Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * miscinit.c
4 : * miscellaneous initialization support stuff
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/init/miscinit.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <sys/param.h>
18 : #include <signal.h>
19 : #include <time.h>
20 : #include <sys/file.h>
21 : #include <sys/stat.h>
22 : #include <sys/time.h>
23 : #include <fcntl.h>
24 : #include <unistd.h>
25 : #include <grp.h>
26 : #include <pwd.h>
27 : #include <netinet/in.h>
28 : #include <arpa/inet.h>
29 : #include <utime.h>
30 :
31 : #include "access/htup_details.h"
32 : #include "access/parallel.h"
33 : #include "catalog/pg_authid.h"
34 : #include "common/file_perm.h"
35 : #include "libpq/libpq.h"
36 : #include "libpq/pqsignal.h"
37 : #include "mb/pg_wchar.h"
38 : #include "miscadmin.h"
39 : #include "pgstat.h"
40 : #include "postmaster/autovacuum.h"
41 : #include "postmaster/interrupt.h"
42 : #include "postmaster/postmaster.h"
43 : #include "replication/slotsync.h"
44 : #include "storage/fd.h"
45 : #include "storage/ipc.h"
46 : #include "storage/latch.h"
47 : #include "storage/pg_shmem.h"
48 : #include "storage/pmsignal.h"
49 : #include "storage/proc.h"
50 : #include "storage/procarray.h"
51 : #include "utils/builtins.h"
52 : #include "utils/guc.h"
53 : #include "utils/inval.h"
54 : #include "utils/memutils.h"
55 : #include "utils/pidfile.h"
56 : #include "utils/syscache.h"
57 : #include "utils/varlena.h"
58 :
59 :
60 : #define DIRECTORY_LOCK_FILE "postmaster.pid"
61 :
62 : ProcessingMode Mode = InitProcessing;
63 :
64 : BackendType MyBackendType;
65 :
66 : /* List of lock files to be removed at proc exit */
67 : static List *lock_files = NIL;
68 :
69 : static Latch LocalLatchData;
70 :
71 : /* ----------------------------------------------------------------
72 : * ignoring system indexes support stuff
73 : *
74 : * NOTE: "ignoring system indexes" means we do not use the system indexes
75 : * for lookups (either in hardwired catalog accesses or in planner-generated
76 : * plans). We do, however, still update the indexes when a catalog
77 : * modification is made.
78 : * ----------------------------------------------------------------
79 : */
80 :
81 : bool IgnoreSystemIndexes = false;
82 :
83 :
84 : /* ----------------------------------------------------------------
85 : * common process startup code
86 : * ----------------------------------------------------------------
87 : */
88 :
89 : /*
90 : * Initialize the basic environment for a postmaster child
91 : *
92 : * Should be called as early as possible after the child's startup. However,
93 : * on EXEC_BACKEND builds it does need to be after read_backend_variables().
94 : */
95 : void
96 45180 : InitPostmasterChild(void)
97 : {
98 45180 : IsUnderPostmaster = true; /* we are a postmaster subprocess now */
99 :
100 : /*
101 : * Start our win32 signal implementation. This has to be done after we
102 : * read the backend variables, because we need to pick up the signal pipe
103 : * from the parent process.
104 : */
105 : #ifdef WIN32
106 : pgwin32_signal_initialize();
107 : #endif
108 :
109 45180 : InitProcessGlobals();
110 :
111 : /*
112 : * make sure stderr is in binary mode before anything can possibly be
113 : * written to it, in case it's actually the syslogger pipe, so the pipe
114 : * chunking protocol isn't disturbed. Non-logpipe data gets translated on
115 : * redirection (e.g. via pg_ctl -l) anyway.
116 : */
117 : #ifdef WIN32
118 : _setmode(fileno(stderr), _O_BINARY);
119 : #endif
120 :
121 : /* We don't want the postmaster's proc_exit() handlers */
122 45180 : on_exit_reset();
123 :
124 : /* In EXEC_BACKEND case we will not have inherited BlockSig etc values */
125 : #ifdef EXEC_BACKEND
126 : pqinitmask();
127 : #endif
128 :
129 : /* Initialize process-local latch support */
130 45180 : InitializeWaitEventSupport();
131 45180 : InitProcessLocalLatch();
132 45180 : InitializeLatchWaitSet();
133 :
134 : /*
135 : * If possible, make this process a group leader, so that the postmaster
136 : * can signal any child processes too. Not all processes will have
137 : * children, but for consistency we make all postmaster child processes do
138 : * this.
139 : */
140 : #ifdef HAVE_SETSID
141 45180 : if (setsid() < 0)
142 0 : elog(FATAL, "setsid() failed: %m");
143 : #endif
144 :
145 : /*
146 : * Every postmaster child process is expected to respond promptly to
147 : * SIGQUIT at all times. Therefore we centrally remove SIGQUIT from
148 : * BlockSig and install a suitable signal handler. (Client-facing
149 : * processes may choose to replace this default choice of handler with
150 : * quickdie().) All other blockable signals remain blocked for now.
151 : */
152 45180 : pqsignal(SIGQUIT, SignalHandlerForCrashExit);
153 :
154 45180 : sigdelset(&BlockSig, SIGQUIT);
155 45180 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
156 :
157 : /* Request a signal if the postmaster dies, if possible. */
158 45180 : PostmasterDeathSignalInit();
159 :
160 : /* Don't give the pipe to subprograms that we execute. */
161 : #ifndef WIN32
162 45180 : if (fcntl(postmaster_alive_fds[POSTMASTER_FD_WATCH], F_SETFD, FD_CLOEXEC) < 0)
163 0 : ereport(FATAL,
164 : (errcode_for_socket_access(),
165 : errmsg_internal("could not set postmaster death monitoring pipe to FD_CLOEXEC mode: %m")));
166 : #endif
167 45180 : }
168 :
169 : /*
170 : * Initialize the basic environment for a standalone process.
171 : *
172 : * argv0 has to be suitable to find the program's executable.
173 : */
174 : void
175 492 : InitStandaloneProcess(const char *argv0)
176 : {
177 : Assert(!IsPostmasterEnvironment);
178 :
179 492 : MyBackendType = B_STANDALONE_BACKEND;
180 :
181 : /*
182 : * Start our win32 signal implementation
183 : */
184 : #ifdef WIN32
185 : pgwin32_signal_initialize();
186 : #endif
187 :
188 492 : InitProcessGlobals();
189 :
190 : /* Initialize process-local latch support */
191 492 : InitializeWaitEventSupport();
192 492 : InitProcessLocalLatch();
193 492 : InitializeLatchWaitSet();
194 :
195 : /*
196 : * For consistency with InitPostmasterChild, initialize signal mask here.
197 : * But we don't unblock SIGQUIT or provide a default handler for it.
198 : */
199 492 : pqinitmask();
200 492 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
201 :
202 : /* Compute paths, no postmaster to inherit from */
203 492 : if (my_exec_path[0] == '\0')
204 : {
205 492 : if (find_my_exec(argv0, my_exec_path) < 0)
206 0 : elog(FATAL, "%s: could not locate my own executable path",
207 : argv0);
208 : }
209 :
210 492 : if (pkglib_path[0] == '\0')
211 492 : get_pkglib_path(my_exec_path, pkglib_path);
212 492 : }
213 :
214 : void
215 44680 : SwitchToSharedLatch(void)
216 : {
217 : Assert(MyLatch == &LocalLatchData);
218 : Assert(MyProc != NULL);
219 :
220 44680 : MyLatch = &MyProc->procLatch;
221 :
222 44680 : if (FeBeWaitSet)
223 26116 : ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetLatchPos, WL_LATCH_SET,
224 : MyLatch);
225 :
226 : /*
227 : * Set the shared latch as the local one might have been set. This
228 : * shouldn't normally be necessary as code is supposed to check the
229 : * condition before waiting for the latch, but a bit care can't hurt.
230 : */
231 44680 : SetLatch(MyLatch);
232 44680 : }
233 :
234 : void
235 47436 : InitProcessLocalLatch(void)
236 : {
237 47436 : MyLatch = &LocalLatchData;
238 47436 : InitLatch(MyLatch);
239 47436 : }
240 :
241 : void
242 44680 : SwitchBackToLocalLatch(void)
243 : {
244 : Assert(MyLatch != &LocalLatchData);
245 : Assert(MyProc != NULL && MyLatch == &MyProc->procLatch);
246 :
247 44680 : MyLatch = &LocalLatchData;
248 :
249 44680 : if (FeBeWaitSet)
250 26116 : ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetLatchPos, WL_LATCH_SET,
251 : MyLatch);
252 :
253 44680 : SetLatch(MyLatch);
254 44680 : }
255 :
256 : /*
257 : * Return a human-readable string representation of a BackendType.
258 : *
259 : * The string is not localized here, but we mark the strings for translation
260 : * so that callers can invoke _() on the result.
