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 1.386 2004/05/13 22:45:02 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 */
152 char *ListenAddresses;
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 const 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;
185 const bool ExecBackend = false;
189 * Set by the -o option
191 static char ExtraOptions[MAXPGPATH];
194 * These globals control the behavior of the postmaster in case some
195 * backend dumps core. Normally, it kills all peers of the dead backend
196 * and reinitializes shared memory. By specifying -s or -n, we can have
197 * the postmaster stop (rather than kill) peers and not reinitialize
198 * shared data structures.
200 static bool Reinit = true;
201 static int SendStop = false;
203 /* still more option variables */
204 bool EnableSSL = false;
205 bool SilentMode = false; /* silent mode (-S) */
207 int PreAuthDelay = 0;
208 int AuthenticationTimeout = 60;
209 int CheckPointTimeout = 300;
210 int CheckPointWarning = 30;
211 time_t LastSignalledCheckpoint = 0;
213 bool log_hostname; /* for ps display and logging */
214 bool Log_connections = false;
215 bool Db_user_namespace = false;
217 char *rendezvous_name;
219 /* list of library:init-function to be preloaded */
220 char *preload_libraries_string = NULL;
222 /* Startup/shutdown state */
223 static pid_t StartupPID = 0,
227 static time_t checkpointed = 0;
230 #define SmartShutdown 1
231 #define FastShutdown 2
233 static int Shutdown = NoShutdown;
235 static bool FatalError = false; /* T if recovering from backend crash */
237 bool ClientAuthInProgress = false; /* T during new-client
241 * State for assigning random salts and cancel keys.
242 * Also, the global MyCancelKey passes the cancel key assigned to a given
243 * backend from the postmaster to that backend (via fork).
246 static unsigned int random_seed = 0;
248 static int debug_flag = 0;
254 #ifdef HAVE_INT_OPTRESET
259 * postmaster.c - function prototypes
261 static void pmdaemonize(int argc, char *argv[]);
262 static Port *ConnCreate(int serverFd);
263 static void ConnFree(Port *port);
264 static void reset_shared(unsigned short port);
265 static void SIGHUP_handler(SIGNAL_ARGS);
266 static void pmdie(SIGNAL_ARGS);
267 static void reaper(SIGNAL_ARGS);
268 static void sigusr1_handler(SIGNAL_ARGS);
269 static void dummy_handler(SIGNAL_ARGS);
270 static void CleanupProc(int pid, int exitstatus);
271 static void LogChildExit(int lev, const char *procname,
272 int pid, int exitstatus);
273 static void BackendInit(Port *port);
274 static int BackendRun(Port *port);
275 static void ExitPostmaster(int status);
276 static void usage(const char *);
277 static int ServerLoop(void);
278 static int BackendStartup(Port *port);
279 static int ProcessStartupPacket(Port *port, bool SSLdone);
280 static void processCancelRequest(Port *port, void *pkt);
281 static int initMasks(fd_set *rmask);
282 static void report_fork_failure_to_client(Port *port, int errnum);
283 static enum CAC_state canAcceptConnections(void);
284 static long PostmasterRandom(void);
285 static void RandomSalt(char *cryptSalt, char *md5Salt);
286 static void SignalChildren(int signal);
287 static int CountChildren(void);
288 static bool CreateOptsFile(int argc, char *argv[], char *fullprogname);
289 NON_EXEC_STATIC void SSDataBaseInit(int xlop);
290 static pid_t SSDataBase(int xlop);
292 postmaster_error(const char *fmt,...)
293 /* This lets gcc check the format string for consistency. */
294 __attribute__((format(printf, 1, 2)));
298 static char postgres_exec_path[MAXPGPATH];
301 pid_t win32_forkexec(const char* path, char *argv[]);
303 static void win32_AddChild(pid_t pid, HANDLE handle);
304 static void win32_RemoveChild(pid_t pid);
305 static pid_t win32_waitpid(int *exitstatus);
306 static DWORD WINAPI win32_sigchld_waiter(LPVOID param);
308 static pid_t *win32_childPIDArray;
309 static HANDLE *win32_childHNDArray;
310 static unsigned long win32_numChildren = 0;
313 static pid_t Backend_forkexec(Port *port);
315 static unsigned long tmpBackendFileNum = 0;
316 void read_backend_variables(unsigned long id, Port *port);
317 static bool write_backend_variables(Port *port);
319 static void ShmemBackendArrayAdd(Backend *bn);
320 static void ShmemBackendArrayRemove(pid_t pid);
323 #define StartupDataBase() SSDataBase(BS_XLOG_STARTUP)
324 #define CheckPointDataBase() SSDataBase(BS_XLOG_CHECKPOINT)
325 #define StartBackgroundWriter() SSDataBase(BS_XLOG_BGWRITER)
326 #define ShutdownDataBase() SSDataBase(BS_XLOG_SHUTDOWN)
329 checkDataDir(const char *checkdir)
331 char path[MAXPGPATH];
333 struct stat stat_buf;
335 if (checkdir == NULL)
338 gettext("%s does not know where to find the database system data.\n"
339 "You must specify the directory that contains the database system\n"
340 "either by specifying the -D invocation option or by setting the\n"
341 "PGDATA environment variable.\n"),
346 if (stat(checkdir, &stat_buf) == -1)
350 (errcode_for_file_access(),
351 errmsg("data directory \"%s\" does not exist",
355 (errcode_for_file_access(),
356 errmsg("could not read permissions of directory \"%s\": %m",
361 * Check if the directory has group or world access. If so, reject.
363 * XXX temporarily suppress check when on Windows, because there may not
364 * be proper support for Unix-y file permissions. Need to think of a
365 * reasonable check to apply on Windows.
367 #if !defined(__CYGWIN__) && !defined(WIN32)
368 if (stat_buf.st_mode & (S_IRWXG | S_IRWXO))
370 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
371 errmsg("data directory \"%s\" has group or world access",
373 errdetail("Permissions should be u=rwx (0700).")));
376 /* Look for PG_VERSION before looking for pg_control */
377 ValidatePgVersion(checkdir);
379 snprintf(path, sizeof(path), "%s/global/pg_control", checkdir);
381 fp = AllocateFile(path, PG_BINARY_R);
385 gettext("%s: could not find the database system\n"
386 "Expected to find it in the directory \"%s\",\n"
387 "but could not open file \"%s\": %s\n"),
388 progname, checkdir, path, strerror(errno));
395 #ifdef USE_RENDEZVOUS
397 /* reg_reply -- empty callback function for DNSServiceRegistrationCreate() */
399 reg_reply(DNSServiceRegistrationReplyErrorType errorCode, void *context)
406 PostmasterMain(int argc, char *argv[])
410 char original_extraoptions[MAXPGPATH];
411 char *potential_DataDir = NULL;
414 *original_extraoptions = '\0';
416 progname = get_progname(argv[0]);
418 IsPostmasterEnvironment = true;
421 * Catch standard options before doing much else. This even works on
422 * systems without getopt_long.
426 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
431 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
433 puts("postmaster (PostgreSQL) " PG_VERSION);
439 * for security, no dir or file created can be group or other
442 umask((mode_t) 0077);
444 MyProcPid = PostmasterPid = getpid();
447 * Fire up essential subsystems: memory management
452 * By default, palloc() requests in the postmaster will be allocated
453 * in the PostmasterContext, which is space that can be recycled by
454 * backends. Allocated data that needs to be available to backends
455 * should be allocated in TopMemoryContext.
457 PostmasterContext = AllocSetContextCreate(TopMemoryContext,
459 ALLOCSET_DEFAULT_MINSIZE,
460 ALLOCSET_DEFAULT_INITSIZE,
461 ALLOCSET_DEFAULT_MAXSIZE);
462 MemoryContextSwitchTo(PostmasterContext);
464 IgnoreSystemIndexes(false);
469 InitializeGUCOptions();
471 potential_DataDir = getenv("PGDATA"); /* default value */
475 while ((opt = getopt(argc, argv, "A:a:B:b:c:D:d:Fh:ik:lm:MN:no:p:Ss-:")) != -1)
480 #ifdef USE_ASSERT_CHECKING
481 SetConfigOption("debug_assertions", optarg, PGC_POSTMASTER, PGC_S_ARGV);
483 postmaster_error("assert checking is not compiled in");
487 /* Can no longer set authentication method. */
490 SetConfigOption("shared_buffers", optarg, PGC_POSTMASTER, PGC_S_ARGV);
493 /* Can no longer set the backend executable file to use. */
496 potential_DataDir = optarg;
500 /* Turn on debugging for the postmaster. */
501 char *debugstr = palloc(strlen("debug") + strlen(optarg) + 1);
503 sprintf(debugstr, "debug%s", optarg);
504 SetConfigOption("log_min_messages", debugstr,
505 PGC_POSTMASTER, PGC_S_ARGV);
507 debug_flag = atoi(optarg);
511 SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
514 SetConfigOption("listen_addresses", optarg, PGC_POSTMASTER, PGC_S_ARGV);
517 SetConfigOption("listen_addresses", "*", PGC_POSTMASTER, PGC_S_ARGV);
520 SetConfigOption("unix_socket_directory", optarg, PGC_POSTMASTER, PGC_S_ARGV);
524 SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
528 /* Multiplexed backends no longer supported. */
533 * ignore this flag. This may be passed in because the
534 * program was run as 'postgres -M' instead of
539 /* The max number of backends to start. */
540 SetConfigOption("max_connections", optarg, PGC_POSTMASTER, PGC_S_ARGV);
543 /* Don't reinit shared mem after abnormal exit */
549 * Other options to pass to the backend on the command
550 * line -- useful only for debugging.
552 strcat(ExtraOptions, " ");
553 strcat(ExtraOptions, optarg);
554 strcpy(original_extraoptions, optarg);
557 SetConfigOption("port", optarg, PGC_POSTMASTER, PGC_S_ARGV);
562 * Start in 'S'ilent mode (disassociate from controlling
563 * tty). You may also think of this as 'S'ysV mode since
564 * it's most badly needed on SysV-derived systems like
567 SetConfigOption("silent_mode", "true", PGC_POSTMASTER, PGC_S_ARGV);
572 * In the event that some backend dumps core, send
573 * SIGSTOP, rather than SIGQUIT, to all its peers. This
574 * lets the wily post_hacker collect core dumps from
585 ParseLongOption(optarg, &name, &value);
590 (errcode(ERRCODE_SYNTAX_ERROR),
591 errmsg("--%s requires a value",
595 (errcode(ERRCODE_SYNTAX_ERROR),
596 errmsg("-c %s requires a value",
600 SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
609 gettext("Try \"%s --help\" for more information.\n"),
616 * Postmaster accepts no non-option switch arguments.
