1 /*-------------------------------------------------------------------------
4 * This program acts as a clearing house for requests to the
5 * POSTGRES system. Frontend programs send a startup message
6 * to the Postmaster and the postmaster uses the info in the
7 * message to setup a backend process.
9 * The postmaster also manages system-wide operations such as
10 * startup, shutdown, and periodic checkpoints. The postmaster
11 * itself doesn't do those operations, mind you --- it just forks
12 * off a subprocess to do them at the right times. It also takes
13 * care of resetting the system if a backend crashes.
15 * The postmaster process creates the shared memory and semaphore
16 * pools during startup, but as a rule does not touch them itself.
17 * In particular, it is not a member of the PGPROC array of backends
18 * and so it cannot participate in lock-manager operations. Keeping
19 * the postmaster away from shared memory operations makes it simpler
20 * and more reliable. The postmaster is almost always able to recover
21 * from crashes of individual backends by resetting shared memory;
22 * if it did much with shared memory then it would be prone to crashing
23 * along with the backends.
25 * When a request message is received, we now fork() immediately.
26 * The child process performs authentication of the request, and
27 * then becomes a backend if successful. This allows the auth code
28 * to be written in a simple single-threaded style (as opposed to the
29 * crufty "poor man's multitasking" code that used to be needed).
30 * More importantly, it ensures that blockages in non-multithreaded
31 * libraries like SSL or PAM cannot cause denial of service to other
35 * Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
36 * Portions Copyright (c) 1994, Regents of the University of California
40 * $PostgreSQL: pgsql/src/backend/postmaster/postmaster.c,v 1.373 2004/03/10 21:12:46 momjian Exp $
45 * The Postmaster sets up a few shared memory data structures
46 * for the backends. It should at the very least initialize the
50 * The Postmaster shares memory with the backends but should avoid
51 * touching shared memory, so as not to become stuck if a crashing
52 * backend screws up locks or shared memory. Likewise, the Postmaster
53 * should never block on messages from frontend clients.
56 * The Postmaster cleans up after backends if they have an emergency
57 * exit and/or core dump.
59 *-------------------------------------------------------------------------
70 #include <sys/socket.h>
74 #include <sys/param.h>
75 #include <netinet/in.h>
76 #include <arpa/inet.h>
80 #ifdef HAVE_SYS_SELECT_H
81 #include <sys/select.h>
89 #include <DNSServiceDiscovery/DNSServiceDiscovery.h>
92 #include "catalog/pg_database.h"
93 #include "commands/async.h"
94 #include "lib/dllist.h"
95 #include "libpq/auth.h"
96 #include "libpq/crypt.h"
97 #include "libpq/libpq.h"
98 #include "libpq/pqcomm.h"
99 #include "libpq/pqsignal.h"
100 #include "miscadmin.h"
101 #include "nodes/nodes.h"
102 #include "storage/fd.h"
103 #include "storage/ipc.h"
104 #include "storage/pg_shmem.h"
105 #include "storage/pmsignal.h"
106 #include "storage/proc.h"
107 #include "storage/bufmgr.h"
108 #include "access/xlog.h"
109 #include "tcop/tcopprot.h"
110 #include "utils/guc.h"
111 #include "utils/memutils.h"
112 #include "utils/ps_status.h"
113 #include "bootstrap/bootstrap.h"
117 #define INVALID_SOCK (-1)
119 #ifdef HAVE_SIGPROCMASK
131 * List of active backends (or child processes anyway; we don't actually
132 * know whether a given child has become a backend or is still in the
133 * authorization phase). This is used mainly to keep track of how many
134 * children we have and send them appropriate signals when necessary.
138 pid_t pid; /* process id of backend */
139 long cancel_key; /* cancel key for cancels for this backend */
142 static Dllist *BackendList;
145 #define NUM_BACKENDARRAY_ELEMS (2*MaxBackends)
146 static Backend *ShmemBackendArray;
149 /* The socket number we are listening for connections on */
155 * MaxBackends is the limit on the number of backends we can start.
156 * Note that a larger MaxBackends value will increase the size of the
157 * shared memory area as well as cause the postmaster to grab more
158 * kernel semaphores, even if you never actually use that many
164 * ReservedBackends is the number of backends reserved for superuser use.
165 * This number is taken out of the pool size given by MaxBackends so
166 * number of backend slots available to non-superusers is
167 * (MaxBackends - ReservedBackends). Note what this really means is
168 * "if there are <= ReservedBackends connections available, only superusers
169 * can make new connections" --- pre-existing superuser connections don't
170 * count against the limit.
172 int ReservedBackends;
175 static char *progname = NULL;
177 /* The socket(s) we're listening to. */
179 static int ListenSocket[MAXLISTEN];
181 /* Used to reduce macros tests */
183 const bool ExecBackend = true;
186 const bool ExecBackend = false;
190 * Set by the -o option
192 static char ExtraOptions[MAXPGPATH];
195 * These globals control the behavior of the postmaster in case some
196 * backend dumps core. Normally, it kills all peers of the dead backend
197 * and reinitializes shared memory. By specifying -s or -n, we can have
198 * the postmaster stop (rather than kill) peers and not reinitialize
199 * shared data structures.
201 static bool Reinit = true;
202 static int SendStop = false;
204 /* still more option variables */
205 bool NetServer = false; /* listen on TCP/IP */
206 bool EnableSSL = false;
207 bool SilentMode = false; /* silent mode (-S) */
209 int PreAuthDelay = 0;
210 int AuthenticationTimeout = 60;
211 int CheckPointTimeout = 300;
212 int CheckPointWarning = 30;
213 time_t LastSignalledCheckpoint = 0;
215 bool log_hostname; /* for ps display */
217 bool Log_connections = false;
218 bool Db_user_namespace = false;
220 char *rendezvous_name;
222 /* list of library:init-function to be preloaded */
223 char *preload_libraries_string = NULL;
225 /* Startup/shutdown state */
226 static pid_t StartupPID = 0,
230 static time_t checkpointed = 0;
233 #define SmartShutdown 1
234 #define FastShutdown 2
236 static int Shutdown = NoShutdown;
238 static bool FatalError = false; /* T if recovering from backend crash */
240 bool ClientAuthInProgress = false; /* T during new-client
244 * State for assigning random salts and cancel keys.
245 * Also, the global MyCancelKey passes the cancel key assigned to a given
246 * backend from the postmaster to that backend (via fork).
249 static unsigned int random_seed = 0;
251 static int debug_flag = 0;
257 #ifdef HAVE_INT_OPTRESET
262 * postmaster.c - function prototypes
264 static void pmdaemonize(int argc, char *argv[]);
265 static Port *ConnCreate(int serverFd);
266 static void ConnFree(Port *port);
267 static void reset_shared(unsigned short port);
268 static void SIGHUP_handler(SIGNAL_ARGS);
269 static void pmdie(SIGNAL_ARGS);
270 static void reaper(SIGNAL_ARGS);
271 static void sigusr1_handler(SIGNAL_ARGS);
272 static void dummy_handler(SIGNAL_ARGS);
273 static void CleanupProc(int pid, int exitstatus);
274 static void LogChildExit(int lev, const char *procname,
275 int pid, int exitstatus);
276 static void BackendInit(Port *port);
277 static int BackendRun(Port *port);
278 static void ExitPostmaster(int status);
279 static void usage(const char *);
280 static int ServerLoop(void);
281 static int BackendStartup(Port *port);
282 static int ProcessStartupPacket(Port *port, bool SSLdone);
283 static void processCancelRequest(Port *port, void *pkt);
284 static int initMasks(fd_set *rmask);
285 static void report_fork_failure_to_client(Port *port, int errnum);
286 static enum CAC_state canAcceptConnections(void);
287 static long PostmasterRandom(void);
288 static void RandomSalt(char *cryptSalt, char *md5Salt);
289 static void SignalChildren(int signal);
290 static int CountChildren(void);
291 static bool CreateOptsFile(int argc, char *argv[]);
292 NON_EXEC_STATIC void SSDataBaseInit(int xlop);
293 static pid_t SSDataBase(int xlop);
295 postmaster_error(const char *fmt,...)
296 /* This lets gcc check the format string for consistency. */
297 __attribute__((format(printf, 1, 2)));
301 pid_t win32_forkexec(const char* path, char *argv[]);
303 static void win32_AddChild(pid_t pid, HANDLE handle);
304 static void win32_RemoveChild(pid_t pid);
305 static pid_t win32_waitpid(int *exitstatus);
306 static DWORD WINAPI win32_sigchld_waiter(LPVOID param);
308 static pid_t *win32_childPIDArray;
309 static HANDLE *win32_childHNDArray;
310 static unsigned long win32_numChildren = 0;
313 static pid_t Backend_forkexec(Port *port);
315 static unsigned long tmpBackendFileNum = 0;
316 void read_backend_variables(unsigned long id, Port *port);
317 static bool write_backend_variables(Port *port);
319 size_t ShmemBackendArraySize(void);
320 void ShmemBackendArrayAllocation(void);
321 static void ShmemBackendArrayAdd(Backend *bn);
322 static void ShmemBackendArrayRemove(pid_t pid);
325 #define StartupDataBase() SSDataBase(BS_XLOG_STARTUP)
326 #define CheckPointDataBase() SSDataBase(BS_XLOG_CHECKPOINT)
327 #define StartBackgroundWriter() SSDataBase(BS_XLOG_BGWRITER)
328 #define ShutdownDataBase() SSDataBase(BS_XLOG_SHUTDOWN)
332 checkDataDir(const char *checkdir)
334 char path[MAXPGPATH];
336 struct stat stat_buf;
338 if (checkdir == NULL)
341 gettext("%s does not know where to find the database system data.\n"
342 "You must specify the directory that contains the database system\n"
343 "either by specifying the -D invocation option or by setting the\n"
344 "PGDATA environment variable.\n"),
349 if (stat(checkdir, &stat_buf) == -1)
353 (errcode_for_file_access(),
354 errmsg("data directory \"%s\" does not exist",
358 (errcode_for_file_access(),
359 errmsg("could not read permissions of directory \"%s\": %m",
364 * Check if the directory has group or world access. If so, reject.