261 : */
262 : const char *
263 1652926 : GetBackendTypeDesc(BackendType backendType)
264 : {
265 1652926 : const char *backendDesc = gettext_noop("unknown process type");
266 :
267 1652926 : switch (backendType)
268 : {
269 : #define PG_PROCTYPE(bktype, description, main_func, shmem_attach) \
270 : case bktype: backendDesc = description; break;
271 : #include "postmaster/proctypelist.h"
272 : #undef PG_PROCTYPE
273 : }
274 1652926 : return backendDesc;
275 : }
276 :
277 : /* ----------------------------------------------------------------
278 : * database path / name support stuff
279 : * ----------------------------------------------------------------
280 : */
281 :
282 : void
283 33328 : SetDatabasePath(const char *path)
284 : {
285 : /* This should happen only once per process */
286 : Assert(!DatabasePath);
287 33328 : DatabasePath = MemoryContextStrdup(TopMemoryContext, path);
288 33328 : }
289 :
290 : /*
291 : * Validate the proposed data directory.
292 : *
293 : * Also initialize file and directory create modes and mode mask.
294 : */
295 : void
296 2192 : checkDataDir(void)
297 : {
298 : struct stat stat_buf;
299 :
300 : Assert(DataDir);
301 :
302 2192 : if (stat(DataDir, &stat_buf) != 0)
303 : {
304 0 : if (errno == ENOENT)
305 0 : ereport(FATAL,
306 : (errcode_for_file_access(),
307 : errmsg("data directory \"%s\" does not exist",
308 : DataDir)));
309 : else
310 0 : ereport(FATAL,
311 : (errcode_for_file_access(),
312 : errmsg("could not read permissions of directory \"%s\": %m",
313 : DataDir)));
314 : }
315 :
316 : /* eventual chdir would fail anyway, but let's test ... */
317 2192 : if (!S_ISDIR(stat_buf.st_mode))
318 0 : ereport(FATAL,
319 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
320 : errmsg("specified data directory \"%s\" is not a directory",
321 : DataDir)));
322 :
323 : /*
324 : * Check that the directory belongs to my userid; if not, reject.
325 : *
326 : * This check is an essential part of the interlock that prevents two
327 : * postmasters from starting in the same directory (see CreateLockFile()).
328 : * Do not remove or weaken it.
329 : *
330 : * XXX can we safely enable this check on Windows?
331 : */
332 : #if !defined(WIN32) && !defined(__CYGWIN__)
333 2192 : if (stat_buf.st_uid != geteuid())
334 0 : ereport(FATAL,
335 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
336 : errmsg("data directory \"%s\" has wrong ownership",
337 : DataDir),
338 : errhint("The server must be started by the user that owns the data directory.")));
339 : #endif
340 :
341 : /*
342 : * Check if the directory has correct permissions. If not, reject.
343 : *
344 : * Only two possible modes are allowed, 0700 and 0750. The latter mode
345 : * indicates that group read/execute should be allowed on all newly
346 : * created files and directories.
347 : *
348 : * XXX temporarily suppress check when on Windows, because there may not
349 : * be proper support for Unix-y file permissions. Need to think of a
350 : * reasonable check to apply on Windows.
351 : */
352 : #if !defined(WIN32) && !defined(__CYGWIN__)
353 2192 : if (stat_buf.st_mode & PG_MODE_MASK_GROUP)
354 0 : ereport(FATAL,
355 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
356 : errmsg("data directory \"%s\" has invalid permissions",
357 : DataDir),
358 : errdetail("Permissions should be u=rwx (0700) or u=rwx,g=rx (0750).")));
359 : #endif
360 :
361 : /*
362 : * Reset creation modes and mask based on the mode of the data directory.
363 : *
364 : * The mask was set earlier in startup to disallow group permissions on
365 : * newly created files and directories. However, if group read/execute
366 : * are present on the data directory then modify the create modes and mask
367 : * to allow group read/execute on newly created files and directories and
368 : * set the data_directory_mode GUC.
369 : *
370 : * Suppress when on Windows, because there may not be proper support for
371 : * Unix-y file permissions.
372 : */
373 : #if !defined(WIN32) && !defined(__CYGWIN__)
374 2192 : SetDataDirectoryCreatePerm(stat_buf.st_mode);
375 :
376 2192 : umask(pg_mode_mask);
377 2192 : data_directory_mode = pg_dir_create_mode;
378 : #endif
379 :
380 : /* Check for PG_VERSION */
381 2192 : ValidatePgVersion(DataDir);
382 2192 : }
383 :
384 : /*
385 : * Set data directory, but make sure it's an absolute path. Use this,
386 : * never set DataDir directly.
387 : */
388 : void
389 2204 : SetDataDir(const char *dir)
390 : {
391 : char *new;
392 :
393 : Assert(dir);
394 :
395 : /* If presented path is relative, convert to absolute */
396 2204 : new = make_absolute_path(dir);
397 :
398 2204 : free(DataDir);
399 2204 : DataDir = new;
400 2204 : }
401 :
402 : /*
403 : * Change working directory to DataDir. Most of the postmaster and backend
404 : * code assumes that we are in DataDir so it can use relative paths to access
405 : * stuff in and under the data directory. For convenience during path
406 : * setup, however, we don't force the chdir to occur during SetDataDir.
407 : */
408 : void
409 2192 : ChangeToDataDir(void)
410 : {
411 : Assert(DataDir);
412 :
413 2192 : if (chdir(DataDir) < 0)
414 0 : ereport(FATAL,
415 : (errcode_for_file_access(),
416 : errmsg("could not change directory to \"%s\": %m",
417 : DataDir)));
418 2192 : }
419 :
420 :
421 : /* ----------------------------------------------------------------
422 : * User ID state
423 : *
424 : * We have to track several different values associated with the concept
425 : * of "user ID".
426 : *
427 : * AuthenticatedUserId is determined at connection start and never changes.
428 : *
429 : * SessionUserId is initially the same as AuthenticatedUserId, but can be
430 : * changed by SET SESSION AUTHORIZATION (if AuthenticatedUserId is a
431 : * superuser). This is the ID reported by the SESSION_USER SQL function.
432 : *
433 : * OuterUserId is the current user ID in effect at the "outer level" (outside
434 : * any transaction or function). This is initially the same as SessionUserId,
435 : * but can be changed by SET ROLE to any role that SessionUserId is a
436 : * member of. (XXX rename to something like CurrentRoleId?)
437 : *
438 : * CurrentUserId is the current effective user ID; this is the one to use
439 : * for all normal permissions-checking purposes. At outer level this will
440 : * be the same as OuterUserId, but it changes during calls to SECURITY
441 : * DEFINER functions, as well as locally in some specialized commands.
442 : *
443 : * SecurityRestrictionContext holds flags indicating reason(s) for changing
444 : * CurrentUserId. In some cases we need to lock down operations that are
445 : * not directly controlled by privilege settings, and this provides a
446 : * convenient way to do it.
447 : * ----------------------------------------------------------------
448 : */
449 : static Oid AuthenticatedUserId = InvalidOid;
450 : static Oid SessionUserId = InvalidOid;
451 : static Oid OuterUserId = InvalidOid;
452 : static Oid CurrentUserId = InvalidOid;
453 : static const char *SystemUser = NULL;
454 :
455 : /* We also have to remember the superuser state of the session user */
456 : static bool SessionUserIsSuperuser = false;
457 :
458 : static int SecurityRestrictionContext = 0;
459 :
460 : /* We also remember if a SET ROLE is currently active */
461 : static bool SetRoleIsActive = false;
462 :
463 : /*
464 : * GetUserId - get the current effective user ID.
465 : *
466 : * Note: there's no SetUserId() anymore; use SetUserIdAndSecContext().
467 : */
468 : Oid
469 13746456 : GetUserId(void)
470 : {
471 : Assert(OidIsValid(CurrentUserId));
472 13746456 : return CurrentUserId;
473 : }
474 :
475 :
476 : /*
477 : * GetOuterUserId/SetOuterUserId - get/set the outer-level user ID.
478 : */
479 : Oid
480 1562 : GetOuterUserId(void)
481 : {
482 : Assert(OidIsValid(OuterUserId));
483 1562 : return OuterUserId;
484 : }
485 :
486 :
487 : static void
488 84474 : SetOuterUserId(Oid userid, bool is_superuser)
489 : {
490 : Assert(SecurityRestrictionContext == 0);
491 : Assert(OidIsValid(userid));
492 84474 : OuterUserId = userid;
493 :
494 : /* We force the effective user ID to match, too */
495 84474 : CurrentUserId = userid;
496 :
497 : /* Also update the is_superuser GUC to match OuterUserId's property */
498 84474 : SetConfigOption("is_superuser",
499 : is_superuser ? "on" : "off",
500 : PGC_INTERNAL, PGC_S_DYNAMIC_DEFAULT);
501 84474 : }
502 :
503 :
504 : /*
505 : * GetSessionUserId/SetSessionUserId - get/set the session user ID.
506 : */
507 : Oid
508 70774 : GetSessionUserId(void)
509 : {
510 : Assert(OidIsValid(SessionUserId));
511 70774 : return SessionUserId;
512 : }
513 :
514 : bool
515 3668 : GetSessionUserIsSuperuser(void)
516 : {
517 : Assert(OidIsValid(SessionUserId));
518 3668 : return SessionUserIsSuperuser;
519 : }
520 :
521 : static void
522 41404 : SetSessionUserId(Oid userid, bool is_superuser)
523 : {
524 : Assert(SecurityRestrictionContext == 0);
525 : Assert(OidIsValid(userid));
526 41404 : SessionUserId = userid;
527 41404 : SessionUserIsSuperuser = is_superuser;
528 41404 : }
529 :
530 : /*
531 : * Return the system user representing the authenticated identity.