620 postmaster_error("invalid argument: \"%s\"", argv[optind]);
622 gettext("Try \"%s --help\" for more information.\n"),
628 * Now we can set the data directory, and then read postgresql.conf.
630 checkDataDir(potential_DataDir); /* issues error messages */
631 SetDataDir(potential_DataDir);
633 ProcessConfigFile(PGC_POSTMASTER);
635 write_nondefault_variables(PGC_POSTMASTER);
639 * Check for invalid combinations of GUC settings.
641 if (NBuffers < 2 * MaxBackends || NBuffers < 16)
644 * Do not accept -B so small that backends are likely to starve
645 * for lack of buffers. The specific choices here are somewhat
648 postmaster_error("the number of buffers (-B) must be at least twice the number of allowed connections (-N) and at least 16");
652 if (ReservedBackends >= MaxBackends)
654 postmaster_error("superuser_reserved_connections must be less than max_connections");
659 * Other one-time internal sanity checks can go here.
661 if (!CheckDateTokenTables())
663 postmaster_error("invalid datetoken tables, please fix");
668 * Now that we are done processing the postmaster arguments, reset
669 * getopt(3) library so that it will work correctly in subprocesses.
672 #ifdef HAVE_INT_OPTRESET
673 optreset = 1; /* some systems need this too */
676 /* For debugging: display postmaster environment */
678 extern char **environ;
682 (errmsg_internal("%s: PostmasterMain: initial environ dump:",
685 (errmsg_internal("-----------------------------------------")));
686 for (p = environ; *p; ++p)
688 (errmsg_internal("\t%s", *p)));
690 (errmsg_internal("-----------------------------------------")));
694 * On some systems our dynloader code needs the executable's pathname.
696 if (find_my_exec(my_exec_path, argv[0]) < 0)
698 (errmsg("%s: could not locate my own executable path",
702 if (find_other_exec(postgres_exec_path, argv[0], "postgres", PG_VERSIONSTR) < 0)
704 (errmsg("%s: could not locate postgres executable or non-matching version",
709 * Initialize SSL library, if specified.
717 * process any libraries that should be preloaded and optionally
720 if (preload_libraries_string)
721 process_preload_libraries(preload_libraries_string);
724 * Fork away from controlling terminal, if -S specified.
726 * Must do this before we grab any interlock files, else the interlocks
727 * will show the wrong PID.
730 pmdaemonize(argc, argv);
733 * Create lockfile for data directory.
735 * We want to do this before we try to grab the input sockets, because
736 * the data directory interlock is more reliable than the socket-file
737 * interlock (thanks to whoever decided to put socket files in /tmp
738 * :-(). For the same reason, it's best to grab the TCP socket before
741 CreateDataDirLockFile(DataDir, true);
744 * Remove old temporary files. At this point there can be no other
745 * Postgres processes running in this directory, so this should be
751 * Establish input sockets.
753 for (i = 0; i < MAXLISTEN; i++)
754 ListenSocket[i] = -1;
762 curhost = ListenAddresses;
765 /* ignore whitespace */
766 while (isspace((unsigned char) *curhost))
768 if (*curhost == '\0')
771 while (*endptr != '\0' && !isspace((unsigned char) *endptr))
775 if (strcmp(curhost,"*") == 0)
776 status = StreamServerPort(AF_UNSPEC, NULL,
777 (unsigned short) PostPortNumber,
779 ListenSocket, MAXLISTEN);
781 status = StreamServerPort(AF_UNSPEC, curhost,
782 (unsigned short) PostPortNumber,
784 ListenSocket, MAXLISTEN);
785 if (status != STATUS_OK)
787 (errmsg("could not create listen socket for \"%s\"",
797 #ifdef USE_RENDEZVOUS
798 /* Register for Rendezvous only if we opened TCP socket(s) */
799 if (ListenSocket[0] != -1 && rendezvous_name != NULL)
801 DNSServiceRegistrationCreate(rendezvous_name,
804 htonl(PostPortNumber),
806 (DNSServiceRegistrationReply) reg_reply,
811 #ifdef HAVE_UNIX_SOCKETS
812 status = StreamServerPort(AF_UNIX, NULL,
813 (unsigned short) PostPortNumber,
815 ListenSocket, MAXLISTEN);
816 if (status != STATUS_OK)
818 (errmsg("could not create Unix-domain socket")));
822 * check that we have some socket to listen on
824 if (ListenSocket[0] == -1)
826 (errmsg("no socket created for listening")));
831 * Set up shared memory and semaphores.
833 reset_shared(PostPortNumber);
836 * Estimate number of openable files. This must happen after setting up
837 * semaphores, because on some platforms semaphores count as open files.
842 * Initialize the list of active backends.
844 BackendList = DLNewList();
848 * Initialize the child pid/HANDLE arrays
850 win32_childPIDArray = (pid_t*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(pid_t));
851 win32_childHNDArray = (HANDLE*)malloc(NUM_BACKENDARRAY_ELEMS*sizeof(HANDLE));
852 if (!win32_childPIDArray || !win32_childHNDArray)
854 (errcode(ERRCODE_OUT_OF_MEMORY),
855 errmsg("out of memory")));
859 * Record postmaster options. We delay this till now to avoid
860 * recording bogus options (eg, NBuffers too high for available
863 if (!CreateOptsFile(argc, argv, my_exec_path))
867 * Set up signal handlers for the postmaster process.
869 * CAUTION: when changing this list, check for side-effects on the signal
870 * handling setup of child processes. See tcop/postgres.c,
871 * bootstrap/bootstrap.c, and postmaster/pgstat.c.
874 PG_SETMASK(&BlockSig);
876 pqsignal(SIGHUP, SIGHUP_handler); /* reread config file and have
877 * children do same */
878 pqsignal(SIGINT, pmdie); /* send SIGTERM and ShutdownDataBase */
879 pqsignal(SIGQUIT, pmdie); /* send SIGQUIT and die */
880 pqsignal(SIGTERM, pmdie); /* wait for children and ShutdownDataBase */
881 pqsignal(SIGALRM, SIG_IGN); /* ignored */
882 pqsignal(SIGPIPE, SIG_IGN); /* ignored */
883 pqsignal(SIGUSR1, sigusr1_handler); /* message from child process */
884 pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
885 pqsignal(SIGCHLD, reaper); /* handle child termination */
886 pqsignal(SIGTTIN, SIG_IGN); /* ignored */
887 pqsignal(SIGTTOU, SIG_IGN); /* ignored */
888 /* ignore SIGXFSZ, so that ulimit violations work like disk full */
890 pqsignal(SIGXFSZ, SIG_IGN); /* ignored */
894 * Reset whereToSendOutput from Debug (its starting state) to None.
895 * This prevents ereport from sending log messages to stderr unless
896 * the syslog/stderr switch permits. We don't do this until the
897 * postmaster is fully launched, since startup failures may as well be
898 * reported to stderr.
900 whereToSendOutput = None;
903 * On many platforms, the first call of localtime() incurs significant
904 * overhead to load timezone info from the system configuration files.
905 * By doing it once in the postmaster, we avoid having to do it in
906 * every started child process. The savings are not huge, but they
910 time_t now = time(NULL);
912 (void) localtime(&now);
916 * Initialize and try to startup the statistics collector process
922 * Load cached files for client authentication.
930 * We're ready to rock and roll...
932 StartupPID = StartupDataBase();
934 status = ServerLoop();
937 * ServerLoop probably shouldn't ever return, but if it does, close
940 ExitPostmaster(status != STATUS_OK);
942 return 0; /* not reached */
946 pmdaemonize(int argc, char *argv[])
950 elog(FATAL,"SilentMode not supported under WIN32");
956 struct itimerval prof_itimer;
960 /* see comments in BackendRun */
961 getitimer(ITIMER_PROF, &prof_itimer);
965 if (pid == (pid_t) -1)
967 postmaster_error("could not fork background process: %s",
973 /* Parent should just exit, without doing any atexit cleanup */
978 setitimer(ITIMER_PROF, &prof_itimer, NULL);
981 MyProcPid = getpid(); /* reset MyProcPid to child */
983 /* GH: If there's no setsid(), we hopefully don't need silent mode.
984 * Until there's a better solution.
989 postmaster_error("could not dissociate from controlling TTY: %s",
994 i = open(NULL_DEV, O_RDWR | PG_BINARY);
1005 * Print out help message
1008 usage(const char *progname)
1010 printf(gettext("%s is the PostgreSQL server.\n\n"), progname);
1011 printf(gettext("Usage:\n %s [OPTION]...\n\n"), progname);
1012 printf(gettext("Options:\n"));
1013 #ifdef USE_ASSERT_CHECKING
1014 printf(gettext(" -A 1|0 enable/disable run-time assert checking\n"));
1016 printf(gettext(" -B NBUFFERS number of shared buffers\n"));
1017 printf(gettext(" -c NAME=VALUE set run-time parameter\n"));
1018 printf(gettext(" -d 1-5 debugging level\n"));
1019 printf(gettext(" -D DATADIR database directory\n"));
1020 printf(gettext(" -F turn fsync off\n"));
1021 printf(gettext(" -h HOSTNAME host name or IP address to listen on\n"));
1022 printf(gettext(" -i enable TCP/IP connections\n"));
1023 printf(gettext(" -k DIRECTORY Unix-domain socket location\n"));
1025 printf(gettext(" -l enable SSL connections\n"));
1027 printf(gettext(" -N MAX-CONNECT maximum number of allowed connections\n"));
1028 printf(gettext(" -o OPTIONS pass \"OPTIONS\" to each server process\n"));
1029 printf(gettext(" -p PORT port number to listen on\n"));
1030 printf(gettext(" -S silent mode (start in background without logging output)\n"));
1031 printf(gettext(" --help show this help, then exit\n"));
1032 printf(gettext(" --version output version information, then exit\n"));
1034 printf(gettext("\nDeveloper options:\n"));
1035 printf(gettext(" -n do not reinitialize shared memory after abnormal exit\n"));
1036 printf(gettext(" -s send SIGSTOP to all backend servers if one dies\n"));
1038 printf(gettext("\nPlease read the documentation for the complete list of run-time\n"
1039 "configuration settings and how to set them on the command line or in\n"
1040 "the configuration file.\n\n"
1041 "Report bugs to <pgsql-bugs@postgresql.org>.\n"));
1054 gettimeofday(&now, &tz);
1056 nSockets = initMasks(&readmask);
1062 struct timeval timeout;
1065 * The timeout for the select() below is normally set on the basis
1066 * of the time to the next checkpoint. However, if for some
1067 * reason we don't have a next-checkpoint time, time out after 60
1068 * seconds. This keeps checkpoint scheduling from locking up when
1069 * we get new connection requests infrequently (since we are
1070 * likely to detect checkpoint completion just after enabling
1071 * signals below, after we've already made the decision about how
1072 * long to wait this time).
