r19426: merge nearly all the differences between Samba3 tdb and Samba4
[Samba/bb.git] / source / tdb / common / open.c
blobe1f21aa8560b3e02bfadc2e0a61914de30f5f7da
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 int size, ret = -1;
55 /* We make it up in memory, then write it out if not internal */
56 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
57 if (!(newdb = (struct tdb_header *)calloc(size, 1)))
58 return TDB_ERRCODE(TDB_ERR_OOM, -1);
60 /* Fill in the header */
61 newdb->version = TDB_VERSION;
62 newdb->hash_size = hash_size;
63 if (tdb->flags & TDB_INTERNAL) {
64 tdb->map_size = size;
65 tdb->map_ptr = (char *)newdb;
66 memcpy(&tdb->header, newdb, sizeof(tdb->header));
67 /* Convert the `ondisk' version if asked. */
68 CONVERT(*newdb);
69 return 0;
71 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
72 goto fail;
74 if (ftruncate(tdb->fd, 0) == -1)
75 goto fail;
77 /* This creates an endian-converted header, as if read from disk */
78 CONVERT(*newdb);
79 memcpy(&tdb->header, newdb, sizeof(tdb->header));
80 /* Don't endian-convert the magic food! */
81 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
82 if (write(tdb->fd, newdb, size) != size) {
83 ret = -1;
84 } else {
85 ret = 0;
88 fail:
89 SAFE_FREE(newdb);
90 return ret;
95 static int tdb_already_open(dev_t device,
96 ino_t ino)
98 struct tdb_context *i;
100 for (i = tdbs; i; i = i->next) {
101 if (i->device == device && i->inode == ino) {
102 return 1;
106 return 0;
109 /* open the database, creating it if necessary
111 The open_flags and mode are passed straight to the open call on the
112 database file. A flags value of O_WRONLY is invalid. The hash size
113 is advisory, use zero for a default value.
115 Return is NULL on error, in which case errno is also set. Don't
116 try to call tdb_error or tdb_errname, just do strerror(errno).
118 @param name may be NULL for internal databases. */
119 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
120 int open_flags, mode_t mode)
122 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
125 /* a default logging function */
126 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
127 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
132 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
133 int open_flags, mode_t mode,
134 const struct tdb_logging_context *log_ctx,
135 tdb_hash_func hash_fn)
137 struct tdb_context *tdb;
138 struct stat st;
139 int rev = 0, locked = 0;
140 unsigned char *vp;
141 u32 vertest;
143 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
144 /* Can't log this */
145 errno = ENOMEM;
146 goto fail;
148 tdb_io_init(tdb);
149 tdb->fd = -1;
150 tdb->name = NULL;
151 tdb->map_ptr = NULL;
152 tdb->flags = tdb_flags;
153 tdb->open_flags = open_flags;
154 if (log_ctx) {
155 tdb->log = *log_ctx;
156 } else {
157 tdb->log.log_fn = null_log_fn;
158 tdb->log.log_private = NULL;
160 tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
162 /* cache the page size */
163 tdb->page_size = getpagesize();
164 if (tdb->page_size <= 0) {
165 tdb->page_size = 0x2000;
168 if ((open_flags & O_ACCMODE) == O_WRONLY) {
169 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
170 name));
171 errno = EINVAL;
172 goto fail;
175 if (hash_size == 0)
176 hash_size = DEFAULT_HASH_SIZE;
177 if ((open_flags & O_ACCMODE) == O_RDONLY) {
178 tdb->read_only = 1;
179 /* read only databases don't do locking or clear if first */
180 tdb->flags |= TDB_NOLOCK;
181 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
184 /* internal databases don't mmap or lock, and start off cleared */
185 if (tdb->flags & TDB_INTERNAL) {
186 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
187 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
188 if (tdb_new_database(tdb, hash_size) != 0) {
189 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
190 goto fail;
192 goto internal;
195 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
196 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
197 name, strerror(errno)));
198 goto fail; /* errno set by open(2) */
201 /* ensure there is only one process initialising at once */
202 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
203 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
204 name, strerror(errno)));
205 goto fail; /* errno set by tdb_brlock */
208 /* we need to zero database if we are the only one with it open */
209 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
210 (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
211 open_flags |= O_CREAT;
212 if (ftruncate(tdb->fd, 0) == -1) {
213 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
214 "failed to truncate %s: %s\n",
215 name, strerror(errno)));
216 goto fail; /* errno set by ftruncate */
220 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
221 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
222 || (tdb->header.version != TDB_VERSION
223 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
224 /* its not a valid database - possibly initialise it */
225 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
226 errno = EIO; /* ie bad format or something */
227 goto fail;
229 rev = (tdb->flags & TDB_CONVERT);
231 vp = (unsigned char *)&tdb->header.version;
232 vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
233 (((u32)vp[2]) << 8) | (u32)vp[3];
234 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
235 if (!rev)