532 : * It is defined in InitializeSystemUser() as auth_method:authn_id.
533 : */
534 : const char *
535 44 : GetSystemUser(void)
536 : {
537 44 : return SystemUser;
538 : }
539 :
540 : /*
541 : * GetAuthenticatedUserId/SetAuthenticatedUserId - get/set the authenticated
542 : * user ID
543 : */
544 : Oid
545 33128 : GetAuthenticatedUserId(void)
546 : {
547 : Assert(OidIsValid(AuthenticatedUserId));
548 33128 : return AuthenticatedUserId;
549 : }
550 :
551 : void
552 29772 : SetAuthenticatedUserId(Oid userid)
553 : {
554 : Assert(OidIsValid(userid));
555 :
556 : /* call only once */
557 : Assert(!OidIsValid(AuthenticatedUserId));
558 :
559 29772 : AuthenticatedUserId = userid;
560 :
561 : /* Also mark our PGPROC entry with the authenticated user id */
562 : /* (We assume this is an atomic store so no lock is needed) */
563 29772 : MyProc->roleId = userid;
564 29772 : }
565 :
566 :
567 : /*
568 : * GetUserIdAndSecContext/SetUserIdAndSecContext - get/set the current user ID
569 : * and the SecurityRestrictionContext flags.
570 : *
571 : * Currently there are three valid bits in SecurityRestrictionContext:
572 : *
573 : * SECURITY_LOCAL_USERID_CHANGE indicates that we are inside an operation
574 : * that is temporarily changing CurrentUserId via these functions. This is
575 : * needed to indicate that the actual value of CurrentUserId is not in sync
576 : * with guc.c's internal state, so SET ROLE has to be disallowed.
577 : *
578 : * SECURITY_RESTRICTED_OPERATION indicates that we are inside an operation
579 : * that does not wish to trust called user-defined functions at all. The
580 : * policy is to use this before operations, e.g. autovacuum and REINDEX, that
581 : * enumerate relations of a database or schema and run functions associated
582 : * with each found relation. The relation owner is the new user ID. Set this
583 : * as soon as possible after locking the relation. Restore the old user ID as
584 : * late as possible before closing the relation; restoring it shortly after
585 : * close is also tolerable. If a command has both relation-enumerating and
586 : * non-enumerating modes, e.g. ANALYZE, both modes set this bit. This bit
587 : * prevents not only SET ROLE, but various other changes of session state that
588 : * normally is unprotected but might possibly be used to subvert the calling
589 : * session later. An example is replacing an existing prepared statement with
590 : * new code, which will then be executed with the outer session's permissions
591 : * when the prepared statement is next used. These restrictions are fairly
592 : * draconian, but the functions called in relation-enumerating operations are
593 : * really supposed to be side-effect-free anyway.
594 : *
595 : * SECURITY_NOFORCE_RLS indicates that we are inside an operation which should
596 : * ignore the FORCE ROW LEVEL SECURITY per-table indication. This is used to
597 : * ensure that FORCE RLS does not mistakenly break referential integrity
598 : * checks. Note that this is intentionally only checked when running as the
599 : * owner of the table (which should always be the case for referential
600 : * integrity checks).
601 : *
602 : * Unlike GetUserId, GetUserIdAndSecContext does *not* Assert that the current
603 : * value of CurrentUserId is valid; nor does SetUserIdAndSecContext require
604 : * the new value to be valid. In fact, these routines had better not
605 : * ever throw any kind of error. This is because they are used by
606 : * StartTransaction and AbortTransaction to save/restore the settings,
607 : * and during the first transaction within a backend, the value to be saved
608 : * and perhaps restored is indeed invalid. We have to be able to get
609 : * through AbortTransaction without asserting in case InitPostgres fails.
610 : */
611 : void
612 1859150 : GetUserIdAndSecContext(Oid *userid, int *sec_context)
613 : {
614 1859150 : *userid = CurrentUserId;
615 1859150 : *sec_context = SecurityRestrictionContext;
616 1859150 : }
617 :
618 : void
619 1509258 : SetUserIdAndSecContext(Oid userid, int sec_context)
620 : {
621 1509258 : CurrentUserId = userid;
622 1509258 : SecurityRestrictionContext = sec_context;
623 1509258 : }
624 :
625 :
626 : /*
627 : * InLocalUserIdChange - are we inside a local change of CurrentUserId?
628 : */
629 : bool
630 68004 : InLocalUserIdChange(void)
631 : {
632 68004 : return (SecurityRestrictionContext & SECURITY_LOCAL_USERID_CHANGE) != 0;
633 : }
634 :
635 : /*
636 : * InSecurityRestrictedOperation - are we inside a security-restricted command?
637 : */
638 : bool
639 80322 : InSecurityRestrictedOperation(void)
640 : {
641 80322 : return (SecurityRestrictionContext & SECURITY_RESTRICTED_OPERATION) != 0;
642 : }
643 :
644 : /*
645 : * InNoForceRLSOperation - are we ignoring FORCE ROW LEVEL SECURITY ?
646 : */
647 : bool
648 264 : InNoForceRLSOperation(void)
649 : {
650 264 : return (SecurityRestrictionContext & SECURITY_NOFORCE_RLS) != 0;
651 : }
652 :
653 :
654 : /*
655 : * These are obsolete versions of Get/SetUserIdAndSecContext that are
656 : * only provided for bug-compatibility with some rather dubious code in
657 : * pljava. We allow the userid to be set, but only when not inside a
658 : * security restriction context.
659 : */
660 : void
661 0 : GetUserIdAndContext(Oid *userid, bool *sec_def_context)
662 : {
663 0 : *userid = CurrentUserId;
664 0 : *sec_def_context = InLocalUserIdChange();
665 0 : }
666 :
667 : void
668 0 : SetUserIdAndContext(Oid userid, bool sec_def_context)
669 : {
670 : /* We throw the same error SET ROLE would. */
671 0 : if (InSecurityRestrictedOperation())
672 0 : ereport(ERROR,
673 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
674 : errmsg("cannot set parameter \"%s\" within security-restricted operation",
675 : "role")));
676 0 : CurrentUserId = userid;
677 0 : if (sec_def_context)
678 0 : SecurityRestrictionContext |= SECURITY_LOCAL_USERID_CHANGE;
679 : else
680 0 : SecurityRestrictionContext &= ~SECURITY_LOCAL_USERID_CHANGE;
681 0 : }
682 :
683 :
684 : /*
685 : * Check whether specified role has explicit REPLICATION privilege
686 : */
687 : bool
688 3472 : has_rolreplication(Oid roleid)
689 : {
690 3472 : bool result = false;
691 : HeapTuple utup;
692 :
693 : /* Superusers bypass all permission checking. */
694 3472 : if (superuser_arg(roleid))
695 3362 : return true;
696 :
697 110 : utup = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid));
698 110 : if (HeapTupleIsValid(utup))
699 : {
700 110 : result = ((Form_pg_authid) GETSTRUCT(utup))->rolreplication;
701 110 : ReleaseSysCache(utup);
702 : }
703 110 : return result;
704 : }
705 :
706 : /*
707 : * Initialize user identity during normal backend startup
708 : */
709 : void
710 29776 : InitializeSessionUserId(const char *rolename, Oid roleid,
711 : bool bypass_login_check)
712 : {
713 : HeapTuple roleTup;
714 : Form_pg_authid rform;
715 : char *rname;
716 : bool is_superuser;
717 :
718 : /*
719 : * In a parallel worker, we don't have to do anything here.
720 : * ParallelWorkerMain already set our output variables, and we aren't
721 : * going to enforce either rolcanlogin or rolconnlimit. Furthermore, we
722 : * don't really want to perform a catalog lookup for the role: we don't
723 : * want to fail if it's been dropped.
724 : */
725 29776 : if (InitializingParallelWorker)
726 : {
727 : Assert(bypass_login_check);
728 2752 : return;
729 : }
730 :
731 : /*
732 : * Don't do scans if we're bootstrapping, none of the system catalogs
733 : * exist yet, and they should be owned by postgres anyway.
734 : */
735 : Assert(!IsBootstrapProcessingMode());
736 :
737 : /*
738 : * Make sure syscache entries are flushed for recent catalog changes. This
739 : * allows us to find roles that were created on-the-fly during
740 : * authentication.
741 : */
742 27024 : AcceptInvalidationMessages();
743 :
744 : /*
745 : * Look up the role, either by name if that's given or by OID if not.
