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 1.367 2004/02/17 03:54:56 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 * Initialize the list of active backends.
845 BackendList = DLNewList();
849 * Initialize the child pid/HANDLE arrays
851 /* FIXME: [fork/exec] Ideally, we would resize these arrays with changes
852 * in MaxBackends, but this'll do as a first order solution.
854 win32_childPIDArray = (pid_t*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(pid_t));
855 win32_childHNDArray = (HANDLE*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(HANDLE));
856 if (!win32_childPIDArray || !win32_childHNDArray)
858 (errcode(ERRCODE_OUT_OF_MEMORY),
859 errmsg("out of memory")));
863 * Record postmaster options. We delay this till now to avoid
864 * recording bogus options (eg, NBuffers too high for available
867 if (!CreateOptsFile(argc, argv))
871 * Set up signal handlers for the postmaster process.
873 * CAUTION: when changing this list, check for side-effects on the signal
874 * handling setup of child processes. See tcop/postgres.c,
875 * bootstrap/bootstrap.c, and postmaster/pgstat.c.
878 PG_SETMASK(&BlockSig);
880 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
881 * children do same */
882 pqsignal(SIGINT, pmdie); /* send SIGTERM and ShutdownDataBase */
883 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
884 pqsignal(SIGTERM, pmdie); /* wait for children and ShutdownDataBase */
885 pqsignal(SIGALRM, SIG_IGN); /* ignored */
886 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
887 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
888 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
889 pqsignal(SIGCHLD, reaper); /* handle child termination */
890 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
891 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
892 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
894 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
898 * Reset whereToSendOutput from Debug (its starting state) to None.
899 * This prevents ereport from sending log messages to stderr unless
900 * the syslog/stderr switch permits. We don't do this until the
901 * postmaster is fully launched, since startup failures may as well be
902 * reported to stderr.
904 whereToSendOutput = None;
907 * On many platforms, the first call of localtime() incurs significant
908 * overhead to load timezone info from the system configuration files.
909 * By doing it once in the postmaster, we avoid having to do it in
910 * every started child process. The savings are not huge, but they
914 time_t now = time(NULL);
916 (void) localtime(&now);
920 * Initialize and try to startup the statistics collector process
926 * Load cached files for client authentication.
934 * We're ready to rock and roll...
936 StartupPID = StartupDataBase();
938 status = ServerLoop();
941 * ServerLoop probably shouldn't ever return, but if it does, close
944 ExitPostmaster(status != STATUS_OK);
946 return 0; /* not reached */
950 pmdaemonize(int argc, char *argv[])
956 struct itimerval prof_itimer;
960 /* see comments in BackendRun */
961 getitimer(ITIMER_PROF, &prof_itimer);
965 /* FIXME: [fork/exec] to be implemented? */
970 if (pid == (pid_t) -1)
972 postmaster_error("could not fork background process: %s",
978 /* Parent should just exit, without doing any atexit cleanup */
983 setitimer(ITIMER_PROF, &prof_itimer, NULL);
986 MyProcPid = getpid(); /* reset MyProcPid to child */
988 /* GH: If there's no setsid(), we hopefully don't need silent mode.
989 * Until there's a better solution.
994 postmaster_error("could not dissociate from controlling TTY: %s",
999 i = open(NULL_DEV, O_RDWR | PG_BINARY);
1009 * Print out help message
1012 usage(const char *progname)
1014 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
1015 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
1016 printf(gettext("Options:\n"));
1017 #ifdef USE_ASSERT_CHECKING
1018 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
1020 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
1021 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
1022 printf(gettext(" -d 1-5 debugging level\n"));
1023 printf(gettext(" -D DATADIR database directory\n"));
1024 printf(gettext(" -F turn fsync off\n"));
1025 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
1026 printf(gettext(" -i enable TCP/IP connections\n"));
1027 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
1029 printf(gettext(" -l enable SSL connections\n"));
1031 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
1032 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
1033 printf(gettext(" -p PORT port number to listen on\n"));
1034 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1035 printf(gettext(" --help show this help, then exit\n"));
1036 printf(gettext(" --version output version information, then exit\n"));
1038 printf(gettext("\nDeveloper options:\n"));
1039 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1040 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1042 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1043 "configuration settings and how to set them on the command line or in\n"
1044 "the configuration file.\n\n"
1045 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1058 gettimeofday(&now, &tz);
1060 nSockets = initMasks(&readmask);
1066 struct timeval timeout;
1069 * The timeout for the select() below is normally set on the basis
1070 * of the time to the next checkpoint. However, if for some
1071 * reason we don't have a next-checkpoint time, time out after 60
1072 * seconds. This keeps checkpoint scheduling from locking up when
1073 * we get new connection requests infrequently (since we are
1074 * likely to detect checkpoint completion just after enabling
1075 * signals below, after we've already made the decision about how
1076 * long to wait this time).
1078 timeout.tv_sec = 60;
1079 timeout.tv_usec = 0;
1081 if (CheckPointPID == 0 && checkpointed &&
1082 StartupPID == 0 && Shutdown == NoShutdown &&
1083 !FatalError && random_seed != 0)
1085 time_t now = time(NULL);
1087 if (CheckPointTimeout + checkpointed > now)
1090 * Not time for checkpoint yet, so set select timeout
1092 timeout.tv_sec = CheckPointTimeout + checkpointed - now;
1096 /* Time to make the checkpoint... */
1097 CheckPointPID = CheckPointDataBase();
1100 * if fork failed, schedule another try at 0.1 normal
1103 if (CheckPointPID == 0)
1105 timeout.tv_sec = CheckPointTimeout / 10;
1106 checkpointed = now + timeout.tv_sec - CheckPointTimeout;
1112 * If no background writer process is running and we should
1113 * do background writing, start one. It doesn't matter if
1114 * this fails, we'll just try again later.
1116 if (BgWriterPID == 0 && BgWriterPercent > 0 &&
1117 StartupPID == 0 && Shutdown == NoShutdown &&
1118 !FatalError && random_seed != 0)
1120 BgWriterPID = StartBackgroundWriter();
1124 * Wait for something to happen.
1126 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1128 PG_SETMASK(&UnBlockSig);
1130 if (select(nSockets, &rmask, NULL, NULL, &timeout) < 0)
1132 PG_SETMASK(&BlockSig);
1133 if (errno == EINTR || errno == EWOULDBLOCK)
1136 (errcode_for_socket_access(),
1137 errmsg("select() failed in postmaster: %m")));
1138 return STATUS_ERROR;
1142 * Block all signals until we wait again. (This makes it safe for
1143 * our signal handlers to do nontrivial work.)
1145 PG_SETMASK(&BlockSig);
1148 * Select a random seed at the time of first receiving a request.
1150 while (random_seed == 0)
1152 gettimeofday(&later, &tz);
1155 * We are not sure how much precision is in tv_usec, so we
1156 * swap the nibbles of 'later' and XOR them with 'now'. On the
1157 * off chance that the result is 0, we loop until it isn't.
1159 random_seed = now.tv_usec ^
1160 ((later.tv_usec << 16) |
1161 ((later.tv_usec >> 16) & 0xffff));
1165 * New connection pending on any of our sockets? If so, fork a
1166 * child process to deal with it.
1168 for (i = 0; i < MAXLISTEN; i++)
1170 if (ListenSocket[i] == -1)
1172 if (FD_ISSET(ListenSocket[i], &rmask))
1174 port = ConnCreate(ListenSocket[i]);
1177 BackendStartup(port);
1180 * We no longer need the open socket or port structure
1183 StreamClose(port->sock);
1189 /* If we have lost the stats collector, try to start a new one */
1190 if (!pgstat_is_running)
1197 * Initialise the masks for select() for the ports
1198 * we are listening on. Return the number of sockets to listen on.
