < 0)
2378 send(port->sock, buffer, strlen(buffer) + 1, 0);
2383 * split_opts -- split a string of options and append it to an argv array
2385 * NB: the string is destructively modified!
2387 * Since no current POSTGRES arguments require any quoting characters,
2388 * we can use the simple-minded tactic of assuming each set of space-
2389 * delimited characters is a separate argv element.
2391 * If you don't like that, well, we *used* to pass the whole option string
2392 * as ONE argument to execl(), which was even less intelligent...
2395 split_opts(char **argv, int *argcp, char *s)
2399 while (isspace((unsigned char) *s))
2403 argv[(*argcp)++] = s;
2404 while (*s && !isspace((unsigned char) *s))
2413 * BackendInit/Run -- perform authentication [BackendInit], and if successful,
2414 * set up the backend's argument list [BackendRun] and invoke
2418 * Shouldn't return at all.
2419 * If PostgresMain() fails, return status.
2422 BackendInit(Port *port)
2427 char remote_host[NI_MAXHOST];
2428 char remote_port[NI_MAXSERV];
2430 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2432 ClientAuthInProgress = true; /* limit visibility of log messages */
2434 /* We don't want the postmaster's proc_exit() handlers */
2438 * Signal handlers setting is moved to tcop/postgres...
2441 /* Save port etc. for ps status */
2444 /* Reset MyProcPid to new backend's pid */
2445 MyProcPid = getpid();
2448 * Initialize libpq and enable reporting of ereport errors to the
2449 * client. Must do this now because authentication uses libpq to send
2452 pq_init(); /* initialize libpq to talk to client */
2453 whereToSendOutput = Remote; /* now safe to ereport to client */
2456 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2457 * during any client authentication related communication. Otherwise
2458 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2459 * if a buggy client blocks a backend during authentication.
2461 pqsignal(SIGTERM, authdie);
2462 pqsignal(SIGQUIT, authdie);
2463 pqsignal(SIGALRM, authdie);
2464 PG_SETMASK(&AuthBlockSig);
2467 * Get the remote host name and port for logging and status display.
2469 remote_host[0] = '\0';
2470 remote_port[0] = '\0';
2471 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2472 remote_host, sizeof(remote_host),
2473 remote_port, sizeof(remote_port),
2474 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2476 getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2477 remote_host, sizeof(remote_host),
2478 remote_port, sizeof(remote_port),
2479 NI_NUMERICHOST | NI_NUMERICSERV);
2482 if (Log_connections)
2484 (errmsg("connection received: host=%s port=%s",
2485 remote_host, remote_port)));
2489 /* modify remote_host for use in ps status */
2490 char tmphost[NI_MAXHOST];
2492 snprintf(tmphost, sizeof(tmphost), "%s:%s", remote_host, remote_port);
2493 StrNCpy(remote_host, tmphost, sizeof(remote_host));
2497 * Ready to begin client interaction. We will give up and exit(0)
2498 * after a time delay, so that a broken client can't hog a connection
2499 * indefinitely. PreAuthDelay doesn't count against the time limit.
2501 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2502 elog(FATAL, "could not set timer for authorization timeout");
2505 * Receive the startup packet (which might turn out to be a cancel
2508 status = ProcessStartupPacket(port, false);
2510 if (status != STATUS_OK)
2514 * Now that we have the user and database name, we can set the process
2515 * title for ps. It's good to do this as early as possible in
2518 init_ps_display(port->user_name, port->database_name, remote_host);
2519 set_ps_display("authentication");
2522 * Now perform authentication exchange.
2524 ClientAuthentication(port); /* might not return, if failure */
2527 * Done with authentication. Disable timeout, and prevent
2528 * SIGTERM/SIGQUIT again until backend startup is complete.
2530 if (!disable_sig_alarm(false))
2531 elog(FATAL, "could not disable timer for authorization timeout");
2532 PG_SETMASK(&BlockSig);
2534 if (Log_connections)
2536 (errmsg("connection authorized: user=%s database=%s",
2537 port->user_name, port->database_name)));
2540 * Don't want backend to be able to see the postmaster random number
2541 * generator state. We have to clobber the static random_seed *and*
2542 * start a new random sequence in the random() library function.