746 : */
747 27024 : if (rolename != NULL)
748 : {
749 25984 : roleTup = SearchSysCache1(AUTHNAME, PointerGetDatum(rolename));
750 25984 : if (!HeapTupleIsValid(roleTup))
751 4 : ereport(FATAL,
752 : (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
753 : errmsg("role \"%s\" does not exist", rolename)));
754 : }
755 : else
756 : {
757 1040 : roleTup = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid));
758 1040 : if (!HeapTupleIsValid(roleTup))
759 0 : ereport(FATAL,
760 : (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
761 : errmsg("role with OID %u does not exist", roleid)));
762 : }
763 :
764 27020 : rform = (Form_pg_authid) GETSTRUCT(roleTup);
765 27020 : roleid = rform->oid;
766 27020 : rname = NameStr(rform->rolname);
767 27020 : is_superuser = rform->rolsuper;
768 :
769 27020 : SetAuthenticatedUserId(roleid);
770 :
771 : /*
772 : * Set SessionUserId and related variables, including "role", via the GUC
773 : * mechanisms.
774 : *
775 : * Note: ideally we would use PGC_S_DYNAMIC_DEFAULT here, so that
776 : * session_authorization could subsequently be changed from
777 : * pg_db_role_setting entries. Instead, session_authorization in
778 : * pg_db_role_setting has no effect. Changing that would require solving
779 : * two problems:
780 : *
781 : * 1. If pg_db_role_setting has values for both session_authorization and
782 : * role, we could not be sure which order those would be applied in, and
783 : * it would matter.
784 : *
785 : * 2. Sites may have years-old session_authorization entries. There's not
786 : * been any particular reason to remove them. Ending the dormancy of
787 : * those entries could seriously change application behavior, so only a
788 : * major release should do that.
789 : */
790 27020 : SetConfigOption("session_authorization", rname,
791 : PGC_BACKEND, PGC_S_OVERRIDE);
792 :
793 : /*
794 : * These next checks are not enforced when in standalone mode, so that
795 : * there is a way to recover from sillinesses like "UPDATE pg_authid SET
796 : * rolcanlogin = false;".
797 : */
798 27020 : if (IsUnderPostmaster)
799 : {
800 : /*
801 : * Is role allowed to login at all? (But background workers can
802 : * override this by setting bypass_login_check.)
803 : */
804 27020 : if (!bypass_login_check && !rform->rolcanlogin)
805 6 : ereport(FATAL,
806 : (errcode(ERRCODE_INVALID_AUTHORIZATION_SPECIFICATION),
807 : errmsg("role \"%s\" is not permitted to log in",
808 : rname)));
809 :
810 : /*
811 : * Check connection limit for this role. We enforce the limit only
812 : * for regular backends, since other process types have their own
813 : * PGPROC pools.
814 : *
815 : * There is a race condition here --- we create our PGPROC before
816 : * checking for other PGPROCs. If two backends did this at about the
817 : * same time, they might both think they were over the limit, while
818 : * ideally one should succeed and one fail. Getting that to work
819 : * exactly seems more trouble than it is worth, however; instead we
820 : * just document that the connection limit is approximate.
821 : */
822 27014 : if (rform->rolconnlimit >= 0 &&
823 0 : AmRegularBackendProcess() &&
824 0 : !is_superuser &&
825 0 : CountUserBackends(roleid) > rform->rolconnlimit)
826 0 : ereport(FATAL,
827 : (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
828 : errmsg("too many connections for role \"%s\"",
829 : rname)));
830 : }
831 :
832 27014 : ReleaseSysCache(roleTup);
833 : }
834 :
835 :
836 : /*
837 : * Initialize user identity during special backend startup
838 : */
839 : void
840 5366 : InitializeSessionUserIdStandalone(void)
841 : {
842 : /*
843 : * This function should only be called in single-user mode, in autovacuum
844 : * workers, in slot sync worker and in background workers.
845 : */
846 : Assert(!IsUnderPostmaster || AmAutoVacuumWorkerProcess() ||
847 : AmLogicalSlotSyncWorkerProcess() || AmBackgroundWorkerProcess());
848 :
849 : /* call only once */
850 : Assert(!OidIsValid(AuthenticatedUserId));
851 :
852 5366 : AuthenticatedUserId = BOOTSTRAP_SUPERUSERID;
853 :
854 : /*
855 : * XXX Ideally we'd do this via SetConfigOption("session_authorization"),
856 : * but we lack the role name needed to do that, and we can't fetch it
857 : * because one reason for this special case is to be able to start up even
858 : * if something's happened to the BOOTSTRAP_SUPERUSERID's pg_authid row.
859 : * Since we don't set the GUC itself, C code will see the value as NULL,
860 : * and current_setting() will report an empty string within this session.
861 : */
862 5366 : SetSessionAuthorization(BOOTSTRAP_SUPERUSERID, true);
863 :
864 : /* We could do SetConfigOption("role"), but let's be consistent */
865 5366 : SetCurrentRoleId(InvalidOid, false);
866 5366 : }
867 :
868 : /*
869 : * Initialize the system user.
870 : *
871 : * This is built as auth_method:authn_id.
872 : */
873 : void
874 252 : InitializeSystemUser(const char *authn_id, const char *auth_method)
875 : {
876 : char *system_user;
877 :
878 : /* call only once */
879 : Assert(SystemUser == NULL);
880 :
881 : /*
882 : * InitializeSystemUser should be called only when authn_id is not NULL,
883 : * meaning that auth_method is valid.
884 : */
885 : Assert(authn_id != NULL);
886 :
887 252 : system_user = psprintf("%s:%s", auth_method, authn_id);
888 :
889 : /* Store SystemUser in long-lived storage */
890 252 : SystemUser = MemoryContextStrdup(TopMemoryContext, system_user);
891 252 : pfree(system_user);
892 252 : }
893 :
894 : /*
895 : * SQL-function SYSTEM_USER
896 : */
897 : Datum
898 44 : system_user(PG_FUNCTION_ARGS)
899 : {
900 44 : const char *sysuser = GetSystemUser();
901 :
902 44 : if (sysuser)
903 22 : PG_RETURN_DATUM(CStringGetTextDatum(sysuser));
904 : else
905 22 : PG_RETURN_NULL();
906 : }
907 :
908 : /*
909 : * Change session auth ID while running
910 : *
911 : * The SQL standard says that SET SESSION AUTHORIZATION implies SET ROLE NONE.
912 : * We mechanize that at higher levels not here, because this is the GUC
913 : * assign hook for "session_authorization", and it must be commutative with
914 : * SetCurrentRoleId (the hook for "role") because guc.c provides no guarantees
915 : * which will run first during cases such as transaction rollback. Therefore,
916 : * we update derived state (OuterUserId/CurrentUserId/is_superuser) only if
917 : * !SetRoleIsActive.
918 : */
919 : void
920 41404 : SetSessionAuthorization(Oid userid, bool is_superuser)
921 : {
922 41404 : SetSessionUserId(userid, is_superuser);
923 :
924 41404 : if (!SetRoleIsActive)
925 41336 : SetOuterUserId(userid, is_superuser);
926 41404 : }
927 :
928 : /*
929 : * Report current role id
930 : * This follows the semantics of SET ROLE, ie return the outer-level ID
931 : * not the current effective ID, and return InvalidOid when the setting
932 : * is logically SET ROLE NONE.
933 : */
934 : Oid
935 988 : GetCurrentRoleId(void)
936 : {
937 988 : if (SetRoleIsActive)
938 60 : return OuterUserId;
939 : else
940 928 : return InvalidOid;
941 : }
942 :
943 : /*
944 : * Change Role ID while running (SET ROLE)
945 : *
946 : * If roleid is InvalidOid, we are doing SET ROLE NONE: revert to the
947 : * session user authorization. In this case the is_superuser argument
948 : * is ignored.
949 : *
950 : * When roleid is not InvalidOid, the caller must have checked whether
951 : * the session user has permission to become that role. (We cannot check
952 : * here because this routine must be able to execute in a failed transaction
953 : * to restore a prior value of the ROLE GUC variable.)
954 : */
955 : void
956 45394 : SetCurrentRoleId(Oid roleid, bool is_superuser)
957 : {
958 : /*
959 : * Get correct info if it's SET ROLE NONE
960 : *
961 : * If SessionUserId hasn't been set yet, do nothing beyond updating
962 : * SetRoleIsActive --- the eventual SetSessionAuthorization call will
963 : * update the derived state. This is needed since we will get called
964 : * during GUC initialization.
965 : */
966 45394 : if (!OidIsValid(roleid))
967 : {
968 44360 : SetRoleIsActive = false;
969 :
970 44360 : if (!OidIsValid(SessionUserId))
971 2256 : return;
972 :
973 42104 : roleid = SessionUserId;
974 42104 : is_superuser = SessionUserIsSuperuser;
975 : }
976 : else
977 1034 : SetRoleIsActive = true;
978 :
979 43138 : SetOuterUserId(roleid, is_superuser);
980 : }
981 :
982 :
983 : /*
984 : * Get user name from user oid, returns NULL for nonexistent roleid if noerr
985 : * is true.
