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 and shutdown. The postmaster itself doesn't do those
11 * operations, mind you --- it just forks off a subprocess to do them
12 * at the right times. It also takes care of resetting the system
13 * if a backend crashes.
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 1.406 2004/07/10 23:29:16 tgl Exp $
45 * The Postmaster sets up shared memory data structures
49 * The Postmaster shares memory with the backends but should avoid
50 * touching shared memory, so as not to become stuck if a crashing
51 * backend screws up locks or shared memory. Likewise, the Postmaster
52 * should never block on messages from frontend clients.
55 * The Postmaster cleans up after backends if they have an emergency
56 * exit and/or core dump.
59 * Use write_stderr() only for reporting "interactive" errors
60 * (essentially, bogus arguments on the command line). Once the
61 * postmaster is launched, use ereport(). In particular, don't use
62 * write_stderr() for anything that occurs after pmdaemonize.
64 *-------------------------------------------------------------------------
75 #include <sys/socket.h>
78 #include <sys/param.h>
79 #include <netinet/in.h>
80 #include <arpa/inet.h>
84 #ifdef HAVE_SYS_SELECT_H
85 #include <sys/select.h>
93 #include <DNSServiceDiscovery/DNSServiceDiscovery.h>
96 #include "catalog/pg_database.h"
97 #include "commands/async.h"
98 #include "lib/dllist.h"
99 #include "libpq/auth.h"
100 #include "libpq/crypt.h"
101 #include "libpq/libpq.h"
102 #include "libpq/pqcomm.h"
103 #include "libpq/pqsignal.h"
104 #include "miscadmin.h"
105 #include "nodes/nodes.h"
106 #include "postmaster/postmaster.h"
107 #include "storage/fd.h"
108 #include "storage/ipc.h"
109 #include "storage/pg_shmem.h"
110 #include "storage/pmsignal.h"
111 #include "storage/proc.h"
112 #include "storage/bufmgr.h"
113 #include "access/xlog.h"
114 #include "tcop/tcopprot.h"
115 #include "utils/guc.h"
116 #include "utils/memutils.h"
117 #include "utils/ps_status.h"
118 #include "bootstrap/bootstrap.h"
123 * List of active backends (or child processes anyway; we don't actually
124 * know whether a given child has become a backend or is still in the
125 * authorization phase). This is used mainly to keep track of how many
126 * children we have and send them appropriate signals when necessary.
128 * "Special" children such as the startup and bgwriter tasks are not in
133 pid_t pid; /* process id of backend */
134 long cancel_key; /* cancel key for cancels for this backend */
137 static Dllist *BackendList;
140 #define NUM_BACKENDARRAY_ELEMS (2*MaxBackends)
141 static Backend *ShmemBackendArray;
144 /* The socket number we are listening for connections on */
147 char *ListenAddresses;
150 * ReservedBackends is the number of backends reserved for superuser use.
151 * This number is taken out of the pool size given by MaxBackends so
152 * number of backend slots available to non-superusers is
153 * (MaxBackends - ReservedBackends). Note what this really means is
154 * "if there are <= ReservedBackends connections available, only superusers
155 * can make new connections" --- pre-existing superuser connections don't
156 * count against the limit.
158 int ReservedBackends;
161 static const char *progname = NULL;
163 /* The socket(s) we're listening to. */
165 static int ListenSocket[MAXLISTEN];
168 * Set by the -o option
170 static char ExtraOptions[MAXPGPATH];
173 * These globals control the behavior of the postmaster in case some
174 * backend dumps core. Normally, it kills all peers of the dead backend
175 * and reinitializes shared memory. By specifying -s or -n, we can have
176 * the postmaster stop (rather than kill) peers and not reinitialize
177 * shared data structures.
179 static bool Reinit = true;
180 static int SendStop = false;
182 /* still more option variables */
183 bool EnableSSL = false;
184 bool SilentMode = false; /* silent mode (-S) */
186 int PreAuthDelay = 0;
187 int AuthenticationTimeout = 60;
189 bool log_hostname; /* for ps display and logging */
190 bool Log_connections = false;
191 bool Db_user_namespace = false;
193 char *rendezvous_name;
195 /* list of library:init-function to be preloaded */
196 char *preload_libraries_string = NULL;
198 /* PIDs of special child processes; 0 when not running */
199 static pid_t StartupPID = 0,
203 /* Startup/shutdown state */
205 #define SmartShutdown 1
206 #define FastShutdown 2
208 static int Shutdown = NoShutdown;
210 static bool FatalError = false; /* T if recovering from backend crash */
212 bool ClientAuthInProgress = false; /* T during new-client
216 * State for assigning random salts and cancel keys.
217 * Also, the global MyCancelKey passes the cancel key assigned to a given
218 * backend from the postmaster to that backend (via fork).
220 static unsigned int random_seed = 0;
222 static int debug_flag = 0;
228 #ifdef HAVE_INT_OPTRESET
233 * postmaster.c - function prototypes
235 static void checkDataDir(const char *checkdir);
236 #ifdef USE_RENDEZVOUS
237 static void reg_reply(DNSServiceRegistrationReplyErrorType errorCode,
240 static void pmdaemonize(void);
241 static Port *ConnCreate(int serverFd);
242 static void ConnFree(Port *port);
243 static void reset_shared(unsigned short port);
244 static void SIGHUP_handler(SIGNAL_ARGS);
245 static void pmdie(SIGNAL_ARGS);
246 static void reaper(SIGNAL_ARGS);
247 static void sigusr1_handler(SIGNAL_ARGS);
248 static void dummy_handler(SIGNAL_ARGS);
249 static void CleanupProc(int pid, int exitstatus);
250 static void HandleChildCrash(int pid, int exitstatus);
251 static void LogChildExit(int lev, const char *procname,
252 int pid, int exitstatus);
253 static int BackendRun(Port *port);
254 static void ExitPostmaster(int status);
255 static void usage(const char *);
256 static int ServerLoop(void);
257 static int BackendStartup(Port *port);
258 static int ProcessStartupPacket(Port *port, bool SSLdone);
259 static void processCancelRequest(Port *port, void *pkt);
260 static int initMasks(fd_set *rmask);
261 static void report_fork_failure_to_client(Port *port, int errnum);
262 static enum CAC_state canAcceptConnections(void);
263 static long PostmasterRandom(void);
264 static void RandomSalt(char *cryptSalt, char *md5Salt);
265 static void SignalChildren(int signal);
266 static int CountChildren(void);
267 static bool CreateOptsFile(int argc, char *argv[], char *fullprogname);
268 static pid_t StartChildProcess(int xlop);
273 static pid_t win32_forkexec(const char *path, char *argv[]);
274 static void win32_AddChild(pid_t pid, HANDLE handle);
275 static void win32_RemoveChild(pid_t pid);
276 static pid_t win32_waitpid(int *exitstatus);
277 static DWORD WINAPI win32_sigchld_waiter(LPVOID param);
279 static pid_t *win32_childPIDArray;
280 static HANDLE *win32_childHNDArray;
281 static unsigned long win32_numChildren = 0;
283 HANDLE PostmasterHandle;
286 static pid_t backend_forkexec(Port *port);
287 static pid_t internal_forkexec(int argc, char *argv[], Port *port);
289 static void read_backend_variables(char *filename, Port *port);
290 static bool write_backend_variables(char *filename, Port *port);
292 static void ShmemBackendArrayAdd(Backend *bn);
293 static void ShmemBackendArrayRemove(pid_t pid);
295 #endif /* EXEC_BACKEND */
297 #define StartupDataBase() StartChildProcess(BS_XLOG_STARTUP)
298 #define StartBackgroundWriter() StartChildProcess(BS_XLOG_BGWRITER)
302 * Postmaster main entry point
305 PostmasterMain(int argc, char *argv[])
309 char *potential_DataDir = NULL;
312 progname = get_progname(argv[0]);
314 MyProcPid = PostmasterPid = getpid();
316 IsPostmasterEnvironment = true;
319 * Catch standard options before doing much else. This even works on
320 * systems without getopt_long.
324 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
329 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
331 puts("postmaster (PostgreSQL) " PG_VERSION);
337 * for security, no dir or file created can be group or other
340 umask((mode_t) 0077);
343 * Fire up essential subsystems: memory management
348 * By default, palloc() requests in the postmaster will be allocated
349 * in the PostmasterContext, which is space that can be recycled by
350 * backends. Allocated data that needs to be available to backends
351 * should be allocated in TopMemoryContext.
353 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
355 ALLOCSET_DEFAULT_MINSIZE,
356 ALLOCSET_DEFAULT_INITSIZE,
357 ALLOCSET_DEFAULT_MAXSIZE);
358 MemoryContextSwitchTo(PostmasterContext);
360 IgnoreSystemIndexes(false);
362 if (find_my_exec(argv[0], my_exec_path) < 0)
363 elog(FATAL, "%s: could not locate my own executable path",
366 get_pkglib_path(my_exec_path, pkglib_path);
371 InitializeGUCOptions();
373 potential_DataDir = getenv("PGDATA"); /* default value */
377 while ((opt = getopt(argc, argv, "A:a:B:b:c:D:d:Fh:ik:lm:MN:no:p:Ss-:")) != -1)
382 #ifdef USE_ASSERT_CHECKING
383 SetConfigOption("debug_assertions", optarg, PGC_POSTMASTER, PGC_S_ARGV);
385 write_stderr("%s: assert checking is not compiled in\n", progname);
389 /* Can no longer set authentication method. */
392 SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
395 /* Can no longer set the backend executable file to use. */
398 potential_DataDir = optarg;
402 /* Turn on debugging for the postmaster. */
403 char *debugstr = palloc(strlen("debug") + strlen(optarg) + 1);
405 sprintf(debugstr, "debug%s", optarg);
406 SetConfigOption("log_min_messages", debugstr,
407 PGC_POSTMASTER, PGC_S_ARGV);
409 debug_flag = atoi(optarg);
413 SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
416 SetConfigOption("listen_addresses", optarg, PGC_POSTMASTER, PGC_S_ARGV);
419 SetConfigOption("listen_addresses", "*", PGC_POSTMASTER, PGC_S_ARGV);
422 SetConfigOption("unix_socket_directory", optarg, PGC_POSTMASTER, PGC_S_ARGV);
426 SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
430 /* Multiplexed backends no longer supported. */
435 * ignore this flag. This may be passed in because the
436 * program was run as 'postgres -M' instead of
441 /* The max number of backends to start. */
442 SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
445 /* Don't reinit shared mem after abnormal exit */
451 * Other options to pass to the backend on the command line
453 snprintf(ExtraOptions + strlen(ExtraOptions),
454 sizeof(ExtraOptions) - strlen(ExtraOptions),
458 SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
463 * Start in 'S'ilent mode (disassociate from controlling
464 * tty). You may also think of this as 'S'ysV mode since
465 * it's most badly needed on SysV-derived systems like
468 SetConfigOption("silent_mode", "true", PGC_POSTMASTER, PGC_S_ARGV);
473 * In the event that some backend dumps core, send
474 * SIGSTOP, rather than SIGQUIT, to all its peers. This
475 * lets the wily post_hacker collect core dumps from
486 ParseLongOption(optarg, &name, &value);
491 (errcode(ERRCODE_SYNTAX_ERROR),
492 errmsg("--%s requires a value",
496 (errcode(ERRCODE_SYNTAX_ERROR),
497 errmsg("-c %s requires a value",
501 SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
509 write_stderr("Try \"%s --help\" for more information.\n",
516 * Postmaster accepts no non-option switch arguments.
