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 1.400 2004/05/29 22:48:19 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.
58 *-------------------------------------------------------------------------
68 #include <sys/socket.h>
71 #include <sys/param.h>
72 #include <netinet/in.h>
73 #include <arpa/inet.h>
77 #ifdef HAVE_SYS_SELECT_H
78 #include <sys/select.h>
86 #include <DNSServiceDiscovery/DNSServiceDiscovery.h>
89 #include "catalog/pg_database.h"
90 #include "commands/async.h"
91 #include "lib/dllist.h"
92 #include "libpq/auth.h"
93 #include "libpq/crypt.h"
94 #include "libpq/libpq.h"
95 #include "libpq/pqcomm.h"
96 #include "libpq/pqsignal.h"
97 #include "miscadmin.h"
98 #include "nodes/nodes.h"
99 #include "postmaster/postmaster.h"
101 #include "storage/fd.h"
102 #include "storage/ipc.h"
103 #include "storage/pg_shmem.h"
104 #include "storage/pmsignal.h"
105 #include "storage/proc.h"
106 #include "storage/bufmgr.h"
107 #include "access/xlog.h"
108 #include "tcop/tcopprot.h"
109 #include "utils/guc.h"
110 #include "utils/memutils.h"
111 #include "utils/ps_status.h"
112 #include "bootstrap/bootstrap.h"
117 * List of active backends (or child processes anyway; we don't actually
118 * know whether a given child has become a backend or is still in the
119 * authorization phase). This is used mainly to keep track of how many
120 * children we have and send them appropriate signals when necessary.
122 * "Special" children such as the startup and bgwriter tasks are not in
127 pid_t pid; /* process id of backend */
128 long cancel_key; /* cancel key for cancels for this backend */
131 static Dllist *BackendList;
134 #define NUM_BACKENDARRAY_ELEMS (2*MaxBackends)
135 static Backend *ShmemBackendArray;
138 /* The socket number we are listening for connections on */
141 char *ListenAddresses;
144 * ReservedBackends is the number of backends reserved for superuser use.
145 * This number is taken out of the pool size given by MaxBackends so
146 * number of backend slots available to non-superusers is
147 * (MaxBackends - ReservedBackends). Note what this really means is
148 * "if there are <= ReservedBackends connections available, only superusers
149 * can make new connections" --- pre-existing superuser connections don't
150 * count against the limit.
152 int ReservedBackends;
155 static const char *progname = NULL;
157 /* The socket(s) we're listening to. */
159 static int ListenSocket[MAXLISTEN];
162 * Set by the -o option
164 static char ExtraOptions[MAXPGPATH];
167 * These globals control the behavior of the postmaster in case some
168 * backend dumps core. Normally, it kills all peers of the dead backend
169 * and reinitializes shared memory. By specifying -s or -n, we can have
170 * the postmaster stop (rather than kill) peers and not reinitialize
171 * shared data structures.
173 static bool Reinit = true;
174 static int SendStop = false;
176 /* still more option variables */
177 bool EnableSSL = false;
178 bool SilentMode = false; /* silent mode (-S) */
180 int PreAuthDelay = 0;
181 int AuthenticationTimeout = 60;
183 bool log_hostname; /* for ps display and logging */
184 bool Log_connections = false;
185 bool Db_user_namespace = false;
187 char *rendezvous_name;
189 /* list of library:init-function to be preloaded */
190 char *preload_libraries_string = NULL;
192 /* PIDs of special child processes; 0 when not running */
193 static pid_t StartupPID = 0,
196 /* Startup/shutdown state */
198 #define SmartShutdown 1
199 #define FastShutdown 2
201 static int Shutdown = NoShutdown;
203 static bool FatalError = false; /* T if recovering from backend crash */
205 bool ClientAuthInProgress = false; /* T during new-client
209 * State for assigning random salts and cancel keys.
210 * Also, the global MyCancelKey passes the cancel key assigned to a given
211 * backend from the postmaster to that backend (via fork).
213 static unsigned int random_seed = 0;
215 static int debug_flag = 0;
221 #ifdef HAVE_INT_OPTRESET
226 * postmaster.c - function prototypes
228 static void checkDataDir(const char *checkdir);
229 #ifdef USE_RENDEZVOUS
230 static void reg_reply(DNSServiceRegistrationReplyErrorType errorCode,
233 static void pmdaemonize(void);
234 static Port *ConnCreate(int serverFd);
235 static void ConnFree(Port *port);
236 static void reset_shared(unsigned short port);
237 static void SIGHUP_handler(SIGNAL_ARGS);
238 static void pmdie(SIGNAL_ARGS);
239 static void reaper(SIGNAL_ARGS);
240 static void sigusr1_handler(SIGNAL_ARGS);
241 static void dummy_handler(SIGNAL_ARGS);
242 static void CleanupProc(int pid, int exitstatus);
243 static void HandleChildCrash(int pid, int exitstatus);
244 static void LogChildExit(int lev, const char *procname,
245 int pid, int exitstatus);
246 static int BackendRun(Port *port);
247 static void ExitPostmaster(int status);
248 static void usage(const char *);
249 static int ServerLoop(void);
250 static int BackendStartup(Port *port);
251 static int ProcessStartupPacket(Port *port, bool SSLdone);
252 static void processCancelRequest(Port *port, void *pkt);
253 static int initMasks(fd_set *rmask);
254 static void report_fork_failure_to_client(Port *port, int errnum);
255 static enum CAC_state canAcceptConnections(void);
256 static long PostmasterRandom(void);
257 static void RandomSalt(char *cryptSalt, char *md5Salt);
258 static void SignalChildren(int signal);
259 static int CountChildren(void);
260 static bool CreateOptsFile(int argc, char *argv[], char *fullprogname);
261 static pid_t StartChildProcess(int xlop);
263 postmaster_error(const char *fmt,...)
264 /* This lets gcc check the format string for consistency. */
265 __attribute__((format(printf, 1, 2)));
270 static pid_t win32_forkexec(const char *path, char *argv[]);
271 static void win32_AddChild(pid_t pid, HANDLE handle);
272 static void win32_RemoveChild(pid_t pid);
273 static pid_t win32_waitpid(int *exitstatus);
274 static DWORD WINAPI win32_sigchld_waiter(LPVOID param);
276 static pid_t *win32_childPIDArray;
277 static HANDLE *win32_childHNDArray;
278 static unsigned long win32_numChildren = 0;
281 static pid_t backend_forkexec(Port *port);
282 static pid_t internal_forkexec(int argc, char *argv[], Port *port);
284 static void read_backend_variables(char *filename, Port *port);
285 static bool write_backend_variables(char *filename, Port *port);
287 static void ShmemBackendArrayAdd(Backend *bn);
288 static void ShmemBackendArrayRemove(pid_t pid);
290 #endif /* EXEC_BACKEND */
292 #define StartupDataBase() StartChildProcess(BS_XLOG_STARTUP)
293 #define StartBackgroundWriter() StartChildProcess(BS_XLOG_BGWRITER)
297 * Postmaster main entry point
300 PostmasterMain(int argc, char *argv[])
304 char *potential_DataDir = NULL;
307 progname = get_progname(argv[0]);
309 MyProcPid = PostmasterPid = getpid();
311 IsPostmasterEnvironment = true;
314 * Catch standard options before doing much else. This even works on
315 * systems without getopt_long.
319 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
324 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
326 puts("postmaster (PostgreSQL) " PG_VERSION);
332 * for security, no dir or file created can be group or other
335 umask((mode_t) 0077);
338 * Fire up essential subsystems: memory management
343 * By default, palloc() requests in the postmaster will be allocated
344 * in the PostmasterContext, which is space that can be recycled by
345 * backends. Allocated data that needs to be available to backends
346 * should be allocated in TopMemoryContext.
348 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
350 ALLOCSET_DEFAULT_MINSIZE,
351 ALLOCSET_DEFAULT_INITSIZE,
352 ALLOCSET_DEFAULT_MAXSIZE);
353 MemoryContextSwitchTo(PostmasterContext);
355 IgnoreSystemIndexes(false);
357 if (find_my_exec(argv[0], my_exec_path) < 0)
358 elog(FATAL, "%s: could not locate my own executable path",
361 get_pkglib_path(my_exec_path, pkglib_path);
366 InitializeGUCOptions();
368 potential_DataDir = getenv("PGDATA"); /* default value */
372 while ((opt = getopt(argc, argv, "A:a:B:b:c:D:d:Fh:ik:lm:MN:no:p:Ss-:")) != -1)
377 #ifdef USE_ASSERT_CHECKING
378 SetConfigOption("debug_assertions", optarg, PGC_POSTMASTER, PGC_S_ARGV);
380 postmaster_error("assert checking is not compiled in");
384 /* Can no longer set authentication method. */
387 SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
390 /* Can no longer set the backend executable file to use. */
393 potential_DataDir = optarg;
397 /* Turn on debugging for the postmaster. */
398 char *debugstr = palloc(strlen("debug") + strlen(optarg) + 1);
400 sprintf(debugstr, "debug%s", optarg);
401 SetConfigOption("log_min_messages", debugstr,
402 PGC_POSTMASTER, PGC_S_ARGV);
404 debug_flag = atoi(optarg);
408 SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
411 SetConfigOption("listen_addresses", optarg, PGC_POSTMASTER, PGC_S_ARGV);
414 SetConfigOption("listen_addresses", "*", PGC_POSTMASTER, PGC_S_ARGV);
417 SetConfigOption("unix_socket_directory", optarg, PGC_POSTMASTER, PGC_S_ARGV);
421 SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
425 /* Multiplexed backends no longer supported. */
430 * ignore this flag. This may be passed in because the
431 * program was run as 'postgres -M' instead of
436 /* The max number of backends to start. */
437 SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
440 /* Don't reinit shared mem after abnormal exit */
446 * Other options to pass to the backend on the command line
448 strcat(ExtraOptions, " ");
449 strcat(ExtraOptions, optarg);
452 SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
457 * Start in 'S'ilent mode (disassociate from controlling
458 * tty). You may also think of this as 'S'ysV mode since
459 * it's most badly needed on SysV-derived systems like
462 SetConfigOption("silent_mode", "true", PGC_POSTMASTER, PGC_S_ARGV);
467 * In the event that some backend dumps core, send
468 * SIGSTOP, rather than SIGQUIT, to all its peers. This
469 * lets the wily post_hacker collect core dumps from
480 ParseLongOption(optarg, &name, &value);
485 (errcode(ERRCODE_SYNTAX_ERROR),
486 errmsg("--%s requires a value",
490 (errcode(ERRCODE_SYNTAX_ERROR),
491 errmsg("-c %s requires a value",
495 SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
504 gettext("Try \"%s --help\" for more information.\n"),
511 * Postmaster accepts no non-option switch arguments.