366 * XXX temporarily suppress check when on Windows, because there may not
367 * be proper support for Unix-y file permissions. Need to think of a
368 * reasonable check to apply on Windows.
370 #if !defined(__CYGWIN__) && !defined(WIN32)
371 if (stat_buf.st_mode & (S_IRWXG | S_IRWXO))
373 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
374 errmsg("data directory \"%s\" has group or world access",
376 errdetail("Permissions should be u=rwx (0700).")));
379 /* Look for PG_VERSION before looking for pg_control */
380 ValidatePgVersion(checkdir);
382 snprintf(path, sizeof(path), "%s/global/pg_control", checkdir);
384 fp = AllocateFile(path, PG_BINARY_R);
388 gettext("%s: could not find the database system\n"
389 "Expected to find it in the directory \"%s\",\n"
390 "but could not open file \"%s\": %s\n"),
391 progname, checkdir, path, strerror(errno));
398 #ifdef USE_RENDEZVOUS
400 /* reg_reply -- empty callback function for DNSServiceRegistrationCreate() */
402 reg_reply(DNSServiceRegistrationReplyErrorType errorCode, void *context)
409 PostmasterMain(int argc, char *argv[])
413 char original_extraoptions[MAXPGPATH];
414 char *potential_DataDir = NULL;
417 *original_extraoptions = '\0';
421 IsPostmasterEnvironment = true;
424 * Catch standard options before doing much else. This even works on
425 * systems without getopt_long.
429 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
434 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
436 puts("postmaster (PostgreSQL) " PG_VERSION);
442 * for security, no dir or file created can be group or other
445 umask((mode_t) 0077);
447 MyProcPid = PostmasterPid = getpid();
450 * Fire up essential subsystems: memory management
455 * By default, palloc() requests in the postmaster will be allocated
456 * in the PostmasterContext, which is space that can be recycled by
457 * backends. Allocated data that needs to be available to backends
458 * should be allocated in TopMemoryContext.
460 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
462 ALLOCSET_DEFAULT_MINSIZE,
463 ALLOCSET_DEFAULT_INITSIZE,
464 ALLOCSET_DEFAULT_MAXSIZE);
465 MemoryContextSwitchTo(PostmasterContext);
467 IgnoreSystemIndexes(false);
472 InitializeGUCOptions();
474 potential_DataDir = getenv("PGDATA"); /* default value */
478 while ((opt = getopt(argc, argv, "A:a:B:b:c:D:d:Fh:ik:lm:MN:no:p:Ss-:")) != -1)
483 #ifdef USE_ASSERT_CHECKING
484 SetConfigOption("debug_assertions", optarg, PGC_POSTMASTER, PGC_S_ARGV);
486 postmaster_error("assert checking is not compiled in");
490 /* Can no longer set authentication method. */
493 SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
496 /* Can no longer set the backend executable file to use. */
499 potential_DataDir = optarg;
503 /* Turn on debugging for the postmaster. */
504 char *debugstr = palloc(strlen("debug") + strlen(optarg) + 1);
506 sprintf(debugstr, "debug%s", optarg);
507 SetConfigOption("log_min_messages", debugstr,
508 PGC_POSTMASTER, PGC_S_ARGV);
510 debug_flag = atoi(optarg);
514 SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
517 SetConfigOption("virtual_host", optarg, PGC_POSTMASTER, PGC_S_ARGV);
520 SetConfigOption("tcpip_socket", "true", PGC_POSTMASTER, PGC_S_ARGV);
523 SetConfigOption("unix_socket_directory", optarg, PGC_POSTMASTER, PGC_S_ARGV);
527 SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
531 /* Multiplexed backends no longer supported. */
536 * ignore this flag. This may be passed in because the
537 * program was run as 'postgres -M' instead of
542 /* The max number of backends to start. */
543 SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
546 /* Don't reinit shared mem after abnormal exit */
552 * Other options to pass to the backend on the command
553 * line -- useful only for debugging.
555 strcat(ExtraOptions, " ");
556 strcat(ExtraOptions, optarg);
557 strcpy(original_extraoptions, optarg);
560 SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
565 * Start in 'S'ilent mode (disassociate from controlling
566 * tty). You may also think of this as 'S'ysV mode since
567 * it's most badly needed on SysV-derived systems like
570 SetConfigOption("silent_mode", "true", PGC_POSTMASTER, PGC_S_ARGV);
575 * In the event that some backend dumps core, send
576 * SIGSTOP, rather than SIGQUIT, to all its peers. This
577 * lets the wily post_hacker collect core dumps from
588 ParseLongOption(optarg, &name, &value);
593 (errcode(ERRCODE_SYNTAX_ERROR),
594 errmsg("--%s requires a value",
598 (errcode(ERRCODE_SYNTAX_ERROR),
599 errmsg("-c %s requires a value",
603 SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
612 gettext("Try \"%s --help\" for more information.\n"),
619 * Postmaster accepts no non-option switch arguments.
623 postmaster_error("invalid argument: \"%s\"", argv[optind]);
625 gettext("Try \"%s --help\" for more information.\n"),
631 * Now we can set the data directory, and then read postgresql.conf.
633 checkDataDir(potential_DataDir); /* issues error messages */
634 SetDataDir(potential_DataDir);
636 ProcessConfigFile(PGC_POSTMASTER);
638 write_nondefault_variables(PGC_POSTMASTER);
642 * Check for invalid combinations of GUC settings.
644 if (NBuffers < 2 * MaxBackends || NBuffers < 16)
647 * Do not accept -B so small that backends are likely to starve
648 * for lack of buffers. The specific choices here are somewhat
651 postmaster_error("the number of buffers (-B) must be at least twice the number of allowed connections (-N) and at least 16");
655 if (ReservedBackends >= MaxBackends)
657 postmaster_error("superuser_reserved_connections must be less than max_connections");
662 * Other one-time internal sanity checks can go here.
664 if (!CheckDateTokenTables())
666 postmaster_error("invalid datetoken tables, please fix");
671 * Now that we are done processing the postmaster arguments, reset
672 * getopt(3) library so that it will work correctly in subprocesses.
675 #ifdef HAVE_INT_OPTRESET
676 optreset = 1; /* some systems need this too */
679 /* For debugging: display postmaster environment */
681 extern char **environ;
685 (errmsg_internal("%s: PostmasterMain: initial environ dump:",
688 (errmsg_internal("-----------------------------------------")));
689 for (p = environ; *p; ++p)
691 (errmsg_internal("\t%s", *p)));
693 (errmsg_internal("-----------------------------------------")));
697 * On some systems our dynloader code needs the executable's pathname.
699 if (FindExec(pg_pathname, progname, "postgres") < 0)
701 (errmsg("%s: could not locate postgres executable",
705 * Initialize SSL library, if specified.
708 if (EnableSSL && !NetServer)
710 postmaster_error("TCP/IP connections must be enabled for SSL");
718 * process any libraries that should be preloaded and optionally
721 if (preload_libraries_string)
722 process_preload_libraries(preload_libraries_string);
725 * Fork away from controlling terminal, if -S specified.
727 * Must do this before we grab any interlock files, else the interlocks
728 * will show the wrong PID.
731 pmdaemonize(argc, argv);
734 * Create lockfile for data directory.
736 * We want to do this before we try to grab the input sockets, because
737 * the data directory interlock is more reliable than the socket-file
738 * interlock (thanks to whoever decided to put socket files in /tmp
739 * :-(). For the same reason, it's best to grab the TCP socket before
742 CreateDataDirLockFile(DataDir, true);
745 * Remove old temporary files. At this point there can be no other
746 * Postgres processes running in this directory, so this should be
752 * Establish input sockets.
754 for (i = 0; i < MAXLISTEN; i++)
755 ListenSocket[i] = -1;
759 if (VirtualHost && VirtualHost[0])
765 curhost = VirtualHost;
768 while (*curhost == ' ') /* skip any extra spaces */
770 if (*curhost == '\0')
772 endptr = strchr(curhost, ' ');
778 status = StreamServerPort(AF_UNSPEC, curhost,
779 (unsigned short) PostPortNumber,
781 ListenSocket, MAXLISTEN);
782 if (status != STATUS_OK)
784 (errmsg("could not create listen socket for \"%s\"",
789 curhost = endptr + 1;
797 status = StreamServerPort(AF_UNSPEC, NULL,
798 (unsigned short) PostPortNumber,
800 ListenSocket, MAXLISTEN);
801 if (status != STATUS_OK)
803 (errmsg("could not create TCP/IP listen socket")));
806 #ifdef USE_RENDEZVOUS
807 if (rendezvous_name != NULL)
809 DNSServiceRegistrationCreate(rendezvous_name,
812 htonl(PostPortNumber),
814 (DNSServiceRegistrationReply) reg_reply,
820 #ifdef HAVE_UNIX_SOCKETS
821 status = StreamServerPort(AF_UNIX, NULL,
822 (unsigned short) PostPortNumber,
824 ListenSocket, MAXLISTEN);
825 if (status != STATUS_OK)
827 (errmsg("could not create Unix-domain socket")));
833 * Set up shared memory and semaphores.
835 reset_shared(PostPortNumber);
838 * Estimate number of openable files. This must happen after setting up
839 * semaphores, because on some platforms semaphores count as open files.
844 * Initialize the list of active backends.
846 BackendList = DLNewList();
850 * Initialize the child pid/HANDLE arrays
852 win32_childPIDArray = (pid_t*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(pid_t));
853 win32_childHNDArray = (HANDLE*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(HANDLE));
854 if (!win32_childPIDArray || !win32_childHNDArray)
856 (errcode(ERRCODE_OUT_OF_MEMORY),
857 errmsg("out of memory")));
861 * Record postmaster options. We delay this till now to avoid
862 * recording bogus options (eg, NBuffers too high for available
865 if (!CreateOptsFile(argc, argv))
869 * Set up signal handlers for the postmaster process.