2545 gettimeofday(&now, &tz);
2546 srandom((unsigned int) now.tv_usec);
2551 BackendRun(Port *port)
2561 * Let's clean up ourselves as the postmaster child, and
2562 * close the postmaster's other sockets
2564 ClosePostmasterPorts(true);
2567 * PreAuthDelay is a debugging aid for investigating problems in the
2568 * authentication cycle: it can be set in postgresql.conf to allow
2569 * time to attach to the newly-forked backend with a debugger. (See
2570 * also the -W backend switch, which we allow clients to pass through
2571 * PGOPTIONS, but it is not honored until after authentication.)
2573 if (PreAuthDelay > 0)
2574 sleep(PreAuthDelay);
2576 /* Will exit on failure */
2581 * Now, build the argv vector that will be given to PostgresMain.
2583 * The layout of the command line is
2584 * postgres [secure switches] -p databasename [insecure switches]
2585 * where the switches after -p come from the client request.
2587 * The maximum possible number of commandline arguments that could come
2588 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2592 maxac = 10; /* for fixed args supplied below */
2593 maxac += (strlen(ExtraOptions) + 1) / 2;
2594 if (port->cmdline_options)
2595 maxac += (strlen(port->cmdline_options) + 1) / 2;
2597 av = (char **) MemoryContextAlloc(TopMemoryContext,
2598 maxac * sizeof(char *));
2601 av[ac++] = "postgres";
2604 * Pass the requested debugging level along to the backend.
2608 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2609 av[ac++] = debugbuf;
2613 * Pass any backend switches specified with -o in the postmaster's own
2614 * command line. We assume these are secure.
2616 split_opts(av, &ac, ExtraOptions);
2618 /* Tell the backend what protocol the frontend is using. */
2619 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2620 av[ac++] = protobuf;
2623 /* pass data dir before end of secure switches (-p) */
2629 * Tell the backend it is being called from the postmaster, and which
2630 * database to use. -p marks the end of secure switches.
2633 av[ac++] = port->database_name;
2636 * Pass the (insecure) option switches from the connection request.
2637 * (It's OK to mangle port->cmdline_options now.)
2639 if (port->cmdline_options)
2640 split_opts(av, &ac, port->cmdline_options);
2647 * Release postmaster's working memory context so that backend can
2648 * recycle the space. Note this does not trash *MyProcPort, because
2649 * ConnCreate() allocated that space with malloc() ... else we'd need
2650 * to copy the Port data here. Also, subsidiary data such as the
2651 * username isn't lost either; see ProcessStartupPacket().
2653 MemoryContextSwitchTo(TopMemoryContext);
2654 #ifndef EXEC_BACKEND
2655 MemoryContextDelete(PostmasterContext);
2657 PostmasterContext = NULL;
2660 * Debug: print arguments being passed to backend
2663 (errmsg_internal("%s child[%d]: starting with (",
2664 progname, getpid())));
2665 for (i = 0; i < ac; ++i)
2667 (errmsg_internal("\t%s", av[i])));
2669 (errmsg_internal(")")));
2671 ClientAuthInProgress = false; /* client_min_messages is active
2674 return (PostgresMain(ac, av, port->user_name));
2682 * SubPostmasterMain -- prepare the fork/exec'd process to be in an equivalent
2683 * state (for calling BackendRun) as a forked process.
2686 * Shouldn't return at all.
