[GLUE] Rsync SAMBA_3_0 SVN r25598 in order to create the v3-0-test branch.
[Samba.git] / source / tdb / common / open.c
blob402352d3af208088627318d19cd3dc1ad61da7e0
1 /*
2 Unix SMB/CIFS implementation.
4 trivial database library
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
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
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 2 of the License, or (at your option) any later version.
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.
24 You should have received a copy of the GNU Lesser General Public
25 License along with this library; if not, write to the Free Software
26 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
29 #include "tdb_private.h"
31 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
32 static struct tdb_context *tdbs = NULL;
35 /* This is based on the hash algorithm from gdbm */
36 static unsigned int default_tdb_hash(TDB_DATA *key)
38 u32 value; /* Used to compute the hash value. */
39 u32 i; /* Used to cycle through random values. */
41 /* Set the initial value from the key size. */
42 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
43 value = (value + (key->dptr[i] << (i*5 % 24)));
45 return (1103515243 * value + 12345);
49 /* initialise a new database with a specified hash size */
50 static int tdb_new_database(struct tdb_context *tdb, int hash_size)
52 struct tdb_header *newdb;
53 size_t size;
54 int ret = -1;
55 ssize_t written;
57 /* We make it up in memory, then write it out if not internal */
58 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
59 if (!(newdb = (struct tdb_header *)calloc(size, 1)))
60 return TDB_ERRCODE(TDB_ERR_OOM, -1);
62 /* Fill in the header */
63 newdb->version = TDB_VERSION;
64 newdb->hash_size = hash_size;
65 if (tdb->flags & TDB_INTERNAL) {
66 tdb->map_size = size;
67 tdb->map_ptr = (char *)newdb;
68 memcpy(&tdb->header, newdb, sizeof(tdb->header));
69 /* Convert the `ondisk' version if asked. */
70 CONVERT(*newdb);
71 return 0;
73 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
74 goto fail;
76 if (ftruncate(tdb->fd, 0) == -1)
77 goto fail;
79 /* This creates an endian-converted header, as if read from disk */
80 CONVERT(*newdb);
81 memcpy(&tdb->header, newdb, sizeof(tdb->header));
82 /* Don't endian-convert the magic food! */
83 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
84 /* we still have "ret == -1" here */
85 written = write(tdb->fd, newdb, size);
86 if (written == size) {
87 ret = 0;
88 } else if (written != -1) {
89 /* call write once again, this usually should return -1 and
90 * set errno appropriately */
91 size -= written;
92 written = write(tdb->fd, newdb+written, size);
93 if (written == size) {
94 ret = 0;
95 } else if (written >= 0) {
96 /* a second incomplete write - we give up.
97 * guessing the errno... */
98 errno = ENOSPC;
102 fail:
103 SAFE_FREE(newdb);
104 return ret;
109 static int tdb_already_open(dev_t device,
110 ino_t ino)
112 struct tdb_context *i;
114 for (i = tdbs; i; i = i->next) {
115 if (i->device == device && i->inode == ino) {
116 return 1;
120 return 0;
123 /* open the database, creating it if necessary
125 The open_flags and mode are passed straight to the open call on the
126 database file. A flags value of O_WRONLY is invalid. The hash size
127 is advisory, use zero for a default value.
129 Return is NULL on error, in which case errno is also set. Don't
130 try to call tdb_error or tdb_errname, just do strerror(errno).