515 postmaster_error("invalid argument: \"%s\"", argv[optind]);
517 gettext("Try \"%s --help\" for more information.\n"),
523 * Now we can set the data directory, and then read postgresql.conf.
525 checkDataDir(potential_DataDir); /* issues error messages */
526 SetDataDir(potential_DataDir);
528 ProcessConfigFile(PGC_POSTMASTER);
530 /* If timezone is not set, determine what the OS uses */
531 pg_timezone_initialize();
534 write_nondefault_variables(PGC_POSTMASTER);
538 * Check for invalid combinations of GUC settings.
540 if (NBuffers < 2 * MaxBackends || NBuffers < 16)
543 * Do not accept -B so small that backends are likely to starve
544 * for lack of buffers. The specific choices here are somewhat
547 postmaster_error("the number of buffers (-B) must be at least twice the number of allowed connections (-N) and at least 16");
551 if (ReservedBackends >= MaxBackends)
553 postmaster_error("superuser_reserved_connections must be less than max_connections");
558 * Other one-time internal sanity checks can go here.
560 if (!CheckDateTokenTables())
562 postmaster_error("invalid datetoken tables, please fix");
567 * Now that we are done processing the postmaster arguments, reset
568 * getopt(3) library so that it will work correctly in subprocesses.
571 #ifdef HAVE_INT_OPTRESET
572 optreset = 1; /* some systems need this too */
575 /* For debugging: display postmaster environment */
577 extern char **environ;
581 (errmsg_internal("%s: PostmasterMain: initial environ dump:",
584 (errmsg_internal("-----------------------------------------")));
585 for (p = environ; *p; ++p)
587 (errmsg_internal("\t%s", *p)));
589 (errmsg_internal("-----------------------------------------")));
593 if (find_other_exec(argv[0], "postgres", PG_VERSIONSTR,
594 postgres_exec_path) < 0)
596 (errmsg("%s: could not locate matching postgres executable",
601 * Initialize SSL library, if specified.
609 * process any libraries that should be preloaded and optionally
612 if (preload_libraries_string)
613 process_preload_libraries(preload_libraries_string);
616 * Fork away from controlling terminal, if -S specified.
618 * Must do this before we grab any interlock files, else the interlocks
619 * will show the wrong PID.
625 * Create lockfile for data directory.
627 * We want to do this before we try to grab the input sockets, because
628 * the data directory interlock is more reliable than the socket-file
629 * interlock (thanks to whoever decided to put socket files in /tmp
630 * :-(). For the same reason, it's best to grab the TCP socket(s) before
633 CreateDataDirLockFile(DataDir, true);
636 * Remove old temporary files. At this point there can be no other
637 * Postgres processes running in this directory, so this should be
643 * Establish input sockets.
645 for (i = 0; i < MAXLISTEN; i++)
646 ListenSocket[i] = -1;
654 curhost = ListenAddresses;
657 /* ignore whitespace */
658 while (isspace((unsigned char) *curhost))
660 if (*curhost == '\0')
663 while (*endptr != '\0' && !isspace((unsigned char) *endptr))
667 if (strcmp(curhost, "*") == 0)
668 status = StreamServerPort(AF_UNSPEC, NULL,
669 (unsigned short) PostPortNumber,
671 ListenSocket, MAXLISTEN);
673 status = StreamServerPort(AF_UNSPEC, curhost,
674 (unsigned short) PostPortNumber,
676 ListenSocket, MAXLISTEN);
677 if (status != STATUS_OK)
679 (errmsg("could not create listen socket for \"%s\"",
683 curhost = endptr + 1;
689 #ifdef USE_RENDEZVOUS
690 /* Register for Rendezvous only if we opened TCP socket(s) */
691 if (ListenSocket[0] != -1 && rendezvous_name != NULL)
693 DNSServiceRegistrationCreate(rendezvous_name,
696 htonl(PostPortNumber),
698 (DNSServiceRegistrationReply) reg_reply,
703 #ifdef HAVE_UNIX_SOCKETS
704 status = StreamServerPort(AF_UNIX, NULL,
705 (unsigned short) PostPortNumber,
707 ListenSocket, MAXLISTEN);
708 if (status != STATUS_OK)
710 (errmsg("could not create Unix-domain socket")));
714 * check that we have some socket to listen on
716 if (ListenSocket[0] == -1)
718 (errmsg("no socket created for listening")));
723 * Set up shared memory and semaphores.
725 reset_shared(PostPortNumber);
728 * Estimate number of openable files. This must happen after setting
729 * up semaphores, because on some platforms semaphores count as open
735 * Initialize the list of active backends.
737 BackendList = DLNewList();
741 * Initialize the child pid/HANDLE arrays for signal handling.
743 win32_childPIDArray = (pid_t *)
744 malloc(NUM_BACKENDARRAY_ELEMS * sizeof(pid_t));
745 win32_childHNDArray = (HANDLE *)
746 malloc(NUM_BACKENDARRAY_ELEMS * sizeof(HANDLE));
747 if (!win32_childPIDArray || !win32_childHNDArray)
749 (errcode(ERRCODE_OUT_OF_MEMORY),
750 errmsg("out of memory")));
754 * Record postmaster options. We delay this till now to avoid
755 * recording bogus options (eg, NBuffers too high for available
758 if (!CreateOptsFile(argc, argv, my_exec_path))
762 * Set up signal handlers for the postmaster process.
764 * CAUTION: when changing this list, check for side-effects on the signal
765 * handling setup of child processes. See tcop/postgres.c,
766 * bootstrap/bootstrap.c, postmaster/bgwriter.c, and postmaster/pgstat.c.
769 PG_SETMASK(&BlockSig);
771 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
772 * children do same */
773 pqsignal(SIGINT, pmdie); /* send SIGTERM and shut down */
774 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
775 pqsignal(SIGTERM, pmdie); /* wait for children and shut down */
776 pqsignal(SIGALRM, SIG_IGN); /* ignored */
777 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
778 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
779 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
780 pqsignal(SIGCHLD, reaper); /* handle child termination */
781 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
782 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
783 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
785 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
789 * Reset whereToSendOutput from Debug (its starting state) to None.
790 * This prevents ereport from sending log messages to stderr unless
791 * the syslog/stderr switch permits. We don't do this until the
792 * postmaster is fully launched, since startup failures may as well be
793 * reported to stderr.
795 whereToSendOutput = None;
798 * Initialize and try to startup the statistics collector process
804 * Load cached files for client authentication.
812 * We're ready to rock and roll...
814 StartupPID = StartupDataBase();
816 status = ServerLoop();
819 * ServerLoop probably shouldn't ever return, but if it does, close
822 ExitPostmaster(status != STATUS_OK);
824 return 0; /* not reached */
829 * Validate the proposed data directory
832 checkDataDir(const char *checkdir)
834 char path[MAXPGPATH];
836 struct stat stat_buf;
838 if (checkdir == NULL)
841 gettext("%s does not know where to find the database system data.\n"
842 "You must specify the directory that contains the database system\n"
843 "either by specifying the -D invocation option or by setting the\n"
844 "PGDATA environment variable.\n"),
849 if (stat(checkdir, &stat_buf) == -1)
853 (errcode_for_file_access(),
854 errmsg("data directory \"%s\" does not exist",
858 (errcode_for_file_access(),
859 errmsg("could not read permissions of directory \"%s\": %m",
864 * Check if the directory has group or world access. If so, reject.
866 * XXX temporarily suppress check when on Windows, because there may not
867 * be proper support for Unix-y file permissions. Need to think of a
868 * reasonable check to apply on Windows.
870 #if !defined(__CYGWIN__) && !defined(WIN32)
871 if (stat_buf.st_mode & (S_IRWXG | S_IRWXO))
873 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
874 errmsg("data directory \"%s\" has group or world access",
876 errdetail("Permissions should be u=rwx (0700).")));
879 /* Look for PG_VERSION before looking for pg_control */
880 ValidatePgVersion(checkdir);
882 snprintf(path, sizeof(path), "%s/global/pg_control", checkdir);
884 fp = AllocateFile(path, PG_BINARY_R);
888 gettext("%s: could not find the database system\n"
889 "Expected to find it in the directory \"%s\",\n"
890 "but could not open file \"%s\": %s\n"),
891 progname, checkdir, path, strerror(errno));
898 #ifdef USE_RENDEZVOUS
901 * empty callback function for DNSServiceRegistrationCreate()
904 reg_reply(DNSServiceRegistrationReplyErrorType errorCode, void *context)
909 #endif /* USE_RENDEZVOUS */
913 * Fork away from the controlling terminal (-S option)
923 struct itimerval prof_itimer;
927 /* see comments in BackendStartup */
928 getitimer(ITIMER_PROF, &prof_itimer);
932 if (pid == (pid_t) -1)
934 postmaster_error("could not fork background process: %s",
940 /* Parent should just exit, without doing any atexit cleanup */
945 setitimer(ITIMER_PROF, &prof_itimer, NULL);
948 MyProcPid = PostmasterPid = getpid(); /* reset PID vars to child */
950 /* GH: If there's no setsid(), we hopefully don't need silent mode.