2689 SubPostmasterMain(int argc, char* argv[])
2691 unsigned long backendID;
2694 memset((void*)&port, 0, sizeof(Port));
2697 /* Do this sooner rather than later... */
2698 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2700 /* Setup global context */
2701 MemoryContextInit();
2702 InitializeGUCOptions();
2704 /* Parse passed-in context */
2706 backendID = (unsigned long)atol(argv[argc++]);
2707 DataDir = strdup(argv[argc++]);
2709 /* Read in file-based context */
2710 read_nondefault_variables();
2711 read_backend_variables(backendID,&port);
2713 /* FIXME: [fork/exec] Ugh */
2719 /* Attach process to shared segments */
2720 AttachSharedMemoryAndSemaphores();
2723 proc_exit(BackendRun(&port));
2728 * Backend_forkexec -- fork/exec off a backend process
2731 * the pid of the fork/exec'd process
2734 Backend_forkexec(Port *port)
2738 int ac = 0, bufc = 0, i;
2739 char buf[2][MAXPGPATH];
2741 if (!write_backend_variables(port))
2742 return -1; /* log made by write_backend_variables */
2744 av[ac++] = "postgres";
2745 av[ac++] = "-forkexec";
2747 /* Format up context to pass to exec'd process */
2748 snprintf(buf[bufc++],MAXPGPATH,"%lu",tmpBackendFileNum);
2749 /* FIXME: [fork/exec] whitespaces in directories? */
2750 snprintf(buf[bufc++],MAXPGPATH,"%s",DataDir);
2752 /* Add to the arg list */
2753 Assert(bufc <= lengthof(buf));
2754 for (i = 0; i < bufc; i++)
2757 /* FIXME: [fork/exec] ExtraOptions? */
2760 Assert(ac <= lengthof(av));
2763 pid = win32_forkexec(pg_pathname,av); /* logs on error */
2765 /* Fire off execv in child */
2766 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
2768 * FIXME: [fork/exec] suggestions for what to do here?
2769 * Probably OK to issue error (unlike pgstat case)
2773 return pid; /* Parent returns pid */
2780 * ExitPostmaster -- cleanup
2782 * Do NOT call exit() directly --- always go through here!
2785 ExitPostmaster(int status)
2787 /* should cleanup shared memory and kill all backends */
2790 * Not sure of the semantics here. When the Postmaster dies, should
2791 * the backends all be killed? probably not.
2793 * MUST -- vadim 05-10-1999
2795 /* Should I use true instead? */
2796 ClosePostmasterPorts(false);
2802 * sigusr1_handler - handle signal conditions from child processes
2805 sigusr1_handler(SIGNAL_ARGS)
2807 int save_errno = errno;
2809 PG_SETMASK(&BlockSig);
2811 if (CheckPostmasterSignal(PMSIGNAL_DO_CHECKPOINT))
2813 if (CheckPointWarning != 0)
2816 * This only times checkpoints forced by running out of
2817 * segment files. Other checkpoints could reduce the
2818 * frequency of forced checkpoints.
2820 time_t now = time(NULL);
2822 if (LastSignalledCheckpoint != 0)
2824 int elapsed_secs = now - LastSignalledCheckpoint;
2826 if (elapsed_secs < CheckPointWarning)
2828 (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
2830 errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
2832 LastSignalledCheckpoint = now;
2836 * Request to schedule a checkpoint
2838 * Ignore request if checkpoint is already running or checkpointing
2839 * is currently disabled
2841 if (CheckPointPID == 0 && checkpointed &&
2842 Shutdown == NoShutdown && !FatalError && random_seed != 0)
2844 CheckPointPID = CheckPointDataBase();
2845 /* note: if fork fails, CheckPointPID stays 0; nothing happens */
2849 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2852 * Password or group file has changed.
2858 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2861 * Send SIGUSR2 to all children (triggers AsyncNotifyHandler). See
2862 * storage/ipc/sinvaladt.c for the use of this.
2864 if (Shutdown == NoShutdown)
2865 SignalChildren(SIGUSR2);
2868 PG_SETMASK(&UnBlockSig);
2875 * Dummy signal handler
2877 * We use this for signals that we don't actually use in the postmaster,
2878 * but we do use in backends. If we SIG_IGN such signals in the postmaster,
2879 * then a newly started backend might drop a signal that arrives before it's
2880 * able to reconfigure its signal processing. (See notes in postgres.c.)
2883 dummy_handler(SIGNAL_ARGS)
2889 * CharRemap: given an int in range 0..61, produce textual encoding of it
2890 * per crypt(3) conventions.