986 : */
987 : char *
988 21352 : GetUserNameFromId(Oid roleid, bool noerr)
989 : {
990 : HeapTuple tuple;
991 : char *result;
992 :
993 21352 : tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid));
994 21352 : if (!HeapTupleIsValid(tuple))
995 : {
996 18 : if (!noerr)
997 0 : ereport(ERROR,
998 : (errcode(ERRCODE_UNDEFINED_OBJECT),
999 : errmsg("invalid role OID: %u", roleid)));
1000 18 : result = NULL;
1001 : }
1002 : else
1003 : {
1004 21334 : result = pstrdup(NameStr(((Form_pg_authid) GETSTRUCT(tuple))->rolname));
1005 21334 : ReleaseSysCache(tuple);
1006 : }
1007 21352 : return result;
1008 : }
1009 :
1010 : /* ------------------------------------------------------------------------
1011 : * Client connection state shared with parallel workers
1012 : *
1013 : * ClientConnectionInfo contains pieces of information about the client that
1014 : * need to be synced to parallel workers when they initialize.
1015 : *-------------------------------------------------------------------------
1016 : */
1017 :
1018 : ClientConnectionInfo MyClientConnectionInfo;
1019 :
1020 : /*
1021 : * Intermediate representation of ClientConnectionInfo for easier
1022 : * serialization. Variable-length fields are allocated right after this
1023 : * header.
1024 : */
1025 : typedef struct SerializedClientConnectionInfo
1026 : {
1027 : int32 authn_id_len; /* strlen(authn_id), or -1 if NULL */
1028 : UserAuth auth_method;
1029 : } SerializedClientConnectionInfo;
1030 :
1031 : /*
1032 : * Calculate the space needed to serialize MyClientConnectionInfo.
1033 : */
1034 : Size
1035 916 : EstimateClientConnectionInfoSpace(void)
1036 : {
1037 916 : Size size = 0;
1038 :
1039 916 : size = add_size(size, sizeof(SerializedClientConnectionInfo));
1040 :
1041 916 : if (MyClientConnectionInfo.authn_id)
1042 2 : size = add_size(size, strlen(MyClientConnectionInfo.authn_id) + 1);
1043 :
1044 916 : return size;
1045 : }
1046 :
1047 : /*
1048 : * Serialize MyClientConnectionInfo for use by parallel workers.
1049 : */
1050 : void
1051 916 : SerializeClientConnectionInfo(Size maxsize PG_USED_FOR_ASSERTS_ONLY,
1052 : char *start_address)
1053 : {
1054 916 : SerializedClientConnectionInfo serialized = {0};
1055 :
1056 916 : serialized.authn_id_len = -1;
1057 916 : serialized.auth_method = MyClientConnectionInfo.auth_method;
1058 :
1059 916 : if (MyClientConnectionInfo.authn_id)
1060 2 : serialized.authn_id_len = strlen(MyClientConnectionInfo.authn_id);
1061 :
1062 : /* Copy serialized representation to buffer */
1063 : Assert(maxsize >= sizeof(serialized));
1064 916 : memcpy(start_address, &serialized, sizeof(serialized));
1065 :
1066 916 : maxsize -= sizeof(serialized);
1067 916 : start_address += sizeof(serialized);
1068 :
1069 : /* Copy authn_id into the space after the struct */
1070 916 : if (serialized.authn_id_len >= 0)
1071 : {
1072 : Assert(maxsize >= (serialized.authn_id_len + 1));
1073 2 : memcpy(start_address,
1074 2 : MyClientConnectionInfo.authn_id,
1075 : /* include the NULL terminator to ease deserialization */
1076 2 : serialized.authn_id_len + 1);
1077 : }
1078 916 : }
1079 :
1080 : /*
1081 : * Restore MyClientConnectionInfo from its serialized representation.
1082 : */
1083 : void
1084 2752 : RestoreClientConnectionInfo(char *conninfo)
1085 : {
1086 : SerializedClientConnectionInfo serialized;
1087 :
1088 2752 : memcpy(&serialized, conninfo, sizeof(serialized));
1089 :
1090 : /* Copy the fields back into place */
1091 2752 : MyClientConnectionInfo.authn_id = NULL;
1092 2752 : MyClientConnectionInfo.auth_method = serialized.auth_method;
1093 :
1094 2752 : if (serialized.authn_id_len >= 0)
1095 : {
1096 : char *authn_id;
1097 :
1098 4 : authn_id = conninfo + sizeof(serialized);
1099 4 : MyClientConnectionInfo.authn_id = MemoryContextStrdup(TopMemoryContext,
1100 : authn_id);
1101 : }
1102 2752 : }
1103 :
1104 :
1105 : /*-------------------------------------------------------------------------
1106 : * Interlock-file support
1107 : *
1108 : * These routines are used to create both a data-directory lockfile
1109 : * ($DATADIR/postmaster.pid) and Unix-socket-file lockfiles ($SOCKFILE.lock).
1110 : * Both kinds of files contain the same info initially, although we can add
1111 : * more information to a data-directory lockfile after it's created, using
1112 : * AddToDataDirLockFile(). See pidfile.h for documentation of the contents
1113 : * of these lockfiles.
1114 : *
1115 : * On successful lockfile creation, a proc_exit callback to remove the
1116 : * lockfile is automatically created.
1117 : *-------------------------------------------------------------------------
1118 : */
1119 :
1120 : /*
1121 : * proc_exit callback to remove lockfiles.
1122 : */
1123 : static void
1124 2186 : UnlinkLockFiles(int status, Datum arg)
1125 : {
1126 : ListCell *l;
1127 :
1128 6102 : foreach(l, lock_files)
1129 : {
1130 3916 : char *curfile = (char *) lfirst(l);
1131 :
1132 3916 : unlink(curfile);
1133 : /* Should we complain if the unlink fails? */
1134 : }
1135 : /* Since we're about to exit, no need to reclaim storage */
1136 :
1137 : /*
1138 : * Lock file removal should always be the last externally visible action
1139 : * of a postmaster or standalone backend, while we won't come here at all
1140 : * when exiting postmaster child processes. Therefore, this is a good
1141 : * place to log completion of shutdown. We could alternatively teach
1142 : * proc_exit() to do it, but that seems uglier. In a standalone backend,
1143 : * use NOTICE elevel to be less chatty.
1144 : */
1145 2186 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
1146 : (errmsg("database system is shut down")));
1147 2186 : }
1148 :
1149 : /*
1150 : * Create a lockfile.
1151 : *
1152 : * filename is the path name of the lockfile to create.
1153 : * amPostmaster is used to determine how to encode the output PID.
1154 : * socketDir is the Unix socket directory path to include (possibly empty).
1155 : * isDDLock and refName are used to determine what error message to produce.
1156 : */
1157 : static void
1158 3922 : CreateLockFile(const char *filename, bool amPostmaster,
1159 : const char *socketDir,
1160 : bool isDDLock, const char *refName)
1161 : {
1162 : int fd;
1163 : char buffer[MAXPGPATH * 2 + 256];
1164 : int ntries;
1165 : int len;
1166 : int encoded_pid;
1167 : pid_t other_pid;
1168 : pid_t my_pid,
1169 : my_p_pid,
1170 : my_gp_pid;
1171 : const char *envvar;
1172 :
1173 : /*
1174 : * If the PID in the lockfile is our own PID or our parent's or
1175 : * grandparent's PID, then the file must be stale (probably left over from
1176 : * a previous system boot cycle). We need to check this because of the
1177 : * likelihood that a reboot will assign exactly the same PID as we had in
1178 : * the previous reboot, or one that's only one or two counts larger and
1179 : * hence the lockfile's PID now refers to an ancestor shell process. We
1180 : * allow pg_ctl to pass down its parent shell PID (our grandparent PID)
1181 : * via the environment variable PG_GRANDPARENT_PID; this is so that
1182 : * launching the postmaster via pg_ctl can be just as reliable as
1183 : * launching it directly. There is no provision for detecting
1184 : * further-removed ancestor processes, but if the init script is written
1185 : * carefully then all but the immediate parent shell will be root-owned
1186 : * processes and so the kill test will fail with EPERM. Note that we
1187 : * cannot get a false negative this way, because an existing postmaster
1188 : * would surely never launch a competing postmaster or pg_ctl process
1189 : * directly.
1190 : */
1191 3922 : my_pid = getpid();
1192 :
1193 : #ifndef WIN32
1194 3922 : my_p_pid = getppid();
1195 : #else
1196 :
1197 : /*
1198 : * Windows hasn't got getppid(), but doesn't need it since it's not using
1199 : * real kill() either...
1200 : */
1201 : my_p_pid = 0;
1202 : #endif
1203 :
1204 3922 : envvar = getenv("PG_GRANDPARENT_PID");
1205 3922 : if (envvar)
1206 3098 : my_gp_pid = atoi(envvar);
1207 : else
1208 824 : my_gp_pid = 0;
1209 :
1210 : /*
1211 : * We need a loop here because of race conditions. But don't loop forever
1212 : * (for example, a non-writable $PGDATA directory might cause a failure
1213 : * that won't go away). 100 tries seems like plenty.
1214 : */
1215 3922 : for (ntries = 0;; ntries++)
1216 : {
1217 : /*
1218 : * Try to create the lock file --- O_EXCL makes this atomic.
1219 : *
1220 : * Think not to make the file protection weaker than 0600/0640. See
1221 : * comments below.