1202 initMasks(fd_set *rmask)
1209 for (i = 0; i < MAXLISTEN; i++)
1211 int fd = ListenSocket[i];
1225 * Read the startup packet and do something according to it.
1227 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1228 * not return at all.
1230 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1231 * if that's what you want. Return STATUS_ERROR if you don't want to
1232 * send anything to the client, which would typically be appropriate
1233 * if we detect a communications failure.)
1236 ProcessStartupPacket(Port *port, bool SSLdone)
1240 ProtocolVersion proto;
1241 MemoryContext oldcontext;
1243 if (pq_getbytes((char *) &len, 4) == EOF)
1246 * EOF after SSLdone probably means the client didn't like our
1247 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1248 * so don't clutter the log with a complaint.
1252 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1253 errmsg("incomplete startup packet")));
1254 return STATUS_ERROR;
1260 if (len < (int32) sizeof(ProtocolVersion) ||
1261 len > MAX_STARTUP_PACKET_LENGTH)
1264 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1265 errmsg("invalid length of startup packet")));
1266 return STATUS_ERROR;
1270 * Allocate at least the size of an old-style startup packet, plus one
1271 * extra byte, and make sure all are zeroes. This ensures we will
1272 * have null termination of all strings, in both fixed- and
1273 * variable-length packet layouts.
1275 if (len <= (int32) sizeof(StartupPacket))
1276 buf = palloc0(sizeof(StartupPacket) + 1);
1278 buf = palloc0(len + 1);
1280 if (pq_getbytes(buf, len) == EOF)
1283 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1284 errmsg("incomplete startup packet")));
1285 return STATUS_ERROR;
1289 * The first field is either a protocol version number or a special
1292 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1294 if (proto == CANCEL_REQUEST_CODE)
1296 processCancelRequest(port, buf);
1297 return 127; /* XXX */
1300 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1305 /* No SSL when disabled or on Unix sockets */
1306 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1309 SSLok = 'S'; /* Support for SSL */
1311 SSLok = 'N'; /* No support for SSL */
1313 if (send(port->sock, &SSLok, 1, 0) != 1)
1316 (errcode_for_socket_access(),
1317 errmsg("failed to send SSL negotiation response: %m")));
1318 return STATUS_ERROR; /* close the connection */
1322 if (SSLok == 'S' && secure_open_server(port) == -1)
1323 return STATUS_ERROR;
1325 /* regular startup packet, cancel, etc packet should follow... */
1326 /* but not another SSL negotiation request */
1327 return ProcessStartupPacket(port, true);
1330 /* Could add additional special packet types here */
1333 * Set FrontendProtocol now so that ereport() knows what format to
1334 * send if we fail during startup.
1336 FrontendProtocol = proto;
1338 /* Check we can handle the protocol the frontend is using. */
1340 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1341 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1342 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1343 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1345 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1346 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1347 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1348 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1349 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1350 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1353 * Now fetch parameters out of startup packet and save them into the
1354 * Port structure. All data structures attached to the Port struct
1355 * must be allocated in TopMemoryContext so that they won't disappear
1356 * when we pass them to PostgresMain (see BackendRun). We need not
1357 * worry about leaking this storage on failure, since we aren't in the
1358 * postmaster process anymore.
1360 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1362 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1364 int32 offset = sizeof(ProtocolVersion);
1367 * Scan packet body for name/option pairs. We can assume any
1368 * string beginning within the packet body is null-terminated,
1369 * thanks to zeroing extra byte above.
1371 port->guc_options = NIL;
1373 while (offset < len)
1375 char *nameptr = ((char *) buf) + offset;
1379 if (*nameptr == '\0')
1380 break; /* found packet terminator */
1381 valoffset = offset + strlen(nameptr) + 1;
1382 if (valoffset >= len)
1383 break; /* missing value, will complain below */
1384 valptr = ((char *) buf) + valoffset;
1386 if (strcmp(nameptr, "database") == 0)
1387 port->database_name = pstrdup(valptr);
1388 else if (strcmp(nameptr, "user") == 0)
1389 port->user_name = pstrdup(valptr);
1390 else if (strcmp(nameptr, "options") == 0)
1391 port->cmdline_options = pstrdup(valptr);
1394 /* Assume it's a generic GUC option */
1395 port->guc_options = lappend(port->guc_options,
1397 port->guc_options = lappend(port->guc_options,
1400 offset = valoffset + strlen(valptr) + 1;
1404 * If we didn't find a packet terminator exactly at the end of the
1405 * given packet length, complain.
1407 if (offset != len - 1)
1409 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1410 errmsg("invalid startup packet layout: expected terminator as last byte")));
1415 * Get the parameters from the old-style, fixed-width-fields
1416 * startup packet as C strings. The packet destination was
1417 * cleared first so a short packet has zeros silently added. We
1418 * have to be prepared to truncate the pstrdup result for oversize
1421 StartupPacket *packet = (StartupPacket *) buf;
1423 port->database_name = pstrdup(packet->database);
1424 if (strlen(port->database_name) > sizeof(packet->database))
1425 port->database_name[sizeof(packet->database)] = '\0';
1426 port->user_name = pstrdup(packet->user);
1427 if (strlen(port->user_name) > sizeof(packet->user))
1428 port->user_name[sizeof(packet->user)] = '\0';
1429 port->cmdline_options = pstrdup(packet->options);
1430 if (strlen(port->cmdline_options) > sizeof(packet->options))
1431 port->cmdline_options[sizeof(packet->options)] = '\0';
1432 port->guc_options = NIL;
1435 /* Check a user name was given. */
1436 if (port->user_name == NULL || port->user_name[0] == '\0')
1438 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1439 errmsg("no PostgreSQL user name specified in startup packet")));
1441 /* The database defaults to the user name. */
1442 if (port->database_name == NULL || port->database_name[0] == '\0')
1443 port->database_name = pstrdup(port->user_name);
1445 if (Db_user_namespace)
1448 * If user@, it is a global user, remove '@'. We only want to do
1449 * this if there is an '@' at the end and no earlier in the user
1450 * string or they may fake as a local user of another database
1451 * attaching to this database.
1453 if (strchr(port->user_name, '@') ==
1454 port->user_name + strlen(port->user_name) - 1)
1455 *strchr(port->user_name, '@') = '\0';
1458 /* Append '@' and dbname */
1461 db_user = palloc(strlen(port->user_name) +
1462 strlen(port->database_name) + 2);
1463 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1464 port->user_name = db_user;
1469 * Truncate given database and user names to length of a Postgres
1470 * name. This avoids lookup failures when overlength names are given.
1472 if (strlen(port->database_name) >= NAMEDATALEN)
1473 port->database_name[NAMEDATALEN - 1] = '\0';
1474 if (strlen(port->user_name) >= NAMEDATALEN)
1475 port->user_name[NAMEDATALEN - 1] = '\0';
1478 * Done putting stuff in TopMemoryContext.
1480 MemoryContextSwitchTo(oldcontext);
1483 * If we're going to reject the connection due to database state, say
1484 * so now instead of wasting cycles on an authentication exchange.
1485 * (This also allows a pg_ping utility to be written.)
1487 switch (port->canAcceptConnections)
1491 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1492 errmsg("the database system is starting up")));
1496 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1497 errmsg("the database system is shutting down")));
1501 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1502 errmsg("the database system is in recovery mode")));
1506 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1507 errmsg("sorry, too many clients already")));
1519 * The client has sent a cancel request packet, not a normal
1520 * start-a-new-connection packet. Perform the necessary processing.
1521 * Nothing is sent back to the client.