2914 RandomSalt(char *cryptSalt, char *md5Salt)
2916 long rand = PostmasterRandom();
2918 cryptSalt[0] = CharRemap(rand % 62);
2919 cryptSalt[1] = CharRemap(rand / 62);
2922 * It's okay to reuse the first random value for one of the MD5 salt
2923 * bytes, since only one of the two salts will be sent to the client.
2924 * After that we need to compute more random bits.
2926 * We use % 255, sacrificing one possible byte value, so as to ensure
2927 * that all bits of the random() value participate in the result.
2928 * While at it, add one to avoid generating any null bytes.
2930 md5Salt[0] = (rand % 255) + 1;
2931 rand = PostmasterRandom();
2932 md5Salt[1] = (rand % 255) + 1;
2933 rand = PostmasterRandom();
2934 md5Salt[2] = (rand % 255) + 1;
2935 rand = PostmasterRandom();
2936 md5Salt[3] = (rand % 255) + 1;
2943 PostmasterRandom(void)
2945 static bool initialized = false;
2949 Assert(random_seed != 0);
2950 srandom(random_seed);
2958 * Count up number of child processes.
2967 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2969 bp = (Backend *) DLE_VAL(curr);
2970 if (bp->pid != MyProcPid)
2973 if (CheckPointPID != 0)
2975 if (BgWriterPID != 0)
2981 * Fire off a subprocess for startup/shutdown/checkpoint.
2983 * Return value of SSDataBase is subprocess' PID, or 0 if failed to start subprocess
2984 * (0 is returned only for checkpoint case).
2986 * note: in the EXEC_BACKEND case, we delay the fork until argument list has been
2989 NON_EXEC_STATIC void
2990 SSDataBaseInit(int xlop)
2992 const char *statmsg;
2994 IsUnderPostmaster = true; /* we are a postmaster subprocess
2997 /* Lose the postmaster's on-exit routines and port connections */
3001 * Identify myself via ps
3005 case BS_XLOG_STARTUP:
3006 statmsg = "startup subprocess";
3008 case BS_XLOG_CHECKPOINT:
3009 statmsg = "checkpoint subprocess";
3011 case BS_XLOG_BGWRITER:
3012 statmsg = "bgwriter subprocess";
3014 case BS_XLOG_SHUTDOWN:
3015 statmsg = "shutdown subprocess";
3018 statmsg = "??? subprocess";
3021 init_ps_display(statmsg, "", "");
3027 SSDataBase(int xlop)
3031 #ifndef EXEC_BACKEND
3032 #ifdef LINUX_PROFILE
3033 struct itimerval prof_itimer;
3042 #ifndef EXEC_BACKEND
3043 #ifdef LINUX_PROFILE
3044 /* see comments in BackendRun */
3045 getitimer(ITIMER_PROF, &prof_itimer);
3049 /* Specific beos actions before backend startup */
3050 beos_before_backend_startup();
3053 /* Non EXEC_BACKEND case; fork here */
3054 if ((pid = fork()) == 0) /* child */
3062 #ifndef EXEC_BACKEND
3063 #ifdef LINUX_PROFILE
3064 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3068 /* Specific beos actions after backend startup */
3069 beos_backend_startup();
3072 /* Close the postmaster's sockets */
3073 ClosePostmasterPorts(true);
3075 SSDataBaseInit(xlop);
3077 if (!write_backend_variables(NULL))
3078 return -1; /* log issued by write_backend_variables */
3081 /* Set up command-line arguments for subprocess */
3082 av[ac++] = "postgres";
3087 snprintf(nbbuf, sizeof(nbbuf), "-B%d", NBuffers);
3090 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
3094 /* pass data dir before end of secure switches (-p) */
3098 /* and the backend identifier + dbname */
3099 snprintf(idbuf, sizeof(idbuf), "-p%lu,template1", tmpBackendFileNum);
3103 av[ac++] = "template1";
3108 Assert(ac < lengthof(av));
3111 /* EXEC_BACKEND case; fork/exec here */
3113 pid = win32_forkexec(pg_pathname,av); /* logs on error */
3115 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
3118 elog(ERROR,"unable to execv in SSDataBase: %m");
3123 BootstrapMain(ac, av);
3131 #ifndef EXEC_BACKEND
3133 /* Specific beos actions before backend startup */
3134 beos_backend_startup_failed();
3139 case BS_XLOG_STARTUP:
3141 (errmsg("could not fork startup process: %m")));
3143 case BS_XLOG_CHECKPOINT:
3145 (errmsg("could not fork checkpoint process: %m")));
3147 case BS_XLOG_BGWRITER:
3149 (errmsg("could not fork bgwriter process: %m")));
3151 case BS_XLOG_SHUTDOWN:
3153 (errmsg("could not fork shutdown process: %m")));
3157 (errmsg("could not fork process: %m")));
3162 * fork failure is fatal during startup/shutdown, but there's no
3163 * need to choke if a routine checkpoint or starting a background
3166 if (xlop == BS_XLOG_CHECKPOINT)
3168 if (xlop == BS_XLOG_BGWRITER)
3174 * The startup and shutdown processes are not considered normal
3175 * backends, but the checkpoint and bgwriter processes are.
