Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * fd.c
4 : : * Virtual file descriptor code.
5 : : *
6 : : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : * IDENTIFICATION
10 : : * src/backend/storage/file/fd.c
11 : : *
12 : : * NOTES:
13 : : *
14 : : * This code manages a cache of 'virtual' file descriptors (VFDs).
15 : : * The server opens many file descriptors for a variety of reasons,
16 : : * including base tables, scratch files (e.g., sort and hash spool
17 : : * files), and random calls to C library routines like system(3); it
18 : : * is quite easy to exceed system limits on the number of open files a
19 : : * single process can have. (This is around 1024 on many modern
20 : : * operating systems, but may be lower on others.)
21 : : *
22 : : * VFDs are managed as an LRU pool, with actual OS file descriptors
23 : : * being opened and closed as needed. Obviously, if a routine is
24 : : * opened using these interfaces, all subsequent operations must also
25 : : * be through these interfaces (the File type is not a real file
26 : : * descriptor).
27 : : *
28 : : * For this scheme to work, most (if not all) routines throughout the
29 : : * server should use these interfaces instead of calling the C library
30 : : * routines (e.g., open(2) and fopen(3)) themselves. Otherwise, we
31 : : * may find ourselves short of real file descriptors anyway.
32 : : *
33 : : * INTERFACE ROUTINES
34 : : *
35 : : * PathNameOpenFile and OpenTemporaryFile are used to open virtual files.
36 : : * A File opened with OpenTemporaryFile is automatically deleted when the
37 : : * File is closed, either explicitly or implicitly at end of transaction or
38 : : * process exit. PathNameOpenFile is intended for files that are held open
39 : : * for a long time, like relation files. It is the caller's responsibility
40 : : * to close them, there is no automatic mechanism in fd.c for that.
41 : : *
42 : : * PathName(Create|Open|Delete)Temporary(File|Dir) are used to manage
43 : : * temporary files that have names so that they can be shared between
44 : : * backends. Such files are automatically closed and count against the
45 : : * temporary file limit of the backend that creates them, but unlike anonymous
46 : : * files they are not automatically deleted. See sharedfileset.c for a shared
47 : : * ownership mechanism that provides automatic cleanup for shared files when
48 : : * the last of a group of backends detaches.
49 : : *
50 : : * AllocateFile, AllocateDir, OpenPipeStream and OpenTransientFile are
51 : : * wrappers around fopen(3), opendir(3), popen(3) and open(2), respectively.
52 : : * They behave like the corresponding native functions, except that the handle
53 : : * is registered with the current subtransaction, and will be automatically
54 : : * closed at abort. These are intended mainly for short operations like
55 : : * reading a configuration file; there is a limit on the number of files that
56 : : * can be opened using these functions at any one time.
57 : : *
58 : : * Finally, BasicOpenFile is just a thin wrapper around open() that can
59 : : * release file descriptors in use by the virtual file descriptors if
60 : : * necessary. There is no automatic cleanup of file descriptors returned by
61 : : * BasicOpenFile, it is solely the caller's responsibility to close the file
62 : : * descriptor by calling close(2).
63 : : *
64 : : * If a non-virtual file descriptor needs to be held open for any length of
65 : : * time, report it to fd.c by calling AcquireExternalFD or ReserveExternalFD
66 : : * (and eventually ReleaseExternalFD), so that we can take it into account
67 : : * while deciding how many VFDs can be open. This applies to FDs obtained
68 : : * with BasicOpenFile as well as those obtained without use of any fd.c API.
69 : : *
70 : : *-------------------------------------------------------------------------
71 : : */
72 : :
73 : : #include "postgres.h"
74 : :
75 : : #include <dirent.h>
76 : : #include <sys/file.h>
77 : : #include <sys/param.h>
78 : : #include <sys/resource.h> /* for getrlimit */
79 : : #include <sys/stat.h>
80 : : #include <sys/types.h>
81 : : #ifndef WIN32
82 : : #include <sys/mman.h>
83 : : #endif
84 : : #include <limits.h>
85 : : #include <unistd.h>
86 : : #include <fcntl.h>
87 : :
88 : : #include "access/xact.h"
89 : : #include "access/xlog.h"
90 : : #include "catalog/pg_tablespace.h"
91 : : #include "common/file_perm.h"
92 : : #include "common/file_utils.h"
93 : : #include "common/pg_prng.h"
94 : : #include "miscadmin.h"
95 : : #include "pgstat.h"
96 : : #include "portability/mem.h"
97 : : #include "postmaster/startup.h"
98 : : #include "storage/fd.h"
99 : : #include "storage/ipc.h"
100 : : #include "utils/guc.h"
101 : : #include "utils/guc_hooks.h"
102 : : #include "utils/resowner.h"
103 : : #include "utils/varlena.h"
104 : :
105 : : /* Define PG_FLUSH_DATA_WORKS if we have an implementation for pg_flush_data */
106 : : #if defined(HAVE_SYNC_FILE_RANGE)
107 : : #define PG_FLUSH_DATA_WORKS 1
108 : : #elif !defined(WIN32) && defined(MS_ASYNC)
109 : : #define PG_FLUSH_DATA_WORKS 1
110 : : #elif defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
111 : : #define PG_FLUSH_DATA_WORKS 1
112 : : #endif
113 : :
114 : : /*
115 : : * We must leave some file descriptors free for system(), the dynamic loader,
116 : : * and other code that tries to open files without consulting fd.c. This
117 : : * is the number left free. (While we try fairly hard to prevent EMFILE
118 : : * errors, there's never any guarantee that we won't get ENFILE due to
119 : : * other processes chewing up FDs. So it's a bad idea to try to open files
120 : : * without consulting fd.c. Nonetheless we cannot control all code.)
121 : : *
122 : : * Because this is just a fixed setting, we are effectively assuming that
123 : : * no such code will leave FDs open over the long term; otherwise the slop
124 : : * is likely to be insufficient. Note in particular that we expect that
125 : : * loading a shared library does not result in any permanent increase in
126 : : * the number of open files. (This appears to be true on most if not
127 : : * all platforms as of Feb 2004.)
128 : : */
129 : : #define NUM_RESERVED_FDS 10
130 : :
131 : : /*
132 : : * If we have fewer than this many usable FDs after allowing for the reserved
133 : : * ones, choke. (This value is chosen to work with "ulimit -n 64", but not
134 : : * much less than that. Note that this value ensures numExternalFDs can be
135 : : * at least 16; as of this writing, the contrib/postgres_fdw regression tests
136 : : * will not pass unless that can grow to at least 14.)
137 : : */
138 : : #define FD_MINFREE 48
139 : :
140 : : /*
141 : : * A number of platforms allow individual processes to open many more files
142 : : * than they can really support when *many* processes do the same thing.
143 : : * This GUC parameter lets the DBA limit max_safe_fds to something less than
144 : : * what the postmaster's initial probe suggests will work.
145 : : */
146 : : int max_files_per_process = 1000;
147 : :
148 : : /*
149 : : * Maximum number of file descriptors to open for operations that fd.c knows
150 : : * about (VFDs, AllocateFile etc, or "external" FDs). This is initialized
151 : : * to a conservative value, and remains that way indefinitely in bootstrap or
152 : : * standalone-backend cases. In normal postmaster operation, the postmaster
153 : : * calls set_max_safe_fds() late in initialization to update the value, and
154 : : * that value is then inherited by forked subprocesses.
155 : : *
156 : : * Note: the value of max_files_per_process is taken into account while
157 : : * setting this variable, and so need not be tested separately.
158 : : */
159 : : int max_safe_fds = FD_MINFREE; /* default if not changed */
160 : :
161 : : /* Whether it is safe to continue running after fsync() fails. */
162 : : bool data_sync_retry = false;
163 : :
164 : : /* How SyncDataDirectory() should do its job. */
165 : : int recovery_init_sync_method = DATA_DIR_SYNC_METHOD_FSYNC;
166 : :
167 : : /* Which kinds of files should be opened with PG_O_DIRECT. */
168 : : int io_direct_flags;
169 : :
170 : : /* Debugging.... */
171 : :
172 : : #ifdef FDDEBUG
173 : : #define DO_DB(A) \
174 : : do { \
175 : : int _do_db_save_errno = errno; \
176 : : A; \
177 : : errno = _do_db_save_errno; \
178 : : } while (0)
179 : : #else
180 : : #define DO_DB(A) \
181 : : ((void) 0)
182 : : #endif
183 : :
184 : : #define VFD_CLOSED (-1)
185 : :
186 : : #define FileIsValid(file) \
187 : : ((file) > 0 && (file) < (int) SizeVfdCache && VfdCache[file].fileName != NULL)
188 : :
189 : : #define FileIsNotOpen(file) (VfdCache[file].fd == VFD_CLOSED)
190 : :
191 : : /* these are the assigned bits in fdstate below: */
192 : : #define FD_DELETE_AT_CLOSE (1 << 0) /* T = delete when closed */
193 : : #define FD_CLOSE_AT_EOXACT (1 << 1) /* T = close at eoXact */
194 : : #define FD_TEMP_FILE_LIMIT (1 << 2) /* T = respect temp_file_limit */
195 : :
196 : : typedef struct vfd
197 : : {
198 : : int fd; /* current FD, or VFD_CLOSED if none */
199 : : unsigned short fdstate; /* bitflags for VFD's state */
200 : : ResourceOwner resowner; /* owner, for automatic cleanup */
201 : : File nextFree; /* link to next free VFD, if in freelist */
202 : : File lruMoreRecently; /* doubly linked recency-of-use list */
203 : : File lruLessRecently;
204 : : off_t fileSize; /* current size of file (0 if not temporary) */
205 : : char *fileName; /* name of file, or NULL for unused VFD */
206 : : /* NB: fileName is malloc'd, and must be free'd when closing the VFD */
207 : : int fileFlags; /* open(2) flags for (re)opening the file */
208 : : mode_t fileMode; /* mode to pass to open(2) */
209 : : } Vfd;
210 : :
211 : : /*
212 : : * Virtual File Descriptor array pointer and size. This grows as
213 : : * needed. 'File' values are indexes into this array.
214 : : * Note that VfdCache[0] is not a usable VFD, just a list header.
215 : : */
216 : : static Vfd *VfdCache;
217 : : static Size SizeVfdCache = 0;
218 : :
219 : : /*
220 : : * Number of file descriptors known to be in use by VFD entries.
221 : : */
222 : : static int nfile = 0;
223 : :
224 : : /*
225 : : * Flag to tell whether it's worth scanning VfdCache looking for temp files
226 : : * to close
227 : : */
228 : : static bool have_xact_temporary_files = false;
229 : :
230 : : /*
231 : : * Tracks the total size of all temporary files. Note: when temp_file_limit
232 : : * is being enforced, this cannot overflow since the limit cannot be more
233 : : * than INT_MAX kilobytes. When not enforcing, it could theoretically
234 : : * overflow, but we don't care.
235 : : */
236 : : static uint64 temporary_files_size = 0;
237 : :
238 : : /* Temporary file access initialized and not yet shut down? */
239 : : #ifdef USE_ASSERT_CHECKING
240 : : static bool temporary_files_allowed = false;
241 : : #endif
242 : :
243 : : /*
244 : : * List of OS handles opened with AllocateFile, AllocateDir and
245 : : * OpenTransientFile.
246 : : */
247 : : typedef enum
248 : : {
249 : : AllocateDescFile,
250 : : AllocateDescPipe,
251 : : AllocateDescDir,
252 : : AllocateDescRawFD,
253 : : } AllocateDescKind;
254 : :
255 : : typedef struct
256 : : {
257 : : AllocateDescKind kind;
258 : : SubTransactionId create_subid;
259 : : union
260 : : {
261 : : FILE *file;
262 : : DIR *dir;
263 : : int fd;
264 : : } desc;
265 : : } AllocateDesc;
266 : :
267 : : static int numAllocatedDescs = 0;
268 : : static int maxAllocatedDescs = 0;
269 : : static AllocateDesc *allocatedDescs = NULL;
270 : :
271 : : /*
272 : : * Number of open "external" FDs reported to Reserve/ReleaseExternalFD.
273 : : */
274 : : static int numExternalFDs = 0;
275 : :
276 : : /*
277 : : * Number of temporary files opened during the current session;
278 : : * this is used in generation of tempfile names.
279 : : */
280 : : static long tempFileCounter = 0;
281 : :
282 : : /*
283 : : * Array of OIDs of temp tablespaces. (Some entries may be InvalidOid,
284 : : * indicating that the current database's default tablespace should be used.)
285 : : * When numTempTableSpaces is -1, this has not been set in the current
286 : : * transaction.
287 : : */
288 : : static Oid *tempTableSpaces = NULL;
289 : : static int numTempTableSpaces = -1;
290 : : static int nextTempTableSpace = 0;
291 : :
292 : :
293 : : /*--------------------
294 : : *
295 : : * Private Routines
296 : : *
297 : : * Delete - delete a file from the Lru ring
298 : : * LruDelete - remove a file from the Lru ring and close its FD
299 : : * Insert - put a file at the front of the Lru ring
300 : : * LruInsert - put a file at the front of the Lru ring and open it
301 : : * ReleaseLruFile - Release an fd by closing the last entry in the Lru ring
302 : : * ReleaseLruFiles - Release fd(s) until we're under the max_safe_fds limit
303 : : * AllocateVfd - grab a free (or new) file record (from VfdCache)
304 : : * FreeVfd - free a file record
305 : : *
306 : : * The Least Recently Used ring is a doubly linked list that begins and
307 : : * ends on element zero. Element zero is special -- it doesn't represent
308 : : * a file and its "fd" field always == VFD_CLOSED. Element zero is just an
309 : : * anchor that shows us the beginning/end of the ring.
310 : : * Only VFD elements that are currently really open (have an FD assigned) are
311 : : * in the Lru ring. Elements that are "virtually" open can be recognized
312 : : * by having a non-null fileName field.
313 : : *
314 : : * example:
315 : : *
316 : : * /--less----\ /---------\
317 : : * v \ v \
318 : : * #0 --more---> LeastRecentlyUsed --more-\ \
319 : : * ^\ | |
320 : : * \\less--> MostRecentlyUsedFile <---/ |
321 : : * \more---/ \--less--/
322 : : *
323 : : *--------------------
324 : : */
325 : : static void Delete(File file);
326 : : static void LruDelete(File file);
327 : : static void Insert(File file);
328 : : static int LruInsert(File file);
329 : : static bool ReleaseLruFile(void);
330 : : static void ReleaseLruFiles(void);
331 : : static File AllocateVfd(void);
332 : : static void FreeVfd(File file);
333 : :
334 : : static int FileAccess(File file);
335 : : static File OpenTemporaryFileInTablespace(Oid tblspcOid, bool rejectError);
336 : : static bool reserveAllocatedDesc(void);
337 : : static int FreeDesc(AllocateDesc *desc);
338 : :
339 : : static void BeforeShmemExit_Files(int code, Datum arg);
340 : : static void CleanupTempFiles(bool isCommit, bool isProcExit);
341 : : static void RemovePgTempRelationFiles(const char *tsdirname);
342 : : static void RemovePgTempRelationFilesInDbspace(const char *dbspacedirname);
343 : :
344 : : static void walkdir(const char *path,
345 : : void (*action) (const char *fname, bool isdir, int elevel),
346 : : bool process_symlinks,
347 : : int elevel);
348 : : #ifdef PG_FLUSH_DATA_WORKS
349 : : static void pre_sync_fname(const char *fname, bool isdir, int elevel);
350 : : #endif
351 : : static void datadir_fsync_fname(const char *fname, bool isdir, int elevel);
352 : : static void unlink_if_exists_fname(const char *fname, bool isdir, int elevel);
353 : :
354 : : static int fsync_parent_path(const char *fname, int elevel);
355 : :
356 : :
357 : : /* ResourceOwner callbacks to hold virtual file descriptors */
358 : : static void ResOwnerReleaseFile(Datum res);
359 : : static char *ResOwnerPrintFile(Datum res);
360 : :
361 : : static const ResourceOwnerDesc file_resowner_desc =
362 : : {
363 : : .name = "File",
364 : : .release_phase = RESOURCE_RELEASE_AFTER_LOCKS,
365 : : .release_priority = RELEASE_PRIO_FILES,
366 : : .ReleaseResource = ResOwnerReleaseFile,
367 : : .DebugPrint = ResOwnerPrintFile
368 : : };
369 : :
370 : : /* Convenience wrappers over ResourceOwnerRemember/Forget */
371 : : static inline void
158 heikki.linnakangas@i 372 :GNC 6135 : ResourceOwnerRememberFile(ResourceOwner owner, File file)
373 : : {
374 : 6135 : ResourceOwnerRemember(owner, Int32GetDatum(file), &file_resowner_desc);
375 : 6135 : }
376 : : static inline void
377 : 6131 : ResourceOwnerForgetFile(ResourceOwner owner, File file)
378 : : {
379 : 6131 : ResourceOwnerForget(owner, Int32GetDatum(file), &file_resowner_desc);
380 : 6131 : }
381 : :
382 : : /*
383 : : * pg_fsync --- do fsync with or without writethrough
384 : : */
385 : : int
8528 tgl@sss.pgh.pa.us 386 :CBC 47682 : pg_fsync(int fd)
387 : : {
388 : : #if !defined(WIN32) && defined(USE_ASSERT_CHECKING)
389 : : struct stat st;
390 : :
391 : : /*
392 : : * Some operating system implementations of fsync() have requirements
393 : : * about the file access modes that were used when their file descriptor
394 : : * argument was opened, and these requirements differ depending on whether
395 : : * the file descriptor is for a directory.
396 : : *
397 : : * For any file descriptor that may eventually be handed to fsync(), we
398 : : * should have opened it with access modes that are compatible with
399 : : * fsync() on all supported systems, otherwise the code may not be
400 : : * portable, even if it runs ok on the current system.
401 : : *
402 : : * We assert here that a descriptor for a file was opened with write
403 : : * permissions (either O_RDWR or O_WRONLY) and for a directory without
404 : : * write permissions (O_RDONLY).
405 : : *
406 : : * Ignore any fstat errors and let the follow-up fsync() do its work.
407 : : * Doing this sanity check here counts for the case where fsync() is
408 : : * disabled.
409 : : */
1601 michael@paquier.xyz 410 [ + - ]: 47682 : if (fstat(fd, &st) == 0)
411 : : {
412 : 47682 : int desc_flags = fcntl(fd, F_GETFL);
413 : :
414 : : /*
415 : : * O_RDONLY is historically 0, so just make sure that for directories
416 : : * no write flags are used.
417 : : */
418 [ + + ]: 47682 : if (S_ISDIR(st.st_mode))
419 [ - + ]: 17319 : Assert((desc_flags & (O_RDWR | O_WRONLY)) == 0);
420 : : else
421 [ - + ]: 30363 : Assert((desc_flags & (O_RDWR | O_WRONLY)) != 0);
422 : : }
423 : 47682 : errno = 0;
424 : : #endif
425 : :
426 : : /* #if is to skip the wal_sync_method test if there's no need for it */
427 : : #if defined(HAVE_FSYNC_WRITETHROUGH)
428 : : if (wal_sync_method == WAL_SYNC_METHOD_FSYNC_WRITETHROUGH)
429 : : return pg_fsync_writethrough(fd);
430 : : else
431 : : #endif
4876 tgl@sss.pgh.pa.us 432 : 47682 : return pg_fsync_no_writethrough(fd);
433 : : }
434 : :
435 : :
436 : : /*
437 : : * pg_fsync_no_writethrough --- same as fsync except does nothing if
438 : : * enableFsync is off
439 : : */
440 : : int
6904 bruce@momjian.us 441 : 47682 : pg_fsync_no_writethrough(int fd)
442 : : {
443 : : int rc;
444 : :
300 andres@anarazel.de 445 [ - + ]: 47682 : if (!enableFsync)
8528 tgl@sss.pgh.pa.us 446 : 47682 : return 0;
447 : :
300 andres@anarazel.de 448 :UBC 0 : retry:
449 : 0 : rc = fsync(fd);
450 : :
451 [ # # # # ]: 0 : if (rc == -1 && errno == EINTR)
452 : 0 : goto retry;
453 : :
454 : 0 : return rc;
455 : : }
456 : :
457 : : /*
458 : : * pg_fsync_writethrough
459 : : */
460 : : int
6904 bruce@momjian.us 461 : 0 : pg_fsync_writethrough(int fd)
462 : : {
463 [ # # ]: 0 : if (enableFsync)
464 : : {
465 : : #if defined(F_FULLFSYNC)
466 : : return (fcntl(fd, F_FULLFSYNC, 0) == -1) ? -1 : 0;
467 : : #else
5165 tgl@sss.pgh.pa.us 468 : 0 : errno = ENOSYS;
6904 bruce@momjian.us 469 : 0 : return -1;
470 : : #endif
471 : : }
472 : : else
473 : 0 : return 0;
474 : : }
475 : :
476 : : /*
477 : : * pg_fdatasync --- same as fdatasync except does nothing if enableFsync is off
478 : : */
479 : : int
8456 tgl@sss.pgh.pa.us 480 : 0 : pg_fdatasync(int fd)
481 : : {
482 : : int rc;
483 : :
300 andres@anarazel.de 484 [ # # ]: 0 : if (!enableFsync)
8456 tgl@sss.pgh.pa.us 485 : 0 : return 0;
486 : :
300 andres@anarazel.de 487 : 0 : retry:
488 : 0 : rc = fdatasync(fd);
489 : :
490 [ # # # # ]: 0 : if (rc == -1 && errno == EINTR)
491 : 0 : goto retry;
492 : :
493 : 0 : return rc;
494 : : }
495 : :
496 : : /*
497 : : * pg_file_exists -- check that a file exists.
498 : : *
499 : : * This requires an absolute path to the file. Returns true if the file is
500 : : * not a directory, false otherwise.
501 : : */
502 : : bool
93 michael@paquier.xyz 503 :GNC 16476 : pg_file_exists(const char *name)
504 : : {
505 : : struct stat st;
506 : :
507 [ - + ]: 16476 : Assert(name != NULL);
508 : :
509 [ + + ]: 16476 : if (stat(name, &st) == 0)
510 : 8138 : return !S_ISDIR(st.st_mode);
511 [ - + - - : 8338 : else if (!(errno == ENOENT || errno == ENOTDIR || errno == EACCES))
- - ]
93 michael@paquier.xyz 512 [ # # ]:UNC 0 : ereport(ERROR,
513 : : (errcode_for_file_access(),
514 : : errmsg("could not access file \"%s\": %m", name)));
515 : :
93 michael@paquier.xyz 516 :GNC 8338 : return false;
517 : : }
518 : :
519 : : /*
520 : : * pg_flush_data --- advise OS that the described dirty data should be flushed
521 : : *
522 : : * offset of 0 with nbytes 0 means that the entire file should be flushed
523 : : */
524 : : void
2977 andres@anarazel.de 525 :CBC 110713 : pg_flush_data(int fd, off_t offset, off_t nbytes)
526 : : {
527 : : /*
528 : : * Right now file flushing is primarily used to avoid making later
529 : : * fsync()/fdatasync() calls have less impact. Thus don't trigger flushes
530 : : * if fsyncs are disabled - that's a decision we might want to make
531 : : * configurable at some point.
532 : : */
533 [ + - ]: 110713 : if (!enableFsync)
534 : 110713 : return;
535 : :
536 : : /*
537 : : * We compile all alternatives that are supported on the current platform,
538 : : * to find portability problems more easily.
539 : : */
540 : : #if defined(HAVE_SYNC_FILE_RANGE)
541 : : {
542 : : int rc;
543 : : static bool not_implemented_by_kernel = false;
544 : :
1876 tmunro@postgresql.or 545 [ # # ]:UBC 0 : if (not_implemented_by_kernel)
546 : 0 : return;
547 : :
300 andres@anarazel.de 548 : 0 : retry:
549 : :
550 : : /*
551 : : * sync_file_range(SYNC_FILE_RANGE_WRITE), currently linux specific,
552 : : * tells the OS that writeback for the specified blocks should be
553 : : * started, but that we don't want to wait for completion. Note that
554 : : * this call might block if too much dirty data exists in the range.