1074 timeout.tv_sec = 60;
1075 timeout.tv_usec = 0;
1077 if (CheckPointPID == 0 && checkpointed &&
1078 StartupPID == 0 && Shutdown == NoShutdown &&
1079 !FatalError && random_seed != 0)
1081 time_t now = time(NULL);
1083 if (CheckPointTimeout + checkpointed > now)
1086 * Not time for checkpoint yet, so set select timeout
1088 timeout.tv_sec = CheckPointTimeout + checkpointed - now;
1092 /* Time to make the checkpoint... */
1093 CheckPointPID = CheckPointDataBase();
1096 * if fork failed, schedule another try at 0.1 normal
1099 if (CheckPointPID == 0)
1101 timeout.tv_sec = CheckPointTimeout / 10;
1102 checkpointed = now + timeout.tv_sec - CheckPointTimeout;
1108 * If no background writer process is running and we should
1109 * do background writing, start one. It doesn't matter if
1110 * this fails, we'll just try again later.
1112 if (BgWriterPID == 0 && BgWriterPercent > 0 &&
1113 StartupPID == 0 && Shutdown == NoShutdown &&
1114 !FatalError && random_seed != 0)
1116 BgWriterPID = StartBackgroundWriter();
1120 * Wait for something to happen.
1122 memcpy((char *) &rmask, (char *) &readmask, sizeof(fd_set));
1124 PG_SETMASK(&UnBlockSig);
1126 if (select(nSockets, &rmask, NULL, NULL, &timeout) < 0)
1128 PG_SETMASK(&BlockSig);
1129 if (errno == EINTR || errno == EWOULDBLOCK)
1132 (errcode_for_socket_access(),
1133 errmsg("select() failed in postmaster: %m")));
1134 return STATUS_ERROR;
1138 * Block all signals until we wait again. (This makes it safe for
1139 * our signal handlers to do nontrivial work.)
1141 PG_SETMASK(&BlockSig);
1144 * Select a random seed at the time of first receiving a request.
1146 while (random_seed == 0)
1148 gettimeofday(&later, &tz);
1151 * We are not sure how much precision is in tv_usec, so we
1152 * swap the nibbles of 'later' and XOR them with 'now'. On the
1153 * off chance that the result is 0, we loop until it isn't.
1155 random_seed = now.tv_usec ^
1156 ((later.tv_usec << 16) |
1157 ((later.tv_usec >> 16) & 0xffff));
1161 * New connection pending on any of our sockets? If so, fork a
1162 * child process to deal with it.
1164 for (i = 0; i < MAXLISTEN; i++)
1166 if (ListenSocket[i] == -1)
1168 if (FD_ISSET(ListenSocket[i], &rmask))
1170 port = ConnCreate(ListenSocket[i]);
1173 BackendStartup(port);
1176 * We no longer need the open socket or port structure
1179 StreamClose(port->sock);
1185 /* If we have lost the stats collector, try to start a new one */
1186 if (!pgstat_is_running)
1193 * Initialise the masks for select() for the ports
1194 * we are listening on. Return the number of sockets to listen on.
1198 initMasks(fd_set *rmask)
1205 for (i = 0; i < MAXLISTEN; i++)
1207 int fd = ListenSocket[i];
1221 * Read the startup packet and do something according to it.
1223 * Returns STATUS_OK or STATUS_ERROR, or might call ereport(FATAL) and
1224 * not return at all.
1226 * (Note that ereport(FATAL) stuff is sent to the client, so only use it
1227 * if that's what you want. Return STATUS_ERROR if you don't want to
1228 * send anything to the client, which would typically be appropriate
1229 * if we detect a communications failure.)
1232 ProcessStartupPacket(Port *port, bool SSLdone)
1236 ProtocolVersion proto;
1237 MemoryContext oldcontext;
1239 if (pq_getbytes((char *) &len, 4) == EOF)
1242 * EOF after SSLdone probably means the client didn't like our
1243 * response to NEGOTIATE_SSL_CODE. That's not an error condition,
1244 * so don't clutter the log with a complaint.
1248 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1249 errmsg("incomplete startup packet")));
1250 return STATUS_ERROR;
1256 if (len < (int32) sizeof(ProtocolVersion) ||
1257 len > MAX_STARTUP_PACKET_LENGTH)
1260 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1261 errmsg("invalid length of startup packet")));
1262 return STATUS_ERROR;
1266 * Allocate at least the size of an old-style startup packet, plus one
1267 * extra byte, and make sure all are zeroes. This ensures we will
1268 * have null termination of all strings, in both fixed- and
1269 * variable-length packet layouts.
1271 if (len <= (int32) sizeof(StartupPacket))
1272 buf = palloc0(sizeof(StartupPacket) + 1);
1274 buf = palloc0(len + 1);
1276 if (pq_getbytes(buf, len) == EOF)
1279 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1280 errmsg("incomplete startup packet")));
1281 return STATUS_ERROR;
1285 * The first field is either a protocol version number or a special
1288 port->proto = proto = ntohl(*((ProtocolVersion *) buf));
1290 if (proto == CANCEL_REQUEST_CODE)
1292 processCancelRequest(port, buf);
1293 return 127; /* XXX */
1296 if (proto == NEGOTIATE_SSL_CODE && !SSLdone)
1301 /* No SSL when disabled or on Unix sockets */
1302 if (!EnableSSL || IS_AF_UNIX(port->laddr.addr.ss_family))
1305 SSLok = 'S'; /* Support for SSL */
1307 SSLok = 'N'; /* No support for SSL */
1309 if (send(port->sock, &SSLok, 1, 0) != 1)
1312 (errcode_for_socket_access(),
1313 errmsg("failed to send SSL negotiation response: %m")));
1314 return STATUS_ERROR; /* close the connection */
1318 if (SSLok == 'S' && secure_open_server(port) == -1)
1319 return STATUS_ERROR;
1321 /* regular startup packet, cancel, etc packet should follow... */
1322 /* but not another SSL negotiation request */
1323 return ProcessStartupPacket(port, true);
1326 /* Could add additional special packet types here */
1329 * Set FrontendProtocol now so that ereport() knows what format to
1330 * send if we fail during startup.
1332 FrontendProtocol = proto;
1334 /* Check we can handle the protocol the frontend is using. */
1336 if (PG_PROTOCOL_MAJOR(proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
1337 PG_PROTOCOL_MAJOR(proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
1338 (PG_PROTOCOL_MAJOR(proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
1339 PG_PROTOCOL_MINOR(proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
1341 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1342 errmsg("unsupported frontend protocol %u.%u: server supports %u.0 to %u.%u",
1343 PG_PROTOCOL_MAJOR(proto), PG_PROTOCOL_MINOR(proto),
1344 PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST),
1345 PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST),
1346 PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST))));
1349 * Now fetch parameters out of startup packet and save them into the
1350 * Port structure. All data structures attached to the Port struct
1351 * must be allocated in TopMemoryContext so that they won't disappear
1352 * when we pass them to PostgresMain (see BackendRun). We need not
1353 * worry about leaking this storage on failure, since we aren't in the
1354 * postmaster process anymore.
1356 oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1358 if (PG_PROTOCOL_MAJOR(proto) >= 3)
1360 int32 offset = sizeof(ProtocolVersion);
1363 * Scan packet body for name/option pairs. We can assume any
1364 * string beginning within the packet body is null-terminated,
1365 * thanks to zeroing extra byte above.
1367 port->guc_options = NIL;
1369 while (offset < len)
1371 char *nameptr = ((char *) buf) + offset;
1375 if (*nameptr == '\0')
1376 break; /* found packet terminator */
1377 valoffset = offset + strlen(nameptr) + 1;
1378 if (valoffset >= len)
1379 break; /* missing value, will complain below */
1380 valptr = ((char *) buf) + valoffset;
1382 if (strcmp(nameptr, "database") == 0)
1383 port->database_name = pstrdup(valptr);
1384 else if (strcmp(nameptr, "user") == 0)
1385 port->user_name = pstrdup(valptr);
1386 else if (strcmp(nameptr, "options") == 0)
1387 port->cmdline_options = pstrdup(valptr);
1390 /* Assume it's a generic GUC option */
1391 port->guc_options = lappend(port->guc_options,
1393 port->guc_options = lappend(port->guc_options,
1396 offset = valoffset + strlen(valptr) + 1;
1400 * If we didn't find a packet terminator exactly at the end of the
1401 * given packet length, complain.