951 * Until there's a better solution.
956 postmaster_error("could not dissociate from controlling TTY: %s",
961 i = open(NULL_DEV, O_RDWR | PG_BINARY);
968 elog(FATAL, "SilentMode not supported under WIN32");
974 * Print out help message
977 usage(const char *progname)
979 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
980 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
981 printf(gettext("Options:\n"));
982 #ifdef USE_ASSERT_CHECKING
983 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
985 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
986 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
987 printf(gettext(" -d 1-5 debugging level\n"));
988 printf(gettext(" -D DATADIR database directory\n"));
989 printf(gettext(" -F turn fsync off\n"));
990 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
991 printf(gettext(" -i enable TCP/IP connections\n"));
992 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
994 printf(gettext(" -l enable SSL connections\n"));
996 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
997 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
998 printf(gettext(" -p PORT port number to listen on\n"));
999 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1000 printf(gettext(" --help show this help, then exit\n"));
1001 printf(gettext(" --version output version information, then exit\n"));
1003 printf(gettext("\nDeveloper options:\n"));
1004 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1005 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1007 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1008 "configuration settings and how to set them on the command line or in\n"
1009 "the configuration file.\n\n"
1010 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1015 * Main idle loop of postmaster
1024 struct timeval earlier,
1028 gettimeofday(&earlier, &tz);
1029 last_touch_time = time(NULL);
1031 nSockets = initMasks(&readmask);
1037 struct timeval timeout;
1042 * Wait for something to happen.
1044 * We wait at most one minute, to ensure that the other background
1045 * tasks handled below get done even when no requests are arriving.
1047 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1049 timeout.tv_sec = 60;
1050 timeout.tv_usec = 0;
1052 PG_SETMASK(&UnBlockSig);
1054 selres = select(nSockets, &rmask, NULL, NULL, &timeout);
1057 * Block all signals until we wait again. (This makes it safe for
1058 * our signal handlers to do nontrivial work.)
1060 PG_SETMASK(&BlockSig);
1064 if (errno == EINTR || errno == EWOULDBLOCK)
1067 (errcode_for_socket_access(),
1068 errmsg("select() failed in postmaster: %m")));
1069 return STATUS_ERROR;
1073 * New connection pending on any of our sockets? If so, fork a
1074 * child process to deal with it.
1079 * Select a random seed at the time of first receiving a request.
1081 while (random_seed == 0)
1083 gettimeofday(&later, &tz);
1086 * We are not sure how much precision is in tv_usec, so we
1087 * swap the nibbles of 'later' and XOR them with 'earlier'. On
1088 * the off chance that the result is 0, we loop until it isn't.
1090 random_seed = earlier.tv_usec ^
1091 ((later.tv_usec << 16) |
1092 ((later.tv_usec >> 16) & 0xffff));
1095 for (i = 0; i < MAXLISTEN; i++)
1097 if (ListenSocket[i] == -1)
1099 if (FD_ISSET(ListenSocket[i], &rmask))
1101 port = ConnCreate(ListenSocket[i]);
1104 BackendStartup(port);
1107 * We no longer need the open socket or port structure
1110 StreamClose(port->sock);
1118 * If no background writer process is running, and we are not in
1119 * a state that prevents it, start one. It doesn't matter if this
1120 * fails, we'll just try again later.
1122 if (BgWriterPID == 0 && StartupPID == 0 && !FatalError)
1124 BgWriterPID = StartBackgroundWriter();
1125 /* If shutdown is pending, set it going */
1126 if (Shutdown > NoShutdown && BgWriterPID != 0)
1127 kill(BgWriterPID, SIGUSR2);
1130 /* If we have lost the stats collector, try to start a new one */
1131 if (!pgstat_is_running)
1135 * Touch the socket and lock file at least every ten minutes, to ensure
1136 * that they are not removed by overzealous /tmp-cleaning tasks.
1139 if (now - last_touch_time >= 10 * 60)
1142 TouchSocketLockFile();
1143 last_touch_time = now;
1150 * Initialise the masks for select() for the ports we are listening on.
1151 * Return the number of sockets to listen on.
1154 initMasks(fd_set *rmask)
1161 for (i = 0; i < MAXLISTEN; i++)
1163 int fd = ListenSocket[i];
1177 * Read the startup packet and do something according to it.
1179 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1180 * not return at all.
1182 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1183 * if that's what you want. Return STATUS_ERROR if you don't want to
1184 * send anything to the client, which would typically be appropriate
1185 * if we detect a communications failure.)
1188 ProcessStartupPacket(Port *port, bool SSLdone)
1192 ProtocolVersion proto;
1193 MemoryContext oldcontext;
1195 if (pq_getbytes((char *) &len, 4) == EOF)
1198 * EOF after SSLdone probably means the client didn't like our
1199 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1200 * so don't clutter the log with a complaint.
1204 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1205 errmsg("incomplete startup packet")));
1206 return STATUS_ERROR;
1212 if (len < (int32) sizeof(ProtocolVersion) ||
1213 len > MAX_STARTUP_PACKET_LENGTH)
1216 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1217 errmsg("invalid length of startup packet")));
1218 return STATUS_ERROR;
1222 * Allocate at least the size of an old-style startup packet, plus one
1223 * extra byte, and make sure all are zeroes. This ensures we will
1224 * have null termination of all strings, in both fixed- and
1225 * variable-length packet layouts.
1227 if (len <= (int32) sizeof(StartupPacket))
1228 buf = palloc0(sizeof(StartupPacket) + 1);
1230 buf = palloc0(len + 1);
1232 if (pq_getbytes(buf, len) == EOF)
1235 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1236 errmsg("incomplete startup packet")));
1237 return STATUS_ERROR;
1241 * The first field is either a protocol version number or a special
1244 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1246 if (proto == CANCEL_REQUEST_CODE)
1248 processCancelRequest(port, buf);
1249 return 127; /* XXX */
1252 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1257 /* No SSL when disabled or on Unix sockets */
1258 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1261 SSLok = 'S'; /* Support for SSL */
1263 SSLok = 'N'; /* No support for SSL */
1265 if (send(port->sock, &SSLok, 1, 0) != 1)
1268 (errcode_for_socket_access(),
1269 errmsg("failed to send SSL negotiation response: %m")));
1270 return STATUS_ERROR; /* close the connection */
1274 if (SSLok == 'S' && secure_open_server(port) == -1)
1275 return STATUS_ERROR;
1277 /* regular startup packet, cancel, etc packet should follow... */
1278 /* but not another SSL negotiation request */
1279 return ProcessStartupPacket(port, true);
1282 /* Could add additional special packet types here */
1285 * Set FrontendProtocol now so that ereport() knows what format to
1286 * send if we fail during startup.
1288 FrontendProtocol = proto;
1290 /* Check we can handle the protocol the frontend is using. */
1292 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1293 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1294 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1295 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1297 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1298 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1299 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1300 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1301 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1302 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1305 * Now fetch parameters out of startup packet and save them into the
1306 * Port structure. All data structures attached to the Port struct
1307 * must be allocated in TopMemoryContext so that they won't disappear
1308 * when we pass them to PostgresMain (see BackendRun). We need not
1309 * worry about leaking this storage on failure, since we aren't in the
1310 * postmaster process anymore.
1312 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1314 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1316 int32 offset = sizeof(ProtocolVersion);
1319 * Scan packet body for name/option pairs. We can assume any
1320 * string beginning within the packet body is null-terminated,
1321 * thanks to zeroing extra byte above.
1323 port->guc_options = NIL;
1325 while (offset < len)
1327 char *nameptr = ((char *) buf) + offset;
1331 if (*nameptr == '\0')
1332 break; /* found packet terminator */
1333 valoffset = offset + strlen(nameptr) + 1;
1334 if (valoffset >= len)
1335 break; /* missing value, will complain below */
1336 valptr = ((char *) buf) + valoffset;
1338 if (strcmp(nameptr, "database") == 0)
1339 port->database_name = pstrdup(valptr);
1340 else if (strcmp(nameptr, "user") == 0)
1341 port->user_name = pstrdup(valptr);
1342 else if (strcmp(nameptr, "options") == 0)
1343 port->cmdline_options = pstrdup(valptr);
1346 /* Assume it's a generic GUC option */
1347 port->guc_options = lappend(port->guc_options,
1349 port->guc_options = lappend(port->guc_options,
1352 offset = valoffset + strlen(valptr) + 1;
1356 * If we didn't find a packet terminator exactly at the end of the
1357 * given packet length, complain.
1359 if (offset != len - 1)
1361 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1362 errmsg("invalid startup packet layout: expected terminator as last byte")));
1367 * Get the parameters from the old-style, fixed-width-fields
1368 * startup packet as C strings. The packet destination was
1369 * cleared first so a short packet has zeros silently added. We
1370 * have to be prepared to truncate the pstrdup result for oversize
1373 StartupPacket *packet = (StartupPacket *) buf;
1375 port->database_name = pstrdup(packet->database);
1376 if (strlen(port->database_name) > sizeof(packet->database))
1377 port->database_name[sizeof(packet->database)] = '\0';
1378 port->user_name = pstrdup(packet->user);
1379 if (strlen(port->user_name) > sizeof(packet->user))
1380 port->user_name[sizeof(packet->user)] = '\0';
1381 port->cmdline_options = pstrdup(packet->options);
1382 if (strlen(port->cmdline_options) > sizeof(packet->options))
1383 port->cmdline_options[sizeof(packet->options)] = '\0';
1384 port->guc_options = NIL;
1387 /* Check a user name was given. */
1388 if (port->user_name == NULL || port->user_name[0] == '\0')
1390 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1391 errmsg("no PostgreSQL user name specified in startup packet")));
1393 /* The database defaults to the user name. */
1394 if (port->database_name == NULL || port->database_name[0] == '\0')
1395 port->database_name = pstrdup(port->user_name);
1397 if (Db_user_namespace)
1400 * If user@, it is a global user, remove '@'. We only want to do
1401 * this if there is an '@' at the end and no earlier in the user
1402 * string or they may fake as a local user of another database
1403 * attaching to this database.