1222 : */
1223 3928 : fd = open(filename, O_RDWR | O_CREAT | O_EXCL, pg_file_create_mode);
1224 3928 : if (fd >= 0)
1225 3916 : break; /* Success; exit the retry loop */
1226 :
1227 : /*
1228 : * Couldn't create the pid file. Probably it already exists.
1229 : */
1230 12 : if ((errno != EEXIST && errno != EACCES) || ntries > 100)
1231 0 : ereport(FATAL,
1232 : (errcode_for_file_access(),
1233 : errmsg("could not create lock file \"%s\": %m",
1234 : filename)));
1235 :
1236 : /*
1237 : * Read the file to get the old owner's PID. Note race condition
1238 : * here: file might have been deleted since we tried to create it.
1239 : */
1240 12 : fd = open(filename, O_RDONLY, pg_file_create_mode);
1241 12 : if (fd < 0)
1242 : {
1243 0 : if (errno == ENOENT)
1244 0 : continue; /* race condition; try again */
1245 0 : ereport(FATAL,
1246 : (errcode_for_file_access(),
1247 : errmsg("could not open lock file \"%s\": %m",
1248 : filename)));
1249 : }
1250 12 : pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_READ);
1251 12 : if ((len = read(fd, buffer, sizeof(buffer) - 1)) < 0)
1252 0 : ereport(FATAL,
1253 : (errcode_for_file_access(),
1254 : errmsg("could not read lock file \"%s\": %m",
1255 : filename)));
1256 12 : pgstat_report_wait_end();
1257 12 : close(fd);
1258 :
1259 12 : if (len == 0)
1260 : {
1261 0 : ereport(FATAL,
1262 : (errcode(ERRCODE_LOCK_FILE_EXISTS),
1263 : errmsg("lock file \"%s\" is empty", filename),
1264 : errhint("Either another server is starting, or the lock file is the remnant of a previous server startup crash.")));
1265 : }
1266 :
1267 12 : buffer[len] = '\0';
1268 12 : encoded_pid = atoi(buffer);
1269 :
1270 : /* if pid < 0, the pid is for postgres, not postmaster */
1271 12 : other_pid = (pid_t) (encoded_pid < 0 ? -encoded_pid : encoded_pid);
1272 :
1273 12 : if (other_pid <= 0)
1274 0 : elog(FATAL, "bogus data in lock file \"%s\": \"%s\"",
1275 : filename, buffer);
1276 :
1277 : /*
1278 : * Check to see if the other process still exists
1279 : *
1280 : * Per discussion above, my_pid, my_p_pid, and my_gp_pid can be
1281 : * ignored as false matches.
1282 : *
1283 : * Normally kill() will fail with ESRCH if the given PID doesn't
1284 : * exist.
1285 : *
1286 : * We can treat the EPERM-error case as okay because that error
1287 : * implies that the existing process has a different userid than we
1288 : * do, which means it cannot be a competing postmaster. A postmaster
1289 : * cannot successfully attach to a data directory owned by a userid
1290 : * other than its own, as enforced in checkDataDir(). Also, since we
1291 : * create the lockfiles mode 0600/0640, we'd have failed above if the
1292 : * lockfile belonged to another userid --- which means that whatever
1293 : * process kill() is reporting about isn't the one that made the
1294 : * lockfile. (NOTE: this last consideration is the only one that
1295 : * keeps us from blowing away a Unix socket file belonging to an
1296 : * instance of Postgres being run by someone else, at least on
1297 : * machines where /tmp hasn't got a stickybit.)
1298 : */
1299 12 : if (other_pid != my_pid && other_pid != my_p_pid &&
1300 : other_pid != my_gp_pid)
1301 : {
1302 12 : if (kill(other_pid, 0) == 0 ||
1303 8 : (errno != ESRCH && errno != EPERM))
1304 : {
1305 : /* lockfile belongs to a live process */
1306 4 : ereport(FATAL,
1307 : (errcode(ERRCODE_LOCK_FILE_EXISTS),
1308 : errmsg("lock file \"%s\" already exists",
1309 : filename),
1310 : isDDLock ?
1311 : (encoded_pid < 0 ?
1312 : errhint("Is another postgres (PID %d) running in data directory \"%s\"?",
1313 : (int) other_pid, refName) :
1314 : errhint("Is another postmaster (PID %d) running in data directory \"%s\"?",
1315 : (int) other_pid, refName)) :
1316 : (encoded_pid < 0 ?
1317 : errhint("Is another postgres (PID %d) using socket file \"%s\"?",
1318 : (int) other_pid, refName) :
1319 : errhint("Is another postmaster (PID %d) using socket file \"%s\"?",
1320 : (int) other_pid, refName))));
1321 : }
1322 : }
1323 :
1324 : /*
1325 : * No, the creating process did not exist. However, it could be that
1326 : * the postmaster crashed (or more likely was kill -9'd by a clueless
1327 : * admin) but has left orphan backends behind. Check for this by
1328 : * looking to see if there is an associated shmem segment that is
1329 : * still in use.
1330 : *
1331 : * Note: because postmaster.pid is written in multiple steps, we might
1332 : * not find the shmem ID values in it; we can't treat that as an
1333 : * error.
1334 : */
1335 8 : if (isDDLock)
1336 : {
1337 4 : char *ptr = buffer;
1338 : unsigned long id1,
1339 : id2;
1340 : int lineno;
1341 :
1342 28 : for (lineno = 1; lineno < LOCK_FILE_LINE_SHMEM_KEY; lineno++)
1343 : {
1344 24 : if ((ptr = strchr(ptr, '\n')) == NULL)
1345 0 : break;
1346 24 : ptr++;
1347 : }
1348 :
1349 4 : if (ptr != NULL &&
1350 4 : sscanf(ptr, "%lu %lu", &id1, &id2) == 2)
1351 : {
1352 4 : if (PGSharedMemoryIsInUse(id1, id2))
1353 2 : ereport(FATAL,
1354 : (errcode(ERRCODE_LOCK_FILE_EXISTS),
1355 : errmsg("pre-existing shared memory block (key %lu, ID %lu) is still in use",
1356 : id1, id2),
1357 : errhint("Terminate any old server processes associated with data directory \"%s\".",
1358 : refName)));
1359 : }
1360 : }
1361 :
1362 : /*
1363 : * Looks like nobody's home. Unlink the file and try again to create
1364 : * it. Need a loop because of possible race condition against other
1365 : * would-be creators.
1366 : */
1367 6 : if (unlink(filename) < 0)
1368 0 : ereport(FATAL,
1369 : (errcode_for_file_access(),
1370 : errmsg("could not remove old lock file \"%s\": %m",
1371 : filename),
1372 : errhint("The file seems accidentally left over, but "
1373 : "it could not be removed. Please remove the file "
1374 : "by hand and try again.")));
1375 : }
1376 :
1377 : /*
1378 : * Successfully created the file, now fill it. See comment in pidfile.h
1379 : * about the contents. Note that we write the same first five lines into
1380 : * both datadir and socket lockfiles; although more stuff may get added to
1381 : * the datadir lockfile later.
1382 : */
1383 3916 : snprintf(buffer, sizeof(buffer), "%d\n%s\n" INT64_FORMAT "\n%d\n%s\n",
1384 : amPostmaster ? (int) my_pid : -((int) my_pid),
1385 : DataDir,
1386 : MyStartTime,
1387 : PostPortNumber,
1388 : socketDir);
1389 :
1390 : /*
1391 : * In a standalone backend, the next line (LOCK_FILE_LINE_LISTEN_ADDR)
1392 : * will never receive data, so fill it in as empty now.
1393 : */
1394 3916 : if (isDDLock && !amPostmaster)
1395 440 : strlcat(buffer, "\n", sizeof(buffer));
1396 :
1397 3916 : errno = 0;
1398 3916 : pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_WRITE);
1399 3916 : if (write(fd, buffer, strlen(buffer)) != strlen(buffer))
1400 : {
1401 0 : int save_errno = errno;
1402 :
1403 0 : close(fd);
1404 0 : unlink(filename);
1405 : /* if write didn't set errno, assume problem is no disk space */
1406 0 : errno = save_errno ? save_errno : ENOSPC;
1407 0 : ereport(FATAL,
1408 : (errcode_for_file_access(),
1409 : errmsg("could not write lock file \"%s\": %m", filename)));
1410 : }
1411 3916 : pgstat_report_wait_end();
1412 :
1413 3916 : pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_CREATE_SYNC);
1414 3916 : if (pg_fsync(fd) != 0)
1415 : {
1416 0 : int save_errno = errno;
1417 :
1418 0 : close(fd);
1419 0 : unlink(filename);
1420 0 : errno = save_errno;
1421 0 : ereport(FATAL,
1422 : (errcode_for_file_access(),
1423 : errmsg("could not write lock file \"%s\": %m", filename)));
1424 : }
1425 3916 : pgstat_report_wait_end();
1426 3916 : if (close(fd) != 0)
1427 : {
1428 0 : int save_errno = errno;
1429 :
1430 0 : unlink(filename);
1431 0 : errno = save_errno;
1432 0 : ereport(FATAL,
1433 : (errcode_for_file_access(),
1434 : errmsg("could not write lock file \"%s\": %m", filename)));
1435 : }
1436 :
1437 : /*
1438 : * Arrange to unlink the lock file(s) at proc_exit. If this is the first
1439 : * one, set up the on_proc_exit function to do it; then add this lock file
1440 : * to the list of files to unlink.
