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 1.369 2004/02/25 19:41:22 momjian Exp $
45 * The Postmaster sets up a few shared memory data structures
46 * for the backends. It should at the very least initialize the
50 * The Postmaster shares memory with the backends but should avoid
51 * touching shared memory, so as not to become stuck if a crashing
52 * backend screws up locks or shared memory. Likewise, the Postmaster
53 * should never block on messages from frontend clients.
56 * The Postmaster cleans up after backends if they have an emergency
57 * exit and/or core dump.
59 *-------------------------------------------------------------------------
70 #include <sys/socket.h>
74 #include <sys/param.h>
75 #include <netinet/in.h>
76 #include <arpa/inet.h>
80 #ifdef HAVE_SYS_SELECT_H
81 #include <sys/select.h>
89 #include <DNSServiceDiscovery/DNSServiceDiscovery.h>
92 #include "catalog/pg_database.h"
93 #include "commands/async.h"
94 #include "lib/dllist.h"
95 #include "libpq/auth.h"
96 #include "libpq/crypt.h"
97 #include "libpq/libpq.h"
98 #include "libpq/pqcomm.h"
99 #include "libpq/pqsignal.h"
100 #include "miscadmin.h"
101 #include "nodes/nodes.h"
102 #include "storage/fd.h"
103 #include "storage/ipc.h"
104 #include "storage/pg_shmem.h"
105 #include "storage/pmsignal.h"
106 #include "storage/proc.h"
107 #include "storage/bufmgr.h"
108 #include "access/xlog.h"
109 #include "tcop/tcopprot.h"
110 #include "utils/guc.h"
111 #include "utils/memutils.h"
112 #include "utils/ps_status.h"
113 #include "bootstrap/bootstrap.h"
117 #define INVALID_SOCK (-1)
119 #ifdef HAVE_SIGPROCMASK
131 * List of active backends (or child processes anyway; we don't actually
132 * know whether a given child has become a backend or is still in the
133 * authorization phase). This is used mainly to keep track of how many
134 * children we have and send them appropriate signals when necessary.
138 pid_t pid; /* process id of backend */
139 long cancel_key; /* cancel key for cancels for this backend */
142 static Dllist *BackendList;
145 #define NUM_BACKENDARRAY_ELEMS (2*MaxBackends)
146 static Backend *ShmemBackendArray;
149 /* The socket number we are listening for connections on */
155 * MaxBackends is the limit on the number of backends we can start.
156 * Note that a larger MaxBackends value will increase the size of the
157 * shared memory area as well as cause the postmaster to grab more
158 * kernel semaphores, even if you never actually use that many
164 * ReservedBackends is the number of backends reserved for superuser use.
165 * This number is taken out of the pool size given by MaxBackends so
166 * number of backend slots available to non-superusers is
167 * (MaxBackends - ReservedBackends). Note what this really means is
168 * "if there are <= ReservedBackends connections available, only superusers
169 * can make new connections" --- pre-existing superuser connections don't
170 * count against the limit.
172 int ReservedBackends;
175 static char *progname = NULL;
177 /* The socket(s) we're listening to. */
179 static int ListenSocket[MAXLISTEN];
181 /* Used to reduce macros tests */
183 const bool ExecBackend = true;
186 const bool ExecBackend = false;
190 * Set by the -o option
192 static char ExtraOptions[MAXPGPATH];
195 * These globals control the behavior of the postmaster in case some
196 * backend dumps core. Normally, it kills all peers of the dead backend
197 * and reinitializes shared memory. By specifying -s or -n, we can have
198 * the postmaster stop (rather than kill) peers and not reinitialize
199 * shared data structures.
201 static bool Reinit = true;
202 static int SendStop = false;
204 /* still more option variables */
205 bool NetServer = false; /* listen on TCP/IP */
206 bool EnableSSL = false;
207 bool SilentMode = false; /* silent mode (-S) */
209 int PreAuthDelay = 0;
210 int AuthenticationTimeout = 60;
211 int CheckPointTimeout = 300;
212 int CheckPointWarning = 30;
213 time_t LastSignalledCheckpoint = 0;
215 bool log_hostname; /* for ps display */
217 bool Log_connections = false;
218 bool Db_user_namespace = false;
220 char *rendezvous_name;
222 /* For FNCTL_NONBLOCK */
223 #if defined(WIN32) || defined(__BEOS__)
224 long ioctlsocket_ret=1;
227 /* list of library:init-function to be preloaded */
228 char *preload_libraries_string = NULL;
230 /* Startup/shutdown state */
231 static pid_t StartupPID = 0,
235 static time_t checkpointed = 0;
238 #define SmartShutdown 1
239 #define FastShutdown 2
241 static int Shutdown = NoShutdown;
243 static bool FatalError = false; /* T if recovering from backend crash */
245 bool ClientAuthInProgress = false; /* T during new-client
249 * State for assigning random salts and cancel keys.
250 * Also, the global MyCancelKey passes the cancel key assigned to a given
251 * backend from the postmaster to that backend (via fork).
254 static unsigned int random_seed = 0;
256 static int debug_flag = 0;
262 #ifdef HAVE_INT_OPTRESET
267 * postmaster.c - function prototypes
269 static void pmdaemonize(int argc, char *argv[]);
270 static Port *ConnCreate(int serverFd);
271 static void ConnFree(Port *port);
272 static void reset_shared(unsigned short port);
273 static void SIGHUP_handler(SIGNAL_ARGS);
274 static void pmdie(SIGNAL_ARGS);
275 static void reaper(SIGNAL_ARGS);
276 static void sigusr1_handler(SIGNAL_ARGS);
277 static void dummy_handler(SIGNAL_ARGS);
278 static void CleanupProc(int pid, int exitstatus);
279 static void LogChildExit(int lev, const char *procname,
280 int pid, int exitstatus);
281 static void BackendInit(Port *port);
282 static int BackendRun(Port *port);
283 static void ExitPostmaster(int status);
284 static void usage(const char *);
285 static int ServerLoop(void);
286 static int BackendStartup(Port *port);
287 static int ProcessStartupPacket(Port *port, bool SSLdone);
288 static void processCancelRequest(Port *port, void *pkt);
289 static int initMasks(fd_set *rmask);
290 static void report_fork_failure_to_client(Port *port, int errnum);
291 static enum CAC_state canAcceptConnections(void);
292 static long PostmasterRandom(void);
293 static void RandomSalt(char *cryptSalt, char *md5Salt);
294 static void SignalChildren(int signal);
295 static int CountChildren(void);
296 static bool CreateOptsFile(int argc, char *argv[]);
297 NON_EXEC_STATIC void SSDataBaseInit(int xlop);
298 static pid_t SSDataBase(int xlop);
300 postmaster_error(const char *fmt,...)
301 /* This lets gcc check the format string for consistency. */
302 __attribute__((format(printf, 1, 2)));
306 pid_t win32_forkexec(const char* path, char *argv[]);
308 static void win32_AddChild(pid_t pid, HANDLE handle);
309 static void win32_RemoveChild(pid_t pid);
310 static pid_t win32_waitpid(int *exitstatus);
311 static DWORD WINAPI win32_sigchld_waiter(LPVOID param);
313 static pid_t *win32_childPIDArray;
314 static HANDLE *win32_childHNDArray;
315 static unsigned long win32_numChildren = 0;
318 static pid_t Backend_forkexec(Port *port);
320 static unsigned long tmpBackendFileNum = 0;
321 void read_backend_variables(unsigned long id, Port *port);
322 static bool write_backend_variables(Port *port);
324 size_t ShmemBackendArraySize(void);
325 void ShmemBackendArrayAllocation(void);
326 static void ShmemBackendArrayAdd(Backend *bn);
327 static void ShmemBackendArrayRemove(pid_t pid);
330 #define StartupDataBase() SSDataBase(BS_XLOG_STARTUP)
331 #define CheckPointDataBase() SSDataBase(BS_XLOG_CHECKPOINT)
332 #define StartBackgroundWriter() SSDataBase(BS_XLOG_BGWRITER)
333 #define ShutdownDataBase() SSDataBase(BS_XLOG_SHUTDOWN)
337 checkDataDir(const char *checkdir)
339 char path[MAXPGPATH];
341 struct stat stat_buf;
343 if (checkdir == NULL)
346 gettext("%s does not know where to find the database system data.\n"
347 "You must specify the directory that contains the database system\n"
348 "either by specifying the -D invocation option or by setting the\n"
349 "PGDATA environment variable.\n"),
354 if (stat(checkdir, &stat_buf) == -1)
358 (errcode_for_file_access(),
359 errmsg("data directory \"%s\" does not exist",
363 (errcode_for_file_access(),
364 errmsg("could not read permissions of directory \"%s\": %m",
369 * Check if the directory has group or world access. If so, reject.
371 * XXX temporarily suppress check when on Windows, because there may not
372 * be proper support for Unix-y file permissions. Need to think of a
373 * reasonable check to apply on Windows.
375 #if !defined(__CYGWIN__) && !defined(WIN32)
376 if (stat_buf.st_mode & (S_IRWXG | S_IRWXO))
378 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
379 errmsg("data directory \"%s\" has group or world access",
381 errdetail("Permissions should be u=rwx (0700).")));
384 /* Look for PG_VERSION before looking for pg_control */
385 ValidatePgVersion(checkdir);
387 snprintf(path, sizeof(path), "%s/global/pg_control", checkdir);
389 fp = AllocateFile(path, PG_BINARY_R);
393 gettext("%s: could not find the database system\n"
394 "Expected to find it in the directory \"%s\",\n"
395 "but could not open file \"%s\": %s\n"),
396 progname, checkdir, path, strerror(errno));
403 #ifdef USE_RENDEZVOUS
405 /* reg_reply -- empty callback function for DNSServiceRegistrationCreate() */
407 reg_reply(DNSServiceRegistrationReplyErrorType errorCode, void *context)
414 PostmasterMain(int argc, char *argv[])
418 char original_extraoptions[MAXPGPATH];
419 char *potential_DataDir = NULL;
422 *original_extraoptions = '\0';
426 IsPostmasterEnvironment = true;
429 * Catch standard options before doing much else. This even works on
430 * systems without getopt_long.