871 * CAUTION: when changing this list, check for side-effects on the signal
872 * handling setup of child processes. See tcop/postgres.c,
873 * bootstrap/bootstrap.c, and postmaster/pgstat.c.
876 PG_SETMASK(&BlockSig);
878 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
879 * children do same */
880 pqsignal(SIGINT, pmdie); /* send SIGTERM and ShutdownDataBase */
881 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
882 pqsignal(SIGTERM, pmdie); /* wait for children and ShutdownDataBase */
883 pqsignal(SIGALRM, SIG_IGN); /* ignored */
884 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
885 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
886 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
887 pqsignal(SIGCHLD, reaper); /* handle child termination */
888 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
889 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
890 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
892 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
896 * Reset whereToSendOutput from Debug (its starting state) to None.
897 * This prevents ereport from sending log messages to stderr unless
898 * the syslog/stderr switch permits. We don't do this until the
899 * postmaster is fully launched, since startup failures may as well be
900 * reported to stderr.
902 whereToSendOutput = None;
905 * On many platforms, the first call of localtime() incurs significant
906 * overhead to load timezone info from the system configuration files.
907 * By doing it once in the postmaster, we avoid having to do it in
908 * every started child process. The savings are not huge, but they
912 time_t now = time(NULL);
914 (void) localtime(&now);
918 * Initialize and try to startup the statistics collector process
924 * Load cached files for client authentication.
932 * We're ready to rock and roll...
934 StartupPID = StartupDataBase();
936 status = ServerLoop();
939 * ServerLoop probably shouldn't ever return, but if it does, close
942 ExitPostmaster(status != STATUS_OK);
944 return 0; /* not reached */
948 pmdaemonize(int argc, char *argv[])
954 struct itimerval prof_itimer;
958 /* see comments in BackendRun */
959 getitimer(ITIMER_PROF, &prof_itimer);
963 /* FIXME: [fork/exec] to be implemented? */
968 if (pid == (pid_t) -1)
970 postmaster_error("could not fork background process: %s",
976 /* Parent should just exit, without doing any atexit cleanup */
981 setitimer(ITIMER_PROF, &prof_itimer, NULL);
984 MyProcPid = getpid(); /* reset MyProcPid to child */
986 /* GH: If there's no setsid(), we hopefully don't need silent mode.
987 * Until there's a better solution.
992 postmaster_error("could not dissociate from controlling TTY: %s",
997 i = open(NULL_DEV, O_RDWR | PG_BINARY);
1007 * Print out help message
1010 usage(const char *progname)
1012 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
1013 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
1014 printf(gettext("Options:\n"));
1015 #ifdef USE_ASSERT_CHECKING
1016 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
1018 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
1019 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
1020 printf(gettext(" -d 1-5 debugging level\n"));
1021 printf(gettext(" -D DATADIR database directory\n"));
1022 printf(gettext(" -F turn fsync off\n"));
1023 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
1024 printf(gettext(" -i enable TCP/IP connections\n"));
1025 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
1027 printf(gettext(" -l enable SSL connections\n"));
1029 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
1030 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
1031 printf(gettext(" -p PORT port number to listen on\n"));
1032 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1033 printf(gettext(" --help show this help, then exit\n"));
1034 printf(gettext(" --version output version information, then exit\n"));
1036 printf(gettext("\nDeveloper options:\n"));
1037 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1038 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1040 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1041 "configuration settings and how to set them on the command line or in\n"
1042 "the configuration file.\n\n"
1043 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1056 gettimeofday(&now, &tz);
1058 nSockets = initMasks(&readmask);
1064 struct timeval timeout;
1067 * The timeout for the select() below is normally set on the basis
1068 * of the time to the next checkpoint. However, if for some
1069 * reason we don't have a next-checkpoint time, time out after 60
1070 * seconds. This keeps checkpoint scheduling from locking up when
1071 * we get new connection requests infrequently (since we are
1072 * likely to detect checkpoint completion just after enabling
1073 * signals below, after we've already made the decision about how
1074 * long to wait this time).
1076 timeout.tv_sec = 60;
1077 timeout.tv_usec = 0;
1079 if (CheckPointPID == 0 && checkpointed &&
1080 StartupPID == 0 && Shutdown == NoShutdown &&
1081 !FatalError && random_seed != 0)
1083 time_t now = time(NULL);
1085 if (CheckPointTimeout + checkpointed > now)
1088 * Not time for checkpoint yet, so set select timeout
1090 timeout.tv_sec = CheckPointTimeout + checkpointed - now;
1094 /* Time to make the checkpoint... */
1095 CheckPointPID = CheckPointDataBase();
1098 * if fork failed, schedule another try at 0.1 normal
1101 if (CheckPointPID == 0)
1103 timeout.tv_sec = CheckPointTimeout / 10;
1104 checkpointed = now + timeout.tv_sec - CheckPointTimeout;
1110 * If no background writer process is running and we should
1111 * do background writing, start one. It doesn't matter if
1112 * this fails, we'll just try again later.
1114 if (BgWriterPID == 0 && BgWriterPercent > 0 &&
1115 StartupPID == 0 && Shutdown == NoShutdown &&
1116 !FatalError && random_seed != 0)
1118 BgWriterPID = StartBackgroundWriter();
1122 * Wait for something to happen.
1124 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1126 PG_SETMASK(&UnBlockSig);
1128 if (select(nSockets, &rmask, NULL, NULL, &timeout) < 0)
1130 PG_SETMASK(&BlockSig);
1131 if (errno == EINTR || errno == EWOULDBLOCK)
1134 (errcode_for_socket_access(),
1135 errmsg("select() failed in postmaster: %m")));
1136 return STATUS_ERROR;
1140 * Block all signals until we wait again. (This makes it safe for
1141 * our signal handlers to do nontrivial work.)
1143 PG_SETMASK(&BlockSig);
1146 * Select a random seed at the time of first receiving a request.
1148 while (random_seed == 0)
1150 gettimeofday(&later, &tz);
1153 * We are not sure how much precision is in tv_usec, so we
1154 * swap the nibbles of 'later' and XOR them with 'now'. On the
1155 * off chance that the result is 0, we loop until it isn't.
1157 random_seed = now.tv_usec ^
1158 ((later.tv_usec << 16) |
1159 ((later.tv_usec >> 16) & 0xffff));
1163 * New connection pending on any of our sockets? If so, fork a
1164 * child process to deal with it.
1166 for (i = 0; i < MAXLISTEN; i++)
1168 if (ListenSocket[i] == -1)
1170 if (FD_ISSET(ListenSocket[i], &rmask))
1172 port = ConnCreate(ListenSocket[i]);
1175 BackendStartup(port);
1178 * We no longer need the open socket or port structure
1181 StreamClose(port->sock);
1187 /* If we have lost the stats collector, try to start a new one */
1188 if (!pgstat_is_running)
1195 * Initialise the masks for select() for the ports
1196 * we are listening on. Return the number of sockets to listen on.
1200 initMasks(fd_set *rmask)
1207 for (i = 0; i < MAXLISTEN; i++)
1209 int fd = ListenSocket[i];
1223 * Read the startup packet and do something according to it.
1225 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1226 * not return at all.
1228 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1229 * if that's what you want. Return STATUS_ERROR if you don't want to
1230 * send anything to the client, which would typically be appropriate
1231 * if we detect a communications failure.)
1234 ProcessStartupPacket(Port *port, bool SSLdone)
1238 ProtocolVersion proto;
1239 MemoryContext oldcontext;
1241 if (pq_getbytes((char *) &len, 4) == EOF)
1244 * EOF after SSLdone probably means the client didn't like our
1245 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1246 * so don't clutter the log with a complaint.
1250 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1251 errmsg("incomplete startup packet")));
1252 return STATUS_ERROR;
1258 if (len < (int32) sizeof(ProtocolVersion) ||
1259 len > MAX_STARTUP_PACKET_LENGTH)
1262 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1263 errmsg("invalid length of startup packet")));
1264 return STATUS_ERROR;
1268 * Allocate at least the size of an old-style startup packet, plus one
1269 * extra byte, and make sure all are zeroes. This ensures we will
1270 * have null termination of all strings, in both fixed- and
1271 * variable-length packet layouts.
1273 if (len <= (int32) sizeof(StartupPacket))
1274 buf = palloc0(sizeof(StartupPacket) + 1);
1276 buf = palloc0(len + 1);
1278 if (pq_getbytes(buf, len) == EOF)
1281 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1282 errmsg("incomplete startup packet")));
1283 return STATUS_ERROR;
1287 * The first field is either a protocol version number or a special
1290 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1292 if (proto == CANCEL_REQUEST_CODE)
1294 processCancelRequest(port, buf);
1295 return 127; /* XXX */
1298 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1303 /* No SSL when disabled or on Unix sockets */
1304 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1307 SSLok = 'S'; /* Support for SSL */
1309 SSLok = 'N'; /* No support for SSL */
1311 if (send(port->sock, &SSLok, 1, 0) != 1)
1314 (errcode_for_socket_access(),
1315 errmsg("failed to send SSL negotiation response: %m")));
1316 return STATUS_ERROR; /* close the connection */
1320 if (SSLok == 'S' && secure_open_server(port) == -1)
1321 return STATUS_ERROR;
1323 /* regular startup packet, cancel, etc packet should follow... */
1324 /* but not another SSL negotiation request */
1325 return ProcessStartupPacket(port, true);
1328 /* Could add additional special packet types here */
1331 * Set FrontendProtocol now so that ereport() knows what format to
1332 * send if we fail during startup.
1334 FrontendProtocol = proto;
1336 /* Check we can handle the protocol the frontend is using. */
1338 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1339 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1340 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1341 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1343 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1344 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1345 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1346 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1347 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1348 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1351 * Now fetch parameters out of startup packet and save them into the
1352 * Port structure. All data structures attached to the Port struct
1353 * must be allocated in TopMemoryContext so that they won't disappear
1354 * when we pass them to PostgresMain (see BackendRun). We need not
1355 * worry about leaking this storage on failure, since we aren't in the
1356 * postmaster process anymore.