1403 if (offset != len - 1)
1405 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1406 errmsg("invalid startup packet layout: expected terminator as last byte")));
1411 * Get the parameters from the old-style, fixed-width-fields
1412 * startup packet as C strings. The packet destination was
1413 * cleared first so a short packet has zeros silently added. We
1414 * have to be prepared to truncate the pstrdup result for oversize
1417 StartupPacket *packet = (StartupPacket *) buf;
1419 port->database_name = pstrdup(packet->database);
1420 if (strlen(port->database_name) > sizeof(packet->database))
1421 port->database_name[sizeof(packet->database)] = '\0';
1422 port->user_name = pstrdup(packet->user);
1423 if (strlen(port->user_name) > sizeof(packet->user))
1424 port->user_name[sizeof(packet->user)] = '\0';
1425 port->cmdline_options = pstrdup(packet->options);
1426 if (strlen(port->cmdline_options) > sizeof(packet->options))
1427 port->cmdline_options[sizeof(packet->options)] = '\0';
1428 port->guc_options = NIL;
1431 /* Check a user name was given. */
1432 if (port->user_name == NULL || port->user_name[0] == '\0')
1434 (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
1435 errmsg("no PostgreSQL user name specified in startup packet")));
1437 /* The database defaults to the user name. */
1438 if (port->database_name == NULL || port->database_name[0] == '\0')
1439 port->database_name = pstrdup(port->user_name);
1441 if (Db_user_namespace)
1444 * If user@, it is a global user, remove '@'. We only want to do
1445 * this if there is an '@' at the end and no earlier in the user
1446 * string or they may fake as a local user of another database
1447 * attaching to this database.
1449 if (strchr(port->user_name, '@') ==
1450 port->user_name + strlen(port->user_name) - 1)
1451 *strchr(port->user_name, '@') = '\0';
1454 /* Append '@' and dbname */
1457 db_user = palloc(strlen(port->user_name) +
1458 strlen(port->database_name) + 2);
1459 sprintf(db_user, "%s@%s", port->user_name, port->database_name);
1460 port->user_name = db_user;
1465 * Truncate given database and user names to length of a Postgres
1466 * name. This avoids lookup failures when overlength names are given.
1468 if (strlen(port->database_name) >= NAMEDATALEN)
1469 port->database_name[NAMEDATALEN - 1] = '\0';
1470 if (strlen(port->user_name) >= NAMEDATALEN)
1471 port->user_name[NAMEDATALEN - 1] = '\0';
1474 * Done putting stuff in TopMemoryContext.
1476 MemoryContextSwitchTo(oldcontext);
1479 * If we're going to reject the connection due to database state, say
1480 * so now instead of wasting cycles on an authentication exchange.
1481 * (This also allows a pg_ping utility to be written.)
1483 switch (port->canAcceptConnections)
1487 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1488 errmsg("the database system is starting up")));
1492 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1493 errmsg("the database system is shutting down")));
1497 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
1498 errmsg("the database system is in recovery mode")));
1502 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
1503 errmsg("sorry, too many clients already")));
1515 * The client has sent a cancel request packet, not a normal
1516 * start-a-new-connection packet. Perform the necessary processing.
1517 * Nothing is sent back to the client.
1520 processCancelRequest(Port *port, void *pkt)
1522 CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
1524 long cancelAuthCode;
1526 #ifndef EXEC_BACKEND
1532 backendPID = (int) ntohl(canc->backendPID);
1533 cancelAuthCode = (long) ntohl(canc->cancelAuthCode);
1535 if (backendPID == CheckPointPID)
1538 (errmsg_internal("ignoring cancel request for checkpoint process %d",
1542 else if (backendPID == BgWriterPID)
1545 (errmsg_internal("ignoring cancel request for bgwriter process %d",
1550 /* See if we have a matching backend */
1551 #ifndef EXEC_BACKEND
1552 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
1554 bp = (Backend *) DLE_VAL(curr);
1556 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
1558 bp = (Backend*) &ShmemBackendArray[i];
1560 if (bp->pid == backendPID)
1562 if (bp->cancel_key == cancelAuthCode)
1564 /* Found a match; signal that backend to cancel current op */
1566 (errmsg_internal("processing cancel request: sending SIGINT to process %d",
1568 kill(bp->pid, SIGINT);
1571 /* Right PID, wrong key: no way, Jose */
1573 (errmsg_internal("bad key in cancel request for process %d",
1579 /* No matching backend */
1581 (errmsg_internal("bad pid in cancel request for process %d",
1586 * canAcceptConnections --- check to see if database state allows connections.
1588 static enum CAC_state
1589 canAcceptConnections(void)
1591 /* Can't start backends when in startup/shutdown/recovery state. */
1592 if (Shutdown > NoShutdown)
1593 return CAC_SHUTDOWN;
1597 return CAC_RECOVERY;
1600 * Don't start too many children.
1602 * We allow more connections than we can have backends here because some
1603 * might still be authenticating; they might fail auth, or some
1604 * existing backend might exit before the auth cycle is completed. The
1605 * exact MaxBackends limit is enforced when a new backend tries to
1606 * join the shared-inval backend array.
1608 if (CountChildren() >= 2 * MaxBackends)
1616 * ConnCreate -- create a local connection data structure
1619 ConnCreate(int serverFd)
1623 if (!(port = (Port *) calloc(1, sizeof(Port))))
1626 (errcode(ERRCODE_OUT_OF_MEMORY),
1627 errmsg("out of memory")));
1631 if (StreamConnection(serverFd, port) != STATUS_OK)
1633 StreamClose(port->sock);
1640 * Precompute password salt values to use for this connection.
1641 * It's slightly annoying to do this long in advance of knowing
1642 * whether we'll need 'em or not, but we must do the random()
1643 * calls before we fork, not after. Else the postmaster's random
1644 * sequence won't get advanced, and all backends would end up
1645 * using the same salt...
1647 RandomSalt(port->cryptSalt, port->md5Salt);
1655 * ConnFree -- free a local connection data structure
1658 ConnFree(Port *conn)
1668 * ClosePostmasterPorts -- close all the postmaster's open sockets
1670 * This is called during child process startup to release file descriptors
1671 * that are not needed by that child process. The postmaster still has
1672 * them open, of course.
1675 ClosePostmasterPorts(bool pgstat_too)
1679 /* Close the listen sockets */
1680 for (i = 0; i < MAXLISTEN; i++)
1682 if (ListenSocket[i] != -1)
1684 StreamClose(ListenSocket[i]);
1685 ListenSocket[i] = -1;
1689 /* Close pgstat control sockets, unless we're starting pgstat itself */
1691 pgstat_close_sockets();
1696 * reset_shared -- reset shared memory and semaphores
1699 reset_shared(unsigned short port)
1702 * Create or re-create shared memory and semaphores.
1704 * Note: in each "cycle of life" we will normally assign the same IPC
1705 * keys (if using SysV shmem and/or semas), since the port number is
1706 * used to determine IPC keys. This helps ensure that we will clean
1707 * up dead IPC objects if the postmaster crashes and is restarted.
1709 CreateSharedMemoryAndSemaphores(false, MaxBackends, port);
1714 * SIGHUP -- reread config files, and tell children to do same
1717 SIGHUP_handler(SIGNAL_ARGS)
1719 int save_errno = errno;
1721 PG_SETMASK(&BlockSig);
1723 if (Shutdown <= SmartShutdown)
1726 (errmsg("received SIGHUP, reloading configuration files")));
1727 ProcessConfigFile(PGC_SIGHUP);
1729 write_nondefault_variables(PGC_SIGHUP);
1731 SignalChildren(SIGHUP);
1736 * Tell the background writer to terminate so that we
1737 * will start a new one with a possibly changed config
1739 if (BgWriterPID != 0)
1740 kill(BgWriterPID, SIGTERM);
1743 PG_SETMASK(&UnBlockSig);
1751 * pmdie -- signal handler for processing various postmaster signals.
1756 int save_errno = errno;
1758 PG_SETMASK(&BlockSig);
1761 (errmsg_internal("postmaster received signal %d",
1762 postgres_signal_arg)));
1764 switch (postgres_signal_arg)
1771 * Wait for children to end their work and ShutdownDataBase.
1773 if (Shutdown >= SmartShutdown)
1775 Shutdown = SmartShutdown;
1777 (errmsg("received smart shutdown request")));
1779 /* Must tell bgwriter to quit, or it never will... */
1780 if (BgWriterPID != 0)
1781 kill(BgWriterPID, SIGTERM);
1783 if (DLGetHead(BackendList)) /* let reaper() handle this */
1787 * No children left. Shutdown data base system.
1789 if (StartupPID > 0 || FatalError) /* let reaper() handle
1792 if (ShutdownPID > 0)
1794 elog(PANIC, "shutdown process %d already running",
1799 ShutdownPID = ShutdownDataBase();
1807 * Abort all children with SIGTERM (rollback active transactions
1808 * and exit) and ShutdownDataBase when they are gone.
1810 if (Shutdown >= FastShutdown)
1812 Shutdown = FastShutdown;
1814 (errmsg("received fast shutdown request")));
1816 if (DLGetHead(BackendList))
1821 (errmsg("aborting any active transactions")));
1822 SignalChildren(SIGTERM);
1823 /* reaper() does the rest */
1829 * No children left. Shutdown data base system.
1831 * Unlike the previous case, it is not an error for the shutdown
1832 * process to be running already (we could get SIGTERM followed
1833 * shortly later by SIGINT).
1835 if (StartupPID > 0 || FatalError) /* let reaper() handle
1838 if (ShutdownPID == 0)
1839 ShutdownPID = ShutdownDataBase();
1845 * Immediate Shutdown:
1847 * abort all children with SIGQUIT and exit without attempt to
1848 * properly shutdown data base system.
1851 (errmsg("received immediate shutdown request")));
1852 if (ShutdownPID > 0)
1853 kill(ShutdownPID, SIGQUIT);
1855 kill(StartupPID, SIGQUIT);
1856 if (DLGetHead(BackendList))
1857 SignalChildren(SIGQUIT);
1862 PG_SETMASK(&UnBlockSig);
1868 * Reaper -- signal handler to cleanup after a backend (child) dies.
1873 int save_errno = errno;
1876 int status; /* backend exit status */
1879 union wait status; /* backend exit status */
1883 int pid; /* process id of dead backend */
1885 PG_SETMASK(&BlockSig);
1888 (errmsg_internal("reaping dead processes")));
1890 while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
1892 exitstatus = status;
1895 while ((pid = wait3(&status, WNOHANG, NULL)) > 0)
1897 exitstatus = status.w_status;
1899 while ((pid = win32_waitpid(&exitstatus)) > 0)
1902 * We need to do this here, and not in CleanupProc, since this
1903 * is to be called on all children when we are done with them.