520 write_stderr("%s: invalid argument: \"%s\"\n",
521 progname, argv[optind]);
522 write_stderr("Try \"%s --help\" for more information.\n",
528 * Now we can set the data directory, and then read postgresql.conf.
530 checkDataDir(potential_DataDir); /* issues error messages */
531 SetDataDir(potential_DataDir);
533 ProcessConfigFile(PGC_POSTMASTER);
535 /* If timezone is not set, determine what the OS uses */
536 pg_timezone_initialize();
539 write_nondefault_variables(PGC_POSTMASTER);
543 * Check for invalid combinations of GUC settings.
545 if (NBuffers < 2 * MaxBackends || NBuffers < 16)
548 * Do not accept -B so small that backends are likely to starve
549 * for lack of buffers. The specific choices here are somewhat
552 write_stderr("%s: the number of buffers (-B) must be at least twice the number of allowed connections (-N) and at least 16\n", progname);
556 if (ReservedBackends >= MaxBackends)
558 write_stderr("%s: superuser_reserved_connections must be less than max_connections\n", progname);
563 * Other one-time internal sanity checks can go here.
565 if (!CheckDateTokenTables())
567 write_stderr("%s: invalid datetoken tables, please fix\n", progname);
572 * Now that we are done processing the postmaster arguments, reset
573 * getopt(3) library so that it will work correctly in subprocesses.
576 #ifdef HAVE_INT_OPTRESET
577 optreset = 1; /* some systems need this too */
580 /* For debugging: display postmaster environment */
582 extern char **environ;
586 (errmsg_internal("%s: PostmasterMain: initial environ dump:",
589 (errmsg_internal("-----------------------------------------")));
590 for (p = environ; *p; ++p)
592 (errmsg_internal("\t%s", *p)));
594 (errmsg_internal("-----------------------------------------")));
598 if (find_other_exec(argv[0], "postgres", PG_VERSIONSTR,
599 postgres_exec_path) < 0)
601 (errmsg("%s: could not locate matching postgres executable",
606 * Initialize SSL library, if specified.
614 * process any libraries that should be preloaded and optionally
617 if (preload_libraries_string)
618 process_preload_libraries(preload_libraries_string);
621 * Fork away from controlling terminal, if -S specified.
623 * Must do this before we grab any interlock files, else the interlocks
624 * will show the wrong PID.
630 * Create lockfile for data directory.
632 * We want to do this before we try to grab the input sockets, because
633 * the data directory interlock is more reliable than the socket-file
634 * interlock (thanks to whoever decided to put socket files in /tmp
635 * :-(). For the same reason, it's best to grab the TCP socket(s) before
638 CreateDataDirLockFile(DataDir, true);
641 * Remove old temporary files. At this point there can be no other
642 * Postgres processes running in this directory, so this should be
648 * Establish input sockets.
650 for (i = 0; i < MAXLISTEN; i++)
651 ListenSocket[i] = -1;
659 curhost = ListenAddresses;
662 /* ignore whitespace */
663 while (isspace((unsigned char) *curhost))
665 if (*curhost == '\0')
668 while (*endptr != '\0' && !isspace((unsigned char) *endptr))
672 if (strcmp(curhost, "*") == 0)
673 status = StreamServerPort(AF_UNSPEC, NULL,
674 (unsigned short) PostPortNumber,
676 ListenSocket, MAXLISTEN);
678 status = StreamServerPort(AF_UNSPEC, curhost,
679 (unsigned short) PostPortNumber,
681 ListenSocket, MAXLISTEN);
682 if (status != STATUS_OK)
684 (errmsg("could not create listen socket for \"%s\"",
688 curhost = endptr + 1;
694 #ifdef USE_RENDEZVOUS
695 /* Register for Rendezvous only if we opened TCP socket(s) */
696 if (ListenSocket[0] != -1 && rendezvous_name != NULL)
698 DNSServiceRegistrationCreate(rendezvous_name,
701 htonl(PostPortNumber),
703 (DNSServiceRegistrationReply) reg_reply,
708 #ifdef HAVE_UNIX_SOCKETS
709 status = StreamServerPort(AF_UNIX, NULL,
710 (unsigned short) PostPortNumber,
712 ListenSocket, MAXLISTEN);
713 if (status != STATUS_OK)
715 (errmsg("could not create Unix-domain socket")));
719 * check that we have some socket to listen on
721 if (ListenSocket[0] == -1)
723 (errmsg("no socket created for listening")));
728 * Set up shared memory and semaphores.
730 reset_shared(PostPortNumber);
733 * Estimate number of openable files. This must happen after setting
734 * up semaphores, because on some platforms semaphores count as open
740 * Initialize the list of active backends.
742 BackendList = DLNewList();
746 * Initialize the child pid/HANDLE arrays for signal handling.
748 win32_childPIDArray = (pid_t *)
749 malloc(NUM_BACKENDARRAY_ELEMS * sizeof(pid_t));
750 win32_childHNDArray = (HANDLE *)
751 malloc(NUM_BACKENDARRAY_ELEMS * sizeof(HANDLE));
752 if (!win32_childPIDArray || !win32_childHNDArray)
754 (errcode(ERRCODE_OUT_OF_MEMORY),
755 errmsg("out of memory")));
758 * Set up a handle that child processes can use to check whether the
759 * postmaster is still running.
761 if (DuplicateHandle(GetCurrentProcess(),
767 DUPLICATE_SAME_ACCESS) == 0)
769 (errmsg_internal("could not duplicate postmaster handle: %d",
770 (int) GetLastError())));
774 * Record postmaster options. We delay this till now to avoid
775 * recording bogus options (eg, NBuffers too high for available
778 if (!CreateOptsFile(argc, argv, my_exec_path))
782 * Set up signal handlers for the postmaster process.
784 * CAUTION: when changing this list, check for side-effects on the signal
785 * handling setup of child processes. See tcop/postgres.c,
786 * bootstrap/bootstrap.c, postmaster/bgwriter.c, and postmaster/pgstat.c.
789 PG_SETMASK(&BlockSig);
791 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
792 * children do same */
793 pqsignal(SIGINT, pmdie); /* send SIGTERM and shut down */
794 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
795 pqsignal(SIGTERM, pmdie); /* wait for children and shut down */
796 pqsignal(SIGALRM, SIG_IGN); /* ignored */
797 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
798 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
799 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
800 pqsignal(SIGCHLD, reaper); /* handle child termination */
801 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
802 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
803 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
805 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
809 * Reset whereToSendOutput from Debug (its starting state) to None.
810 * This prevents ereport from sending log messages to stderr unless
811 * the syslog/stderr switch permits. We don't do this until the
812 * postmaster is fully launched, since startup failures may as well be
813 * reported to stderr.
815 whereToSendOutput = None;
818 * Initialize the statistics collector stuff
823 * Load cached files for client authentication.
831 * We're ready to rock and roll...
833 StartupPID = StartupDataBase();
836 write_nondefault_variables(PGC_POSTMASTER);
839 status = ServerLoop();
842 * ServerLoop probably shouldn't ever return, but if it does, close
845 ExitPostmaster(status != STATUS_OK);
847 return 0; /* not reached */
852 * Validate the proposed data directory
855 checkDataDir(const char *checkdir)
857 char path[MAXPGPATH];
859 struct stat stat_buf;
861 if (checkdir == NULL)
863 write_stderr("%s does not know where to find the database system data.\n"
864 "You must specify the directory that contains the database system\n"
865 "either by specifying the -D invocation option or by setting the\n"
866 "PGDATA environment variable.\n",
871 if (stat(checkdir, &stat_buf) == -1)
875 (errcode_for_file_access(),
876 errmsg("data directory \"%s\" does not exist",
880 (errcode_for_file_access(),
881 errmsg("could not read permissions of directory \"%s\": %m",
886 * Check if the directory has group or world access. If so, reject.
888 * XXX temporarily suppress check when on Windows, because there may not
889 * be proper support for Unix-y file permissions. Need to think of a
890 * reasonable check to apply on Windows.
