2 Trivial Database 2: opening and closing TDBs
3 Copyright (C) Rusty Russell 2010
5 This library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 3 of the License, or (at your option) any later version.
10 This library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with this library; if not, see <http://www.gnu.org/licenses/>.
21 /* all tdbs, to detect double-opens (fcntl file don't nest!) */
22 static struct tdb_context *tdbs = NULL;
24 static struct tdb_file *find_file(dev_t device, ino_t ino)
26 struct tdb_context *i;
28 for (i = tdbs; i; i = i->next) {
29 if (i->file->device == device && i->file->inode == ino) {
37 static bool read_all(int fd, void *buf, size_t len)
41 ret = read(fd, buf, len);
49 buf = (char *)buf + ret;
55 static uint64_t random_number(struct tdb_context *tdb)
61 fd = open("/dev/urandom", O_RDONLY);
63 if (read_all(fd, &ret, sizeof(ret))) {
69 /* FIXME: Untested! Based on Wikipedia protocol description! */
70 fd = open("/dev/egd-pool", O_RDWR);
72 /* Command is 1, next byte is size we want to read. */
73 char cmd[2] = { 1, sizeof(uint64_t) };
74 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
75 char reply[1 + sizeof(uint64_t)];
76 int r = read(fd, reply, sizeof(reply));
78 /* Copy at least some bytes. */
79 memcpy(&ret, reply+1, r - 1);
80 if (reply[0] == sizeof(uint64_t)
81 && r == sizeof(reply)) {
90 /* Fallback: pid and time. */
91 gettimeofday(&now, NULL);
92 ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
93 tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
94 "tdb_open: random from getpid and time");
99 struct tdb_header hdr;
100 struct tdb_freetable ftable;
103 /* initialise a new database */
104 static enum TDB_ERROR tdb_new_database(struct tdb_context *tdb,
105 struct tdb_attribute_seed *seed,
106 struct tdb_header *hdr)
108 /* We make it up in memory, then write it out if not internal */
109 struct new_database newdb;
110 unsigned int magic_len;
112 enum TDB_ERROR ecode;
114 /* Fill in the header */
115 newdb.hdr.version = TDB_VERSION;
117 newdb.hdr.hash_seed = seed->seed;
119 newdb.hdr.hash_seed = random_number(tdb);
120 newdb.hdr.hash_test = TDB_HASH_MAGIC;
121 newdb.hdr.hash_test = tdb->hash_fn(&newdb.hdr.hash_test,
122 sizeof(newdb.hdr.hash_test),
125 newdb.hdr.recovery = 0;
126 newdb.hdr.features_used = newdb.hdr.features_offered = TDB_FEATURE_MASK;
127 newdb.hdr.seqnum = 0;
128 memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
129 /* Initial hashes are empty. */
130 memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
133 newdb.hdr.free_table = offsetof(struct new_database, ftable);
134 memset(&newdb.ftable, 0, sizeof(newdb.ftable));
135 ecode = set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
136 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
137 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
139 if (ecode != TDB_SUCCESS) {
144 memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
145 strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
147 /* This creates an endian-converted database, as if read from disk */
148 magic_len = sizeof(newdb.hdr.magic_food);
150 (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
154 if (tdb->flags & TDB_INTERNAL) {
155 tdb->file->map_size = sizeof(newdb);
156 tdb->file->map_ptr = malloc(tdb->file->map_size);
157 if (!tdb->file->map_ptr) {
158 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
160 " failed to allocate");
162 memcpy(tdb->file->map_ptr, &newdb, tdb->file->map_size);