555 : : * This is the preferable method on OSs supporting it, as it works
556 : : * reliably when available (contrast to msync()) and doesn't flush out
557 : : * clean data (like FADV_DONTNEED).
558 : : */
2977 559 : 0 : rc = sync_file_range(fd, offset, nbytes,
560 : : SYNC_FILE_RANGE_WRITE);
561 [ # # ]: 0 : if (rc != 0)
562 : : {
563 : : int elevel;
564 : :
300 565 [ # # ]: 0 : if (rc == EINTR)
566 : 0 : goto retry;
567 : :
568 : : /*
569 : : * For systems that don't have an implementation of
570 : : * sync_file_range() such as Windows WSL, generate only one
571 : : * warning and then suppress all further attempts by this process.
572 : : */
1876 tmunro@postgresql.or 573 [ # # ]: 0 : if (errno == ENOSYS)
574 : : {
575 : 0 : elevel = WARNING;
576 : 0 : not_implemented_by_kernel = true;
577 : : }
578 : : else
579 : 0 : elevel = data_sync_elevel(WARNING);
580 : :
581 [ # # ]: 0 : ereport(elevel,
582 : : (errcode_for_file_access(),
583 : : errmsg("could not flush dirty data: %m")));
584 : : }
585 : :
2977 andres@anarazel.de 586 : 0 : return;
587 : : }
588 : : #endif
589 : : #if !defined(WIN32) && defined(MS_ASYNC)
590 : : {
591 : : void *p;
592 : : static int pagesize = 0;
593 : :
594 : : /*
595 : : * On several OSs msync(MS_ASYNC) on a mmap'ed file triggers
596 : : * writeback. On linux it only does so if MS_SYNC is specified, but
597 : : * then it does the writeback synchronously. Luckily all common linux
598 : : * systems have sync_file_range(). This is preferable over
599 : : * FADV_DONTNEED because it doesn't flush out clean data.
600 : : *
601 : : * We map the file (mmap()), tell the kernel to sync back the contents
602 : : * (msync()), and then remove the mapping again (munmap()).
603 : : */
604 : :
605 : : /* mmap() needs actual length if we want to map whole file */
606 : : if (offset == 0 && nbytes == 0)
607 : : {
608 : : nbytes = lseek(fd, 0, SEEK_END);
609 : : if (nbytes < 0)
610 : : {
611 : : ereport(WARNING,
612 : : (errcode_for_file_access(),
613 : : errmsg("could not determine dirty data size: %m")));
614 : : return;
615 : : }
616 : : }
617 : :
618 : : /*
619 : : * Some platforms reject partial-page mmap() attempts. To deal with
620 : : * that, just truncate the request to a page boundary. If any extra
621 : : * bytes don't get flushed, well, it's only a hint anyway.
622 : : */
623 : :
624 : : /* fetch pagesize only once */
625 : : if (pagesize == 0)
626 : : pagesize = sysconf(_SC_PAGESIZE);
627 : :
628 : : /* align length to pagesize, dropping any fractional page */
629 : : if (pagesize > 0)
630 : : nbytes = (nbytes / pagesize) * pagesize;
631 : :
632 : : /* fractional-page request is a no-op */
633 : : if (nbytes <= 0)
634 : : return;
635 : :
636 : : /*
637 : : * mmap could well fail, particularly on 32-bit platforms where there
638 : : * may simply not be enough address space. If so, silently fall
639 : : * through to the next implementation.
640 : : */
641 : : if (nbytes <= (off_t) SSIZE_MAX)
642 : : p = mmap(NULL, nbytes, PROT_READ, MAP_SHARED, fd, offset);
643 : : else
644 : : p = MAP_FAILED;
645 : :
646 : : if (p != MAP_FAILED)
647 : : {
648 : : int rc;
649 : :
650 : : rc = msync(p, (size_t) nbytes, MS_ASYNC);
651 : : if (rc != 0)
652 : : {
653 : : ereport(data_sync_elevel(WARNING),
654 : : (errcode_for_file_access(),
655 : : errmsg("could not flush dirty data: %m")));
656 : : /* NB: need to fall through to munmap()! */
657 : : }
658 : :
659 : : rc = munmap(p, (size_t) nbytes);
660 : : if (rc != 0)
661 : : {
662 : : /* FATAL error because mapping would remain */
663 : : ereport(FATAL,
664 : : (errcode_for_file_access(),
665 : : errmsg("could not munmap() while flushing data: %m")));
666 : : }
667 : :
668 : : return;
669 : : }
670 : : }
671 : : #endif
672 : : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
673 : : {
674 : : int rc;
675 : :
676 : : /*
677 : : * Signal the kernel that the passed in range should not be cached
678 : : * anymore. This has the, desired, side effect of writing out dirty
679 : : * data, and the, undesired, side effect of likely discarding useful
680 : : * clean cached blocks. For the latter reason this is the least
681 : : * preferable method.
682 : : */
683 : :
684 : : rc = posix_fadvise(fd, offset, nbytes, POSIX_FADV_DONTNEED);
685 : :
686 : : if (rc != 0)
687 : : {
688 : : /* don't error out, this is just a performance optimization */
689 : : ereport(WARNING,
690 : : (errcode_for_file_access(),
691 : : errmsg("could not flush dirty data: %m")));
692 : : }
693 : :
694 : : return;
695 : : }
696 : : #endif
697 : : }
698 : :
699 : : /*
700 : : * Truncate an open file to a given length.
701 : : */
702 : : static int
300 andres@anarazel.de 703 :CBC 520 : pg_ftruncate(int fd, off_t length)
704 : : {
705 : : int ret;
706 : :
707 : 520 : retry:
708 : 520 : ret = ftruncate(fd, length);
709 : :
710 [ - + - - ]: 520 : if (ret == -1 && errno == EINTR)
300 andres@anarazel.de 711 :UBC 0 : goto retry;
712 : :
300 andres@anarazel.de 713 :CBC 520 : return ret;
714 : : }
715 : :
716 : : /*
717 : : * Truncate a file to a given length by name.
718 : : */
719 : : int
1230 tmunro@postgresql.or 720 : 199495 : pg_truncate(const char *path, off_t length)
721 : : {
722 : : int ret;
723 : : #ifdef WIN32
724 : : int save_errno;
725 : : int fd;
726 : :
727 : : fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
728 : : if (fd >= 0)
729 : : {
730 : : ret = pg_ftruncate(fd, length);
731 : : save_errno = errno;
732 : : CloseTransientFile(fd);
733 : : errno = save_errno;
734 : : }
735 : : else
736 : : ret = -1;
737 : : #else
738 : :
300 andres@anarazel.de 739 : 199495 : retry:
740 : 199495 : ret = truncate(path, length);
741 : :
742 [ + + - + ]: 199495 : if (ret == -1 && errno == EINTR)
300 andres@anarazel.de 743 :UBC 0 : goto retry;
744 : : #endif
745 : :
300 andres@anarazel.de 746 :CBC 199495 : return ret;
747 : : }
748 : :
749 : : /*
750 : : * fsync_fname -- fsync a file or directory, handling errors properly
751 : : *
752 : : * Try to fsync a file or directory. When doing the latter, ignore errors that
753 : : * indicate the OS just doesn't allow/require fsyncing directories.
754 : : */
755 : : void
2958 756 : 16284 : fsync_fname(const char *fname, bool isdir)
757 : : {
1973 tmunro@postgresql.or 758 : 16284 : fsync_fname_ext(fname, isdir, false, data_sync_elevel(ERROR));
2958 andres@anarazel.de 759 : 16284 : }
760 : :
761 : : /*
762 : : * durable_rename -- rename(2) wrapper, issuing fsyncs required for durability
763 : : *
764 : : * This routine ensures that, after returning, the effect of renaming file
765 : : * persists in case of a crash. A crash while this routine is running will
766 : : * leave you with either the pre-existing or the moved file in place of the
767 : : * new file; no mixed state or truncated files are possible.
768 : : *
769 : : * It does so by using fsync on the old filename and the possibly existing
770 : : * target filename before the rename, and the target file and directory after.
771 : : *
772 : : * Note that rename() cannot be used across arbitrary directories, as they
773 : : * might not be on the same filesystem. Therefore this routine does not
774 : : * support renaming across directories.
775 : : *
776 : : * Log errors with the caller specified severity.
777 : : *
778 : : * Returns 0 if the operation succeeded, -1 otherwise. Note that errno is not
779 : : * valid upon return.
780 : : */
781 : : int
782 : 3257 : durable_rename(const char *oldfile, const char *newfile, int elevel)
783 : : {
784 : : int fd;
785 : :
786 : : /*
787 : : * First fsync the old and target path (if it exists), to ensure that they
788 : : * are properly persistent on disk. Syncing the target file is not
789 : : * strictly necessary, but it makes it easier to reason about crashes;
790 : : * because it's then guaranteed that either source or target file exists
791 : : * after a crash.
792 : : */
793 [ - + ]: 3257 : if (fsync_fname_ext(oldfile, false, false, elevel) != 0)
2958 andres@anarazel.de 794 :UBC 0 : return -1;
795 : :
2395 peter_e@gmx.net 796 :CBC 3257 : fd = OpenTransientFile(newfile, PG_BINARY | O_RDWR);
2958 andres@anarazel.de 797 [ + + ]: 3257 : if (fd < 0)
798 : : {
799 [ - + ]: 1909 : if (errno != ENOENT)
800 : : {
2958 andres@anarazel.de 801 [ # # ]:UBC 0 : ereport(elevel,
802 : : (errcode_for_file_access(),
803 : : errmsg("could not open file \"%s\": %m", newfile)));
804 : 0 : return -1;
805 : : }
806 : : }
807 : : else
808 : : {
2958 andres@anarazel.de 809 [ - + ]:CBC 1348 : if (pg_fsync(fd) != 0)
810 : : {
811 : : int save_errno;
812 : :
813 : : /* close file upon error, might not be in transaction context */
2958 andres@anarazel.de 814 :UBC 0 : save_errno = errno;
815 : 0 : CloseTransientFile(fd);
816 : 0 : errno = save_errno;
817 : :
818 [ # # ]: 0 : ereport(elevel,
819 : : (errcode_for_file_access(),
820 : : errmsg("could not fsync file \"%s\": %m", newfile)));
821 : 0 : return -1;
822 : : }
823 : :
1744 peter@eisentraut.org 824 [ - + ]:CBC 1348 : if (CloseTransientFile(fd) != 0)
825 : : {
1863 michael@paquier.xyz 826 [ # # ]:UBC 0 : ereport(elevel,
827 : : (errcode_for_file_access(),
828 : : errmsg("could not close file \"%s\": %m", newfile)));
829 : 0 : return -1;
830 : : }
831 : : }
832 : :
833 : : /* Time to do the real deal... */
2958 andres@anarazel.de 834 [ - + ]:CBC 3257 : if (rename(oldfile, newfile) < 0)
835 : : {
2958 andres@anarazel.de 836 [ # # ]:UBC 0 : ereport(elevel,
837 : : (errcode_for_file_access(),
838 : : errmsg("could not rename file \"%s\" to \"%s\": %m",
839 : : oldfile, newfile)));
840 : 0 : return -1;
841 : : }
842 : :
843 : : /*
844 : : * To guarantee renaming the file is persistent, fsync the file with its
845 : : * new name, and its containing directory.
846 : : */
2958 andres@anarazel.de 847 [ - + ]:CBC 3257 : if (fsync_fname_ext(newfile, false, false, elevel) != 0)
2958 andres@anarazel.de 848 :UBC 0 : return -1;
849 : :
2958 andres@anarazel.de 850 [ - + ]:CBC 3257 : if (fsync_parent_path(newfile, elevel) != 0)
2958 andres@anarazel.de 851 :UBC 0 : return -1;
852 : :
2958 andres@anarazel.de 853 :CBC 3257 : return 0;
854 : : }
855 : :
856 : : /*
857 : : * durable_unlink -- remove a file in a durable manner
858 : : *
859 : : * This routine ensures that, after returning, the effect of removing file
860 : : * persists in case of a crash. A crash while this routine is running will
861 : : * leave the system in no mixed state.
862 : : *
863 : : * It does so by using fsync on the parent directory of the file after the
864 : : * actual removal is done.
865 : : *
866 : : * Log errors with the severity specified by caller.
867 : : *
868 : : * Returns 0 if the operation succeeded, -1 otherwise. Note that errno is not
869 : : * valid upon return.
870 : : */
871 : : int
2575 teodor@sigaev.ru 872 : 140 : durable_unlink(const char *fname, int elevel)
873 : : {
874 [ + + ]: 140 : if (unlink(fname) < 0)
875 : : {
876 [ + + ]: 36 : ereport(elevel,
877 : : (errcode_for_file_access(),
878 : : errmsg("could not remove file \"%s\": %m",
879 : : fname)));
880 : 36 : return -1;
881 : : }
882 : :
883 : : /*
884 : : * To guarantee that the removal of the file is persistent, fsync its
885 : : * parent directory.
886 : : */
887 [ - + ]: 104 : if (fsync_parent_path(fname, elevel) != 0)
2575 teodor@sigaev.ru 888 :UBC 0 : return -1;
889 : :
2575 teodor@sigaev.ru 890 :CBC 104 : return 0;
891 : : }
892 : :
893 : : /*
894 : : * InitFileAccess --- initialize this module during backend startup
895 : : *
896 : : * This is called during either normal or standalone backend start.
897 : : * It is *not* called in the postmaster.
898 : : *
899 : : * Note that this does not initialize temporary file access, that is
900 : : * separately initialized via InitTemporaryFileAccess().
901 : : */
902 : : void
6824 tgl@sss.pgh.pa.us 903 : 19575 : InitFileAccess(void)
904 : : {
6756 bruce@momjian.us 905 [ - + ]: 19575 : Assert(SizeVfdCache == 0); /* call me only once */
906 : :
907 : : /* initialize cache header entry */
6824 tgl@sss.pgh.pa.us 908 : 19575 : VfdCache = (Vfd *) malloc(sizeof(Vfd));
909 [ - + ]: 19575 : if (VfdCache == NULL)
6824 tgl@sss.pgh.pa.us 910 [ # # ]:UBC 0 : ereport(FATAL,
911 : : (errcode(ERRCODE_OUT_OF_MEMORY),
912 : : errmsg("out of memory")));
913 : :
6824 tgl@sss.pgh.pa.us 914 [ + - + - :CBC 156600 : MemSet((char *) &(VfdCache[0]), 0, sizeof(Vfd));
+ - + - +
+ ]
915 : 19575 : VfdCache->fd = VFD_CLOSED;
916 : :
917 : 19575 : SizeVfdCache = 1;
981 andres@anarazel.de 918 : 19575 : }
919 : :
920 : : /*
921 : : * InitTemporaryFileAccess --- initialize temporary file access during startup
922 : : *
923 : : * This is called during either normal or standalone backend start.
924 : : * It is *not* called in the postmaster.
925 : : *
926 : : * This is separate from InitFileAccess() because temporary file cleanup can
927 : : * cause pgstat reporting. As pgstat is shut down during before_shmem_exit(),
928 : : * our reporting has to happen before that. Low level file access should be
929 : : * available for longer, hence the separate initialization / shutdown of
930 : : * temporary file handling.
931 : : */
932 : : void
933 : 19575 : InitTemporaryFileAccess(void)
934 : : {
731 drowley@postgresql.o 935 [ - + ]: 19575 : Assert(SizeVfdCache != 0); /* InitFileAccess() needs to have run */
981 andres@anarazel.de 936 [ - + ]: 19575 : Assert(!temporary_files_allowed); /* call me only once */
937 : :
938 : : /*
939 : : * Register before-shmem-exit hook to ensure temp files are dropped while
940 : : * we can still report stats.
941 : : */
942 : 19575 : before_shmem_exit(BeforeShmemExit_Files, 0);
943 : :
944 : : #ifdef USE_ASSERT_CHECKING
945 : 19575 : temporary_files_allowed = true;
946 : : #endif
6824 tgl@sss.pgh.pa.us 947 : 19575 : }
948 : :
949 : : /*
950 : : * count_usable_fds --- count how many FDs the system will let us open,
951 : : * and estimate how many are already open.
952 : : *
953 : : * We stop counting if usable_fds reaches max_to_probe. Note: a small
954 : : * value of max_to_probe might result in an underestimate of already_open;
955 : : * we must fill in any "gaps" in the set of used FDs before the calculation
956 : : * of already_open will give the right answer. In practice, max_to_probe
957 : : * of a couple of dozen should be enough to ensure good results.
958 : : *
959 : : * We assume stderr (FD 2) is available for dup'ing. While the calling
960 : : * script could theoretically close that, it would be a really bad idea,
961 : : * since then one risks loss of error messages from, e.g., libc.
962 : : */
963 : : static void
6825 964 : 728 : count_usable_fds(int max_to_probe, int *usable_fds, int *already_open)
965 : : {
966 : : int *fd;
967 : : int size;
7356 968 : 728 : int used = 0;
969 : 728 : int highestfd = 0;
970 : : int j;
971 : :
972 : : #ifdef HAVE_GETRLIMIT
973 : : struct rlimit rlim;
974 : : int getrlimit_status;
975 : : #endif
976 : :
977 : 728 : size = 1024;
978 : 728 : fd = (int *) palloc(size * sizeof(int));
979 : :
980 : : #ifdef HAVE_GETRLIMIT
5520 peter_e@gmx.net 981 : 728 : getrlimit_status = getrlimit(RLIMIT_NOFILE, &rlim);
982 [ - + ]: 728 : if (getrlimit_status != 0)
5520 peter_e@gmx.net 983 [ # # ]:UBC 0 : ereport(WARNING, (errmsg("getrlimit failed: %m")));
984 : : #endif /* HAVE_GETRLIMIT */
985 : :
986 : : /* dup until failure or probe limit reached */
987 : : for (;;)
7356 tgl@sss.pgh.pa.us 988 :CBC 727272 : {
989 : : int thisfd;
990 : :
991 : : #ifdef HAVE_GETRLIMIT
992 : :
993 : : /*
994 : : * don't go beyond RLIMIT_NOFILE; causes irritating kernel logs on
995 : : * some platforms
996 : : */
5520 peter_e@gmx.net 997 [ + - - + ]: 728000 : if (getrlimit_status == 0 && highestfd >= rlim.rlim_cur - 1)
5520 peter_e@gmx.net 998 :UBC 0 : break;
999 : : #endif
1000 : :
955 tgl@sss.pgh.pa.us 1001 :CBC 728000 : thisfd = dup(2);
7356 1002 [ - + ]: 728000 : if (thisfd < 0)
1003 : : {
1004 : : /* Expect EMFILE or ENFILE, else it's fishy */
7356 tgl@sss.pgh.pa.us 1005 [ # # # # ]:UBC 0 : if (errno != EMFILE && errno != ENFILE)
955 1006 [ # # ]: 0 : elog(WARNING, "duplicating stderr file descriptor failed after %d successes: %m", used);
7356 1007 : 0 : break;
1008 : : }
1009 : :
7356 tgl@sss.pgh.pa.us 1010 [ - + ]:CBC 728000 : if (used >= size)
1011 : : {
7356 tgl@sss.pgh.pa.us 1012 :UBC 0 : size *= 2;
1013 : 0 : fd = (int *) repalloc(fd, size * sizeof(int));
1014 : : }
7356 tgl@sss.pgh.pa.us 1015 :CBC 728000 : fd[used++] = thisfd;
1016 : :
1017 [ + - ]: 728000 : if (highestfd < thisfd)
1018 : 728000 : highestfd = thisfd;
1019 : :
6825 1020 [ + + ]: 728000 : if (used >= max_to_probe)
1021 : 728 : break;
1022 : : }
1023 : :
1024 : : /* release the files we opened */
7356 1025 [ + + ]: 728728 : for (j = 0; j < used; j++)
1026 : 728000 : close(fd[j]);
1027 : :
1028 : 728 : pfree(fd);
1029 : :
1030 : : /*
1031 : : * Return results. usable_fds is just the number of successful dups. We
1032 : : * assume that the system limit is highestfd+1 (remember 0 is a legal FD
1033 : : * number) and so already_open is highestfd+1 - usable_fds.
1034 : : */
1035 : 728 : *usable_fds = used;
7168 bruce@momjian.us 1036 : 728 : *already_open = highestfd + 1 - used;
7356 tgl@sss.pgh.pa.us 1037 : 728 : }
1038 : :
1039 : : /*
1040 : : * set_max_safe_fds
1041 : : * Determine number of file descriptors that fd.c is allowed to use
1042 : : */
1043 : : void
1044 : 728 : set_max_safe_fds(void)
1045 : : {
1046 : : int usable_fds;
1047 : : int already_open;
1048 : :
1049 : : /*----------
1050 : : * We want to set max_safe_fds to
1051 : : * MIN(usable_fds, max_files_per_process - already_open)
1052 : : * less the slop factor for files that are opened without consulting
1053 : : * fd.c. This ensures that we won't exceed either max_files_per_process
1054 : : * or the experimentally-determined EMFILE limit.
1055 : : *----------
1056 : : */
6825 1057 : 728 : count_usable_fds(max_files_per_process,
1058 : : &usable_fds, &already_open);
1059 : :
7356 1060 : 728 : max_safe_fds = Min(usable_fds, max_files_per_process - already_open);
1061 : :
1062 : : /*
1063 : : * Take off the FDs reserved for system() etc.
1064 : : */
1065 : 728 : max_safe_fds -= NUM_RESERVED_FDS;
1066 : :
1067 : : /*
1068 : : * Make sure we still have enough to get by.
1069 : : */
1070 [ - + ]: 728 : if (max_safe_fds < FD_MINFREE)
7356 tgl@sss.pgh.pa.us 1071 [ # # ]:UBC 0 : ereport(FATAL,
1072 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
1073 : : errmsg("insufficient file descriptors available to start server process"),
1074 : : errdetail("System allows %d, server needs at least %d.",
1075 : : max_safe_fds + NUM_RESERVED_FDS,
1076 : : FD_MINFREE + NUM_RESERVED_FDS)));
1077 : :
7356 tgl@sss.pgh.pa.us 1078 [ + + ]:CBC 728 : elog(DEBUG2, "max_safe_fds = %d, usable_fds = %d, already_open = %d",
1079 : : max_safe_fds, usable_fds, already_open);
1080 : 728 : }
1081 : :
1082 : : /*
1083 : : * Open a file with BasicOpenFilePerm() and pass default file mode for the
1084 : : * fileMode parameter.
1085 : : */
1086 : : int
2395 peter_e@gmx.net 1087 : 32743 : BasicOpenFile(const char *fileName, int fileFlags)
1088 : : {
2199 sfrost@snowman.net 1089 : 32743 : return BasicOpenFilePerm(fileName, fileFlags, pg_file_create_mode);
1090 : : }
1091 : :
1092 : : /*
1093 : : * BasicOpenFilePerm --- same as open(2) except can free other FDs if needed
1094 : : *
1095 : : * This is exported for use by places that really want a plain kernel FD,
1096 : : * but need to be proof against running out of FDs. Once an FD has been
1097 : : * successfully returned, it is the caller's responsibility to ensure that
1098 : : * it will not be leaked on ereport()! Most users should *not* call this
1099 : : * routine directly, but instead use the VFD abstraction level, which
1100 : : * provides protection against descriptor leaks as well as management of
1101 : : * files that need to be open for more than a short period of time.
1102 : : *
1103 : : * Ideally this should be the *only* direct call of open() in the backend.
1104 : : * In practice, the postmaster calls open() directly, and there are some
1105 : : * direct open() calls done early in backend startup. Those are OK since
1106 : : * this module wouldn't have any open files to close at that point anyway.
1107 : : */
1108 : : int
2395 peter_e@gmx.net 1109 : 8945591 : BasicOpenFilePerm(const char *fileName, int fileFlags, mode_t fileMode)
1110 : : {
1111 : : int fd;
1112 : :
8717 tgl@sss.pgh.pa.us 1113 : 8945591 : tryAgain:
1114 : : #ifdef PG_O_DIRECT_USE_F_NOCACHE
1115 : :
1116 : : /*
1117 : : * The value we defined to stand in for O_DIRECT when simulating it with
1118 : : * F_NOCACHE had better not collide with any of the standard flags.