1405 if (strchr(port->user_name, '@') ==
1406 port->user_name + strlen(port->user_name) - 1)
1407 *strchr(port->user_name, '@') = '\0';
1410 /* Append '@' and dbname */
1413 db_user = palloc(strlen(port->user_name) +
1414 strlen(port->database_name) + 2);
1415 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1416 port->user_name = db_user;
1421 * Truncate given database and user names to length of a Postgres
1422 * name. This avoids lookup failures when overlength names are given.
1424 if (strlen(port->database_name) >= NAMEDATALEN)
1425 port->database_name[NAMEDATALEN - 1] = '\0';
1426 if (strlen(port->user_name) >= NAMEDATALEN)
1427 port->user_name[NAMEDATALEN - 1] = '\0';
1430 * Done putting stuff in TopMemoryContext.
1432 MemoryContextSwitchTo(oldcontext);
1435 * If we're going to reject the connection due to database state, say
1436 * so now instead of wasting cycles on an authentication exchange.
1437 * (This also allows a pg_ping utility to be written.)
1439 switch (port->canAcceptConnections)
1443 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1444 errmsg("the database system is starting up")));
1448 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1449 errmsg("the database system is shutting down")));
1453 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1454 errmsg("the database system is in recovery mode")));
1458 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1459 errmsg("sorry, too many clients already")));
1471 * The client has sent a cancel request packet, not a normal
1472 * start-a-new-connection packet. Perform the necessary processing.
1473 * Nothing is sent back to the client.
1476 processCancelRequest(Port *port, void *pkt)
1478 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1480 long cancelAuthCode;
1482 #ifndef EXEC_BACKEND
1488 backendPID = (int) ntohl(canc->backendPID);
1489 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1491 if (backendPID == BgWriterPID)
1494 (errmsg_internal("ignoring cancel request for bgwriter process %d",
1500 * See if we have a matching backend. In the EXEC_BACKEND case, we
1501 * can no longer access the postmaster's own backend list, and must
1502 * rely on the duplicate array in shared memory.
1504 #ifndef EXEC_BACKEND
1505 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1507 bp = (Backend *) DLE_VAL(curr);
1509 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1511 bp = (Backend *) &ShmemBackendArray[i];
1513 if (bp->pid == backendPID)
1515 if (bp->cancel_key == cancelAuthCode)
1517 /* Found a match; signal that backend to cancel current op */
1519 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1521 kill(bp->pid, SIGINT);
1524 /* Right PID, wrong key: no way, Jose */
1526 (errmsg_internal("bad key in cancel request for process %d",
1532 /* No matching backend */
1534 (errmsg_internal("bad pid in cancel request for process %d",
1539 * canAcceptConnections --- check to see if database state allows connections.
1541 static enum CAC_state
1542 canAcceptConnections(void)
1544 /* Can't start backends when in startup/shutdown/recovery state. */
1545 if (Shutdown > NoShutdown)
1546 return CAC_SHUTDOWN;
1550 return CAC_RECOVERY;
1553 * Don't start too many children.
1555 * We allow more connections than we can have backends here because some
1556 * might still be authenticating; they might fail auth, or some
1557 * existing backend might exit before the auth cycle is completed. The
1558 * exact MaxBackends limit is enforced when a new backend tries to
1559 * join the shared-inval backend array.
1561 if (CountChildren() >= 2 * MaxBackends)
1569 * ConnCreate -- create a local connection data structure
1572 ConnCreate(int serverFd)
1576 if (!(port = (Port *) calloc(1, sizeof(Port))))
1579 (errcode(ERRCODE_OUT_OF_MEMORY),
1580 errmsg("out of memory")));
1584 if (StreamConnection(serverFd, port) != STATUS_OK)
1586 StreamClose(port->sock);
1593 * Precompute password salt values to use for this connection.
1594 * It's slightly annoying to do this long in advance of knowing
1595 * whether we'll need 'em or not, but we must do the random()
1596 * calls before we fork, not after. Else the postmaster's random
1597 * sequence won't get advanced, and all backends would end up
1598 * using the same salt...
1600 RandomSalt(port->cryptSalt, port->md5Salt);
1608 * ConnFree -- free a local connection data structure
1611 ConnFree(Port *conn)
1621 * ClosePostmasterPorts -- close all the postmaster's open sockets
1623 * This is called during child process startup to release file descriptors
1624 * that are not needed by that child process. The postmaster still has
1625 * them open, of course.
1628 ClosePostmasterPorts(void)
1632 /* Close the listen sockets */
1633 for (i = 0; i < MAXLISTEN; i++)
1635 if (ListenSocket[i] != -1)
1637 StreamClose(ListenSocket[i]);
1638 ListenSocket[i] = -1;
1645 * reset_shared -- reset shared memory and semaphores
1648 reset_shared(unsigned short port)
1651 * Create or re-create shared memory and semaphores.
1653 * Note: in each "cycle of life" we will normally assign the same IPC
1654 * keys (if using SysV shmem and/or semas), since the port number is
1655 * used to determine IPC keys. This helps ensure that we will clean
1656 * up dead IPC objects if the postmaster crashes and is restarted.
1658 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1663 * SIGHUP -- reread config files, and tell children to do same
1666 SIGHUP_handler(SIGNAL_ARGS)
1668 int save_errno = errno;
1670 PG_SETMASK(&BlockSig);
1672 if (Shutdown <= SmartShutdown)
1675 (errmsg("received SIGHUP, reloading configuration files")));
1676 ProcessConfigFile(PGC_SIGHUP);
1677 SignalChildren(SIGHUP);
1678 if (BgWriterPID != 0)
1679 kill(BgWriterPID, SIGHUP);
1684 /* Update the starting-point file for future children */
1685 write_nondefault_variables(PGC_SIGHUP);
1689 PG_SETMASK(&UnBlockSig);
1696 * pmdie -- signal handler for processing various postmaster signals.
1701 int save_errno = errno;
1703 PG_SETMASK(&BlockSig);
1706 (errmsg_internal("postmaster received signal %d",
1707 postgres_signal_arg)));
1709 switch (postgres_signal_arg)
1715 * Wait for children to end their work, then shut down.
1717 if (Shutdown >= SmartShutdown)
1719 Shutdown = SmartShutdown;
1721 (errmsg("received smart shutdown request")));
1723 if (DLGetHead(BackendList))
1724 break; /* let reaper() handle this */
1727 * No children left. Begin shutdown of data base system.
1729 if (StartupPID != 0 || FatalError)
1730 break; /* let reaper() handle this */
1731 /* Start the bgwriter if not running */
1732 if (BgWriterPID == 0)
1733 BgWriterPID = StartBackgroundWriter();
1734 /* And tell it to shut down */
1735 if (BgWriterPID != 0)
1736 kill(BgWriterPID, SIGUSR2);
1743 * Abort all children with SIGTERM (rollback active transactions
1744 * and exit) and shut down when they are gone.
1746 if (Shutdown >= FastShutdown)
1748 Shutdown = FastShutdown;
1750 (errmsg("received fast shutdown request")));
1752 if (DLGetHead(BackendList))
1757 (errmsg("aborting any active transactions")));
1758 SignalChildren(SIGTERM);
1759 /* reaper() does the rest */
1765 * No children left. Begin shutdown of data base system.
1767 * Note: if we previously got SIGTERM then we may send SIGUSR2
1768 * to the bgwriter a second time here. This should be harmless.
1770 if (StartupPID != 0 || FatalError)
1771 break; /* let reaper() handle this */
1772 /* Start the bgwriter if not running */
1773 if (BgWriterPID == 0)
1774 BgWriterPID = StartBackgroundWriter();
1775 /* And tell it to shut down */
1776 if (BgWriterPID != 0)
1777 kill(BgWriterPID, SIGUSR2);
1782 * Immediate Shutdown:
1784 * abort all children with SIGQUIT and exit without attempt to
1785 * properly shut down data base system.
1788 (errmsg("received immediate shutdown request")));
1789 if (StartupPID != 0)
1790 kill(StartupPID, SIGQUIT);
1791 if (BgWriterPID != 0)
1792 kill(BgWriterPID, SIGQUIT);
1793 if (DLGetHead(BackendList))
1794 SignalChildren(SIGQUIT);
1799 PG_SETMASK(&UnBlockSig);
1805 * Reaper -- signal handler to cleanup after a backend (child) dies.
1810 int save_errno = errno;
1813 int status; /* backend exit status */
1817 union wait status; /* backend exit status */
1821 int pid; /* process id of dead backend */
1823 PG_SETMASK(&BlockSig);
1826 (errmsg_internal("reaping dead processes")));
1828 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1830 exitstatus = status;
1833 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1835 exitstatus = status.w_status;
1837 while ((pid = win32_waitpid(&exitstatus)) > 0)
1840 * We need to do this here, and not in CleanupProc, since this is
1841 * to be called on all children when we are done with them. Could
1842 * move to LogChildExit, but that seems like asking for future
1845 win32_RemoveChild(pid);
1847 #endif /* HAVE_WAITPID */
1850 * Check if this child was the statistics collector. If so, try to
1851 * start a new one. (If fail, we'll try again in future cycles of
1854 if (pgstat_ispgstat(pid))
1856 LogChildExit(LOG, gettext("statistics collector process"),
1863 * Check if this child was a startup process.