1904 * Could move to LogChildExit, but that seems like asking for
1907 win32_RemoveChild(pid);
1912 * Check if this child was the statistics collector. If so, try to
1913 * start a new one. (If fail, we'll try again in future cycles of
1916 if (pgstat_ispgstat(pid))
1918 LogChildExit(LOG, gettext("statistics collector process"),
1925 * Check if this child was a shutdown or startup process.
1927 if (ShutdownPID > 0 && pid == ShutdownPID)
1929 if (exitstatus != 0)
1931 LogChildExit(LOG, gettext("shutdown process"),
1935 /* Normal postmaster exit is here */
1939 if (StartupPID > 0 && pid == StartupPID)
1941 if (exitstatus != 0)
1943 LogChildExit(LOG, gettext("startup process"),
1946 (errmsg("aborting startup due to startup process failure")));
1952 * Startup succeeded - remember its ID and RedoRecPtr.
1954 * NB: this MUST happen before we fork a checkpoint or shutdown
1955 * subprocess, else they will have wrong local ThisStartUpId.
1959 FatalError = false; /* done with recovery */
1962 * Arrange for first checkpoint to occur after standard delay.
1965 checkpointed = time(NULL);
1968 * Go to shutdown mode if a shutdown request was pending.
1970 if (Shutdown > NoShutdown)
1972 if (ShutdownPID > 0)
1974 elog(PANIC, "startup process %d died while shutdown process %d already running",
1975 pid, (int) ShutdownPID);
1978 ShutdownPID = ShutdownDataBase();
1985 * Else do standard child cleanup.
1987 CleanupProc(pid, exitstatus);
1989 } /* loop over pending child-death reports */
1994 * Wait for all children exit, then reset shmem and
1997 if (DLGetHead(BackendList) || StartupPID > 0 || ShutdownPID > 0)
2000 (errmsg("all server processes terminated; reinitializing")));
2003 reset_shared(PostPortNumber);
2005 StartupPID = StartupDataBase();
2010 if (Shutdown > NoShutdown)
2012 if (DLGetHead(BackendList))
2014 if (StartupPID > 0 || ShutdownPID > 0)
2016 ShutdownPID = ShutdownDataBase();
2020 PG_SETMASK(&UnBlockSig);
2027 * CleanupProc -- cleanup after terminated backend.
2029 * Remove all local state associated with backend.
2032 CleanupProc(int pid,
2033 int exitstatus) /* child's exit status. */
2039 LogChildExit(DEBUG2, gettext("child process"), pid, exitstatus);
2042 * If a backend dies in an ugly way (i.e. exit status not 0) then we
2043 * must signal all other backends to quickdie. If exit status is zero
2044 * we assume everything is hunky dory and simply remove the backend
2045 * from the active backend list.
2047 if (exitstatus == 0)
2049 curr = DLGetHead(BackendList);
2052 bp = (Backend *) DLE_VAL(curr);
2056 ShmemBackendArrayRemove(bp->pid);
2063 curr = DLGetSucc(curr);
2066 if (pid == CheckPointPID)
2071 checkpointed = time(NULL);
2072 /* Update RedoRecPtr for future child backends */
2073 GetSavedRedoRecPtr();
2076 else if (pid == BgWriterPID)
2084 /* below here we're dealing with a non-normal exit */
2086 /* Make log entry unless we did so already */
2090 (pid == CheckPointPID) ? gettext("checkpoint process") :
2091 (pid == BgWriterPID) ? gettext("bgwriter process") :
2092 gettext("server process"),
2095 (errmsg("terminating any other active server processes")));
2098 curr = DLGetHead(BackendList);
2101 next = DLGetSucc(curr);
2102 bp = (Backend *) DLE_VAL(curr);
2106 * This backend is still alive. Unless we did so already,
2107 * tell it to commit hara-kiri.
2109 * SIGQUIT is the special signal that says exit without proc_exit
2110 * and let the user know what's going on. But if SendStop is
2111 * set (-s on command line), then we send SIGSTOP instead, so
2112 * that we can get core dumps from all backends by hand.
2117 (errmsg_internal("sending %s to process %d",
2118 (SendStop ? "SIGSTOP" : "SIGQUIT"),
2120 kill(bp->pid, (SendStop ? SIGSTOP : SIGQUIT));
2126 * Found entry for freshly-dead backend, so remove it.
2129 ShmemBackendArrayRemove(bp->pid);
2138 if (pid == CheckPointPID)
2143 else if (pid == BgWriterPID)
2150 * Tell the collector about backend termination
2159 * Log the death of a child process.
2162 LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2164 if (WIFEXITED(exitstatus))
2168 * translator: %s is a noun phrase describing a child process,
2169 * such as "server process"
2171 (errmsg("%s (PID %d) exited with exit code %d",
2172 procname, pid, WEXITSTATUS(exitstatus))));
2173 else if (WIFSIGNALED(exitstatus))
2177 * translator: %s is a noun phrase describing a child process,
2178 * such as "server process"
2180 (errmsg("%s (PID %d) was terminated by signal %d",
2181 procname, pid, WTERMSIG(exitstatus))));
2186 * translator: %s is a noun phrase describing a child process,
2187 * such as "server process"
2189 (errmsg("%s (PID %d) exited with unexpected status %d",
2190 procname, pid, exitstatus)));
2194 * Send a signal to all backend children.
2197 SignalChildren(int signal)
2203 curr = DLGetHead(BackendList);
2206 next = DLGetSucc(curr);
2207 bp = (Backend *) DLE_VAL(curr);
2209 if (bp->pid != MyProcPid)
2212 (errmsg_internal("sending signal %d to process %d",
2215 kill(bp->pid, signal);
2223 * BackendStartup -- start backend process
2225 * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
2228 BackendStartup(Port *port)
2230 Backend *bn; /* for backend cleanup */
2233 #ifdef LINUX_PROFILE
2234 struct itimerval prof_itimer;
2238 * Compute the cancel key that will be assigned to this backend. The
2239 * backend will have its own copy in the forked-off process' value of
2240 * MyCancelKey, so that it can transmit the key to the frontend.
2242 MyCancelKey = PostmasterRandom();
2245 * Make room for backend data structure. Better before the fork() so
2246 * we can handle failure cleanly.
2248 bn = (Backend *) malloc(sizeof(Backend));
2252 (errcode(ERRCODE_OUT_OF_MEMORY),
2253 errmsg("out of memory")));
2254 return STATUS_ERROR;
2258 * Flush stdio channels just before fork, to avoid double-output
2259 * problems. Ideally we'd use fflush(NULL) here, but there are still a
2260 * few non-ANSI stdio libraries out there (like SunOS 4.1.x) that
2261 * coredump if we do. Presently stdout and stderr are the only stdio
2262 * output channels used by the postmaster, so fflush'ing them should
2268 #ifdef LINUX_PROFILE
2271 * Linux's fork() resets the profiling timer in the child process. If
2272 * we want to profile child processes then we need to save and restore
2273 * the timer setting. This is a waste of time if not profiling,
2274 * however, so only do it if commanded by specific -DLINUX_PROFILE
2277 getitimer(ITIMER_PROF, &prof_itimer);
2281 /* Specific beos actions before backend startup */
2282 beos_before_backend_startup();
2285 port->canAcceptConnections = canAcceptConnections();
2287 pid = Backend_forkexec(port);
2291 if (pid == 0) /* child */
2293 #ifdef LINUX_PROFILE
2294 setitimer(ITIMER_PROF, &prof_itimer, NULL);
2298 /* Specific beos backend startup actions */
2299 beos_backend_startup();
2303 proc_exit(BackendRun(port));
2307 /* in parent, error */
2310 int save_errno = errno;
2313 /* Specific beos backend startup actions */
2314 beos_backend_startup_failed();
2319 (errmsg("could not fork new process for connection: %m")));
2320 report_fork_failure_to_client(port, save_errno);
2321 return STATUS_ERROR;
2324 /* in parent, normal */
2326 (errmsg_internal("forked new backend, pid=%d socket=%d",
2327 (int) pid, port->sock)));
2330 * Everything's been successful, it's safe to add this backend to our
2334 bn->cancel_key = MyCancelKey;
2336 ShmemBackendArrayAdd(bn);
2338 DLAddHead(BackendList, DLNewElem(bn));
2344 * Try to report backend fork() failure to client before we close the
2345 * connection. Since we do not care to risk blocking the postmaster on
2346 * this connection, we set the connection to non-blocking and try only once.
2348 * This is grungy special-purpose code; we cannot use backend libpq since
2349 * it's not up and running.
2352 report_fork_failure_to_client(Port *port, int errnum)
2356 /* Format the error message packet (always V2 protocol) */
2357 snprintf(buffer, sizeof(buffer), "E%s%s\n",
2358 gettext("could not fork new process for connection: "),
2361 /* Set port to non-blocking. Don't do send() if this fails */
2362 if (!set_noblock(port->sock))
2365 send(port->sock, buffer, strlen(buffer) + 1, 0);
2370 * split_opts -- split a string of options and append it to an argv array
2372 * NB: the string is destructively modified!
2374 * Since no current POSTGRES arguments require any quoting characters,
2375 * we can use the simple-minded tactic of assuming each set of space-
2376 * delimited characters is a separate argv element.
2378 * If you don't like that, well, we *used* to pass the whole option string
2379 * as ONE argument to execl(), which was even less intelligent...
2382 split_opts(char **argv, int *argcp, char *s)
2386 while (isspace((unsigned char) *s))
2390 argv[(*argcp)++] = s;
2391 while (*s && !isspace((unsigned char) *s))
2400 * BackendInit/Run -- perform authentication [BackendInit], and if successful,
2401 * set up the backend's argument list [BackendRun] and invoke
2405 * Shouldn't return at all.
2406 * If PostgresMain() fails, return status.
2409 BackendInit(Port *port)
2414 char remote_host[NI_MAXHOST];
2415 char remote_port[NI_MAXSERV];
2416 char remote_ps_data[NI_MAXHOST];
2418 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2420 ClientAuthInProgress = true; /* limit visibility of log messages */
2422 /* We don't want the postmaster's proc_exit() handlers */
2426 * Signal handlers setting is moved to tcop/postgres...