434 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
439 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
441 puts("postmaster (PostgreSQL) " PG_VERSION);
447 * for security, no dir or file created can be group or other
450 umask((mode_t) 0077);
452 MyProcPid = PostmasterPid = getpid();
455 * Fire up essential subsystems: memory management
460 * By default, palloc() requests in the postmaster will be allocated
461 * in the PostmasterContext, which is space that can be recycled by
462 * backends. Allocated data that needs to be available to backends
463 * should be allocated in TopMemoryContext.
465 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
467 ALLOCSET_DEFAULT_MINSIZE,
468 ALLOCSET_DEFAULT_INITSIZE,
469 ALLOCSET_DEFAULT_MAXSIZE);
470 MemoryContextSwitchTo(PostmasterContext);
472 IgnoreSystemIndexes(false);
477 InitializeGUCOptions();
479 potential_DataDir = getenv("PGDATA"); /* default value */
483 while ((opt = getopt(argc, argv, "A:a:B:b:c:D:d:Fh:ik:lm:MN:no:p:Ss-:")) != -1)
488 #ifdef USE_ASSERT_CHECKING
489 SetConfigOption("debug_assertions", optarg, PGC_POSTMASTER, PGC_S_ARGV);
491 postmaster_error("assert checking is not compiled in");
495 /* Can no longer set authentication method. */
498 SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
501 /* Can no longer set the backend executable file to use. */
504 potential_DataDir = optarg;
508 /* Turn on debugging for the postmaster. */
509 char *debugstr = palloc(strlen("debug") + strlen(optarg) + 1);
511 sprintf(debugstr, "debug%s", optarg);
512 SetConfigOption("log_min_messages", debugstr,
513 PGC_POSTMASTER, PGC_S_ARGV);
515 debug_flag = atoi(optarg);
519 SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
522 SetConfigOption("virtual_host", optarg, PGC_POSTMASTER, PGC_S_ARGV);
525 SetConfigOption("tcpip_socket", "true", PGC_POSTMASTER, PGC_S_ARGV);
528 SetConfigOption("unix_socket_directory", optarg, PGC_POSTMASTER, PGC_S_ARGV);
532 SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
536 /* Multiplexed backends no longer supported. */
541 * ignore this flag. This may be passed in because the
542 * program was run as 'postgres -M' instead of
547 /* The max number of backends to start. */
548 SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
551 /* Don't reinit shared mem after abnormal exit */
557 * Other options to pass to the backend on the command
558 * line -- useful only for debugging.
560 strcat(ExtraOptions, " ");
561 strcat(ExtraOptions, optarg);
562 strcpy(original_extraoptions, optarg);
565 SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
570 * Start in 'S'ilent mode (disassociate from controlling
571 * tty). You may also think of this as 'S'ysV mode since
572 * it's most badly needed on SysV-derived systems like
575 SetConfigOption("silent_mode", "true", PGC_POSTMASTER, PGC_S_ARGV);
580 * In the event that some backend dumps core, send
581 * SIGSTOP, rather than SIGQUIT, to all its peers. This
582 * lets the wily post_hacker collect core dumps from
593 ParseLongOption(optarg, &name, &value);
598 (errcode(ERRCODE_SYNTAX_ERROR),
599 errmsg("--%s requires a value",
603 (errcode(ERRCODE_SYNTAX_ERROR),
604 errmsg("-c %s requires a value",
608 SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
617 gettext("Try \"%s --help\" for more information.\n"),
624 * Postmaster accepts no non-option switch arguments.
628 postmaster_error("invalid argument: \"%s\"", argv[optind]);
630 gettext("Try \"%s --help\" for more information.\n"),
636 * Now we can set the data directory, and then read postgresql.conf.
638 checkDataDir(potential_DataDir); /* issues error messages */
639 SetDataDir(potential_DataDir);
641 ProcessConfigFile(PGC_POSTMASTER);
643 write_nondefault_variables(PGC_POSTMASTER);
647 * Check for invalid combinations of GUC settings.
649 if (NBuffers < 2 * MaxBackends || NBuffers < 16)
652 * Do not accept -B so small that backends are likely to starve
653 * for lack of buffers. The specific choices here are somewhat
656 postmaster_error("the number of buffers (-B) must be at least twice the number of allowed connections (-N) and at least 16");
660 if (ReservedBackends >= MaxBackends)
662 postmaster_error("superuser_reserved_connections must be less than max_connections");
667 * Other one-time internal sanity checks can go here.
669 if (!CheckDateTokenTables())
671 postmaster_error("invalid datetoken tables, please fix");
676 * Now that we are done processing the postmaster arguments, reset
677 * getopt(3) library so that it will work correctly in subprocesses.
680 #ifdef HAVE_INT_OPTRESET
681 optreset = 1; /* some systems need this too */
684 /* For debugging: display postmaster environment */
686 extern char **environ;
690 (errmsg_internal("%s: PostmasterMain: initial environ dump:",
693 (errmsg_internal("-----------------------------------------")));
694 for (p = environ; *p; ++p)
696 (errmsg_internal("\t%s", *p)));
698 (errmsg_internal("-----------------------------------------")));
702 * On some systems our dynloader code needs the executable's pathname.
704 if (FindExec(pg_pathname, progname, "postgres") < 0)
706 (errmsg("%s: could not locate postgres executable",
710 * Initialize SSL library, if specified.
713 if (EnableSSL && !NetServer)
715 postmaster_error("TCP/IP connections must be enabled for SSL");
723 * process any libraries that should be preloaded and optionally
726 if (preload_libraries_string)
727 process_preload_libraries(preload_libraries_string);
730 * Fork away from controlling terminal, if -S specified.
732 * Must do this before we grab any interlock files, else the interlocks
733 * will show the wrong PID.
736 pmdaemonize(argc, argv);
739 * Create lockfile for data directory.
741 * We want to do this before we try to grab the input sockets, because
742 * the data directory interlock is more reliable than the socket-file
743 * interlock (thanks to whoever decided to put socket files in /tmp
744 * :-(). For the same reason, it's best to grab the TCP socket before
747 CreateDataDirLockFile(DataDir, true);
750 * Remove old temporary files. At this point there can be no other
751 * Postgres processes running in this directory, so this should be
757 * Establish input sockets.
759 for (i = 0; i < MAXLISTEN; i++)
760 ListenSocket[i] = -1;
764 if (VirtualHost && VirtualHost[0])
770 curhost = VirtualHost;
773 while (*curhost == ' ') /* skip any extra spaces */
775 if (*curhost == '\0')
777 endptr = strchr(curhost, ' ');
783 status = StreamServerPort(AF_UNSPEC, curhost,
784 (unsigned short) PostPortNumber,
786 ListenSocket, MAXLISTEN);
787 if (status != STATUS_OK)
789 (errmsg("could not create listen socket for \"%s\"",
794 curhost = endptr + 1;
802 status = StreamServerPort(AF_UNSPEC, NULL,
803 (unsigned short) PostPortNumber,
805 ListenSocket, MAXLISTEN);
806 if (status != STATUS_OK)
808 (errmsg("could not create TCP/IP listen socket")));
811 #ifdef USE_RENDEZVOUS
812 if (rendezvous_name != NULL)
814 DNSServiceRegistrationCreate(rendezvous_name,
817 htonl(PostPortNumber),
819 (DNSServiceRegistrationReply) reg_reply,
825 #ifdef HAVE_UNIX_SOCKETS
826 status = StreamServerPort(AF_UNIX, NULL,
827 (unsigned short) PostPortNumber,
829 ListenSocket, MAXLISTEN);
830 if (status != STATUS_OK)
832 (errmsg("could not create Unix-domain socket")));
838 * Set up shared memory and semaphores.
840 reset_shared(PostPortNumber);
843 * Estimate number of openable files. This must happen after setting up
844 * semaphores, because on some platforms semaphores count as open files.
849 * Initialize the list of active backends.
851 BackendList = DLNewList();
855 * Initialize the child pid/HANDLE arrays
857 win32_childPIDArray = (pid_t*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(pid_t));
858 win32_childHNDArray = (HANDLE*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(HANDLE));
859 if (!win32_childPIDArray || !win32_childHNDArray)
861 (errcode(ERRCODE_OUT_OF_MEMORY),
862 errmsg("out of memory")));
866 * Record postmaster options. We delay this till now to avoid
867 * recording bogus options (eg, NBuffers too high for available
870 if (!CreateOptsFile(argc, argv))
874 * Set up signal handlers for the postmaster process.
876 * CAUTION: when changing this list, check for side-effects on the signal
877 * handling setup of child processes. See tcop/postgres.c,
878 * bootstrap/bootstrap.c, and postmaster/pgstat.c.
881 PG_SETMASK(&BlockSig);
883 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
884 * children do same */
885 pqsignal(SIGINT, pmdie); /* send SIGTERM and ShutdownDataBase */
886 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
887 pqsignal(SIGTERM, pmdie); /* wait for children and ShutdownDataBase */
888 pqsignal(SIGALRM, SIG_IGN); /* ignored */
889 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
890 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
891 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
892 pqsignal(SIGCHLD, reaper); /* handle child termination */
893 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
894 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
895 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
897 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
901 * Reset whereToSendOutput from Debug (its starting state) to None.
902 * This prevents ereport from sending log messages to stderr unless
903 * the syslog/stderr switch permits. We don't do this until the
904 * postmaster is fully launched, since startup failures may as well be
905 * reported to stderr.
907 whereToSendOutput = None;
910 * On many platforms, the first call of localtime() incurs significant
911 * overhead to load timezone info from the system configuration files.