1865 if (StartupPID != 0 && pid == StartupPID)
1868 if (exitstatus != 0)
1870 LogChildExit(LOG, gettext("startup process"),
1873 (errmsg("aborting startup due to startup process failure")));
1878 * Startup succeeded - we are done with system startup or recovery.
1883 * Crank up the background writer. It doesn't matter if this
1884 * fails, we'll just try again later.
1886 Assert(BgWriterPID == 0);
1887 BgWriterPID = StartBackgroundWriter();
1890 * Go to shutdown mode if a shutdown request was pending.
1892 if (Shutdown > NoShutdown && BgWriterPID != 0)
1893 kill(BgWriterPID, SIGUSR2);
1899 * Was it the bgwriter?
1901 if (BgWriterPID != 0 && pid == BgWriterPID)
1903 if (exitstatus == 0 && Shutdown > NoShutdown &&
1904 !FatalError && !DLGetHead(BackendList))
1907 * Normal postmaster exit is here: we've seen normal
1908 * exit of the bgwriter after it's been told to shut down.
1909 * We expect that it wrote a shutdown checkpoint. (If
1910 * for some reason it didn't, recovery will occur on next
1911 * postmaster start.)
1916 * Any unexpected exit of the bgwriter is treated as a crash.
1918 LogChildExit(DEBUG2, gettext("background writer process"),
1920 HandleChildCrash(pid, exitstatus);
1925 * Else do standard backend child cleanup.
1927 CleanupProc(pid, exitstatus);
1928 } /* loop over pending child-death reports */
1933 * Wait for all children exit, then reset shmem and
1936 if (DLGetHead(BackendList) || StartupPID != 0 || BgWriterPID != 0)
1939 (errmsg("all server processes terminated; reinitializing")));
1942 reset_shared(PostPortNumber);
1944 StartupPID = StartupDataBase();
1949 if (Shutdown > NoShutdown)
1951 if (DLGetHead(BackendList) || StartupPID != 0)
1953 /* Start the bgwriter if not running */
1954 if (BgWriterPID == 0)
1955 BgWriterPID = StartBackgroundWriter();
1956 /* And tell it to shut down */
1957 if (BgWriterPID != 0)
1958 kill(BgWriterPID, SIGUSR2);
1962 PG_SETMASK(&UnBlockSig);
1969 * CleanupProc -- cleanup after terminated backend.
1971 * Remove all local state associated with backend.
1974 CleanupProc(int pid,
1975 int exitstatus) /* child's exit status. */
1979 LogChildExit(DEBUG2, gettext("server process"), pid, exitstatus);
1982 * If a backend dies in an ugly way (i.e. exit status not 0) then we
1983 * must signal all other backends to quickdie. If exit status is zero
1984 * we assume everything is hunky dory and simply remove the backend
1985 * from the active backend list.
1987 if (exitstatus != 0)
1989 HandleChildCrash(pid, exitstatus);
1993 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1995 Backend *bp = (Backend *) DLE_VAL(curr);
2003 ShmemBackendArrayRemove(pid);
2005 /* Tell the collector about backend termination */
2013 * HandleChildCrash -- cleanup after failed backend or bgwriter.
2015 * The objectives here are to clean up our local state about the child
2016 * process, and to signal all other remaining children to quickdie.
2019 HandleChildCrash(int pid,
2020 int exitstatus) /* child's exit status. */
2027 * Make log entry unless there was a previous crash (if so, nonzero
2028 * exit status is to be expected in SIGQUIT response; don't clutter log)
2033 (pid == BgWriterPID) ?
2034 gettext("background writer process") :
2035 gettext("server process"),
2038 (errmsg("terminating any other active server processes")));
2041 /* Process regular backends */
2042 for (curr = DLGetHead(BackendList); curr; curr = next)
2044 next = DLGetSucc(curr);
2045 bp = (Backend *) DLE_VAL(curr);
2049 * Found entry for freshly-dead backend, so remove it.
2055 ShmemBackendArrayRemove(pid);
2057 /* Tell the collector about backend termination */
2059 /* Keep looping so we can signal remaining backends */
2064 * This backend is still alive. Unless we did so already,
2065 * tell it to commit hara-kiri.
2067 * SIGQUIT is the special signal that says exit without proc_exit
2068 * and let the user know what's going on. But if SendStop is
2069 * set (-s on command line), then we send SIGSTOP instead, so
2070 * that we can get core dumps from all backends by hand.
2075 (errmsg_internal("sending %s to process %d",
2076 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2078 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2083 /* Take care of the bgwriter too */
2084 if (pid == BgWriterPID)
2086 else if (BgWriterPID != 0 && !FatalError)
2089 (errmsg_internal("sending %s to process %d",
2090 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2091 (int) BgWriterPID)));
2092 kill(BgWriterPID, (SendStop ? SIGSTOP : SIGQUIT));
2099 * Log the death of a child process.
2102 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2104 if (WIFEXITED(exitstatus))
2108 * translator: %s is a noun phrase describing a child process,
2109 * such as "server process"
2111 (errmsg("%s (PID %d) exited with exit code %d",
2112 procname, pid, WEXITSTATUS(exitstatus))));
2113 else if (WIFSIGNALED(exitstatus))
2117 * translator: %s is a noun phrase describing a child process,
2118 * such as "server process"
2120 (errmsg("%s (PID %d) was terminated by signal %d",
2121 procname, pid, WTERMSIG(exitstatus))));
2126 * translator: %s is a noun phrase describing a child process,
2127 * such as "server process"
2129 (errmsg("%s (PID %d) exited with unexpected status %d",
2130 procname, pid, exitstatus)));
2134 * Send a signal to all backend children.
2137 SignalChildren(int signal)
2141 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2143 Backend *bp = (Backend *) DLE_VAL(curr);
2146 (errmsg_internal("sending signal %d to process %d",
2147 signal, (int) bp->pid)));
2148 kill(bp->pid, signal);
2153 * BackendStartup -- start backend process
2155 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2158 BackendStartup(Port *port)
2160 Backend *bn; /* for backend cleanup */
2163 #ifdef LINUX_PROFILE
2164 struct itimerval prof_itimer;
2168 * Compute the cancel key that will be assigned to this backend. The
2169 * backend will have its own copy in the forked-off process' value of
2170 * MyCancelKey, so that it can transmit the key to the frontend.
2172 MyCancelKey = PostmasterRandom();
2175 * Make room for backend data structure. Better before the fork() so
2176 * we can handle failure cleanly.
2178 bn = (Backend *) malloc(sizeof(Backend));
2182 (errcode(ERRCODE_OUT_OF_MEMORY),
2183 errmsg("out of memory")));
2184 return STATUS_ERROR;
2187 /* Pass down canAcceptConnections state (kluge for EXEC_BACKEND case) */
2188 port->canAcceptConnections = canAcceptConnections();
2191 * Flush stdio channels just before fork, to avoid double-output
2192 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2193 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2194 * coredump if we do. Presently stdout and stderr are the only stdio
2195 * output channels used by the postmaster, so fflush'ing them should
2203 pid = backend_forkexec(port);
2205 #else /* !EXEC_BACKEND */
2207 #ifdef LINUX_PROFILE
2210 * Linux's fork() resets the profiling timer in the child process. If
2211 * we want to profile child processes then we need to save and restore
2212 * the timer setting. This is a waste of time if not profiling,
2213 * however, so only do it if commanded by specific -DLINUX_PROFILE
2216 getitimer(ITIMER_PROF, &prof_itimer);
2220 /* Specific beos actions before backend startup */
2221 beos_before_backend_startup();
2226 if (pid == 0) /* child */
2228 #ifdef LINUX_PROFILE
2229 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2233 /* Specific beos backend startup actions */
2234 beos_backend_startup();
2238 proc_exit(BackendRun(port));
2241 #endif /* EXEC_BACKEND */
2245 /* in parent, fork failed */
2246 int save_errno = errno;
2249 /* Specific beos backend startup actions */
2250 beos_backend_startup_failed();
2255 (errmsg("could not fork new process for connection: %m")));
2256 report_fork_failure_to_client(port, save_errno);
2257 return STATUS_ERROR;
2260 /* in parent, successful fork */
2262 (errmsg_internal("forked new backend, pid=%d socket=%d",
2263 (int) pid, port->sock)));
2266 * Everything's been successful, it's safe to add this backend to our
2270 bn->cancel_key = MyCancelKey;
2271 DLAddHead(BackendList, DLNewElem(bn));
2273 ShmemBackendArrayAdd(bn);
2280 * Try to report backend fork() failure to client before we close the
2281 * connection. Since we do not care to risk blocking the postmaster on
2282 * this connection, we set the connection to non-blocking and try only once.
2284 * This is grungy special-purpose code; we cannot use backend libpq since
2285 * it's not up and running.
2288 report_fork_failure_to_client(Port *port, int errnum)
2292 /* Format the error message packet (always V2 protocol) */
2293 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2294 gettext("could not fork new process for connection: "),
2297 /* Set port to non-blocking. Don't do send() if this fails */
2298 if (!set_noblock(port->sock))
2301 send(port->sock, buffer, strlen(buffer) + 1, 0);
2306 * split_opts -- split a string of options and append it to an argv array
2308 * NB: the string is destructively modified!
2310 * Since no current POSTGRES arguments require any quoting characters,
2311 * we can use the simple-minded tactic of assuming each set of space-
2312 * delimited characters is a separate argv element.
2314 * If you don't like that, well, we *used* to pass the whole option string
2315 * as ONE argument to execl(), which was even less intelligent...