2429 /* save start time for end of session reporting */
2430 gettimeofday(&(port->session_start),NULL);
2432 /* set these to empty in case they are needed before we set them up */
2433 port->remote_host = "";
2434 port->remote_port = "";
2435 port->commandTag = "";
2437 /* Save port etc. for ps status */
2440 /* Reset MyProcPid to new backend's pid */
2441 MyProcPid = getpid();
2444 * Initialize libpq and enable reporting of ereport errors to the
2445 * client. Must do this now because authentication uses libpq to send
2448 pq_init(); /* initialize libpq to talk to client */
2449 whereToSendOutput = Remote; /* now safe to ereport to client */
2452 * We arrange for a simple exit(0) if we receive SIGTERM or SIGQUIT
2453 * during any client authentication related communication. Otherwise
2454 * the postmaster cannot shutdown the database FAST or IMMED cleanly
2455 * if a buggy client blocks a backend during authentication.
2457 pqsignal(SIGTERM, authdie);
2458 pqsignal(SIGQUIT, authdie);
2459 pqsignal(SIGALRM, authdie);
2460 PG_SETMASK(&AuthBlockSig);
2463 * Get the remote host name and port for logging and status display.
2465 remote_host[0] = '\0';
2466 remote_port[0] = '\0';
2467 if (getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2468 remote_host, sizeof(remote_host),
2469 remote_port, sizeof(remote_port),
2470 (log_hostname ? 0 : NI_NUMERICHOST) | NI_NUMERICSERV))
2472 getnameinfo_all(&port->raddr.addr, port->raddr.salen,
2473 remote_host, sizeof(remote_host),
2474 remote_port, sizeof(remote_port),
2475 NI_NUMERICHOST | NI_NUMERICSERV);
2477 snprintf(remote_ps_data, sizeof(remote_ps_data),
2478 remote_port[0] == '\0' ? "%s" : "%s(%s)",
2479 remote_host, remote_port);
2481 if (Log_connections)
2483 (errmsg("connection received: host=%s port=%s",
2484 remote_host, remote_port)));
2487 * save remote_host and remote_port in port stucture
2489 port->remote_host = strdup(remote_host);
2490 port->remote_port = strdup(remote_port);
2493 * Ready to begin client interaction. We will give up and exit(0)
2494 * after a time delay, so that a broken client can't hog a connection
2495 * indefinitely. PreAuthDelay doesn't count against the time limit.
2497 if (!enable_sig_alarm(AuthenticationTimeout * 1000, false))
2498 elog(FATAL, "could not set timer for authorization timeout");
2501 * Receive the startup packet (which might turn out to be a cancel
2504 status = ProcessStartupPacket(port, false);
2506 if (status != STATUS_OK)
2510 * Now that we have the user and database name, we can set the process
2511 * title for ps. It's good to do this as early as possible in
2514 init_ps_display(port->user_name, port->database_name, remote_ps_data);
2515 set_ps_display("authentication");
2518 * Now perform authentication exchange.
2520 ClientAuthentication(port); /* might not return, if failure */
2523 * Done with authentication. Disable timeout, and prevent
2524 * SIGTERM/SIGQUIT again until backend startup is complete.
2526 if (!disable_sig_alarm(false))
2527 elog(FATAL, "could not disable timer for authorization timeout");
2528 PG_SETMASK(&BlockSig);
2530 if (Log_connections)
2532 (errmsg("connection authorized: user=%s database=%s",
2533 port->user_name, port->database_name)));
2536 * Don't want backend to be able to see the postmaster random number
2537 * generator state. We have to clobber the static random_seed *and*
2538 * start a new random sequence in the random() library function.
2541 gettimeofday(&now, &tz);
2542 srandom((unsigned int) now.tv_usec);
2547 BackendRun(Port *port)
2557 * Let's clean up ourselves as the postmaster child, and
2558 * close the postmaster's other sockets
2560 ClosePostmasterPorts(true);
2563 * PreAuthDelay is a debugging aid for investigating problems in the
2564 * authentication cycle: it can be set in postgresql.conf to allow
2565 * time to attach to the newly-forked backend with a debugger. (See
2566 * also the -W backend switch, which we allow clients to pass through
2567 * PGOPTIONS, but it is not honored until after authentication.)
2569 if (PreAuthDelay > 0)
2570 pg_usleep(PreAuthDelay*1000000L);
2572 /* Will exit on failure */
2577 * Now, build the argv vector that will be given to PostgresMain.
2579 * The layout of the command line is
2580 * postgres [secure switches] -p databasename [insecure switches]
2581 * where the switches after -p come from the client request.
2583 * The maximum possible number of commandline arguments that could come
2584 * from ExtraOptions or port->cmdline_options is (strlen + 1) / 2; see
2588 maxac = 10; /* for fixed args supplied below */
2589 maxac += (strlen(ExtraOptions) + 1) / 2;
2590 if (port->cmdline_options)
2591 maxac += (strlen(port->cmdline_options) + 1) / 2;
2593 av = (char **) MemoryContextAlloc(TopMemoryContext,
2594 maxac * sizeof(char *));
2597 av[ac++] = "postgres";
2600 * Pass the requested debugging level along to the backend.
2604 snprintf(debugbuf, sizeof(debugbuf), "-d%d", debug_flag);
2605 av[ac++] = debugbuf;
2609 * Pass any backend switches specified with -o in the postmaster's own
2610 * command line. We assume these are secure.
2612 split_opts(av, &ac, ExtraOptions);
2614 /* Tell the backend what protocol the frontend is using. */
2615 snprintf(protobuf, sizeof(protobuf), "-v%u", port->proto);
2616 av[ac++] = protobuf;
2619 /* pass data dir before end of secure switches (-p) */
2625 * Tell the backend it is being called from the postmaster, and which
2626 * database to use. -p marks the end of secure switches.
2629 av[ac++] = port->database_name;
2632 * Pass the (insecure) option switches from the connection request.
2633 * (It's OK to mangle port->cmdline_options now.)
2635 if (port->cmdline_options)
2636 split_opts(av, &ac, port->cmdline_options);
2643 * Release postmaster's working memory context so that backend can
2644 * recycle the space. Note this does not trash *MyProcPort, because
2645 * ConnCreate() allocated that space with malloc() ... else we'd need
2646 * to copy the Port data here. Also, subsidiary data such as the
2647 * username isn't lost either; see ProcessStartupPacket().
2649 MemoryContextSwitchTo(TopMemoryContext);
2650 #ifndef EXEC_BACKEND
2651 MemoryContextDelete(PostmasterContext);
2653 PostmasterContext = NULL;
2656 * Debug: print arguments being passed to backend
2659 (errmsg_internal("%s child[%d]: starting with (",
2660 progname, getpid())));
2661 for (i = 0; i < ac; ++i)
2663 (errmsg_internal("\t%s", av[i])));
2665 (errmsg_internal(")")));
2667 ClientAuthInProgress = false; /* client_min_messages is active
2670 return (PostgresMain(ac, av, port->user_name));
2678 * SubPostmasterMain -- prepare the fork/exec'd process to be in an equivalent
2679 * state (for calling BackendRun) as a forked process.
2682 * Shouldn't return at all.
2685 SubPostmasterMain(int argc, char* argv[])
2687 unsigned long backendID;
2690 memset((void*)&port, 0, sizeof(Port));
2693 /* Do this sooner rather than later... */
2694 IsUnderPostmaster = true; /* we are a postmaster subprocess now */
2696 /* In EXEC case we will not have inherited these settings */
2697 IsPostmasterEnvironment = true;
2698 whereToSendOutput = None;
2700 /* Setup global context */
2701 MemoryContextInit();
2702 InitializeGUCOptions();
2704 /* Parse passed-in context */
2706 backendID = (unsigned long)atol(argv[argc++]);
2707 DataDir = strdup(argv[argc++]);
2709 /* Read in file-based context */
2710 read_nondefault_variables();
2711 read_backend_variables(backendID,&port);
2713 /* Remaining initialization */
2714 pgstat_init_forkexec_backend();
2716 /* FIXME: [fork/exec] Ugh */
2722 /* Attach process to shared segments */
2723 CreateSharedMemoryAndSemaphores(false, MaxBackends, 0);
2726 proc_exit(BackendRun(&port));
2731 * Backend_forkexec -- fork/exec off a backend process
2734 * the pid of the fork/exec'd process
2737 Backend_forkexec(Port *port)
2741 int ac = 0, bufc = 0, i;
2742 char buf[2][MAXPGPATH];
2744 if (!write_backend_variables(port))
2745 return -1; /* log made by write_backend_variables */
2747 av[ac++] = "postgres";
2748 av[ac++] = "-forkexec";
2750 /* Format up context to pass to exec'd process */
2751 snprintf(buf[bufc++],MAXPGPATH,"%lu",tmpBackendFileNum);
2752 snprintf(buf[bufc++],MAXPGPATH,"\"%s\"",DataDir);
2754 /* Add to the arg list */
2755 Assert(bufc <= lengthof(buf));
2756 for (i = 0; i < bufc; i++)
2759 /* FIXME: [fork/exec] ExtraOptions? */
2762 Assert(ac <= lengthof(av));
2765 pid = win32_forkexec(postgres_exec_path, av); /* logs on error */
2767 /* Fire off execv in child */
2768 if ((pid = fork()) == 0 && (execv(postgres_exec_path, av) == -1))
2770 * FIXME: [fork/exec] suggestions for what to do here?
2771 * Probably OK to issue error (unlike pgstat case)
2775 return pid; /* Parent returns pid */
2782 * ExitPostmaster -- cleanup
2784 * Do NOT call exit() directly --- always go through here!
2787 ExitPostmaster(int status)
2789 /* should cleanup shared memory and kill all backends */
2792 * Not sure of the semantics here. When the Postmaster dies, should
2793 * the backends all be killed? probably not.