1524 processCancelRequest(Port *port, void *pkt)
1526 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1528 long cancelAuthCode;
1530 #ifndef EXEC_BACKEND
1536 backendPID = (int) ntohl(canc->backendPID);
1537 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1539 if (backendPID == CheckPointPID)
1542 (errmsg_internal("ignoring cancel request for checkpoint process %d",
1546 else if (backendPID == BgWriterPID)
1549 (errmsg_internal("ignoring cancel request for bgwriter process %d",
1554 /* See if we have a matching backend */
1555 #ifndef EXEC_BACKEND
1556 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1558 bp = (Backend *) DLE_VAL(curr);
1560 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1562 bp = (Backend*) &ShmemBackendArray[i];
1564 if (bp->pid == backendPID)
1566 if (bp->cancel_key == cancelAuthCode)
1568 /* Found a match; signal that backend to cancel current op */
1570 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1572 kill(bp->pid, SIGINT);
1575 /* Right PID, wrong key: no way, Jose */
1577 (errmsg_internal("bad key in cancel request for process %d",
1583 /* No matching backend */
1585 (errmsg_internal("bad pid in cancel request for process %d",
1590 * canAcceptConnections --- check to see if database state allows connections.
1592 static enum CAC_state
1593 canAcceptConnections(void)
1595 /* Can't start backends when in startup/shutdown/recovery state. */
1596 if (Shutdown > NoShutdown)
1597 return CAC_SHUTDOWN;
1601 return CAC_RECOVERY;
1604 * Don't start too many children.
1606 * We allow more connections than we can have backends here because some
1607 * might still be authenticating; they might fail auth, or some
1608 * existing backend might exit before the auth cycle is completed. The
1609 * exact MaxBackends limit is enforced when a new backend tries to
1610 * join the shared-inval backend array.
1612 if (CountChildren() >= 2 * MaxBackends)
1620 * ConnCreate -- create a local connection data structure
1623 ConnCreate(int serverFd)
1627 if (!(port = (Port *) calloc(1, sizeof(Port))))
1630 (errcode(ERRCODE_OUT_OF_MEMORY),
1631 errmsg("out of memory")));
1635 if (StreamConnection(serverFd, port) != STATUS_OK)
1637 StreamClose(port->sock);
1644 * Precompute password salt values to use for this connection.
1645 * It's slightly annoying to do this long in advance of knowing
1646 * whether we'll need 'em or not, but we must do the random()
1647 * calls before we fork, not after. Else the postmaster's random
1648 * sequence won't get advanced, and all backends would end up
1649 * using the same salt...
1651 RandomSalt(port->cryptSalt, port->md5Salt);
1659 * ConnFree -- free a local connection data structure
1662 ConnFree(Port *conn)
1672 * ClosePostmasterPorts -- close all the postmaster's open sockets
1674 * This is called during child process startup to release file descriptors
1675 * that are not needed by that child process. The postmaster still has
1676 * them open, of course.
1679 ClosePostmasterPorts(bool pgstat_too)
1683 /* Close the listen sockets */
1684 for (i = 0; i < MAXLISTEN; i++)
1686 if (ListenSocket[i] != -1)
1688 StreamClose(ListenSocket[i]);
1689 ListenSocket[i] = -1;
1693 /* Close pgstat control sockets, unless we're starting pgstat itself */
1695 pgstat_close_sockets();
1700 * reset_shared -- reset shared memory and semaphores
1703 reset_shared(unsigned short port)
1706 * Create or re-create shared memory and semaphores.
1708 * Note: in each "cycle of life" we will normally assign the same IPC
1709 * keys (if using SysV shmem and/or semas), since the port number is
1710 * used to determine IPC keys. This helps ensure that we will clean
1711 * up dead IPC objects if the postmaster crashes and is restarted.
1713 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1718 * SIGHUP -- reread config files, and tell children to do same
1721 SIGHUP_handler(SIGNAL_ARGS)
1723 int save_errno = errno;
1725 PG_SETMASK(&BlockSig);
1727 if (Shutdown <= SmartShutdown)
1730 (errmsg("received SIGHUP, reloading configuration files")));
1731 ProcessConfigFile(PGC_SIGHUP);
1733 write_nondefault_variables(PGC_SIGHUP);
1735 SignalChildren(SIGHUP);
1740 * Tell the background writer to terminate so that we
1741 * will start a new one with a possibly changed config
1743 if (BgWriterPID != 0)
1744 kill(BgWriterPID, SIGTERM);
1747 PG_SETMASK(&UnBlockSig);
1755 * pmdie -- signal handler for processing various postmaster signals.
1760 int save_errno = errno;
1762 PG_SETMASK(&BlockSig);
1765 (errmsg_internal("postmaster received signal %d",
1766 postgres_signal_arg)));
1768 switch (postgres_signal_arg)
1775 * Wait for children to end their work and ShutdownDataBase.
1777 if (Shutdown >= SmartShutdown)
1779 Shutdown = SmartShutdown;
1781 (errmsg("received smart shutdown request")));
1783 /* Must tell bgwriter to quit, or it never will... */
1784 if (BgWriterPID != 0)
1785 kill(BgWriterPID, SIGTERM);
1787 if (DLGetHead(BackendList)) /* let reaper() handle this */
1791 * No children left. Shutdown data base system.
1793 if (StartupPID > 0 || FatalError) /* let reaper() handle
1796 if (ShutdownPID > 0)
1798 elog(PANIC, "shutdown process %d already running",
1803 ShutdownPID = ShutdownDataBase();
1811 * Abort all children with SIGTERM (rollback active transactions
1812 * and exit) and ShutdownDataBase when they are gone.
1814 if (Shutdown >= FastShutdown)
1816 Shutdown = FastShutdown;
1818 (errmsg("received fast shutdown request")));
1820 if (DLGetHead(BackendList))
1825 (errmsg("aborting any active transactions")));
1826 SignalChildren(SIGTERM);
1827 /* reaper() does the rest */
1833 * No children left. Shutdown data base system.
1835 * Unlike the previous case, it is not an error for the shutdown
1836 * process to be running already (we could get SIGTERM followed
1837 * shortly later by SIGINT).
1839 if (StartupPID > 0 || FatalError) /* let reaper() handle
1842 if (ShutdownPID == 0)
1843 ShutdownPID = ShutdownDataBase();
1849 * Immediate Shutdown:
1851 * abort all children with SIGQUIT and exit without attempt to
1852 * properly shutdown data base system.
1855 (errmsg("received immediate shutdown request")));
1856 if (ShutdownPID > 0)
1857 kill(ShutdownPID, SIGQUIT);
1859 kill(StartupPID, SIGQUIT);
1860 if (DLGetHead(BackendList))
1861 SignalChildren(SIGQUIT);
1866 PG_SETMASK(&UnBlockSig);
1872 * Reaper -- signal handler to cleanup after a backend (child) dies.
1877 int save_errno = errno;
1880 int status; /* backend exit status */
1883 union wait status; /* backend exit status */
1887 int pid; /* process id of dead backend */
1889 PG_SETMASK(&BlockSig);
1892 (errmsg_internal("reaping dead processes")));
1894 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1896 exitstatus = status;
1899 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1901 exitstatus = status.w_status;
1903 while ((pid = win32_waitpid(&exitstatus)) > 0)
1906 * We need to do this here, and not in CleanupProc, since this
1907 * is to be called on all children when we are done with them.
1908 * Could move to LogChildExit, but that seems like asking for
1911 win32_RemoveChild(pid);
1916 * Check if this child was the statistics collector. If so, try to
1917 * start a new one. (If fail, we'll try again in future cycles of
1920 if (pgstat_ispgstat(pid))
1922 LogChildExit(LOG, gettext("statistics collector process"),
1929 * Check if this child was a shutdown or startup process.