2318 split_opts(char **argv, int *argcp, char *s)
2322 while (isspace((unsigned char) *s))
2326 argv[(*argcp)++] = s;
2327 while (*s && !isspace((unsigned char) *s))
2336 * BackendRun -- perform authentication, and if successful,
2337 * set up the backend's argument list and invoke PostgresMain()
2340 * Shouldn't return at all.
2341 * If PostgresMain() fails, return status.
2344 BackendRun(Port *port)
2349 char remote_host[NI_MAXHOST];
2350 char remote_port[NI_MAXSERV];
2351 char remote_ps_data[NI_MAXHOST];
2359 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2362 * Let's clean up ourselves as the postmaster child, and close the
2363 * postmaster's listen sockets
2365 ClosePostmasterPorts();
2367 /* We don't want the postmaster's proc_exit() handlers */
2371 * Signal handlers setting is moved to tcop/postgres...
2374 /* Save port etc. for ps status */
2377 /* Reset MyProcPid to new backend's pid */
2378 MyProcPid = getpid();
2381 * PreAuthDelay is a debugging aid for investigating problems in the
2382 * authentication cycle: it can be set in postgresql.conf to allow
2383 * time to attach to the newly-forked backend with a debugger. (See
2384 * also the -W backend switch, which we allow clients to pass through
2385 * PGOPTIONS, but it is not honored until after authentication.)
2387 if (PreAuthDelay > 0)
2388 pg_usleep(PreAuthDelay * 1000000L);
2390 ClientAuthInProgress = true; /* limit visibility of log messages */
2392 /* save start time for end of session reporting */
2393 gettimeofday(&(port->session_start), NULL);
2395 /* set these to empty in case they are needed before we set them up */
2396 port->remote_host = "";
2397 port->remote_port = "";
2398 port->commandTag = "";
2401 * Initialize libpq and enable reporting of ereport errors to the
2402 * client. Must do this now because authentication uses libpq to send
2405 pq_init(); /* initialize libpq to talk to client */
2406 whereToSendOutput = Remote; /* now safe to ereport to client */
2409 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2410 * during any client authentication related communication. Otherwise
2411 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2412 * if a buggy client blocks a backend during authentication.
2414 pqsignal(SIGTERM, authdie);
2415 pqsignal(SIGQUIT, authdie);
2416 pqsignal(SIGALRM, authdie);
2417 PG_SETMASK(&AuthBlockSig);
2420 * Get the remote host name and port for logging and status display.
2422 remote_host[0] = '\0';
2423 remote_port[0] = '\0';
2424 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2425 remote_host, sizeof(remote_host),
2426 remote_port, sizeof(remote_port),
2427 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2429 int ret = getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2430 remote_host, sizeof(remote_host),
2431 remote_port, sizeof(remote_port),
2432 NI_NUMERICHOST | NI_NUMERICSERV);
2436 (errmsg("getnameinfo_all() failed: %s",
2437 gai_strerror(ret))));
2439 snprintf(remote_ps_data, sizeof(remote_ps_data),
2440 remote_port[0] == '\0' ? "%s" : "%s(%s)",
2441 remote_host, remote_port);
2443 if (Log_connections)
2445 (errmsg("connection received: host=%s port=%s",
2446 remote_host, remote_port)));
2449 * save remote_host and remote_port in port stucture
2451 port->remote_host = strdup(remote_host);
2452 port->remote_port = strdup(remote_port);
2455 * In EXEC_BACKEND case, we didn't inherit the contents of pg_hba.c
2456 * etcetera from the postmaster, and have to load them ourselves.
2457 * Build the PostmasterContext (which didn't exist before, in this
2458 * process) to contain the data.
2460 * FIXME: [fork/exec] Ugh. Is there a way around this overhead?
2463 Assert(PostmasterContext == NULL);
2464 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
2466 ALLOCSET_DEFAULT_MINSIZE,
2467 ALLOCSET_DEFAULT_INITSIZE,
2468 ALLOCSET_DEFAULT_MAXSIZE);
2469 MemoryContextSwitchTo(PostmasterContext);
2478 * Ready to begin client interaction. We will give up and exit(0)
2479 * after a time delay, so that a broken client can't hog a connection
2480 * indefinitely. PreAuthDelay doesn't count against the time limit.
2482 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2483 elog(FATAL, "could not set timer for authorization timeout");
2486 * Receive the startup packet (which might turn out to be a cancel
2489 status = ProcessStartupPacket(port, false);
2491 if (status != STATUS_OK)
2495 * Now that we have the user and database name, we can set the process
2496 * title for ps. It's good to do this as early as possible in
2499 init_ps_display(port->user_name, port->database_name, remote_ps_data);
2500 set_ps_display("authentication");
2503 * Now perform authentication exchange.
2505 ClientAuthentication(port); /* might not return, if failure */
2508 * Done with authentication. Disable timeout, and prevent
2509 * SIGTERM/SIGQUIT again until backend startup is complete.
2511 if (!disable_sig_alarm(false))
2512 elog(FATAL, "could not disable timer for authorization timeout");
2513 PG_SETMASK(&BlockSig);
2515 if (Log_connections)
2517 (errmsg("connection authorized: user=%s database=%s",
2518 port->user_name, port->database_name)));
2521 * Don't want backend to be able to see the postmaster random number
2522 * generator state. We have to clobber the static random_seed *and*
2523 * start a new random sequence in the random() library function.
2526 gettimeofday(&now, &tz);
2527 srandom((unsigned int) now.tv_usec);
2531 * Now, build the argv vector that will be given to PostgresMain.
2533 * The layout of the command line is
2534 * postgres [secure switches] -p databasename [insecure switches]
2535 * where the switches after -p come from the client request.
2537 * The maximum possible number of commandline arguments that could come
2538 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2542 maxac = 10; /* for fixed args supplied below */
2543 maxac += (strlen(ExtraOptions) + 1) / 2;
2544 if (port->cmdline_options)
2545 maxac += (strlen(port->cmdline_options) + 1) / 2;
2547 av = (char **) MemoryContextAlloc(TopMemoryContext,
2548 maxac * sizeof(char *));
2551 av[ac++] = "postgres";
2554 * Pass the requested debugging level along to the backend.
2558 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2559 av[ac++] = debugbuf;
2563 * Pass any backend switches specified with -o in the postmaster's own
2564 * command line. We assume these are secure. (It's OK to mangle
2565 * ExtraOptions now, since we're safely inside a subprocess.)
2567 split_opts(av, &ac, ExtraOptions);
2569 /* Tell the backend what protocol the frontend is using. */
2570 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2571 av[ac++] = protobuf;
2574 * Tell the backend it is being called from the postmaster, and which
2575 * database to use. -p marks the end of secure switches.
2578 av[ac++] = port->database_name;
2581 * Pass the (insecure) option switches from the connection request.
2582 * (It's OK to mangle port->cmdline_options now.)
2584 if (port->cmdline_options)
2585 split_opts(av, &ac, port->cmdline_options);
2592 * Release postmaster's working memory context so that backend can
2593 * recycle the space. Note this does not trash *MyProcPort, because
2594 * ConnCreate() allocated that space with malloc() ... else we'd need
2595 * to copy the Port data here. Also, subsidiary data such as the
2596 * username isn't lost either; see ProcessStartupPacket().
2598 MemoryContextSwitchTo(TopMemoryContext);
2599 MemoryContextDelete(PostmasterContext);
2600 PostmasterContext = NULL;
2603 * Debug: print arguments being passed to backend
2606 (errmsg_internal("%s child[%d]: starting with (",
2607 progname, getpid())));
2608 for (i = 0; i < ac; ++i)
2610 (errmsg_internal("\t%s", av[i])));
2612 (errmsg_internal(")")));
2614 ClientAuthInProgress = false; /* client_min_messages is active
2617 return (PostgresMain(ac, av, port->user_name));
2624 * postmaster_forkexec -- fork and exec a postmaster subprocess
2626 * The caller must have set up the argv array already, except for argv[2]
2627 * which will be filled with the name of the temp variable file.
2629 * Returns the child process PID, or -1 on fork failure (a suitable error
2630 * message has been logged on failure).
2632 * All uses of this routine will dispatch to SubPostmasterMain in the
2636 postmaster_forkexec(int argc, char *argv[])
2640 /* This entry point passes dummy values for the Port variables */
2641 memset(&port, 0, sizeof(port));
2642 return internal_forkexec(argc, argv, &port);
2646 * backend_forkexec -- fork/exec off a backend process
2648 * returns the pid of the fork/exec'd process, or -1 on failure
2651 backend_forkexec(Port *port)
2656 av[ac++] = "postgres";
2657 av[ac++] = "-forkbackend";
2658 av[ac++] = NULL; /* filled in by internal_forkexec */
2661 Assert(ac < lengthof(av));
2663 return internal_forkexec(ac, av, port);
2667 internal_forkexec(int argc, char *argv[], Port *port)
2670 char tmpfilename[MAXPGPATH];
2672 if (!write_backend_variables(tmpfilename, port))
2673 return -1; /* log made by write_backend_variables */
2675 /* Make sure caller set up argv properly */
2677 Assert(argv[argc] == NULL);
2678 Assert(strncmp(argv[1], "-fork", 5) == 0);
2679 Assert(argv[2] == NULL);
2681 /* Insert temp file name after -fork argument */
2682 argv[2] = tmpfilename;
2685 pid = win32_forkexec(postgres_exec_path, argv);
2687 /* Fire off execv in child */
2688 if ((pid = fork()) == 0)
2690 if (execv(postgres_exec_path, argv) < 0)
2693 (errmsg("could not exec backend process \"%s\": %m",
2694 postgres_exec_path)));
2695 /* We're already in the child process here, can't return */
2701 return pid; /* Parent returns pid, or -1 on fork failure */
2705 * SubPostmasterMain -- Get the fork/exec'd process into a state equivalent
2706 * to what it would be if we'd simply forked on Unix, and then
2707 * dispatch to the appropriate place.