892 #if !defined(__CYGWIN__) && !defined(WIN32)
893 if (stat_buf.st_mode & (S_IRWXG | S_IRWXO))
895 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
896 errmsg("data directory \"%s\" has group or world access",
898 errdetail("Permissions should be u=rwx (0700).")));
901 /* Look for PG_VERSION before looking for pg_control */
902 ValidatePgVersion(checkdir);
904 snprintf(path, sizeof(path), "%s/global/pg_control", checkdir);
906 fp = AllocateFile(path, PG_BINARY_R);
909 write_stderr("%s: could not find the database system\n"
910 "Expected to find it in the directory \"%s\",\n"
911 "but could not open file \"%s\": %s\n",
912 progname, checkdir, path, strerror(errno));
919 #ifdef USE_RENDEZVOUS
922 * empty callback function for DNSServiceRegistrationCreate()
925 reg_reply(DNSServiceRegistrationReplyErrorType errorCode, void *context)
930 #endif /* USE_RENDEZVOUS */
934 * Fork away from the controlling terminal (-S option)
944 struct itimerval prof_itimer;
948 /* see comments in BackendStartup */
949 getitimer(ITIMER_PROF, &prof_itimer);
953 if (pid == (pid_t) -1)
955 write_stderr("%s: could not fork background process: %s\n",
956 progname, strerror(errno));
961 /* Parent should just exit, without doing any atexit cleanup */
966 setitimer(ITIMER_PROF, &prof_itimer, NULL);
969 MyProcPid = PostmasterPid = getpid(); /* reset PID vars to child */
971 /* GH: If there's no setsid(), we hopefully don't need silent mode.
972 * Until there's a better solution.
977 write_stderr("%s: could not dissociate from controlling TTY: %s\n",
978 progname, strerror(errno));
982 i = open(NULL_DEV, O_RDWR | PG_BINARY);
989 elog(FATAL, "SilentMode not supported under WIN32");
995 * Print out help message
998 usage(const char *progname)
1000 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
1001 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
1002 printf(gettext("Options:\n"));
1003 #ifdef USE_ASSERT_CHECKING
1004 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
1006 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
1007 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
1008 printf(gettext(" -d 1-5 debugging level\n"));
1009 printf(gettext(" -D DATADIR database directory\n"));
1010 printf(gettext(" -F turn fsync off\n"));
1011 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
1012 printf(gettext(" -i enable TCP/IP connections\n"));
1013 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
1015 printf(gettext(" -l enable SSL connections\n"));
1017 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
1018 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
1019 printf(gettext(" -p PORT port number to listen on\n"));
1020 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1021 printf(gettext(" --help show this help, then exit\n"));
1022 printf(gettext(" --version output version information, then exit\n"));
1024 printf(gettext("\nDeveloper options:\n"));
1025 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1026 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1028 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1029 "configuration settings and how to set them on the command line or in\n"
1030 "the configuration file.\n\n"
1031 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1036 * Main idle loop of postmaster
1045 struct timeval earlier,
1049 gettimeofday(&earlier, &tz);
1050 last_touch_time = time(NULL);
1052 nSockets = initMasks(&readmask);
1058 struct timeval timeout;
1063 * Wait for something to happen.
1065 * We wait at most one minute, to ensure that the other background
1066 * tasks handled below get done even when no requests are arriving.
1068 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1070 timeout.tv_sec = 60;
1071 timeout.tv_usec = 0;
1073 PG_SETMASK(&UnBlockSig);
1075 selres = select(nSockets, &rmask, NULL, NULL, &timeout);
1078 * Block all signals until we wait again. (This makes it safe for
1079 * our signal handlers to do nontrivial work.)
1081 PG_SETMASK(&BlockSig);
1085 if (errno == EINTR || errno == EWOULDBLOCK)
1088 (errcode_for_socket_access(),
1089 errmsg("select() failed in postmaster: %m")));
1090 return STATUS_ERROR;
1094 * New connection pending on any of our sockets? If so, fork a
1095 * child process to deal with it.
1100 * Select a random seed at the time of first receiving a request.
1102 while (random_seed == 0)
1104 gettimeofday(&later, &tz);
1107 * We are not sure how much precision is in tv_usec, so we
1108 * swap the nibbles of 'later' and XOR them with 'earlier'. On
1109 * the off chance that the result is 0, we loop until it isn't.
1111 random_seed = earlier.tv_usec ^
1112 ((later.tv_usec << 16) |
1113 ((later.tv_usec >> 16) & 0xffff));
1116 for (i = 0; i < MAXLISTEN; i++)
1118 if (ListenSocket[i] == -1)
1120 if (FD_ISSET(ListenSocket[i], &rmask))
1122 port = ConnCreate(ListenSocket[i]);
1125 BackendStartup(port);
1128 * We no longer need the open socket or port structure
1131 StreamClose(port->sock);
1139 * If no background writer process is running, and we are not in
1140 * a state that prevents it, start one. It doesn't matter if this
1141 * fails, we'll just try again later.
1143 if (BgWriterPID == 0 && StartupPID == 0 && !FatalError)
1145 BgWriterPID = StartBackgroundWriter();
1146 /* If shutdown is pending, set it going */
1147 if (Shutdown > NoShutdown && BgWriterPID != 0)
1148 kill(BgWriterPID, SIGUSR2);
1151 /* If we have lost the stats collector, try to start a new one */
1152 if (PgStatPID == 0 &&
1153 StartupPID == 0 && !FatalError && Shutdown == NoShutdown)
1154 PgStatPID = pgstat_start();
1157 * Touch the socket and lock file at least every ten minutes, to ensure
1158 * that they are not removed by overzealous /tmp-cleaning tasks.
1161 if (now - last_touch_time >= 10 * 60)
1164 TouchSocketLockFile();
1165 last_touch_time = now;
1172 * Initialise the masks for select() for the ports we are listening on.
1173 * Return the number of sockets to listen on.
1176 initMasks(fd_set *rmask)
1183 for (i = 0; i < MAXLISTEN; i++)
1185 int fd = ListenSocket[i];
1199 * Read the startup packet and do something according to it.
1201 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1202 * not return at all.
1204 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1205 * if that's what you want. Return STATUS_ERROR if you don't want to
1206 * send anything to the client, which would typically be appropriate
1207 * if we detect a communications failure.)
1210 ProcessStartupPacket(Port *port, bool SSLdone)
1214 ProtocolVersion proto;
1215 MemoryContext oldcontext;
1217 if (pq_getbytes((char *) &len, 4) == EOF)
1220 * EOF after SSLdone probably means the client didn't like our
1221 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1222 * so don't clutter the log with a complaint.
1226 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1227 errmsg("incomplete startup packet")));
1228 return STATUS_ERROR;
1234 if (len < (int32) sizeof(ProtocolVersion) ||
1235 len > MAX_STARTUP_PACKET_LENGTH)
1238 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1239 errmsg("invalid length of startup packet")));
1240 return STATUS_ERROR;
1244 * Allocate at least the size of an old-style startup packet, plus one
1245 * extra byte, and make sure all are zeroes. This ensures we will
1246 * have null termination of all strings, in both fixed- and
1247 * variable-length packet layouts.
1249 if (len <= (int32) sizeof(StartupPacket))
1250 buf = palloc0(sizeof(StartupPacket) + 1);
1252 buf = palloc0(len + 1);
1254 if (pq_getbytes(buf, len) == EOF)
1257 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1258 errmsg("incomplete startup packet")));
1259 return STATUS_ERROR;
1263 * The first field is either a protocol version number or a special
1266 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1268 if (proto == CANCEL_REQUEST_CODE)
1270 processCancelRequest(port, buf);
1271 return 127; /* XXX */
1274 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1279 /* No SSL when disabled or on Unix sockets */
1280 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1283 SSLok = 'S'; /* Support for SSL */
1285 SSLok = 'N'; /* No support for SSL */
1287 if (send(port->sock, &SSLok, 1, 0) != 1)
1290 (errcode_for_socket_access(),
1291 errmsg("failed to send SSL negotiation response: %m")));
1292 return STATUS_ERROR; /* close the connection */
1296 if (SSLok == 'S' && secure_open_server(port) == -1)
1297 return STATUS_ERROR;
1299 /* regular startup packet, cancel, etc packet should follow... */
1300 /* but not another SSL negotiation request */
1301 return ProcessStartupPacket(port, true);
1304 /* Could add additional special packet types here */
1307 * Set FrontendProtocol now so that ereport() knows what format to
1308 * send if we fail during startup.
1310 FrontendProtocol = proto;
1312 /* Check we can handle the protocol the frontend is using. */
1314 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1315 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1316 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1317 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1319 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1320 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1321 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1322 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1323 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1324 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1327 * Now fetch parameters out of startup packet and save them into the
1328 * Port structure. All data structures attached to the Port struct
1329 * must be allocated in TopMemoryContext so that they won't disappear
1330 * when we pass them to PostgresMain (see BackendRun). We need not
1331 * worry about leaking this storage on failure, since we aren't in the
1332 * postmaster process anymore.
1334 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1336 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1338 int32 offset = sizeof(ProtocolVersion);
1341 * Scan packet body for name/option pairs. We can assume any
1342 * string beginning within the packet body is null-terminated,
1343 * thanks to zeroing extra byte above.
1345 port->guc_options = NIL;
1347 while (offset < len)
1349 char *nameptr = ((char *) buf) + offset;
1353 if (*nameptr == '\0')
1354 break; /* found packet terminator */
1355 valoffset = offset + strlen(nameptr) + 1;
1356 if (valoffset >= len)
1357 break; /* missing value, will complain below */
1358 valptr = ((char *) buf) + valoffset;
1360 if (strcmp(nameptr, "database") == 0)
1361 port->database_name = pstrdup(valptr);
1362 else if (strcmp(nameptr, "user") == 0)
1363 port->user_name = pstrdup(valptr);
1364 else if (strcmp(nameptr, "options") == 0)
1365 port->cmdline_options = pstrdup(valptr);
1368 /* Assume it's a generic GUC option */
1369 port->guc_options = lappend(port->guc_options,
1371 port->guc_options = lappend(port->guc_options,
1374 offset = valoffset + strlen(valptr) + 1;
1378 * If we didn't find a packet terminator exactly at the end of the
1379 * given packet length, complain.