1358 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1360 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1362 int32 offset = sizeof(ProtocolVersion);
1365 * Scan packet body for name/option pairs. We can assume any
1366 * string beginning within the packet body is null-terminated,
1367 * thanks to zeroing extra byte above.
1369 port->guc_options = NIL;
1371 while (offset < len)
1373 char *nameptr = ((char *) buf) + offset;
1377 if (*nameptr == '\0')
1378 break; /* found packet terminator */
1379 valoffset = offset + strlen(nameptr) + 1;
1380 if (valoffset >= len)
1381 break; /* missing value, will complain below */
1382 valptr = ((char *) buf) + valoffset;
1384 if (strcmp(nameptr, "database") == 0)
1385 port->database_name = pstrdup(valptr);
1386 else if (strcmp(nameptr, "user") == 0)
1387 port->user_name = pstrdup(valptr);
1388 else if (strcmp(nameptr, "options") == 0)
1389 port->cmdline_options = pstrdup(valptr);
1392 /* Assume it's a generic GUC option */
1393 port->guc_options = lappend(port->guc_options,
1395 port->guc_options = lappend(port->guc_options,
1398 offset = valoffset + strlen(valptr) + 1;
1402 * If we didn't find a packet terminator exactly at the end of the
1403 * given packet length, complain.
1405 if (offset != len - 1)
1407 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1408 errmsg("invalid startup packet layout: expected terminator as last byte")));
1413 * Get the parameters from the old-style, fixed-width-fields
1414 * startup packet as C strings. The packet destination was
1415 * cleared first so a short packet has zeros silently added. We
1416 * have to be prepared to truncate the pstrdup result for oversize
1419 StartupPacket *packet = (StartupPacket *) buf;
1421 port->database_name = pstrdup(packet->database);
1422 if (strlen(port->database_name) > sizeof(packet->database))
1423 port->database_name[sizeof(packet->database)] = '\0';
1424 port->user_name = pstrdup(packet->user);
1425 if (strlen(port->user_name) > sizeof(packet->user))
1426 port->user_name[sizeof(packet->user)] = '\0';
1427 port->cmdline_options = pstrdup(packet->options);
1428 if (strlen(port->cmdline_options) > sizeof(packet->options))
1429 port->cmdline_options[sizeof(packet->options)] = '\0';
1430 port->guc_options = NIL;
1433 /* Check a user name was given. */
1434 if (port->user_name == NULL || port->user_name[0] == '\0')
1436 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1437 errmsg("no PostgreSQL user name specified in startup packet")));
1439 /* The database defaults to the user name. */
1440 if (port->database_name == NULL || port->database_name[0] == '\0')
1441 port->database_name = pstrdup(port->user_name);
1443 if (Db_user_namespace)
1446 * If user@, it is a global user, remove '@'. We only want to do
1447 * this if there is an '@' at the end and no earlier in the user
1448 * string or they may fake as a local user of another database
1449 * attaching to this database.
1451 if (strchr(port->user_name, '@') ==
1452 port->user_name + strlen(port->user_name) - 1)
1453 *strchr(port->user_name, '@') = '\0';
1456 /* Append '@' and dbname */
1459 db_user = palloc(strlen(port->user_name) +
1460 strlen(port->database_name) + 2);
1461 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1462 port->user_name = db_user;
1467 * Truncate given database and user names to length of a Postgres
1468 * name. This avoids lookup failures when overlength names are given.
1470 if (strlen(port->database_name) >= NAMEDATALEN)
1471 port->database_name[NAMEDATALEN - 1] = '\0';
1472 if (strlen(port->user_name) >= NAMEDATALEN)
1473 port->user_name[NAMEDATALEN - 1] = '\0';
1476 * Done putting stuff in TopMemoryContext.
1478 MemoryContextSwitchTo(oldcontext);
1481 * If we're going to reject the connection due to database state, say
1482 * so now instead of wasting cycles on an authentication exchange.
1483 * (This also allows a pg_ping utility to be written.)
1485 switch (port->canAcceptConnections)
1489 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1490 errmsg("the database system is starting up")));
1494 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1495 errmsg("the database system is shutting down")));
1499 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1500 errmsg("the database system is in recovery mode")));
1504 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1505 errmsg("sorry, too many clients already")));
1517 * The client has sent a cancel request packet, not a normal
1518 * start-a-new-connection packet. Perform the necessary processing.
1519 * Nothing is sent back to the client.
1522 processCancelRequest(Port *port, void *pkt)
1524 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1526 long cancelAuthCode;
1528 #ifndef EXEC_BACKEND
1534 backendPID = (int) ntohl(canc->backendPID);
1535 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1537 if (backendPID == CheckPointPID)
1540 (errmsg_internal("ignoring cancel request for checkpoint process %d",
1544 else if (backendPID == BgWriterPID)
1547 (errmsg_internal("ignoring cancel request for bgwriter process %d",
1552 /* See if we have a matching backend */
1553 #ifndef EXEC_BACKEND
1554 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1556 bp = (Backend *) DLE_VAL(curr);
1558 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1560 bp = (Backend*) &ShmemBackendArray[i];
1562 if (bp->pid == backendPID)
1564 if (bp->cancel_key == cancelAuthCode)
1566 /* Found a match; signal that backend to cancel current op */
1568 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1570 kill(bp->pid, SIGINT);
1573 /* Right PID, wrong key: no way, Jose */
1575 (errmsg_internal("bad key in cancel request for process %d",
1581 /* No matching backend */
1583 (errmsg_internal("bad pid in cancel request for process %d",
1588 * canAcceptConnections --- check to see if database state allows connections.
1590 static enum CAC_state
1591 canAcceptConnections(void)
1593 /* Can't start backends when in startup/shutdown/recovery state. */
1594 if (Shutdown > NoShutdown)
1595 return CAC_SHUTDOWN;
1599 return CAC_RECOVERY;
1602 * Don't start too many children.
1604 * We allow more connections than we can have backends here because some
1605 * might still be authenticating; they might fail auth, or some
1606 * existing backend might exit before the auth cycle is completed. The
1607 * exact MaxBackends limit is enforced when a new backend tries to
1608 * join the shared-inval backend array.
1610 if (CountChildren() >= 2 * MaxBackends)
1618 * ConnCreate -- create a local connection data structure
1621 ConnCreate(int serverFd)
1625 if (!(port = (Port *) calloc(1, sizeof(Port))))
1628 (errcode(ERRCODE_OUT_OF_MEMORY),
1629 errmsg("out of memory")));
1633 if (StreamConnection(serverFd, port) != STATUS_OK)
1635 StreamClose(port->sock);
1642 * Precompute password salt values to use for this connection.
1643 * It's slightly annoying to do this long in advance of knowing
1644 * whether we'll need 'em or not, but we must do the random()
1645 * calls before we fork, not after. Else the postmaster's random
1646 * sequence won't get advanced, and all backends would end up
1647 * using the same salt...
1649 RandomSalt(port->cryptSalt, port->md5Salt);
1657 * ConnFree -- free a local connection data structure
1660 ConnFree(Port *conn)
1670 * ClosePostmasterPorts -- close all the postmaster's open sockets
1672 * This is called during child process startup to release file descriptors
1673 * that are not needed by that child process. The postmaster still has
1674 * them open, of course.
1677 ClosePostmasterPorts(bool pgstat_too)
1681 /* Close the listen sockets */
1682 for (i = 0; i < MAXLISTEN; i++)
1684 if (ListenSocket[i] != -1)
1686 StreamClose(ListenSocket[i]);
1687 ListenSocket[i] = -1;
1691 /* Close pgstat control sockets, unless we're starting pgstat itself */
1693 pgstat_close_sockets();
1698 * reset_shared -- reset shared memory and semaphores
1701 reset_shared(unsigned short port)
1704 * Create or re-create shared memory and semaphores.
1706 * Note: in each "cycle of life" we will normally assign the same IPC
1707 * keys (if using SysV shmem and/or semas), since the port number is
1708 * used to determine IPC keys. This helps ensure that we will clean
1709 * up dead IPC objects if the postmaster crashes and is restarted.
1711 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1716 * SIGHUP -- reread config files, and tell children to do same
1719 SIGHUP_handler(SIGNAL_ARGS)
1721 int save_errno = errno;
1723 PG_SETMASK(&BlockSig);
1725 if (Shutdown <= SmartShutdown)
1728 (errmsg("received SIGHUP, reloading configuration files")));
1729 ProcessConfigFile(PGC_SIGHUP);
1731 write_nondefault_variables(PGC_SIGHUP);
1733 SignalChildren(SIGHUP);
1738 * Tell the background writer to terminate so that we
1739 * will start a new one with a possibly changed config
1741 if (BgWriterPID != 0)
1742 kill(BgWriterPID, SIGTERM);
1745 PG_SETMASK(&UnBlockSig);
1753 * pmdie -- signal handler for processing various postmaster signals.
1758 int save_errno = errno;
1760 PG_SETMASK(&BlockSig);
1763 (errmsg_internal("postmaster received signal %d",
1764 postgres_signal_arg)));
1766 switch (postgres_signal_arg)
1773 * Wait for children to end their work and ShutdownDataBase.
1775 if (Shutdown >= SmartShutdown)
1777 Shutdown = SmartShutdown;
1779 (errmsg("received smart shutdown request")));
1781 /* Must tell bgwriter to quit, or it never will... */
1782 if (BgWriterPID != 0)
1783 kill(BgWriterPID, SIGTERM);
1785 if (DLGetHead(BackendList)) /* let reaper() handle this */
1789 * No children left. Shutdown data base system.
1791 if (StartupPID > 0 || FatalError) /* let reaper() handle
1794 if (ShutdownPID > 0)
1796 elog(PANIC, "shutdown process %d already running",
1801 ShutdownPID = ShutdownDataBase();
1809 * Abort all children with SIGTERM (rollback active transactions
1810 * and exit) and ShutdownDataBase when they are gone.