2709 * The first two command line arguments are expected to be "-forkFOO"
2710 * (where FOO indicates which postmaster child we are to become), and
2711 * the name of a variables file that we can read to load data that would
2712 * have been inherited by fork() on Unix. Remaining arguments go to the
2713 * subprocess FooMain() routine.
2716 SubPostmasterMain(int argc, char *argv[])
2720 /* Do this sooner rather than later... */
2721 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2723 MyProcPid = getpid(); /* reset MyProcPid */
2725 /* In EXEC_BACKEND case we will not have inherited these settings */
2726 IsPostmasterEnvironment = true;
2727 whereToSendOutput = None;
2729 PG_SETMASK(&BlockSig);
2731 /* Setup essential subsystems */
2732 MemoryContextInit();
2733 InitializeGUCOptions();
2735 /* Check we got appropriate args */
2737 elog(FATAL, "invalid subpostmaster invocation");
2739 /* Read in file-based context */
2740 memset(&port, 0, sizeof(Port));
2741 read_backend_variables(argv[2], &port);
2742 read_nondefault_variables();
2744 /* Run backend or appropriate child */
2745 if (strcmp(argv[1], "-forkbackend") == 0)
2747 /* BackendRun will close sockets */
2749 /* Attach process to shared segments */
2750 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2752 Assert(argc == 3); /* shouldn't be any more args */
2753 proc_exit(BackendRun(&port));
2755 if (strcmp(argv[1], "-forkboot") == 0)
2757 /* Close the postmaster's sockets */
2758 ClosePostmasterPorts();
2760 /* Attach process to shared segments */
2761 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2763 BootstrapMain(argc - 2, argv + 2);
2766 if (strcmp(argv[1], "-forkbuf") == 0)
2768 /* Close the postmaster's sockets */
2769 ClosePostmasterPorts();
2771 /* Do not want to attach to shared memory */
2773 PgstatBufferMain(argc, argv);
2776 if (strcmp(argv[1], "-forkcol") == 0)
2779 * Do NOT close postmaster sockets here, because we are forking from
2780 * pgstat buffer process, which already did it.
2783 /* Do not want to attach to shared memory */
2785 PgstatCollectorMain(argc, argv);
2789 return 1; /* shouldn't get here */
2792 #endif /* EXEC_BACKEND */
2796 * ExitPostmaster -- cleanup
2798 * Do NOT call exit() directly --- always go through here!
2801 ExitPostmaster(int status)
2803 /* should cleanup shared memory and kill all backends */
2806 * Not sure of the semantics here. When the Postmaster dies, should
2807 * the backends all be killed? probably not.
2809 * MUST -- vadim 05-10-1999
2816 * sigusr1_handler - handle signal conditions from child processes
2819 sigusr1_handler(SIGNAL_ARGS)
2821 int save_errno = errno;
2823 PG_SETMASK(&BlockSig);
2825 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2828 * Password or group file has changed.
2834 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2837 * Send SIGUSR1 to all children (triggers
2838 * CatchupInterruptHandler). See storage/ipc/sinval[adt].c for the
2841 if (Shutdown <= SmartShutdown)
2842 SignalChildren(SIGUSR1);
2845 PG_SETMASK(&UnBlockSig);
2852 * Dummy signal handler
2854 * We use this for signals that we don't actually use in the postmaster,
2855 * but we do use in backends. If we were to SIG_IGN such signals in the
2856 * postmaster, then a newly started backend might drop a signal that arrives
2857 * before it's able to reconfigure its signal processing. (See notes in
2861 dummy_handler(SIGNAL_ARGS)
2867 * CharRemap: given an int in range 0..61, produce textual encoding of it
2868 * per crypt(3) conventions.
2892 RandomSalt(char *cryptSalt, char *md5Salt)
2894 long rand = PostmasterRandom();
2896 cryptSalt[0] = CharRemap(rand % 62);
2897 cryptSalt[1] = CharRemap(rand / 62);
2900 * It's okay to reuse the first random value for one of the MD5 salt
2901 * bytes, since only one of the two salts will be sent to the client.
2902 * After that we need to compute more random bits.
2904 * We use % 255, sacrificing one possible byte value, so as to ensure
2905 * that all bits of the random() value participate in the result.
2906 * While at it, add one to avoid generating any null bytes.
2908 md5Salt[0] = (rand % 255) + 1;
2909 rand = PostmasterRandom();
2910 md5Salt[1] = (rand % 255) + 1;
2911 rand = PostmasterRandom();
2912 md5Salt[2] = (rand % 255) + 1;
2913 rand = PostmasterRandom();
2914 md5Salt[3] = (rand % 255) + 1;
2921 PostmasterRandom(void)
2923 static bool initialized = false;
2927 Assert(random_seed != 0);
2928 srandom(random_seed);
2936 * Count up number of child processes.
2944 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2953 * StartChildProcess -- start a non-backend child process for the postmaster
2955 * xlog determines what kind of child will be started. All child types
2956 * initially go to BootstrapMain, which will handle common setup.
2958 * Return value of StartChildProcess is subprocess' PID, or 0 if failed
2959 * to start subprocess.
2962 StartChildProcess(int xlop)
2968 #ifdef LINUX_PROFILE
2969 struct itimerval prof_itimer;
2973 * Set up command-line arguments for subprocess
2975 av[ac++] = "postgres";
2978 av[ac++] = "-forkboot";
2979 av[ac++] = NULL; /* filled in by postmaster_forkexec */
2982 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
2986 av[ac++] = "template1";
2989 Assert(ac < lengthof(av));
2992 * Flush stdio channels (see comments in BackendStartup)
2999 pid = postmaster_forkexec(ac, av);
3001 #else /* !EXEC_BACKEND */
3003 #ifdef LINUX_PROFILE
3004 /* see comments in BackendStartup */
3005 getitimer(ITIMER_PROF, &prof_itimer);
3009 /* Specific beos actions before backend startup */
3010 beos_before_backend_startup();
3015 if (pid == 0) /* child */
3017 #ifdef LINUX_PROFILE
3018 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3022 /* Specific beos actions after backend startup */
3023 beos_backend_startup();
3026 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
3028 /* Close the postmaster's sockets */
3029 ClosePostmasterPorts();
3031 /* Lose the postmaster's on-exit routines and port connections */
3034 BootstrapMain(ac, av);
3038 #endif /* EXEC_BACKEND */
3042 /* in parent, fork failed */
3043 int save_errno = errno;
3046 /* Specific beos actions before backend startup */
3047 beos_backend_startup_failed();
3052 case BS_XLOG_STARTUP:
3054 (errmsg("could not fork startup process: %m")));
3056 case BS_XLOG_BGWRITER:
3058 (errmsg("could not fork background writer process: %m")));
3062 (errmsg("could not fork process: %m")));
3067 * fork failure is fatal during startup, but there's no need
3068 * to choke immediately if starting other child types fails.
3070 if (xlop == BS_XLOG_STARTUP)
3076 * in parent, successful fork
3083 * Create the opts file
3086 CreateOptsFile(int argc, char *argv[], char *fullprogname)
3088 char filename[MAXPGPATH];
3092 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3094 if ((fp = fopen(filename, "w")) == NULL)
3096 elog(LOG, "could not create file \"%s\": %m", filename);
3100 fprintf(fp, "%s", fullprogname);
3101 for (i = 1; i < argc; i++)
3102 fprintf(fp, " '%s'", argv[i]);
3107 elog(LOG, "could not write file \"%s\": %m", filename);
3115 * This should be used only for reporting "interactive" errors (essentially,
3116 * bogus arguments on the command line). Once the postmaster is launched,
3117 * use ereport. In particular, don't use this for anything that occurs
3118 * after pmdaemonize.
3121 postmaster_error(const char *fmt,...)