912 * By doing it once in the postmaster, we avoid having to do it in
913 * every started child process. The savings are not huge, but they
917 time_t now = time(NULL);
919 (void) localtime(&now);
923 * Initialize and try to startup the statistics collector process
929 * Load cached files for client authentication.
937 * We're ready to rock and roll...
939 StartupPID = StartupDataBase();
941 status = ServerLoop();
944 * ServerLoop probably shouldn't ever return, but if it does, close
947 ExitPostmaster(status != STATUS_OK);
949 return 0; /* not reached */
953 pmdaemonize(int argc, char *argv[])
959 struct itimerval prof_itimer;
963 /* see comments in BackendRun */
964 getitimer(ITIMER_PROF, &prof_itimer);
968 /* FIXME: [fork/exec] to be implemented? */
973 if (pid == (pid_t) -1)
975 postmaster_error("could not fork background process: %s",
981 /* Parent should just exit, without doing any atexit cleanup */
986 setitimer(ITIMER_PROF, &prof_itimer, NULL);
989 MyProcPid = getpid(); /* reset MyProcPid to child */
991 /* GH: If there's no setsid(), we hopefully don't need silent mode.
992 * Until there's a better solution.
997 postmaster_error("could not dissociate from controlling TTY: %s",
1002 i = open(NULL_DEV, O_RDWR | PG_BINARY);
1012 * Print out help message
1015 usage(const char *progname)
1017 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
1018 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
1019 printf(gettext("Options:\n"));
1020 #ifdef USE_ASSERT_CHECKING
1021 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
1023 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
1024 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
1025 printf(gettext(" -d 1-5 debugging level\n"));
1026 printf(gettext(" -D DATADIR database directory\n"));
1027 printf(gettext(" -F turn fsync off\n"));
1028 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
1029 printf(gettext(" -i enable TCP/IP connections\n"));
1030 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
1032 printf(gettext(" -l enable SSL connections\n"));
1034 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
1035 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
1036 printf(gettext(" -p PORT port number to listen on\n"));
1037 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1038 printf(gettext(" --help show this help, then exit\n"));
1039 printf(gettext(" --version output version information, then exit\n"));
1041 printf(gettext("\nDeveloper options:\n"));
1042 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1043 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1045 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1046 "configuration settings and how to set them on the command line or in\n"
1047 "the configuration file.\n\n"
1048 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1061 gettimeofday(&now, &tz);
1063 nSockets = initMasks(&readmask);
1069 struct timeval timeout;
1072 * The timeout for the select() below is normally set on the basis
1073 * of the time to the next checkpoint. However, if for some
1074 * reason we don't have a next-checkpoint time, time out after 60
1075 * seconds. This keeps checkpoint scheduling from locking up when
1076 * we get new connection requests infrequently (since we are
1077 * likely to detect checkpoint completion just after enabling
1078 * signals below, after we've already made the decision about how
1079 * long to wait this time).
1081 timeout.tv_sec = 60;
1082 timeout.tv_usec = 0;
1084 if (CheckPointPID == 0 && checkpointed &&
1085 StartupPID == 0 && Shutdown == NoShutdown &&
1086 !FatalError && random_seed != 0)
1088 time_t now = time(NULL);
1090 if (CheckPointTimeout + checkpointed > now)
1093 * Not time for checkpoint yet, so set select timeout
1095 timeout.tv_sec = CheckPointTimeout + checkpointed - now;
1099 /* Time to make the checkpoint... */
1100 CheckPointPID = CheckPointDataBase();
1103 * if fork failed, schedule another try at 0.1 normal
1106 if (CheckPointPID == 0)
1108 timeout.tv_sec = CheckPointTimeout / 10;
1109 checkpointed = now + timeout.tv_sec - CheckPointTimeout;
1115 * If no background writer process is running and we should
1116 * do background writing, start one. It doesn't matter if
1117 * this fails, we'll just try again later.
1119 if (BgWriterPID == 0 && BgWriterPercent > 0 &&
1120 StartupPID == 0 && Shutdown == NoShutdown &&
1121 !FatalError && random_seed != 0)
1123 BgWriterPID = StartBackgroundWriter();
1127 * Wait for something to happen.
1129 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1131 PG_SETMASK(&UnBlockSig);
1133 if (select(nSockets, &rmask, NULL, NULL, &timeout) < 0)
1135 PG_SETMASK(&BlockSig);
1136 if (errno == EINTR || errno == EWOULDBLOCK)
1139 (errcode_for_socket_access(),
1140 errmsg("select() failed in postmaster: %m")));
1141 return STATUS_ERROR;
1145 * Block all signals until we wait again. (This makes it safe for
1146 * our signal handlers to do nontrivial work.)
1148 PG_SETMASK(&BlockSig);
1151 * Select a random seed at the time of first receiving a request.
1153 while (random_seed == 0)
1155 gettimeofday(&later, &tz);
1158 * We are not sure how much precision is in tv_usec, so we
1159 * swap the nibbles of 'later' and XOR them with 'now'. On the
1160 * off chance that the result is 0, we loop until it isn't.
1162 random_seed = now.tv_usec ^
1163 ((later.tv_usec << 16) |
1164 ((later.tv_usec >> 16) & 0xffff));
1168 * New connection pending on any of our sockets? If so, fork a
1169 * child process to deal with it.
1171 for (i = 0; i < MAXLISTEN; i++)
1173 if (ListenSocket[i] == -1)
1175 if (FD_ISSET(ListenSocket[i], &rmask))
1177 port = ConnCreate(ListenSocket[i]);
1180 BackendStartup(port);
1183 * We no longer need the open socket or port structure
1186 StreamClose(port->sock);
1192 /* If we have lost the stats collector, try to start a new one */
1193 if (!pgstat_is_running)
1200 * Initialise the masks for select() for the ports
1201 * we are listening on. Return the number of sockets to listen on.
1205 initMasks(fd_set *rmask)
1212 for (i = 0; i < MAXLISTEN; i++)
1214 int fd = ListenSocket[i];
1228 * Read the startup packet and do something according to it.
1230 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1231 * not return at all.
1233 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1234 * if that's what you want. Return STATUS_ERROR if you don't want to
1235 * send anything to the client, which would typically be appropriate
1236 * if we detect a communications failure.)
1239 ProcessStartupPacket(Port *port, bool SSLdone)
1243 ProtocolVersion proto;
1244 MemoryContext oldcontext;
1246 if (pq_getbytes((char *) &len, 4) == EOF)
1249 * EOF after SSLdone probably means the client didn't like our
1250 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1251 * so don't clutter the log with a complaint.
1255 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1256 errmsg("incomplete startup packet")));
1257 return STATUS_ERROR;
1263 if (len < (int32) sizeof(ProtocolVersion) ||
1264 len > MAX_STARTUP_PACKET_LENGTH)
1267 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1268 errmsg("invalid length of startup packet")));
1269 return STATUS_ERROR;
1273 * Allocate at least the size of an old-style startup packet, plus one
1274 * extra byte, and make sure all are zeroes. This ensures we will
1275 * have null termination of all strings, in both fixed- and
1276 * variable-length packet layouts.
1278 if (len <= (int32) sizeof(StartupPacket))
1279 buf = palloc0(sizeof(StartupPacket) + 1);
1281 buf = palloc0(len + 1);
1283 if (pq_getbytes(buf, len) == EOF)
1286 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1287 errmsg("incomplete startup packet")));
1288 return STATUS_ERROR;
1292 * The first field is either a protocol version number or a special
1295 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1297 if (proto == CANCEL_REQUEST_CODE)
1299 processCancelRequest(port, buf);
1300 return 127; /* XXX */
1303 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1308 /* No SSL when disabled or on Unix sockets */
1309 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1312 SSLok = 'S'; /* Support for SSL */
1314 SSLok = 'N'; /* No support for SSL */
1316 if (send(port->sock, &SSLok, 1, 0) != 1)
1319 (errcode_for_socket_access(),
1320 errmsg("failed to send SSL negotiation response: %m")));
1321 return STATUS_ERROR; /* close the connection */
1325 if (SSLok == 'S' && secure_open_server(port) == -1)
1326 return STATUS_ERROR;
1328 /* regular startup packet, cancel, etc packet should follow... */
1329 /* but not another SSL negotiation request */
1330 return ProcessStartupPacket(port, true);
1333 /* Could add additional special packet types here */
1336 * Set FrontendProtocol now so that ereport() knows what format to
1337 * send if we fail during startup.
1339 FrontendProtocol = proto;
1341 /* Check we can handle the protocol the frontend is using. */
1343 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1344 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1345 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1346 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1348 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1349 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1350 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1351 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1352 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1353 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1356 * Now fetch parameters out of startup packet and save them into the
1357 * Port structure. All data structures attached to the Port struct
1358 * must be allocated in TopMemoryContext so that they won't disappear
1359 * when we pass them to PostgresMain (see BackendRun). We need not
1360 * worry about leaking this storage on failure, since we aren't in the
1361 * postmaster process anymore.
1363 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1365 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1367 int32 offset = sizeof(ProtocolVersion);
1370 * Scan packet body for name/option pairs. We can assume any
1371 * string beginning within the packet body is null-terminated,
1372 * thanks to zeroing extra byte above.
1374 port->guc_options = NIL;
1376 while (offset < len)
1378 char *nameptr = ((char *) buf) + offset;
1382 if (*nameptr == '\0')
1383 break; /* found packet terminator */
1384 valoffset = offset + strlen(nameptr) + 1;
1385 if (valoffset >= len)
1386 break; /* missing value, will complain below */
1387 valptr = ((char *) buf) + valoffset;
1389 if (strcmp(nameptr, "database") == 0)
1390 port->database_name = pstrdup(valptr);
1391 else if (strcmp(nameptr, "user") == 0)
1392 port->user_name = pstrdup(valptr);
1393 else if (strcmp(nameptr, "options") == 0)
1394 port->cmdline_options = pstrdup(valptr);
1397 /* Assume it's a generic GUC option */
1398 port->guc_options = lappend(port->guc_options,
1400 port->guc_options = lappend(port->guc_options,
1403 offset = valoffset + strlen(valptr) + 1;
1407 * If we didn't find a packet terminator exactly at the end of the
1408 * given packet length, complain.