1931 if (ShutdownPID > 0 && pid == ShutdownPID)
1933 if (exitstatus != 0)
1935 LogChildExit(LOG, gettext("shutdown process"),
1939 /* Normal postmaster exit is here */
1943 if (StartupPID > 0 && pid == StartupPID)
1945 if (exitstatus != 0)
1947 LogChildExit(LOG, gettext("startup process"),
1950 (errmsg("aborting startup due to startup process failure")));
1956 * Startup succeeded - remember its ID and RedoRecPtr.
1958 * NB: this MUST happen before we fork a checkpoint or shutdown
1959 * subprocess, else they will have wrong local ThisStartUpId.
1963 FatalError = false; /* done with recovery */
1966 * Arrange for first checkpoint to occur after standard delay.
1969 checkpointed = time(NULL);
1972 * Go to shutdown mode if a shutdown request was pending.
1974 if (Shutdown > NoShutdown)
1976 if (ShutdownPID > 0)
1978 elog(PANIC, "startup process %d died while shutdown process %d already running",
1979 pid, (int) ShutdownPID);
1982 ShutdownPID = ShutdownDataBase();
1989 * Else do standard child cleanup.
1991 CleanupProc(pid, exitstatus);
1993 } /* loop over pending child-death reports */
1998 * Wait for all children exit, then reset shmem and
2001 if (DLGetHead(BackendList) || StartupPID > 0 || ShutdownPID > 0)
2004 (errmsg("all server processes terminated; reinitializing")));
2007 reset_shared(PostPortNumber);
2009 StartupPID = StartupDataBase();
2014 if (Shutdown > NoShutdown)
2016 if (DLGetHead(BackendList))
2018 if (StartupPID > 0 || ShutdownPID > 0)
2020 ShutdownPID = ShutdownDataBase();
2024 PG_SETMASK(&UnBlockSig);
2030 * CleanupProc -- cleanup after terminated backend.
2032 * Remove all local state associated with backend.
2035 CleanupProc(int pid,
2036 int exitstatus) /* child's exit status. */
2042 LogChildExit(DEBUG2, gettext("child process"), pid, exitstatus);
2045 * If a backend dies in an ugly way (i.e. exit status not 0) then we
2046 * must signal all other backends to quickdie. If exit status is zero
2047 * we assume everything is hunky dory and simply remove the backend
2048 * from the active backend list.
2050 if (exitstatus == 0)
2052 curr = DLGetHead(BackendList);
2055 bp = (Backend *) DLE_VAL(curr);
2059 ShmemBackendArrayRemove(bp->pid);
2066 curr = DLGetSucc(curr);
2069 if (pid == CheckPointPID)
2074 checkpointed = time(NULL);
2075 /* Update RedoRecPtr for future child backends */
2076 GetSavedRedoRecPtr();
2079 else if (pid == BgWriterPID)
2087 /* below here we're dealing with a non-normal exit */
2089 /* Make log entry unless we did so already */
2093 (pid == CheckPointPID) ? gettext("checkpoint process") :
2094 (pid == BgWriterPID) ? gettext("bgwriter process") :
2095 gettext("server process"),
2098 (errmsg("terminating any other active server processes")));
2101 curr = DLGetHead(BackendList);
2104 next = DLGetSucc(curr);
2105 bp = (Backend *) DLE_VAL(curr);
2109 * This backend is still alive. Unless we did so already,
2110 * tell it to commit hara-kiri.
2112 * SIGQUIT is the special signal that says exit without proc_exit
2113 * and let the user know what's going on. But if SendStop is
2114 * set (-s on command line), then we send SIGSTOP instead, so
2115 * that we can get core dumps from all backends by hand.
2120 (errmsg_internal("sending %s to process %d",
2121 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2123 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2129 * Found entry for freshly-dead backend, so remove it.
2132 ShmemBackendArrayRemove(bp->pid);
2141 if (pid == CheckPointPID)
2146 else if (pid == BgWriterPID)
2153 * Tell the collector about backend termination
2162 * Log the death of a child process.
2165 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2167 if (WIFEXITED(exitstatus))
2171 * translator: %s is a noun phrase describing a child process,
2172 * such as "server process"
2174 (errmsg("%s (PID %d) exited with exit code %d",
2175 procname, pid, WEXITSTATUS(exitstatus))));
2176 else if (WIFSIGNALED(exitstatus))
2180 * translator: %s is a noun phrase describing a child process,
2181 * such as "server process"
2183 (errmsg("%s (PID %d) was terminated by signal %d",
2184 procname, pid, WTERMSIG(exitstatus))));
2189 * translator: %s is a noun phrase describing a child process,
2190 * such as "server process"
2192 (errmsg("%s (PID %d) exited with unexpected status %d",
2193 procname, pid, exitstatus)));
2197 * Send a signal to all backend children.
2200 SignalChildren(int signal)
2206 curr = DLGetHead(BackendList);
2209 next = DLGetSucc(curr);
2210 bp = (Backend *) DLE_VAL(curr);
2212 if (bp->pid != MyProcPid)
2215 (errmsg_internal("sending signal %d to process %d",
2218 kill(bp->pid, signal);
2226 * BackendStartup -- start backend process
2228 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2231 BackendStartup(Port *port)
2233 Backend *bn; /* for backend cleanup */
2236 #ifdef LINUX_PROFILE
2237 struct itimerval prof_itimer;
2241 * Compute the cancel key that will be assigned to this backend. The
2242 * backend will have its own copy in the forked-off process' value of
2243 * MyCancelKey, so that it can transmit the key to the frontend.
2245 MyCancelKey = PostmasterRandom();
2248 * Make room for backend data structure. Better before the fork() so
2249 * we can handle failure cleanly.
2251 bn = (Backend *) malloc(sizeof(Backend));
2255 (errcode(ERRCODE_OUT_OF_MEMORY),
2256 errmsg("out of memory")));
2257 return STATUS_ERROR;
2261 * Flush stdio channels just before fork, to avoid double-output
2262 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2263 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2264 * coredump if we do. Presently stdout and stderr are the only stdio
2265 * output channels used by the postmaster, so fflush'ing them should
2271 #ifdef LINUX_PROFILE
2274 * Linux's fork() resets the profiling timer in the child process. If
2275 * we want to profile child processes then we need to save and restore
2276 * the timer setting. This is a waste of time if not profiling,
2277 * however, so only do it if commanded by specific -DLINUX_PROFILE
2280 getitimer(ITIMER_PROF, &prof_itimer);
2284 /* Specific beos actions before backend startup */
2285 beos_before_backend_startup();
2288 port->canAcceptConnections = canAcceptConnections();
2290 pid = Backend_forkexec(port);
2294 if (pid == 0) /* child */
2296 #ifdef LINUX_PROFILE
2297 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2301 /* Specific beos backend startup actions */
2302 beos_backend_startup();
2306 proc_exit(BackendRun(port));
2310 /* in parent, error */
2313 int save_errno = errno;
2316 /* Specific beos backend startup actions */
2317 beos_backend_startup_failed();
2322 (errmsg("could not fork new process for connection: %m")));
2323 report_fork_failure_to_client(port, save_errno);
2324 return STATUS_ERROR;
2327 /* in parent, normal */
2329 (errmsg_internal("forked new backend, pid=%d socket=%d",
2330 (int) pid, port->sock)));
2333 * Everything's been successful, it's safe to add this backend to our
2337 bn->cancel_key = MyCancelKey;
2339 ShmemBackendArrayAdd(bn);
2341 DLAddHead(BackendList, DLNewElem(bn));
2347 * Try to report backend fork() failure to client before we close the
2348 * connection. Since we do not care to risk blocking the postmaster on
2349 * this connection, we set the connection to non-blocking and try only once.