3176 * They must be added to the list of backends.
3178 if (xlop == BS_XLOG_CHECKPOINT || xlop == BS_XLOG_BGWRITER)
3180 if (!(bn = (Backend *) malloc(sizeof(Backend))))
3183 (errcode(ERRCODE_OUT_OF_MEMORY),
3184 errmsg("out of memory")));
3189 bn->cancel_key = PostmasterRandom();
3191 ShmemBackendArrayAdd(bn);
3193 DLAddHead(BackendList, DLNewElem(bn));
3196 * Since this code is executed periodically, it's a fine place to
3197 * do other actions that should happen every now and then on no
3198 * particular schedule. Such as...
3201 TouchSocketLockFile();
3209 * Create the opts file
3212 CreateOptsFile(int argc, char *argv[])
3214 char fullprogname[MAXPGPATH];
3215 char filename[MAXPGPATH];
3219 if (FindExec(fullprogname, argv[0], "postmaster") < 0)
3222 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3224 if ((fp = fopen(filename, "w")) == NULL)
3226 elog(LOG, "could not create file \"%s\": %m", filename);
3230 fprintf(fp, "%s", fullprogname);
3231 for (i = 1; i < argc; i++)
3232 fprintf(fp, " '%s'", argv[i]);
3237 elog(LOG, "could not write file \"%s\": %m", filename);
3245 * This should be used only for reporting "interactive" errors (essentially,
3246 * bogus arguments on the command line). Once the postmaster is launched,
3247 * use ereport. In particular, don't use this for anything that occurs
3248 * after pmdaemonize.
3251 postmaster_error(const char *fmt,...)
3255 fprintf(stderr, "%s: ", progname);
3257 vfprintf(stderr, gettext(fmt), ap);
3259 fprintf(stderr, "\n");
3266 * The following need to be available to the read/write_backend_variables
3269 #include "storage/spin.h"
3270 extern XLogRecPtr RedoRecPtr;
3271 extern XLogwrtResult LogwrtResult;
3272 extern slock_t *ShmemLock;
3273 extern slock_t *ShmemIndexLock;
3274 extern void *ShmemIndexAlloc;
3275 typedef struct LWLock LWLock;
3276 extern LWLock *LWLockArray;
3277 extern slock_t *ProcStructLock;
3278 extern int pgStatSock;
3280 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3281 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3282 #define get_tmp_backend_file_name(buf,id) \
3286 "%s/%s/%s.backend_var.%lu", \
3288 PG_TEMP_FILES_DIR, \
3289 PG_TEMP_FILE_PREFIX, \
3294 write_backend_variables(Port *port)
3296 char filename[MAXPGPATH];
3298 get_tmp_backend_file_name(filename,++tmpBackendFileNum);
3301 fp = AllocateFile(filename, PG_BINARY_W);
3304 /* As per OpenTemporaryFile... */
3305 char dirname[MAXPGPATH];
3306 sprintf(dirname,"%s/%s",DataDir,PG_TEMP_FILES_DIR);
3307 mkdir(dirname, S_IRWXU);
3309 fp = AllocateFile(filename, PG_BINARY_W);
3313 (errcode_for_file_access(),
3314 errmsg("could not write to file \"%s\": %m", filename)));
3322 write_var(port->sock,fp);
3323 write_var(port->proto,fp);
3324 write_var(port->laddr,fp);
3325 write_var(port->raddr,fp);
3326 write_var(port->canAcceptConnections,fp);
3327 write_var(port->cryptSalt,fp);
3328 write_var(port->md5Salt,fp);
3330 write_var(MyCancelKey,fp);
3332 write_var(RedoRecPtr,fp);
3333 write_var(LogwrtResult,fp);
3335 write_var(UsedShmemSegID,fp);