1410 if (offset != len - 1)
1412 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1413 errmsg("invalid startup packet layout: expected terminator as last byte")));
1418 * Get the parameters from the old-style, fixed-width-fields
1419 * startup packet as C strings. The packet destination was
1420 * cleared first so a short packet has zeros silently added. We
1421 * have to be prepared to truncate the pstrdup result for oversize
1424 StartupPacket *packet = (StartupPacket *) buf;
1426 port->database_name = pstrdup(packet->database);
1427 if (strlen(port->database_name) > sizeof(packet->database))
1428 port->database_name[sizeof(packet->database)] = '\0';
1429 port->user_name = pstrdup(packet->user);
1430 if (strlen(port->user_name) > sizeof(packet->user))
1431 port->user_name[sizeof(packet->user)] = '\0';
1432 port->cmdline_options = pstrdup(packet->options);
1433 if (strlen(port->cmdline_options) > sizeof(packet->options))
1434 port->cmdline_options[sizeof(packet->options)] = '\0';
1435 port->guc_options = NIL;
1438 /* Check a user name was given. */
1439 if (port->user_name == NULL || port->user_name[0] == '\0')
1441 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1442 errmsg("no PostgreSQL user name specified in startup packet")));
1444 /* The database defaults to the user name. */
1445 if (port->database_name == NULL || port->database_name[0] == '\0')
1446 port->database_name = pstrdup(port->user_name);
1448 if (Db_user_namespace)
1451 * If user@, it is a global user, remove '@'. We only want to do
1452 * this if there is an '@' at the end and no earlier in the user
1453 * string or they may fake as a local user of another database
1454 * attaching to this database.
1456 if (strchr(port->user_name, '@') ==
1457 port->user_name + strlen(port->user_name) - 1)
1458 *strchr(port->user_name, '@') = '\0';
1461 /* Append '@' and dbname */
1464 db_user = palloc(strlen(port->user_name) +
1465 strlen(port->database_name) + 2);
1466 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1467 port->user_name = db_user;
1472 * Truncate given database and user names to length of a Postgres
1473 * name. This avoids lookup failures when overlength names are given.
1475 if (strlen(port->database_name) >= NAMEDATALEN)
1476 port->database_name[NAMEDATALEN - 1] = '\0';
1477 if (strlen(port->user_name) >= NAMEDATALEN)
1478 port->user_name[NAMEDATALEN - 1] = '\0';
1481 * Done putting stuff in TopMemoryContext.
1483 MemoryContextSwitchTo(oldcontext);
1486 * If we're going to reject the connection due to database state, say
1487 * so now instead of wasting cycles on an authentication exchange.
1488 * (This also allows a pg_ping utility to be written.)
1490 switch (port->canAcceptConnections)
1494 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1495 errmsg("the database system is starting up")));
1499 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1500 errmsg("the database system is shutting down")));
1504 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1505 errmsg("the database system is in recovery mode")));
1509 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1510 errmsg("sorry, too many clients already")));
1522 * The client has sent a cancel request packet, not a normal
1523 * start-a-new-connection packet. Perform the necessary processing.
1524 * Nothing is sent back to the client.
1527 processCancelRequest(Port *port, void *pkt)
1529 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1531 long cancelAuthCode;
1533 #ifndef EXEC_BACKEND
1539 backendPID = (int) ntohl(canc->backendPID);
1540 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1542 if (backendPID == CheckPointPID)
1545 (errmsg_internal("ignoring cancel request for checkpoint process %d",
1549 else if (backendPID == BgWriterPID)
1552 (errmsg_internal("ignoring cancel request for bgwriter process %d",
1557 /* See if we have a matching backend */
1558 #ifndef EXEC_BACKEND
1559 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1561 bp = (Backend *) DLE_VAL(curr);
1563 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1565 bp = (Backend*) &ShmemBackendArray[i];
1567 if (bp->pid == backendPID)
1569 if (bp->cancel_key == cancelAuthCode)
1571 /* Found a match; signal that backend to cancel current op */
1573 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1575 kill(bp->pid, SIGINT);
1578 /* Right PID, wrong key: no way, Jose */
1580 (errmsg_internal("bad key in cancel request for process %d",
1586 /* No matching backend */
1588 (errmsg_internal("bad pid in cancel request for process %d",
1593 * canAcceptConnections --- check to see if database state allows connections.
1595 static enum CAC_state
1596 canAcceptConnections(void)
1598 /* Can't start backends when in startup/shutdown/recovery state. */
1599 if (Shutdown > NoShutdown)
1600 return CAC_SHUTDOWN;
1604 return CAC_RECOVERY;
1607 * Don't start too many children.
1609 * We allow more connections than we can have backends here because some
1610 * might still be authenticating; they might fail auth, or some
1611 * existing backend might exit before the auth cycle is completed. The
1612 * exact MaxBackends limit is enforced when a new backend tries to
1613 * join the shared-inval backend array.
1615 if (CountChildren() >= 2 * MaxBackends)
1623 * ConnCreate -- create a local connection data structure
1626 ConnCreate(int serverFd)
1630 if (!(port = (Port *) calloc(1, sizeof(Port))))
1633 (errcode(ERRCODE_OUT_OF_MEMORY),
1634 errmsg("out of memory")));
1638 if (StreamConnection(serverFd, port) != STATUS_OK)
1640 StreamClose(port->sock);
1647 * Precompute password salt values to use for this connection.
1648 * It's slightly annoying to do this long in advance of knowing
1649 * whether we'll need 'em or not, but we must do the random()
1650 * calls before we fork, not after. Else the postmaster's random
1651 * sequence won't get advanced, and all backends would end up
1652 * using the same salt...
1654 RandomSalt(port->cryptSalt, port->md5Salt);
1662 * ConnFree -- free a local connection data structure
1665 ConnFree(Port *conn)
1675 * ClosePostmasterPorts -- close all the postmaster's open sockets
1677 * This is called during child process startup to release file descriptors
1678 * that are not needed by that child process. The postmaster still has
1679 * them open, of course.
1682 ClosePostmasterPorts(bool pgstat_too)
1686 /* Close the listen sockets */
1687 for (i = 0; i < MAXLISTEN; i++)
1689 if (ListenSocket[i] != -1)
1691 StreamClose(ListenSocket[i]);
1692 ListenSocket[i] = -1;
1696 /* Close pgstat control sockets, unless we're starting pgstat itself */
1698 pgstat_close_sockets();
1703 * reset_shared -- reset shared memory and semaphores
1706 reset_shared(unsigned short port)
1709 * Create or re-create shared memory and semaphores.
1711 * Note: in each "cycle of life" we will normally assign the same IPC
1712 * keys (if using SysV shmem and/or semas), since the port number is
1713 * used to determine IPC keys. This helps ensure that we will clean
1714 * up dead IPC objects if the postmaster crashes and is restarted.
1716 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1721 * SIGHUP -- reread config files, and tell children to do same
1724 SIGHUP_handler(SIGNAL_ARGS)
1726 int save_errno = errno;
1728 PG_SETMASK(&BlockSig);
1730 if (Shutdown <= SmartShutdown)
1733 (errmsg("received SIGHUP, reloading configuration files")));
1734 ProcessConfigFile(PGC_SIGHUP);
1736 write_nondefault_variables(PGC_SIGHUP);
1738 SignalChildren(SIGHUP);
1743 * Tell the background writer to terminate so that we
1744 * will start a new one with a possibly changed config
1746 if (BgWriterPID != 0)
1747 kill(BgWriterPID, SIGTERM);
1750 PG_SETMASK(&UnBlockSig);
1758 * pmdie -- signal handler for processing various postmaster signals.
1763 int save_errno = errno;
1765 PG_SETMASK(&BlockSig);
1768 (errmsg_internal("postmaster received signal %d",
1769 postgres_signal_arg)));
1771 switch (postgres_signal_arg)
1778 * Wait for children to end their work and ShutdownDataBase.
1780 if (Shutdown >= SmartShutdown)
1782 Shutdown = SmartShutdown;
1784 (errmsg("received smart shutdown request")));
1786 /* Must tell bgwriter to quit, or it never will... */
1787 if (BgWriterPID != 0)
1788 kill(BgWriterPID, SIGTERM);
1790 if (DLGetHead(BackendList)) /* let reaper() handle this */
1794 * No children left. Shutdown data base system.
1796 if (StartupPID > 0 || FatalError) /* let reaper() handle
1799 if (ShutdownPID > 0)
1801 elog(PANIC, "shutdown process %d already running",
1806 ShutdownPID = ShutdownDataBase();
1814 * Abort all children with SIGTERM (rollback active transactions
1815 * and exit) and ShutdownDataBase when they are gone.
1817 if (Shutdown >= FastShutdown)
1819 Shutdown = FastShutdown;
1821 (errmsg("received fast shutdown request")));
1823 if (DLGetHead(BackendList))
1828 (errmsg("aborting any active transactions")));
1829 SignalChildren(SIGTERM);
1830 /* reaper() does the rest */
1836 * No children left. Shutdown data base system.
1838 * Unlike the previous case, it is not an error for the shutdown
1839 * process to be running already (we could get SIGTERM followed
1840 * shortly later by SIGINT).
1842 if (StartupPID > 0 || FatalError) /* let reaper() handle
1845 if (ShutdownPID == 0)
1846 ShutdownPID = ShutdownDataBase();
1852 * Immediate Shutdown:
1854 * abort all children with SIGQUIT and exit without attempt to
1855 * properly shutdown data base system.