1381 if (offset != len - 1)
1383 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1384 errmsg("invalid startup packet layout: expected terminator as last byte")));
1389 * Get the parameters from the old-style, fixed-width-fields
1390 * startup packet as C strings. The packet destination was
1391 * cleared first so a short packet has zeros silently added. We
1392 * have to be prepared to truncate the pstrdup result for oversize
1395 StartupPacket *packet = (StartupPacket *) buf;
1397 port->database_name = pstrdup(packet->database);
1398 if (strlen(port->database_name) > sizeof(packet->database))
1399 port->database_name[sizeof(packet->database)] = '\0';
1400 port->user_name = pstrdup(packet->user);
1401 if (strlen(port->user_name) > sizeof(packet->user))
1402 port->user_name[sizeof(packet->user)] = '\0';
1403 port->cmdline_options = pstrdup(packet->options);
1404 if (strlen(port->cmdline_options) > sizeof(packet->options))
1405 port->cmdline_options[sizeof(packet->options)] = '\0';
1406 port->guc_options = NIL;
1409 /* Check a user name was given. */
1410 if (port->user_name == NULL || port->user_name[0] == '\0')
1412 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1413 errmsg("no PostgreSQL user name specified in startup packet")));
1415 /* The database defaults to the user name. */
1416 if (port->database_name == NULL || port->database_name[0] == '\0')
1417 port->database_name = pstrdup(port->user_name);
1419 if (Db_user_namespace)
1422 * If user@, it is a global user, remove '@'. We only want to do
1423 * this if there is an '@' at the end and no earlier in the user
1424 * string or they may fake as a local user of another database
1425 * attaching to this database.
1427 if (strchr(port->user_name, '@') ==
1428 port->user_name + strlen(port->user_name) - 1)
1429 *strchr(port->user_name, '@') = '\0';
1432 /* Append '@' and dbname */
1435 db_user = palloc(strlen(port->user_name) +
1436 strlen(port->database_name) + 2);
1437 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1438 port->user_name = db_user;
1443 * Truncate given database and user names to length of a Postgres
1444 * name. This avoids lookup failures when overlength names are given.
1446 if (strlen(port->database_name) >= NAMEDATALEN)
1447 port->database_name[NAMEDATALEN - 1] = '\0';
1448 if (strlen(port->user_name) >= NAMEDATALEN)
1449 port->user_name[NAMEDATALEN - 1] = '\0';
1452 * Done putting stuff in TopMemoryContext.
1454 MemoryContextSwitchTo(oldcontext);
1457 * If we're going to reject the connection due to database state, say
1458 * so now instead of wasting cycles on an authentication exchange.
1459 * (This also allows a pg_ping utility to be written.)
1461 switch (port->canAcceptConnections)
1465 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1466 errmsg("the database system is starting up")));
1470 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1471 errmsg("the database system is shutting down")));
1475 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1476 errmsg("the database system is in recovery mode")));
1480 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1481 errmsg("sorry, too many clients already")));
1493 * The client has sent a cancel request packet, not a normal
1494 * start-a-new-connection packet. Perform the necessary processing.
1495 * Nothing is sent back to the client.
1498 processCancelRequest(Port *port, void *pkt)
1500 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1502 long cancelAuthCode;
1504 #ifndef EXEC_BACKEND
1510 backendPID = (int) ntohl(canc->backendPID);
1511 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1514 * See if we have a matching backend. In the EXEC_BACKEND case, we
1515 * can no longer access the postmaster's own backend list, and must
1516 * rely on the duplicate array in shared memory.
1518 #ifndef EXEC_BACKEND
1519 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1521 bp = (Backend *) DLE_VAL(curr);
1523 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1525 bp = (Backend *) &ShmemBackendArray[i];
1527 if (bp->pid == backendPID)
1529 if (bp->cancel_key == cancelAuthCode)
1531 /* Found a match; signal that backend to cancel current op */
1533 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1535 kill(bp->pid, SIGINT);
1538 /* Right PID, wrong key: no way, Jose */
1540 (errmsg_internal("bad key in cancel request for process %d",
1546 /* No matching backend */
1548 (errmsg_internal("bad pid in cancel request for process %d",
1553 * canAcceptConnections --- check to see if database state allows connections.
1555 static enum CAC_state
1556 canAcceptConnections(void)
1558 /* Can't start backends when in startup/shutdown/recovery state. */
1559 if (Shutdown > NoShutdown)
1560 return CAC_SHUTDOWN;
1564 return CAC_RECOVERY;
1567 * Don't start too many children.
1569 * We allow more connections than we can have backends here because some
1570 * might still be authenticating; they might fail auth, or some
1571 * existing backend might exit before the auth cycle is completed. The
1572 * exact MaxBackends limit is enforced when a new backend tries to
1573 * join the shared-inval backend array.
1575 if (CountChildren() >= 2 * MaxBackends)
1583 * ConnCreate -- create a local connection data structure
1586 ConnCreate(int serverFd)
1590 if (!(port = (Port *) calloc(1, sizeof(Port))))
1593 (errcode(ERRCODE_OUT_OF_MEMORY),
1594 errmsg("out of memory")));
1598 if (StreamConnection(serverFd, port) != STATUS_OK)
1600 StreamClose(port->sock);
1607 * Precompute password salt values to use for this connection.
1608 * It's slightly annoying to do this long in advance of knowing
1609 * whether we'll need 'em or not, but we must do the random()
1610 * calls before we fork, not after. Else the postmaster's random
1611 * sequence won't get advanced, and all backends would end up
1612 * using the same salt...
1614 RandomSalt(port->cryptSalt, port->md5Salt);
1622 * ConnFree -- free a local connection data structure
1625 ConnFree(Port *conn)
1635 * ClosePostmasterPorts -- close all the postmaster's open sockets
1637 * This is called during child process startup to release file descriptors
1638 * that are not needed by that child process. The postmaster still has
1639 * them open, of course.
1642 ClosePostmasterPorts(void)
1646 /* Close the listen sockets */
1647 for (i = 0; i < MAXLISTEN; i++)
1649 if (ListenSocket[i] != -1)
1651 StreamClose(ListenSocket[i]);
1652 ListenSocket[i] = -1;
1659 * reset_shared -- reset shared memory and semaphores
1662 reset_shared(unsigned short port)
1665 * Create or re-create shared memory and semaphores.
1667 * Note: in each "cycle of life" we will normally assign the same IPC
1668 * keys (if using SysV shmem and/or semas), since the port number is
1669 * used to determine IPC keys. This helps ensure that we will clean
1670 * up dead IPC objects if the postmaster crashes and is restarted.
1672 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1677 * SIGHUP -- reread config files, and tell children to do same
1680 SIGHUP_handler(SIGNAL_ARGS)
1682 int save_errno = errno;
1684 PG_SETMASK(&BlockSig);
1686 if (Shutdown <= SmartShutdown)
1689 (errmsg("received SIGHUP, reloading configuration files")));
1690 ProcessConfigFile(PGC_SIGHUP);
1691 SignalChildren(SIGHUP);
1692 if (BgWriterPID != 0)
1693 kill(BgWriterPID, SIGHUP);
1694 /* PgStatPID does not currently need SIGHUP */
1699 /* Update the starting-point file for future children */
1700 write_nondefault_variables(PGC_SIGHUP);
1704 PG_SETMASK(&UnBlockSig);
1711 * pmdie -- signal handler for processing various postmaster signals.
1716 int save_errno = errno;
1718 PG_SETMASK(&BlockSig);
1721 (errmsg_internal("postmaster received signal %d",
1722 postgres_signal_arg)));
1724 switch (postgres_signal_arg)
1730 * Wait for children to end their work, then shut down.
1732 if (Shutdown >= SmartShutdown)
1734 Shutdown = SmartShutdown;
1736 (errmsg("received smart shutdown request")));
1738 if (DLGetHead(BackendList))
1739 break; /* let reaper() handle this */
1742 * No children left. Begin shutdown of data base system.
1744 if (StartupPID != 0 || FatalError)
1745 break; /* let reaper() handle this */
1746 /* Start the bgwriter if not running */
1747 if (BgWriterPID == 0)
1748 BgWriterPID = StartBackgroundWriter();
1749 /* And tell it to shut down */
1750 if (BgWriterPID != 0)
1751 kill(BgWriterPID, SIGUSR2);
1752 /* Tell pgstat to shut down too; nothing left for it to do */
1754 kill(PgStatPID, SIGQUIT);
1761 * Abort all children with SIGTERM (rollback active transactions
1762 * and exit) and shut down when they are gone.
1764 if (Shutdown >= FastShutdown)
1766 Shutdown = FastShutdown;
1768 (errmsg("received fast shutdown request")));
1770 if (DLGetHead(BackendList))
1775 (errmsg("aborting any active transactions")));
1776 SignalChildren(SIGTERM);
1777 /* reaper() does the rest */
1783 * No children left. Begin shutdown of data base system.
1785 * Note: if we previously got SIGTERM then we may send SIGUSR2
1786 * to the bgwriter a second time here. This should be harmless.
1788 if (StartupPID != 0 || FatalError)
1789 break; /* let reaper() handle this */
1790 /* Start the bgwriter if not running */
1791 if (BgWriterPID == 0)
1792 BgWriterPID = StartBackgroundWriter();
1793 /* And tell it to shut down */
1794 if (BgWriterPID != 0)
1795 kill(BgWriterPID, SIGUSR2);
1796 /* Tell pgstat to shut down too; nothing left for it to do */
1798 kill(PgStatPID, SIGQUIT);
1803 * Immediate Shutdown:
1805 * abort all children with SIGQUIT and exit without attempt to
1806 * properly shut down data base system.
1809 (errmsg("received immediate shutdown request")));
1810 if (StartupPID != 0)
1811 kill(StartupPID, SIGQUIT);
1812 if (BgWriterPID != 0)
1813 kill(BgWriterPID, SIGQUIT);
1815 kill(PgStatPID, SIGQUIT);
1816 if (DLGetHead(BackendList))
1817 SignalChildren(SIGQUIT);
1822 PG_SETMASK(&UnBlockSig);
1828 * Reaper -- signal handler to cleanup after a backend (child) dies.