1119 : : */
1120 : : StaticAssertStmt((PG_O_DIRECT &
1121 : : (O_APPEND |
1122 : : O_CLOEXEC |
1123 : : O_CREAT |
1124 : : O_DSYNC |
1125 : : O_EXCL |
1126 : : O_RDWR |
1127 : : O_RDONLY |
1128 : : O_SYNC |
1129 : : O_TRUNC |
1130 : : O_WRONLY)) == 0,
1131 : : "PG_O_DIRECT value collides with standard flag");
1132 : : fd = open(fileName, fileFlags & ~PG_O_DIRECT, fileMode);
1133 : : #else
1134 : 8945591 : fd = open(fileName, fileFlags, fileMode);
1135 : : #endif
1136 : :
1137 [ + + ]: 8945590 : if (fd >= 0)
1138 : : {
1139 : : #ifdef PG_O_DIRECT_USE_F_NOCACHE
1140 : : if (fileFlags & PG_O_DIRECT)
1141 : : {
1142 : : if (fcntl(fd, F_NOCACHE, 1) < 0)
1143 : : {
1144 : : int save_errno = errno;
1145 : :
1146 : : close(fd);
1147 : : errno = save_errno;
1148 : : return -1;
1149 : : }
1150 : : }
1151 : : #endif
1152 : :
1153 : 8552684 : return fd; /* success! */
1154 : : }
1155 : :
8631 1156 [ + - - + ]: 392906 : if (errno == EMFILE || errno == ENFILE)
1157 : : {
8424 bruce@momjian.us 1158 :UBC 0 : int save_errno = errno;
1159 : :
7570 tgl@sss.pgh.pa.us 1160 [ # # ]: 0 : ereport(LOG,
1161 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
1162 : : errmsg("out of file descriptors: %m; release and retry")));
8717 1163 : 0 : errno = 0;
8631 1164 [ # # ]: 0 : if (ReleaseLruFile())
1165 : 0 : goto tryAgain;
1166 : 0 : errno = save_errno;
1167 : : }
1168 : :
8717 tgl@sss.pgh.pa.us 1169 :CBC 392906 : return -1; /* failure */
1170 : : }
1171 : :
1172 : : /*
1173 : : * AcquireExternalFD - attempt to reserve an external file descriptor
1174 : : *
1175 : : * This should be used by callers that need to hold a file descriptor open
1176 : : * over more than a short interval, but cannot use any of the other facilities
1177 : : * provided by this module.
1178 : : *
1179 : : * The difference between this and the underlying ReserveExternalFD function
1180 : : * is that this will report failure (by setting errno and returning false)
1181 : : * if "too many" external FDs are already reserved. This should be used in
1182 : : * any code where the total number of FDs to be reserved is not predictable
1183 : : * and small.
1184 : : */
1185 : : bool
1511 1186 : 152009 : AcquireExternalFD(void)
1187 : : {
1188 : : /*
1189 : : * We don't want more than max_safe_fds / 3 FDs to be consumed for
1190 : : * "external" FDs.
1191 : : */
1192 [ + - ]: 152009 : if (numExternalFDs < max_safe_fds / 3)
1193 : : {
1194 : 152009 : ReserveExternalFD();
1195 : 152009 : return true;
1196 : : }
1511 tgl@sss.pgh.pa.us 1197 :UBC 0 : errno = EMFILE;
1198 : 0 : return false;
1199 : : }
1200 : :
1201 : : /*
1202 : : * ReserveExternalFD - report external consumption of a file descriptor
1203 : : *
1204 : : * This should be used by callers that need to hold a file descriptor open
1205 : : * over more than a short interval, but cannot use any of the other facilities
1206 : : * provided by this module. This just tracks the use of the FD and closes
1207 : : * VFDs if needed to ensure we keep NUM_RESERVED_FDS FDs available.
1208 : : *
1209 : : * Call this directly only in code where failure to reserve the FD would be
1210 : : * fatal; for example, the WAL-writing code does so, since the alternative is
1211 : : * session failure. Also, it's very unwise to do so in code that could
1212 : : * consume more than one FD per process.
1213 : : *
1214 : : * Note: as long as everybody plays nice so that NUM_RESERVED_FDS FDs remain
1215 : : * available, it doesn't matter too much whether this is called before or
1216 : : * after actually opening the FD; but doing so beforehand reduces the risk of
1217 : : * an EMFILE failure if not everybody played nice. In any case, it's solely
1218 : : * caller's responsibility to keep the external-FD count in sync with reality.
1219 : : */
1220 : : void
1511 tgl@sss.pgh.pa.us 1221 :CBC 215196 : ReserveExternalFD(void)
1222 : : {
1223 : : /*
1224 : : * Release VFDs if needed to stay safe. Because we do this before
1225 : : * incrementing numExternalFDs, the final state will be as desired, i.e.,
1226 : : * nfile + numAllocatedDescs + numExternalFDs <= max_safe_fds.
1227 : : */
1228 : 215196 : ReleaseLruFiles();
1229 : :
1230 : 215196 : numExternalFDs++;
1231 : 215196 : }
1232 : :
1233 : : /*
1234 : : * ReleaseExternalFD - report release of an external file descriptor
1235 : : *
1236 : : * This is guaranteed not to change errno, so it can be used in failure paths.
1237 : : */
1238 : : void
1239 : 200312 : ReleaseExternalFD(void)
1240 : : {
1241 [ - + ]: 200312 : Assert(numExternalFDs > 0);
1242 : 200312 : numExternalFDs--;
1243 : 200312 : }
1244 : :
1245 : :
1246 : : #if defined(FDDEBUG)
1247 : :
1248 : : static void
1249 : : _dump_lru(void)
1250 : : {
1251 : : int mru = VfdCache[0].lruLessRecently;
1252 : : Vfd *vfdP = &VfdCache[mru];
1253 : : char buf[2048];
1254 : :
1255 : : snprintf(buf, sizeof(buf), "LRU: MOST %d ", mru);
1256 : : while (mru != 0)
1257 : : {
1258 : : mru = vfdP->lruLessRecently;
1259 : : vfdP = &VfdCache[mru];
1260 : : snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "%d ", mru);
1261 : : }
1262 : : snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "LEAST");
1263 : : elog(LOG, "%s", buf);
1264 : : }
1265 : : #endif /* FDDEBUG */
1266 : :
1267 : : static void
10141 scrappy@hub.org 1268 : 1330680 : Delete(File file)
1269 : : {
1270 : : Vfd *vfdP;
1271 : :
9107 tgl@sss.pgh.pa.us 1272 [ - + ]: 1330680 : Assert(file != 0);
1273 : :
1274 : : DO_DB(elog(LOG, "Delete %d (%s)",
1275 : : file, VfdCache[file].fileName));
1276 : : DO_DB(_dump_lru());
1277 : :
1278 : 1330680 : vfdP = &VfdCache[file];
1279 : :
1280 : 1330680 : VfdCache[vfdP->lruLessRecently].lruMoreRecently = vfdP->lruMoreRecently;
1281 : 1330680 : VfdCache[vfdP->lruMoreRecently].lruLessRecently = vfdP->lruLessRecently;
1282 : :
1283 : : DO_DB(_dump_lru());
10141 scrappy@hub.org 1284 : 1330680 : }
1285 : :
1286 : : static void
1287 : 39607 : LruDelete(File file)
1288 : : {
1289 : : Vfd *vfdP;
1290 : :
9107 tgl@sss.pgh.pa.us 1291 [ - + ]: 39607 : Assert(file != 0);
1292 : :
1293 : : DO_DB(elog(LOG, "LruDelete %d (%s)",
1294 : : file, VfdCache[file].fileName));
1295 : :
1296 : 39607 : vfdP = &VfdCache[file];
1297 : :
1298 : : /*
1299 : : * Close the file. We aren't expecting this to fail; if it does, better
1300 : : * to leak the FD than to mess up our internal state.
1301 : : */
1744 peter@eisentraut.org 1302 [ - + ]: 39607 : if (close(vfdP->fd) != 0)
1973 tmunro@postgresql.or 1303 [ # # # # ]:UBC 0 : elog(vfdP->fdstate & FD_TEMP_FILE_LIMIT ? LOG : data_sync_elevel(LOG),
1304 : : "could not close file \"%s\": %m", vfdP->fileName);
9107 tgl@sss.pgh.pa.us 1305 :CBC 39607 : vfdP->fd = VFD_CLOSED;
2609 1306 : 39607 : --nfile;
1307 : :
1308 : : /* delete the vfd record from the LRU ring */
1309 : 39607 : Delete(file);
10141 scrappy@hub.org 1310 : 39607 : }
1311 : :
1312 : : static void
1313 : 1691083 : Insert(File file)
1314 : : {
1315 : : Vfd *vfdP;
1316 : :
9107 tgl@sss.pgh.pa.us 1317 [ - + ]: 1691083 : Assert(file != 0);
1318 : :
1319 : : DO_DB(elog(LOG, "Insert %d (%s)",
1320 : : file, VfdCache[file].fileName));
1321 : : DO_DB(_dump_lru());
1322 : :
9716 bruce@momjian.us 1323 : 1691083 : vfdP = &VfdCache[file];
1324 : :
1325 : 1691083 : vfdP->lruMoreRecently = 0;
1326 : 1691083 : vfdP->lruLessRecently = VfdCache[0].lruLessRecently;
1327 : 1691083 : VfdCache[0].lruLessRecently = file;
1328 : 1691083 : VfdCache[vfdP->lruLessRecently].lruMoreRecently = file;
1329 : :
1330 : : DO_DB(_dump_lru());
10141 scrappy@hub.org 1331 : 1691083 : }
1332 : :
1333 : : /* returns 0 on success, -1 on re-open failure (with errno set) */
1334 : : static int
9716 bruce@momjian.us 1335 : 14391 : LruInsert(File file)
1336 : : {
1337 : : Vfd *vfdP;
1338 : :
9107 tgl@sss.pgh.pa.us 1339 [ - + ]: 14391 : Assert(file != 0);
1340 : :
1341 : : DO_DB(elog(LOG, "LruInsert %d (%s)",
1342 : : file, VfdCache[file].fileName));
1343 : :
9716 bruce@momjian.us 1344 : 14391 : vfdP = &VfdCache[file];
1345 : :
1346 [ + - ]: 14391 : if (FileIsNotOpen(file))
1347 : : {
1348 : : /* Close excess kernel FDs. */
3962 tgl@sss.pgh.pa.us 1349 : 14391 : ReleaseLruFiles();
1350 : :
1351 : : /*
1352 : : * The open could still fail for lack of file descriptors, eg due to
1353 : : * overall system file table being full. So, be prepared to release
1354 : : * another FD if necessary...
1355 : : */
2395 peter_e@gmx.net 1356 : 14391 : vfdP->fd = BasicOpenFilePerm(vfdP->fileName, vfdP->fileFlags,
1357 : : vfdP->fileMode);
9716 bruce@momjian.us 1358 [ - + ]: 14391 : if (vfdP->fd < 0)
1359 : : {
1360 : : DO_DB(elog(LOG, "re-open failed: %m"));
3986 tgl@sss.pgh.pa.us 1361 :UBC 0 : return -1;
1362 : : }
1363 : : else
1364 : : {
9716 bruce@momjian.us 1365 :CBC 14391 : ++nfile;
1366 : : }
1367 : : }
1368 : :
1369 : : /*
1370 : : * put it at the head of the Lru ring
1371 : : */
1372 : :
1373 : 14391 : Insert(file);
1374 : :
9357 1375 : 14391 : return 0;
1376 : : }
1377 : :
1378 : : /*
1379 : : * Release one kernel FD by closing the least-recently-used VFD.
1380 : : */
1381 : : static bool
8631 tgl@sss.pgh.pa.us 1382 : 39479 : ReleaseLruFile(void)
1383 : : {
1384 : : DO_DB(elog(LOG, "ReleaseLruFile. Opened %d", nfile));
1385 : :
1386 [ + - ]: 39479 : if (nfile > 0)
1387 : : {
1388 : : /*
1389 : : * There are opened files and so there should be at least one used vfd
1390 : : * in the ring.
1391 : : */
1392 [ - + ]: 39479 : Assert(VfdCache[0].lruMoreRecently != 0);
1393 : 39479 : LruDelete(VfdCache[0].lruMoreRecently);
1394 : 39479 : return true; /* freed a file */
1395 : : }
8631 tgl@sss.pgh.pa.us 1396 :UBC 0 : return false; /* no files available to free */
1397 : : }
1398 : :
1399 : : /*
1400 : : * Release kernel FDs as needed to get under the max_safe_fds limit.
1401 : : * After calling this, it's OK to try to open another file.
1402 : : */
1403 : : static void
3962 tgl@sss.pgh.pa.us 1404 :CBC 9234179 : ReleaseLruFiles(void)
1405 : : {
1511 1406 [ + + ]: 9273658 : while (nfile + numAllocatedDescs + numExternalFDs >= max_safe_fds)
1407 : : {
3962 1408 [ - + ]: 39479 : if (!ReleaseLruFile())
3962 tgl@sss.pgh.pa.us 1409 :UBC 0 : break;
1410 : : }
3962 tgl@sss.pgh.pa.us 1411 :CBC 9234179 : }
1412 : :
1413 : : static File
8631 1414 : 1247209 : AllocateVfd(void)
1415 : : {
1416 : : Index i;
1417 : : File file;
1418 : :
1419 : : DO_DB(elog(LOG, "AllocateVfd. Size %zu", SizeVfdCache));
1420 : :
6756 bruce@momjian.us 1421 [ - + ]: 1247209 : Assert(SizeVfdCache > 0); /* InitFileAccess not called? */
1422 : :
9716 1423 [ + + ]: 1247209 : if (VfdCache[0].nextFree == 0)
1424 : : {
1425 : : /*
1426 : : * The free list is empty so it is time to increase the size of the
1427 : : * array. We choose to double it each time this happens. However,
1428 : : * there's not much point in starting *real* small.
1429 : : */
9091 1430 : 22403 : Size newCacheSize = SizeVfdCache * 2;
1431 : : Vfd *newVfdCache;
1432 : :
9107 tgl@sss.pgh.pa.us 1433 [ + + ]: 22403 : if (newCacheSize < 32)
1434 : 16145 : newCacheSize = 32;
1435 : :
1436 : : /*
1437 : : * Be careful not to clobber VfdCache ptr if realloc fails.
1438 : : */
8412 1439 : 22403 : newVfdCache = (Vfd *) realloc(VfdCache, sizeof(Vfd) * newCacheSize);
1440 [ - + ]: 22403 : if (newVfdCache == NULL)
7570 tgl@sss.pgh.pa.us 1441 [ # # ]:UBC 0 : ereport(ERROR,
1442 : : (errcode(ERRCODE_OUT_OF_MEMORY),
1443 : : errmsg("out of memory")));
8412 tgl@sss.pgh.pa.us 1444 :CBC 22403 : VfdCache = newVfdCache;
1445 : :
1446 : : /*
1447 : : * Initialize the new entries and link them into the free list.
1448 : : */
9107 1449 [ + + ]: 1094674 : for (i = SizeVfdCache; i < newCacheSize; i++)
1450 : : {
1451 [ + - + - : 8578168 : MemSet((char *) &(VfdCache[i]), 0, sizeof(Vfd));
+ - + - +
+ ]
9716 bruce@momjian.us 1452 : 1072271 : VfdCache[i].nextFree = i + 1;
1453 : 1072271 : VfdCache[i].fd = VFD_CLOSED;
1454 : : }
9107 tgl@sss.pgh.pa.us 1455 : 22403 : VfdCache[newCacheSize - 1].nextFree = 0;
9716 bruce@momjian.us 1456 : 22403 : VfdCache[0].nextFree = SizeVfdCache;
1457 : :
1458 : : /*
1459 : : * Record the new size
1460 : : */
9107 tgl@sss.pgh.pa.us 1461 : 22403 : SizeVfdCache = newCacheSize;
1462 : : }
1463 : :
9716 bruce@momjian.us 1464 : 1247209 : file = VfdCache[0].nextFree;
1465 : :
1466 : 1247209 : VfdCache[0].nextFree = VfdCache[file].nextFree;
1467 : :
1468 : 1247209 : return file;
1469 : : }
1470 : :
1471 : : static void
10141 scrappy@hub.org 1472 : 870663 : FreeVfd(File file)
1473 : : {
9107 tgl@sss.pgh.pa.us 1474 : 870663 : Vfd *vfdP = &VfdCache[file];
1475 : :
1476 : : DO_DB(elog(LOG, "FreeVfd: %d (%s)",
1477 : : file, vfdP->fileName ? vfdP->fileName : ""));
1478 : :
1479 [ + + ]: 870663 : if (vfdP->fileName != NULL)
1480 : : {
1481 : 480600 : free(vfdP->fileName);
1482 : 480600 : vfdP->fileName = NULL;
1483 : : }
8412 1484 : 870663 : vfdP->fdstate = 0x0;
1485 : :
9107 1486 : 870663 : vfdP->nextFree = VfdCache[0].nextFree;
9716 bruce@momjian.us 1487 : 870663 : VfdCache[0].nextFree = file;
10141 scrappy@hub.org 1488 : 870663 : }
1489 : :
1490 : : /* returns 0 on success, -1 on re-open failure (with errno set) */
1491 : : static int
1492 : 2640399 : FileAccess(File file)
1493 : : {
1494 : : int returnValue;
1495 : :
1496 : : DO_DB(elog(LOG, "FileAccess %d (%s)",
1497 : : file, VfdCache[file].fileName));
1498 : :
1499 : : /*
1500 : : * Is the file open? If not, open it and put it at the head of the LRU
1501 : : * ring (possibly closing the least recently used file to get an FD).
1502 : : */
1503 : :
9716 bruce@momjian.us 1504 [ + + ]: 2640399 : if (FileIsNotOpen(file))
1505 : : {
1506 : 14391 : returnValue = LruInsert(file);
1507 [ - + ]: 14391 : if (returnValue != 0)
9716 bruce@momjian.us 1508 :UBC 0 : return returnValue;
1509 : : }
9107 tgl@sss.pgh.pa.us 1510 [ + + ]:CBC 2626008 : else if (VfdCache[0].lruLessRecently != file)
1511 : : {
1512 : : /*
1513 : : * We now know that the file is open and that it is not the last one
1514 : : * accessed, so we need to move it to the head of the Lru ring.
1515 : : */
1516 : :
9716 bruce@momjian.us 1517 : 819547 : Delete(file);
1518 : 819547 : Insert(file);
1519 : : }
1520 : :
9357 1521 : 2640399 : return 0;
1522 : : }
1523 : :
1524 : : /*
1525 : : * Called whenever a temporary file is deleted to report its size.
1526 : : */
1527 : : static void
2326 andres@anarazel.de 1528 : 3798 : ReportTemporaryFileUsage(const char *path, off_t size)
1529 : : {
1530 : 3798 : pgstat_report_tempfile(size);
1531 : :
1532 [ + + ]: 3798 : if (log_temp_files >= 0)
1533 : : {
1534 [ + + ]: 1244 : if ((size / 1024) >= log_temp_files)
1535 [ + - ]: 109 : ereport(LOG,
1536 : : (errmsg("temporary file: path \"%s\", size %lu",
1537 : : path, (unsigned long) size)));
1538 : : }
1539 : 3798 : }
1540 : :
1541 : : /*
1542 : : * Called to register a temporary file for automatic close.
1543 : : * ResourceOwnerEnlarge(CurrentResourceOwner) must have been called
1544 : : * before the file was opened.
1545 : : */
1546 : : static void
1547 : 6135 : RegisterTemporaryFile(File file)
1548 : : {
1549 : 6135 : ResourceOwnerRememberFile(CurrentResourceOwner, file);
1550 : 6135 : VfdCache[file].resowner = CurrentResourceOwner;
1551 : :
1552 : : /* Backup mechanism for closing at end of xact. */
1553 : 6135 : VfdCache[file].fdstate |= FD_CLOSE_AT_EOXACT;
1554 : 6135 : have_xact_temporary_files = true;
1555 : 6135 : }
1556 : :
1557 : : /*
1558 : : * Called when we get a shared invalidation message on some relation.
1559 : : */
1560 : : #ifdef NOT_USED
1561 : : void
1562 : : FileInvalidate(File file)
1563 : : {
1564 : : Assert(FileIsValid(file));
1565 : : if (!FileIsNotOpen(file))
1566 : : LruDelete(file);
1567 : : }
1568 : : #endif
1569 : :
1570 : : /*
1571 : : * Open a file with PathNameOpenFilePerm() and pass default file mode for the
1572 : : * fileMode parameter.
1573 : : */
1574 : : File
2395 peter_e@gmx.net 1575 : 1247209 : PathNameOpenFile(const char *fileName, int fileFlags)
1576 : : {
2199 sfrost@snowman.net 1577 : 1247209 : return PathNameOpenFilePerm(fileName, fileFlags, pg_file_create_mode);
1578 : : }
1579 : :
1580 : : /*
1581 : : * open a file in an arbitrary directory
1582 : : *
1583 : : * NB: if the passed pathname is relative (which it usually is),
1584 : : * it will be interpreted relative to the process' working directory
1585 : : * (which should always be $PGDATA when this code is running).
1586 : : */
1587 : : File
2395 peter_e@gmx.net 1588 : 1247209 : PathNameOpenFilePerm(const char *fileName, int fileFlags, mode_t fileMode)
1589 : : {
1590 : : char *fnamecopy;
1591 : : File file;
1592 : : Vfd *vfdP;
1593 : :
1594 : : DO_DB(elog(LOG, "PathNameOpenFilePerm: %s %x %o",
1595 : : fileName, fileFlags, fileMode));
1596 : :
1597 : : /*
1598 : : * We need a malloc'd copy of the file name; fail cleanly if no room.
1599 : : */
7570 tgl@sss.pgh.pa.us 1600 : 1247209 : fnamecopy = strdup(fileName);
1601 [ - + ]: 1247209 : if (fnamecopy == NULL)
7570 tgl@sss.pgh.pa.us 1602 [ # # ]:UBC 0 : ereport(ERROR,
1603 : : (errcode(ERRCODE_OUT_OF_MEMORY),
1604 : : errmsg("out of memory")));
1605 : :
9716 bruce@momjian.us 1606 :CBC 1247209 : file = AllocateVfd();
1607 : 1247209 : vfdP = &VfdCache[file];
1608 : :
1609 : : /* Close excess kernel FDs. */
3962 tgl@sss.pgh.pa.us 1610 : 1247209 : ReleaseLruFiles();
1611 : :
1612 : : /*
1613 : : * Descriptors managed by VFDs are implicitly marked O_CLOEXEC. The
1614 : : * client shouldn't be expected to know which kernel descriptors are
1615 : : * currently open, so it wouldn't make sense for them to be inherited by
1616 : : * executed subprograms.
1617 : : */
408 tmunro@postgresql.or 1618 : 1247209 : fileFlags |= O_CLOEXEC;
1619 : :
2395 peter_e@gmx.net 1620 : 1247209 : vfdP->fd = BasicOpenFilePerm(fileName, fileFlags, fileMode);
1621 : :
9716 bruce@momjian.us 1622 [ + + ]: 1247208 : if (vfdP->fd < 0)
1623 : : {
3986 tgl@sss.pgh.pa.us 1624 : 390063 : int save_errno = errno;
1625 : :
9716 bruce@momjian.us 1626 : 390063 : FreeVfd(file);
7570 tgl@sss.pgh.pa.us 1627 : 390063 : free(fnamecopy);
3986 1628 : 390063 : errno = save_errno;
9716 bruce@momjian.us 1629 : 390063 : return -1;
1630 : : }
1631 : 857145 : ++nfile;
1632 : : DO_DB(elog(LOG, "PathNameOpenFile: success %d",
1633 : : vfdP->fd));
1634 : :
7570 tgl@sss.pgh.pa.us 1635 : 857145 : vfdP->fileName = fnamecopy;
1636 : : /* Saved flags are adjusted to be OK for re-opening file */
8412 1637 : 857145 : vfdP->fileFlags = fileFlags & ~(O_CREAT | O_TRUNC | O_EXCL);
9716 bruce@momjian.us 1638 : 857145 : vfdP->fileMode = fileMode;
4655 tgl@sss.pgh.pa.us 1639 : 857145 : vfdP->fileSize = 0;
7922 1640 : 857145 : vfdP->fdstate = 0x0;
5246 heikki.linnakangas@i 1641 : 857145 : vfdP->resowner = NULL;
1642 : :
1245 tgl@sss.pgh.pa.us 1643 : 857145 : Insert(file);
1644 : :
9716 bruce@momjian.us 1645 : 857145 : return file;
1646 : : }
1647 : :
1648 : : /*
1649 : : * Create directory 'directory'. If necessary, create 'basedir', which must
1650 : : * be the directory above it. This is designed for creating the top-level
1651 : : * temporary directory on demand before creating a directory underneath it.