1858 (errmsg("received immediate shutdown request")));
1859 if (ShutdownPID > 0)
1860 kill(ShutdownPID, SIGQUIT);
1862 kill(StartupPID, SIGQUIT);
1863 if (DLGetHead(BackendList))
1864 SignalChildren(SIGQUIT);
1869 PG_SETMASK(&UnBlockSig);
1875 * Reaper -- signal handler to cleanup after a backend (child) dies.
1880 int save_errno = errno;
1883 int status; /* backend exit status */
1886 union wait status; /* backend exit status */
1890 int pid; /* process id of dead backend */
1892 PG_SETMASK(&BlockSig);
1895 (errmsg_internal("reaping dead processes")));
1897 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1899 exitstatus = status;
1902 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1904 exitstatus = status.w_status;
1906 while ((pid = win32_waitpid(&exitstatus)) > 0)
1909 * We need to do this here, and not in CleanupProc, since this
1910 * is to be called on all children when we are done with them.
1911 * Could move to LogChildExit, but that seems like asking for
1914 win32_RemoveChild(pid);
1919 * Check if this child was the statistics collector. If so, try to
1920 * start a new one. (If fail, we'll try again in future cycles of
1923 if (pgstat_ispgstat(pid))
1925 LogChildExit(LOG, gettext("statistics collector process"),
1932 * Check if this child was a shutdown or startup process.
1934 if (ShutdownPID > 0 && pid == ShutdownPID)
1936 if (exitstatus != 0)
1938 LogChildExit(LOG, gettext("shutdown process"),
1942 /* Normal postmaster exit is here */
1946 if (StartupPID > 0 && pid == StartupPID)
1948 if (exitstatus != 0)
1950 LogChildExit(LOG, gettext("startup process"),
1953 (errmsg("aborting startup due to startup process failure")));
1959 * Startup succeeded - remember its ID and RedoRecPtr.
1961 * NB: this MUST happen before we fork a checkpoint or shutdown
1962 * subprocess, else they will have wrong local ThisStartUpId.
1966 FatalError = false; /* done with recovery */
1969 * Arrange for first checkpoint to occur after standard delay.
1972 checkpointed = time(NULL);
1975 * Go to shutdown mode if a shutdown request was pending.
1977 if (Shutdown > NoShutdown)
1979 if (ShutdownPID > 0)
1981 elog(PANIC, "startup process %d died while shutdown process %d already running",
1982 pid, (int) ShutdownPID);
1985 ShutdownPID = ShutdownDataBase();
1992 * Else do standard child cleanup.
1994 CleanupProc(pid, exitstatus);
1996 } /* loop over pending child-death reports */
2001 * Wait for all children exit, then reset shmem and
2004 if (DLGetHead(BackendList) || StartupPID > 0 || ShutdownPID > 0)
2007 (errmsg("all server processes terminated; reinitializing")));
2010 reset_shared(PostPortNumber);
2012 StartupPID = StartupDataBase();
2017 if (Shutdown > NoShutdown)
2019 if (DLGetHead(BackendList))
2021 if (StartupPID > 0 || ShutdownPID > 0)
2023 ShutdownPID = ShutdownDataBase();
2027 PG_SETMASK(&UnBlockSig);
2033 * CleanupProc -- cleanup after terminated backend.
2035 * Remove all local state associated with backend.
2038 CleanupProc(int pid,
2039 int exitstatus) /* child's exit status. */
2045 LogChildExit(DEBUG2, gettext("child process"), pid, exitstatus);
2048 * If a backend dies in an ugly way (i.e. exit status not 0) then we
2049 * must signal all other backends to quickdie. If exit status is zero
2050 * we assume everything is hunky dory and simply remove the backend
2051 * from the active backend list.
2053 if (exitstatus == 0)
2055 curr = DLGetHead(BackendList);
2058 bp = (Backend *) DLE_VAL(curr);
2062 ShmemBackendArrayRemove(bp->pid);
2069 curr = DLGetSucc(curr);
2072 if (pid == CheckPointPID)
2077 checkpointed = time(NULL);
2078 /* Update RedoRecPtr for future child backends */
2079 GetSavedRedoRecPtr();
2082 else if (pid == BgWriterPID)
2090 /* below here we're dealing with a non-normal exit */
2092 /* Make log entry unless we did so already */
2096 (pid == CheckPointPID) ? gettext("checkpoint process") :
2097 (pid == BgWriterPID) ? gettext("bgwriter process") :
2098 gettext("server process"),
2101 (errmsg("terminating any other active server processes")));
2104 curr = DLGetHead(BackendList);
2107 next = DLGetSucc(curr);
2108 bp = (Backend *) DLE_VAL(curr);
2112 * This backend is still alive. Unless we did so already,
2113 * tell it to commit hara-kiri.
2115 * SIGQUIT is the special signal that says exit without proc_exit
2116 * and let the user know what's going on. But if SendStop is
2117 * set (-s on command line), then we send SIGSTOP instead, so
2118 * that we can get core dumps from all backends by hand.
2123 (errmsg_internal("sending %s to process %d",
2124 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2126 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2132 * Found entry for freshly-dead backend, so remove it.
2135 ShmemBackendArrayRemove(bp->pid);
2144 if (pid == CheckPointPID)
2149 else if (pid == BgWriterPID)
2156 * Tell the collector about backend termination
2165 * Log the death of a child process.
2168 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2170 if (WIFEXITED(exitstatus))
2174 * translator: %s is a noun phrase describing a child process,
2175 * such as "server process"
2177 (errmsg("%s (PID %d) exited with exit code %d",
2178 procname, pid, WEXITSTATUS(exitstatus))));
2179 else if (WIFSIGNALED(exitstatus))
2183 * translator: %s is a noun phrase describing a child process,
2184 * such as "server process"
2186 (errmsg("%s (PID %d) was terminated by signal %d",
2187 procname, pid, WTERMSIG(exitstatus))));
2192 * translator: %s is a noun phrase describing a child process,
2193 * such as "server process"
2195 (errmsg("%s (PID %d) exited with unexpected status %d",
2196 procname, pid, exitstatus)));
2200 * Send a signal to all backend children.
2203 SignalChildren(int signal)
2209 curr = DLGetHead(BackendList);
2212 next = DLGetSucc(curr);
2213 bp = (Backend *) DLE_VAL(curr);
2215 if (bp->pid != MyProcPid)
2218 (errmsg_internal("sending signal %d to process %d",
2221 kill(bp->pid, signal);
2229 * BackendStartup -- start backend process
2231 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2234 BackendStartup(Port *port)
2236 Backend *bn; /* for backend cleanup */
2239 #ifdef LINUX_PROFILE
2240 struct itimerval prof_itimer;
2244 * Compute the cancel key that will be assigned to this backend. The
2245 * backend will have its own copy in the forked-off process' value of
2246 * MyCancelKey, so that it can transmit the key to the frontend.
2248 MyCancelKey = PostmasterRandom();
2251 * Make room for backend data structure. Better before the fork() so
2252 * we can handle failure cleanly.
2254 bn = (Backend *) malloc(sizeof(Backend));
2258 (errcode(ERRCODE_OUT_OF_MEMORY),
2259 errmsg("out of memory")));
2260 return STATUS_ERROR;
2264 * Flush stdio channels just before fork, to avoid double-output
2265 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2266 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2267 * coredump if we do. Presently stdout and stderr are the only stdio
2268 * output channels used by the postmaster, so fflush'ing them should
2274 #ifdef LINUX_PROFILE
2277 * Linux's fork() resets the profiling timer in the child process. If
2278 * we want to profile child processes then we need to save and restore
2279 * the timer setting. This is a waste of time if not profiling,
2280 * however, so only do it if commanded by specific -DLINUX_PROFILE
2283 getitimer(ITIMER_PROF, &prof_itimer);
2287 /* Specific beos actions before backend startup */
2288 beos_before_backend_startup();
2291 port->canAcceptConnections = canAcceptConnections();
2293 pid = Backend_forkexec(port);
2297 if (pid == 0) /* child */
2299 #ifdef LINUX_PROFILE
2300 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2304 /* Specific beos backend startup actions */
2305 beos_backend_startup();
2309 proc_exit(BackendRun(port));
2313 /* in parent, error */
2316 int save_errno = errno;
2319 /* Specific beos backend startup actions */
2320 beos_backend_startup_failed();
2325 (errmsg("could not fork new process for connection: %m")));
2326 report_fork_failure_to_client(port, save_errno);
2327 return STATUS_ERROR;
2330 /* in parent, normal */
2332 (errmsg_internal("forked new backend, pid=%d socket=%d",
2333 (int) pid, port->sock)));
2336 * Everything's been successful, it's safe to add this backend to our
2340 bn->cancel_key = MyCancelKey;
2342 ShmemBackendArrayAdd(bn);
2344 DLAddHead(BackendList, DLNewElem(bn));
2350 * Try to report backend fork() failure to client before we close the
2351 * connection. Since we do not care to risk blocking the postmaster on
2352 * this connection, we set the connection to non-blocking and try only once.
2354 * This is grungy special-purpose code; we cannot use backend libpq since
2355 * it's not up and running.
2358 report_fork_failure_to_client(Port *port, int errnum)
2362 /* Format the error message packet (always V2 protocol) */
2363 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2364 gettext("could not fork new process for connection: "),
2367 /* Set port to non-blocking. Don't do send() if this fails */
2368 if (FCNTL_NONBLOCK(port->sock) < 0)
2371 send(port->sock, buffer, strlen(buffer) + 1, 0);
2376 * split_opts -- split a string of options and append it to an argv array
2378 * NB: the string is destructively modified!
2380 * Since no current POSTGRES arguments require any quoting characters,
2381 * we can use the simple-minded tactic of assuming each set of space-
2382 * delimited characters is a separate argv element.