3336 write_var(UsedShmemSegAddr,fp);
3338 write_var(ShmemLock,fp);
3339 write_var(ShmemIndexLock,fp);
3340 write_var(ShmemVariableCache,fp);
3341 write_var(ShmemIndexAlloc,fp);
3342 write_var(ShmemBackendArray,fp);
3344 write_var(LWLockArray,fp);
3345 write_var(ProcStructLock,fp);
3346 write_var(pgStatSock,fp);
3348 write_var(PreAuthDelay,fp);
3349 write_var(debug_flag,fp);
3350 write_var(PostmasterPid,fp);
3356 (errcode_for_file_access(),
3357 errmsg("could not write to file \"%s\": %m", filename)));
3365 read_backend_variables(unsigned long id, Port *port)
3367 char filename[MAXPGPATH];
3369 get_tmp_backend_file_name(filename,id);
3372 fp = AllocateFile(filename, PG_BINARY_R);
3376 (errcode_for_file_access(),
3377 errmsg("could not read from backend_variables file \"%s\": %m", filename)));
3384 read_var(port->sock,fp);
3385 read_var(port->proto,fp);
3386 read_var(port->laddr,fp);
3387 read_var(port->raddr,fp);
3388 read_var(port->canAcceptConnections,fp);
3389 read_var(port->cryptSalt,fp);
3390 read_var(port->md5Salt,fp);
3392 read_var(MyCancelKey,fp);
3394 read_var(RedoRecPtr,fp);
3395 read_var(LogwrtResult,fp);
3397 read_var(UsedShmemSegID,fp);
3398 read_var(UsedShmemSegAddr,fp);
3400 read_var(ShmemLock,fp);
3401 read_var(ShmemIndexLock,fp);
3402 read_var(ShmemVariableCache,fp);
3403 read_var(ShmemIndexAlloc,fp);
3404 read_var(ShmemBackendArray,fp);
3406 read_var(LWLockArray,fp);
3407 read_var(ProcStructLock,fp);
3408 read_var(pgStatSock,fp);
3410 read_var(PreAuthDelay,fp);
3411 read_var(debug_flag,fp);
3412 read_var(PostmasterPid,fp);
3416 if (unlink(filename) != 0)
3418 (errcode_for_file_access(),
3419 errmsg("could not remove file \"%s\": %m", filename)));
3423 size_t ShmemBackendArraySize(void)
3425 return (NUM_BACKENDARRAY_ELEMS*sizeof(Backend));
3428 void ShmemBackendArrayAllocation(void)
3430 size_t size = ShmemBackendArraySize();
3431 ShmemBackendArray = (Backend*)ShmemAlloc(size);
3432 memset(ShmemBackendArray, 0, size);
3435 static void ShmemBackendArrayAdd(Backend *bn)
3438 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3440 /* Find an empty slot */
3441 if (ShmemBackendArray[i].pid == 0)
3443 ShmemBackendArray[i] = *bn;
3448 /* FIXME: [fork/exec] some sort of error */
3452 static void ShmemBackendArrayRemove(pid_t pid)
3455 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3457 if (ShmemBackendArray[i].pid == pid)
3459 /* Mark the slot as empty */
3460 ShmemBackendArray[i].pid = 0;
3465 /* Something stronger than WARNING here? */
3467 (errmsg_internal("unable to find backend entry with pid %d",
3475 pid_t win32_forkexec(const char* path, char *argv[])
3478 PROCESS_INFORMATION pi;
3481 char cmdLine[MAXPGPATH];
3483 /* Format the cmd line */
3484 snprintf(cmdLine,sizeof(cmdLine),"%s",path);
3486 while (argv[++i] != NULL)
3488 /* FIXME: [fork/exec] some strlen checks might be prudent here */
3489 strcat(cmdLine," ");
3490 strcat(cmdLine,argv[i]);
3494 * The following snippet can disappear when we consistently
3495 * use forward slashes.