1652 : : * Do nothing if the directory already exists.
1653 : : *
1654 : : * Directories created within the top-level temporary directory should begin
1655 : : * with PG_TEMP_FILE_PREFIX, so that they can be identified as temporary and
1656 : : * deleted at startup by RemovePgTempFiles(). Further subdirectories below
1657 : : * that do not need any particular prefix.
1658 : : */
1659 : : void
2326 andres@anarazel.de 1660 : 172 : PathNameCreateTemporaryDir(const char *basedir, const char *directory)
1661 : : {
2199 sfrost@snowman.net 1662 [ + + ]: 172 : if (MakePGDirectory(directory) < 0)
1663 : : {
2326 andres@anarazel.de 1664 [ + + ]: 19 : if (errno == EEXIST)
1665 : 8 : return;
1666 : :
1667 : : /*
1668 : : * Failed. Try to create basedir first in case it's missing. Tolerate
1669 : : * EEXIST to close a race against another process following the same
1670 : : * algorithm.
1671 : : */
2199 sfrost@snowman.net 1672 [ - + - - ]: 11 : if (MakePGDirectory(basedir) < 0 && errno != EEXIST)
2326 andres@anarazel.de 1673 [ # # ]:UBC 0 : ereport(ERROR,
1674 : : (errcode_for_file_access(),
1675 : : errmsg("cannot create temporary directory \"%s\": %m",
1676 : : basedir)));
1677 : :
1678 : : /* Try again. */
2199 sfrost@snowman.net 1679 [ - + - - ]:CBC 11 : if (MakePGDirectory(directory) < 0 && errno != EEXIST)
2326 andres@anarazel.de 1680 [ # # ]:UBC 0 : ereport(ERROR,
1681 : : (errcode_for_file_access(),
1682 : : errmsg("cannot create temporary subdirectory \"%s\": %m",
1683 : : directory)));
1684 : : }
1685 : : }
1686 : :
1687 : : /*
1688 : : * Delete a directory and everything in it, if it exists.
1689 : : */
1690 : : void
2326 andres@anarazel.de 1691 :CBC 203 : PathNameDeleteTemporaryDir(const char *dirname)
1692 : : {
1693 : : struct stat statbuf;
1694 : :
1695 : : /* Silently ignore missing directory. */
1696 [ + + + - ]: 203 : if (stat(dirname, &statbuf) != 0 && errno == ENOENT)
1697 : 40 : return;
1698 : :
1699 : : /*
1700 : : * Currently, walkdir doesn't offer a way for our passed in function to
1701 : : * maintain state. Perhaps it should, so that we could tell the caller
1702 : : * whether this operation succeeded or failed. Since this operation is
1703 : : * used in a cleanup path, we wouldn't actually behave differently: we'll
1704 : : * just log failures.
1705 : : */
1706 : 163 : walkdir(dirname, unlink_if_exists_fname, false, LOG);
1707 : : }
1708 : :
1709 : : /*
1710 : : * Open a temporary file that will disappear when we close it.
1711 : : *
1712 : : * This routine takes care of generating an appropriate tempfile name.
1713 : : * There's no need to pass in fileFlags or fileMode either, since only
1714 : : * one setting makes any sense for a temp file.
1715 : : *
1716 : : * Unless interXact is true, the file is remembered by CurrentResourceOwner
1717 : : * to ensure it's closed and deleted when it's no longer needed, typically at
1718 : : * the end-of-transaction. In most cases, you don't want temporary files to
1719 : : * outlive the transaction that created them, so this should be false -- but
1720 : : * if you need "somewhat" temporary storage, this might be useful. In either
1721 : : * case, the file is removed when the File is explicitly closed.
1722 : : */
1723 : : File
6156 tgl@sss.pgh.pa.us 1724 : 1927 : OpenTemporaryFile(bool interXact)
1725 : : {
6160 1726 : 1927 : File file = 0;
1727 : :
981 andres@anarazel.de 1728 [ - + ]: 1927 : Assert(temporary_files_allowed); /* check temp file access is up */
1729 : :
1730 : : /*
1731 : : * Make sure the current resource owner has space for this File before we
1732 : : * open it, if we'll be registering it below.
1733 : : */
2349 tgl@sss.pgh.pa.us 1734 [ + + ]: 1927 : if (!interXact)
158 heikki.linnakangas@i 1735 :GNC 1921 : ResourceOwnerEnlarge(CurrentResourceOwner);
1736 : :
1737 : : /*
1738 : : * If some temp tablespace(s) have been given to us, try to use the next
1739 : : * one. If a given tablespace can't be found, we silently fall back to
1740 : : * the database's default tablespace.
1741 : : *
1742 : : * BUT: if the temp file is slated to outlive the current transaction,
1743 : : * force it into the database's default tablespace, so that it will not
1744 : : * pose a threat to possible tablespace drop attempts.
1745 : : */
6156 tgl@sss.pgh.pa.us 1746 [ + + + - ]:CBC 1927 : if (numTempTableSpaces > 0 && !interXact)
1747 : : {
5995 bruce@momjian.us 1748 :GBC 1 : Oid tblspcOid = GetNextTempTableSpace();
1749 : :
6156 tgl@sss.pgh.pa.us 1750 [ + - ]: 1 : if (OidIsValid(tblspcOid))
1751 : 1 : file = OpenTemporaryFileInTablespace(tblspcOid, false);
1752 : : }
1753 : :
1754 : : /*
1755 : : * If not, or if tablespace is bad, create in database's default
1756 : : * tablespace. MyDatabaseTableSpace should normally be set before we get
1757 : : * here, but just in case it isn't, fall back to pg_default tablespace.
1758 : : */
6160 tgl@sss.pgh.pa.us 1759 [ + + ]:CBC 1927 : if (file <= 0)
1760 [ + + ]: 1926 : file = OpenTemporaryFileInTablespace(MyDatabaseTableSpace ?
1761 : : MyDatabaseTableSpace :
1762 : : DEFAULTTABLESPACE_OID,
1763 : : true);
1764 : :
1765 : : /* Mark it for deletion at close and temporary file size limit */
2326 andres@anarazel.de 1766 : 1927 : VfdCache[file].fdstate |= FD_DELETE_AT_CLOSE | FD_TEMP_FILE_LIMIT;
1767 : :
1768 : : /* Register it with the current resource owner */
6160 tgl@sss.pgh.pa.us 1769 [ + + ]: 1927 : if (!interXact)
2326 andres@anarazel.de 1770 : 1921 : RegisterTemporaryFile(file);
1771 : :
6160 tgl@sss.pgh.pa.us 1772 : 1927 : return file;
1773 : : }
1774 : :
1775 : : /*
1776 : : * Return the path of the temp directory in a given tablespace.
1777 : : */
1778 : : void
2326 andres@anarazel.de 1779 : 11616 : TempTablespacePath(char *path, Oid tablespace)
1780 : : {
1781 : : /*
1782 : : * Identify the tempfile directory for this tablespace.
1783 : : *
1784 : : * If someone tries to specify pg_global, use pg_default instead.
1785 : : */
1786 [ + - + + ]: 11616 : if (tablespace == InvalidOid ||
2326 andres@anarazel.de 1787 [ - + ]:GBC 1 : tablespace == DEFAULTTABLESPACE_OID ||
1788 : : tablespace == GLOBALTABLESPACE_OID)
2326 andres@anarazel.de 1789 :CBC 11615 : snprintf(path, MAXPGPATH, "base/%s", PG_TEMP_FILES_DIR);
1790 : : else
1791 : : {
1792 : : /* All other tablespaces are accessed via symlinks */
2326 andres@anarazel.de 1793 :GBC 1 : snprintf(path, MAXPGPATH, "pg_tblspc/%u/%s/%s",
1794 : : tablespace, TABLESPACE_VERSION_DIRECTORY,
1795 : : PG_TEMP_FILES_DIR);
1796 : : }
2326 andres@anarazel.de 1797 :CBC 11616 : }
1798 : :
1799 : : /*
1800 : : * Open a temporary file in a specific tablespace.
1801 : : * Subroutine for OpenTemporaryFile, which see for details.
1802 : : */
1803 : : static File
1804 : 1927 : OpenTemporaryFileInTablespace(Oid tblspcOid, bool rejectError)
1805 : : {
1806 : : char tempdirpath[MAXPGPATH];
1807 : : char tempfilepath[MAXPGPATH];
1808 : : File file;
1809 : :
1810 : 1927 : TempTablespacePath(tempdirpath, tblspcOid);
1811 : :
1812 : : /*
1813 : : * Generate a tempfile name that should be unique within the current
1814 : : * database instance.
1815 : : */
6160 tgl@sss.pgh.pa.us 1816 : 1927 : snprintf(tempfilepath, sizeof(tempfilepath), "%s/%s%d.%ld",
1817 : : tempdirpath, PG_TEMP_FILE_PREFIX, MyProcPid, tempFileCounter++);
1818 : :
1819 : : /*
1820 : : * Open the file. Note: we don't use O_EXCL, in case there is an orphaned
1821 : : * temp file that can be reused.
1822 : : */
1823 : 1927 : file = PathNameOpenFile(tempfilepath,
1824 : : O_RDWR | O_CREAT | O_TRUNC | PG_BINARY);
9107 1825 [ + + ]: 1927 : if (file <= 0)
1826 : : {
1827 : : /*
1828 : : * We might need to create the tablespace's tempfile directory, if no
1829 : : * one has yet done so.
1830 : : *
1831 : : * Don't check for an error from MakePGDirectory; it could fail if
1832 : : * someone else just did the same thing. If it doesn't work then
1833 : : * we'll bomb out on the second create attempt, instead.
1834 : : */
2199 sfrost@snowman.net 1835 : 85 : (void) MakePGDirectory(tempdirpath);
1836 : :
6160 tgl@sss.pgh.pa.us 1837 : 85 : file = PathNameOpenFile(tempfilepath,
1838 : : O_RDWR | O_CREAT | O_TRUNC | PG_BINARY);
1839 [ - + - - ]: 85 : if (file <= 0 && rejectError)
7570 tgl@sss.pgh.pa.us 1840 [ # # ]:UBC 0 : elog(ERROR, "could not create temporary file \"%s\": %m",
1841 : : tempfilepath);
1842 : : }
1843 : :
9107 tgl@sss.pgh.pa.us 1844 :CBC 1927 : return file;
1845 : : }
1846 : :
1847 : :
1848 : : /*
1849 : : * Create a new file. The directory containing it must already exist. Files
1850 : : * created this way are subject to temp_file_limit and are automatically
1851 : : * closed at end of transaction, but are not automatically deleted on close
1852 : : * because they are intended to be shared between cooperating backends.
1853 : : *
1854 : : * If the file is inside the top-level temporary directory, its name should
1855 : : * begin with PG_TEMP_FILE_PREFIX so that it can be identified as temporary
1856 : : * and deleted at startup by RemovePgTempFiles(). Alternatively, it can be
1857 : : * inside a directory created with PathNameCreateTemporaryDir(), in which case
1858 : : * the prefix isn't needed.
1859 : : */
1860 : : File
2326 andres@anarazel.de 1861 : 2044 : PathNameCreateTemporaryFile(const char *path, bool error_on_failure)
1862 : : {
1863 : : File file;
1864 : :
981 1865 [ - + ]: 2044 : Assert(temporary_files_allowed); /* check temp file access is up */
1866 : :
158 heikki.linnakangas@i 1867 :GNC 2044 : ResourceOwnerEnlarge(CurrentResourceOwner);
1868 : :
1869 : : /*
1870 : : * Open the file. Note: we don't use O_EXCL, in case there is an orphaned
1871 : : * temp file that can be reused.
1872 : : */
2326 andres@anarazel.de 1873 :CBC 2044 : file = PathNameOpenFile(path, O_RDWR | O_CREAT | O_TRUNC | PG_BINARY);
1874 [ + + ]: 2043 : if (file <= 0)
1875 : : {
1876 [ - + ]: 172 : if (error_on_failure)
2326 andres@anarazel.de 1877 [ # # ]:UBC 0 : ereport(ERROR,
1878 : : (errcode_for_file_access(),
1879 : : errmsg("could not create temporary file \"%s\": %m",
1880 : : path)));
1881 : : else
2326 andres@anarazel.de 1882 :CBC 172 : return file;
1883 : : }
1884 : :
1885 : : /* Mark it for temp_file_limit accounting. */
1886 : 1871 : VfdCache[file].fdstate |= FD_TEMP_FILE_LIMIT;
1887 : :
1888 : : /* Register it for automatic close. */
1889 : 1871 : RegisterTemporaryFile(file);
1890 : :
1891 : 1871 : return file;
1892 : : }
1893 : :
1894 : : /*
1895 : : * Open a file that was created with PathNameCreateTemporaryFile, possibly in
1896 : : * another backend. Files opened this way don't count against the
1897 : : * temp_file_limit of the caller, are automatically closed at the end of the
1898 : : * transaction but are not deleted on close.
1899 : : */
1900 : : File
1327 akapila@postgresql.o 1901 : 4975 : PathNameOpenTemporaryFile(const char *path, int mode)
1902 : : {
1903 : : File file;
1904 : :
981 andres@anarazel.de 1905 [ - + ]: 4975 : Assert(temporary_files_allowed); /* check temp file access is up */
1906 : :
158 heikki.linnakangas@i 1907 :GNC 4975 : ResourceOwnerEnlarge(CurrentResourceOwner);
1908 : :
1327 akapila@postgresql.o 1909 :CBC 4975 : file = PathNameOpenFile(path, mode | PG_BINARY);
1910 : :
1911 : : /* If no such file, then we don't raise an error. */
2326 andres@anarazel.de 1912 [ + + - + ]: 4975 : if (file <= 0 && errno != ENOENT)
2326 andres@anarazel.de 1913 [ # # ]:UBC 0 : ereport(ERROR,
1914 : : (errcode_for_file_access(),
1915 : : errmsg("could not open temporary file \"%s\": %m",
1916 : : path)));
1917 : :
2326 andres@anarazel.de 1918 [ + + ]:CBC 4975 : if (file > 0)
1919 : : {
1920 : : /* Register it for automatic close. */
1921 : 2343 : RegisterTemporaryFile(file);
1922 : : }
1923 : :
1924 : 4975 : return file;
1925 : : }
1926 : :
1927 : : /*
1928 : : * Delete a file by pathname. Return true if the file existed, false if
1929 : : * didn't.
1930 : : */
1931 : : bool
1932 : 4123 : PathNameDeleteTemporaryFile(const char *path, bool error_on_failure)
1933 : : {
1934 : : struct stat filestats;
1935 : : int stat_errno;
1936 : :
1937 : : /* Get the final size for pgstat reporting. */
1938 [ + + ]: 4123 : if (stat(path, &filestats) != 0)
1939 : 2252 : stat_errno = errno;
1940 : : else
1941 : 1871 : stat_errno = 0;
1942 : :
1943 : : /*
1944 : : * Unlike FileClose's automatic file deletion code, we tolerate
1945 : : * non-existence to support BufFileDeleteFileSet which doesn't know how
1946 : : * many segments it has to delete until it runs out.
1947 : : */
1948 [ + + ]: 4123 : if (stat_errno == ENOENT)
1949 : 2252 : return false;
1950 : :
1951 [ - + ]: 1871 : if (unlink(path) < 0)
1952 : : {
2326 andres@anarazel.de 1953 [ # # ]:UBC 0 : if (errno != ENOENT)
1954 [ # # # # ]: 0 : ereport(error_on_failure ? ERROR : LOG,
1955 : : (errcode_for_file_access(),
1956 : : errmsg("could not unlink temporary file \"%s\": %m",
1957 : : path)));
1958 : 0 : return false;
1959 : : }
1960 : :
2326 andres@anarazel.de 1961 [ + - ]:CBC 1871 : if (stat_errno == 0)
1962 : 1871 : ReportTemporaryFileUsage(path, filestats.st_size);
1963 : : else
1964 : : {
2326 andres@anarazel.de 1965 :UBC 0 : errno = stat_errno;
1966 [ # # ]: 0 : ereport(LOG,
1967 : : (errcode_for_file_access(),
1968 : : errmsg("could not stat file \"%s\": %m", path)));
1969 : : }
1970 : :
2326 andres@anarazel.de 1971 :CBC 1871 : return true;
1972 : : }
1973 : :
1974 : : /*
1975 : : * close a file when done with it
1976 : : */
1977 : : void
10141 scrappy@hub.org 1978 : 480600 : FileClose(File file)
1979 : : {
1980 : : Vfd *vfdP;
1981 : :
9107 tgl@sss.pgh.pa.us 1982 [ + - + - : 480600 : Assert(FileIsValid(file));
- + ]
1983 : :
1984 : : DO_DB(elog(LOG, "FileClose: %d (%s)",
1985 : : file, VfdCache[file].fileName));
1986 : :
8099 1987 : 480600 : vfdP = &VfdCache[file];
1988 : :
9716 bruce@momjian.us 1989 [ + + ]: 480600 : if (!FileIsNotOpen(file))
1990 : : {
1991 : : /* close the file */
1744 peter@eisentraut.org 1992 [ - + ]: 471526 : if (close(vfdP->fd) != 0)
1993 : : {
1994 : : /*
1995 : : * We may need to panic on failure to close non-temporary files;
1996 : : * see LruDelete.
1997 : : */
1973 tmunro@postgresql.or 1998 [ # # # # ]:UBC 0 : elog(vfdP->fdstate & FD_TEMP_FILE_LIMIT ? LOG : data_sync_elevel(LOG),
1999 : : "could not close file \"%s\": %m", vfdP->fileName);
2000 : : }
2001 : :
9716 bruce@momjian.us 2002 :CBC 471526 : --nfile;
8099 tgl@sss.pgh.pa.us 2003 : 471526 : vfdP->fd = VFD_CLOSED;
2004 : :
2005 : : /* remove the file from the lru ring */
2609 2006 : 471526 : Delete(file);
2007 : : }
2008 : :
2326 andres@anarazel.de 2009 [ + + ]: 480600 : if (vfdP->fdstate & FD_TEMP_FILE_LIMIT)
2010 : : {
2011 : : /* Subtract its size from current usage (do first in case of error) */
2012 : 3798 : temporary_files_size -= vfdP->fileSize;
2013 : 3798 : vfdP->fileSize = 0;
2014 : : }
2015 : :
2016 : : /*
2017 : : * Delete the file if it was temporary, and make a log entry if wanted
2018 : : */
2019 [ + + ]: 480600 : if (vfdP->fdstate & FD_DELETE_AT_CLOSE)
2020 : : {
2021 : : struct stat filestats;
2022 : : int stat_errno;
2023 : :
2024 : : /*
2025 : : * If we get an error, as could happen within the ereport/elog calls,
2026 : : * we'll come right back here during transaction abort. Reset the
2027 : : * flag to ensure that we can't get into an infinite loop. This code
2028 : : * is arranged to ensure that the worst-case consequence is failing to
2029 : : * emit log message(s), not failing to attempt the unlink.
2030 : : */
2031 : 1927 : vfdP->fdstate &= ~FD_DELETE_AT_CLOSE;
2032 : :
2033 : :
2034 : : /* first try the stat() */
4462 magnus@hagander.net 2035 [ - + ]: 1927 : if (stat(vfdP->fileName, &filestats))
4462 magnus@hagander.net 2036 :UBC 0 : stat_errno = errno;
2037 : : else
4462 magnus@hagander.net 2038 :CBC 1927 : stat_errno = 0;
2039 : :
2040 : : /* in any case do the unlink */
2041 [ - + ]: 1927 : if (unlink(vfdP->fileName))
1227 peter@eisentraut.org 2042 [ # # ]:UBC 0 : ereport(LOG,
2043 : : (errcode_for_file_access(),
2044 : : errmsg("could not delete file \"%s\": %m", vfdP->fileName)));
2045 : :
2046 : : /* and last report the stat results */
4462 magnus@hagander.net 2047 [ + - ]:CBC 1927 : if (stat_errno == 0)
2326 andres@anarazel.de 2048 : 1927 : ReportTemporaryFileUsage(vfdP->fileName, filestats.st_size);
2049 : : else
2050 : : {
4460 magnus@hagander.net 2051 :UBC 0 : errno = stat_errno;
1227 peter@eisentraut.org 2052 [ # # ]: 0 : ereport(LOG,
2053 : : (errcode_for_file_access(),
2054 : : errmsg("could not stat file \"%s\": %m", vfdP->fileName)));
2055 : : }
2056 : : }
2057 : :
2058 : : /* Unregister it from the resource owner */
5246 heikki.linnakangas@i 2059 [ + + ]:CBC 480600 : if (vfdP->resowner)
2060 : 6131 : ResourceOwnerForgetFile(vfdP->resowner, file);
2061 : :
2062 : : /*
2063 : : * Return the Vfd slot to the free list
2064 : : */
9107 tgl@sss.pgh.pa.us 2065 : 480600 : FreeVfd(file);
10141 scrappy@hub.org 2066 : 480600 : }
2067 : :
2068 : : /*
2069 : : * FilePrefetch - initiate asynchronous read of a given range of the file.
2070 : : *
2071 : : * Currently the only implementation of this function is using posix_fadvise
2072 : : * which is the simplest standardized interface that accomplishes this.
2073 : : * We could add an implementation using libaio in the future; but note that
2074 : : * this API is inappropriate for libaio, which wants to have a buffer provided
2075 : : * to read into.