2795 * MUST -- vadim 05-10-1999
2797 /* Should I use true instead? */
2798 ClosePostmasterPorts(false);
2804 * sigusr1_handler - handle signal conditions from child processes
2807 sigusr1_handler(SIGNAL_ARGS)
2809 int save_errno = errno;
2811 PG_SETMASK(&BlockSig);
2813 if (CheckPostmasterSignal(PMSIGNAL_DO_CHECKPOINT))
2815 if (CheckPointWarning != 0)
2818 * This only times checkpoints forced by running out of
2819 * segment files. Other checkpoints could reduce the
2820 * frequency of forced checkpoints.
2822 time_t now = time(NULL);
2824 if (LastSignalledCheckpoint != 0)
2826 int elapsed_secs = now - LastSignalledCheckpoint;
2828 if (elapsed_secs < CheckPointWarning)
2830 (errmsg("checkpoints are occurring too frequently (%d seconds apart)",
2832 errhint("Consider increasing the configuration parameter \"checkpoint_segments\".")));
2834 LastSignalledCheckpoint = now;
2838 * Request to schedule a checkpoint
2840 * Ignore request if checkpoint is already running or checkpointing
2841 * is currently disabled
2843 if (CheckPointPID == 0 && checkpointed &&
2844 StartupPID == 0 && Shutdown == NoShutdown &&
2845 !FatalError && random_seed != 0)
2847 CheckPointPID = CheckPointDataBase();
2848 /* note: if fork fails, CheckPointPID stays 0; nothing happens */
2852 if (CheckPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE))
2855 * Password or group file has changed.
2861 if (CheckPostmasterSignal(PMSIGNAL_WAKEN_CHILDREN))
2864 * Send SIGUSR2 to all children (triggers AsyncNotifyHandler). See
2865 * storage/ipc/sinvaladt.c for the use of this.
2867 if (Shutdown == NoShutdown)
2868 SignalChildren(SIGUSR2);
2871 PG_SETMASK(&UnBlockSig);
2878 * Dummy signal handler
2880 * We use this for signals that we don't actually use in the postmaster,
2881 * but we do use in backends. If we SIG_IGN such signals in the postmaster,
2882 * then a newly started backend might drop a signal that arrives before it's
2883 * able to reconfigure its signal processing. (See notes in postgres.c.)
2886 dummy_handler(SIGNAL_ARGS)
2892 * CharRemap: given an int in range 0..61, produce textual encoding of it
2893 * per crypt(3) conventions.
2917 RandomSalt(char *cryptSalt, char *md5Salt)
2919 long rand = PostmasterRandom();
2921 cryptSalt[0] = CharRemap(rand % 62);
2922 cryptSalt[1] = CharRemap(rand / 62);
2925 * It's okay to reuse the first random value for one of the MD5 salt
2926 * bytes, since only one of the two salts will be sent to the client.
2927 * After that we need to compute more random bits.
2929 * We use % 255, sacrificing one possible byte value, so as to ensure
2930 * that all bits of the random() value participate in the result.
2931 * While at it, add one to avoid generating any null bytes.
2933 md5Salt[0] = (rand % 255) + 1;
2934 rand = PostmasterRandom();
2935 md5Salt[1] = (rand % 255) + 1;
2936 rand = PostmasterRandom();
2937 md5Salt[2] = (rand % 255) + 1;
2938 rand = PostmasterRandom();
2939 md5Salt[3] = (rand % 255) + 1;
2946 PostmasterRandom(void)
2948 static bool initialized = false;
2952 Assert(random_seed != 0);
2953 srandom(random_seed);
2961 * Count up number of child processes.
2970 for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
2972 bp = (Backend *) DLE_VAL(curr);
2973 if (bp->pid != MyProcPid)
2976 /* Checkpoint and bgwriter will be in the list, discount them */
2977 if (CheckPointPID != 0)
2979 if (BgWriterPID != 0)
2985 * Fire off a subprocess for startup/shutdown/checkpoint/bgwriter.
2987 * Return value of SSDataBase is subprocess' PID, or 0 if failed to start subprocess
2988 * (0 is returned only for checkpoint/bgwriter cases).
2990 * note: in the EXEC_BACKEND case, we delay the fork until argument list has been
2993 NON_EXEC_STATIC void
2994 SSDataBaseInit(int xlop)
2996 const char *statmsg;
2998 IsUnderPostmaster = true; /* we are a postmaster subprocess
3002 /* In EXEC case we will not have inherited these settings */
3003 IsPostmasterEnvironment = true;
3004 whereToSendOutput = None;
3007 MyProcPid = getpid(); /* reset MyProcPid */
3009 /* Lose the postmaster's on-exit routines and port connections */
3013 * Identify myself via ps
3017 case BS_XLOG_STARTUP:
3018 statmsg = "startup subprocess";
3020 case BS_XLOG_CHECKPOINT:
3021 statmsg = "checkpoint subprocess";
3023 case BS_XLOG_BGWRITER:
3024 statmsg = "bgwriter subprocess";
3026 case BS_XLOG_SHUTDOWN:
3027 statmsg = "shutdown subprocess";
3030 statmsg = "??? subprocess";
3033 init_ps_display(statmsg, "", "");
3039 SSDataBase(int xlop)
3043 #ifndef EXEC_BACKEND
3044 #ifdef LINUX_PROFILE
3045 struct itimerval prof_itimer;
3049 char ddirbuf[MAXPGPATH];
3055 #ifndef EXEC_BACKEND
3056 #ifdef LINUX_PROFILE
3057 /* see comments in BackendRun */
3058 getitimer(ITIMER_PROF, &prof_itimer);
3062 /* Specific beos actions before backend startup */
3063 beos_before_backend_startup();
3066 /* Non EXEC_BACKEND case; fork here */
3067 if ((pid = fork()) == 0) /* child */
3075 #ifndef EXEC_BACKEND
3076 #ifdef LINUX_PROFILE
3077 setitimer(ITIMER_PROF, &prof_itimer, NULL);
3081 /* Specific beos actions after backend startup */
3082 beos_backend_startup();
3085 /* Close the postmaster's sockets */
3086 ClosePostmasterPorts(true);
3088 SSDataBaseInit(xlop);
3090 if (!write_backend_variables(NULL))
3091 return -1; /* log issued by write_backend_variables */
3094 /* Set up command-line arguments for subprocess */
3095 av[ac++] = "postgres";
3100 snprintf(nbbuf, sizeof(nbbuf), "-B%d", NBuffers);
3103 snprintf(xlbuf, sizeof(xlbuf), "-x%d", xlop);
3107 /* pass data dir before end of secure switches (-p) */
3108 snprintf(ddirbuf, MAXPGPATH, "\"%s\"", DataDir);
3112 /* and the backend identifier + dbname */
3113 snprintf(idbuf, sizeof(idbuf), "-p%lu,template1", tmpBackendFileNum);
3117 av[ac++] = "template1";
3122 Assert(ac < lengthof(av));
3125 /* EXEC_BACKEND case; fork/exec here */
3127 pid = win32_forkexec(postgres_exec_path, av); /* logs on error */
3129 if ((pid = fork()) == 0 && (execv(postgres_exec_path, av) == -1))
3132 elog(ERROR, "unable to execv in SSDataBase: %m");
3137 BootstrapMain(ac, av);
3145 #ifndef EXEC_BACKEND
3147 /* Specific beos actions before backend startup */
3148 beos_backend_startup_failed();
3153 case BS_XLOG_STARTUP:
3155 (errmsg("could not fork startup process: %m")));
3157 case BS_XLOG_CHECKPOINT:
3159 (errmsg("could not fork checkpoint process: %m")));
3161 case BS_XLOG_BGWRITER:
3163 (errmsg("could not fork bgwriter process: %m")));
3165 case BS_XLOG_SHUTDOWN:
3167 (errmsg("could not fork shutdown process: %m")));
3171 (errmsg("could not fork process: %m")));
3176 * fork failure is fatal during startup/shutdown, but there's no
3177 * need to choke if a routine checkpoint or starting a background
3180 if (xlop == BS_XLOG_CHECKPOINT)
3182 if (xlop == BS_XLOG_BGWRITER)
3188 * The startup and shutdown processes are not considered normal
3189 * backends, but the checkpoint and bgwriter processes are.
3190 * They must be added to the list of backends.
3192 if (xlop == BS_XLOG_CHECKPOINT || xlop == BS_XLOG_BGWRITER)
3194 if (!(bn = (Backend *) malloc(sizeof(Backend))))
3197 (errcode(ERRCODE_OUT_OF_MEMORY),
3198 errmsg("out of memory")));
3203 bn->cancel_key = PostmasterRandom();
3205 ShmemBackendArrayAdd(bn);
3207 DLAddHead(BackendList, DLNewElem(bn));
3210 * Since this code is executed periodically, it's a fine place to
3211 * do other actions that should happen every now and then on no
3212 * particular schedule. Such as...
3215 TouchSocketLockFile();
3223 * Create the opts file
3226 CreateOptsFile(int argc, char *argv[], char *fullprogname)
3228 char filename[MAXPGPATH];
3232 snprintf(filename, sizeof(filename), "%s/postmaster.opts", DataDir);
3234 if ((fp = fopen(filename, "w")) == NULL)
3236 elog(LOG, "could not create file \"%s\": %m", filename);
3240 fprintf(fp, "%s", fullprogname);
3241 for (i = 1; i < argc; i++)
3242 fprintf(fp, " '%s'", argv[i]);
3247 elog(LOG, "could not write file \"%s\": %m", filename);
3255 * This should be used only for reporting "interactive" errors (essentially,
3256 * bogus arguments on the command line). Once the postmaster is launched,
3257 * use ereport. In particular, don't use this for anything that occurs
3258 * after pmdaemonize.
3261 postmaster_error(const char *fmt,...)