2351 * This is grungy special-purpose code; we cannot use backend libpq since
2352 * it's not up and running.
2355 report_fork_failure_to_client(Port *port, int errnum)
2359 /* Format the error message packet (always V2 protocol) */
2360 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2361 gettext("could not fork new process for connection: "),
2364 /* Set port to non-blocking. Don't do send() if this fails */
2365 if (FCNTL_NONBLOCK(port->sock) < 0)
2368 send(port->sock, buffer, strlen(buffer) + 1, 0);
2373 * split_opts -- split a string of options and append it to an argv array
2375 * NB: the string is destructively modified!
2377 * Since no current POSTGRES arguments require any quoting characters,
2378 * we can use the simple-minded tactic of assuming each set of space-
2379 * delimited characters is a separate argv element.
2381 * If you don't like that, well, we *used* to pass the whole option string
2382 * as ONE argument to execl(), which was even less intelligent...
2385 split_opts(char **argv, int *argcp, char *s)
2389 while (isspace((unsigned char) *s))
2393 argv[(*argcp)++] = s;
2394 while (*s && !isspace((unsigned char) *s))
2403 * BackendInit/Run -- perform authentication [BackendInit], and if successful,
2404 * set up the backend's argument list [BackendRun] and invoke
2408 * Shouldn't return at all.
2409 * If PostgresMain() fails, return status.
2412 BackendInit(Port *port)
2417 char remote_host[NI_MAXHOST];
2418 char remote_port[NI_MAXSERV];
2420 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2422 ClientAuthInProgress = true; /* limit visibility of log messages */
2424 /* We don't want the postmaster's proc_exit() handlers */
2428 * Signal handlers setting is moved to tcop/postgres...
2431 /* save start time for end of session reporting */
2432 gettimeofday(&(port->session_start),NULL);
2434 /* set these to empty in case they are needed before we set them up */
2435 port->remote_host = "";
2436 port->remote_port = "";
2438 /* Save port etc. for ps status */
2441 /* Reset MyProcPid to new backend's pid */
2442 MyProcPid = getpid();
2445 * Initialize libpq and enable reporting of ereport errors to the
2446 * client. Must do this now because authentication uses libpq to send
2449 pq_init(); /* initialize libpq to talk to client */
2450 whereToSendOutput = Remote; /* now safe to ereport to client */
2453 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2454 * during any client authentication related communication. Otherwise
2455 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2456 * if a buggy client blocks a backend during authentication.
2458 pqsignal(SIGTERM, authdie);
2459 pqsignal(SIGQUIT, authdie);
2460 pqsignal(SIGALRM, authdie);
2461 PG_SETMASK(&AuthBlockSig);
2464 * Get the remote host name and port for logging and status display.
2466 remote_host[0] = '\0';
2467 remote_port[0] = '\0';
2468 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2469 remote_host, sizeof(remote_host),
2470 remote_port, sizeof(remote_port),
2471 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2473 getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2474 remote_host, sizeof(remote_host),
2475 remote_port, sizeof(remote_port),
2476 NI_NUMERICHOST | NI_NUMERICSERV);
2479 if (Log_connections)
2481 (errmsg("connection received: host=%s port=%s",
2482 remote_host, remote_port)));
2486 /* modify remote_host for use in ps status */
2487 char tmphost[NI_MAXHOST];
2489 snprintf(tmphost, sizeof(tmphost), "%s:%s", remote_host, remote_port);
2490 StrNCpy(remote_host, tmphost, sizeof(remote_host));
2494 * save remote_host and remote_port in port stucture
2496 port->remote_host = strdup(remote_host);
2497 port->remote_port = strdup(remote_port);
2500 * Ready to begin client interaction. We will give up and exit(0)
2501 * after a time delay, so that a broken client can't hog a connection
2502 * indefinitely. PreAuthDelay doesn't count against the time limit.
2504 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2505 elog(FATAL, "could not set timer for authorization timeout");
2508 * Receive the startup packet (which might turn out to be a cancel
2511 status = ProcessStartupPacket(port, false);
2513 if (status != STATUS_OK)
2517 * Now that we have the user and database name, we can set the process
2518 * title for ps. It's good to do this as early as possible in
2521 init_ps_display(port->user_name, port->database_name, remote_host);
2522 set_ps_display("authentication");
2525 * Now perform authentication exchange.
2527 ClientAuthentication(port); /* might not return, if failure */
2530 * Done with authentication. Disable timeout, and prevent
2531 * SIGTERM/SIGQUIT again until backend startup is complete.
2533 if (!disable_sig_alarm(false))
2534 elog(FATAL, "could not disable timer for authorization timeout");
2535 PG_SETMASK(&BlockSig);
2537 if (Log_connections)
2539 (errmsg("connection authorized: user=%s database=%s",
2540 port->user_name, port->database_name)));
2543 * Don't want backend to be able to see the postmaster random number
2544 * generator state. We have to clobber the static random_seed *and*
2545 * start a new random sequence in the random() library function.
2548 gettimeofday(&now, &tz);
2549 srandom((unsigned int) now.tv_usec);
2554 BackendRun(Port *port)
2564 * Let's clean up ourselves as the postmaster child, and
2565 * close the postmaster's other sockets
2567 ClosePostmasterPorts(true);
2570 * PreAuthDelay is a debugging aid for investigating problems in the
2571 * authentication cycle: it can be set in postgresql.conf to allow
2572 * time to attach to the newly-forked backend with a debugger. (See
2573 * also the -W backend switch, which we allow clients to pass through
2574 * PGOPTIONS, but it is not honored until after authentication.)
2576 if (PreAuthDelay > 0)
2577 sleep(PreAuthDelay);
2579 /* Will exit on failure */
2584 * Now, build the argv vector that will be given to PostgresMain.
2586 * The layout of the command line is
2587 * postgres [secure switches] -p databasename [insecure switches]
2588 * where the switches after -p come from the client request.
2590 * The maximum possible number of commandline arguments that could come
2591 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2595 maxac = 10; /* for fixed args supplied below */
2596 maxac += (strlen(ExtraOptions) + 1) / 2;
2597 if (port->cmdline_options)
2598 maxac += (strlen(port->cmdline_options) + 1) / 2;
2600 av = (char **) MemoryContextAlloc(TopMemoryContext,
2601 maxac * sizeof(char *));
2604 av[ac++] = "postgres";
2607 * Pass the requested debugging level along to the backend.
2611 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2612 av[ac++] = debugbuf;
2616 * Pass any backend switches specified with -o in the postmaster's own
2617 * command line. We assume these are secure.
2619 split_opts(av, &ac, ExtraOptions);
2621 /* Tell the backend what protocol the frontend is using. */
2622 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2623 av[ac++] = protobuf;
2626 /* pass data dir before end of secure switches (-p) */
2632 * Tell the backend it is being called from the postmaster, and which
2633 * database to use. -p marks the end of secure switches.
2636 av[ac++] = port->database_name;
2639 * Pass the (insecure) option switches from the connection request.
2640 * (It's OK to mangle port->cmdline_options now.)
2642 if (port->cmdline_options)
2643 split_opts(av, &ac, port->cmdline_options);
2650 * Release postmaster's working memory context so that backend can
2651 * recycle the space. Note this does not trash *MyProcPort, because
2652 * ConnCreate() allocated that space with malloc() ... else we'd need
2653 * to copy the Port data here. Also, subsidiary data such as the
2654 * username isn't lost either; see ProcessStartupPacket().