165 if (lseek(tdb->file->fd, 0, SEEK_SET) == -1) {
166 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
168 " failed to seek: %s", strerror(errno));
171 if (ftruncate(tdb->file->fd, 0) == -1) {
172 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
174 " failed to truncate: %s", strerror(errno));
177 rlen = write(tdb->file->fd, &newdb, sizeof(newdb));
178 if (rlen != sizeof(newdb)) {
181 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
182 "tdb_new_database: %zi writing header: %s",
183 rlen, strerror(errno));
188 static enum TDB_ERROR tdb_new_file(struct tdb_context *tdb)
190 tdb->file = malloc(sizeof(*tdb->file));
192 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
193 "tdb_open: cannot alloc tdb_file structure");
194 tdb->file->num_lockrecs = 0;
195 tdb->file->lockrecs = NULL;
196 tdb->file->allrecord_lock.count = 0;
197 tdb->file->refcnt = 1;
201 enum TDB_ERROR tdb_set_attribute(struct tdb_context *tdb,
202 const union tdb_attribute *attr)
204 switch (attr->base.attr) {
205 case TDB_ATTRIBUTE_LOG:
206 tdb->log_fn = attr->log.fn;
207 tdb->log_data = attr->log.data;
209 case TDB_ATTRIBUTE_HASH:
210 case TDB_ATTRIBUTE_SEED:
211 case TDB_ATTRIBUTE_OPENHOOK:
212 return tdb->last_error
213 = tdb_logerr(tdb, TDB_ERR_EINVAL,
216 " cannot set %s after opening",
217 attr->base.attr == TDB_ATTRIBUTE_HASH
218 ? "TDB_ATTRIBUTE_HASH"
219 : attr->base.attr == TDB_ATTRIBUTE_SEED
220 ? "TDB_ATTRIBUTE_SEED"
221 : "TDB_ATTRIBUTE_OPENHOOK");
222 case TDB_ATTRIBUTE_STATS:
223 return tdb->last_error
224 = tdb_logerr(tdb, TDB_ERR_EINVAL,
227 " cannot set TDB_ATTRIBUTE_STATS");
228 case TDB_ATTRIBUTE_FLOCK:
229 tdb->lock_fn = attr->flock.lock;
230 tdb->unlock_fn = attr->flock.unlock;
231 tdb->lock_data = attr->flock.data;
234 return tdb->last_error
235 = tdb_logerr(tdb, TDB_ERR_EINVAL,
238 " unknown attribute type %u",
244 enum TDB_ERROR tdb_get_attribute(struct tdb_context *tdb,
245 union tdb_attribute *attr)
247 switch (attr->base.attr) {
248 case TDB_ATTRIBUTE_LOG:
250 return tdb->last_error = TDB_ERR_NOEXIST;
251 attr->log.fn = tdb->log_fn;
252 attr->log.data = tdb->log_data;
254 case TDB_ATTRIBUTE_HASH:
255 attr->hash.fn = tdb->hash_fn;
256 attr->hash.data = tdb->hash_data;
258 case TDB_ATTRIBUTE_SEED:
259 attr->seed.seed = tdb->hash_seed;
261 case TDB_ATTRIBUTE_OPENHOOK:
263 return tdb->last_error = TDB_ERR_NOEXIST;
264 attr->openhook.fn = tdb->openhook;
265 attr->openhook.data = tdb->openhook_data;
267 case TDB_ATTRIBUTE_STATS: {
268 size_t size = attr->stats.size;
269 if (size > tdb->stats.size)
270 size = tdb->stats.size;
271 memcpy(&attr->stats, &tdb->stats, size);
274 case TDB_ATTRIBUTE_FLOCK:
275 attr->flock.lock = tdb->lock_fn;
276 attr->flock.unlock = tdb->unlock_fn;
277 attr->flock.data = tdb->lock_data;
280 return tdb->last_error
281 = tdb_logerr(tdb, TDB_ERR_EINVAL,
284 " unknown attribute type %u",
287 attr->base.next = NULL;
291 void tdb_unset_attribute(struct tdb_context *tdb,
292 enum tdb_attribute_type type)
295 case TDB_ATTRIBUTE_LOG:
298 case TDB_ATTRIBUTE_OPENHOOK:
299 tdb->openhook = NULL;
301 case TDB_ATTRIBUTE_HASH:
302 case TDB_ATTRIBUTE_SEED:
303 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
304 "tdb_unset_attribute: cannot unset %s after opening",
305 type == TDB_ATTRIBUTE_HASH
306 ? "TDB_ATTRIBUTE_HASH"
307 : "TDB_ATTRIBUTE_SEED");
309 case TDB_ATTRIBUTE_STATS:
310 tdb_logerr(tdb, TDB_ERR_EINVAL,
312 "tdb_unset_attribute:"
313 "cannot unset TDB_ATTRIBUTE_STATS");
315 case TDB_ATTRIBUTE_FLOCK:
316 tdb->lock_fn = tdb_fcntl_lock;
317 tdb->unlock_fn = tdb_fcntl_unlock;
320 tdb_logerr(tdb, TDB_ERR_EINVAL,
322 "tdb_unset_attribute: unknown attribute type %u",
327 struct tdb_context *tdb_open(const char *name, int tdb_flags,
328 int open_flags, mode_t mode,
329 union tdb_attribute *attr)
331 struct tdb_context *tdb;
337 struct tdb_header hdr;
338 struct tdb_attribute_seed *seed = NULL;
340 enum TDB_ERROR ecode;
343 tdb = malloc(sizeof(*tdb) + (name ? strlen(name) + 1 : 0));
349 /* Set name immediately for logging functions. */
351 tdb->name = strcpy((char *)(tdb + 1), name);
355 tdb->flags = tdb_flags;
357 tdb->open_flags = open_flags;
358 tdb->last_error = TDB_SUCCESS;
360 tdb->openhook = NULL;
361 tdb->lock_fn = tdb_fcntl_lock;
362 tdb->unlock_fn = tdb_fcntl_unlock;
363 tdb->hash_fn = tdb_jenkins_hash;
364 memset(&tdb->stats, 0, sizeof(tdb->stats));
365 tdb->stats.base.attr = TDB_ATTRIBUTE_STATS;
366 tdb->stats.size = sizeof(tdb->stats);
368 tdb->tdb2.direct_access = 0;
369 tdb->tdb2.transaction = NULL;
370 tdb->tdb2.access = NULL;
373 switch (attr->base.attr) {
374 case TDB_ATTRIBUTE_HASH:
375 tdb->hash_fn = attr->hash.fn;
376 tdb->hash_data = attr->hash.data;
378 case TDB_ATTRIBUTE_SEED:
381 case TDB_ATTRIBUTE_OPENHOOK:
382 tdb->openhook = attr->openhook.fn;
383 tdb->openhook_data = attr->openhook.data;
386 /* These are set as normal. */
387 ecode = tdb_set_attribute(tdb, attr);
388 if (ecode != TDB_SUCCESS)
391 attr = attr->base.next;
394 if (tdb_flags & ~(TDB_INTERNAL | TDB_NOLOCK | TDB_NOMMAP | TDB_CONVERT
395 | TDB_NOSYNC | TDB_SEQNUM | TDB_ALLOW_NESTING
397 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
398 "tdb_open: unknown flags %u", tdb_flags);
402 if ((open_flags & O_ACCMODE) == O_WRONLY) {
403 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
404 "tdb_open: can't open tdb %s write-only",
409 if ((open_flags & O_ACCMODE) == O_RDONLY) {
411 tdb->flags |= TDB_RDONLY;
413 if (tdb_flags & TDB_RDONLY) {
414 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL,
416 "tdb_open: can't use TDB_RDONLY"
417 " without O_RDONLY");
423 /* internal databases don't need any of the rest. */
424 if (tdb->flags & TDB_INTERNAL) {
425 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
426 ecode = tdb_new_file(tdb);
427 if (ecode != TDB_SUCCESS) {
431 ecode = tdb_new_database(tdb, seed, &hdr);
432 if (ecode != TDB_SUCCESS) {
435 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
436 tdb->hash_seed = hdr.hash_seed;
437 tdb_ftable_init(tdb);
441 if (stat(name, &st) != -1)
442 tdb->file = find_file(st.st_dev, st.st_ino);
447 if ((fd = open(name, open_flags, mode)) == -1) {
448 /* errno set by open(2) */
450 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
451 "tdb_open: could not open file %s: %s",
452 name, strerror(errno));
456 /* on exec, don't inherit the fd */
457 v = fcntl(fd, F_GETFD, 0);
458 fcntl(fd, F_SETFD, v | FD_CLOEXEC);
460 if (fstat(fd, &st) == -1) {
462 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
463 "tdb_open: could not stat open %s: %s",
464 name, strerror(errno));
469 ecode = tdb_new_file(tdb);
470 if (ecode != TDB_SUCCESS) {
476 tdb->file->device = st.st_dev;
477 tdb->file->inode = st.st_ino;
478 tdb->file->map_ptr = NULL;
479 tdb->file->map_size = sizeof(struct tdb_header);
482 /* ensure there is only one process initialising at once */
483 ecode = tdb_lock_open(tdb, openlock, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
484 if (ecode != TDB_SUCCESS) {
489 /* call their open hook if they gave us one. */
