source: branches/samba-3.5.x/lib/tdb/common/open.c

Last change on this file was 847, checked in by Silvan Scherrer, 12 years ago

samba server 3.5: bring 3.5 code base in line with 3.6
File size: 16.3 KB
Line 
1 /*
2 Unix SMB/CIFS implementation.
3
4 trivial database library
5
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
9
10 ** NOTE! The following LGPL license applies to the tdb
11 ** library. This does NOT imply that all of Samba is released
12 ** under the LGPL
13
14 This library is free software; you can redistribute it and/or
15 modify it under the terms of the GNU Lesser General Public
16 License as published by the Free Software Foundation; either
17 version 3 of the License, or (at your option) any later version.
18
19 This library is distributed in the hope that it will be useful,
20 but WITHOUT ANY WARRANTY; without even the implied warranty of
21 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 Lesser General Public License for more details.
23
24 You should have received a copy of the GNU Lesser General Public
25 License along with this library; if not, see <http://www.gnu.org/licenses/>.
26*/
27
28#include "tdb_private.h"
29
30/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
31static struct tdb_context *tdbs = NULL;
32
33
34/* This is based on the hash algorithm from gdbm */
35static unsigned int default_tdb_hash(TDB_DATA *key)
36{
37 uint32_t value; /* Used to compute the hash value. */
38 uint32_t i; /* Used to cycle through random values. */
39
40 /* Set the initial value from the key size. */
41 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
42 value = (value + (key->dptr[i] << (i*5 % 24)));
43
44 return (1103515243 * value + 12345);
45}
46
47
48/* initialise a new database with a specified hash size */
49static int tdb_new_database(struct tdb_context *tdb, int hash_size)
50{
51 struct tdb_header *newdb;
52 size_t size;
53 int ret = -1;
54 ssize_t written;
55
56 /* We make it up in memory, then write it out if not internal */
57 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
58 if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
59 tdb->ecode = TDB_ERR_OOM;
60 return -1;
61 }
62
63 /* Fill in the header */
64 newdb->version = TDB_VERSION;
65 newdb->hash_size = hash_size;
66 if (tdb->flags & TDB_INTERNAL) {
67 tdb->map_size = size;
68 tdb->map_ptr = (char *)newdb;
69 memcpy(&tdb->header, newdb, sizeof(tdb->header));
70 /* Convert the `ondisk' version if asked. */
71 CONVERT(*newdb);
72 return 0;
73 }
74 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
75 goto fail;
76
77 if (ftruncate(tdb->fd, 0) == -1)
78 goto fail;
79
80 /* This creates an endian-converted header, as if read from disk */
81 CONVERT(*newdb);
82 memcpy(&tdb->header, newdb, sizeof(tdb->header));
83 /* Don't endian-convert the magic food! */
84 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
85 /* we still have "ret == -1" here */
86 written = write(tdb->fd, newdb, size);
87 if (written == size) {
88 ret = 0;
89 } else if (written != -1) {
90 /* call write once again, this usually should return -1 and
91 * set errno appropriately */
92 size -= written;
93 written = write(tdb->fd, newdb+written, size);
94 if (written == size) {
95 ret = 0;
96 } else if (written >= 0) {
97 /* a second incomplete write - we give up.
98 * guessing the errno... */
99 errno = ENOSPC;
100 }
101 }
102
103 fail:
104 SAFE_FREE(newdb);
105 return ret;
106}
107
108
109
110static int tdb_already_open(dev_t device,
111 ino_t ino)
112{
113 struct tdb_context *i;
114
115 for (i = tdbs; i; i = i->next) {
116 if (i->device == device && i->inode == ino) {
117 return 1;
118 }
119 }
120
121 return 0;
122}
123
124/* open the database, creating it if necessary
125
126 The open_flags and mode are passed straight to the open call on the
127 database file. A flags value of O_WRONLY is invalid. The hash size
128 is advisory, use zero for a default value.
129
130 Return is NULL on error, in which case errno is also set. Don't
131 try to call tdb_error or tdb_errname, just do strerror(errno).