1833 int save_errno = errno;
1836 int status; /* backend exit status */
1840 union wait status; /* backend exit status */
1844 int pid; /* process id of dead backend */
1846 PG_SETMASK(&BlockSig);
1849 (errmsg_internal("reaping dead processes")));
1851 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1853 exitstatus = status;
1856 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1858 exitstatus = status.w_status;
1860 while ((pid = win32_waitpid(&exitstatus)) > 0)
1863 * We need to do this here, and not in CleanupProc, since this is
1864 * to be called on all children when we are done with them. Could
1865 * move to LogChildExit, but that seems like asking for future
1868 win32_RemoveChild(pid);
1870 #endif /* HAVE_WAITPID */
1873 * Check if this child was a startup process.
1875 if (StartupPID != 0 && pid == StartupPID)
1878 if (exitstatus != 0)
1880 LogChildExit(LOG, gettext("startup process"),
1883 (errmsg("aborting startup due to startup process failure")));
1888 * Startup succeeded - we are done with system startup or recovery.
1893 * Crank up the background writer. It doesn't matter if this
1894 * fails, we'll just try again later.
1896 Assert(BgWriterPID == 0);
1897 BgWriterPID = StartBackgroundWriter();
1900 * Go to shutdown mode if a shutdown request was pending.
1901 * Otherwise, try to start the stats collector too.
1903 if (Shutdown > NoShutdown && BgWriterPID != 0)
1904 kill(BgWriterPID, SIGUSR2);
1905 else if (PgStatPID == 0 && Shutdown == NoShutdown)
1906 PgStatPID = pgstat_start();
1912 * Was it the bgwriter?
1914 if (BgWriterPID != 0 && pid == BgWriterPID)
1917 if (exitstatus == 0 && Shutdown > NoShutdown &&
1918 !FatalError && !DLGetHead(BackendList))
1921 * Normal postmaster exit is here: we've seen normal
1922 * exit of the bgwriter after it's been told to shut down.
1923 * We expect that it wrote a shutdown checkpoint. (If
1924 * for some reason it didn't, recovery will occur on next
1925 * postmaster start.)
1930 * Any unexpected exit of the bgwriter is treated as a crash.
1932 LogChildExit(LOG, gettext("background writer process"),
1934 HandleChildCrash(pid, exitstatus);
1939 * Was it the statistics collector? If so, just try to start a new
1940 * one; no need to force reset of the rest of the system. (If fail,
1941 * we'll try again in future cycles of the main loop.)
1943 if (PgStatPID != 0 && pid == PgStatPID)
1946 if (exitstatus != 0)
1947 LogChildExit(LOG, gettext("statistics collector process"),
1949 if (StartupPID == 0 && !FatalError && Shutdown == NoShutdown)
1950 PgStatPID = pgstat_start();
1955 * Else do standard backend child cleanup.
1957 CleanupProc(pid, exitstatus);
1958 } /* loop over pending child-death reports */
1963 * Wait for all children exit, then reset shmem and
1966 if (DLGetHead(BackendList) || StartupPID != 0 || BgWriterPID != 0)
1969 (errmsg("all server processes terminated; reinitializing")));
1972 reset_shared(PostPortNumber);
1974 StartupPID = StartupDataBase();
1979 if (Shutdown > NoShutdown)
1981 if (DLGetHead(BackendList) || StartupPID != 0)
1983 /* Start the bgwriter if not running */
1984 if (BgWriterPID == 0)
1985 BgWriterPID = StartBackgroundWriter();
1986 /* And tell it to shut down */
1987 if (BgWriterPID != 0)
1988 kill(BgWriterPID, SIGUSR2);
1992 PG_SETMASK(&UnBlockSig);
1999 * CleanupProc -- cleanup after terminated backend.
2001 * Remove all local state associated with backend.
2004 CleanupProc(int pid,
2005 int exitstatus) /* child's exit status. */
2009 LogChildExit(DEBUG2, gettext("server process"), pid, exitstatus);
2012 * If a backend dies in an ugly way (i.e. exit status not 0) then we
2013 * must signal all other backends to quickdie. If exit status is zero
2014 * we assume everything is hunky dory and simply remove the backend
2015 * from the active backend list.
2017 if (exitstatus != 0)
2019 HandleChildCrash(pid, exitstatus);
2023 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2025 Backend *bp = (Backend *) DLE_VAL(curr);
2033 ShmemBackendArrayRemove(pid);
2035 /* Tell the collector about backend termination */
2043 * HandleChildCrash -- cleanup after failed backend or bgwriter.
2045 * The objectives here are to clean up our local state about the child
2046 * process, and to signal all other remaining children to quickdie.
2049 HandleChildCrash(int pid,
2050 int exitstatus) /* child's exit status. */
2057 * Make log entry unless there was a previous crash (if so, nonzero
2058 * exit status is to be expected in SIGQUIT response; don't clutter log)
2063 (pid == BgWriterPID) ?
2064 gettext("background writer process") :
2065 gettext("server process"),
2068 (errmsg("terminating any other active server processes")));
2071 /* Process regular backends */
2072 for (curr = DLGetHead(BackendList); curr; curr = next)
2074 next = DLGetSucc(curr);
2075 bp = (Backend *) DLE_VAL(curr);
2079 * Found entry for freshly-dead backend, so remove it.
2085 ShmemBackendArrayRemove(pid);
2087 /* Tell the collector about backend termination */
2089 /* Keep looping so we can signal remaining backends */
2094 * This backend is still alive. Unless we did so already,
2095 * tell it to commit hara-kiri.
2097 * SIGQUIT is the special signal that says exit without proc_exit
2098 * and let the user know what's going on. But if SendStop is
2099 * set (-s on command line), then we send SIGSTOP instead, so
2100 * that we can get core dumps from all backends by hand.
2105 (errmsg_internal("sending %s to process %d",
2106 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2108 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2113 /* Take care of the bgwriter too */
2114 if (pid == BgWriterPID)
2116 else if (BgWriterPID != 0 && !FatalError)
2119 (errmsg_internal("sending %s to process %d",
2120 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2121 (int) BgWriterPID)));
2122 kill(BgWriterPID, (SendStop ? SIGSTOP : SIGQUIT));
2125 /* Force a power-cycle of the pgstat processes too */
2126 /* (Shouldn't be necessary, but just for luck) */
2127 if (PgStatPID != 0 && !FatalError)
2130 (errmsg_internal("sending %s to process %d",
2133 kill(PgStatPID, SIGQUIT);
2140 * Log the death of a child process.
2143 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2145 if (WIFEXITED(exitstatus))
2149 * translator: %s is a noun phrase describing a child process,
2150 * such as "server process"
2152 (errmsg("%s (PID %d) exited with exit code %d",
2153 procname, pid, WEXITSTATUS(exitstatus))));
2154 else if (WIFSIGNALED(exitstatus))
2158 * translator: %s is a noun phrase describing a child process,
2159 * such as "server process"
2161 (errmsg("%s (PID %d) was terminated by signal %d",
2162 procname, pid, WTERMSIG(exitstatus))));
2167 * translator: %s is a noun phrase describing a child process,
2168 * such as "server process"
2170 (errmsg("%s (PID %d) exited with unexpected status %d",
2171 procname, pid, exitstatus)));
2175 * Send a signal to all backend children (but NOT special children)
2178 SignalChildren(int signal)
2182 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2184 Backend *bp = (Backend *) DLE_VAL(curr);
2187 (errmsg_internal("sending signal %d to process %d",
2188 signal, (int) bp->pid)));
2189 kill(bp->pid, signal);
2194 * BackendStartup -- start backend process
2196 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2199 BackendStartup(Port *port)
2201 Backend *bn; /* for backend cleanup */
2204 #ifdef LINUX_PROFILE
2205 struct itimerval prof_itimer;
2209 * Compute the cancel key that will be assigned to this backend. The
2210 * backend will have its own copy in the forked-off process' value of
2211 * MyCancelKey, so that it can transmit the key to the frontend.
2213 MyCancelKey = PostmasterRandom();
2216 * Make room for backend data structure. Better before the fork() so
2217 * we can handle failure cleanly.
2219 bn = (Backend *) malloc(sizeof(Backend));
2223 (errcode(ERRCODE_OUT_OF_MEMORY),
2224 errmsg("out of memory")));
2225 return STATUS_ERROR;
2228 /* Pass down canAcceptConnections state (kluge for EXEC_BACKEND case) */
2229 port->canAcceptConnections = canAcceptConnections();
2232 * Flush stdio channels just before fork, to avoid double-output
2233 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2234 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2235 * coredump if we do. Presently stdout and stderr are the only stdio
2236 * output channels used by the postmaster, so fflush'ing them should
2244 pid = backend_forkexec(port);
2246 #else /* !EXEC_BACKEND */
2248 #ifdef LINUX_PROFILE
2251 * Linux's fork() resets the profiling timer in the child process. If
2252 * we want to profile child processes then we need to save and restore
2253 * the timer setting. This is a waste of time if not profiling,
2254 * however, so only do it if commanded by specific -DLINUX_PROFILE
2257 getitimer(ITIMER_PROF, &prof_itimer);
2261 /* Specific beos actions before backend startup */
2262 beos_before_backend_startup();
2267 if (pid == 0) /* child */
2269 #ifdef LINUX_PROFILE
2270 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2274 /* Specific beos backend startup actions */
2275 beos_backend_startup();
2279 proc_exit(BackendRun(port));
2282 #endif /* EXEC_BACKEND */
2286 /* in parent, fork failed */
2287 int save_errno = errno;
2290 /* Specific beos backend startup actions */
2291 beos_backend_startup_failed();
2296 (errmsg("could not fork new process for connection: %m")));
2297 report_fork_failure_to_client(port, save_errno);
2298 return STATUS_ERROR;
2301 /* in parent, successful fork */
2303 (errmsg_internal("forked new backend, pid=%d socket=%d",
2304 (int) pid, port->sock)));
2307 * Everything's been successful, it's safe to add this backend to our
2311 bn->cancel_key = MyCancelKey;
2312 DLAddHead(BackendList, DLNewElem(bn));
2314 ShmemBackendArrayAdd(bn);
2321 * Try to report backend fork() failure to client before we close the
2322 * connection. Since we do not care to risk blocking the postmaster on
2323 * this connection, we set the connection to non-blocking and try only once.