1812 if (Shutdown >= FastShutdown)
1814 Shutdown = FastShutdown;
1816 (errmsg("received fast shutdown request")));
1818 if (DLGetHead(BackendList))
1823 (errmsg("aborting any active transactions")));
1824 SignalChildren(SIGTERM);
1825 /* reaper() does the rest */
1831 * No children left. Shutdown data base system.
1833 * Unlike the previous case, it is not an error for the shutdown
1834 * process to be running already (we could get SIGTERM followed
1835 * shortly later by SIGINT).
1837 if (StartupPID > 0 || FatalError) /* let reaper() handle
1840 if (ShutdownPID == 0)
1841 ShutdownPID = ShutdownDataBase();
1847 * Immediate Shutdown:
1849 * abort all children with SIGQUIT and exit without attempt to
1850 * properly shutdown data base system.
1853 (errmsg("received immediate shutdown request")));
1854 if (ShutdownPID > 0)
1855 kill(ShutdownPID, SIGQUIT);
1857 kill(StartupPID, SIGQUIT);
1858 if (DLGetHead(BackendList))
1859 SignalChildren(SIGQUIT);
1864 PG_SETMASK(&UnBlockSig);
1870 * Reaper -- signal handler to cleanup after a backend (child) dies.
1875 int save_errno = errno;
1878 int status; /* backend exit status */
1881 union wait status; /* backend exit status */
1885 int pid; /* process id of dead backend */
1887 PG_SETMASK(&BlockSig);
1890 (errmsg_internal("reaping dead processes")));
1892 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1894 exitstatus = status;
1897 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1899 exitstatus = status.w_status;
1901 while ((pid = win32_waitpid(&exitstatus)) > 0)
1904 * We need to do this here, and not in CleanupProc, since this
1905 * is to be called on all children when we are done with them.
1906 * Could move to LogChildExit, but that seems like asking for
1909 win32_RemoveChild(pid);
1914 * Check if this child was the statistics collector. If so, try to
1915 * start a new one. (If fail, we'll try again in future cycles of
1918 if (pgstat_ispgstat(pid))
1920 LogChildExit(LOG, gettext("statistics collector process"),
1927 * Check if this child was a shutdown or startup process.
1929 if (ShutdownPID > 0 && pid == ShutdownPID)
1931 if (exitstatus != 0)
1933 LogChildExit(LOG, gettext("shutdown process"),
1937 /* Normal postmaster exit is here */
1941 if (StartupPID > 0 && pid == StartupPID)
1943 if (exitstatus != 0)
1945 LogChildExit(LOG, gettext("startup process"),
1948 (errmsg("aborting startup due to startup process failure")));
1954 * Startup succeeded - remember its ID and RedoRecPtr.
1956 * NB: this MUST happen before we fork a checkpoint or shutdown
1957 * subprocess, else they will have wrong local ThisStartUpId.
1961 FatalError = false; /* done with recovery */
1964 * Arrange for first checkpoint to occur after standard delay.
1967 checkpointed = time(NULL);
1970 * Go to shutdown mode if a shutdown request was pending.
1972 if (Shutdown > NoShutdown)
1974 if (ShutdownPID > 0)
1976 elog(PANIC, "startup process %d died while shutdown process %d already running",
1977 pid, (int) ShutdownPID);
1980 ShutdownPID = ShutdownDataBase();
1987 * Else do standard child cleanup.
1989 CleanupProc(pid, exitstatus);
1991 } /* loop over pending child-death reports */
1996 * Wait for all children exit, then reset shmem and
1999 if (DLGetHead(BackendList) || StartupPID > 0 || ShutdownPID > 0)
2002 (errmsg("all server processes terminated; reinitializing")));
2005 reset_shared(PostPortNumber);
2007 StartupPID = StartupDataBase();
2012 if (Shutdown > NoShutdown)
2014 if (DLGetHead(BackendList))
2016 if (StartupPID > 0 || ShutdownPID > 0)
2018 ShutdownPID = ShutdownDataBase();
2022 PG_SETMASK(&UnBlockSig);
2028 * CleanupProc -- cleanup after terminated backend.
2030 * Remove all local state associated with backend.
2033 CleanupProc(int pid,
2034 int exitstatus) /* child's exit status. */
2040 LogChildExit(DEBUG2, gettext("child process"), pid, exitstatus);
2043 * If a backend dies in an ugly way (i.e. exit status not 0) then we
2044 * must signal all other backends to quickdie. If exit status is zero
2045 * we assume everything is hunky dory and simply remove the backend
2046 * from the active backend list.
2048 if (exitstatus == 0)
2050 curr = DLGetHead(BackendList);
2053 bp = (Backend *) DLE_VAL(curr);
2057 ShmemBackendArrayRemove(bp->pid);
2064 curr = DLGetSucc(curr);
2067 if (pid == CheckPointPID)
2072 checkpointed = time(NULL);
2073 /* Update RedoRecPtr for future child backends */
2074 GetSavedRedoRecPtr();
2077 else if (pid == BgWriterPID)
2085 /* below here we're dealing with a non-normal exit */
2087 /* Make log entry unless we did so already */
2091 (pid == CheckPointPID) ? gettext("checkpoint process") :
2092 (pid == BgWriterPID) ? gettext("bgwriter process") :
2093 gettext("server process"),
2096 (errmsg("terminating any other active server processes")));
2099 curr = DLGetHead(BackendList);
2102 next = DLGetSucc(curr);
2103 bp = (Backend *) DLE_VAL(curr);
2107 * This backend is still alive. Unless we did so already,
2108 * tell it to commit hara-kiri.
2110 * SIGQUIT is the special signal that says exit without proc_exit
2111 * and let the user know what's going on. But if SendStop is
2112 * set (-s on command line), then we send SIGSTOP instead, so
2113 * that we can get core dumps from all backends by hand.
2118 (errmsg_internal("sending %s to process %d",
2119 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2121 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2127 * Found entry for freshly-dead backend, so remove it.
2130 ShmemBackendArrayRemove(bp->pid);
2139 if (pid == CheckPointPID)
2144 else if (pid == BgWriterPID)
2151 * Tell the collector about backend termination
2160 * Log the death of a child process.
2163 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2165 if (WIFEXITED(exitstatus))
2169 * translator: %s is a noun phrase describing a child process,
2170 * such as "server process"
2172 (errmsg("%s (PID %d) exited with exit code %d",
2173 procname, pid, WEXITSTATUS(exitstatus))));
2174 else if (WIFSIGNALED(exitstatus))
2178 * translator: %s is a noun phrase describing a child process,
2179 * such as "server process"
2181 (errmsg("%s (PID %d) was terminated by signal %d",
2182 procname, pid, WTERMSIG(exitstatus))));
2187 * translator: %s is a noun phrase describing a child process,
2188 * such as "server process"
2190 (errmsg("%s (PID %d) exited with unexpected status %d",
2191 procname, pid, exitstatus)));
2195 * Send a signal to all backend children.
2198 SignalChildren(int signal)
2204 curr = DLGetHead(BackendList);
2207 next = DLGetSucc(curr);
2208 bp = (Backend *) DLE_VAL(curr);
2210 if (bp->pid != MyProcPid)
2213 (errmsg_internal("sending signal %d to process %d",
2216 kill(bp->pid, signal);
2224 * BackendStartup -- start backend process
2226 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2229 BackendStartup(Port *port)
2231 Backend *bn; /* for backend cleanup */
2234 #ifdef LINUX_PROFILE
2235 struct itimerval prof_itimer;
2239 * Compute the cancel key that will be assigned to this backend. The
2240 * backend will have its own copy in the forked-off process' value of
2241 * MyCancelKey, so that it can transmit the key to the frontend.
2243 MyCancelKey = PostmasterRandom();
2246 * Make room for backend data structure. Better before the fork() so
2247 * we can handle failure cleanly.
2249 bn = (Backend *) malloc(sizeof(Backend));
2253 (errcode(ERRCODE_OUT_OF_MEMORY),
2254 errmsg("out of memory")));
2255 return STATUS_ERROR;
2259 * Flush stdio channels just before fork, to avoid double-output
2260 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2261 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2262 * coredump if we do. Presently stdout and stderr are the only stdio
2263 * output channels used by the postmaster, so fflush'ing them should
2269 #ifdef LINUX_PROFILE
2272 * Linux's fork() resets the profiling timer in the child process. If
2273 * we want to profile child processes then we need to save and restore
2274 * the timer setting. This is a waste of time if not profiling,
2275 * however, so only do it if commanded by specific -DLINUX_PROFILE
2278 getitimer(ITIMER_PROF, &prof_itimer);
2282 /* Specific beos actions before backend startup */
2283 beos_before_backend_startup();
2286 port->canAcceptConnections = canAcceptConnections();
2288 pid = Backend_forkexec(port);
2292 if (pid == 0) /* child */
2294 #ifdef LINUX_PROFILE
2295 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2299 /* Specific beos backend startup actions */
2300 beos_backend_startup();
2304 proc_exit(BackendRun(port));
2308 /* in parent, error */
2311 int save_errno = errno;
2314 /* Specific beos backend startup actions */
2315 beos_backend_startup_failed();
2320 (errmsg("could not fork new process for connection: %m")));
2321 report_fork_failure_to_client(port, save_errno);
2322 return STATUS_ERROR;
2325 /* in parent, normal */
2327 (errmsg_internal("forked new backend, pid=%d socket=%d",
2328 (int) pid, port->sock)));
2331 * Everything's been successful, it's safe to add this backend to our
2335 bn->cancel_key = MyCancelKey;
2337 ShmemBackendArrayAdd(bn);
2339 DLAddHead(BackendList, DLNewElem(bn));
2345 * Try to report backend fork() failure to client before we close the
2346 * connection. Since we do not care to risk blocking the postmaster on
2347 * this connection, we set the connection to non-blocking and try only once.
2349 * This is grungy special-purpose code; we cannot use backend libpq since
2350 * it's not up and running.