3265 fprintf(stderr, "%s: ", progname);
3267 vfprintf(stderr, gettext(fmt), ap);
3269 fprintf(stderr, "\n");
3276 * The following need to be available to the read/write_backend_variables
3279 #include "storage/spin.h"
3280 extern XLogRecPtr RedoRecPtr;
3281 extern XLogwrtResult LogwrtResult;
3282 extern slock_t *ShmemLock;
3283 extern slock_t *ShmemIndexLock;
3284 extern void *ShmemIndexAlloc;
3285 typedef struct LWLock LWLock;
3286 extern LWLock *LWLockArray;
3287 extern slock_t *ProcStructLock;
3288 extern int pgStatSock;
3290 #define write_var(var,fp) fwrite((void*)&(var),sizeof(var),1,fp)
3291 #define read_var(var,fp) fread((void*)&(var),sizeof(var),1,fp)
3292 #define get_tmp_backend_file_name(buf,id) \
3296 "%s/%s/%s.backend_var.%lu", \
3298 PG_TEMP_FILES_DIR, \
3299 PG_TEMP_FILE_PREFIX, \
3304 write_backend_variables(Port *port)
3306 char filename[MAXPGPATH];
3308 get_tmp_backend_file_name(filename,++tmpBackendFileNum);
3311 fp = AllocateFile(filename, PG_BINARY_W);
3314 /* As per OpenTemporaryFile... */
3315 char dirname[MAXPGPATH];
3316 sprintf(dirname,"%s/%s",DataDir,PG_TEMP_FILES_DIR);
3317 mkdir(dirname, S_IRWXU);
3319 fp = AllocateFile(filename, PG_BINARY_W);
3323 (errcode_for_file_access(),
3324 errmsg("could not write to file \"%s\": %m", filename)));
3332 write_var(port->sock,fp);
3333 write_var(port->proto,fp);
3334 write_var(port->laddr,fp);
3335 write_var(port->raddr,fp);
3336 write_var(port->canAcceptConnections,fp);
3337 write_var(port->cryptSalt,fp);
3338 write_var(port->md5Salt,fp);
3340 write_var(MyCancelKey,fp);
3342 write_var(RedoRecPtr,fp);
3343 write_var(LogwrtResult,fp);
3345 write_var(UsedShmemSegID,fp);
3346 write_var(UsedShmemSegAddr,fp);
3348 write_var(ShmemLock,fp);
3349 write_var(ShmemIndexLock,fp);
3350 write_var(ShmemVariableCache,fp);
3351 write_var(ShmemIndexAlloc,fp);
3352 write_var(ShmemBackendArray,fp);
3354 write_var(LWLockArray,fp);
3355 write_var(ProcStructLock,fp);
3356 write_var(pgStatSock,fp);
3358 write_var(PreAuthDelay,fp);
3359 write_var(debug_flag,fp);
3360 write_var(PostmasterPid,fp);
3366 (errcode_for_file_access(),
3367 errmsg("could not write to file \"%s\": %m", filename)));
3375 read_backend_variables(unsigned long id, Port *port)
3377 char filename[MAXPGPATH];
3379 get_tmp_backend_file_name(filename,id);
3382 fp = AllocateFile(filename, PG_BINARY_R);
3386 (errcode_for_file_access(),
3387 errmsg("could not read from backend_variables file \"%s\": %m", filename)));
3394 read_var(port->sock,fp);
3395 read_var(port->proto,fp);
3396 read_var(port->laddr,fp);
3397 read_var(port->raddr,fp);
3398 read_var(port->canAcceptConnections,fp);
3399 read_var(port->cryptSalt,fp);
3400 read_var(port->md5Salt,fp);
3402 read_var(MyCancelKey,fp);
3404 read_var(RedoRecPtr,fp);
3405 read_var(LogwrtResult,fp);
3407 read_var(UsedShmemSegID,fp);
3408 read_var(UsedShmemSegAddr,fp);
3410 read_var(ShmemLock,fp);
3411 read_var(ShmemIndexLock,fp);
3412 read_var(ShmemVariableCache,fp);
3413 read_var(ShmemIndexAlloc,fp);
3414 read_var(ShmemBackendArray,fp);
3416 read_var(LWLockArray,fp);
3417 read_var(ProcStructLock,fp);
3418 read_var(pgStatSock,fp);
3420 read_var(PreAuthDelay,fp);
3421 read_var(debug_flag,fp);
3422 read_var(PostmasterPid,fp);
3426 if (unlink(filename) != 0)
3428 (errcode_for_file_access(),
3429 errmsg("could not remove file \"%s\": %m", filename)));
3433 size_t ShmemBackendArraySize(void)
3435 return (NUM_BACKENDARRAY_ELEMS*sizeof(Backend));
3438 void ShmemBackendArrayAllocation(void)
3440 size_t size = ShmemBackendArraySize();
3441 ShmemBackendArray = (Backend*)ShmemAlloc(size);
3442 memset(ShmemBackendArray, 0, size);
3445 static void ShmemBackendArrayAdd(Backend *bn)
3448 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3450 /* Find an empty slot */
3451 if (ShmemBackendArray[i].pid == 0)
3453 ShmemBackendArray[i] = *bn;
3459 (errmsg_internal("unable to add backend entry")));
3462 static void ShmemBackendArrayRemove(pid_t pid)
3465 for (i = 0; i < NUM_BACKENDARRAY_ELEMS; i++)
3467 if (ShmemBackendArray[i].pid == pid)
3469 /* Mark the slot as empty */
3470 ShmemBackendArray[i].pid = 0;
3476 (errmsg_internal("unable to find backend entry with pid %d",
3484 pid_t win32_forkexec(const char* path, char *argv[])
3487 PROCESS_INFORMATION pi;
3490 char cmdLine[MAXPGPATH];
3491 HANDLE childHandleCopy;
3492 HANDLE waiterThread;
3494 /* Format the cmd line */
3495 snprintf(cmdLine,sizeof(cmdLine),"\"%s\"",path);
3497 while (argv[++i] != NULL)
3499 /* FIXME: [fork/exec] some strlen checks might be prudent here */
3500 strcat(cmdLine," ");
3501 strcat(cmdLine,argv[i]);
3505 * The following snippet can disappear when we consistently
3506 * use forward slashes.
3509 while (*(p++) != '\0')
3510 if (*p == '/') *p = '\\';
3512 memset(&pi,0,sizeof(pi));
3513 memset(&si,0,sizeof(si));
3515 if (!CreateProcess(NULL,cmdLine,NULL,NULL,TRUE,0,NULL,NULL,&si,&pi))
3517 elog(ERROR,"CreateProcess call failed (%i): %m",(int)GetLastError());
3521 if (!IsUnderPostmaster)
3522 /* We are the Postmaster creating a child... */
3523 win32_AddChild(pi.dwProcessId,pi.hProcess);
3525 if (!DuplicateHandle(GetCurrentProcess(),
3527 GetCurrentProcess(),
3531 DUPLICATE_SAME_ACCESS))
3533 (errmsg_internal("failed to duplicate child handle: %i",(int)GetLastError())));
3534 waiterThread = CreateThread(NULL, 64*1024, win32_sigchld_waiter, (LPVOID)childHandleCopy, 0, NULL);
3537 (errmsg_internal("failed to create sigchld waiter thread: %i",(int)GetLastError())));
3538 CloseHandle(waiterThread);
3540 if (IsUnderPostmaster)
3541 CloseHandle(pi.hProcess);
3542 CloseHandle(pi.hThread);
3544 return pi.dwProcessId;
3548 * Note: The following three functions must not be interrupted (eg. by signals).
3549 * As the Postgres Win32 signalling architecture (currently) requires polling,
3550 * or APC checking functions which aren't used here, this is not an issue.
3552 * We keep two separate arrays, instead of a single array of pid/HANDLE structs,
3553 * to avoid having to re-create a handle array for WaitForMultipleObjects on
3554 * each call to win32_waitpid.
3557 static void win32_AddChild(pid_t pid, HANDLE handle)
3559 Assert(win32_childPIDArray && win32_childHNDArray);
3560 if (win32_numChildren < NUM_BACKENDARRAY_ELEMS)
3562 win32_childPIDArray[win32_numChildren] = pid;
3563 win32_childHNDArray[win32_numChildren] = handle;
3564 ++win32_numChildren;
3568 (errmsg_internal("unable to add child entry with pid %lu",
3572 static void win32_RemoveChild(pid_t pid)
3575 Assert(win32_childPIDArray && win32_childHNDArray);
3577 for (i = 0; i < win32_numChildren; i++)
3579 if (win32_childPIDArray[i] == pid)
3581 CloseHandle(win32_childHNDArray[i]);
3583 /* Swap last entry into the "removed" one */
3584 --win32_numChildren;
3585 win32_childPIDArray[i] = win32_childPIDArray[win32_numChildren];
3586 win32_childHNDArray[i] = win32_childHNDArray[win32_numChildren];
3592 (errmsg_internal("unable to find child entry with pid %lu",
3596 static pid_t win32_waitpid(int *exitstatus)
3598 Assert(win32_childPIDArray && win32_childHNDArray);
3599 elog(DEBUG3,"waiting on %lu children",win32_numChildren);
3601 if (win32_numChildren > 0)
3604 * Note: Do NOT use WaitForMultipleObjectsEx, as we don't
3605 * want to run queued APCs here.
3609 DWORD ret = WaitForMultipleObjects(win32_numChildren,win32_childHNDArray,FALSE,0);
3615 (errmsg_internal("failed to wait on %lu children: %i",
3616 win32_numChildren,(int)GetLastError())));
3617 /* Fall through to WAIT_TIMEOUTs return */
3620 /* No children have finished */
3624 /* Get the exit code, and return the PID of, the respective process */
3625 index = ret-WAIT_OBJECT_0;
3626 Assert(index >= 0 && index < win32_numChildren);
3627 if (!GetExitCodeProcess(win32_childHNDArray[index],&exitCode))
3629 * If we get this far, this should never happen, but, then again...
3630 * No choice other than to assume a catastrophic failure.
3633 (errmsg_internal("failed to get exit code for child %lu",
3634 win32_childPIDArray[index])));
3635 *exitstatus = (int)exitCode;
3636 return win32_childPIDArray[index];
3644 /* Note! Code belows executes on separate threads, one for
3645 each child process created */
3646 static DWORD WINAPI win32_sigchld_waiter(LPVOID param) {
3647 HANDLE procHandle = (HANDLE)param;
3649 DWORD r = WaitForSingleObject(procHandle, INFINITE);
3650 if (r == WAIT_OBJECT_0)
3651 pg_queue_signal(SIGCHLD);
3653 fprintf(stderr,"ERROR: Failed to wait on child process handle: %i\n",(int)GetLastError());
3654 CloseHandle(procHandle);