2656 MemoryContextSwitchTo(TopMemoryContext);
2657 #ifndef EXEC_BACKEND
2658 MemoryContextDelete(PostmasterContext);
2660 PostmasterContext = NULL;
2663 * Debug: print arguments being passed to backend
2666 (errmsg_internal("%s child[%d]: starting with (",
2667 progname, getpid())));
2668 for (i = 0; i < ac; ++i)
2670 (errmsg_internal("\t%s", av[i])));
2672 (errmsg_internal(")")));
2674 ClientAuthInProgress = false; /* client_min_messages is active
2677 return (PostgresMain(ac, av, port->user_name));
2685 * SubPostmasterMain -- prepare the fork/exec'd process to be in an equivalent
2686 * state (for calling BackendRun) as a forked process.
2689 * Shouldn't return at all.
2692 SubPostmasterMain(int argc, char* argv[])
2694 unsigned long backendID;
2697 memset((void*)&port, 0, sizeof(Port));
2700 /* Do this sooner rather than later... */
2701 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2703 /* In EXEC case we will not have inherited these settings */
2704 IsPostmasterEnvironment = true;
2705 whereToSendOutput = None;
2707 /* Setup global context */
2708 MemoryContextInit();
2709 InitializeGUCOptions();
2711 /* Parse passed-in context */
2713 backendID = (unsigned long)atol(argv[argc++]);
2714 DataDir = strdup(argv[argc++]);
2716 /* Read in file-based context */
2717 read_nondefault_variables();
2718 read_backend_variables(backendID,&port);
2720 /* FIXME: [fork/exec] Ugh */
2726 /* Attach process to shared segments */
2727 AttachSharedMemoryAndSemaphores();
2730 proc_exit(BackendRun(&port));
2735 * Backend_forkexec -- fork/exec off a backend process
2738 * the pid of the fork/exec'd process
2741 Backend_forkexec(Port *port)
2745 int ac = 0, bufc = 0, i;
2746 char buf[2][MAXPGPATH];
2748 if (!write_backend_variables(port))
2749 return -1; /* log made by write_backend_variables */
2751 av[ac++] = "postgres";
2752 av[ac++] = "-forkexec";
2754 /* Format up context to pass to exec'd process */
2755 snprintf(buf[bufc++],MAXPGPATH,"%lu",tmpBackendFileNum);
2756 /* FIXME: [fork/exec] whitespaces in directories? */
2757 snprintf(buf[bufc++],MAXPGPATH,"%s",DataDir);
2759 /* Add to the arg list */
2760 Assert(bufc <= lengthof(buf));
2761 for (i = 0; i < bufc; i++)
2764 /* FIXME: [fork/exec] ExtraOptions? */
2767 Assert(ac <= lengthof(av));
2770 pid = win32_forkexec(pg_pathname,av); /* logs on error */
2772 /* Fire off execv in child */
2773 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
2775 * FIXME: [fork/exec] suggestions for what to do here?
2776 * Probably OK to issue error (unlike pgstat case)
2780 return pid; /* Parent returns pid */
2787 * ExitPostmaster -- cleanup
2789 * Do NOT call exit() directly --- always go through here!
2792 ExitPostmaster(int status)
2794 /* should cleanup shared memory and kill all backends */
2797 * Not sure of the semantics here. When the Postmaster dies, should
2798 * the backends all be killed? probably not.
2800 * MUST -- vadim 05-10-1999
2802 /* Should I use true instead? */
2803 ClosePostmasterPorts(false);
2809 * sigusr1_handler - handle signal conditions from child processes
2812 sigusr1_handler(SIGNAL_ARGS)
2814 int save_errno = errno;
2816 PG_SETMASK(&BlockSig);
2818 if (CheckPostmasterSignal(PMSIGNAL_DO_CHECKPOINT))
2820 if (CheckPointWarning != 0)
2823 * This only times checkpoints forced by running out of
2824 * segment files. Other checkpoints could reduce the
2825 * frequency of forced checkpoints.
2827 time_t now = time(NULL);
2829 if (LastSignalledCheckpoint != 0)
2831 int elapsed_secs = now - LastSignalledCheckpoint;
2833 if (elapsed_secs < CheckPointWarning)
2835 (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
2837 errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
2839 LastSignalledCheckpoint = now;
2843 * Request to schedule a checkpoint
2845 * Ignore request if checkpoint is already running or checkpointing
2846 * is currently disabled
2848 if (CheckPointPID == 0 && checkpointed &&
2849 StartupPID == 0 && Shutdown == NoShutdown &&
2850 !FatalError && random_seed != 0)
2852 CheckPointPID = CheckPointDataBase();
2853 /* note: if fork fails, CheckPointPID stays 0; nothing happens */
2857 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2860 * Password or group file has changed.
2866 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2869 * Send SIGUSR2 to all children (triggers AsyncNotifyHandler). See
2870 * storage/ipc/sinvaladt.c for the use of this.
2872 if (Shutdown == NoShutdown)
2873 SignalChildren(SIGUSR2);
2876 PG_SETMASK(&UnBlockSig);
2883 * Dummy signal handler
2885 * We use this for signals that we don't actually use in the postmaster,
2886 * but we do use in backends. If we SIG_IGN such signals in the postmaster,
2887 * then a newly started backend might drop a signal that arrives before it's
2888 * able to reconfigure its signal processing. (See notes in postgres.c.)
2891 dummy_handler(SIGNAL_ARGS)
2897 * CharRemap: given an int in range 0..61, produce textual encoding of it
2898 * per crypt(3) conventions.
2922 RandomSalt(char *cryptSalt, char *md5Salt)
2924 long rand = PostmasterRandom();
2926 cryptSalt[0] = CharRemap(rand % 62);
2927 cryptSalt[1] = CharRemap(rand / 62);
2930 * It's okay to reuse the first random value for one of the MD5 salt
2931 * bytes, since only one of the two salts will be sent to the client.
2932 * After that we need to compute more random bits.
2934 * We use % 255, sacrificing one possible byte value, so as to ensure
2935 * that all bits of the random() value participate in the result.
2936 * While at it, add one to avoid generating any null bytes.
2938 md5Salt[0] = (rand % 255) + 1;
2939 rand = PostmasterRandom();
2940 md5Salt[1] = (rand % 255) + 1;
2941 rand = PostmasterRandom();
2942 md5Salt[2] = (rand % 255) + 1;
2943 rand = PostmasterRandom();
2944 md5Salt[3] = (rand % 255) + 1;
2951 PostmasterRandom(void)
2953 static bool initialized = false;
2957 Assert(random_seed != 0);
2958 srandom(random_seed);
2966 * Count up number of child processes.
2975 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2977 bp = (Backend *) DLE_VAL(curr);
2978 if (bp->pid != MyProcPid)
2981 /* Checkpoint and bgwriter will be in the list, discount them */
2982 if (CheckPointPID != 0)
2984 if (BgWriterPID != 0)
2990 * Fire off a subprocess for startup/shutdown/checkpoint/bgwriter.
2992 * Return value of SSDataBase is subprocess' PID, or 0 if failed to start subprocess
2993 * (0 is returned only for checkpoint/bgwriter cases).