2353 report_fork_failure_to_client(Port *port, int errnum)
2357 /* Format the error message packet (always V2 protocol) */
2358 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2359 gettext("could not fork new process for connection: "),
2362 /* Set port to non-blocking. Don't do send() if this fails */
2363 if (!set_noblock(port->sock))
2366 send(port->sock, buffer, strlen(buffer) + 1, 0);
2371 * split_opts -- split a string of options and append it to an argv array
2373 * NB: the string is destructively modified!
2375 * Since no current POSTGRES arguments require any quoting characters,
2376 * we can use the simple-minded tactic of assuming each set of space-
2377 * delimited characters is a separate argv element.
2379 * If you don't like that, well, we *used* to pass the whole option string
2380 * as ONE argument to execl(), which was even less intelligent...
2383 split_opts(char **argv, int *argcp, char *s)
2387 while (isspace((unsigned char) *s))
2391 argv[(*argcp)++] = s;
2392 while (*s && !isspace((unsigned char) *s))
2401 * BackendInit/Run -- perform authentication [BackendInit], and if successful,
2402 * set up the backend's argument list [BackendRun] and invoke
2406 * Shouldn't return at all.
2407 * If PostgresMain() fails, return status.
2410 BackendInit(Port *port)
2415 char remote_host[NI_MAXHOST];
2416 char remote_port[NI_MAXSERV];
2418 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2420 ClientAuthInProgress = true; /* limit visibility of log messages */
2422 /* We don't want the postmaster's proc_exit() handlers */
2426 * Signal handlers setting is moved to tcop/postgres...
2429 /* save start time for end of session reporting */
2430 gettimeofday(&(port->session_start),NULL);
2432 /* set these to empty in case they are needed before we set them up */
2433 port->remote_host = "";
2434 port->remote_port = "";
2435 port->commandTag = "";
2437 /* Save port etc. for ps status */
2440 /* Reset MyProcPid to new backend's pid */
2441 MyProcPid = getpid();
2444 * Initialize libpq and enable reporting of ereport errors to the
2445 * client. Must do this now because authentication uses libpq to send
2448 pq_init(); /* initialize libpq to talk to client */
2449 whereToSendOutput = Remote; /* now safe to ereport to client */
2452 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2453 * during any client authentication related communication. Otherwise
2454 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2455 * if a buggy client blocks a backend during authentication.
2457 pqsignal(SIGTERM, authdie);
2458 pqsignal(SIGQUIT, authdie);
2459 pqsignal(SIGALRM, authdie);
2460 PG_SETMASK(&AuthBlockSig);
2463 * Get the remote host name and port for logging and status display.
2465 remote_host[0] = '\0';
2466 remote_port[0] = '\0';
2467 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2468 remote_host, sizeof(remote_host),
2469 remote_port, sizeof(remote_port),
2470 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2472 getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2473 remote_host, sizeof(remote_host),
2474 remote_port, sizeof(remote_port),
2475 NI_NUMERICHOST | NI_NUMERICSERV);
2478 if (Log_connections)
2480 (errmsg("connection received: host=%s port=%s",
2481 remote_host, remote_port)));
2485 /* modify remote_host for use in ps status */
2486 char tmphost[NI_MAXHOST];
2488 snprintf(tmphost, sizeof(tmphost), "%s(%s)", remote_host, remote_port);
2489 StrNCpy(remote_host, tmphost, sizeof(remote_host));
2493 * save remote_host and remote_port in port stucture
2495 port->remote_host = strdup(remote_host);
2496 port->remote_port = strdup(remote_port);
2499 * Ready to begin client interaction. We will give up and exit(0)
2500 * after a time delay, so that a broken client can't hog a connection
2501 * indefinitely. PreAuthDelay doesn't count against the time limit.
2503 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2504 elog(FATAL, "could not set timer for authorization timeout");
2507 * Receive the startup packet (which might turn out to be a cancel
2510 status = ProcessStartupPacket(port, false);
2512 if (status != STATUS_OK)
2516 * Now that we have the user and database name, we can set the process
2517 * title for ps. It's good to do this as early as possible in
2520 init_ps_display(port->user_name, port->database_name, remote_host);
2521 set_ps_display("authentication");
2524 * Now perform authentication exchange.
2526 ClientAuthentication(port); /* might not return, if failure */
2529 * Done with authentication. Disable timeout, and prevent
2530 * SIGTERM/SIGQUIT again until backend startup is complete.
2532 if (!disable_sig_alarm(false))
2533 elog(FATAL, "could not disable timer for authorization timeout");
2534 PG_SETMASK(&BlockSig);
2536 if (Log_connections)
2538 (errmsg("connection authorized: user=%s database=%s",
2539 port->user_name, port->database_name)));
2542 * Don't want backend to be able to see the postmaster random number
2543 * generator state. We have to clobber the static random_seed *and*
2544 * start a new random sequence in the random() library function.
2547 gettimeofday(&now, &tz);
2548 srandom((unsigned int) now.tv_usec);
2553 BackendRun(Port *port)
2563 * Let's clean up ourselves as the postmaster child, and
2564 * close the postmaster's other sockets
2566 ClosePostmasterPorts(true);
2569 * PreAuthDelay is a debugging aid for investigating problems in the
2570 * authentication cycle: it can be set in postgresql.conf to allow
2571 * time to attach to the newly-forked backend with a debugger. (See
2572 * also the -W backend switch, which we allow clients to pass through
2573 * PGOPTIONS, but it is not honored until after authentication.)
2575 if (PreAuthDelay > 0)
2576 sleep(PreAuthDelay);
2578 /* Will exit on failure */
2583 * Now, build the argv vector that will be given to PostgresMain.
2585 * The layout of the command line is
2586 * postgres [secure switches] -p databasename [insecure switches]
2587 * where the switches after -p come from the client request.
2589 * The maximum possible number of commandline arguments that could come
2590 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2594 maxac = 10; /* for fixed args supplied below */
2595 maxac += (strlen(ExtraOptions) + 1) / 2;
2596 if (port->cmdline_options)
2597 maxac += (strlen(port->cmdline_options) + 1) / 2;
2599 av = (char **) MemoryContextAlloc(TopMemoryContext,
2600 maxac * sizeof(char *));
2603 av[ac++] = "postgres";
2606 * Pass the requested debugging level along to the backend.
2610 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2611 av[ac++] = debugbuf;
2615 * Pass any backend switches specified with -o in the postmaster's own
2616 * command line. We assume these are secure.
2618 split_opts(av, &ac, ExtraOptions);
2620 /* Tell the backend what protocol the frontend is using. */
2621 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2622 av[ac++] = protobuf;
2625 /* pass data dir before end of secure switches (-p) */
2631 * Tell the backend it is being called from the postmaster, and which
2632 * database to use. -p marks the end of secure switches.
2635 av[ac++] = port->database_name;
2638 * Pass the (insecure) option switches from the connection request.
2639 * (It's OK to mangle port->cmdline_options now.)
2641 if (port->cmdline_options)
2642 split_opts(av, &ac, port->cmdline_options);
2649 * Release postmaster's working memory context so that backend can
2650 * recycle the space. Note this does not trash *MyProcPort, because
2651 * ConnCreate() allocated that space with malloc() ... else we'd need
2652 * to copy the Port data here. Also, subsidiary data such as the
2653 * username isn't lost either; see ProcessStartupPacket().
2655 MemoryContextSwitchTo(TopMemoryContext);
2656 #ifndef EXEC_BACKEND
2657 MemoryContextDelete(PostmasterContext);
2659 PostmasterContext = NULL;
2662 * Debug: print arguments being passed to backend
2665 (errmsg_internal("%s child[%d]: starting with (",
2666 progname, getpid())));
2667 for (i = 0; i < ac; ++i)
2669 (errmsg_internal("\t%s", av[i])));
2671 (errmsg_internal(")")));
2673 ClientAuthInProgress = false; /* client_min_messages is active
2676 return (PostgresMain(ac, av, port->user_name));
2684 * SubPostmasterMain -- prepare the fork/exec'd process to be in an equivalent
2685 * state (for calling BackendRun) as a forked process.
2688 * Shouldn't return at all.
2691 SubPostmasterMain(int argc, char* argv[])
2693 unsigned long backendID;
2696 memset((void*)&port, 0, sizeof(Port));
2699 /* Do this sooner rather than later... */
2700 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2702 /* In EXEC case we will not have inherited these settings */
2703 IsPostmasterEnvironment = true;
2704 whereToSendOutput = None;
2706 /* Setup global context */
2707 MemoryContextInit();
2708 InitializeGUCOptions();
2710 /* Parse passed-in context */
2712 backendID = (unsigned long)atol(argv[argc++]);
2713 DataDir = strdup(argv[argc++]);
2715 /* Read in file-based context */
2716 read_nondefault_variables();
2717 read_backend_variables(backendID,&port);
2719 /* Remaining initialization */
2720 pgstat_init_forkexec_backend();
2722 /* FIXME: [fork/exec] Ugh */
2728 /* Attach process to shared segments */
2729 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2732 proc_exit(BackendRun(&port));
2737 * Backend_forkexec -- fork/exec off a backend process
2740 * the pid of the fork/exec'd process
2743 Backend_forkexec(Port *port)
2747 int ac = 0, bufc = 0, i;
2748 char buf[2][MAXPGPATH];
2750 if (!write_backend_variables(port))
2751 return -1; /* log made by write_backend_variables */
2753 av[ac++] = "postgres";
2754 av[ac++] = "-forkexec";
2756 /* Format up context to pass to exec'd process */
2757 snprintf(buf[bufc++],MAXPGPATH,"%lu",tmpBackendFileNum);
2758 /* FIXME: [fork/exec] whitespaces in directories? */
2759 snprintf(buf[bufc++],MAXPGPATH,"%s",DataDir);
2761 /* Add to the arg list */
2762 Assert(bufc <= lengthof(buf));
2763 for (i = 0; i < bufc; i++)
2766 /* FIXME: [fork/exec] ExtraOptions? */
2769 Assert(ac <= lengthof(av));
2772 pid = win32_forkexec(pg_pathname,av); /* logs on error */
2774 /* Fire off execv in child */
2775 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
2777 * FIXME: [fork/exec] suggestions for what to do here?