132
133 @param name may be NULL for internal databases. */
134struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
135 int open_flags, mode_t mode)
136{
137 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
138}
139
140/* a default logging function */
141static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
142static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
143{
144}
145
146
147struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
148 int open_flags, mode_t mode,
149 const struct tdb_logging_context *log_ctx,
150 tdb_hash_func hash_fn)
151{
152 struct tdb_context *tdb;
153 struct stat st;
154 int rev = 0, locked = 0;
155 unsigned char *vp;
156 uint32_t vertest;
157 unsigned v;
158
159 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
160 /* Can't log this */
161 errno = ENOMEM;
162 goto fail;
163 }
164 tdb_io_init(tdb);
165 tdb->fd = -1;
166#ifdef TDB_TRACE
167 tdb->tracefd = -1;
168#endif
169 tdb->name = NULL;
170 tdb->map_ptr = NULL;
171 tdb->flags = tdb_flags;
172 tdb->open_flags = open_flags;
173 if (log_ctx) {
174 tdb->log = *log_ctx;
175 } else {
176 tdb->log.log_fn = null_log_fn;
177 tdb->log.log_private = NULL;
178 }
179 tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
180
181 /* cache the page size */
182 tdb->page_size = getpagesize();
183 if (tdb->page_size <= 0) {
184 tdb->page_size = 0x2000;
185 }
186
187 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
188
189 if ((open_flags & O_ACCMODE) == O_WRONLY) {
190 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
191 name));
192 errno = EINVAL;
193 goto fail;
194 }
195
196 if (hash_size == 0)
197 hash_size = DEFAULT_HASH_SIZE;
198 if ((open_flags & O_ACCMODE) == O_RDONLY) {
199 tdb->read_only = 1;
200 /* read only databases don't do locking or clear if first */
201 tdb->flags |= TDB_NOLOCK;
202 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
203 }
204
205 if ((tdb->flags & TDB_ALLOW_NESTING) &&
206 (tdb->flags & TDB_DISALLOW_NESTING)) {
207 tdb->ecode = TDB_ERR_NESTING;
208 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
209 "allow_nesting and disallow_nesting are not allowed together!"));
210 errno = EINVAL;
211 goto fail;
212 }
213
214 /*
215 * TDB_ALLOW_NESTING is the default behavior.
216 * Note: this may change in future versions!
217 */
218 if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
219 tdb->flags |= TDB_ALLOW_NESTING;
220 }
221
222 /* internal databases don't mmap or lock, and start off cleared */
223 if (tdb->flags & TDB_INTERNAL) {
224 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
225 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
226 if (tdb_new_database(tdb, hash_size) != 0) {
227 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
228 goto fail;
229 }
230 goto internal;
231 }
232
233#ifdef __OS2__
234 if (os2_crtSem(tdb, name, "tdb_open_ex") != 0)
235 goto fail;
236#endif
237
238 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
239 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
240 name, strerror(errno)));
241 goto fail; /* errno set by open(2) */
242 }
243
244 /* on exec, don't inherit the fd */
245 v = fcntl(tdb->fd, F_GETFD, 0);
246 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
247
248 /* ensure there is only one process initialising at once */
249 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
250 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
251 name, strerror(errno)));
252 goto fail; /* errno set by tdb_brlock */
253 }
254
255 /* we need to zero database if we are the only one with it open */
256 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
257 (!tdb->read_only)
258 && (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))
259 ) {
260 open_flags |= O_CREAT;
261 if (ftruncate(tdb->fd, 0) == -1) {
262 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
263 "failed to truncate %s: %s\n",
264 name, strerror(errno)));