491 ecode = tdb->openhook(tdb->file->fd, tdb->openhook_data);
492 if (ecode != TDB_SUCCESS) {
493 tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
494 "tdb_open: open hook failed");
497 open_flags |= O_CREAT;
500 /* If they used O_TRUNC, read will return 0. */
501 rlen = pread(tdb->file->fd, &hdr, sizeof(hdr), 0);
502 if (rlen == 0 && (open_flags & O_CREAT)) {
503 ecode = tdb_new_database(tdb, seed, &hdr);
504 if (ecode != TDB_SUCCESS) {
507 } else if (rlen < 0) {
508 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
509 "tdb_open: error %s reading %s",
510 strerror(errno), name);
512 } else if (rlen < sizeof(hdr)
513 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
514 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
515 "tdb_open: %s is not a tdb file", name);
519 if (hdr.version != TDB_VERSION) {
520 if (hdr.version == bswap_64(TDB_VERSION))
521 tdb->flags |= TDB_CONVERT;
524 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
526 " %s is unknown version 0x%llx",
527 name, (long long)hdr.version);
530 } else if (tdb->flags & TDB_CONVERT) {
531 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
533 " %s does not need TDB_CONVERT",
538 tdb_convert(tdb, &hdr, sizeof(hdr));
539 tdb->hash_seed = hdr.hash_seed;
540 hash_test = TDB_HASH_MAGIC;
541 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
542 if (hdr.hash_test != hash_test) {
543 /* wrong hash variant */
544 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
546 " %s uses a different hash function",
551 /* Clear any features we don't understand. */
552 if ((open_flags & O_ACCMODE) != O_RDONLY) {
553 hdr.features_used &= TDB_FEATURE_MASK;
554 ecode = tdb_write_convert(tdb, offsetof(struct tdb_header,
557 sizeof(hdr.features_used));
558 if (ecode != TDB_SUCCESS)
562 tdb_unlock_open(tdb, openlock);
564 /* This make sure we have current map_size and mmap. */
565 ecode = tdb->tdb2.io->oob(tdb, tdb->file->map_size + 1, true);
566 if (unlikely(ecode != TDB_SUCCESS))
569 /* Now it's fully formed, recover if necessary. */
570 berr = tdb_needs_recovery(tdb);
571 if (unlikely(berr != false)) {
576 ecode = tdb_lock_and_recover(tdb);
577 if (ecode != TDB_SUCCESS) {
582 ecode = tdb_ftable_init(tdb);
583 if (ecode != TDB_SUCCESS) {
592 /* Map ecode to some logical errno. */
594 case TDB_ERR_CORRUPT:
599 saved_errno = EWOULDBLOCK;
602 saved_errno = ENOMEM;
605 saved_errno = EINVAL;
608 saved_errno = EINVAL;
617 tdb_lock_cleanup(tdb);
618 if (--tdb->file->refcnt == 0) {
619 assert(tdb->file->num_lockrecs == 0);
620 if (tdb->file->map_ptr) {
621 if (tdb->flags & TDB_INTERNAL) {
622 free(tdb->file->map_ptr);
624 tdb_munmap(tdb->file);
626 if (close(tdb->file->fd) != 0)
627 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
628 "tdb_open: failed to close tdb fd"
629 " on error: %s", strerror(errno));
630 free(tdb->file->lockrecs);
640 int tdb_close(struct tdb_context *tdb)
643 struct tdb_context **i;
645 tdb_trace(tdb, "tdb_close");
647 if (tdb->tdb2.transaction) {
648 tdb_transaction_cancel(tdb);
651 if (tdb->file->map_ptr) {
652 if (tdb->flags & TDB_INTERNAL)
653 free(tdb->file->map_ptr);
655 tdb_munmap(tdb->file);
658 tdb_lock_cleanup(tdb);
659 if (--tdb->file->refcnt == 0) {
660 ret = close(tdb->file->fd);
661 free(tdb->file->lockrecs);
666 /* Remove from tdbs list */
667 for (i = &tdbs; *i; i = &(*i)->next) {
682 void tdb_foreach_(int (*fn)(struct tdb_context *, void *), void *p)
684 struct tdb_context *i;
686 for (i = tdbs; i; i = i->next) {