2995 * note: in the EXEC_BACKEND case, we delay the fork until argument list has been
2998 NON_EXEC_STATIC void
2999 SSDataBaseInit(int xlop)
3001 const char *statmsg;
3003 IsUnderPostmaster = true; /* we are a postmaster subprocess
3007 /* In EXEC case we will not have inherited these settings */
3008 IsPostmasterEnvironment = true;
3009 whereToSendOutput = None;
3012 MyProcPid = getpid(); /* reset MyProcPid */
3014 /* Lose the postmaster's on-exit routines and port connections */
3018 * Identify myself via ps
3022 case BS_XLOG_STARTUP:
3023 statmsg = "startup subprocess";
3025 case BS_XLOG_CHECKPOINT:
3026 statmsg = "checkpoint subprocess";
3028 case BS_XLOG_BGWRITER:
3029 statmsg = "bgwriter subprocess";
3031 case BS_XLOG_SHUTDOWN:
3032 statmsg = "shutdown subprocess";
3035 statmsg = "??? subprocess";
3038 init_ps_display(statmsg, "", "");
3044 SSDataBase(int xlop)
3048 #ifndef EXEC_BACKEND
3049 #ifdef LINUX_PROFILE
3050 struct itimerval prof_itimer;
3059 #ifndef EXEC_BACKEND
3060 #ifdef LINUX_PROFILE
3061 /* see comments in BackendRun */
3062 getitimer(ITIMER_PROF, &prof_itimer);
3066 /* Specific beos actions before backend startup */
3067 beos_before_backend_startup();
3070 /* Non EXEC_BACKEND case; fork here */
3071 if ((pid = fork()) == 0) /* child */
3079 #ifndef EXEC_BACKEND
3080 #ifdef LINUX_PROFILE
3081 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3085 /* Specific beos actions after backend startup */
3086 beos_backend_startup();
3089 /* Close the postmaster's sockets */
3090 ClosePostmasterPorts(true);
3092 SSDataBaseInit(xlop);
3094 if (!write_backend_variables(NULL))
3095 return -1; /* log issued by write_backend_variables */
3098 /* Set up command-line arguments for subprocess */
3099 av[ac++] = "postgres";
3104 snprintf(nbbuf, sizeof(nbbuf), "-B%d", NBuffers);
3107 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
3111 /* pass data dir before end of secure switches (-p) */
3115 /* and the backend identifier + dbname */
3116 snprintf(idbuf, sizeof(idbuf), "-p%lu,template1", tmpBackendFileNum);
3120 av[ac++] = "template1";
3125 Assert(ac < lengthof(av));
3128 /* EXEC_BACKEND case; fork/exec here */
3130 pid = win32_forkexec(pg_pathname,av); /* logs on error */
3132 if ((pid = fork()) == 0 && (execv(pg_pathname,av) == -1))
3135 elog(ERROR,"unable to execv in SSDataBase: %m");
3140 BootstrapMain(ac, av);
3148 #ifndef EXEC_BACKEND
3150 /* Specific beos actions before backend startup */
3151 beos_backend_startup_failed();
3156 case BS_XLOG_STARTUP:
3158 (errmsg("could not fork startup process: %m")));
3160 case BS_XLOG_CHECKPOINT:
3162 (errmsg("could not fork checkpoint process: %m")));
3164 case BS_XLOG_BGWRITER:
3166 (errmsg("could not fork bgwriter process: %m")));
3168 case BS_XLOG_SHUTDOWN:
3170 (errmsg("could not fork shutdown process: %m")));
3174 (errmsg("could not fork process: %m")));
3179 * fork failure is fatal during startup/shutdown, but there's no
3180 * need to choke if a routine checkpoint or starting a background
3183 if (xlop == BS_XLOG_CHECKPOINT)
3185 if (xlop == BS_XLOG_BGWRITER)
3191 * The startup and shutdown processes are not considered normal
3192 * backends, but the checkpoint and bgwriter processes are.
3193 * They must be added to the list of backends.
3195 if (xlop == BS_XLOG_CHECKPOINT || xlop == BS_XLOG_BGWRITER)
3197 if (!(bn = (Backend *) malloc(sizeof(Backend))))
3200 (errcode(ERRCODE_OUT_OF_MEMORY),
3201 errmsg("out of memory")));
3206 bn->cancel_key = PostmasterRandom();
3208 ShmemBackendArrayAdd(bn);
3210 DLAddHead(BackendList, DLNewElem(bn));
3213 * Since this code is executed periodically, it's a fine place to
3214 * do other actions that should happen every now and then on no
3215 * particular schedule. Such as...
3218 TouchSocketLockFile();
3226 * Create the opts file
3229 CreateOptsFile(int argc, char *argv[])
3231 char fullprogname[MAXPGPATH];
3232 char filename[MAXPGPATH];
3236 if (FindExec(fullprogname, argv[0], "postmaster") < 0)
3239 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3241 if ((fp = fopen(filename, "w")) == NULL)
3243 elog(LOG, "could not create file \"%s\": %m", filename);
3247 fprintf(fp, "%s", fullprogname);
3248 for (i = 1; i < argc; i++)
3249 fprintf(fp, " '%s'", argv[i]);
3254 elog(LOG, "could not write file \"%s\": %m", filename);
3262 * This should be used only for reporting "interactive" errors (essentially,
3263 * bogus arguments on the command line). Once the postmaster is launched,
3264 * use ereport. In particular, don't use this for anything that occurs
3265 * after pmdaemonize.
3268 postmaster_error(const char *fmt,...)
3272 fprintf(stderr, "%s: ", progname);
3274 vfprintf(stderr, gettext(fmt), ap);
3276 fprintf(stderr, "\n");
3283 * The following need to be available to the read/write_backend_variables
3286 #include "storage/spin.h"
3287 extern XLogRecPtr RedoRecPtr;
3288 extern XLogwrtResult LogwrtResult;
3289 extern slock_t *ShmemLock;
3290 extern slock_t *ShmemIndexLock;
3291 extern void *ShmemIndexAlloc;
3292 typedef struct LWLock LWLock;
3293 extern LWLock *LWLockArray;
3294 extern slock_t *ProcStructLock;
3295 extern int pgStatSock;
3297 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3298 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3299 #define get_tmp_backend_file_name(buf,id) \
3303 "%s/%s/%s.backend_var.%lu", \
3305 PG_TEMP_FILES_DIR, \
3306 PG_TEMP_FILE_PREFIX, \
3311 write_backend_variables(Port *port)
3313 char filename[MAXPGPATH];
3315 get_tmp_backend_file_name(filename,++tmpBackendFileNum);
3318 fp = AllocateFile(filename, PG_BINARY_W);
3321 /* As per OpenTemporaryFile... */
3322 char dirname[MAXPGPATH];
3323 sprintf(dirname,"%s/%s",DataDir,PG_TEMP_FILES_DIR);
3324 mkdir(dirname, S_IRWXU);
3326 fp = AllocateFile(filename, PG_BINARY_W);
3330 (errcode_for_file_access(),
3331 errmsg("could not write to file \"%s\": %m", filename)));
3339 write_var(port->sock,fp);
3340 write_var(port->proto,fp);
3341 write_var(port->laddr,fp);
3342 write_var(port->raddr,fp);
3343 write_var(port->canAcceptConnections,fp);
3344 write_var(port->cryptSalt,fp);
3345 write_var(port->md5Salt,fp);
3347 write_var(MyCancelKey,fp);
3349 write_var(RedoRecPtr,fp);
3350 write_var(LogwrtResult,fp);
3352 write_var(UsedShmemSegID,fp);
3353 write_var(UsedShmemSegAddr,fp);
3355 write_var(ShmemLock,fp);
3356 write_var(ShmemIndexLock,fp);
3357 write_var(ShmemVariableCache,fp);
3358 write_var(ShmemIndexAlloc,fp);
3359 write_var(ShmemBackendArray,fp);
3361 write_var(LWLockArray,fp);
3362 write_var(ProcStructLock,fp);
3363 write_var(pgStatSock,fp);