2778 * Probably OK to issue error (unlike pgstat case)
2782 return pid; /* Parent returns pid */
2789 * ExitPostmaster -- cleanup
2791 * Do NOT call exit() directly --- always go through here!
2794 ExitPostmaster(int status)
2796 /* should cleanup shared memory and kill all backends */
2799 * Not sure of the semantics here. When the Postmaster dies, should
2800 * the backends all be killed? probably not.
2802 * MUST -- vadim 05-10-1999
2804 /* Should I use true instead? */
2805 ClosePostmasterPorts(false);
2811 * sigusr1_handler - handle signal conditions from child processes
2814 sigusr1_handler(SIGNAL_ARGS)
2816 int save_errno = errno;
2818 PG_SETMASK(&BlockSig);
2820 if (CheckPostmasterSignal(PMSIGNAL_DO_CHECKPOINT))
2822 if (CheckPointWarning != 0)
2825 * This only times checkpoints forced by running out of
2826 * segment files. Other checkpoints could reduce the
2827 * frequency of forced checkpoints.
2829 time_t now = time(NULL);
2831 if (LastSignalledCheckpoint != 0)
2833 int elapsed_secs = now - LastSignalledCheckpoint;
2835 if (elapsed_secs < CheckPointWarning)
2837 (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
2839 errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
2841 LastSignalledCheckpoint = now;
2845 * Request to schedule a checkpoint
2847 * Ignore request if checkpoint is already running or checkpointing
2848 * is currently disabled
2850 if (CheckPointPID == 0 && checkpointed &&
2851 StartupPID == 0 && Shutdown == NoShutdown &&
2852 !FatalError && random_seed != 0)
2854 CheckPointPID = CheckPointDataBase();
2855 /* note: if fork fails, CheckPointPID stays 0; nothing happens */
2859 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2862 * Password or group file has changed.
2868 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2871 * Send SIGUSR2 to all children (triggers AsyncNotifyHandler). See
2872 * storage/ipc/sinvaladt.c for the use of this.
2874 if (Shutdown == NoShutdown)
2875 SignalChildren(SIGUSR2);
2878 PG_SETMASK(&UnBlockSig);
2885 * Dummy signal handler
2887 * We use this for signals that we don't actually use in the postmaster,
2888 * but we do use in backends. If we SIG_IGN such signals in the postmaster,
2889 * then a newly started backend might drop a signal that arrives before it's
2890 * able to reconfigure its signal processing. (See notes in postgres.c.)
2893 dummy_handler(SIGNAL_ARGS)
2899 * CharRemap: given an int in range 0..61, produce textual encoding of it
2900 * per crypt(3) conventions.
2924 RandomSalt(char *cryptSalt, char *md5Salt)
2926 long rand = PostmasterRandom();
2928 cryptSalt[0] = CharRemap(rand % 62);
2929 cryptSalt[1] = CharRemap(rand / 62);
2932 * It's okay to reuse the first random value for one of the MD5 salt
2933 * bytes, since only one of the two salts will be sent to the client.
2934 * After that we need to compute more random bits.
2936 * We use % 255, sacrificing one possible byte value, so as to ensure
2937 * that all bits of the random() value participate in the result.
2938 * While at it, add one to avoid generating any null bytes.
2940 md5Salt[0] = (rand % 255) + 1;
2941 rand = PostmasterRandom();
2942 md5Salt[1] = (rand % 255) + 1;
2943 rand = PostmasterRandom();
2944 md5Salt[2] = (rand % 255) + 1;
2945 rand = PostmasterRandom();
2946 md5Salt[3] = (rand % 255) + 1;
2953 PostmasterRandom(void)
2955 static bool initialized = false;
2959 Assert(random_seed != 0);
2960 srandom(random_seed);
2968 * Count up number of child processes.
2977 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2979 bp = (Backend *) DLE_VAL(curr);
2980 if (bp->pid != MyProcPid)
2983 /* Checkpoint and bgwriter will be in the list, discount them */
2984 if (CheckPointPID != 0)
2986 if (BgWriterPID != 0)
2992 * Fire off a subprocess for startup/shutdown/checkpoint/bgwriter.
2994 * Return value of SSDataBase is subprocess' PID, or 0 if failed to start subprocess
2995 * (0 is returned only for checkpoint/bgwriter cases).
2997 * note: in the EXEC_BACKEND case, we delay the fork until argument list has been
3000 NON_EXEC_STATIC void
3001 SSDataBaseInit(int xlop)
3003 const char *statmsg;
3005 IsUnderPostmaster = true; /* we are a postmaster subprocess
3009 /* In EXEC case we will not have inherited these settings */
3010 IsPostmasterEnvironment = true;
3011 whereToSendOutput = None;
3014 MyProcPid = getpid(); /* reset MyProcPid */
3016 /* Lose the postmaster's on-exit routines and port connections */
3020 * Identify myself via ps
3024 case BS_XLOG_STARTUP:
3025 statmsg = "startup subprocess";
3027 case BS_XLOG_CHECKPOINT:
3028 statmsg = "checkpoint subprocess";
3030 case BS_XLOG_BGWRITER:
3031 statmsg = "bgwriter subprocess";
3033 case BS_XLOG_SHUTDOWN:
3034 statmsg = "shutdown subprocess";
3037 statmsg = "??? subprocess";
3040 init_ps_display(statmsg, "", "");
3046 SSDataBase(int xlop)
3050 #ifndef EXEC_BACKEND
3051 #ifdef LINUX_PROFILE
3052 struct itimerval prof_itimer;
3061 #ifndef EXEC_BACKEND
3062 #ifdef LINUX_PROFILE
3063 /* see comments in BackendRun */
3064 getitimer(ITIMER_PROF, &prof_itimer);
3068 /* Specific beos actions before backend startup */
3069 beos_before_backend_startup();
3072 /* Non EXEC_BACKEND case; fork here */
3073 if ((pid = fork()) == 0) /* child */
3081 #ifndef EXEC_BACKEND
3082 #ifdef LINUX_PROFILE
3083 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3087 /* Specific beos actions after backend startup */
3088 beos_backend_startup();
3091 /* Close the postmaster's sockets */
3092 ClosePostmasterPorts(true);
3094 SSDataBaseInit(xlop);
3096 if (!write_backend_variables(NULL))
3097 return -1; /* log issued by write_backend_variables */
3100 /* Set up command-line arguments for subprocess */
3101 av[ac++] = "postgres";
3106 snprintf(nbbuf, sizeof(nbbuf), "-B%d", NBuffers);
3109 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
3113 /* pass data dir before end of secure switches (-p) */
3117 /* and the backend identifier + dbname */
3118 snprintf(idbuf, sizeof(idbuf), "-p%lu,template1", tmpBackendFileNum);
3122 av[ac++] = "template1";
3127 Assert(ac < lengthof(av));
3130 /* EXEC_BACKEND case; fork/exec here */
3132 pid = win32_forkexec(pg_pathname,av); /* logs on error */
3134 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
3137 elog(ERROR,"unable to execv in SSDataBase: %m");
3142 BootstrapMain(ac, av);
3150 #ifndef EXEC_BACKEND
3152 /* Specific beos actions before backend startup */
3153 beos_backend_startup_failed();
3158 case BS_XLOG_STARTUP:
3160 (errmsg("could not fork startup process: %m")));
3162 case BS_XLOG_CHECKPOINT:
3164 (errmsg("could not fork checkpoint process: %m")));
3166 case BS_XLOG_BGWRITER:
3168 (errmsg("could not fork bgwriter process: %m")));
3170 case BS_XLOG_SHUTDOWN:
3172 (errmsg("could not fork shutdown process: %m")));
3176 (errmsg("could not fork process: %m")));
3181 * fork failure is fatal during startup/shutdown, but there's no
3182 * need to choke if a routine checkpoint or starting a background
3185 if (xlop == BS_XLOG_CHECKPOINT)
3187 if (xlop == BS_XLOG_BGWRITER)
3193 * The startup and shutdown processes are not considered normal
3194 * backends, but the checkpoint and bgwriter processes are.
3195 * They must be added to the list of backends.
3197 if (xlop == BS_XLOG_CHECKPOINT || xlop == BS_XLOG_BGWRITER)
3199 if (!(bn = (Backend *) malloc(sizeof(Backend))))
3202 (errcode(ERRCODE_OUT_OF_MEMORY),
3203 errmsg("out of memory")));
3208 bn->cancel_key = PostmasterRandom();
3210 ShmemBackendArrayAdd(bn);
3212 DLAddHead(BackendList, DLNewElem(bn));
3215 * Since this code is executed periodically, it's a fine place to
3216 * do other actions that should happen every now and then on no
3217 * particular schedule. Such as...
3220 TouchSocketLockFile();
3228 * Create the opts file
3231 CreateOptsFile(int argc, char *argv[])
3233 char fullprogname[MAXPGPATH];
3234 char filename[MAXPGPATH];
3238 if (FindExec(fullprogname, argv[0], "postmaster") < 0)
3241 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3243 if ((fp = fopen(filename, "w")) == NULL)
3245 elog(LOG, "could not create file \"%s\": %m", filename);
3249 fprintf(fp, "%s", fullprogname);
3250 for (i = 1; i < argc; i++)
3251 fprintf(fp, " '%s'", argv[i]);
3256 elog(LOG, "could not write file \"%s\": %m", filename);
3264 * This should be used only for reporting "interactive" errors (essentially,
3265 * bogus arguments on the command line). Once the postmaster is launched,
3266 * use ereport. In particular, don't use this for anything that occurs
3267 * after pmdaemonize.
3270 postmaster_error(const char *fmt,...)