2076 : : */
2077 : : int
493 peter@eisentraut.org 2078 : 137257 : FilePrefetch(File file, off_t offset, off_t amount, uint32 wait_event_info)
2079 : : {
2080 : : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_WILLNEED)
2081 : : int returnCode;
2082 : :
5571 tgl@sss.pgh.pa.us 2083 [ + - + - : 137257 : Assert(FileIsValid(file));
- + ]
2084 : :
2085 : : DO_DB(elog(LOG, "FilePrefetch: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2086 : : file, VfdCache[file].fileName,
2087 : : (int64) offset, (int64) amount));
2088 : :
2089 : 137257 : returnCode = FileAccess(file);
2090 [ + - ]: 137257 : if (returnCode < 0)
5571 tgl@sss.pgh.pa.us 2091 :UBC 0 : return returnCode;
2092 : :
300 andres@anarazel.de 2093 :CBC 137257 : retry:
2584 rhaas@postgresql.org 2094 : 137257 : pgstat_report_wait_start(wait_event_info);
5571 tgl@sss.pgh.pa.us 2095 : 137257 : returnCode = posix_fadvise(VfdCache[file].fd, offset, amount,
2096 : : POSIX_FADV_WILLNEED);
2584 rhaas@postgresql.org 2097 : 137257 : pgstat_report_wait_end();
2098 : :
300 andres@anarazel.de 2099 [ - + ]: 137257 : if (returnCode == EINTR)
300 andres@anarazel.de 2100 :UBC 0 : goto retry;
2101 : :
5571 tgl@sss.pgh.pa.us 2102 :CBC 137257 : return returnCode;
2103 : : #else
2104 : : Assert(FileIsValid(file));
2105 : : return 0;
2106 : : #endif
2107 : : }
2108 : :
2109 : : void
2584 rhaas@postgresql.org 2110 : 85187 : FileWriteback(File file, off_t offset, off_t nbytes, uint32 wait_event_info)
2111 : : {
2112 : : int returnCode;
2113 : :
2977 andres@anarazel.de 2114 [ + - + - : 85187 : Assert(FileIsValid(file));
- + ]
2115 : :
2116 : : DO_DB(elog(LOG, "FileWriteback: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2117 : : file, VfdCache[file].fileName,
2118 : : (int64) offset, (int64) nbytes));
2119 : :
2923 tgl@sss.pgh.pa.us 2120 [ - + ]: 85187 : if (nbytes <= 0)
2923 tgl@sss.pgh.pa.us 2121 :UBC 0 : return;
2122 : :
372 tmunro@postgresql.or 2123 [ - + ]:CBC 85187 : if (VfdCache[file].fileFlags & PG_O_DIRECT)
372 tmunro@postgresql.or 2124 :UBC 0 : return;
2125 : :
2977 andres@anarazel.de 2126 :CBC 85187 : returnCode = FileAccess(file);
2127 [ - + ]: 85187 : if (returnCode < 0)
2977 andres@anarazel.de 2128 :UBC 0 : return;
2129 : :
2584 rhaas@postgresql.org 2130 :CBC 85187 : pgstat_report_wait_start(wait_event_info);
2923 tgl@sss.pgh.pa.us 2131 : 85187 : pg_flush_data(VfdCache[file].fd, offset, nbytes);
2584 rhaas@postgresql.org 2132 : 85187 : pgstat_report_wait_end();
2133 : : }
2134 : :
2135 : : ssize_t
124 tmunro@postgresql.or 2136 :GNC 1505916 : FileReadV(File file, const struct iovec *iov, int iovcnt, off_t offset,
2137 : : uint32 wait_event_info)
2138 : : {
2139 : : ssize_t returnCode;
2140 : : Vfd *vfdP;
2141 : :
9107 tgl@sss.pgh.pa.us 2142 [ + - + - :CBC 1505916 : Assert(FileIsValid(file));
- + ]
2143 : :
2144 : : DO_DB(elog(LOG, "FileReadV: %d (%s) " INT64_FORMAT " %d",
2145 : : file, VfdCache[file].fileName,
2146 : : (int64) offset,
2147 : : iovcnt));
2148 : :
7258 2149 : 1505916 : returnCode = FileAccess(file);
2150 [ - + ]: 1505916 : if (returnCode < 0)
7258 tgl@sss.pgh.pa.us 2151 :UBC 0 : return returnCode;
2152 : :
2609 tgl@sss.pgh.pa.us 2153 :CBC 1505916 : vfdP = &VfdCache[file];
2154 : :
6709 2155 : 1505916 : retry:
2584 rhaas@postgresql.org 2156 : 1505916 : pgstat_report_wait_start(wait_event_info);
124 tmunro@postgresql.or 2157 :GNC 1505916 : returnCode = pg_preadv(vfdP->fd, iov, iovcnt, offset);
2584 rhaas@postgresql.org 2158 :CBC 1505916 : pgstat_report_wait_end();
2159 : :
1985 tmunro@postgresql.or 2160 [ - + ]: 1505916 : if (returnCode < 0)
2161 : : {
2162 : : /*
2163 : : * Windows may run out of kernel buffers and return "Insufficient
2164 : : * system resources" error. Wait a bit and retry to solve it.
2165 : : *
2166 : : * It is rumored that EINTR is also possible on some Unix filesystems,
2167 : : * in which case immediate retry is indicated.
2168 : : */
2169 : : #ifdef WIN32
2170 : : DWORD error = GetLastError();
2171 : :
2172 : : switch (error)
2173 : : {
2174 : : case ERROR_NO_SYSTEM_RESOURCES:
2175 : : pg_usleep(1000L);
2176 : : errno = EINTR;
2177 : : break;
2178 : : default:
2179 : : _dosmaperr(error);
2180 : : break;
2181 : : }
2182 : : #endif
2183 : : /* OK to retry if interrupted */
6709 tgl@sss.pgh.pa.us 2184 [ # # ]:UBC 0 : if (errno == EINTR)
2185 : 0 : goto retry;
2186 : : }
2187 : :
9716 bruce@momjian.us 2188 :CBC 1505916 : return returnCode;
2189 : : }
2190 : :
2191 : : ssize_t
124 tmunro@postgresql.or 2192 :GNC 705444 : FileWriteV(File file, const struct iovec *iov, int iovcnt, off_t offset,
2193 : : uint32 wait_event_info)
2194 : : {
2195 : : ssize_t returnCode;
2196 : : Vfd *vfdP;
2197 : :
9107 tgl@sss.pgh.pa.us 2198 [ + - + - :CBC 705444 : Assert(FileIsValid(file));
- + ]
2199 : :
2200 : : DO_DB(elog(LOG, "FileWriteV: %d (%s) " INT64_FORMAT " %d",
2201 : : file, VfdCache[file].fileName,
2202 : : (int64) offset,
2203 : : iovcnt));
2204 : :
7258 2205 : 705444 : returnCode = FileAccess(file);
2206 [ - + ]: 705444 : if (returnCode < 0)
7258 tgl@sss.pgh.pa.us 2207 :UBC 0 : return returnCode;
2208 : :
2609 tgl@sss.pgh.pa.us 2209 :CBC 705444 : vfdP = &VfdCache[file];
2210 : :
2211 : : /*
2212 : : * If enforcing temp_file_limit and it's a temp file, check to see if the
2213 : : * write would overrun temp_file_limit, and throw error if so. Note: it's
2214 : : * really a modularity violation to throw error here; we should set errno
2215 : : * and return -1. However, there's no way to report a suitable error
2216 : : * message if we do that. All current callers would just throw error
2217 : : * immediately anyway, so this is safe at present.
2218 : : */
2326 andres@anarazel.de 2219 [ + - - - ]: 705444 : if (temp_file_limit >= 0 && (vfdP->fdstate & FD_TEMP_FILE_LIMIT))
2220 : : {
124 tmunro@postgresql.or 2221 :UNC 0 : off_t past_write = offset;
2222 : :
2223 [ # # ]: 0 : for (int i = 0; i < iovcnt; ++i)
2224 : 0 : past_write += iov[i].iov_len;
2225 : :
1985 tmunro@postgresql.or 2226 [ # # ]:UBC 0 : if (past_write > vfdP->fileSize)
2227 : : {
4326 bruce@momjian.us 2228 : 0 : uint64 newTotal = temporary_files_size;
2229 : :
1985 tmunro@postgresql.or 2230 : 0 : newTotal += past_write - vfdP->fileSize;
4655 tgl@sss.pgh.pa.us 2231 [ # # ]: 0 : if (newTotal > (uint64) temp_file_limit * (uint64) 1024)
2232 [ # # ]: 0 : ereport(ERROR,
2233 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
2234 : : errmsg("temporary file size exceeds temp_file_limit (%dkB)",
2235 : : temp_file_limit)));
2236 : : }
2237 : : }
2238 : :
6709 tgl@sss.pgh.pa.us 2239 :CBC 705444 : retry:
2584 rhaas@postgresql.org 2240 : 705444 : pgstat_report_wait_start(wait_event_info);
124 tmunro@postgresql.or 2241 :GNC 705444 : returnCode = pg_pwritev(vfdP->fd, iov, iovcnt, offset);
2584 rhaas@postgresql.org 2242 :CBC 705444 : pgstat_report_wait_end();
2243 : :
6709 tgl@sss.pgh.pa.us 2244 [ + - ]: 705444 : if (returnCode >= 0)
2245 : : {
2246 : : /*
2247 : : * Some callers expect short writes to set errno, and traditionally we
2248 : : * have assumed that they imply disk space shortage. We don't want to
2249 : : * waste CPU cycles adding up the total size here, so we'll just set
2250 : : * it for all successful writes in case such a caller determines that
2251 : : * the write was short and ereports "%m".
2252 : : */
124 tmunro@postgresql.or 2253 :GNC 705444 : errno = ENOSPC;
2254 : :
2255 : : /*
2256 : : * Maintain fileSize and temporary_files_size if it's a temp file.
2257 : : */
2326 andres@anarazel.de 2258 [ + + ]:CBC 705444 : if (vfdP->fdstate & FD_TEMP_FILE_LIMIT)
2259 : : {
124 tmunro@postgresql.or 2260 :GNC 58649 : off_t past_write = offset + returnCode;
2261 : :
1985 tmunro@postgresql.or 2262 [ + + ]:CBC 58649 : if (past_write > vfdP->fileSize)
2263 : : {
2264 : 42536 : temporary_files_size += past_write - vfdP->fileSize;
2265 : 42536 : vfdP->fileSize = past_write;
2266 : : }
2267 : : }
2268 : : }
2269 : : else
2270 : : {
2271 : : /*
2272 : : * See comments in FileReadV()
2273 : : */
2274 : : #ifdef WIN32
2275 : : DWORD error = GetLastError();
2276 : :
2277 : : switch (error)
2278 : : {
2279 : : case ERROR_NO_SYSTEM_RESOURCES:
2280 : : pg_usleep(1000L);
2281 : : errno = EINTR;
2282 : : break;
2283 : : default:
2284 : : _dosmaperr(error);
2285 : : break;
2286 : : }
2287 : : #endif
2288 : : /* OK to retry if interrupted */
6709 tgl@sss.pgh.pa.us 2289 [ # # ]:UBC 0 : if (errno == EINTR)
2290 : 0 : goto retry;
2291 : : }
2292 : :
9716 bruce@momjian.us 2293 :CBC 705444 : return returnCode;
2294 : : }
2295 : :
2296 : : int
2584 rhaas@postgresql.org 2297 : 107 : FileSync(File file, uint32 wait_event_info)
2298 : : {
2299 : : int returnCode;
2300 : :
7258 tgl@sss.pgh.pa.us 2301 [ + - + - : 107 : Assert(FileIsValid(file));
- + ]
2302 : :
2303 : : DO_DB(elog(LOG, "FileSync: %d (%s)",
2304 : : file, VfdCache[file].fileName));
2305 : :
2306 : 107 : returnCode = FileAccess(file);
2307 [ - + ]: 107 : if (returnCode < 0)
7258 tgl@sss.pgh.pa.us 2308 :UBC 0 : return returnCode;
2309 : :
2584 rhaas@postgresql.org 2310 :CBC 107 : pgstat_report_wait_start(wait_event_info);
2311 : 107 : returnCode = pg_fsync(VfdCache[file].fd);
2312 : 107 : pgstat_report_wait_end();
2313 : :
2314 : 107 : return returnCode;
2315 : : }
2316 : :
2317 : : /*
2318 : : * Zero a region of the file.
2319 : : *
2320 : : * Returns 0 on success, -1 otherwise. In the latter case errno is set to the
2321 : : * appropriate error.
2322 : : */
2323 : : int
375 andres@anarazel.de 2324 : 195275 : FileZero(File file, off_t offset, off_t amount, uint32 wait_event_info)
2325 : : {
2326 : : int returnCode;
2327 : : ssize_t written;
2328 : :
2329 [ + - + - : 195275 : Assert(FileIsValid(file));
- + ]
2330 : :
2331 : : DO_DB(elog(LOG, "FileZero: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2332 : : file, VfdCache[file].fileName,
2333 : : (int64) offset, (int64) amount));
2334 : :
2335 : 195275 : returnCode = FileAccess(file);
2336 [ - + ]: 195275 : if (returnCode < 0)
375 andres@anarazel.de 2337 :UBC 0 : return returnCode;
2338 : :
375 andres@anarazel.de 2339 :CBC 195275 : pgstat_report_wait_start(wait_event_info);
2340 : 195275 : written = pg_pwrite_zeros(VfdCache[file].fd, amount, offset);
2341 : 195275 : pgstat_report_wait_end();
2342 : :
2343 [ - + ]: 195275 : if (written < 0)
375 andres@anarazel.de 2344 :UBC 0 : return -1;
375 andres@anarazel.de 2345 [ - + ]:CBC 195275 : else if (written != amount)
2346 : : {
2347 : : /* if errno is unset, assume problem is no disk space */
375 andres@anarazel.de 2348 [ # # ]:UBC 0 : if (errno == 0)
2349 : 0 : errno = ENOSPC;
2350 : 0 : return -1;
2351 : : }
2352 : :
375 andres@anarazel.de 2353 :CBC 195275 : return 0;
2354 : : }
2355 : :
2356 : : /*
2357 : : * Try to reserve file space with posix_fallocate(). If posix_fallocate() is
2358 : : * not implemented on the operating system or fails with EINVAL / EOPNOTSUPP,
2359 : : * use FileZero() instead.
2360 : : *
2361 : : * Note that at least glibc() implements posix_fallocate() in userspace if not
2362 : : * implemented by the filesystem. That's not the case for all environments
2363 : : * though.
2364 : : *
2365 : : * Returns 0 on success, -1 otherwise. In the latter case errno is set to the
2366 : : * appropriate error.
2367 : : */
2368 : : int
2369 : 509 : FileFallocate(File file, off_t offset, off_t amount, uint32 wait_event_info)
2370 : : {
2371 : : #ifdef HAVE_POSIX_FALLOCATE
2372 : : int returnCode;
2373 : :
2374 [ + - + - : 509 : Assert(FileIsValid(file));
- + ]
2375 : :
2376 : : DO_DB(elog(LOG, "FileFallocate: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2377 : : file, VfdCache[file].fileName,
2378 : : (int64) offset, (int64) amount));
2379 : :
2380 : 509 : returnCode = FileAccess(file);
2381 [ + - ]: 509 : if (returnCode < 0)
375 andres@anarazel.de 2382 :UBC 0 : return -1;
2383 : :
300 andres@anarazel.de 2384 :CBC 509 : retry:
375 2385 : 509 : pgstat_report_wait_start(wait_event_info);
2386 : 509 : returnCode = posix_fallocate(VfdCache[file].fd, offset, amount);
2387 : 509 : pgstat_report_wait_end();
2388 : :
2389 [ + - ]: 509 : if (returnCode == 0)
2390 : 509 : return 0;
300 andres@anarazel.de 2391 [ # # ]:UBC 0 : else if (returnCode == EINTR)
2392 : 0 : goto retry;
2393 : :
2394 : : /* for compatibility with %m printing etc */
375 2395 : 0 : errno = returnCode;
2396 : :
2397 : : /*
2398 : : * Return in cases of a "real" failure, if fallocate is not supported,
2399 : : * fall through to the FileZero() backed implementation.
2400 : : */
2401 [ # # # # ]: 0 : if (returnCode != EINVAL && returnCode != EOPNOTSUPP)
2402 : 0 : return -1;
2403 : : #endif
2404 : :
2405 : 0 : return FileZero(file, offset, amount, wait_event_info);
2406 : : }
2407 : :
2408 : : off_t
1985 tmunro@postgresql.or 2409 :CBC 2946535 : FileSize(File file)
2410 : : {
9107 tgl@sss.pgh.pa.us 2411 [ + - + - : 2946535 : Assert(FileIsValid(file));
- + ]
2412 : :
2413 : : DO_DB(elog(LOG, "FileSize %d (%s)",
2414 : : file, VfdCache[file].fileName));
2415 : :
9716 bruce@momjian.us 2416 [ + + ]: 2946535 : if (FileIsNotOpen(file))
2417 : : {
1985 tmunro@postgresql.or 2418 [ - + ]: 10184 : if (FileAccess(file) < 0)
1985 tmunro@postgresql.or 2419 :UBC 0 : return (off_t) -1;
2420 : : }
2421 : :
1985 tmunro@postgresql.or 2422 :CBC 2946535 : return lseek(VfdCache[file].fd, 0, SEEK_END);
2423 : : }
2424 : :
2425 : : int
2584 rhaas@postgresql.org 2426 : 520 : FileTruncate(File file, off_t offset, uint32 wait_event_info)
2427 : : {
2428 : : int returnCode;
2429 : :
9107 tgl@sss.pgh.pa.us 2430 [ + - + - : 520 : Assert(FileIsValid(file));
- + ]
2431 : :
2432 : : DO_DB(elog(LOG, "FileTruncate %d (%s)",
2433 : : file, VfdCache[file].fileName));
2434 : :
7258 2435 : 520 : returnCode = FileAccess(file);
2436 [ - + ]: 520 : if (returnCode < 0)
7258 tgl@sss.pgh.pa.us 2437 :UBC 0 : return returnCode;
2438 : :
2584 rhaas@postgresql.org 2439 :CBC 520 : pgstat_report_wait_start(wait_event_info);
300 andres@anarazel.de 2440 : 520 : returnCode = pg_ftruncate(VfdCache[file].fd, offset);
2584 rhaas@postgresql.org 2441 : 520 : pgstat_report_wait_end();
2442 : :
4655 tgl@sss.pgh.pa.us 2443 [ + - - + ]: 520 : if (returnCode == 0 && VfdCache[file].fileSize > offset)
2444 : : {
2445 : : /* adjust our state for truncation of a temp file */
2326 andres@anarazel.de 2446 [ # # ]:UBC 0 : Assert(VfdCache[file].fdstate & FD_TEMP_FILE_LIMIT);
4655 tgl@sss.pgh.pa.us 2447 : 0 : temporary_files_size -= VfdCache[file].fileSize - offset;
2448 : 0 : VfdCache[file].fileSize = offset;
2449 : : }
2450 : :
9357 bruce@momjian.us 2451 :CBC 520 : return returnCode;
2452 : : }
2453 : :
2454 : : /*
2455 : : * Return the pathname associated with an open file.
2456 : : *
2457 : : * The returned string points to an internal buffer, which is valid until
2458 : : * the file is closed.
2459 : : */
2460 : : char *
5366 heikki.linnakangas@i 2461 :GBC 12 : FilePathName(File file)
2462 : : {
2463 [ + - + - : 12 : Assert(FileIsValid(file));
- + ]
2464 : :
2465 : 12 : return VfdCache[file].fileName;
2466 : : }
2467 : :
2468 : : /*
2469 : : * Return the raw file descriptor of an opened file.
2470 : : *
2471 : : * The returned file descriptor will be valid until the file is closed, but
2472 : : * there are a lot of things that can make that happen. So the caller should
2473 : : * be careful not to do much of anything else before it finishes using the
2474 : : * returned file descriptor.
2475 : : */
2476 : : int
2959 rhaas@postgresql.org 2477 :UBC 0 : FileGetRawDesc(File file)
2478 : : {
2479 [ # # # # : 0 : Assert(FileIsValid(file));
# # ]
2480 : 0 : return VfdCache[file].fd;
2481 : : }
2482 : :
2483 : : /*
2484 : : * FileGetRawFlags - returns the file flags on open(2)
2485 : : */
2486 : : int
2487 : 0 : FileGetRawFlags(File file)
2488 : : {
2489 [ # # # # : 0 : Assert(FileIsValid(file));
# # ]
2490 : 0 : return VfdCache[file].fileFlags;
2491 : : }
2492 : :
2493 : : /*
2494 : : * FileGetRawMode - returns the mode bitmask passed to open(2)
2495 : : */
2496 : : mode_t
2497 : 0 : FileGetRawMode(File file)
2498 : : {
2499 [ # # # # : 0 : Assert(FileIsValid(file));
# # ]
2500 : 0 : return VfdCache[file].fileMode;
2501 : : }
2502 : :
2503 : : /*
2504 : : * Make room for another allocatedDescs[] array entry if needed and possible.
2505 : : * Returns true if an array element is available.
2506 : : */
2507 : : static bool
3962 tgl@sss.pgh.pa.us 2508 :CBC 7757383 : reserveAllocatedDesc(void)
2509 : : {
2510 : : AllocateDesc *newDescs;
2511 : : int newMax;
2512 : :
2513 : : /* Quick out if array already has a free slot. */
2514 [ + + ]: 7757383 : if (numAllocatedDescs < maxAllocatedDescs)
2515 : 7756479 : return true;
2516 : :
2517 : : /*
2518 : : * If the array hasn't yet been created in the current process, initialize
2519 : : * it with FD_MINFREE / 3 elements. In many scenarios this is as many as
2520 : : * we will ever need, anyway. We don't want to look at max_safe_fds
2521 : : * immediately because set_max_safe_fds() may not have run yet.
2522 : : */
2523 [ + - ]: 904 : if (allocatedDescs == NULL)
2524 : : {
1511 2525 : 904 : newMax = FD_MINFREE / 3;
3962 2526 : 904 : newDescs = (AllocateDesc *) malloc(newMax * sizeof(AllocateDesc));
2527 : : /* Out of memory already? Treat as fatal error. */
2528 [ - + ]: 904 : if (newDescs == NULL)
3962 tgl@sss.pgh.pa.us 2529 [ # # ]:UBC 0 : ereport(ERROR,
2530 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2531 : : errmsg("out of memory")));
3962 tgl@sss.pgh.pa.us 2532 :CBC 904 : allocatedDescs = newDescs;
2533 : 904 : maxAllocatedDescs = newMax;
2534 : 904 : return true;
2535 : : }
2536 : :
2537 : : /*
2538 : : * Consider enlarging the array beyond the initial allocation used above.
2539 : : * By the time this happens, max_safe_fds should be known accurately.
2540 : : *
2541 : : * We mustn't let allocated descriptors hog all the available FDs, and in
2542 : : * practice we'd better leave a reasonable number of FDs for VFD use. So
2543 : : * set the maximum to max_safe_fds / 3. (This should certainly be at
2544 : : * least as large as the initial size, FD_MINFREE / 3, so we aren't
2545 : : * tightening the restriction here.) Recall that "external" FDs are
2546 : : * allowed to consume another third of max_safe_fds.
2547 : : */
1511 tgl@sss.pgh.pa.us 2548 :UBC 0 : newMax = max_safe_fds / 3;
3962 2549 [ # # ]: 0 : if (newMax > maxAllocatedDescs)
2550 : : {
2551 : 0 : newDescs = (AllocateDesc *) realloc(allocatedDescs,
2552 : : newMax * sizeof(AllocateDesc));
2553 : : /* Treat out-of-memory as a non-fatal error. */
2554 [ # # ]: 0 : if (newDescs == NULL)
2555 : 0 : return false;
2556 : 0 : allocatedDescs = newDescs;
2557 : 0 : maxAllocatedDescs = newMax;
2558 : 0 : return true;
2559 : : }
2560 : :
2561 : : /* Can't enlarge allocatedDescs[] any more. */
2562 : 0 : return false;
2563 : : }
2564 : :
2565 : : /*
2566 : : * Routines that want to use stdio (ie, FILE*) should use AllocateFile
2567 : : * rather than plain fopen(). This lets fd.c deal with freeing FDs if
2568 : : * necessary to open the file. When done, call FreeFile rather than fclose.
2569 : : *
2570 : : * Note that files that will be open for any significant length of time
2571 : : * should NOT be handled this way, since they cannot share kernel file
2572 : : * descriptors with other files; there is grave risk of running out of FDs
2573 : : * if anyone locks down too many FDs. Most callers of this routine are
2574 : : * simply reading a config file that they will read and close immediately.
2575 : : *
2576 : : * fd.c will automatically close all files opened with AllocateFile at
2577 : : * transaction commit or abort; this prevents FD leakage if a routine
2578 : : * that calls AllocateFile is terminated prematurely by ereport(ERROR).
2579 : : *
2580 : : * Ideally this should be the *only* direct call of fopen() in the backend.