1441 : */
1442 3916 : if (lock_files == NIL)
1443 2186 : on_proc_exit(UnlinkLockFiles, 0);
1444 :
1445 : /*
1446 : * Use lcons so that the lock files are unlinked in reverse order of
1447 : * creation; this is critical!
1448 : */
1449 3916 : lock_files = lcons(pstrdup(filename), lock_files);
1450 3916 : }
1451 :
1452 : /*
1453 : * Create the data directory lockfile.
1454 : *
1455 : * When this is called, we must have already switched the working
1456 : * directory to DataDir, so we can just use a relative path. This
1457 : * helps ensure that we are locking the directory we should be.
1458 : *
1459 : * Note that the socket directory path line is initially written as empty.
1460 : * postmaster.c will rewrite it upon creating the first Unix socket.
1461 : */
1462 : void
1463 2192 : CreateDataDirLockFile(bool amPostmaster)
1464 : {
1465 2192 : CreateLockFile(DIRECTORY_LOCK_FILE, amPostmaster, "", true, DataDir);
1466 2186 : }
1467 :
1468 : /*
1469 : * Create a lockfile for the specified Unix socket file.
1470 : */
1471 : void
1472 1730 : CreateSocketLockFile(const char *socketfile, bool amPostmaster,
1473 : const char *socketDir)
1474 : {
1475 : char lockfile[MAXPGPATH];
1476 :
1477 1730 : snprintf(lockfile, sizeof(lockfile), "%s.lock", socketfile);
1478 1730 : CreateLockFile(lockfile, amPostmaster, socketDir, false, socketfile);
1479 1730 : }
1480 :
1481 : /*
1482 : * TouchSocketLockFiles -- mark socket lock files as recently accessed
1483 : *
1484 : * This routine should be called every so often to ensure that the socket
1485 : * lock files have a recent mod or access date. That saves them
1486 : * from being removed by overenthusiastic /tmp-directory-cleaner daemons.
1487 : * (Another reason we should never have put the socket file in /tmp...)
1488 : */
1489 : void
1490 0 : TouchSocketLockFiles(void)
1491 : {
1492 : ListCell *l;
1493 :
1494 0 : foreach(l, lock_files)
1495 : {
1496 0 : char *socketLockFile = (char *) lfirst(l);
1497 :
1498 : /* No need to touch the data directory lock file, we trust */
1499 0 : if (strcmp(socketLockFile, DIRECTORY_LOCK_FILE) == 0)
1500 0 : continue;
1501 :
1502 : /* we just ignore any error here */
1503 0 : (void) utime(socketLockFile, NULL);
1504 : }
1505 0 : }
1506 :
1507 :
1508 : /*
1509 : * Add (or replace) a line in the data directory lock file.
1510 : * The given string should not include a trailing newline.
1511 : *
1512 : * Note: because we don't truncate the file, if we were to rewrite a line
1513 : * with less data than it had before, there would be garbage after the last
1514 : * line. While we could fix that by adding a truncate call, that would make
1515 : * the file update non-atomic, which we'd rather avoid. Therefore, callers
1516 : * should endeavor never to shorten a line once it's been written.
1517 : */
1518 : void
1519 10932 : AddToDataDirLockFile(int target_line, const char *str)
1520 : {
1521 : int fd;
1522 : int len;
1523 : int lineno;
1524 : char *srcptr;
1525 : char *destptr;
1526 : char srcbuffer[BLCKSZ];
1527 : char destbuffer[BLCKSZ];
1528 :
1529 10932 : fd = open(DIRECTORY_LOCK_FILE, O_RDWR | PG_BINARY, 0);
1530 10932 : if (fd < 0)
1531 : {
1532 0 : ereport(LOG,
1533 : (errcode_for_file_access(),
1534 : errmsg("could not open file \"%s\": %m",
1535 : DIRECTORY_LOCK_FILE)));
1536 0 : return;
1537 : }
1538 10932 : pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_ADDTODATADIR_READ);
1539 10932 : len = read(fd, srcbuffer, sizeof(srcbuffer) - 1);
1540 10932 : pgstat_report_wait_end();
1541 10932 : if (len < 0)
1542 : {
1543 0 : ereport(LOG,
1544 : (errcode_for_file_access(),
1545 : errmsg("could not read from file \"%s\": %m",
1546 : DIRECTORY_LOCK_FILE)));
1547 0 : close(fd);
1548 0 : return;
1549 : }
1550 10932 : srcbuffer[len] = '\0';
1551 :
1552 : /*
1553 : * Advance over lines we are not supposed to rewrite, then copy them to
1554 : * destbuffer.
1555 : */
1556 10932 : srcptr = srcbuffer;
1557 74890 : for (lineno = 1; lineno < target_line; lineno++)
1558 : {
1559 65690 : char *eol = strchr(srcptr, '\n');
1560 :
1561 65690 : if (eol == NULL)
1562 1732 : break; /* not enough lines in file yet */
1563 63958 : srcptr = eol + 1;
1564 : }
1565 10932 : memcpy(destbuffer, srcbuffer, srcptr - srcbuffer);
1566 10932 : destptr = destbuffer + (srcptr - srcbuffer);
1567 :
1568 : /*
1569 : * Fill in any missing lines before the target line, in case lines are
1570 : * added to the file out of order.
1571 : */
1572 12664 : for (; lineno < target_line; lineno++)
1573 : {
1574 1732 : if (destptr < destbuffer + sizeof(destbuffer))
1575 1732 : *destptr++ = '\n';
1576 : }
1577 :
1578 : /*
1579 : * Write or rewrite the target line.
1580 : */
1581 10932 : snprintf(destptr, destbuffer + sizeof(destbuffer) - destptr, "%s\n", str);
1582 10932 : destptr += strlen(destptr);
1583 :
1584 : /*
1585 : * If there are more lines in the old file, append them to destbuffer.
1586 : */
1587 10932 : if ((srcptr = strchr(srcptr, '\n')) != NULL)
1588 : {
1589 7030 : srcptr++;
1590 7030 : snprintf(destptr, destbuffer + sizeof(destbuffer) - destptr, "%s",
1591 : srcptr);
1592 : }
1593 :
1594 : /*
1595 : * And rewrite the data. Since we write in a single kernel call, this
1596 : * update should appear atomic to onlookers.
1597 : */
1598 10932 : len = strlen(destbuffer);
1599 10932 : errno = 0;
1600 10932 : pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_ADDTODATADIR_WRITE);
1601 10932 : if (pg_pwrite(fd, destbuffer, len, 0) != len)
1602 : {
1603 0 : pgstat_report_wait_end();
1604 : /* if write didn't set errno, assume problem is no disk space */
1605 0 : if (errno == 0)
1606 0 : errno = ENOSPC;
1607 0 : ereport(LOG,
1608 : (errcode_for_file_access(),
1609 : errmsg("could not write to file \"%s\": %m",
1610 : DIRECTORY_LOCK_FILE)));
1611 0 : close(fd);
1612 0 : return;
1613 : }
1614 10932 : pgstat_report_wait_end();
1615 10932 : pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_ADDTODATADIR_SYNC);
1616 10932 : if (pg_fsync(fd) != 0)
1617 : {
1618 0 : ereport(LOG,
1619 : (errcode_for_file_access(),
1620 : errmsg("could not write to file \"%s\": %m",
1621 : DIRECTORY_LOCK_FILE)));
1622 : }
1623 10932 : pgstat_report_wait_end();
1624 10932 : if (close(fd) != 0)
1625 : {
1626 0 : ereport(LOG,
1627 : (errcode_for_file_access(),
1628 : errmsg("could not write to file \"%s\": %m",
1629 : DIRECTORY_LOCK_FILE)));
1630 : }
1631 : }
1632 :
1633 :
1634 : /*
1635 : * Recheck that the data directory lock file still exists with expected
1636 : * content. Return true if the lock file appears OK, false if it isn't.
1637 : *
1638 : * We call this periodically in the postmaster. The idea is that if the
1639 : * lock file has been removed or replaced by another postmaster, we should
1640 : * do a panic database shutdown. Therefore, we should return true if there
1641 : * is any doubt: we do not want to cause a panic shutdown unnecessarily.
1642 : * Transient failures like EINTR or ENFILE should not cause us to fail.
1643 : * (If there really is something wrong, we'll detect it on a future recheck.)
1644 : */
1645 : bool
1646 62 : RecheckDataDirLockFile(void)
1647 : {
1648 : int fd;
1649 : int len;
1650 : long file_pid;
1651 : char buffer[BLCKSZ];
1652 :
1653 62 : fd = open(DIRECTORY_LOCK_FILE, O_RDWR | PG_BINARY, 0);
1654 62 : if (fd < 0)
1655 : {
1656 : /*
1657 : * There are many foreseeable false-positive error conditions. For
1658 : * safety, fail only on enumerated clearly-something-is-wrong
1659 : * conditions.