265 goto fail; /* errno set by ftruncate */
266 }
267 }
268
269 errno = 0;
270 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
271 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
272 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
273 if (errno == 0) {
274 errno = EIO; /* ie bad format or something */
275 }
276 goto fail;
277 }
278 rev = (tdb->flags & TDB_CONVERT);
279 } else if (tdb->header.version != TDB_VERSION
280 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
281 /* wrong version */
282 errno = EIO;
283 goto fail;
284 }
285 vp = (unsigned char *)&tdb->header.version;
286 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
287 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
288 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
289 if (!rev)
290 tdb->flags &= ~TDB_CONVERT;
291 else {
292 tdb->flags |= TDB_CONVERT;
293 tdb_convert(&tdb->header, sizeof(tdb->header));
294 }
295 if (fstat(tdb->fd, &st) == -1)
296 goto fail;
297
298 if (tdb->header.rwlocks != 0) {
299 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
300 goto fail;
301 }
302
303 /* Is it already in the open list? If so, fail. */
304 if (tdb_already_open(st.st_dev, st.st_ino)) {
305 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
306 "%s (%d,%d) is already open in this process\n",
307 name, (int)st.st_dev, (int)st.st_ino));
308 errno = EBUSY;
309 goto fail;
310 }
311
312 if (!(tdb->name = (char *)strdup(name))) {
313 errno = ENOMEM;
314 goto fail;
315 }
316
317 tdb->map_size = st.st_size;
318 tdb->device = st.st_dev;
319 tdb->inode = st.st_ino;
320 tdb_mmap(tdb);
321 if (locked) {
322 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
323 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
324 "failed to take ACTIVE_LOCK on %s: %s\n",
325 name, strerror(errno)));
326 goto fail;
327 }
328 }
329
330 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
331 we didn't get the initial exclusive lock as we need to let all other
332 users know we're using it. */
333
334 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
335 /* leave this lock in place to indicate it's in use */
336 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
337 goto fail;
338 }
339
340 /* if needed, run recovery */
341 if (tdb_transaction_recover(tdb) == -1) {
342 goto fail;
343 }
344
345#ifdef TDB_TRACE
346 {
347 char tracefile[strlen(name) + 32];
348
349 snprintf(tracefile, sizeof(tracefile),
350 "%s.trace.%li", name, (long)getpid());
351 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
352 if (tdb->tracefd >= 0) {
353 tdb_enable_seqnum(tdb);
354 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
355 open_flags);
356 } else
357 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
358 }
359#endif
360
361 internal:
362 /* Internal (memory-only) databases skip all the code above to
363 * do with disk files, and resume here by releasing their
364 * global lock and hooking into the active list. */
365 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
366 goto fail;
367 tdb->next = tdbs;
368 tdbs = tdb;
369 return tdb;
370
371 fail:
372 { int save_errno = errno;
373
374#ifdef __OS2__
375 DosCloseMutexSem(tdb->hActiveLock);
376 tdb->hActiveLock = 0;
377#endif
378 if (!tdb)
379 return NULL;
380
381#ifdef TDB_TRACE
382 close(tdb->tracefd);
383#endif
384 if (tdb->map_ptr) {
385 if (tdb->flags & TDB_INTERNAL)
386 SAFE_FREE(tdb->map_ptr);
387 else
388 tdb_munmap(tdb);
389 }
390 SAFE_FREE(tdb->name);
391 if (tdb->fd != -1)
392 if (close(tdb->fd) != 0)
393 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
394 SAFE_FREE(tdb);
395 errno = save_errno;
396 return NULL;
397 }
398}
399
400/*
401 * Set the maximum number of dead records per hash chain
402 */
403
404void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
405{
406 tdb->max_dead_records = max_dead;
407}
408
409/**
410 * Close a database.
411 *
412 * @returns -1 for error; 0 for success.