2384 * If you don't like that, well, we *used* to pass the whole option string
2385 * as ONE argument to execl(), which was even less intelligent...
2388 split_opts(char **argv, int *argcp, char *s)
2392 while (isspace((unsigned char) *s))
2396 argv[(*argcp)++] = s;
2397 while (*s && !isspace((unsigned char) *s))
2406 * BackendInit/Run -- perform authentication [BackendInit], and if successful,
2407 * set up the backend's argument list [BackendRun] and invoke
2411 * Shouldn't return at all.
2412 * If PostgresMain() fails, return status.
2415 BackendInit(Port *port)
2420 char remote_host[NI_MAXHOST];
2421 char remote_port[NI_MAXSERV];
2423 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2425 ClientAuthInProgress = true; /* limit visibility of log messages */
2427 /* We don't want the postmaster's proc_exit() handlers */
2431 * Signal handlers setting is moved to tcop/postgres...
2434 /* save start time for end of session reporting */
2435 gettimeofday(&(port->session_start),NULL);
2437 /* set these to empty in case they are needed before we set them up */
2438 port->remote_host = "";
2439 port->remote_port = "";
2441 /* Save port etc. for ps status */
2444 /* Reset MyProcPid to new backend's pid */
2445 MyProcPid = getpid();
2448 * Initialize libpq and enable reporting of ereport errors to the
2449 * client. Must do this now because authentication uses libpq to send
2452 pq_init(); /* initialize libpq to talk to client */
2453 whereToSendOutput = Remote; /* now safe to ereport to client */
2456 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2457 * during any client authentication related communication. Otherwise
2458 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2459 * if a buggy client blocks a backend during authentication.
2461 pqsignal(SIGTERM, authdie);
2462 pqsignal(SIGQUIT, authdie);
2463 pqsignal(SIGALRM, authdie);
2464 PG_SETMASK(&AuthBlockSig);
2467 * Get the remote host name and port for logging and status display.
2469 remote_host[0] = '\0';
2470 remote_port[0] = '\0';
2471 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2472 remote_host, sizeof(remote_host),
2473 remote_port, sizeof(remote_port),
2474 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2476 getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2477 remote_host, sizeof(remote_host),
2478 remote_port, sizeof(remote_port),
2479 NI_NUMERICHOST | NI_NUMERICSERV);
2482 if (Log_connections)
2484 (errmsg("connection received: host=%s port=%s",
2485 remote_host, remote_port)));
2489 /* modify remote_host for use in ps status */
2490 char tmphost[NI_MAXHOST];
2492 snprintf(tmphost, sizeof(tmphost), "%s:%s", remote_host, remote_port);
2493 StrNCpy(remote_host, tmphost, sizeof(remote_host));
2497 * save remote_host and remote_port in port stucture
2499 port->remote_host = strdup(remote_host);
2500 port->remote_port = strdup(remote_port);
2503 * Ready to begin client interaction. We will give up and exit(0)
2504 * after a time delay, so that a broken client can't hog a connection
2505 * indefinitely. PreAuthDelay doesn't count against the time limit.
2507 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2508 elog(FATAL, "could not set timer for authorization timeout");
2511 * Receive the startup packet (which might turn out to be a cancel
2514 status = ProcessStartupPacket(port, false);
2516 if (status != STATUS_OK)
2520 * Now that we have the user and database name, we can set the process
2521 * title for ps. It's good to do this as early as possible in
2524 init_ps_display(port->user_name, port->database_name, remote_host);
2525 set_ps_display("authentication");
2528 * Now perform authentication exchange.
2530 ClientAuthentication(port); /* might not return, if failure */
2533 * Done with authentication. Disable timeout, and prevent
2534 * SIGTERM/SIGQUIT again until backend startup is complete.
2536 if (!disable_sig_alarm(false))
2537 elog(FATAL, "could not disable timer for authorization timeout");
2538 PG_SETMASK(&BlockSig);
2540 if (Log_connections)
2542 (errmsg("connection authorized: user=%s database=%s",
2543 port->user_name, port->database_name)));
2546 * Don't want backend to be able to see the postmaster random number
2547 * generator state. We have to clobber the static random_seed *and*
2548 * start a new random sequence in the random() library function.
2551 gettimeofday(&now, &tz);
2552 srandom((unsigned int) now.tv_usec);
2557 BackendRun(Port *port)
2567 * Let's clean up ourselves as the postmaster child, and
2568 * close the postmaster's other sockets
2570 ClosePostmasterPorts(true);
2573 * PreAuthDelay is a debugging aid for investigating problems in the
2574 * authentication cycle: it can be set in postgresql.conf to allow
2575 * time to attach to the newly-forked backend with a debugger. (See
2576 * also the -W backend switch, which we allow clients to pass through
2577 * PGOPTIONS, but it is not honored until after authentication.)
2579 if (PreAuthDelay > 0)
2580 sleep(PreAuthDelay);
2582 /* Will exit on failure */
2587 * Now, build the argv vector that will be given to PostgresMain.
2589 * The layout of the command line is
2590 * postgres [secure switches] -p databasename [insecure switches]
2591 * where the switches after -p come from the client request.
2593 * The maximum possible number of commandline arguments that could come
2594 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2598 maxac = 10; /* for fixed args supplied below */
2599 maxac += (strlen(ExtraOptions) + 1) / 2;
2600 if (port->cmdline_options)
2601 maxac += (strlen(port->cmdline_options) + 1) / 2;
2603 av = (char **) MemoryContextAlloc(TopMemoryContext,
2604 maxac * sizeof(char *));
2607 av[ac++] = "postgres";
2610 * Pass the requested debugging level along to the backend.
2614 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2615 av[ac++] = debugbuf;
2619 * Pass any backend switches specified with -o in the postmaster's own
2620 * command line. We assume these are secure.
2622 split_opts(av, &ac, ExtraOptions);
2624 /* Tell the backend what protocol the frontend is using. */
2625 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2626 av[ac++] = protobuf;
2629 /* pass data dir before end of secure switches (-p) */
2635 * Tell the backend it is being called from the postmaster, and which
2636 * database to use. -p marks the end of secure switches.
2639 av[ac++] = port->database_name;
2642 * Pass the (insecure) option switches from the connection request.
2643 * (It's OK to mangle port->cmdline_options now.)
2645 if (port->cmdline_options)
2646 split_opts(av, &ac, port->cmdline_options);
2653 * Release postmaster's working memory context so that backend can
2654 * recycle the space. Note this does not trash *MyProcPort, because
2655 * ConnCreate() allocated that space with malloc() ... else we'd need
2656 * to copy the Port data here. Also, subsidiary data such as the
2657 * username isn't lost either; see ProcessStartupPacket().
2659 MemoryContextSwitchTo(TopMemoryContext);
2660 #ifndef EXEC_BACKEND
2661 MemoryContextDelete(PostmasterContext);
2663 PostmasterContext = NULL;
2666 * Debug: print arguments being passed to backend
2669 (errmsg_internal("%s child[%d]: starting with (",
2670 progname, getpid())));
2671 for (i = 0; i < ac; ++i)
2673 (errmsg_internal("\t%s", av[i])));
2675 (errmsg_internal(")")));
2677 ClientAuthInProgress = false; /* client_min_messages is active
2680 return (PostgresMain(ac, av, port->user_name));
2688 * SubPostmasterMain -- prepare the fork/exec'd process to be in an equivalent
2689 * state (for calling BackendRun) as a forked process.
2692 * Shouldn't return at all.
2695 SubPostmasterMain(int argc, char* argv[])
2697 unsigned long backendID;
2700 memset((void*)&port, 0, sizeof(Port));
2703 /* Do this sooner rather than later... */
2704 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2706 /* In EXEC case we will not have inherited these settings */
2707 IsPostmasterEnvironment = true;
2708 whereToSendOutput = None;
2710 /* Setup global context */
2711 MemoryContextInit();
2712 InitializeGUCOptions();
2714 /* Parse passed-in context */
2716 backendID = (unsigned long)atol(argv[argc++]);
2717 DataDir = strdup(argv[argc++]);
2719 /* Read in file-based context */
2720 read_nondefault_variables();
2721 read_backend_variables(backendID,&port);
2723 /* FIXME: [fork/exec] Ugh */
2729 /* Attach process to shared segments */
2730 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2733 proc_exit(BackendRun(&port));
2738 * Backend_forkexec -- fork/exec off a backend process
2741 * the pid of the fork/exec'd process
2744 Backend_forkexec(Port *port)
2748 int ac = 0, bufc = 0, i;
2749 char buf[2][MAXPGPATH];
2751 if (!write_backend_variables(port))
2752 return -1; /* log made by write_backend_variables */
2754 av[ac++] = "postgres";
2755 av[ac++] = "-forkexec";
2757 /* Format up context to pass to exec'd process */
2758 snprintf(buf[bufc++],MAXPGPATH,"%lu",tmpBackendFileNum);
2759 /* FIXME: [fork/exec] whitespaces in directories? */
2760 snprintf(buf[bufc++],MAXPGPATH,"%s",DataDir);
2762 /* Add to the arg list */
2763 Assert(bufc <= lengthof(buf));
2764 for (i = 0; i < bufc; i++)
2767 /* FIXME: [fork/exec] ExtraOptions? */
2770 Assert(ac <= lengthof(av));
2773 pid = win32_forkexec(pg_pathname,av); /* logs on error */
2775 /* Fire off execv in child */
2776 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
2778 * FIXME: [fork/exec] suggestions for what to do here?
2779 * Probably OK to issue error (unlike pgstat case)
2783 return pid; /* Parent returns pid */
2790 * ExitPostmaster -- cleanup
2792 * Do NOT call exit() directly --- always go through here!
2795 ExitPostmaster(int status)
2797 /* should cleanup shared memory and kill all backends */
2800 * Not sure of the semantics here. When the Postmaster dies, should
2801 * the backends all be killed? probably not.