2325 * This is grungy special-purpose code; we cannot use backend libpq since
2326 * it's not up and running.
2329 report_fork_failure_to_client(Port *port, int errnum)
2333 /* Format the error message packet (always V2 protocol) */
2334 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2335 gettext("could not fork new process for connection: "),
2338 /* Set port to non-blocking. Don't do send() if this fails */
2339 if (!set_noblock(port->sock))
2342 send(port->sock, buffer, strlen(buffer) + 1, 0);
2347 * split_opts -- split a string of options and append it to an argv array
2349 * NB: the string is destructively modified!
2351 * Since no current POSTGRES arguments require any quoting characters,
2352 * we can use the simple-minded tactic of assuming each set of space-
2353 * delimited characters is a separate argv element.
2355 * If you don't like that, well, we *used* to pass the whole option string
2356 * as ONE argument to execl(), which was even less intelligent...
2359 split_opts(char **argv, int *argcp, char *s)
2363 while (isspace((unsigned char) *s))
2367 argv[(*argcp)++] = s;
2368 while (*s && !isspace((unsigned char) *s))
2377 * BackendRun -- perform authentication, and if successful,
2378 * set up the backend's argument list and invoke PostgresMain()
2381 * Shouldn't return at all.
2382 * If PostgresMain() fails, return status.
2385 BackendRun(Port *port)
2390 char remote_host[NI_MAXHOST];
2391 char remote_port[NI_MAXSERV];
2392 char remote_ps_data[NI_MAXHOST];
2400 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2403 * Let's clean up ourselves as the postmaster child, and close the
2404 * postmaster's listen sockets
2406 ClosePostmasterPorts();
2408 /* We don't want the postmaster's proc_exit() handlers */
2412 * Signal handlers setting is moved to tcop/postgres...
2415 /* Save port etc. for ps status */
2418 /* Reset MyProcPid to new backend's pid */
2419 MyProcPid = getpid();
2422 * PreAuthDelay is a debugging aid for investigating problems in the
2423 * authentication cycle: it can be set in postgresql.conf to allow
2424 * time to attach to the newly-forked backend with a debugger. (See
2425 * also the -W backend switch, which we allow clients to pass through
2426 * PGOPTIONS, but it is not honored until after authentication.)
2428 if (PreAuthDelay > 0)
2429 pg_usleep(PreAuthDelay * 1000000L);
2431 ClientAuthInProgress = true; /* limit visibility of log messages */
2433 /* save start time for end of session reporting */
2434 gettimeofday(&(port->session_start), NULL);
2436 /* set these to empty in case they are needed before we set them up */
2437 port->remote_host = "";
2438 port->remote_port = "";
2439 port->commandTag = "";
2442 * Initialize libpq and enable reporting of ereport errors to the
2443 * client. Must do this now because authentication uses libpq to send
2446 pq_init(); /* initialize libpq to talk to client */
2447 whereToSendOutput = Remote; /* now safe to ereport to client */
2450 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2451 * during any client authentication related communication. Otherwise
2452 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2453 * if a buggy client blocks a backend during authentication.
2455 pqsignal(SIGTERM, authdie);
2456 pqsignal(SIGQUIT, authdie);
2457 pqsignal(SIGALRM, authdie);
2458 PG_SETMASK(&AuthBlockSig);
2461 * Get the remote host name and port for logging and status display.
2463 remote_host[0] = '\0';
2464 remote_port[0] = '\0';
2465 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2466 remote_host, sizeof(remote_host),
2467 remote_port, sizeof(remote_port),
2468 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2470 int ret = getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2471 remote_host, sizeof(remote_host),
2472 remote_port, sizeof(remote_port),
2473 NI_NUMERICHOST | NI_NUMERICSERV);
2477 (errmsg("getnameinfo_all() failed: %s",
2478 gai_strerror(ret))));
2480 snprintf(remote_ps_data, sizeof(remote_ps_data),
2481 remote_port[0] == '\0' ? "%s" : "%s(%s)",
2482 remote_host, remote_port);
2484 if (Log_connections)
2486 (errmsg("connection received: host=%s port=%s",
2487 remote_host, remote_port)));
2490 * save remote_host and remote_port in port stucture
2492 port->remote_host = strdup(remote_host);
2493 port->remote_port = strdup(remote_port);
2496 * In EXEC_BACKEND case, we didn't inherit the contents of pg_hba.c
2497 * etcetera from the postmaster, and have to load them ourselves.
2498 * Build the PostmasterContext (which didn't exist before, in this
2499 * process) to contain the data.
2501 * FIXME: [fork/exec] Ugh. Is there a way around this overhead?
2504 Assert(PostmasterContext == NULL);
2505 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
2507 ALLOCSET_DEFAULT_MINSIZE,
2508 ALLOCSET_DEFAULT_INITSIZE,
2509 ALLOCSET_DEFAULT_MAXSIZE);
2510 MemoryContextSwitchTo(PostmasterContext);
2519 * Ready to begin client interaction. We will give up and exit(0)
2520 * after a time delay, so that a broken client can't hog a connection
2521 * indefinitely. PreAuthDelay doesn't count against the time limit.
2523 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2524 elog(FATAL, "could not set timer for authorization timeout");
2527 * Receive the startup packet (which might turn out to be a cancel
2530 status = ProcessStartupPacket(port, false);
2532 if (status != STATUS_OK)
2536 * Now that we have the user and database name, we can set the process
2537 * title for ps. It's good to do this as early as possible in
2540 init_ps_display(port->user_name, port->database_name, remote_ps_data);
2541 set_ps_display("authentication");
2544 * Now perform authentication exchange.
2546 ClientAuthentication(port); /* might not return, if failure */
2549 * Done with authentication. Disable timeout, and prevent
2550 * SIGTERM/SIGQUIT again until backend startup is complete.
2552 if (!disable_sig_alarm(false))
2553 elog(FATAL, "could not disable timer for authorization timeout");
2554 PG_SETMASK(&BlockSig);
2556 if (Log_connections)
2558 (errmsg("connection authorized: user=%s database=%s",
2559 port->user_name, port->database_name)));
2562 * Don't want backend to be able to see the postmaster random number
2563 * generator state. We have to clobber the static random_seed *and*
2564 * start a new random sequence in the random() library function.
2567 gettimeofday(&now, &tz);
2568 srandom((unsigned int) now.tv_usec);
2572 * Now, build the argv vector that will be given to PostgresMain.
2574 * The layout of the command line is
2575 * postgres [secure switches] -p databasename [insecure switches]
2576 * where the switches after -p come from the client request.
2578 * The maximum possible number of commandline arguments that could come
2579 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2583 maxac = 10; /* for fixed args supplied below */
2584 maxac += (strlen(ExtraOptions) + 1) / 2;
2585 if (port->cmdline_options)
2586 maxac += (strlen(port->cmdline_options) + 1) / 2;
2588 av = (char **) MemoryContextAlloc(TopMemoryContext,
2589 maxac * sizeof(char *));
2592 av[ac++] = "postgres";
2595 * Pass the requested debugging level along to the backend.
2599 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2600 av[ac++] = debugbuf;
2604 * Pass any backend switches specified with -o in the postmaster's own
2605 * command line. We assume these are secure. (It's OK to mangle
2606 * ExtraOptions now, since we're safely inside a subprocess.)
2608 split_opts(av, &ac, ExtraOptions);
2610 /* Tell the backend what protocol the frontend is using. */
2611 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2612 av[ac++] = protobuf;
2615 * Tell the backend it is being called from the postmaster, and which
2616 * database to use. -p marks the end of secure switches.
2619 av[ac++] = port->database_name;
2622 * Pass the (insecure) option switches from the connection request.
2623 * (It's OK to mangle port->cmdline_options now.)
2625 if (port->cmdline_options)
2626 split_opts(av, &ac, port->cmdline_options);
2633 * Release postmaster's working memory context so that backend can
2634 * recycle the space. Note this does not trash *MyProcPort, because
2635 * ConnCreate() allocated that space with malloc() ... else we'd need
2636 * to copy the Port data here. Also, subsidiary data such as the
2637 * username isn't lost either; see ProcessStartupPacket().
2639 MemoryContextSwitchTo(TopMemoryContext);
2640 MemoryContextDelete(PostmasterContext);
2641 PostmasterContext = NULL;
2644 * Debug: print arguments being passed to backend
2647 (errmsg_internal("%s child[%d]: starting with (",
2648 progname, getpid())));
2649 for (i = 0; i < ac; ++i)
2651 (errmsg_internal("\t%s", av[i])));
2653 (errmsg_internal(")")));
2655 ClientAuthInProgress = false; /* client_min_messages is active
2658 return (PostgresMain(ac, av, port->user_name));
2665 * postmaster_forkexec -- fork and exec a postmaster subprocess
2667 * The caller must have set up the argv array already, except for argv[2]
2668 * which will be filled with the name of the temp variable file.
2670 * Returns the child process PID, or -1 on fork failure (a suitable error
2671 * message has been logged on failure).