2581 : : */
2582 : : FILE *
6616 tgl@sss.pgh.pa.us 2583 :CBC 70612 : AllocateFile(const char *name, const char *mode)
2584 : : {
2585 : : FILE *file;
2586 : :
2587 : : DO_DB(elog(LOG, "AllocateFile: Allocated %d (%s)",
2588 : : numAllocatedDescs, name));
2589 : :
2590 : : /* Can we allocate another non-virtual FD? */
3962 2591 [ - + ]: 70612 : if (!reserveAllocatedDesc())
3962 tgl@sss.pgh.pa.us 2592 [ # # ]:UBC 0 : ereport(ERROR,
2593 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2594 : : errmsg("exceeded maxAllocatedDescs (%d) while trying to open file \"%s\"",
2595 : : maxAllocatedDescs, name)));
2596 : :
2597 : : /* Close excess kernel FDs. */
3962 tgl@sss.pgh.pa.us 2598 :CBC 70612 : ReleaseLruFiles();
2599 : :
9736 bruce@momjian.us 2600 : 70612 : TryAgain:
8631 tgl@sss.pgh.pa.us 2601 [ + + ]: 70612 : if ((file = fopen(name, mode)) != NULL)
2602 : : {
7200 2603 : 63918 : AllocateDesc *desc = &allocatedDescs[numAllocatedDescs];
2604 : :
2605 : 63918 : desc->kind = AllocateDescFile;
2606 : 63918 : desc->desc.file = file;
7150 2607 : 63918 : desc->create_subid = GetCurrentSubTransactionId();
7200 2608 : 63918 : numAllocatedDescs++;
2609 : 63918 : return desc->desc.file;
2610 : : }
2611 : :
8631 2612 [ + - - + ]: 6694 : if (errno == EMFILE || errno == ENFILE)
2613 : : {
8424 bruce@momjian.us 2614 :UBC 0 : int save_errno = errno;
2615 : :
7570 tgl@sss.pgh.pa.us 2616 [ # # ]: 0 : ereport(LOG,
2617 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2618 : : errmsg("out of file descriptors: %m; release and retry")));
8631 2619 : 0 : errno = 0;
2620 [ # # ]: 0 : if (ReleaseLruFile())
9716 bruce@momjian.us 2621 : 0 : goto TryAgain;
8631 tgl@sss.pgh.pa.us 2622 : 0 : errno = save_errno;
2623 : : }
2624 : :
8631 tgl@sss.pgh.pa.us 2625 :CBC 6694 : return NULL;
2626 : : }
2627 : :
2628 : : /*
2629 : : * Open a file with OpenTransientFilePerm() and pass default file mode for
2630 : : * the fileMode parameter.
2631 : : */
2632 : : int
2395 peter_e@gmx.net 2633 : 7651102 : OpenTransientFile(const char *fileName, int fileFlags)
2634 : : {
2199 sfrost@snowman.net 2635 : 7651102 : return OpenTransientFilePerm(fileName, fileFlags, pg_file_create_mode);
2636 : : }
2637 : :
2638 : : /*
2639 : : * Like AllocateFile, but returns an unbuffered fd like open(2)
2640 : : */
2641 : : int
2395 peter_e@gmx.net 2642 : 7651108 : OpenTransientFilePerm(const char *fileName, int fileFlags, mode_t fileMode)
2643 : : {
2644 : : int fd;
2645 : :
2646 : : DO_DB(elog(LOG, "OpenTransientFile: Allocated %d (%s)",
2647 : : numAllocatedDescs, fileName));
2648 : :
2649 : : /* Can we allocate another non-virtual FD? */
3962 tgl@sss.pgh.pa.us 2650 [ - + ]: 7651108 : if (!reserveAllocatedDesc())
3962 tgl@sss.pgh.pa.us 2651 [ # # ]:UBC 0 : ereport(ERROR,
2652 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2653 : : errmsg("exceeded maxAllocatedDescs (%d) while trying to open file \"%s\"",
2654 : : maxAllocatedDescs, fileName)));
2655 : :
2656 : : /* Close excess kernel FDs. */
3962 tgl@sss.pgh.pa.us 2657 :CBC 7651108 : ReleaseLruFiles();
2658 : :
2395 peter_e@gmx.net 2659 : 7651108 : fd = BasicOpenFilePerm(fileName, fileFlags, fileMode);
2660 : :
4156 heikki.linnakangas@i 2661 [ + + ]: 7651108 : if (fd >= 0)
2662 : : {
2663 : 7649039 : AllocateDesc *desc = &allocatedDescs[numAllocatedDescs];
2664 : :
2665 : 7649039 : desc->kind = AllocateDescRawFD;
2666 : 7649039 : desc->desc.fd = fd;
2667 : 7649039 : desc->create_subid = GetCurrentSubTransactionId();
2668 : 7649039 : numAllocatedDescs++;
2669 : :
2670 : 7649039 : return fd;
2671 : : }
2672 : :
2673 : 2069 : return -1; /* failure */
2674 : : }
2675 : :
2676 : : /*
2677 : : * Routines that want to initiate a pipe stream should use OpenPipeStream
2678 : : * rather than plain popen(). This lets fd.c deal with freeing FDs if
2679 : : * necessary. When done, call ClosePipeStream rather than pclose.
2680 : : *
2681 : : * This function also ensures that the popen'd program is run with default
2682 : : * SIGPIPE processing, rather than the SIG_IGN setting the backend normally
2683 : : * uses. This ensures desirable response to, eg, closing a read pipe early.
2684 : : */
2685 : : FILE *
4064 2686 : 45 : OpenPipeStream(const char *command, const char *mode)
2687 : : {
2688 : : FILE *file;
2689 : : int save_errno;
2690 : :
2691 : : DO_DB(elog(LOG, "OpenPipeStream: Allocated %d (%s)",
2692 : : numAllocatedDescs, command));
2693 : :
2694 : : /* Can we allocate another non-virtual FD? */
3962 tgl@sss.pgh.pa.us 2695 [ - + ]: 45 : if (!reserveAllocatedDesc())
3962 tgl@sss.pgh.pa.us 2696 [ # # ]:UBC 0 : ereport(ERROR,
2697 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2698 : : errmsg("exceeded maxAllocatedDescs (%d) while trying to execute command \"%s\"",
2699 : : maxAllocatedDescs, command)));
2700 : :
2701 : : /* Close excess kernel FDs. */
3962 tgl@sss.pgh.pa.us 2702 :CBC 45 : ReleaseLruFiles();
2703 : :
4064 heikki.linnakangas@i 2704 : 45 : TryAgain:
594 tgl@sss.pgh.pa.us 2705 : 45 : fflush(NULL);
1973 2706 : 45 : pqsignal(SIGPIPE, SIG_DFL);
4064 heikki.linnakangas@i 2707 : 45 : errno = 0;
1973 tgl@sss.pgh.pa.us 2708 : 45 : file = popen(command, mode);
2709 : 45 : save_errno = errno;
2710 : 45 : pqsignal(SIGPIPE, SIG_IGN);
2711 : 45 : errno = save_errno;
2712 [ + - ]: 45 : if (file != NULL)
2713 : : {
4064 heikki.linnakangas@i 2714 : 45 : AllocateDesc *desc = &allocatedDescs[numAllocatedDescs];
2715 : :
2716 : 45 : desc->kind = AllocateDescPipe;
2717 : 45 : desc->desc.file = file;
2718 : 45 : desc->create_subid = GetCurrentSubTransactionId();
2719 : 45 : numAllocatedDescs++;
2720 : 45 : return desc->desc.file;
2721 : : }
2722 : :
4064 heikki.linnakangas@i 2723 [ # # # # ]:UBC 0 : if (errno == EMFILE || errno == ENFILE)
2724 : : {
2725 [ # # ]: 0 : ereport(LOG,
2726 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2727 : : errmsg("out of file descriptors: %m; release and retry")));
2728 [ # # ]: 0 : if (ReleaseLruFile())
2729 : 0 : goto TryAgain;
2730 : 0 : errno = save_errno;
2731 : : }
2732 : :
2733 : 0 : return NULL;
2734 : : }
2735 : :
2736 : : /*
2737 : : * Free an AllocateDesc of any type.
2738 : : *
2739 : : * The argument *must* point into the allocatedDescs[] array.
2740 : : */
2741 : : static int
7200 tgl@sss.pgh.pa.us 2742 :CBC 7747826 : FreeDesc(AllocateDesc *desc)
2743 : : {
2744 : : int result;
2745 : :
2746 : : /* Close the underlying object */
2747 [ + + + + : 7747826 : switch (desc->kind)
- ]
2748 : : {
2749 : 63917 : case AllocateDescFile:
2750 : 63917 : result = fclose(desc->desc.file);
2751 : 63917 : break;
4064 heikki.linnakangas@i 2752 : 45 : case AllocateDescPipe:
2753 : 45 : result = pclose(desc->desc.file);
2754 : 45 : break;
7200 tgl@sss.pgh.pa.us 2755 : 34825 : case AllocateDescDir:
2756 : 34825 : result = closedir(desc->desc.dir);
2757 : 34825 : break;
4156 heikki.linnakangas@i 2758 : 7649039 : case AllocateDescRawFD:
2759 : 7649039 : result = close(desc->desc.fd);
2760 : 7649039 : break;
7200 tgl@sss.pgh.pa.us 2761 :UBC 0 : default:
2762 [ # # ]: 0 : elog(ERROR, "AllocateDesc kind not recognized");
2763 : : result = 0; /* keep compiler quiet */
2764 : : break;
2765 : : }
2766 : :
2767 : : /* Compact storage in the allocatedDescs array */
7200 tgl@sss.pgh.pa.us 2768 :CBC 7747826 : numAllocatedDescs--;
2769 : 7747826 : *desc = allocatedDescs[numAllocatedDescs];
2770 : :
2771 : 7747826 : return result;
2772 : : }
2773 : :
2774 : : /*
2775 : : * Close a file returned by AllocateFile.
2776 : : *
2777 : : * Note we do not check fclose's return value --- it is up to the caller
2778 : : * to handle close errors.
2779 : : */
2780 : : int
9715 bruce@momjian.us 2781 : 63908 : FreeFile(FILE *file)
2782 : : {
2783 : : int i;
2784 : :
2785 : : DO_DB(elog(LOG, "FreeFile: Allocated %d", numAllocatedDescs));
2786 : :
2787 : : /* Remove file from list of allocated files, if it's present */
7200 tgl@sss.pgh.pa.us 2788 [ + - ]: 63909 : for (i = numAllocatedDescs; --i >= 0;)
2789 : : {
2790 : 63909 : AllocateDesc *desc = &allocatedDescs[i];
2791 : :
2792 [ + - + + ]: 63909 : if (desc->kind == AllocateDescFile && desc->desc.file == file)
2793 : 63908 : return FreeDesc(desc);
2794 : : }
2795 : :
2796 : : /* Only get here if someone passes us a file not in allocatedDescs */
7200 tgl@sss.pgh.pa.us 2797 [ # # ]:UBC 0 : elog(WARNING, "file passed to FreeFile was not obtained from AllocateFile");
2798 : :
7384 2799 : 0 : return fclose(file);
2800 : : }
2801 : :
2802 : : /*
2803 : : * Close a file returned by OpenTransientFile.
2804 : : *
2805 : : * Note we do not check close's return value --- it is up to the caller
2806 : : * to handle close errors.
2807 : : */
2808 : : int
4156 heikki.linnakangas@i 2809 :CBC 7649038 : CloseTransientFile(int fd)
2810 : : {
2811 : : int i;
2812 : :
2813 : : DO_DB(elog(LOG, "CloseTransientFile: Allocated %d", numAllocatedDescs));
2814 : :
2815 : : /* Remove fd from list of allocated files, if it's present */
2816 [ + - ]: 7649045 : for (i = numAllocatedDescs; --i >= 0;)
2817 : : {
2818 : 7649045 : AllocateDesc *desc = &allocatedDescs[i];
2819 : :
2820 [ + - + + ]: 7649045 : if (desc->kind == AllocateDescRawFD && desc->desc.fd == fd)
2821 : 7649038 : return FreeDesc(desc);
2822 : : }
2823 : :
2824 : : /* Only get here if someone passes us a file not in allocatedDescs */
4156 heikki.linnakangas@i 2825 [ # # ]:UBC 0 : elog(WARNING, "fd passed to CloseTransientFile was not obtained from OpenTransientFile");
2826 : :
2827 : 0 : return close(fd);
2828 : : }
2829 : :
2830 : : /*
2831 : : * Routines that want to use <dirent.h> (ie, DIR*) should use AllocateDir
2832 : : * rather than plain opendir(). This lets fd.c deal with freeing FDs if
2833 : : * necessary to open the directory, and with closing it after an elog.
2834 : : * When done, call FreeDir rather than closedir.
2835 : : *
2836 : : * Returns NULL, with errno set, on failure. Note that failure detection
2837 : : * is commonly left to the following call of ReadDir or ReadDirExtended;
2838 : : * see the comments for ReadDir.
2839 : : *
2840 : : * Ideally this should be the *only* direct call of opendir() in the backend.
2841 : : */
2842 : : DIR *
7356 tgl@sss.pgh.pa.us 2843 :CBC 35618 : AllocateDir(const char *dirname)
2844 : : {
2845 : : DIR *dir;
2846 : :
2847 : : DO_DB(elog(LOG, "AllocateDir: Allocated %d (%s)",
2848 : : numAllocatedDescs, dirname));
2849 : :
2850 : : /* Can we allocate another non-virtual FD? */
3962 2851 [ - + ]: 35618 : if (!reserveAllocatedDesc())
3962 tgl@sss.pgh.pa.us 2852 [ # # ]:UBC 0 : ereport(ERROR,
2853 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2854 : : errmsg("exceeded maxAllocatedDescs (%d) while trying to open directory \"%s\"",
2855 : : maxAllocatedDescs, dirname)));
2856 : :
2857 : : /* Close excess kernel FDs. */
3962 tgl@sss.pgh.pa.us 2858 :CBC 35618 : ReleaseLruFiles();
2859 : :
7356 2860 : 35618 : TryAgain:
2861 [ + + ]: 35618 : if ((dir = opendir(dirname)) != NULL)
2862 : : {
7200 2863 : 34871 : AllocateDesc *desc = &allocatedDescs[numAllocatedDescs];
2864 : :
2865 : 34871 : desc->kind = AllocateDescDir;
2866 : 34871 : desc->desc.dir = dir;
7150 2867 : 34871 : desc->create_subid = GetCurrentSubTransactionId();
7200 2868 : 34871 : numAllocatedDescs++;
2869 : 34871 : return desc->desc.dir;
2870 : : }
2871 : :
7356 2872 [ + - - + ]: 747 : if (errno == EMFILE || errno == ENFILE)
2873 : : {
7356 tgl@sss.pgh.pa.us 2874 :UBC 0 : int save_errno = errno;
2875 : :
2876 [ # # ]: 0 : ereport(LOG,
2877 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2878 : : errmsg("out of file descriptors: %m; release and retry")));
2879 : 0 : errno = 0;
2880 [ # # ]: 0 : if (ReleaseLruFile())
2881 : 0 : goto TryAgain;
2882 : 0 : errno = save_errno;
2883 : : }
2884 : :
7356 tgl@sss.pgh.pa.us 2885 :CBC 747 : return NULL;
2886 : : }
2887 : :
2888 : : /*
2889 : : * Read a directory opened with AllocateDir, ereport'ing any error.
2890 : : *
2891 : : * This is easier to use than raw readdir() since it takes care of some
2892 : : * otherwise rather tedious and error-prone manipulation of errno. Also,
2893 : : * if you are happy with a generic error message for AllocateDir failure,
2894 : : * you can just do
2895 : : *
2896 : : * dir = AllocateDir(path);
2897 : : * while ((dirent = ReadDir(dir, path)) != NULL)
2898 : : * process dirent;
2899 : : * FreeDir(dir);
2900 : : *
2901 : : * since a NULL dir parameter is taken as indicating AllocateDir failed.
2902 : : * (Make sure errno isn't changed between AllocateDir and ReadDir if you
2903 : : * use this shortcut.)
2904 : : *
2905 : : * The pathname passed to AllocateDir must be passed to this routine too,
2906 : : * but it is only used for error reporting.
2907 : : */
2908 : : struct dirent *
6874 2909 : 1871629 : ReadDir(DIR *dir, const char *dirname)
2910 : : {
3244 2911 : 1871629 : return ReadDirExtended(dir, dirname, ERROR);
2912 : : }
2913 : :
2914 : : /*
2915 : : * Alternate version of ReadDir that allows caller to specify the elevel
2916 : : * for any error report (whether it's reporting an initial failure of
2917 : : * AllocateDir or a subsequent directory read failure).
2918 : : *
2919 : : * If elevel < ERROR, returns NULL after any error. With the normal coding
2920 : : * pattern, this will result in falling out of the loop immediately as
2921 : : * though the directory contained no (more) entries.
2922 : : */
2923 : : struct dirent *
2924 : 2862425 : ReadDirExtended(DIR *dir, const char *dirname, int elevel)
2925 : : {
2926 : : struct dirent *dent;
2927 : :
2928 : : /* Give a generic message for AllocateDir failure, if caller didn't */
6874 2929 [ + + ]: 2862425 : if (dir == NULL)
2930 : : {
3244 2931 [ + - ]: 3 : ereport(elevel,
2932 : : (errcode_for_file_access(),
2933 : : errmsg("could not open directory \"%s\": %m",
2934 : : dirname)));
3244 tgl@sss.pgh.pa.us 2935 :UBC 0 : return NULL;
2936 : : }
2937 : :
6874 tgl@sss.pgh.pa.us 2938 :CBC 2862422 : errno = 0;
2939 [ + + ]: 2862422 : if ((dent = readdir(dir)) != NULL)
2940 : 2834320 : return dent;
2941 : :
2942 [ - + ]: 28102 : if (errno)
3244 tgl@sss.pgh.pa.us 2943 [ # # ]:UBC 0 : ereport(elevel,
2944 : : (errcode_for_file_access(),
2945 : : errmsg("could not read directory \"%s\": %m",
2946 : : dirname)));
6874 tgl@sss.pgh.pa.us 2947 :CBC 28102 : return NULL;
2948 : : }
2949 : :
2950 : : /*
2951 : : * Close a directory opened with AllocateDir.
2952 : : *
2953 : : * Returns closedir's return value (with errno set if it's not 0).
2954 : : * Note we do not check the return value --- it is up to the caller
2955 : : * to handle close errors if wanted.
2956 : : *
2957 : : * Does nothing if dir == NULL; we assume that directory open failure was
2958 : : * already reported if desired.
2959 : : */
2960 : : int
7356 2961 : 34713 : FreeDir(DIR *dir)
2962 : : {
2963 : : int i;
2964 : :
2965 : : /* Nothing to do if AllocateDir failed */
2323 2966 [ - + ]: 34713 : if (dir == NULL)
2323 tgl@sss.pgh.pa.us 2967 :UBC 0 : return 0;
2968 : :
2969 : : DO_DB(elog(LOG, "FreeDir: Allocated %d", numAllocatedDescs));
2970 : :
2971 : : /* Remove dir from list of allocated dirs, if it's present */
7200 tgl@sss.pgh.pa.us 2972 [ + - ]:CBC 34713 : for (i = numAllocatedDescs; --i >= 0;)
2973 : : {
2974 : 34713 : AllocateDesc *desc = &allocatedDescs[i];
2975 : :
2976 [ + - + - ]: 34713 : if (desc->kind == AllocateDescDir && desc->desc.dir == dir)
2977 : 34713 : return FreeDesc(desc);
2978 : : }
2979 : :
2980 : : /* Only get here if someone passes us a dir not in allocatedDescs */
7200 tgl@sss.pgh.pa.us 2981 [ # # ]:UBC 0 : elog(WARNING, "dir passed to FreeDir was not obtained from AllocateDir");
2982 : :
7356 2983 : 0 : return closedir(dir);
2984 : : }
2985 : :
2986 : :
2987 : : /*
2988 : : * Close a pipe stream returned by OpenPipeStream.
2989 : : */
2990 : : int
4064 heikki.linnakangas@i 2991 :CBC 45 : ClosePipeStream(FILE *file)
2992 : : {
2993 : : int i;
2994 : :
2995 : : DO_DB(elog(LOG, "ClosePipeStream: Allocated %d", numAllocatedDescs));
2996 : :
2997 : : /* Remove file from list of allocated files, if it's present */
2998 [ + - ]: 45 : for (i = numAllocatedDescs; --i >= 0;)
2999 : : {
3000 : 45 : AllocateDesc *desc = &allocatedDescs[i];
3001 : :
3002 [ + - + - ]: 45 : if (desc->kind == AllocateDescPipe && desc->desc.file == file)
3003 : 45 : return FreeDesc(desc);
3004 : : }
3005 : :
3006 : : /* Only get here if someone passes us a file not in allocatedDescs */
4064 heikki.linnakangas@i 3007 [ # # ]:UBC 0 : elog(WARNING, "file passed to ClosePipeStream was not obtained from OpenPipeStream");
3008 : :
3009 : 0 : return pclose(file);
3010 : : }
3011 : :
3012 : : /*
3013 : : * closeAllVfds
3014 : : *
3015 : : * Force all VFDs into the physically-closed state, so that the fewest
3016 : : * possible number of kernel file descriptors are in use. There is no
3017 : : * change in the logical state of the VFDs.
3018 : : */
3019 : : void
8631 tgl@sss.pgh.pa.us 3020 :CBC 29 : closeAllVfds(void)
3021 : : {
3022 : : Index i;
3023 : :
9107 3024 [ + - ]: 29 : if (SizeVfdCache > 0)
3025 : : {
2489 3026 [ - + ]: 29 : Assert(FileIsNotOpen(0)); /* Make sure ring not corrupted */
9107 3027 [ + + ]: 928 : for (i = 1; i < SizeVfdCache; i++)
3028 : : {
3029 [ + + ]: 899 : if (!FileIsNotOpen(i))
3030 : 128 : LruDelete(i);
3031 : : }
3032 : : }
3033 : 29 : }
3034 : :
3035 : :
3036 : : /*
3037 : : * SetTempTablespaces
3038 : : *
3039 : : * Define a list (actually an array) of OIDs of tablespaces to use for
3040 : : * temporary files. This list will be used until end of transaction,
3041 : : * unless this function is called again before then. It is caller's
3042 : : * responsibility that the passed-in array has adequate lifespan (typically
3043 : : * it'd be allocated in TopTransactionContext).
3044 : : *
3045 : : * Some entries of the array may be InvalidOid, indicating that the current
3046 : : * database's default tablespace should be used.
3047 : : */
3048 : : void
6156 3049 : 2772 : SetTempTablespaces(Oid *tableSpaces, int numSpaces)
3050 : : {
3051 [ - + ]: 2772 : Assert(numSpaces >= 0);
3052 : 2772 : tempTableSpaces = tableSpaces;
3053 : 2772 : numTempTableSpaces = numSpaces;
3054 : :
3055 : : /*
3056 : : * Select a random starting point in the list. This is to minimize
3057 : : * conflicts between backends that are most likely sharing the same list
3058 : : * of temp tablespaces. Note that if we create multiple temp files in the
3059 : : * same transaction, we'll advance circularly through the list --- this
3060 : : * ensures that large temporary sort files are nicely spread across all
3061 : : * available tablespaces.
3062 : : */
3063 [ - + ]: 2772 : if (numSpaces > 1)
868 tgl@sss.pgh.pa.us 3064 :UBC 0 : nextTempTableSpace = pg_prng_uint64_range(&pg_global_prng_state,
3065 : 0 : 0, numSpaces - 1);
3066 : : else
6156 tgl@sss.pgh.pa.us 3067 :CBC 2772 : nextTempTableSpace = 0;
3068 : 2772 : }
3069 : :
3070 : : /*
3071 : : * TempTablespacesAreSet
3072 : : *
3073 : : * Returns true if SetTempTablespaces has been called in current transaction.
3074 : : * (This is just so that tablespaces.c doesn't need its own per-transaction
3075 : : * state.)
3076 : : */
3077 : : bool
3078 : 4752 : TempTablespacesAreSet(void)
3079 : : {
3080 : 4752 : return (numTempTableSpaces >= 0);
3081 : : }
3082 : :
3083 : : /*
3084 : : * GetTempTablespaces
3085 : : *
3086 : : * Populate an array with the OIDs of the tablespaces that should be used for
3087 : : * temporary files. (Some entries may be InvalidOid, indicating that the
3088 : : * current database's default tablespace should be used.) At most numSpaces
3089 : : * entries will be filled.
3090 : : * Returns the number of OIDs that were copied into the output array.