2803 * MUST -- vadim 05-10-1999
2805 /* Should I use true instead? */
2806 ClosePostmasterPorts(false);
2812 * sigusr1_handler - handle signal conditions from child processes
2815 sigusr1_handler(SIGNAL_ARGS)
2817 int save_errno = errno;
2819 PG_SETMASK(&BlockSig);
2821 if (CheckPostmasterSignal(PMSIGNAL_DO_CHECKPOINT))
2823 if (CheckPointWarning != 0)
2826 * This only times checkpoints forced by running out of
2827 * segment files. Other checkpoints could reduce the
2828 * frequency of forced checkpoints.
2830 time_t now = time(NULL);
2832 if (LastSignalledCheckpoint != 0)
2834 int elapsed_secs = now - LastSignalledCheckpoint;
2836 if (elapsed_secs < CheckPointWarning)
2838 (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
2840 errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
2842 LastSignalledCheckpoint = now;
2846 * Request to schedule a checkpoint
2848 * Ignore request if checkpoint is already running or checkpointing
2849 * is currently disabled
2851 if (CheckPointPID == 0 && checkpointed &&
2852 StartupPID == 0 && Shutdown == NoShutdown &&
2853 !FatalError && random_seed != 0)
2855 CheckPointPID = CheckPointDataBase();
2856 /* note: if fork fails, CheckPointPID stays 0; nothing happens */
2860 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2863 * Password or group file has changed.
2869 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2872 * Send SIGUSR2 to all children (triggers AsyncNotifyHandler). See
2873 * storage/ipc/sinvaladt.c for the use of this.
2875 if (Shutdown == NoShutdown)
2876 SignalChildren(SIGUSR2);
2879 PG_SETMASK(&UnBlockSig);
2886 * Dummy signal handler
2888 * We use this for signals that we don't actually use in the postmaster,
2889 * but we do use in backends. If we SIG_IGN such signals in the postmaster,
2890 * then a newly started backend might drop a signal that arrives before it's
2891 * able to reconfigure its signal processing. (See notes in postgres.c.)
2894 dummy_handler(SIGNAL_ARGS)
2900 * CharRemap: given an int in range 0..61, produce textual encoding of it
2901 * per crypt(3) conventions.
2925 RandomSalt(char *cryptSalt, char *md5Salt)
2927 long rand = PostmasterRandom();
2929 cryptSalt[0] = CharRemap(rand % 62);
2930 cryptSalt[1] = CharRemap(rand / 62);
2933 * It's okay to reuse the first random value for one of the MD5 salt
2934 * bytes, since only one of the two salts will be sent to the client.
2935 * After that we need to compute more random bits.
2937 * We use % 255, sacrificing one possible byte value, so as to ensure
2938 * that all bits of the random() value participate in the result.
2939 * While at it, add one to avoid generating any null bytes.
2941 md5Salt[0] = (rand % 255) + 1;
2942 rand = PostmasterRandom();
2943 md5Salt[1] = (rand % 255) + 1;
2944 rand = PostmasterRandom();
2945 md5Salt[2] = (rand % 255) + 1;
2946 rand = PostmasterRandom();
2947 md5Salt[3] = (rand % 255) + 1;
2954 PostmasterRandom(void)
2956 static bool initialized = false;
2960 Assert(random_seed != 0);
2961 srandom(random_seed);
2969 * Count up number of child processes.
2978 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2980 bp = (Backend *) DLE_VAL(curr);
2981 if (bp->pid != MyProcPid)
2984 /* Checkpoint and bgwriter will be in the list, discount them */
2985 if (CheckPointPID != 0)
2987 if (BgWriterPID != 0)
2993 * Fire off a subprocess for startup/shutdown/checkpoint/bgwriter.
2995 * Return value of SSDataBase is subprocess' PID, or 0 if failed to start subprocess
2996 * (0 is returned only for checkpoint/bgwriter cases).
2998 * note: in the EXEC_BACKEND case, we delay the fork until argument list has been
3001 NON_EXEC_STATIC void
3002 SSDataBaseInit(int xlop)
3004 const char *statmsg;
3006 IsUnderPostmaster = true; /* we are a postmaster subprocess
3010 /* In EXEC case we will not have inherited these settings */
3011 IsPostmasterEnvironment = true;
3012 whereToSendOutput = None;
3015 MyProcPid = getpid(); /* reset MyProcPid */
3017 /* Lose the postmaster's on-exit routines and port connections */
3021 * Identify myself via ps
3025 case BS_XLOG_STARTUP:
3026 statmsg = "startup subprocess";
3028 case BS_XLOG_CHECKPOINT:
3029 statmsg = "checkpoint subprocess";
3031 case BS_XLOG_BGWRITER:
3032 statmsg = "bgwriter subprocess";
3034 case BS_XLOG_SHUTDOWN:
3035 statmsg = "shutdown subprocess";
3038 statmsg = "??? subprocess";
3041 init_ps_display(statmsg, "", "");
3047 SSDataBase(int xlop)
3051 #ifndef EXEC_BACKEND
3052 #ifdef LINUX_PROFILE
3053 struct itimerval prof_itimer;
3062 #ifndef EXEC_BACKEND
3063 #ifdef LINUX_PROFILE
3064 /* see comments in BackendRun */
3065 getitimer(ITIMER_PROF, &prof_itimer);
3069 /* Specific beos actions before backend startup */
3070 beos_before_backend_startup();
3073 /* Non EXEC_BACKEND case; fork here */
3074 if ((pid = fork()) == 0) /* child */
3082 #ifndef EXEC_BACKEND
3083 #ifdef LINUX_PROFILE
3084 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3088 /* Specific beos actions after backend startup */
3089 beos_backend_startup();
3092 /* Close the postmaster's sockets */
3093 ClosePostmasterPorts(true);
3095 SSDataBaseInit(xlop);
3097 if (!write_backend_variables(NULL))
3098 return -1; /* log issued by write_backend_variables */
3101 /* Set up command-line arguments for subprocess */
3102 av[ac++] = "postgres";
3107 snprintf(nbbuf, sizeof(nbbuf), "-B%d", NBuffers);
3110 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
3114 /* pass data dir before end of secure switches (-p) */
3118 /* and the backend identifier + dbname */
3119 snprintf(idbuf, sizeof(idbuf), "-p%lu,template1", tmpBackendFileNum);
3123 av[ac++] = "template1";
3128 Assert(ac < lengthof(av));
3131 /* EXEC_BACKEND case; fork/exec here */
3133 pid = win32_forkexec(pg_pathname,av); /* logs on error */
3135 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
3138 elog(ERROR,"unable to execv in SSDataBase: %m");
3143 BootstrapMain(ac, av);
3151 #ifndef EXEC_BACKEND
3153 /* Specific beos actions before backend startup */
3154 beos_backend_startup_failed();
3159 case BS_XLOG_STARTUP:
3161 (errmsg("could not fork startup process: %m")));
3163 case BS_XLOG_CHECKPOINT:
3165 (errmsg("could not fork checkpoint process: %m")));
3167 case BS_XLOG_BGWRITER:
3169 (errmsg("could not fork bgwriter process: %m")));
3171 case BS_XLOG_SHUTDOWN:
3173 (errmsg("could not fork shutdown process: %m")));
3177 (errmsg("could not fork process: %m")));
3182 * fork failure is fatal during startup/shutdown, but there's no
3183 * need to choke if a routine checkpoint or starting a background
3186 if (xlop == BS_XLOG_CHECKPOINT)
3188 if (xlop == BS_XLOG_BGWRITER)
3194 * The startup and shutdown processes are not considered normal
3195 * backends, but the checkpoint and bgwriter processes are.
3196 * They must be added to the list of backends.
3198 if (xlop == BS_XLOG_CHECKPOINT || xlop == BS_XLOG_BGWRITER)
3200 if (!(bn = (Backend *) malloc(sizeof(Backend))))
3203 (errcode(ERRCODE_OUT_OF_MEMORY),
3204 errmsg("out of memory")));
3209 bn->cancel_key = PostmasterRandom();
3211 ShmemBackendArrayAdd(bn);
3213 DLAddHead(BackendList, DLNewElem(bn));
3216 * Since this code is executed periodically, it's a fine place to
3217 * do other actions that should happen every now and then on no
3218 * particular schedule. Such as...
3221 TouchSocketLockFile();
3229 * Create the opts file
3232 CreateOptsFile(int argc, char *argv[])
3234 char fullprogname[MAXPGPATH];
3235 char filename[MAXPGPATH];
3239 if (FindExec(fullprogname, argv[0], "postmaster") < 0)
3242 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3244 if ((fp = fopen(filename, "w")) == NULL)
3246 elog(LOG, "could not create file \"%s\": %m", filename);
3250 fprintf(fp, "%s", fullprogname);
3251 for (i = 1; i < argc; i++)
3252 fprintf(fp, " '%s'", argv[i]);
3257 elog(LOG, "could not write file \"%s\": %m", filename);
3265 * This should be used only for reporting "interactive" errors (essentially,
3266 * bogus arguments on the command line). Once the postmaster is launched,
3267 * use ereport. In particular, don't use this for anything that occurs
3268 * after pmdaemonize.
3271 postmaster_error(const char *fmt,...)