413 **/
414int tdb_close(struct tdb_context *tdb)
415{
416 struct tdb_context **i;
417 int ret = 0;
418
419 tdb_trace(tdb, "tdb_close");
420 if (tdb->transaction) {
421 _tdb_transaction_cancel(tdb);
422 }
423
424 if (tdb->map_ptr) {
425 if (tdb->flags & TDB_INTERNAL)
426 SAFE_FREE(tdb->map_ptr);
427 else
428 tdb_munmap(tdb);
429 }
430 SAFE_FREE(tdb->name);
431 if (tdb->fd != -1) {
432 ret = close(tdb->fd);
433 tdb->fd = -1;
434 }
435 SAFE_FREE(tdb->lockrecs);
436
437 /* Remove from contexts list */
438 for (i = &tdbs; *i; i = &(*i)->next) {
439 if (*i == tdb) {
440 *i = tdb->next;
441 break;
442 }
443 }
444
445#ifdef __OS2__
446 DosCloseMutexSem(tdb->hActiveLock);
447 tdb->hActiveLock = 0;
448#endif
449
450#ifdef TDB_TRACE
451 close(tdb->tracefd);
452#endif
453 memset(tdb, 0, sizeof(*tdb));
454 SAFE_FREE(tdb);
455
456 return ret;
457}
458
459/* register a loging function */
460void tdb_set_logging_function(struct tdb_context *tdb,
461 const struct tdb_logging_context *log_ctx)
462{
463 tdb->log = *log_ctx;
464}
465
466void *tdb_get_logging_private(struct tdb_context *tdb)
467{
468 return tdb->log.log_private;
469}
470
471static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
472{
473#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
474 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
475 struct stat st;
476#endif
477
478 if (tdb->flags & TDB_INTERNAL) {
479 return 0; /* Nothing to do. */
480 }
481
482 if (tdb->num_locks != 0 || tdb->global_lock.count) {
483 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
484 goto fail;
485 }
486
487 if (tdb->transaction != 0) {
488 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
489 goto fail;
490 }
491
492/* If we have real pread & pwrite, we can skip reopen. */
493#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
494 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
495 if (tdb_munmap(tdb) != 0) {
496 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
497 goto fail;
498 }
499 if (close(tdb->fd) != 0)
500 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
501 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
502 if (tdb->fd == -1) {
503 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
504 goto fail;
505 }
506 if (fstat(tdb->fd, &st) != 0) {
507 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
508 goto fail;
509 }
510 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
511 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
512 goto fail;
513 }
514 tdb_mmap(tdb);
515#endif /* fake pread or pwrite */
516
517#ifdef __OS2__
518 DosCloseMutexSem(tdb->hActiveLock);
519 tdb->hActiveLock = 0;
520
521 if (os2_crtSem(tdb, tdb->name, "tdb_reopen") != 0)
522 goto fail;
523#endif
524
525 if (active_lock &&
526 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
527 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
528 goto fail;
529 }
530
531 return 0;
532
533fail:
534 tdb_close(tdb);
535 return -1;
536}
537
538/* reopen a tdb - this can be used after a fork to ensure that we have an independent
539 seek pointer from our parent and to re-establish locks */
540int tdb_reopen(struct tdb_context *tdb)
541{
542 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
543}
544
545/* reopen all tdb's */
546int tdb_reopen_all(int parent_longlived)
547{
548 struct tdb_context *tdb;
549
550 for (tdb=tdbs; tdb; tdb = tdb->next) {
551 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
552
553 /*
554 * If the parent is longlived (ie. a
555 * parent daemon architecture), we know
556 * it will keep it's active lock on a
557 * tdb opened with CLEAR_IF_FIRST. Thus
558 * for child processes we don't have to
559 * add an active lock. This is essential
560 * to improve performance on systems that
561 * keep POSIX locks as a non-scalable data
562 * structure in the kernel.
563 */
564 if (parent_longlived) {
565 /* Ensure no clear-if-first. */
566 active_lock = false;
567 }
568
569 if (tdb_reopen_internal(tdb, active_lock) != 0)
570 return -1;
571 }
572
573 return 0;
574}
575#ifdef __OS2__
576int os2_crtSem(struct tdb_context *tdb, const char *name, const char *caller)
577{
578 // name could be null, so handle it
579 if (name == NULL)
580 return -1;
581
582 char szSem[_MAX_PATH];
583 char drive[_MAX_DRIVE], dir[_MAX_DIR];
584 char fname[_MAX_FNAME], ext[_MAX_EXT];
585 APIRET rc;
586 // extract path info
587 _splitpath(name, drive, dir, fname, ext);
588 sprintf(szSem, "\\SEM32\\TDB_AL_%s%s%s", dir, fname, ext);
589 rc = DosCreateMutexSem(szSem, &tdb->hActiveLock, 0, FALSE);
590 if (rc == ERROR_DUPLICATE_NAME)
591 rc = DosOpenMutexSem(szSem, &tdb->hActiveLock);
592 if (rc != NO_ERROR) {
593 TDB_LOG((tdb, TDB_DEBUG_ERROR, "os2_crtSem: cannot create sem32 %s (rc=%d) called from %s\n",
594 szSem, rc, caller));
595 errno = EINVAL;
596 return -1;
597 }
598 return 0;
599}
600#endif
Note: See TracBrowser for help on using the repository browser.