3091 : : */
3092 : : int
2326 andres@anarazel.de 3093 : 180 : GetTempTablespaces(Oid *tableSpaces, int numSpaces)
3094 : : {
3095 : : int i;
3096 : :
3097 [ - + ]: 180 : Assert(TempTablespacesAreSet());
3098 [ - + - - ]: 180 : for (i = 0; i < numTempTableSpaces && i < numSpaces; ++i)
2326 andres@anarazel.de 3099 :UBC 0 : tableSpaces[i] = tempTableSpaces[i];
3100 : :
2326 andres@anarazel.de 3101 :CBC 180 : return i;
3102 : : }
3103 : :
3104 : : /*
3105 : : * GetNextTempTableSpace
3106 : : *
3107 : : * Select the next temp tablespace to use. A result of InvalidOid means
3108 : : * to use the current database's default tablespace.
3109 : : */
3110 : : Oid
6156 tgl@sss.pgh.pa.us 3111 : 1956 : GetNextTempTableSpace(void)
3112 : : {
3113 [ + + ]: 1956 : if (numTempTableSpaces > 0)
3114 : : {
3115 : : /* Advance nextTempTableSpace counter with wraparound */
6156 tgl@sss.pgh.pa.us 3116 [ + - ]:GBC 1 : if (++nextTempTableSpace >= numTempTableSpaces)
3117 : 1 : nextTempTableSpace = 0;
3118 : 1 : return tempTableSpaces[nextTempTableSpace];
3119 : : }
6156 tgl@sss.pgh.pa.us 3120 :CBC 1955 : return InvalidOid;
3121 : : }
3122 : :
3123 : :
3124 : : /*
3125 : : * AtEOSubXact_Files
3126 : : *
3127 : : * Take care of subtransaction commit/abort. At abort, we close temp files
3128 : : * that the subtransaction may have opened. At commit, we reassign the
3129 : : * files that were opened to the parent subtransaction.
3130 : : */
3131 : : void
7150 3132 : 9927 : AtEOSubXact_Files(bool isCommit, SubTransactionId mySubid,
3133 : : SubTransactionId parentSubid)
3134 : : {
3135 : : Index i;
3136 : :
7200 3137 [ - + ]: 9927 : for (i = 0; i < numAllocatedDescs; i++)
3138 : : {
7150 tgl@sss.pgh.pa.us 3139 [ # # ]:UBC 0 : if (allocatedDescs[i].create_subid == mySubid)
3140 : : {
7200 3141 [ # # ]: 0 : if (isCommit)
7150 3142 : 0 : allocatedDescs[i].create_subid = parentSubid;
3143 : : else
3144 : : {
3145 : : /* have to recheck the item after FreeDesc (ugly) */
7200 3146 : 0 : FreeDesc(&allocatedDescs[i--]);
3147 : : }
3148 : : }
3149 : : }
7200 tgl@sss.pgh.pa.us 3150 :CBC 9927 : }
3151 : :
3152 : : /*
3153 : : * AtEOXact_Files
3154 : : *
3155 : : * This routine is called during transaction commit or abort. All still-open
3156 : : * per-transaction temporary file VFDs are closed, which also causes the
3157 : : * underlying files to be deleted (although they should've been closed already
3158 : : * by the ResourceOwner cleanup). Furthermore, all "allocated" stdio files are
3159 : : * closed. We also forget any transaction-local temp tablespace list.
3160 : : *
3161 : : * The isCommit flag is used only to decide whether to emit warnings about
3162 : : * unclosed files.
3163 : : */
3164 : : void
2178 3165 : 432909 : AtEOXact_Files(bool isCommit)
3166 : : {
3167 : 432909 : CleanupTempFiles(isCommit, false);
6156 3168 : 432909 : tempTableSpaces = NULL;
3169 : 432909 : numTempTableSpaces = -1;
7656 3170 : 432909 : }
3171 : :
3172 : : /*
3173 : : * BeforeShmemExit_Files
3174 : : *
3175 : : * before_shmem_exit hook to clean up temp files during backend shutdown.
3176 : : * Here, we want to clean up *all* temp files including interXact ones.
3177 : : */
3178 : : static void
981 andres@anarazel.de 3179 : 18045 : BeforeShmemExit_Files(int code, Datum arg)
3180 : : {
2178 tgl@sss.pgh.pa.us 3181 : 18045 : CleanupTempFiles(false, true);
3182 : :
3183 : : /* prevent further temp files from being created */
3184 : : #ifdef USE_ASSERT_CHECKING
981 andres@anarazel.de 3185 : 18045 : temporary_files_allowed = false;
3186 : : #endif
7656 tgl@sss.pgh.pa.us 3187 : 18045 : }
3188 : :
3189 : : /*
3190 : : * Close temporary files and delete their underlying files.
3191 : : *
3192 : : * isCommit: if true, this is normal transaction commit, and we don't
3193 : : * expect any remaining files; warn if there are some.
3194 : : *
3195 : : * isProcExit: if true, this is being called as the backend process is
3196 : : * exiting. If that's the case, we should remove all temporary files; if
3197 : : * that's not the case, we are being called for transaction commit/abort
3198 : : * and should only remove transaction-local temp files. In either case,
3199 : : * also clean up "allocated" stdio files, dirs and fds.
3200 : : */
3201 : : static void
2178 3202 : 450954 : CleanupTempFiles(bool isCommit, bool isProcExit)
3203 : : {
3204 : : Index i;
3205 : :
3206 : : /*
3207 : : * Careful here: at proc_exit we need extra cleanup, not just
3208 : : * xact_temporary files.
3209 : : */
4197 3210 [ + + + + ]: 450954 : if (isProcExit || have_xact_temporary_files)
3211 : : {
2489 3212 [ - + ]: 18832 : Assert(FileIsNotOpen(0)); /* Make sure ring not corrupted */
9107 3213 [ + + ]: 1093305 : for (i = 1; i < SizeVfdCache; i++)
3214 : : {
7656 3215 : 1074473 : unsigned short fdstate = VfdCache[i].fdstate;
3216 : :
2326 andres@anarazel.de 3217 [ + + - + ]: 1074473 : if (((fdstate & FD_DELETE_AT_CLOSE) || (fdstate & FD_CLOSE_AT_EOXACT)) &&
3218 [ + - ]: 5 : VfdCache[i].fileName != NULL)
3219 : : {
3220 : : /*
3221 : : * If we're in the process of exiting a backend process, close
3222 : : * all temporary files. Otherwise, only close temporary files
3223 : : * local to the current transaction. They should be closed by
3224 : : * the ResourceOwner mechanism already, so this is just a
3225 : : * debugging cross-check.
3226 : : */
4197 tgl@sss.pgh.pa.us 3227 [ + - ]: 5 : if (isProcExit)
3228 : 5 : FileClose(i);
2326 andres@anarazel.de 3229 [ # # ]:UBC 0 : else if (fdstate & FD_CLOSE_AT_EOXACT)
3230 : : {
4197 tgl@sss.pgh.pa.us 3231 [ # # ]: 0 : elog(WARNING,
3232 : : "temporary file %s not closed at end-of-transaction",
3233 : : VfdCache[i].fileName);
3234 : 0 : FileClose(i);
3235 : : }
3236 : : }
3237 : : }
3238 : :
4197 tgl@sss.pgh.pa.us 3239 :CBC 18832 : have_xact_temporary_files = false;
3240 : : }
3241 : :
3242 : : /* Complain if any allocated files remain open at commit. */
2178 3243 [ + + - + ]: 450954 : if (isCommit && numAllocatedDescs > 0)
2178 tgl@sss.pgh.pa.us 3244 [ # # ]:UBC 0 : elog(WARNING, "%d temporary files and directories not closed at end-of-transaction",
3245 : : numAllocatedDescs);
3246 : :
3247 : : /* Clean up "allocated" stdio files, dirs and fds. */
7200 tgl@sss.pgh.pa.us 3248 [ + + ]:CBC 451076 : while (numAllocatedDescs > 0)
3249 : 122 : FreeDesc(&allocatedDescs[0]);
9107 3250 : 450954 : }
3251 : :
3252 : :
3253 : : /*
3254 : : * Remove temporary and temporary relation files left over from a prior
3255 : : * postmaster session
3256 : : *
3257 : : * This should be called during postmaster startup. It will forcibly
3258 : : * remove any leftover files created by OpenTemporaryFile and any leftover
3259 : : * temporary relation files created by mdcreate.
3260 : : *
3261 : : * During post-backend-crash restart cycle, this routine is called when
3262 : : * remove_temp_files_after_crash GUC is enabled. Multiple crashes while
3263 : : * queries are using temp files could result in useless storage usage that can
3264 : : * only be reclaimed by a service restart. The argument against enabling it is
3265 : : * that someone might want to examine the temporary files for debugging
3266 : : * purposes. This does however mean that OpenTemporaryFile had better allow for
3267 : : * collision with an existing temp file name.
3268 : : *
3269 : : * NOTE: this function and its subroutines generally report syscall failures
3270 : : * with ereport(LOG) and keep going. Removing temp files is not so critical
3271 : : * that we should fail to start the database when we can't do it.
3272 : : */
3273 : : void
8343 3274 : 732 : RemovePgTempFiles(void)
3275 : : {
3276 : : char temp_path[MAXPGPATH + 10 + sizeof(TABLESPACE_VERSION_DIRECTORY) + sizeof(PG_TEMP_FILES_DIR)];
3277 : : DIR *spc_dir;
3278 : : struct dirent *spc_de;
3279 : :
3280 : : /*
3281 : : * First process temp files in pg_default ($PGDATA/base)
3282 : : */
6160 3283 : 732 : snprintf(temp_path, sizeof(temp_path), "base/%s", PG_TEMP_FILES_DIR);
2289 3284 : 732 : RemovePgTempFilesInDir(temp_path, true, false);
4993 rhaas@postgresql.org 3285 : 732 : RemovePgTempRelationFiles("base");
3286 : :
3287 : : /*
3288 : : * Cycle through temp directories for all non-default tablespaces.
3289 : : */
6160 tgl@sss.pgh.pa.us 3290 : 732 : spc_dir = AllocateDir("pg_tblspc");
3291 : :
2323 3292 [ + + ]: 2257 : while ((spc_de = ReadDirExtended(spc_dir, "pg_tblspc", LOG)) != NULL)
3293 : : {
6160 3294 [ + + ]: 1525 : if (strcmp(spc_de->d_name, ".") == 0 ||
3295 [ + + ]: 793 : strcmp(spc_de->d_name, "..") == 0)
7046 3296 : 1464 : continue;
3297 : :
5206 bruce@momjian.us 3298 : 61 : snprintf(temp_path, sizeof(temp_path), "pg_tblspc/%s/%s/%s",
2489 tgl@sss.pgh.pa.us 3299 : 61 : spc_de->d_name, TABLESPACE_VERSION_DIRECTORY, PG_TEMP_FILES_DIR);
2289 3300 : 61 : RemovePgTempFilesInDir(temp_path, true, false);
3301 : :
4993 rhaas@postgresql.org 3302 : 61 : snprintf(temp_path, sizeof(temp_path), "pg_tblspc/%s/%s",
4753 bruce@momjian.us 3303 : 61 : spc_de->d_name, TABLESPACE_VERSION_DIRECTORY);
4993 rhaas@postgresql.org 3304 : 61 : RemovePgTempRelationFiles(temp_path);
3305 : : }
3306 : :
6160 tgl@sss.pgh.pa.us 3307 : 732 : FreeDir(spc_dir);
3308 : :
3309 : : /*
3310 : : * In EXEC_BACKEND case there is a pgsql_tmp directory at the top level of
3311 : : * DataDir as well. However, that is *not* cleaned here because doing so
3312 : : * would create a race condition. It's done separately, earlier in
3313 : : * postmaster startup.
3314 : : */
7046 3315 : 732 : }
3316 : :
3317 : : /*
3318 : : * Process one pgsql_tmp directory for RemovePgTempFiles.
3319 : : *
3320 : : * If missing_ok is true, it's all right for the named directory to not exist.
3321 : : * Any other problem results in a LOG message. (missing_ok should be true at
3322 : : * the top level, since pgsql_tmp directories are not created until needed.)
3323 : : *
3324 : : * At the top level, this should be called with unlink_all = false, so that
3325 : : * only files matching the temporary name prefix will be unlinked. When
3326 : : * recursing it will be called with unlink_all = true to unlink everything
3327 : : * under a top-level temporary directory.
3328 : : *
3329 : : * (These two flags could be replaced by one, but it seems clearer to keep
3330 : : * them separate.)
3331 : : */
3332 : : void
2289 3333 : 794 : RemovePgTempFilesInDir(const char *tmpdirname, bool missing_ok, bool unlink_all)
3334 : : {
3335 : : DIR *temp_dir;
3336 : : struct dirent *temp_de;
3337 : : char rm_path[MAXPGPATH * 2];
3338 : :
7046 3339 : 794 : temp_dir = AllocateDir(tmpdirname);
3340 : :
2289 3341 [ + + + - : 794 : if (temp_dir == NULL && errno == ENOENT && missing_ok)
+ - ]
3342 : 731 : return;
3343 : :
2323 3344 [ + + ]: 193 : while ((temp_de = ReadDirExtended(temp_dir, tmpdirname, LOG)) != NULL)
3345 : : {
7046 3346 [ + + ]: 130 : if (strcmp(temp_de->d_name, ".") == 0 ||
3347 [ + + ]: 67 : strcmp(temp_de->d_name, "..") == 0)
3348 : 126 : continue;
3349 : :
3350 : 4 : snprintf(rm_path, sizeof(rm_path), "%s/%s",
3351 : 4 : tmpdirname, temp_de->d_name);
3352 : :
2326 andres@anarazel.de 3353 [ + + ]: 4 : if (unlink_all ||
3354 [ + - ]: 3 : strncmp(temp_de->d_name,
3355 : : PG_TEMP_FILE_PREFIX,
3356 : : strlen(PG_TEMP_FILE_PREFIX)) == 0)
3357 : 4 : {
590 michael@paquier.xyz 3358 : 4 : PGFileType type = get_dirent_type(rm_path, temp_de, false, LOG);
3359 : :
3360 [ - + ]: 4 : if (type == PGFILETYPE_ERROR)
2326 andres@anarazel.de 3361 :UBC 0 : continue;
590 michael@paquier.xyz 3362 [ + + ]:CBC 4 : else if (type == PGFILETYPE_DIR)
3363 : : {
3364 : : /* recursively remove contents, then directory itself */
2289 tgl@sss.pgh.pa.us 3365 : 1 : RemovePgTempFilesInDir(rm_path, false, true);
3366 : :
2323 3367 [ - + ]: 1 : if (rmdir(rm_path) < 0)
2323 tgl@sss.pgh.pa.us 3368 [ # # ]:UBC 0 : ereport(LOG,
3369 : : (errcode_for_file_access(),
3370 : : errmsg("could not remove directory \"%s\": %m",
3371 : : rm_path)));
3372 : : }
3373 : : else
3374 : : {
2323 tgl@sss.pgh.pa.us 3375 [ - + ]:CBC 3 : if (unlink(rm_path) < 0)
2323 tgl@sss.pgh.pa.us 3376 [ # # ]:UBC 0 : ereport(LOG,
3377 : : (errcode_for_file_access(),
3378 : : errmsg("could not remove file \"%s\": %m",
3379 : : rm_path)));
3380 : : }
3381 : : }
3382 : : else
3383 [ # # ]: 0 : ereport(LOG,
3384 : : (errmsg("unexpected file found in temporary-files directory: \"%s\"",
3385 : : rm_path)));
3386 : : }
3387 : :
7046 tgl@sss.pgh.pa.us 3388 :CBC 63 : FreeDir(temp_dir);
3389 : : }
3390 : :
3391 : : /* Process one tablespace directory, look for per-DB subdirectories */
3392 : : static void
4993 rhaas@postgresql.org 3393 : 793 : RemovePgTempRelationFiles(const char *tsdirname)
3394 : : {
3395 : : DIR *ts_dir;
3396 : : struct dirent *de;
3397 : : char dbspace_path[MAXPGPATH * 2];
3398 : :
3399 : 793 : ts_dir = AllocateDir(tsdirname);
3400 : :
2323 tgl@sss.pgh.pa.us 3401 [ + + ]: 4934 : while ((de = ReadDirExtended(ts_dir, tsdirname, LOG)) != NULL)
3402 : : {
3403 : : /*
3404 : : * We're only interested in the per-database directories, which have
3405 : : * numeric names. Note that this code will also (properly) ignore "."
3406 : : * and "..".
3407 : : */
3408 [ + + ]: 4141 : if (strspn(de->d_name, "0123456789") != strlen(de->d_name))
4993 rhaas@postgresql.org 3409 : 1648 : continue;
3410 : :
3411 : 2493 : snprintf(dbspace_path, sizeof(dbspace_path), "%s/%s",
3412 : 2493 : tsdirname, de->d_name);
3413 : 2493 : RemovePgTempRelationFilesInDbspace(dbspace_path);
3414 : : }
3415 : :
3416 : 793 : FreeDir(ts_dir);
3417 : 793 : }
3418 : :
3419 : : /* Process one per-dbspace directory for RemovePgTempRelationFiles */
3420 : : static void
3421 : 2493 : RemovePgTempRelationFilesInDbspace(const char *dbspacedirname)
3422 : : {
3423 : : DIR *dbspace_dir;
3424 : : struct dirent *de;
3425 : : char rm_path[MAXPGPATH * 2];
3426 : :
3427 : 2493 : dbspace_dir = AllocateDir(dbspacedirname);
3428 : :
2323 tgl@sss.pgh.pa.us 3429 [ + + ]: 749723 : while ((de = ReadDirExtended(dbspace_dir, dbspacedirname, LOG)) != NULL)
3430 : : {
4993 rhaas@postgresql.org 3431 [ + + ]: 747230 : if (!looks_like_temp_rel_name(de->d_name))
3432 : 747216 : continue;
3433 : :
3434 : 14 : snprintf(rm_path, sizeof(rm_path), "%s/%s",
3435 : 14 : dbspacedirname, de->d_name);
3436 : :
2323 tgl@sss.pgh.pa.us 3437 [ - + ]: 14 : if (unlink(rm_path) < 0)
2323 tgl@sss.pgh.pa.us 3438 [ # # ]:UBC 0 : ereport(LOG,
3439 : : (errcode_for_file_access(),
3440 : : errmsg("could not remove file \"%s\": %m",
3441 : : rm_path)));
3442 : : }
3443 : :
4993 rhaas@postgresql.org 3444 :CBC 2493 : FreeDir(dbspace_dir);
3445 : 2493 : }
3446 : :
3447 : : /* t<digits>_<digits>, or t<digits>_<digits>_<forkname> */
3448 : : bool
3449 : 1019180 : looks_like_temp_rel_name(const char *name)
3450 : : {
3451 : : int pos;
3452 : : int savepos;
3453 : :
3454 : : /* Must start with "t". */
3455 [ + + ]: 1019180 : if (name[0] != 't')
3456 : 1019126 : return false;
3457 : :
3458 : : /* Followed by a non-empty string of digits and then an underscore. */
3459 [ + + ]: 235 : for (pos = 1; isdigit((unsigned char) name[pos]); ++pos)
3460 : : ;
3461 [ + - - + ]: 54 : if (pos == 1 || name[pos] != '_')
4993 rhaas@postgresql.org 3462 :UBC 0 : return false;
3463 : :
3464 : : /* Followed by another nonempty string of digits. */
4993 rhaas@postgresql.org 3465 [ + + ]:CBC 285 : for (savepos = ++pos; isdigit((unsigned char) name[pos]); ++pos)
3466 : : ;
3467 [ - + ]: 54 : if (savepos == pos)
4993 rhaas@postgresql.org 3468 :UBC 0 : return false;
3469 : :
3470 : : /* We might have _forkname or .segment or both. */
4993 rhaas@postgresql.org 3471 [ + + ]:CBC 54 : if (name[pos] == '_')
3472 : : {
4753 bruce@momjian.us 3473 : 22 : int forkchar = forkname_chars(&name[pos + 1], NULL);
3474 : :
4993 rhaas@postgresql.org 3475 [ - + ]: 22 : if (forkchar <= 0)
4993 rhaas@postgresql.org 3476 :UBC 0 : return false;
4993 rhaas@postgresql.org 3477 :CBC 22 : pos += forkchar + 1;
3478 : : }
3479 [ + + ]: 54 : if (name[pos] == '.')
3480 : : {
3481 : : int segchar;
3482 : :
4753 bruce@momjian.us 3483 [ + + ]: 44 : for (segchar = 1; isdigit((unsigned char) name[pos + segchar]); ++segchar)
3484 : : ;
4993 rhaas@postgresql.org 3485 [ - + ]: 22 : if (segchar <= 1)
4993 rhaas@postgresql.org 3486 :UBC 0 : return false;
4993 rhaas@postgresql.org 3487 :CBC 22 : pos += segchar;
3488 : : }
3489 : :
3490 : : /* Now we should be at the end. */
3491 [ - + ]: 54 : if (name[pos] != '\0')
4993 rhaas@postgresql.org 3492 :UBC 0 : return false;
4993 rhaas@postgresql.org 3493 :CBC 54 : return true;
3494 : : }
3495 : :
3496 : : #ifdef HAVE_SYNCFS
3497 : : static void
1121 tmunro@postgresql.or 3498 :UBC 0 : do_syncfs(const char *path)
3499 : : {
3500 : : int fd;
3501 : :
902 rhaas@postgresql.org 3502 [ # # # # ]: 0 : ereport_startup_progress("syncing data directory (syncfs), elapsed time: %ld.%02d s, current path: %s",
3503 : : path);
3504 : :
1121 tmunro@postgresql.or 3505 : 0 : fd = OpenTransientFile(path, O_RDONLY);
3506 [ # # ]: 0 : if (fd < 0)
3507 : : {
3508 [ # # ]: 0 : ereport(LOG,
3509 : : (errcode_for_file_access(),
3510 : : errmsg("could not open file \"%s\": %m", path)));
3511 : 0 : return;
3512 : : }
3513 [ # # ]: 0 : if (syncfs(fd) < 0)
3514 [ # # ]: 0 : ereport(LOG,
3515 : : (errcode_for_file_access(),
3516 : : errmsg("could not synchronize file system for file \"%s\": %m", path)));
3517 : 0 : CloseTransientFile(fd);
3518 : : }
3519 : : #endif
3520 : :
3521 : : /*
3522 : : * Issue fsync recursively on PGDATA and all its contents, or issue syncfs for
3523 : : * all potential filesystem, depending on recovery_init_sync_method setting.
3524 : : *
3525 : : * We fsync regular files and directories wherever they are, but we
3526 : : * follow symlinks only for pg_wal and immediately under pg_tblspc.
3527 : : * Other symlinks are presumed to point at files we're not responsible
3528 : : * for fsyncing, and might not have privileges to write at all.
3529 : : *
3530 : : * Errors are logged but not considered fatal; that's because this is used
3531 : : * only during database startup, to deal with the possibility that there are
3532 : : * issued-but-unsynced writes pending against the data directory. We want to
3533 : : * ensure that such writes reach disk before anything that's done in the new
3534 : : * run. However, aborting on error would result in failure to start for
3535 : : * harmless cases such as read-only files in the data directory, and that's
3536 : : * not good either.
3537 : : *
3538 : : * Note that if we previously crashed due to a PANIC on fsync(), we'll be
3539 : : * rewriting all changes again during recovery.
3540 : : *
3541 : : * Note we assume we're chdir'd into PGDATA to begin with.
3542 : : */
3543 : : void
3244 tgl@sss.pgh.pa.us 3544 :CBC 201 : SyncDataDirectory(void)
3545 : : {
3546 : : bool xlog_is_symlink;
3547 : :
3548 : : /* We can skip this whole thing if fsync is disabled. */
3549 [ + - ]: 201 : if (!enableFsync)
3550 : 201 : return;
3551 : :
3552 : : /*
3553 : : * If pg_wal is a symlink, we'll need to recurse into it separately,
3554 : : * because the first walkdir below will ignore it.