236 tdb->flags &= ~TDB_CONVERT;
237 else {
238 tdb->flags |= TDB_CONVERT;
239 tdb_convert(&tdb->header, sizeof(tdb->header));
241 if (fstat(tdb->fd, &st) == -1)
242 goto fail;
244 if (tdb->header.rwlocks != 0) {
245 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
246 goto fail;
249 /* Is it already in the open list? If so, fail. */
250 if (tdb_already_open(st.st_dev, st.st_ino)) {
251 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
252 "%s (%d,%d) is already open in this process\n",
253 name, (int)st.st_dev, (int)st.st_ino));
254 errno = EBUSY;
255 goto fail;
258 if (!(tdb->name = (char *)strdup(name))) {
259 errno = ENOMEM;
260 goto fail;
263 tdb->map_size = st.st_size;
264 tdb->device = st.st_dev;
265 tdb->inode = st.st_ino;
266 tdb->locked = (struct tdb_lock_type *)calloc(tdb->header.hash_size+1,
267 sizeof(tdb->locked[0]));
268 if (!tdb->locked) {
269 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
270 "failed to allocate lock structure for %s\n",
271 name));
272 errno = ENOMEM;
273 goto fail;
275 tdb_mmap(tdb);
276 if (locked) {
277 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
278 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
279 "failed to take ACTIVE_LOCK on %s: %s\n",
280 name, strerror(errno)));
281 goto fail;
286 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
287 we didn't get the initial exclusive lock as we need to let all other
288 users know we're using it. */
290 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
291 /* leave this lock in place to indicate it's in use */
292 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
293 goto fail;
296 /* if needed, run recovery */
297 if (tdb_transaction_recover(tdb) == -1) {
298 goto fail;
301 internal:
302 /* Internal (memory-only) databases skip all the code above to
303 * do with disk files, and resume here by releasing their
304 * global lock and hooking into the active list. */
305 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
306 goto fail;
307 tdb->next = tdbs;
308 tdbs = tdb;
309 return tdb;
311 fail:
312 { int save_errno = errno;
314 if (!tdb)
315 return NULL;
317 if (tdb->map_ptr) {
318 if (tdb->flags & TDB_INTERNAL)
319 SAFE_FREE(tdb->map_ptr);
320 else
321 tdb_munmap(tdb);
323 SAFE_FREE(tdb->name);
324 if (tdb->fd != -1)
325 if (close(tdb->fd) != 0)
326 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
327 SAFE_FREE(tdb->locked);
328 SAFE_FREE(tdb);
329 errno = save_errno;
330 return NULL;
335 * Close a database.
337 * @returns -1 for error; 0 for success.
339 int tdb_close(struct tdb_context *tdb)
341 struct tdb_context **i;
342 int ret = 0;
344 if (tdb->transaction) {
345 tdb_transaction_cancel(tdb);
348 if (tdb->map_ptr) {
349 if (tdb->flags & TDB_INTERNAL)
350 SAFE_FREE(tdb->map_ptr);
351 else
352 tdb_munmap(tdb);
354 SAFE_FREE(tdb->name);
355 if (tdb->fd != -1)
356 ret = close(tdb->fd);
357 SAFE_FREE(tdb->locked);
359 /* Remove from contexts list */
360 for (i = &tdbs; *i; i = &(*i)->next) {
361 if (*i == tdb) {
362 *i = tdb->next;
363 break;
367 memset(tdb, 0, sizeof(*tdb));
368 SAFE_FREE(tdb);
370 return ret;
373 /* register a loging function */
374 void tdb_set_logging_function(struct tdb_context *tdb,
375 const struct tdb_logging_context *log)
377 tdb->log = *log;
380 void *tdb_get_logging_private(struct tdb_context *tdb)
382 return tdb->log.log_private;
385 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
386 seek pointer from our parent and to re-establish locks */
387 int tdb_reopen(struct tdb_context *tdb)
389 struct stat st;
391 if (tdb->flags & TDB_INTERNAL) {
392 return 0; /* Nothing to do. */
395 if (tdb->num_locks != 0 || tdb->global_lock.count) {
396 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
397 goto fail;
400 if (tdb->transaction != 0) {
401 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
402 goto fail;
405 if (tdb_munmap(tdb) != 0) {
406 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
407 goto fail;
409 if (close(tdb->fd) != 0)
410 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
411 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
412 if (tdb->fd == -1) {
413 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
414 goto fail;
416 if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
417 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
418 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
419 goto fail;
421 if (fstat(tdb->fd, &st) != 0) {
422 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
423 goto fail;
425 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
426 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
427 goto fail;
429 tdb_mmap(tdb);
431 return 0;
433 fail:
434 tdb_close(tdb);
435 return -1;
438 /* reopen all tdb's */
439 int tdb_reopen_all(int parent_longlived)
441 struct tdb_context *tdb;
443 for (tdb=tdbs; tdb; tdb = tdb->next) {
445 * If the parent is longlived (ie. a
446 * parent daemon architecture), we know
447 * it will keep it's active lock on a
448 * tdb opened with CLEAR_IF_FIRST. Thus
449 * for child processes we don't have to
450 * add an active lock. This is essential
451 * to improve performance on systems that
452 * keep POSIX locks as a non-scalable data
453 * structure in the kernel.
455 if (parent_longlived) {
456 /* Ensure no clear-if-first. */
457 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
460 if (tdb_reopen(tdb) != 0)
461 return -1;
464 return 0;