1660 : */
1661 0 : switch (errno)
1662 : {
1663 0 : case ENOENT:
1664 : case ENOTDIR:
1665 : /* disaster */
1666 0 : ereport(LOG,
1667 : (errcode_for_file_access(),
1668 : errmsg("could not open file \"%s\": %m",
1669 : DIRECTORY_LOCK_FILE)));
1670 0 : return false;
1671 0 : default:
1672 : /* non-fatal, at least for now */
1673 0 : ereport(LOG,
1674 : (errcode_for_file_access(),
1675 : errmsg("could not open file \"%s\": %m; continuing anyway",
1676 : DIRECTORY_LOCK_FILE)));
1677 0 : return true;
1678 : }
1679 : }
1680 62 : pgstat_report_wait_start(WAIT_EVENT_LOCK_FILE_RECHECKDATADIR_READ);
1681 62 : len = read(fd, buffer, sizeof(buffer) - 1);
1682 62 : pgstat_report_wait_end();
1683 62 : if (len < 0)
1684 : {
1685 0 : ereport(LOG,
1686 : (errcode_for_file_access(),
1687 : errmsg("could not read from file \"%s\": %m",
1688 : DIRECTORY_LOCK_FILE)));
1689 0 : close(fd);
1690 0 : return true; /* treat read failure as nonfatal */
1691 : }
1692 62 : buffer[len] = '\0';
1693 62 : close(fd);
1694 62 : file_pid = atol(buffer);
1695 62 : if (file_pid == getpid())
1696 62 : return true; /* all is well */
1697 :
1698 : /* Trouble: someone's overwritten the lock file */
1699 0 : ereport(LOG,
1700 : (errmsg("lock file \"%s\" contains wrong PID: %ld instead of %ld",
1701 : DIRECTORY_LOCK_FILE, file_pid, (long) getpid())));
1702 0 : return false;
1703 : }
1704 :
1705 :
1706 : /*-------------------------------------------------------------------------
1707 : * Version checking support
1708 : *-------------------------------------------------------------------------
1709 : */
1710 :
1711 : /*
1712 : * Determine whether the PG_VERSION file in directory `path' indicates
1713 : * a data version compatible with the version of this program.
1714 : *
1715 : * If compatible, return. Otherwise, ereport(FATAL).
1716 : */
1717 : void
1718 35420 : ValidatePgVersion(const char *path)
1719 : {
1720 : char full_path[MAXPGPATH];
1721 : FILE *file;
1722 : int ret;
1723 : long file_major;
1724 : long my_major;
1725 : char *endptr;
1726 : char file_version_string[64];
1727 35420 : const char *my_version_string = PG_VERSION;
1728 :
1729 35420 : my_major = strtol(my_version_string, &endptr, 10);
1730 :
1731 35420 : snprintf(full_path, sizeof(full_path), "%s/PG_VERSION", path);
1732 :
1733 35420 : file = AllocateFile(full_path, "r");
1734 35420 : if (!file)
1735 : {
1736 0 : if (errno == ENOENT)
1737 0 : ereport(FATAL,
1738 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1739 : errmsg("\"%s\" is not a valid data directory",
1740 : path),
1741 : errdetail("File \"%s\" is missing.", full_path)));
1742 : else
1743 0 : ereport(FATAL,
1744 : (errcode_for_file_access(),
1745 : errmsg("could not open file \"%s\": %m", full_path)));
1746 : }
1747 :
1748 35420 : file_version_string[0] = '\0';
1749 35420 : ret = fscanf(file, "%63s", file_version_string);
1750 35420 : file_major = strtol(file_version_string, &endptr, 10);
1751 :
1752 35420 : if (ret != 1 || endptr == file_version_string)
1753 0 : ereport(FATAL,
1754 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1755 : errmsg("\"%s\" is not a valid data directory",
1756 : path),
1757 : errdetail("File \"%s\" does not contain valid data.",
1758 : full_path),
1759 : errhint("You might need to initdb.")));
1760 :
1761 35420 : FreeFile(file);
1762 :
1763 35420 : if (my_major != file_major)
1764 0 : ereport(FATAL,
1765 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1766 : errmsg("database files are incompatible with server"),
1767 : errdetail("The data directory was initialized by PostgreSQL version %s, "
1768 : "which is not compatible with this version %s.",
1769 : file_version_string, my_version_string)));
1770 35420 : }
1771 :
1772 : /*-------------------------------------------------------------------------
1773 : * Library preload support
1774 : *-------------------------------------------------------------------------
1775 : */
1776 :
1777 : /*
1778 : * GUC variables: lists of library names to be preloaded at postmaster
1779 : * start and at backend start
1780 : */
1781 : char *session_preload_libraries_string = NULL;
1782 : char *shared_preload_libraries_string = NULL;
1783 : char *local_preload_libraries_string = NULL;
1784 :
1785 : /* Flag telling that we are loading shared_preload_libraries */
1786 : bool process_shared_preload_libraries_in_progress = false;
1787 : bool process_shared_preload_libraries_done = false;
1788 :
1789 : shmem_request_hook_type shmem_request_hook = NULL;
1790 : bool process_shmem_requests_in_progress = false;
1791 :
1792 : /*
1793 : * load the shared libraries listed in 'libraries'
1794 : *
1795 : * 'gucname': name of GUC variable, for error reports
1796 : * 'restricted': if true, force libraries to be in $libdir/plugins/
1797 : */
1798 : static void
1799 49422 : load_libraries(const char *libraries, const char *gucname, bool restricted)
1800 : {
1801 : char *rawstring;
1802 : List *elemlist;
1803 : ListCell *l;
1804 :
1805 49422 : if (libraries == NULL || libraries[0] == '\0')
1806 49342 : return; /* nothing to do */
1807 :
1808 : /* Need a modifiable copy of string */
1809 80 : rawstring = pstrdup(libraries);
1810 :
1811 : /* Parse string into list of filename paths */
1812 80 : if (!SplitDirectoriesString(rawstring, ',', &elemlist))
1813 : {
1814 : /* syntax error in list */
1815 0 : list_free_deep(elemlist);
1816 0 : pfree(rawstring);
1817 0 : ereport(LOG,
1818 : (errcode(ERRCODE_SYNTAX_ERROR),
1819 : errmsg("invalid list syntax in parameter \"%s\"",
1820 : gucname)));
1821 0 : return;
1822 : }
1823 :
1824 162 : foreach(l, elemlist)
1825 : {
1826 : /* Note that filename was already canonicalized */
1827 82 : char *filename = (char *) lfirst(l);
1828 82 : char *expanded = NULL;
1829 :
1830 : /* If restricting, insert $libdir/plugins if not mentioned already */
1831 82 : if (restricted && first_dir_separator(filename) == NULL)
1832 : {
1833 0 : expanded = psprintf("$libdir/plugins/%s", filename);
1834 0 : filename = expanded;
1835 : }
1836 82 : load_file(filename, restricted);
1837 82 : ereport(DEBUG1,
1838 : (errmsg_internal("loaded library \"%s\"", filename)));
1839 82 : if (expanded)
1840 0 : pfree(expanded);
1841 : }
1842 :
1843 80 : list_free_deep(elemlist);
1844 80 : pfree(rawstring);
1845 : }
1846 :
1847 : /*
1848 : * process any libraries that should be preloaded at postmaster start
1849 : */
1850 : void
1851 1886 : process_shared_preload_libraries(void)
1852 : {
1853 1886 : process_shared_preload_libraries_in_progress = true;
1854 1886 : load_libraries(shared_preload_libraries_string,
1855 : "shared_preload_libraries",
1856 : false);
1857 1886 : process_shared_preload_libraries_in_progress = false;
1858 1886 : process_shared_preload_libraries_done = true;
1859 1886 : }
1860 :
1861 : /*
1862 : * process any libraries that should be preloaded at backend start
1863 : */
1864 : void
1865 23768 : process_session_preload_libraries(void)
1866 : {
1867 23768 : load_libraries(session_preload_libraries_string,
1868 : "session_preload_libraries",
1869 : false);
1870 23768 : load_libraries(local_preload_libraries_string,
1871 : "local_preload_libraries",
1872 : true);
1873 23768 : }
1874 :
1875 : /*
1876 : * process any shared memory requests from preloaded libraries
1877 : */
1878 : void
1879 1876 : process_shmem_requests(void)
1880 : {
1881 1876 : process_shmem_requests_in_progress = true;
1882 1876 : if (shmem_request_hook)
1883 32 : shmem_request_hook();
1884 1876 : process_shmem_requests_in_progress = false;
1885 1876 : }
1886 :
1887 : void
1888 3650 : pg_bindtextdomain(const char *domain)
1889 : {
1890 : #ifdef ENABLE_NLS
1891 3650 : if (my_exec_path[0] != '\0')
1892 : {
1893 : char locale_path[MAXPGPATH];
1894 :
1895 3650 : get_locale_path(my_exec_path, locale_path);
1896 3650 : bindtextdomain(domain, locale_path);
1897 3650 : pg_bind_textdomain_codeset(domain);
1898 : }
1899 : #endif
1900 3650 : }
|