3555 : : */
3244 tgl@sss.pgh.pa.us 3556 :UBC 0 : xlog_is_symlink = false;
3557 : :
3558 : : {
3559 : : struct stat st;
3560 : :
2733 rhaas@postgresql.org 3561 [ # # ]: 0 : if (lstat("pg_wal", &st) < 0)
3244 tgl@sss.pgh.pa.us 3562 [ # # ]: 0 : ereport(LOG,
3563 : : (errcode_for_file_access(),
3564 : : errmsg("could not stat file \"%s\": %m",
3565 : : "pg_wal")));
3566 [ # # ]: 0 : else if (S_ISLNK(st.st_mode))
3567 : 0 : xlog_is_symlink = true;
3568 : : }
3569 : :
3570 : : #ifdef HAVE_SYNCFS
221 nathan@postgresql.or 3571 [ # # ]:UNC 0 : if (recovery_init_sync_method == DATA_DIR_SYNC_METHOD_SYNCFS)
3572 : : {
3573 : : DIR *dir;
3574 : : struct dirent *de;
3575 : :
3576 : : /*
3577 : : * On Linux, we don't have to open every single file one by one. We
3578 : : * can use syncfs() to sync whole filesystems. We only expect
3579 : : * filesystem boundaries to exist where we tolerate symlinks, namely
3580 : : * pg_wal and the tablespaces, so we call syncfs() for each of those
3581 : : * directories.
3582 : : */
3583 : :
3584 : : /* Prepare to report progress syncing the data directory via syncfs. */
902 rhaas@postgresql.org 3585 :UBC 0 : begin_startup_progress_phase();
3586 : :
3587 : : /* Sync the top level pgdata directory. */
1121 tmunro@postgresql.or 3588 : 0 : do_syncfs(".");
3589 : : /* If any tablespaces are configured, sync each of those. */
3590 : 0 : dir = AllocateDir("pg_tblspc");
3591 [ # # ]: 0 : while ((de = ReadDirExtended(dir, "pg_tblspc", LOG)))
3592 : : {
3593 : : char path[MAXPGPATH];
3594 : :
3595 [ # # # # ]: 0 : if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
3596 : 0 : continue;
3597 : :
3598 : 0 : snprintf(path, MAXPGPATH, "pg_tblspc/%s", de->d_name);
3599 : 0 : do_syncfs(path);
3600 : : }
3601 : 0 : FreeDir(dir);
3602 : : /* If pg_wal is a symlink, process that too. */
3603 [ # # ]: 0 : if (xlog_is_symlink)
3604 : 0 : do_syncfs("pg_wal");
3605 : 0 : return;
3606 : : }
3607 : : #endif /* !HAVE_SYNCFS */
3608 : :
3609 : : #ifdef PG_FLUSH_DATA_WORKS
3610 : : /* Prepare to report progress of the pre-fsync phase. */
902 rhaas@postgresql.org 3611 : 0 : begin_startup_progress_phase();
3612 : :
3613 : : /*
3614 : : * If possible, hint to the kernel that we're soon going to fsync the data
3615 : : * directory and its contents. Errors in this step are even less
3616 : : * interesting than normal, so log them only at DEBUG1.
3617 : : */
3244 tgl@sss.pgh.pa.us 3618 : 0 : walkdir(".", pre_sync_fname, false, DEBUG1);
3619 [ # # ]: 0 : if (xlog_is_symlink)
2733 rhaas@postgresql.org 3620 : 0 : walkdir("pg_wal", pre_sync_fname, false, DEBUG1);
3244 tgl@sss.pgh.pa.us 3621 : 0 : walkdir("pg_tblspc", pre_sync_fname, true, DEBUG1);
3622 : : #endif
3623 : :
3624 : : /* Prepare to report progress syncing the data directory via fsync. */
902 rhaas@postgresql.org 3625 : 0 : begin_startup_progress_phase();
3626 : :
3627 : : /*
3628 : : * Now we do the fsync()s in the same order.
3629 : : *
3630 : : * The main call ignores symlinks, so in addition to specially processing
3631 : : * pg_wal if it's a symlink, pg_tblspc has to be visited separately with
3632 : : * process_symlinks = true. Note that if there are any plain directories
3633 : : * in pg_tblspc, they'll get fsync'd twice. That's not an expected case
3634 : : * so we don't worry about optimizing it.
3635 : : */
2958 andres@anarazel.de 3636 : 0 : walkdir(".", datadir_fsync_fname, false, LOG);
3244 tgl@sss.pgh.pa.us 3637 [ # # ]: 0 : if (xlog_is_symlink)
2733 rhaas@postgresql.org 3638 : 0 : walkdir("pg_wal", datadir_fsync_fname, false, LOG);
2958 andres@anarazel.de 3639 : 0 : walkdir("pg_tblspc", datadir_fsync_fname, true, LOG);
3640 : : }
3641 : :
3642 : : /*
3643 : : * walkdir: recursively walk a directory, applying the action to each
3644 : : * regular file and directory (including the named directory itself).
3645 : : *
3646 : : * If process_symlinks is true, the action and recursion are also applied
3647 : : * to regular files and directories that are pointed to by symlinks in the
3648 : : * given directory; otherwise symlinks are ignored. Symlinks are always
3649 : : * ignored in subdirectories, ie we intentionally don't pass down the
3650 : : * process_symlinks flag to recursive calls.
3651 : : *
3652 : : * Errors are reported at level elevel, which might be ERROR or less.
3653 : : *
3654 : : * See also walkdir in file_utils.c, which is a frontend version of this
3655 : : * logic.
3656 : : */
3657 : : static void
3244 tgl@sss.pgh.pa.us 3658 :CBC 163 : walkdir(const char *path,
3659 : : void (*action) (const char *fname, bool isdir, int elevel),
3660 : : bool process_symlinks,
3661 : : int elevel)
3662 : : {
3663 : : DIR *dir;
3664 : : struct dirent *de;
3665 : :
3268 rhaas@postgresql.org 3666 : 163 : dir = AllocateDir(path);
3667 : :
3244 tgl@sss.pgh.pa.us 3668 [ + + ]: 2317 : while ((de = ReadDirExtended(dir, path, elevel)) != NULL)
3669 : : {
3670 : : char subpath[MAXPGPATH * 2];
3671 : :
3268 rhaas@postgresql.org 3672 [ - + ]: 2154 : CHECK_FOR_INTERRUPTS();
3673 : :
3674 [ + + ]: 2154 : if (strcmp(de->d_name, ".") == 0 ||
3675 [ + + ]: 1991 : strcmp(de->d_name, "..") == 0)
3676 : 326 : continue;
3677 : :
2560 peter_e@gmx.net 3678 : 1828 : snprintf(subpath, sizeof(subpath), "%s/%s", path, de->d_name);
3679 : :
1315 tmunro@postgresql.or 3680 [ + - - ]: 1828 : switch (get_dirent_type(subpath, de, process_symlinks, elevel))
3681 : : {
3682 : 1828 : case PGFILETYPE_REG:
3683 : 1828 : (*action) (subpath, false, elevel);
3684 : 1828 : break;
1315 tmunro@postgresql.or 3685 :UBC 0 : case PGFILETYPE_DIR:
3686 : 0 : walkdir(subpath, action, false, elevel);
3687 : 0 : break;
3688 : 0 : default:
3689 : :
3690 : : /*
3691 : : * Errors are already reported directly by get_dirent_type(),
3692 : : * and any remaining symlinks and unknown file types are
3693 : : * ignored.
3694 : : */
3695 : 0 : break;
3696 : : }
3697 : : }
3698 : :
3244 tgl@sss.pgh.pa.us 3699 :CBC 163 : FreeDir(dir); /* we ignore any error here */
3700 : :
3701 : : /*
3702 : : * It's important to fsync the destination directory itself as individual
3703 : : * file fsyncs don't guarantee that the directory entry for the file is
3704 : : * synced. However, skip this if AllocateDir failed; the action function
3705 : : * might not be robust against that.
3706 : : */
2323 3707 [ + - ]: 163 : if (dir)
3708 : 163 : (*action) (path, true, elevel);
3244 3709 : 163 : }
3710 : :
3711 : :
3712 : : /*
3713 : : * Hint to the OS that it should get ready to fsync() this file.
3714 : : *
3715 : : * Ignores errors trying to open unreadable files, and logs other errors at a
3716 : : * caller-specified level.
3717 : : */
3718 : : #ifdef PG_FLUSH_DATA_WORKS
3719 : :
3720 : : static void
3244 tgl@sss.pgh.pa.us 3721 :UBC 0 : pre_sync_fname(const char *fname, bool isdir, int elevel)
3722 : : {
3723 : : int fd;
3724 : :
3725 : : /* Don't try to flush directories, it'll likely just fail */
2923 3726 [ # # ]: 0 : if (isdir)
3727 : 0 : return;
3728 : :
902 rhaas@postgresql.org 3729 [ # # # # ]: 0 : ereport_startup_progress("syncing data directory (pre-fsync), elapsed time: %ld.%02d s, current path: %s",
3730 : : fname);
3731 : :
2395 peter_e@gmx.net 3732 : 0 : fd = OpenTransientFile(fname, O_RDONLY | PG_BINARY);
3733 : :
3244 tgl@sss.pgh.pa.us 3734 [ # # ]: 0 : if (fd < 0)
3735 : : {
2923 3736 [ # # ]: 0 : if (errno == EACCES)
3244 3737 : 0 : return;
3738 [ # # ]: 0 : ereport(elevel,
3739 : : (errcode_for_file_access(),
3740 : : errmsg("could not open file \"%s\": %m", fname)));
3741 : 0 : return;
3742 : : }
3743 : :
3744 : : /*
3745 : : * pg_flush_data() ignores errors, which is ok because this is only a
3746 : : * hint.
3747 : : */
2977 andres@anarazel.de 3748 : 0 : pg_flush_data(fd, 0, 0);
3749 : :
1744 peter@eisentraut.org 3750 [ # # ]: 0 : if (CloseTransientFile(fd) != 0)
1863 michael@paquier.xyz 3751 [ # # ]: 0 : ereport(elevel,
3752 : : (errcode_for_file_access(),
3753 : : errmsg("could not close file \"%s\": %m", fname)));
3754 : : }
3755 : :
3756 : : #endif /* PG_FLUSH_DATA_WORKS */
3757 : :
3758 : : static void
2958 andres@anarazel.de 3759 : 0 : datadir_fsync_fname(const char *fname, bool isdir, int elevel)
3760 : : {
902 rhaas@postgresql.org 3761 [ # # # # ]: 0 : ereport_startup_progress("syncing data directory (fsync), elapsed time: %ld.%02d s, current path: %s",
3762 : : fname);
3763 : :
3764 : : /*
3765 : : * We want to silently ignoring errors about unreadable files. Pass that
3766 : : * desire on to fsync_fname_ext().
3767 : : */
2958 andres@anarazel.de 3768 : 0 : fsync_fname_ext(fname, isdir, true, elevel);
3769 : 0 : }
3770 : :
3771 : : static void
2326 andres@anarazel.de 3772 :CBC 1991 : unlink_if_exists_fname(const char *fname, bool isdir, int elevel)
3773 : : {
3774 [ + + ]: 1991 : if (isdir)
3775 : : {
3776 [ - + - - ]: 163 : if (rmdir(fname) != 0 && errno != ENOENT)
2326 andres@anarazel.de 3777 [ # # ]:UBC 0 : ereport(elevel,
3778 : : (errcode_for_file_access(),
3779 : : errmsg("could not remove directory \"%s\": %m", fname)));
3780 : : }
3781 : : else
3782 : : {
3783 : : /* Use PathNameDeleteTemporaryFile to report filesize */
2326 andres@anarazel.de 3784 :CBC 1828 : PathNameDeleteTemporaryFile(fname, false);
3785 : : }
3786 : 1991 : }
3787 : :
3788 : : /*
3789 : : * fsync_fname_ext -- Try to fsync a file or directory
3790 : : *
3791 : : * If ignore_perm is true, ignore errors upon trying to open unreadable
3792 : : * files. Logs other errors at a caller-specified level.
3793 : : *
3794 : : * Returns 0 if the operation succeeded, -1 otherwise.
3795 : : */
3796 : : int
2958 3797 : 26159 : fsync_fname_ext(const char *fname, bool isdir, bool ignore_perm, int elevel)
3798 : : {
3799 : : int fd;
3800 : : int flags;
3801 : : int returncode;
3802 : :
3803 : : /*
3804 : : * Some OSs require directories to be opened read-only whereas other
3805 : : * systems don't allow us to fsync files opened read-only; so we need both
3806 : : * cases here. Using O_RDWR will cause us to fail to fsync files that are
3807 : : * not writable by our userid, but we assume that's OK.
3808 : : */
3244 tgl@sss.pgh.pa.us 3809 : 26159 : flags = PG_BINARY;
3810 [ + + ]: 26159 : if (!isdir)
3811 : 8840 : flags |= O_RDWR;
3812 : : else
3813 : 17319 : flags |= O_RDONLY;
3814 : :
2395 peter_e@gmx.net 3815 : 26159 : fd = OpenTransientFile(fname, flags);
3816 : :
3817 : : /*
3818 : : * Some OSs don't allow us to open directories at all (Windows returns
3819 : : * EACCES), just ignore the error in that case. If desired also silently
3820 : : * ignoring errors about unreadable files. Log others.
3821 : : */
2958 andres@anarazel.de 3822 [ - + - - : 26159 : if (fd < 0 && isdir && (errno == EISDIR || errno == EACCES))
- - - - ]
2958 andres@anarazel.de 3823 :UBC 0 : return 0;
2958 andres@anarazel.de 3824 [ - + - - :CBC 26159 : else if (fd < 0 && ignore_perm && errno == EACCES)
- - ]
2958 andres@anarazel.de 3825 :UBC 0 : return 0;
2958 andres@anarazel.de 3826 [ - + ]:CBC 26159 : else if (fd < 0)
3827 : : {
3244 tgl@sss.pgh.pa.us 3828 [ # # ]:UBC 0 : ereport(elevel,
3829 : : (errcode_for_file_access(),
3830 : : errmsg("could not open file \"%s\": %m", fname)));
2958 andres@anarazel.de 3831 : 0 : return -1;
3832 : : }
3833 : :
3244 tgl@sss.pgh.pa.us 3834 :CBC 26159 : returncode = pg_fsync(fd);
3835 : :
3836 : : /*
3837 : : * Some OSes don't allow us to fsync directories at all, so we can ignore
3838 : : * those errors. Anything else needs to be logged.
3839 : : */
1876 tmunro@postgresql.or 3840 [ - + - - : 26159 : if (returncode != 0 && !(isdir && (errno == EBADF || errno == EINVAL)))
- - - - ]
3841 : : {
3842 : : int save_errno;
3843 : :
3844 : : /* close file upon error, might not be in transaction context */
2958 andres@anarazel.de 3845 :UBC 0 : save_errno = errno;
3846 : 0 : (void) CloseTransientFile(fd);
3847 : 0 : errno = save_errno;
3848 : :
3244 tgl@sss.pgh.pa.us 3849 [ # # ]: 0 : ereport(elevel,
3850 : : (errcode_for_file_access(),
3851 : : errmsg("could not fsync file \"%s\": %m", fname)));
2958 andres@anarazel.de 3852 : 0 : return -1;
3853 : : }
3854 : :
1744 peter@eisentraut.org 3855 [ - + ]:CBC 26159 : if (CloseTransientFile(fd) != 0)
3856 : : {
1863 michael@paquier.xyz 3857 [ # # ]:UBC 0 : ereport(elevel,
3858 : : (errcode_for_file_access(),
3859 : : errmsg("could not close file \"%s\": %m", fname)));
3860 : 0 : return -1;
3861 : : }
3862 : :
2958 andres@anarazel.de 3863 :CBC 26159 : return 0;
3864 : : }
3865 : :
3866 : : /*
3867 : : * fsync_parent_path -- fsync the parent path of a file or directory
3868 : : *
3869 : : * This is aimed at making file operations persistent on disk in case of
3870 : : * an OS crash or power failure.
3871 : : */
3872 : : static int
3873 : 3361 : fsync_parent_path(const char *fname, int elevel)
3874 : : {
3875 : : char parentpath[MAXPGPATH];
3876 : :
3877 : 3361 : strlcpy(parentpath, fname, MAXPGPATH);
3878 : 3361 : get_parent_directory(parentpath);
3879 : :
3880 : : /*
3881 : : * get_parent_directory() returns an empty string if the input argument is
3882 : : * just a file name (see comments in path.c), so handle that as being the
3883 : : * current directory.
3884 : : */
3885 [ + + ]: 3361 : if (strlen(parentpath) == 0)
3886 : 203 : strlcpy(parentpath, ".", MAXPGPATH);
3887 : :
3888 [ - + ]: 3361 : if (fsync_fname_ext(parentpath, true, false, elevel) != 0)
2958 andres@anarazel.de 3889 :UBC 0 : return -1;
3890 : :
2958 andres@anarazel.de 3891 :CBC 3361 : return 0;
3892 : : }
3893 : :
3894 : : /*
3895 : : * Create a PostgreSQL data sub-directory
3896 : : *
3897 : : * The data directory itself, and most of its sub-directories, are created at
3898 : : * initdb time, but we do have some occasions when we create directories in
3899 : : * the backend (CREATE TABLESPACE, for example). In those cases, we want to
3900 : : * make sure that those directories are created consistently. Today, that means
3901 : : * making sure that the created directory has the correct permissions, which is
3902 : : * what pg_dir_create_mode tracks for us.
3903 : : *
3904 : : * Note that we also set the umask() based on what we understand the correct
3905 : : * permissions to be (see file_perm.c).
3906 : : *
3907 : : * For permissions other than the default, mkdir() can be used directly, but
3908 : : * be sure to consider carefully such cases -- a sub-directory with incorrect
3909 : : * permissions in a PostgreSQL data directory could cause backups and other
3910 : : * processes to fail.
3911 : : */
3912 : : int
2199 sfrost@snowman.net 3913 : 1297 : MakePGDirectory(const char *directoryName)
3914 : : {
3915 : 1297 : return mkdir(directoryName, pg_dir_create_mode);
3916 : : }
3917 : :
3918 : : /*
3919 : : * Return the passed-in error level, or PANIC if data_sync_retry is off.
3920 : : *
3921 : : * Failure to fsync any data file is cause for immediate panic, unless
3922 : : * data_sync_retry is enabled. Data may have been written to the operating
3923 : : * system and removed from our buffer pool already, and if we are running on
3924 : : * an operating system that forgets dirty data on write-back failure, there
3925 : : * may be only one copy of the data remaining: in the WAL. A later attempt to
3926 : : * fsync again might falsely report success. Therefore we must not allow any
3927 : : * further checkpoints to be attempted. data_sync_retry can in theory be
3928 : : * enabled on systems known not to drop dirty buffered data on write-back
3929 : : * failure (with the likely outcome that checkpoints will continue to fail
3930 : : * until the underlying problem is fixed).
3931 : : *
3932 : : * Any code that reports a failure from fsync() or related functions should
3933 : : * filter the error level with this function.
3934 : : */
3935 : : int
1973 tmunro@postgresql.or 3936 : 16284 : data_sync_elevel(int elevel)
3937 : : {
3938 [ - + ]: 16284 : return data_sync_retry ? elevel : PANIC;
3939 : : }
3940 : :
3941 : : bool
236 peter@eisentraut.org 3942 : 930 : check_debug_io_direct(char **newval, void **extra, GucSource source)
3943 : : {
372 tmunro@postgresql.or 3944 : 930 : bool result = true;
3945 : : int flags;
3946 : :
3947 : : #if PG_O_DIRECT == 0
3948 : : if (strcmp(*newval, "") != 0)
3949 : : {
3950 : : GUC_check_errdetail("debug_io_direct is not supported on this platform.");
3951 : : result = false;
3952 : : }
3953 : : flags = 0;
3954 : : #else
3955 : : List *elemlist;
3956 : : ListCell *l;
3957 : : char *rawstring;
3958 : :
3959 : : /* Need a modifiable copy of string */
3960 : 930 : rawstring = pstrdup(*newval);
3961 : :
3962 [ - + ]: 930 : if (!SplitGUCList(rawstring, ',', &elemlist))
3963 : : {
87 peter@eisentraut.org 3964 :UNC 0 : GUC_check_errdetail("Invalid list syntax in parameter %s",
3965 : : "debug_io_direct");
372 tmunro@postgresql.or 3966 :UBC 0 : pfree(rawstring);
3967 : 0 : list_free(elemlist);
3968 : 0 : return false;
3969 : : }
3970 : :
372 tmunro@postgresql.or 3971 :CBC 930 : flags = 0;
3972 [ + + + + : 936 : foreach(l, elemlist)
+ + ]
3973 : : {
3974 : 6 : char *item = (char *) lfirst(l);
3975 : :
3976 [ + + ]: 6 : if (pg_strcasecmp(item, "data") == 0)
3977 : 2 : flags |= IO_DIRECT_DATA;
3978 [ + + ]: 4 : else if (pg_strcasecmp(item, "wal") == 0)
3979 : 2 : flags |= IO_DIRECT_WAL;
3980 [ + - ]: 2 : else if (pg_strcasecmp(item, "wal_init") == 0)
3981 : 2 : flags |= IO_DIRECT_WAL_INIT;
3982 : : else
3983 : : {
87 peter@eisentraut.org 3984 :UNC 0 : GUC_check_errdetail("Invalid option \"%s\"", item);
372 tmunro@postgresql.or 3985 :UBC 0 : result = false;
3986 : 0 : break;
3987 : : }
3988 : : }
3989 : :
3990 : : /*
3991 : : * It's possible to configure block sizes smaller than our assumed I/O
3992 : : * alignment size, which could result in invalid I/O requests.
3993 : : */
3994 : : #if XLOG_BLCKSZ < PG_IO_ALIGN_SIZE
3995 : : if (result && (flags & (IO_DIRECT_WAL | IO_DIRECT_WAL_INIT)))
3996 : : {
3997 : : GUC_check_errdetail("debug_io_direct is not supported for WAL because XLOG_BLCKSZ is too small");
3998 : : result = false;
3999 : : }
4000 : : #endif
4001 : : #if BLCKSZ < PG_IO_ALIGN_SIZE
4002 : : if (result && (flags & IO_DIRECT_DATA))
4003 : : {
4004 : : GUC_check_errdetail("debug_io_direct is not supported for data because BLCKSZ is too small");
4005 : : result = false;
4006 : : }
4007 : : #endif
4008 : :
372 tmunro@postgresql.or 4009 :CBC 930 : pfree(rawstring);
4010 : 930 : list_free(elemlist);
4011 : : #endif
4012 : :
4013 [ - + ]: 930 : if (!result)
372 tmunro@postgresql.or 4014 :UBC 0 : return result;
4015 : :
4016 : : /* Save the flags in *extra, for use by assign_debug_io_direct */
372 tmunro@postgresql.or 4017 :CBC 930 : *extra = guc_malloc(ERROR, sizeof(int));
4018 : 930 : *((int *) *extra) = flags;
4019 : :
4020 : 930 : return result;
4021 : : }
4022 : :
4023 : : extern void
236 peter@eisentraut.org 4024 : 930 : assign_debug_io_direct(const char *newval, void *extra)
4025 : : {
372 tmunro@postgresql.or 4026 : 930 : int *flags = (int *) extra;
4027 : :
4028 : 930 : io_direct_flags = *flags;
4029 : 930 : }
4030 : :
4031 : : /* ResourceOwner callbacks */
4032 : :
4033 : : static void
158 heikki.linnakangas@i 4034 :GNC 4 : ResOwnerReleaseFile(Datum res)
4035 : : {
4036 : 4 : File file = (File) DatumGetInt32(res);
4037 : : Vfd *vfdP;
4038 : :
4039 [ + - + - : 4 : Assert(FileIsValid(file));
- + ]
4040 : :
4041 : 4 : vfdP = &VfdCache[file];
4042 : 4 : vfdP->resowner = NULL;
4043 : :
4044 : 4 : FileClose(file);
4045 : 4 : }
4046 : :
4047 : : static char *
158 heikki.linnakangas@i 4048 :UNC 0 : ResOwnerPrintFile(Datum res)
4049 : : {
4050 : 0 : return psprintf("File %d", DatumGetInt32(res));
4051 : : }
|