3274 fprintf(stderr, "%s: ", progname);
3276 vfprintf(stderr, gettext(fmt), ap);
3278 fprintf(stderr, "\n");
3285 * The following need to be available to the read/write_backend_variables
3288 #include "storage/spin.h"
3289 extern XLogRecPtr RedoRecPtr;
3290 extern XLogwrtResult LogwrtResult;
3291 extern slock_t *ShmemLock;
3292 extern slock_t *ShmemIndexLock;
3293 extern void *ShmemIndexAlloc;
3294 typedef struct LWLock LWLock;
3295 extern LWLock *LWLockArray;
3296 extern slock_t *ProcStructLock;
3297 extern int pgStatSock;
3299 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3300 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3301 #define get_tmp_backend_file_name(buf,id) \
3305 "%s/%s/%s.backend_var.%lu", \
3307 PG_TEMP_FILES_DIR, \
3308 PG_TEMP_FILE_PREFIX, \
3313 write_backend_variables(Port *port)
3315 char filename[MAXPGPATH];
3317 get_tmp_backend_file_name(filename,++tmpBackendFileNum);
3320 fp = AllocateFile(filename, PG_BINARY_W);
3323 /* As per OpenTemporaryFile... */
3324 char dirname[MAXPGPATH];
3325 sprintf(dirname,"%s/%s",DataDir,PG_TEMP_FILES_DIR);
3326 mkdir(dirname, S_IRWXU);
3328 fp = AllocateFile(filename, PG_BINARY_W);
3332 (errcode_for_file_access(),
3333 errmsg("could not write to file \"%s\": %m", filename)));
3341 write_var(port->sock,fp);
3342 write_var(port->proto,fp);
3343 write_var(port->laddr,fp);
3344 write_var(port->raddr,fp);
3345 write_var(port->canAcceptConnections,fp);
3346 write_var(port->cryptSalt,fp);
3347 write_var(port->md5Salt,fp);
3349 write_var(MyCancelKey,fp);
3351 write_var(RedoRecPtr,fp);
3352 write_var(LogwrtResult,fp);
3354 write_var(UsedShmemSegID,fp);
3355 write_var(UsedShmemSegAddr,fp);
3357 write_var(ShmemLock,fp);
3358 write_var(ShmemIndexLock,fp);
3359 write_var(ShmemVariableCache,fp);
3360 write_var(ShmemIndexAlloc,fp);
3361 write_var(ShmemBackendArray,fp);
3363 write_var(LWLockArray,fp);
3364 write_var(ProcStructLock,fp);
3365 write_var(pgStatSock,fp);
3367 write_var(PreAuthDelay,fp);
3368 write_var(debug_flag,fp);
3369 write_var(PostmasterPid,fp);
3375 (errcode_for_file_access(),
3376 errmsg("could not write to file \"%s\": %m", filename)));
3384 read_backend_variables(unsigned long id, Port *port)
3386 char filename[MAXPGPATH];
3388 get_tmp_backend_file_name(filename,id);
3391 fp = AllocateFile(filename, PG_BINARY_R);
3395 (errcode_for_file_access(),
3396 errmsg("could not read from backend_variables file \"%s\": %m", filename)));
3403 read_var(port->sock,fp);
3404 read_var(port->proto,fp);
3405 read_var(port->laddr,fp);
3406 read_var(port->raddr,fp);
3407 read_var(port->canAcceptConnections,fp);
3408 read_var(port->cryptSalt,fp);
3409 read_var(port->md5Salt,fp);
3411 read_var(MyCancelKey,fp);
3413 read_var(RedoRecPtr,fp);
3414 read_var(LogwrtResult,fp);
3416 read_var(UsedShmemSegID,fp);
3417 read_var(UsedShmemSegAddr,fp);
3419 read_var(ShmemLock,fp);
3420 read_var(ShmemIndexLock,fp);
3421 read_var(ShmemVariableCache,fp);
3422 read_var(ShmemIndexAlloc,fp);
3423 read_var(ShmemBackendArray,fp);
3425 read_var(LWLockArray,fp);
3426 read_var(ProcStructLock,fp);
3427 read_var(pgStatSock,fp);
3429 read_var(PreAuthDelay,fp);
3430 read_var(debug_flag,fp);
3431 read_var(PostmasterPid,fp);
3435 if (unlink(filename) != 0)
3437 (errcode_for_file_access(),
3438 errmsg("could not remove file \"%s\": %m", filename)));
3442 size_t ShmemBackendArraySize(void)
3444 return (NUM_BACKENDARRAY_ELEMS*sizeof(Backend));
3447 void ShmemBackendArrayAllocation(void)
3449 size_t size = ShmemBackendArraySize();
3450 ShmemBackendArray = (Backend*)ShmemAlloc(size);
3451 memset(ShmemBackendArray, 0, size);
3454 static void ShmemBackendArrayAdd(Backend *bn)
3457 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3459 /* Find an empty slot */
3460 if (ShmemBackendArray[i].pid == 0)
3462 ShmemBackendArray[i] = *bn;
3467 /* FIXME: [fork/exec] some sort of error */
3471 static void ShmemBackendArrayRemove(pid_t pid)
3474 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3476 if (ShmemBackendArray[i].pid == pid)
3478 /* Mark the slot as empty */
3479 ShmemBackendArray[i].pid = 0;
3484 /* Something stronger than WARNING here? */
3486 (errmsg_internal("unable to find backend entry with pid %d",
3494 pid_t win32_forkexec(const char* path, char *argv[])
3497 PROCESS_INFORMATION pi;
3500 char cmdLine[MAXPGPATH];
3501 HANDLE childHandleCopy;
3502 HANDLE waiterThread;
3504 /* Format the cmd line */
3505 snprintf(cmdLine,sizeof(cmdLine),"%s",path);
3507 while (argv[++i] != NULL)
3509 /* FIXME: [fork/exec] some strlen checks might be prudent here */
3510 strcat(cmdLine," ");
3511 strcat(cmdLine,argv[i]);
3515 * The following snippet can disappear when we consistently
3516 * use forward slashes.
3519 while (*(p++) != '\0')
3520 if (*p == '/') *p = '\\';
3522 memset(&pi,0,sizeof(pi));
3523 memset(&si,0,sizeof(si));
3525 if (!CreateProcess(NULL,cmdLine,NULL,NULL,TRUE,0,NULL,NULL,&si,&pi))
3527 elog(ERROR,"CreateProcess call failed (%i): %m",(int)GetLastError());
3531 if (!IsUnderPostmaster)
3532 /* We are the Postmaster creating a child... */
3533 win32_AddChild(pi.dwProcessId,pi.hProcess);
3535 if (!DuplicateHandle(GetCurrentProcess(),
3537 GetCurrentProcess(),
3541 DUPLICATE_SAME_ACCESS))
3543 (errmsg_internal("failed to duplicate child handle: %i",(int)GetLastError())));
3544 waiterThread = CreateThread(NULL, 64*1024, win32_sigchld_waiter, (LPVOID)childHandleCopy, 0, NULL);
3547 (errmsg_internal("failed to create sigchld waiter thread: %i",(int)GetLastError())));
3548 CloseHandle(waiterThread);
3550 if (IsUnderPostmaster)
3551 CloseHandle(pi.hProcess);
3552 CloseHandle(pi.hThread);
3554 return pi.dwProcessId;
3558 * Note: The following three functions must not be interrupted (eg. by signals).
3559 * As the Postgres Win32 signalling architecture (currently) requires polling,
3560 * or APC checking functions which aren't used here, this is not an issue.
3562 * We keep two separate arrays, instead of a single array of pid/HANDLE structs,
3563 * to avoid having to re-create a handle array for WaitForMultipleObjects on
3564 * each call to win32_waitpid.
3567 static void win32_AddChild(pid_t pid, HANDLE handle)
3569 Assert(win32_childPIDArray && win32_childHNDArray);
3570 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3572 win32_childPIDArray[win32_numChildren] = pid;
3573 win32_childHNDArray[win32_numChildren] = handle;
3574 ++win32_numChildren;
3577 /* FIXME: [fork/exec] some sort of error */
3581 static void win32_RemoveChild(pid_t pid)
3584 Assert(win32_childPIDArray && win32_childHNDArray);
3586 for (i = 0; i < win32_numChildren; i++)
3588 if (win32_childPIDArray[i] == pid)
3590 CloseHandle(win32_childHNDArray[i]);
3592 /* Swap last entry into the "removed" one */
3593 --win32_numChildren;
3594 win32_childPIDArray[i] = win32_childPIDArray[win32_numChildren];
3595 win32_childHNDArray[i] = win32_childHNDArray[win32_numChildren];
3600 /* Something stronger than WARNING here? */
3602 (errmsg_internal("unable to find child entry with pid %lu",
3606 static pid_t win32_waitpid(int *exitstatus)
3608 Assert(win32_childPIDArray && win32_childHNDArray);
3609 elog(DEBUG3,"waiting on %lu children",win32_numChildren);
3611 if (win32_numChildren > 0)
3614 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't
3615 * want to run queued APCs here.
3619 DWORD ret = WaitForMultipleObjects(win32_numChildren,win32_childHNDArray,FALSE,0);
3625 (errmsg_internal("failed to wait on %lu children: %i",
3626 win32_numChildren,(int)GetLastError())));
3627 /* Fall through to WAIT_TIMEOUTs return */
3630 /* No children have finished */
3634 /* Get the exit code, and return the PID of, the respective process */
3635 index = ret-WAIT_OBJECT_0;
3636 Assert(index >= 0 && index < win32_numChildren);
3637 if (!GetExitCodeProcess(win32_childHNDArray[index],&exitCode))
3639 * If we get this far, this should never happen, but, then again...
3640 * No choice other than to assume a catastrophic failure.
3643 (errmsg_internal("failed to get exit code for child %lu",
3644 win32_childPIDArray[index])));
3645 *exitstatus = (int)exitCode;
3646 return win32_childPIDArray[index];
3654 /* Note! Code belows executes on separate threads, one for
3655 each child process created */
3656 static DWORD WINAPI win32_sigchld_waiter(LPVOID param) {
3657 HANDLE procHandle = (HANDLE)param;
3659 DWORD r = WaitForSingleObject(procHandle, INFINITE);
3660 if (r == WAIT_OBJECT_0)
3661 pg_queue_signal(SIGCHLD);
3663 fprintf(stderr,"ERROR: Failed to wait on child process handle: %i\n",(int)GetLastError());
3664 CloseHandle(procHandle);