3125 fprintf(stderr, "%s: ", progname);
3127 vfprintf(stderr, gettext(fmt), ap);
3129 fprintf(stderr, "\n");
3136 * The following need to be available to the read/write_backend_variables
3139 #include "storage/spin.h"
3141 extern slock_t *ShmemLock;
3142 extern slock_t *ShmemIndexLock;
3143 extern void *ShmemIndexAlloc;
3144 typedef struct LWLock LWLock;
3145 extern LWLock *LWLockArray;
3146 extern slock_t *ProcStructLock;
3147 extern int pgStatSock;
3149 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3150 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3151 #define write_array_var(var,fp) fwrite((void*)(var),sizeof(var),1,fp)
3152 #define read_array_var(var,fp) fread((void*)(var),sizeof(var),1,fp)
3155 write_backend_variables(char *filename, Port *port)
3157 static unsigned long tmpBackendFileNum = 0;
3159 char str_buf[MAXPGPATH];
3161 /* Calculate name for temp file in caller's buffer */
3163 snprintf(filename, MAXPGPATH, "%s/%s/%s.backend_var.%d.%lu",
3164 DataDir, PG_TEMP_FILES_DIR, PG_TEMP_FILE_PREFIX,
3165 MyProcPid, ++tmpBackendFileNum);
3168 fp = AllocateFile(filename, PG_BINARY_W);
3171 /* As per OpenTemporaryFile... */
3172 char dirname[MAXPGPATH];
3174 snprintf(dirname, MAXPGPATH, "%s/%s", DataDir, PG_TEMP_FILES_DIR);
3175 mkdir(dirname, S_IRWXU);
3177 fp = AllocateFile(filename, PG_BINARY_W);
3181 (errcode_for_file_access(),
3182 errmsg("could not create file \"%s\": %m",
3189 write_var(port->sock, fp);
3190 write_var(port->proto, fp);
3191 write_var(port->laddr, fp);
3192 write_var(port->raddr, fp);
3193 write_var(port->canAcceptConnections, fp);
3194 write_var(port->cryptSalt, fp);
3195 write_var(port->md5Salt, fp);
3198 * XXX FIXME later: writing these strings as MAXPGPATH bytes always is
3199 * probably a waste of resources
3202 StrNCpy(str_buf, DataDir, MAXPGPATH);
3203 write_array_var(str_buf, fp);
3205 write_array_var(ListenSocket, fp);
3207 write_var(MyCancelKey, fp);
3209 write_var(UsedShmemSegID, fp);
3210 write_var(UsedShmemSegAddr, fp);
3212 write_var(ShmemLock, fp);
3213 write_var(ShmemIndexLock, fp);
3214 write_var(ShmemVariableCache, fp);
3215 write_var(ShmemIndexAlloc, fp);
3216 write_var(ShmemBackendArray, fp);
3218 write_var(LWLockArray, fp);
3219 write_var(ProcStructLock, fp);
3220 write_var(pgStatSock, fp);
3222 write_var(debug_flag, fp);
3223 write_var(PostmasterPid, fp);
3225 StrNCpy(str_buf, my_exec_path, MAXPGPATH);
3226 write_array_var(str_buf, fp);
3228 write_array_var(ExtraOptions, fp);
3230 StrNCpy(str_buf, setlocale(LC_COLLATE, NULL), MAXPGPATH);
3231 write_array_var(str_buf, fp);
3232 StrNCpy(str_buf, setlocale(LC_CTYPE, NULL), MAXPGPATH);
3233 write_array_var(str_buf, fp);
3239 (errcode_for_file_access(),
3240 errmsg("could not write to file \"%s\": %m", filename)));
3248 read_backend_variables(char *filename, Port *port)
3251 char str_buf[MAXPGPATH];
3254 fp = AllocateFile(filename, PG_BINARY_R);
3257 (errcode_for_file_access(),
3258 errmsg("could not read from backend variables file \"%s\": %m",
3262 read_var(port->sock, fp);
3263 read_var(port->proto, fp);
3264 read_var(port->laddr, fp);
3265 read_var(port->raddr, fp);
3266 read_var(port->canAcceptConnections, fp);
3267 read_var(port->cryptSalt, fp);
3268 read_var(port->md5Salt, fp);
3270 read_array_var(str_buf, fp);
3271 SetDataDir(str_buf);
3273 read_array_var(ListenSocket, fp);
3275 read_var(MyCancelKey, fp);
3277 read_var(UsedShmemSegID, fp);
3278 read_var(UsedShmemSegAddr, fp);
3280 read_var(ShmemLock, fp);
3281 read_var(ShmemIndexLock, fp);
3282 read_var(ShmemVariableCache, fp);
3283 read_var(ShmemIndexAlloc, fp);
3284 read_var(ShmemBackendArray, fp);
3286 read_var(LWLockArray, fp);
3287 read_var(ProcStructLock, fp);
3288 read_var(pgStatSock, fp);
3290 read_var(debug_flag, fp);
3291 read_var(PostmasterPid, fp);
3293 read_array_var(str_buf, fp);
3294 StrNCpy(my_exec_path, str_buf, MAXPGPATH);
3296 read_array_var(ExtraOptions, fp);
3298 read_array_var(str_buf, fp);
3299 setlocale(LC_COLLATE, str_buf);
3300 read_array_var(str_buf, fp);
3301 setlocale(LC_CTYPE, str_buf);
3305 if (unlink(filename) != 0)
3307 (errcode_for_file_access(),
3308 errmsg("could not remove file \"%s\": %m", filename)));
3313 ShmemBackendArraySize(void)
3315 return (NUM_BACKENDARRAY_ELEMS * sizeof(Backend));
3319 ShmemBackendArrayAllocation(void)
3321 size_t size = ShmemBackendArraySize();
3323 ShmemBackendArray = (Backend *) ShmemAlloc(size);
3324 /* Mark all slots as empty */
3325 memset(ShmemBackendArray, 0, size);
3329 ShmemBackendArrayAdd(Backend *bn)
3333 /* Find an empty slot */
3334 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3336 if (ShmemBackendArray[i].pid == 0)
3338 ShmemBackendArray[i] = *bn;
3344 (errmsg_internal("no free slots in shmem backend array")));
3348 ShmemBackendArrayRemove(pid_t pid)
3352 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3354 if (ShmemBackendArray[i].pid == pid)
3356 /* Mark the slot as empty */
3357 ShmemBackendArray[i].pid = 0;
3363 (errmsg_internal("unable to find backend entry with pid %d",
3367 #endif /* EXEC_BACKEND */
3373 win32_forkexec(const char *path, char *argv[])
3376 PROCESS_INFORMATION pi;
3379 char cmdLine[MAXPGPATH * 2];
3380 HANDLE childHandleCopy;
3381 HANDLE waiterThread;
3383 /* Format the cmd line */
3384 cmdLine[sizeof(cmdLine)-1] = '\0';
3385 cmdLine[sizeof(cmdLine)-2] = '\0';
3386 snprintf(cmdLine, sizeof(cmdLine)-1, "\"%s\"", path);
3388 while (argv[++i] != NULL)
3390 j = strlen(cmdLine);
3391 snprintf(cmdLine+j, sizeof(cmdLine)-1-j, " \"%s\"", argv[i]);
3393 if (cmdLine[sizeof(cmdLine)-2] != '\0')
3395 elog(LOG, "subprocess command line too long");
3399 memset(&pi, 0, sizeof(pi));
3400 memset(&si, 0, sizeof(si));
3402 if (!CreateProcess(NULL, cmdLine, NULL, NULL, TRUE, 0, NULL, NULL, &si, &pi))
3404 elog(LOG, "CreateProcess call failed (%d): %m", (int) GetLastError());
3408 if (!IsUnderPostmaster)
3410 /* We are the Postmaster creating a child... */
3411 win32_AddChild(pi.dwProcessId, pi.hProcess);
3414 if (!DuplicateHandle(GetCurrentProcess(),
3416 GetCurrentProcess(),
3420 DUPLICATE_SAME_ACCESS))
3422 (errmsg_internal("failed to duplicate child handle: %d",
3423 (int) GetLastError())));
3425 waiterThread = CreateThread(NULL, 64 * 1024, win32_sigchld_waiter,
3426 (LPVOID) childHandleCopy, 0, NULL);
3429 (errmsg_internal("failed to create sigchld waiter thread: %d",
3430 (int) GetLastError())));
3431 CloseHandle(waiterThread);
3433 if (IsUnderPostmaster)
3434 CloseHandle(pi.hProcess);
3435 CloseHandle(pi.hThread);
3437 return pi.dwProcessId;
3441 * Note: The following three functions must not be interrupted (eg. by
3442 * signals). As the Postgres Win32 signalling architecture (currently)
3443 * requires polling, or APC checking functions which aren't used here, this
3446 * We keep two separate arrays, instead of a single array of pid/HANDLE
3447 * structs, to avoid having to re-create a handle array for
3448 * WaitForMultipleObjects on each call to win32_waitpid.
3452 win32_AddChild(pid_t pid, HANDLE handle)
3454 Assert(win32_childPIDArray && win32_childHNDArray);
3455 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3457 win32_childPIDArray[win32_numChildren] = pid;
3458 win32_childHNDArray[win32_numChildren] = handle;
3459 ++win32_numChildren;
3463 (errmsg_internal("unable to add child entry with pid %lu",
3464 (unsigned long) pid)));
3468 win32_RemoveChild(pid_t pid)
3472 Assert(win32_childPIDArray && win32_childHNDArray);
3474 for (i = 0; i < win32_numChildren; i++)
3476 if (win32_childPIDArray[i] == pid)
3478 CloseHandle(win32_childHNDArray[i]);
3480 /* Swap last entry into the "removed" one */
3481 --win32_numChildren;
3482 win32_childPIDArray[i] = win32_childPIDArray[win32_numChildren];
3483 win32_childHNDArray[i] = win32_childHNDArray[win32_numChildren];
3489 (errmsg_internal("unable to find child entry with pid %lu",
3490 (unsigned long) pid)));
3494 win32_waitpid(int *exitstatus)
3496 Assert(win32_childPIDArray && win32_childHNDArray);
3497 elog(DEBUG3, "waiting on %lu children", win32_numChildren);
3499 if (win32_numChildren > 0)
3502 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't want to
3503 * run queued APCs here.
3509 ret = WaitForMultipleObjects(win32_numChildren, win32_childHNDArray,
3515 (errmsg_internal("failed to wait on %lu children: %d",
3516 win32_numChildren, (int) GetLastError())));
3520 /* No children have finished */
3526 * Get the exit code, and return the PID of, the
3527 * respective process
3529 index = ret - WAIT_OBJECT_0;
3530 Assert(index >= 0 && index < win32_numChildren);
3531 if (!GetExitCodeProcess(win32_childHNDArray[index], &exitCode))
3534 * If we get this far, this should never happen, but,
3535 * then again... No choice other than to assume a
3536 * catastrophic failure.
3539 (errmsg_internal("failed to get exit code for child %lu",
3540 win32_childPIDArray[index])));
3542 *exitstatus = (int) exitCode;
3543 return win32_childPIDArray[index];
3552 * Note! Code below executes on separate threads, one for
3553 * each child process created
3556 win32_sigchld_waiter(LPVOID param)
3558 HANDLE procHandle = (HANDLE) param;
3560 DWORD r = WaitForSingleObject(procHandle, INFINITE);
3562 if (r == WAIT_OBJECT_0)
3563 pg_queue_signal(SIGCHLD);
3565 fprintf(stderr, "ERROR: Failed to wait on child process handle: %i\n",
3566 (int) GetLastError());
3567 CloseHandle(procHandle);