132 @param name may be NULL for internal databases. */
133 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
134 int open_flags, mode_t mode)
136 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
139 /* a default logging function */
140 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
141 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
146 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
147 int open_flags, mode_t mode,
148 const struct tdb_logging_context *log_ctx,
149 tdb_hash_func hash_fn)
151 struct tdb_context *tdb;
152 struct stat st;
153 int rev = 0, locked = 0;
154 unsigned char *vp;
155 u32 vertest;
157 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
158 /* Can't log this */
159 errno = ENOMEM;
160 goto fail;
162 tdb_io_init(tdb);
163 tdb->fd = -1;
164 tdb->name = NULL;
165 tdb->map_ptr = NULL;
166 tdb->flags = tdb_flags;
167 tdb->open_flags = open_flags;
168 if (log_ctx) {
169 tdb->log = *log_ctx;
170 } else {
171 tdb->log.log_fn = null_log_fn;
172 tdb->log.log_private = NULL;
174 tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
176 /* cache the page size */
177 tdb->page_size = getpagesize();
178 if (tdb->page_size <= 0) {
179 tdb->page_size = 0x2000;
182 if ((open_flags & O_ACCMODE) == O_WRONLY) {
183 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
184 name));
185 errno = EINVAL;
186 goto fail;
189 if (hash_size == 0)
190 hash_size = DEFAULT_HASH_SIZE;
191 if ((open_flags & O_ACCMODE) == O_RDONLY) {
192 tdb->read_only = 1;
193 /* read only databases don't do locking or clear if first */
194 tdb->flags |= TDB_NOLOCK;
195 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
198 /* internal databases don't mmap or lock, and start off cleared */
199 if (tdb->flags & TDB_INTERNAL) {
200 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
201 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
202 if (tdb_new_database(tdb, hash_size) != 0) {
203 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
204 goto fail;
206 goto internal;
209 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
210 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
211 name, strerror(errno)));
212 goto fail; /* errno set by open(2) */
215 /* ensure there is only one process initialising at once */
216 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
217 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
218 name, strerror(errno)));
219 goto fail; /* errno set by tdb_brlock */
222 /* we need to zero database if we are the only one with it open */
223 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
224 (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
225 open_flags |= O_CREAT;
226 if (ftruncate(tdb->fd, 0) == -1) {
227 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
228 "failed to truncate %s: %s\n",
229 name, strerror(errno)));
230 goto fail; /* errno set by ftruncate */
234 errno = 0;
235 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
236 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
237 || (tdb->header.version != TDB_VERSION
238 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
239 /* its not a valid database - possibly initialise it */
240 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
241 if (errno == 0) {
242 errno = EIO; /* ie bad format or something */
244 goto fail;
246 rev = (tdb->flags & TDB_CONVERT);
248 vp = (unsigned char *)&tdb->header.version;
249 vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
250 (((u32)vp[2]) << 8) | (u32)vp[3];
251 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
252 if (!rev)
253 tdb->flags &= ~TDB_CONVERT;
254 else {
255 tdb->flags |= TDB_CONVERT;
256 tdb_convert(&tdb->header, sizeof(tdb->header));
258 if (fstat(tdb->fd, &st) == -1)
259 goto fail;
261 if (tdb->header.rwlocks != 0) {
262 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
263 goto fail;
266 /* Is it already in the open list? If so, fail. */
267 if (tdb_already_open(st.st_dev, st.st_ino)) {
268 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
269 "%s (%d,%d) is already open in this process\n",
270 name, (int)st.st_dev, (int)st.st_ino));
271 errno = EBUSY;
272 goto fail;
275 if (!(tdb->name = (char *)strdup(name))) {
276 errno = ENOMEM;
277 goto fail;
280 tdb->map_size = st.st_size;
281 tdb->device = st.st_dev;
282 tdb->inode = st.st_ino;
283 tdb->max_dead_records = 0;
284 tdb_mmap(tdb);
285 if (locked) {
286 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
287 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
288 "failed to take ACTIVE_LOCK on %s: %s\n",
289 name, strerror(errno)));
290 goto fail;
295 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
296 we didn't get the initial exclusive lock as we need to let all other
297 users know we're using it. */
299 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
300 /* leave this lock in place to indicate it's in use */
301 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
302 goto fail;
305 /* if needed, run recovery */
306 if (tdb_transaction_recover(tdb) == -1) {
307 goto fail;
310 internal:
311 /* Internal (memory-only) databases skip all the code above to
312 * do with disk files, and resume here by releasing their
313 * global lock and hooking into the active list. */
314 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
315 goto fail;
316 tdb->next = tdbs;
317 tdbs = tdb;
318 return tdb;
320 fail:
321 { int save_errno = errno;
323 if (!tdb)
324 return NULL;
326 if (tdb->map_ptr) {
327 if (tdb->flags & TDB_INTERNAL)
328 SAFE_FREE(tdb->map_ptr);
329 else
330 tdb_munmap(tdb);
332 SAFE_FREE(tdb->name);
333 if (tdb->fd != -1)
334 if (close(tdb->fd) != 0)
335 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
336 SAFE_FREE(tdb);
337 errno = save_errno;
338 return NULL;
343 * Set the maximum number of dead records per hash chain
346 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
348 tdb->max_dead_records = max_dead;
352 * Close a database.