3365 write_var(PreAuthDelay,fp);
3366 write_var(debug_flag,fp);
3367 write_var(PostmasterPid,fp);
3373 (errcode_for_file_access(),
3374 errmsg("could not write to file \"%s\": %m", filename)));
3382 read_backend_variables(unsigned long id, Port *port)
3384 char filename[MAXPGPATH];
3386 get_tmp_backend_file_name(filename,id);
3389 fp = AllocateFile(filename, PG_BINARY_R);
3393 (errcode_for_file_access(),
3394 errmsg("could not read from backend_variables file \"%s\": %m", filename)));
3401 read_var(port->sock,fp);
3402 read_var(port->proto,fp);
3403 read_var(port->laddr,fp);
3404 read_var(port->raddr,fp);
3405 read_var(port->canAcceptConnections,fp);
3406 read_var(port->cryptSalt,fp);
3407 read_var(port->md5Salt,fp);
3409 read_var(MyCancelKey,fp);
3411 read_var(RedoRecPtr,fp);
3412 read_var(LogwrtResult,fp);
3414 read_var(UsedShmemSegID,fp);
3415 read_var(UsedShmemSegAddr,fp);
3417 read_var(ShmemLock,fp);
3418 read_var(ShmemIndexLock,fp);
3419 read_var(ShmemVariableCache,fp);
3420 read_var(ShmemIndexAlloc,fp);
3421 read_var(ShmemBackendArray,fp);
3423 read_var(LWLockArray,fp);
3424 read_var(ProcStructLock,fp);
3425 read_var(pgStatSock,fp);
3427 read_var(PreAuthDelay,fp);
3428 read_var(debug_flag,fp);
3429 read_var(PostmasterPid,fp);
3433 if (unlink(filename) != 0)
3435 (errcode_for_file_access(),
3436 errmsg("could not remove file \"%s\": %m", filename)));
3440 size_t ShmemBackendArraySize(void)
3442 return (NUM_BACKENDARRAY_ELEMS*sizeof(Backend));
3445 void ShmemBackendArrayAllocation(void)
3447 size_t size = ShmemBackendArraySize();
3448 ShmemBackendArray = (Backend*)ShmemAlloc(size);
3449 memset(ShmemBackendArray, 0, size);
3452 static void ShmemBackendArrayAdd(Backend *bn)
3455 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3457 /* Find an empty slot */
3458 if (ShmemBackendArray[i].pid == 0)
3460 ShmemBackendArray[i] = *bn;
3465 /* FIXME: [fork/exec] some sort of error */
3469 static void ShmemBackendArrayRemove(pid_t pid)
3472 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3474 if (ShmemBackendArray[i].pid == pid)
3476 /* Mark the slot as empty */
3477 ShmemBackendArray[i].pid = 0;
3482 /* Something stronger than WARNING here? */
3484 (errmsg_internal("unable to find backend entry with pid %d",
3492 pid_t win32_forkexec(const char* path, char *argv[])
3495 PROCESS_INFORMATION pi;
3498 char cmdLine[MAXPGPATH];
3499 HANDLE childHandleCopy;
3500 HANDLE waiterThread;
3502 /* Format the cmd line */
3503 snprintf(cmdLine,sizeof(cmdLine),"%s",path);
3505 while (argv[++i] != NULL)
3507 /* FIXME: [fork/exec] some strlen checks might be prudent here */
3508 strcat(cmdLine," ");
3509 strcat(cmdLine,argv[i]);
3513 * The following snippet can disappear when we consistently
3514 * use forward slashes.
3517 while (*(p++) != '\0')
3518 if (*p == '/') *p = '\\';
3520 memset(&pi,0,sizeof(pi));
3521 memset(&si,0,sizeof(si));
3523 if (!CreateProcess(NULL,cmdLine,NULL,NULL,TRUE,0,NULL,NULL,&si,&pi))
3525 elog(ERROR,"CreateProcess call failed (%d): %m",GetLastError());
3529 if (!IsUnderPostmaster)
3530 /* We are the Postmaster creating a child... */
3531 win32_AddChild(pi.dwProcessId,pi.hProcess);
3533 if (!DuplicateHandle(GetCurrentProcess(),
3535 GetCurrentProcess(),
3539 DUPLICATE_SAME_ACCESS))
3541 (errmsg_internal("failed to duplicate child handle: %i",GetLastError())));
3542 waiterThread = CreateThread(NULL, 64*1024, win32_sigchld_waiter, (LPVOID)childHandleCopy, 0, NULL);
3545 (errmsg_internal("failed to create sigchld waiter thread: %i",GetLastError())));
3546 CloseHandle(waiterThread);
3548 if (IsUnderPostmaster)
3549 CloseHandle(pi.hProcess);
3550 CloseHandle(pi.hThread);
3552 return pi.dwProcessId;
3556 * Note: The following three functions must not be interrupted (eg. by signals).
3557 * As the Postgres Win32 signalling architecture (currently) requires polling,
3558 * or APC checking functions which aren't used here, this is not an issue.
3560 * We keep two separate arrays, instead of a single array of pid/HANDLE structs,
3561 * to avoid having to re-create a handle array for WaitForMultipleObjects on
3562 * each call to win32_waitpid.
3565 static void win32_AddChild(pid_t pid, HANDLE handle)
3567 Assert(win32_childPIDArray && win32_childHNDArray);
3568 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3570 win32_childPIDArray[win32_numChildren] = pid;
3571 win32_childHNDArray[win32_numChildren] = handle;
3572 ++win32_numChildren;
3575 /* FIXME: [fork/exec] some sort of error */
3579 static void win32_RemoveChild(pid_t pid)
3582 Assert(win32_childPIDArray && win32_childHNDArray);
3584 for (i = 0; i < win32_numChildren; i++)
3586 if (win32_childPIDArray[i] == pid)
3588 CloseHandle(win32_childHNDArray[i]);
3590 /* Swap last entry into the "removed" one */
3591 --win32_numChildren;
3592 win32_childPIDArray[i] = win32_childPIDArray[win32_numChildren];
3593 win32_childHNDArray[i] = win32_childHNDArray[win32_numChildren];
3598 /* Something stronger than WARNING here? */
3600 (errmsg_internal("unable to find child entry with pid %d",
3604 static pid_t win32_waitpid(int *exitstatus)
3606 Assert(win32_childPIDArray && win32_childHNDArray);
3607 elog(DEBUG3,"waiting on %d children",win32_numChildren);
3609 if (win32_numChildren > 0)
3612 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't
3613 * want to run queued APCs here.
3617 DWORD ret = WaitForMultipleObjects(win32_numChildren,win32_childHNDArray,FALSE,0);
3623 (errmsg_internal("failed to wait on %d children: %i",
3624 win32_numChildren,GetLastError())));
3625 /* Fall through to WAIT_TIMEOUTs return */
3628 /* No children have finished */
3632 /* Get the exit code, and return the PID of, the respective process */
3633 index = ret-WAIT_OBJECT_0;
3634 Assert(index >= 0 && index < win32_numChildren);
3635 if (!GetExitCodeProcess(win32_childHNDArray[index],&exitCode))
3637 * If we get this far, this should never happen, but, then again...
3638 * No choice other than to assume a catastrophic failure.
3641 (errmsg_internal("failed to get exit code for child %d",
3642 win32_childPIDArray[index])));
3643 *exitstatus = (int)exitCode;
3644 return win32_childPIDArray[index];
3652 /* Note! Code belows executes on separate threads, one for
3653 each child process created */
3654 static DWORD WINAPI win32_sigchld_waiter(LPVOID param) {
3655 HANDLE procHandle = (HANDLE)param;
3657 DWORD r = WaitForSingleObject(procHandle, INFINITE);
3658 if (r == WAIT_OBJECT_0)
3659 pg_queue_signal(SIGCHLD);
3661 fprintf(stderr,"ERROR: Failed to wait on child process handle: %i\n",GetLastError());
3662 CloseHandle(procHandle);