2673 * All uses of this routine will dispatch to SubPostmasterMain in the
2677 postmaster_forkexec(int argc, char *argv[])
2681 /* This entry point passes dummy values for the Port variables */
2682 memset(&port, 0, sizeof(port));
2683 return internal_forkexec(argc, argv, &port);
2687 * backend_forkexec -- fork/exec off a backend process
2689 * returns the pid of the fork/exec'd process, or -1 on failure
2692 backend_forkexec(Port *port)
2697 av[ac++] = "postgres";
2698 av[ac++] = "-forkbackend";
2699 av[ac++] = NULL; /* filled in by internal_forkexec */
2702 Assert(ac < lengthof(av));
2704 return internal_forkexec(ac, av, port);
2708 internal_forkexec(int argc, char *argv[], Port *port)
2711 char tmpfilename[MAXPGPATH];
2713 if (!write_backend_variables(tmpfilename, port))
2714 return -1; /* log made by write_backend_variables */
2716 /* Make sure caller set up argv properly */
2718 Assert(argv[argc] == NULL);
2719 Assert(strncmp(argv[1], "-fork", 5) == 0);
2720 Assert(argv[2] == NULL);
2722 /* Insert temp file name after -fork argument */
2723 argv[2] = tmpfilename;
2726 pid = win32_forkexec(postgres_exec_path, argv);
2728 /* Fire off execv in child */
2729 if ((pid = fork()) == 0)
2731 if (execv(postgres_exec_path, argv) < 0)
2734 (errmsg("could not exec backend process \"%s\": %m",
2735 postgres_exec_path)));
2736 /* We're already in the child process here, can't return */
2742 return pid; /* Parent returns pid, or -1 on fork failure */
2746 * SubPostmasterMain -- Get the fork/exec'd process into a state equivalent
2747 * to what it would be if we'd simply forked on Unix, and then
2748 * dispatch to the appropriate place.
2750 * The first two command line arguments are expected to be "-forkFOO"
2751 * (where FOO indicates which postmaster child we are to become), and
2752 * the name of a variables file that we can read to load data that would
2753 * have been inherited by fork() on Unix. Remaining arguments go to the
2754 * subprocess FooMain() routine.
2757 SubPostmasterMain(int argc, char *argv[])
2761 /* Do this sooner rather than later... */
2762 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2764 MyProcPid = getpid(); /* reset MyProcPid */
2766 /* In EXEC_BACKEND case we will not have inherited these settings */
2767 IsPostmasterEnvironment = true;
2768 whereToSendOutput = None;
2770 PG_SETMASK(&BlockSig);
2772 /* Setup essential subsystems */
2773 MemoryContextInit();
2774 InitializeGUCOptions();
2776 /* Check we got appropriate args */
2778 elog(FATAL, "invalid subpostmaster invocation");
2780 /* Read in file-based context */
2781 memset(&port, 0, sizeof(Port));
2782 read_backend_variables(argv[2], &port);
2783 read_nondefault_variables();
2785 /* Run backend or appropriate child */
2786 if (strcmp(argv[1], "-forkbackend") == 0)
2788 /* BackendRun will close sockets */
2790 /* Attach process to shared segments */
2791 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2793 Assert(argc == 3); /* shouldn't be any more args */
2794 proc_exit(BackendRun(&port));
2796 if (strcmp(argv[1], "-forkboot") == 0)
2798 /* Close the postmaster's sockets */
2799 ClosePostmasterPorts();
2801 /* Attach process to shared segments */
2802 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2804 BootstrapMain(argc - 2, argv + 2);
2807 if (strcmp(argv[1], "-forkbuf") == 0)
2809 /* Close the postmaster's sockets */
2810 ClosePostmasterPorts();
2812 /* Do not want to attach to shared memory */
2814 PgstatBufferMain(argc, argv);
2817 if (strcmp(argv[1], "-forkcol") == 0)
2820 * Do NOT close postmaster sockets here, because we are forking from
2821 * pgstat buffer process, which already did it.
2824 /* Do not want to attach to shared memory */
2826 PgstatCollectorMain(argc, argv);
2830 return 1; /* shouldn't get here */
2833 #endif /* EXEC_BACKEND */
2837 * ExitPostmaster -- cleanup
2839 * Do NOT call exit() directly --- always go through here!
2842 ExitPostmaster(int status)
2844 /* should cleanup shared memory and kill all backends */
2847 * Not sure of the semantics here. When the Postmaster dies, should
2848 * the backends all be killed? probably not.
2850 * MUST -- vadim 05-10-1999
2857 * sigusr1_handler - handle signal conditions from child processes
2860 sigusr1_handler(SIGNAL_ARGS)
2862 int save_errno = errno;
2864 PG_SETMASK(&BlockSig);
2866 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2869 * Password or group file has changed.
2875 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2878 * Send SIGUSR1 to all children (triggers
2879 * CatchupInterruptHandler). See storage/ipc/sinval[adt].c for the
2882 if (Shutdown <= SmartShutdown)
2883 SignalChildren(SIGUSR1);
2886 PG_SETMASK(&UnBlockSig);
2893 * Dummy signal handler
2895 * We use this for signals that we don't actually use in the postmaster,
2896 * but we do use in backends. If we were to SIG_IGN such signals in the
2897 * postmaster, then a newly started backend might drop a signal that arrives
2898 * before it's able to reconfigure its signal processing. (See notes in
2902 dummy_handler(SIGNAL_ARGS)
2908 * CharRemap: given an int in range 0..61, produce textual encoding of it
2909 * per crypt(3) conventions.
2933 RandomSalt(char *cryptSalt, char *md5Salt)
2935 long rand = PostmasterRandom();
2937 cryptSalt[0] = CharRemap(rand % 62);
2938 cryptSalt[1] = CharRemap(rand / 62);
2941 * It's okay to reuse the first random value for one of the MD5 salt
2942 * bytes, since only one of the two salts will be sent to the client.
2943 * After that we need to compute more random bits.
2945 * We use % 255, sacrificing one possible byte value, so as to ensure
2946 * that all bits of the random() value participate in the result.
2947 * While at it, add one to avoid generating any null bytes.
2949 md5Salt[0] = (rand % 255) + 1;
2950 rand = PostmasterRandom();
2951 md5Salt[1] = (rand % 255) + 1;
2952 rand = PostmasterRandom();
2953 md5Salt[2] = (rand % 255) + 1;
2954 rand = PostmasterRandom();
2955 md5Salt[3] = (rand % 255) + 1;
2962 PostmasterRandom(void)
2964 static bool initialized = false;
2968 Assert(random_seed != 0);
2969 srandom(random_seed);
2977 * Count up number of child processes (regular backends only)
2985 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2994 * StartChildProcess -- start a non-backend child process for the postmaster
2996 * xlog determines what kind of child will be started. All child types
2997 * initially go to BootstrapMain, which will handle common setup.
2999 * Return value of StartChildProcess is subprocess' PID, or 0 if failed
3000 * to start subprocess.
3003 StartChildProcess(int xlop)
3009 #ifdef LINUX_PROFILE
3010 struct itimerval prof_itimer;
3014 * Set up command-line arguments for subprocess
3016 av[ac++] = "postgres";
3019 av[ac++] = "-forkboot";
3020 av[ac++] = NULL; /* filled in by postmaster_forkexec */
3023 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
3027 av[ac++] = "template1";
3030 Assert(ac < lengthof(av));
3033 * Flush stdio channels (see comments in BackendStartup)
3040 pid = postmaster_forkexec(ac, av);
3042 #else /* !EXEC_BACKEND */
3044 #ifdef LINUX_PROFILE
3045 /* see comments in BackendStartup */
3046 getitimer(ITIMER_PROF, &prof_itimer);
3050 /* Specific beos actions before backend startup */
3051 beos_before_backend_startup();
3056 if (pid == 0) /* child */
3058 #ifdef LINUX_PROFILE
3059 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3063 /* Specific beos actions after backend startup */
3064 beos_backend_startup();
3067 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
3069 /* Close the postmaster's sockets */
3070 ClosePostmasterPorts();
3072 /* Lose the postmaster's on-exit routines and port connections */
3075 BootstrapMain(ac, av);
3079 #endif /* EXEC_BACKEND */
3083 /* in parent, fork failed */
3084 int save_errno = errno;
3087 /* Specific beos actions before backend startup */
3088 beos_backend_startup_failed();
3093 case BS_XLOG_STARTUP:
3095 (errmsg("could not fork startup process: %m")));
3097 case BS_XLOG_BGWRITER:
3099 (errmsg("could not fork background writer process: %m")));
3103 (errmsg("could not fork process: %m")));
3108 * fork failure is fatal during startup, but there's no need
3109 * to choke immediately if starting other child types fails.