354 * @returns -1 for error; 0 for success.
356 int tdb_close(struct tdb_context *tdb)
358 struct tdb_context **i;
359 int ret = 0;
361 if (tdb->transaction) {
362 tdb_transaction_cancel(tdb);
365 if (tdb->map_ptr) {
366 if (tdb->flags & TDB_INTERNAL)
367 SAFE_FREE(tdb->map_ptr);
368 else
369 tdb_munmap(tdb);
371 SAFE_FREE(tdb->name);
372 if (tdb->fd != -1)
373 ret = close(tdb->fd);
374 SAFE_FREE(tdb->lockrecs);
376 /* Remove from contexts list */
377 for (i = &tdbs; *i; i = &(*i)->next) {
378 if (*i == tdb) {
379 *i = tdb->next;
380 break;
384 memset(tdb, 0, sizeof(*tdb));
385 SAFE_FREE(tdb);
387 return ret;
390 /* register a loging function */
391 void tdb_set_logging_function(struct tdb_context *tdb,
392 const struct tdb_logging_context *log_ctx)
394 tdb->log = *log_ctx;
397 void *tdb_get_logging_private(struct tdb_context *tdb)
399 return tdb->log.log_private;
402 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
403 seek pointer from our parent and to re-establish locks */
404 int tdb_reopen(struct tdb_context *tdb)
406 struct stat st;
408 if (tdb->flags & TDB_INTERNAL) {
409 return 0; /* Nothing to do. */
412 if (tdb->num_locks != 0 || tdb->global_lock.count) {
413 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
414 goto fail;
417 if (tdb->transaction != 0) {
418 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
419 goto fail;
422 if (tdb_munmap(tdb) != 0) {
423 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
424 goto fail;
426 if (close(tdb->fd) != 0)
427 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
428 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
429 if (tdb->fd == -1) {
430 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
431 goto fail;
433 if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
434 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
435 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
436 goto fail;
438 if (fstat(tdb->fd, &st) != 0) {
439 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
440 goto fail;
442 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
443 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
444 goto fail;
446 tdb_mmap(tdb);
448 return 0;
450 fail:
451 tdb_close(tdb);
452 return -1;
455 /* reopen all tdb's */
456 int tdb_reopen_all(int parent_longlived)
458 struct tdb_context *tdb;
460 for (tdb=tdbs; tdb; tdb = tdb->next) {
462 * If the parent is longlived (ie. a
463 * parent daemon architecture), we know
464 * it will keep it's active lock on a
465 * tdb opened with CLEAR_IF_FIRST. Thus
466 * for child processes we don't have to
467 * add an active lock. This is essential
468 * to improve performance on systems that
469 * keep POSIX locks as a non-scalable data
470 * structure in the kernel.
472 if (parent_longlived) {
473 /* Ensure no clear-if-first. */
474 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
477 if (tdb_reopen(tdb) != 0)
478 return -1;
481 return 0;