3275 fprintf(stderr, "%s: ", progname);
3277 vfprintf(stderr, gettext(fmt), ap);
3279 fprintf(stderr, "\n");
3286 * The following need to be available to the read/write_backend_variables
3289 #include "storage/spin.h"
3290 extern XLogRecPtr RedoRecPtr;
3291 extern XLogwrtResult LogwrtResult;
3292 extern slock_t *ShmemLock;
3293 extern slock_t *ShmemIndexLock;
3294 extern void *ShmemIndexAlloc;
3295 typedef struct LWLock LWLock;
3296 extern LWLock *LWLockArray;
3297 extern slock_t *ProcStructLock;
3298 extern int pgStatSock;
3300 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3301 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3302 #define get_tmp_backend_file_name(buf,id) \
3306 "%s/%s/%s.backend_var.%lu", \
3308 PG_TEMP_FILES_DIR, \
3309 PG_TEMP_FILE_PREFIX, \
3314 write_backend_variables(Port *port)
3316 char filename[MAXPGPATH];
3318 get_tmp_backend_file_name(filename,++tmpBackendFileNum);
3321 fp = AllocateFile(filename, PG_BINARY_W);
3324 /* As per OpenTemporaryFile... */
3325 char dirname[MAXPGPATH];
3326 sprintf(dirname,"%s/%s",DataDir,PG_TEMP_FILES_DIR);
3327 mkdir(dirname, S_IRWXU);
3329 fp = AllocateFile(filename, PG_BINARY_W);
3333 (errcode_for_file_access(),
3334 errmsg("could not write to file \"%s\": %m", filename)));
3342 write_var(port->sock,fp);
3343 write_var(port->proto,fp);
3344 write_var(port->laddr,fp);
3345 write_var(port->raddr,fp);
3346 write_var(port->canAcceptConnections,fp);
3347 write_var(port->cryptSalt,fp);
3348 write_var(port->md5Salt,fp);
3350 write_var(MyCancelKey,fp);
3352 write_var(RedoRecPtr,fp);
3353 write_var(LogwrtResult,fp);
3355 write_var(UsedShmemSegID,fp);
3356 write_var(UsedShmemSegAddr,fp);
3358 write_var(ShmemLock,fp);
3359 write_var(ShmemIndexLock,fp);
3360 write_var(ShmemVariableCache,fp);
3361 write_var(ShmemIndexAlloc,fp);
3362 write_var(ShmemBackendArray,fp);
3364 write_var(LWLockArray,fp);
3365 write_var(ProcStructLock,fp);
3366 write_var(pgStatSock,fp);
3368 write_var(PreAuthDelay,fp);
3369 write_var(debug_flag,fp);
3370 write_var(PostmasterPid,fp);
3376 (errcode_for_file_access(),
3377 errmsg("could not write to file \"%s\": %m", filename)));
3385 read_backend_variables(unsigned long id, Port *port)
3387 char filename[MAXPGPATH];
3389 get_tmp_backend_file_name(filename,id);
3392 fp = AllocateFile(filename, PG_BINARY_R);
3396 (errcode_for_file_access(),
3397 errmsg("could not read from backend_variables file \"%s\": %m", filename)));
3404 read_var(port->sock,fp);
3405 read_var(port->proto,fp);
3406 read_var(port->laddr,fp);
3407 read_var(port->raddr,fp);
3408 read_var(port->canAcceptConnections,fp);
3409 read_var(port->cryptSalt,fp);
3410 read_var(port->md5Salt,fp);
3412 read_var(MyCancelKey,fp);
3414 read_var(RedoRecPtr,fp);
3415 read_var(LogwrtResult,fp);
3417 read_var(UsedShmemSegID,fp);
3418 read_var(UsedShmemSegAddr,fp);
3420 read_var(ShmemLock,fp);
3421 read_var(ShmemIndexLock,fp);
3422 read_var(ShmemVariableCache,fp);
3423 read_var(ShmemIndexAlloc,fp);
3424 read_var(ShmemBackendArray,fp);
3426 read_var(LWLockArray,fp);
3427 read_var(ProcStructLock,fp);
3428 read_var(pgStatSock,fp);
3430 read_var(PreAuthDelay,fp);
3431 read_var(debug_flag,fp);
3432 read_var(PostmasterPid,fp);
3436 if (unlink(filename) != 0)
3438 (errcode_for_file_access(),
3439 errmsg("could not remove file \"%s\": %m", filename)));
3443 size_t ShmemBackendArraySize(void)
3445 return (NUM_BACKENDARRAY_ELEMS*sizeof(Backend));
3448 void ShmemBackendArrayAllocation(void)
3450 size_t size = ShmemBackendArraySize();
3451 ShmemBackendArray = (Backend*)ShmemAlloc(size);
3452 memset(ShmemBackendArray, 0, size);
3455 static void ShmemBackendArrayAdd(Backend *bn)
3458 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3460 /* Find an empty slot */
3461 if (ShmemBackendArray[i].pid == 0)
3463 ShmemBackendArray[i] = *bn;
3468 /* FIXME: [fork/exec] some sort of error */
3472 static void ShmemBackendArrayRemove(pid_t pid)
3475 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3477 if (ShmemBackendArray[i].pid == pid)
3479 /* Mark the slot as empty */
3480 ShmemBackendArray[i].pid = 0;
3485 /* Something stronger than WARNING here? */
3487 (errmsg_internal("unable to find backend entry with pid %d",
3495 pid_t win32_forkexec(const char* path, char *argv[])
3498 PROCESS_INFORMATION pi;
3501 char cmdLine[MAXPGPATH];
3502 HANDLE childHandleCopy;
3503 HANDLE waiterThread;
3505 /* Format the cmd line */
3506 snprintf(cmdLine,sizeof(cmdLine),"%s",path);
3508 while (argv[++i] != NULL)
3510 /* FIXME: [fork/exec] some strlen checks might be prudent here */
3511 strcat(cmdLine," ");
3512 strcat(cmdLine,argv[i]);
3516 * The following snippet can disappear when we consistently
3517 * use forward slashes.
3520 while (*(p++) != '\0')
3521 if (*p == '/') *p = '\\';
3523 memset(&pi,0,sizeof(pi));
3524 memset(&si,0,sizeof(si));
3526 if (!CreateProcess(NULL,cmdLine,NULL,NULL,TRUE,0,NULL,NULL,&si,&pi))
3528 elog(ERROR,"CreateProcess call failed (%d): %m",GetLastError());
3532 if (!IsUnderPostmaster)
3533 /* We are the Postmaster creating a child... */
3534 win32_AddChild(pi.dwProcessId,pi.hProcess);
3536 if (!DuplicateHandle(GetCurrentProcess(),
3538 GetCurrentProcess(),
3542 DUPLICATE_SAME_ACCESS))
3544 (errmsg_internal("failed to duplicate child handle: %i",GetLastError())));
3545 waiterThread = CreateThread(NULL, 64*1024, win32_sigchld_waiter, (LPVOID)childHandleCopy, 0, NULL);
3548 (errmsg_internal("failed to create sigchld waiter thread: %i",GetLastError())));
3549 CloseHandle(waiterThread);
3551 if (IsUnderPostmaster)
3552 CloseHandle(pi.hProcess);
3553 CloseHandle(pi.hThread);
3555 return pi.dwProcessId;
3559 * Note: The following three functions must not be interrupted (eg. by signals).
3560 * As the Postgres Win32 signalling architecture (currently) requires polling,
3561 * or APC checking functions which aren't used here, this is not an issue.
3563 * We keep two separate arrays, instead of a single array of pid/HANDLE structs,
3564 * to avoid having to re-create a handle array for WaitForMultipleObjects on
3565 * each call to win32_waitpid.
3568 static void win32_AddChild(pid_t pid, HANDLE handle)
3570 Assert(win32_childPIDArray && win32_childHNDArray);
3571 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3573 win32_childPIDArray[win32_numChildren] = pid;
3574 win32_childHNDArray[win32_numChildren] = handle;
3575 ++win32_numChildren;
3578 /* FIXME: [fork/exec] some sort of error */
3582 static void win32_RemoveChild(pid_t pid)
3585 Assert(win32_childPIDArray && win32_childHNDArray);
3587 for (i = 0; i < win32_numChildren; i++)
3589 if (win32_childPIDArray[i] == pid)
3591 CloseHandle(win32_childHNDArray[i]);
3593 /* Swap last entry into the "removed" one */
3594 --win32_numChildren;
3595 win32_childPIDArray[i] = win32_childPIDArray[win32_numChildren];
3596 win32_childHNDArray[i] = win32_childHNDArray[win32_numChildren];
3601 /* Something stronger than WARNING here? */
3603 (errmsg_internal("unable to find child entry with pid %d",
3607 static pid_t win32_waitpid(int *exitstatus)
3609 Assert(win32_childPIDArray && win32_childHNDArray);
3610 elog(DEBUG3,"waiting on %d children",win32_numChildren);
3612 if (win32_numChildren > 0)
3615 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't
3616 * want to run queued APCs here.
3620 DWORD ret = WaitForMultipleObjects(win32_numChildren,win32_childHNDArray,FALSE,0);
3626 (errmsg_internal("failed to wait on %d children: %i",
3627 win32_numChildren,GetLastError())));
3628 /* Fall through to WAIT_TIMEOUTs return */
3631 /* No children have finished */
3635 /* Get the exit code, and return the PID of, the respective process */
3636 index = ret-WAIT_OBJECT_0;
3637 Assert(index >= 0 && index < win32_numChildren);
3638 if (!GetExitCodeProcess(win32_childHNDArray[index],&exitCode))
3640 * If we get this far, this should never happen, but, then again...
3641 * No choice other than to assume a catastrophic failure.
3644 (errmsg_internal("failed to get exit code for child %d",
3645 win32_childPIDArray[index])));
3646 *exitstatus = (int)exitCode;
3647 return win32_childPIDArray[index];
3655 /* Note! Code belows executes on separate threads, one for
3656 each child process created */
3657 static DWORD WINAPI win32_sigchld_waiter(LPVOID param) {
3658 HANDLE procHandle = (HANDLE)param;
3660 DWORD r = WaitForSingleObject(procHandle, INFINITE);
3661 if (r == WAIT_OBJECT_0)
3662 pg_queue_signal(SIGCHLD);
3664 fprintf(stderr,"ERROR: Failed to wait on child process handle: %i\n",GetLastError());
3665 CloseHandle(procHandle);