3498 while (*(p++) != '\0')
3499 if (*p == '/') *p = '\\';
3501 memset(&pi,0,sizeof(pi));
3502 memset(&si,0,sizeof(si));
3504 if (!CreateProcess(NULL,cmdLine,NULL,NULL,TRUE,0,NULL,NULL,&si,&pi))
3506 elog(ERROR,"CreateProcess call failed (%d): %m",GetLastError());
3510 if (!IsUnderPostmaster)
3511 /* We are the Postmaster creating a child... */
3512 win32_AddChild(pi.dwProcessId,pi.hProcess);
3514 CloseHandle(pi.hProcess);
3515 CloseHandle(pi.hThread);
3517 return pi.dwProcessId;
3521 * Note: The following three functions must not be interrupted (eg. by signals).
3522 * As the Postgres Win32 signalling architecture (currently) requires polling,
3523 * or APC checking functions which aren't used here, this is not an issue.
3525 * We keep two separate arrays, instead of a single array of pid/HANDLE structs,
3526 * to avoid having to re-create a handle array for WaitForMultipleObjects on
3527 * each call to win32_waitpid.
3530 static void win32_AddChild(pid_t pid, HANDLE handle)
3532 Assert(win32_childPIDArray && win32_childHNDArray);
3533 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3535 win32_childPIDArray[win32_numChildren] = pid;
3536 win32_childHNDArray[win32_numChildren] = handle;
3537 ++win32_numChildren;
3540 /* FIXME: [fork/exec] some sort of error */
3544 static void win32_RemoveChild(pid_t pid)
3547 Assert(win32_childPIDArray && win32_childHNDArray);
3549 for (i = 0; i < win32_numChildren; i++)
3551 if (win32_childPIDArray[i] == pid)
3553 CloseHandle(win32_childHNDArray[i]);
3555 /* Swap last entry into the "removed" one */
3556 --win32_numChildren;
3557 win32_childPIDArray[win32_numChildren] = win32_childPIDArray[i];
3558 win32_childHNDArray[win32_numChildren] = win32_childHNDArray[i];
3563 /* Something stronger than WARNING here? */
3565 (errmsg_internal("unable to find child entry with pid %d",
3569 static pid_t win32_waitpid(int *exitstatus)
3571 Assert(win32_childPIDArray && win32_childHNDArray);
3572 elog(DEBUG3,"waiting on %d children",win32_numChildren);
3574 if (win32_numChildren > 0)
3577 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't
3578 * want to run queued APCs here.
3582 DWORD ret = WaitForMultipleObjects(win32_numChildren,win32_childHNDArray,FALSE,0);
3588 (errmsg_internal("failed to wait on %d children",
3589 win32_numChildren)));
3590 /* Fall through to WAIT_TIMEOUTs return */
3593 /* No children have finished */
3597 /* Get the exit code, and return the PID of, the respective process */
3598 index = ret-WAIT_OBJECT_0;
3599 Assert(index >= 0 && index < win32_numChildren);
3600 if (!GetExitCodeProcess(win32_childHNDArray[index],&exitCode))
3602 * If we get this far, this should never happen, but, then again...
3603 * No choice other than to assume a catastrophic failure.
3606 (errmsg_internal("failed to get exit code for child %d",
3607 win32_childPIDArray[index])));
3608 *exitstatus = (int)exitCode;
3609 return win32_childPIDArray[index];