3111 if (xlop == BS_XLOG_STARTUP)
3117 * in parent, successful fork
3124 * Create the opts file
3127 CreateOptsFile(int argc, char *argv[], char *fullprogname)
3129 char filename[MAXPGPATH];
3133 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3135 if ((fp = fopen(filename, "w")) == NULL)
3137 elog(LOG, "could not create file \"%s\": %m", filename);
3141 fprintf(fp, "%s", fullprogname);
3142 for (i = 1; i < argc; i++)
3143 fprintf(fp, " '%s'", argv[i]);
3148 elog(LOG, "could not write file \"%s\": %m", filename);
3159 * The following need to be available to the read/write_backend_variables
3162 #include "storage/spin.h"
3164 extern slock_t *ShmemLock;
3165 extern slock_t *ShmemIndexLock;
3166 extern void *ShmemIndexAlloc;
3167 typedef struct LWLock LWLock;
3168 extern LWLock *LWLockArray;
3169 extern slock_t *ProcStructLock;
3170 extern int pgStatSock;
3172 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3173 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3174 #define write_array_var(var,fp) fwrite((void*)(var),sizeof(var),1,fp)
3175 #define read_array_var(var,fp) fread((void*)(var),sizeof(var),1,fp)
3178 write_backend_variables(char *filename, Port *port)
3180 static unsigned long tmpBackendFileNum = 0;
3182 char str_buf[MAXPGPATH];
3184 /* Calculate name for temp file in caller's buffer */
3186 snprintf(filename, MAXPGPATH, "%s/%s/%s.backend_var.%d.%lu",
3187 DataDir, PG_TEMP_FILES_DIR, PG_TEMP_FILE_PREFIX,
3188 MyProcPid, ++tmpBackendFileNum);
3191 fp = AllocateFile(filename, PG_BINARY_W);
3194 /* As per OpenTemporaryFile... */
3195 char dirname[MAXPGPATH];
3197 snprintf(dirname, MAXPGPATH, "%s/%s", DataDir, PG_TEMP_FILES_DIR);
3198 mkdir(dirname, S_IRWXU);
3200 fp = AllocateFile(filename, PG_BINARY_W);
3204 (errcode_for_file_access(),
3205 errmsg("could not create file \"%s\": %m",
3212 write_var(port->sock, fp);
3213 write_var(port->proto, fp);
3214 write_var(port->laddr, fp);
3215 write_var(port->raddr, fp);
3216 write_var(port->canAcceptConnections, fp);
3217 write_var(port->cryptSalt, fp);
3218 write_var(port->md5Salt, fp);
3221 * XXX FIXME later: writing these strings as MAXPGPATH bytes always is
3222 * probably a waste of resources
3225 StrNCpy(str_buf, DataDir, MAXPGPATH);
3226 write_array_var(str_buf, fp);
3228 write_array_var(ListenSocket, fp);
3230 write_var(MyCancelKey, fp);
3232 write_var(UsedShmemSegID, fp);
3233 write_var(UsedShmemSegAddr, fp);
3235 write_var(ShmemLock, fp);
3236 write_var(ShmemIndexLock, fp);
3237 write_var(ShmemVariableCache, fp);
3238 write_var(ShmemIndexAlloc, fp);
3239 write_var(ShmemBackendArray, fp);
3241 write_var(LWLockArray, fp);
3242 write_var(ProcStructLock, fp);
3243 write_var(pgStatSock, fp);
3245 write_var(debug_flag, fp);
3246 write_var(PostmasterPid, fp);
3248 write_var(PostmasterHandle, fp);
3251 StrNCpy(str_buf, my_exec_path, MAXPGPATH);
3252 write_array_var(str_buf, fp);
3254 write_array_var(ExtraOptions, fp);
3256 StrNCpy(str_buf, setlocale(LC_COLLATE, NULL), MAXPGPATH);
3257 write_array_var(str_buf, fp);
3258 StrNCpy(str_buf, setlocale(LC_CTYPE, NULL), MAXPGPATH);
3259 write_array_var(str_buf, fp);
3265 (errcode_for_file_access(),
3266 errmsg("could not write to file \"%s\": %m", filename)));
3274 read_backend_variables(char *filename, Port *port)
3277 char str_buf[MAXPGPATH];
3280 fp = AllocateFile(filename, PG_BINARY_R);
3283 (errcode_for_file_access(),
3284 errmsg("could not read from backend variables file \"%s\": %m",
3288 read_var(port->sock, fp);
3289 read_var(port->proto, fp);
3290 read_var(port->laddr, fp);
3291 read_var(port->raddr, fp);
3292 read_var(port->canAcceptConnections, fp);
3293 read_var(port->cryptSalt, fp);
3294 read_var(port->md5Salt, fp);
3296 read_array_var(str_buf, fp);
3297 SetDataDir(str_buf);
3299 read_array_var(ListenSocket, fp);
3301 read_var(MyCancelKey, fp);
3303 read_var(UsedShmemSegID, fp);
3304 read_var(UsedShmemSegAddr, fp);
3306 read_var(ShmemLock, fp);
3307 read_var(ShmemIndexLock, fp);
3308 read_var(ShmemVariableCache, fp);
3309 read_var(ShmemIndexAlloc, fp);
3310 read_var(ShmemBackendArray, fp);
3312 read_var(LWLockArray, fp);
3313 read_var(ProcStructLock, fp);
3314 read_var(pgStatSock, fp);
3316 read_var(debug_flag, fp);
3317 read_var(PostmasterPid, fp);
3319 read_var(PostmasterHandle, fp);
3322 read_array_var(str_buf, fp);
3323 StrNCpy(my_exec_path, str_buf, MAXPGPATH);
3325 read_array_var(ExtraOptions, fp);
3327 read_array_var(str_buf, fp);
3328 setlocale(LC_COLLATE, str_buf);
3329 read_array_var(str_buf, fp);
3330 setlocale(LC_CTYPE, str_buf);
3334 if (unlink(filename) != 0)
3336 (errcode_for_file_access(),
3337 errmsg("could not remove file \"%s\": %m", filename)));
3342 ShmemBackendArraySize(void)
3344 return (NUM_BACKENDARRAY_ELEMS * sizeof(Backend));
3348 ShmemBackendArrayAllocation(void)
3350 size_t size = ShmemBackendArraySize();
3352 ShmemBackendArray = (Backend *) ShmemAlloc(size);
3353 /* Mark all slots as empty */
3354 memset(ShmemBackendArray, 0, size);
3358 ShmemBackendArrayAdd(Backend *bn)
3362 /* Find an empty slot */
3363 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3365 if (ShmemBackendArray[i].pid == 0)
3367 ShmemBackendArray[i] = *bn;
3373 (errmsg_internal("no free slots in shmem backend array")));
3377 ShmemBackendArrayRemove(pid_t pid)
3381 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3383 if (ShmemBackendArray[i].pid == pid)
3385 /* Mark the slot as empty */
3386 ShmemBackendArray[i].pid = 0;
3392 (errmsg_internal("could not find backend entry with pid %d",
3396 #endif /* EXEC_BACKEND */
3402 win32_forkexec(const char *path, char *argv[])
3405 PROCESS_INFORMATION pi;
3408 char cmdLine[MAXPGPATH * 2];
3409 HANDLE childHandleCopy;
3410 HANDLE waiterThread;
3412 /* Format the cmd line */
3413 cmdLine[sizeof(cmdLine)-1] = '\0';
3414 cmdLine[sizeof(cmdLine)-2] = '\0';
3415 snprintf(cmdLine, sizeof(cmdLine)-1, "\"%s\"", path);
3417 while (argv[++i] != NULL)
3419 j = strlen(cmdLine);
3420 snprintf(cmdLine+j, sizeof(cmdLine)-1-j, " \"%s\"", argv[i]);
3422 if (cmdLine[sizeof(cmdLine)-2] != '\0')
3424 elog(LOG, "subprocess command line too long");
3428 memset(&pi, 0, sizeof(pi));
3429 memset(&si, 0, sizeof(si));
3431 if (!CreateProcess(NULL, cmdLine, NULL, NULL, TRUE, 0, NULL, NULL, &si, &pi))
3433 elog(LOG, "CreateProcess call failed (%d): %m", (int) GetLastError());
3437 if (!IsUnderPostmaster)
3439 /* We are the Postmaster creating a child... */
3440 win32_AddChild(pi.dwProcessId, pi.hProcess);
3443 if (DuplicateHandle(GetCurrentProcess(),
3445 GetCurrentProcess(),
3449 DUPLICATE_SAME_ACCESS) == 0)
3451 (errmsg_internal("could not duplicate child handle: %d",
3452 (int) GetLastError())));
3454 waiterThread = CreateThread(NULL, 64 * 1024, win32_sigchld_waiter,
3455 (LPVOID) childHandleCopy, 0, NULL);
3458 (errmsg_internal("could not create sigchld waiter thread: %d",
3459 (int) GetLastError())));
3460 CloseHandle(waiterThread);
3462 if (IsUnderPostmaster)
3463 CloseHandle(pi.hProcess);
3464 CloseHandle(pi.hThread);
3466 return pi.dwProcessId;
3470 * Note: The following three functions must not be interrupted (eg. by
3471 * signals). As the Postgres Win32 signalling architecture (currently)
3472 * requires polling, or APC checking functions which aren't used here, this
3475 * We keep two separate arrays, instead of a single array of pid/HANDLE
3476 * structs, to avoid having to re-create a handle array for
3477 * WaitForMultipleObjects on each call to win32_waitpid.
3481 win32_AddChild(pid_t pid, HANDLE handle)
3483 Assert(win32_childPIDArray && win32_childHNDArray);
3484 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3486 win32_childPIDArray[win32_numChildren] = pid;
3487 win32_childHNDArray[win32_numChildren] = handle;
3488 ++win32_numChildren;
3492 (errmsg_internal("no room for child entry with pid %lu",
3493 (unsigned long) pid)));
3497 win32_RemoveChild(pid_t pid)
3501 Assert(win32_childPIDArray && win32_childHNDArray);
3503 for (i = 0; i < win32_numChildren; i++)
3505 if (win32_childPIDArray[i] == pid)
3507 CloseHandle(win32_childHNDArray[i]);
3509 /* Swap last entry into the "removed" one */
3510 --win32_numChildren;
3511 win32_childPIDArray[i] = win32_childPIDArray[win32_numChildren];
3512 win32_childHNDArray[i] = win32_childHNDArray[win32_numChildren];
3518 (errmsg_internal("could not find child entry with pid %lu",
3519 (unsigned long) pid)));
3523 win32_waitpid(int *exitstatus)
3525 Assert(win32_childPIDArray && win32_childHNDArray);
3526 elog(DEBUG3, "waiting on %lu children", win32_numChildren);
3528 if (win32_numChildren > 0)
3531 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't want to
3532 * run queued APCs here.
3538 ret = WaitForMultipleObjects(win32_numChildren, win32_childHNDArray,
3544 (errmsg_internal("failed to wait on %lu children: %d",
3545 win32_numChildren, (int) GetLastError())));
3549 /* No children have finished */
3555 * Get the exit code, and return the PID of, the
3556 * respective process
3558 index = ret - WAIT_OBJECT_0;
3559 Assert(index >= 0 && index < win32_numChildren);
3560 if (!GetExitCodeProcess(win32_childHNDArray[index], &exitCode))
3563 * If we get this far, this should never happen, but,
3564 * then again... No choice other than to assume a
3565 * catastrophic failure.
3568 (errmsg_internal("failed to get exit code for child %lu",
3569 win32_childPIDArray[index])));
3571 *exitstatus = (int) exitCode;
3572 return win32_childPIDArray[index];
3581 * Note! Code below executes on separate threads, one for
3582 * each child process created
3585 win32_sigchld_waiter(LPVOID param)
3587 HANDLE procHandle = (HANDLE) param;
3589 DWORD r = WaitForSingleObject(procHandle, INFINITE);
3591 if (r == WAIT_OBJECT_0)
3592 pg_queue_signal(SIGCHLD);
3594 write_stderr("ERROR: failed to wait on child process handle: %d\n",
3595 (int) GetLastError());
3596 CloseHandle(procHandle);