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/>.
19 #include <ccan/hash/hash.h>
22 /* all lock info, to detect double-opens (fcntl file don't nest!) */
23 static struct tdb_file *files = NULL;
25 static struct tdb_file *find_file(dev_t device, ino_t ino)
29 for (i = files; i; i = i->next) {
30 if (i->device == device && i->inode == ino) {
38 static bool read_all(int fd, void *buf, size_t len)
42 ret = read(fd, buf, len);
50 buf = (char *)buf + ret;
56 static uint64_t random_number(struct tdb_context *tdb)
62 fd = open("/dev/urandom", O_RDONLY);
64 if (read_all(fd, &ret, sizeof(ret))) {
70 /* FIXME: Untested! Based on Wikipedia protocol description! */
71 fd = open("/dev/egd-pool", O_RDWR);
73 /* Command is 1, next byte is size we want to read. */
74 char cmd[2] = { 1, sizeof(uint64_t) };
75 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
76 char reply[1 + sizeof(uint64_t)];
77 int r = read(fd, reply, sizeof(reply));
79 /* Copy at least some bytes. */
80 memcpy(&ret, reply+1, r - 1);
81 if (reply[0] == sizeof(uint64_t)
82 && r == sizeof(reply)) {
91 /* Fallback: pid and time. */
92 gettimeofday(&now, NULL);
93 ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
94 tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
95 "tdb_open: random from getpid and time");
100 struct tdb_header hdr;
101 struct tdb_freetable ftable;
104 /* initialise a new database */
105 static enum TDB_ERROR tdb_new_database(struct tdb_context *tdb,
106 struct tdb_attribute_seed *seed,
107 struct tdb_header *hdr)
109 /* We make it up in memory, then write it out if not internal */
110 struct new_database newdb;
111 unsigned int magic_len;
113 enum TDB_ERROR ecode;
115 /* Fill in the header */
116 newdb.hdr.version = TDB_VERSION;
118 newdb.hdr.hash_seed = seed->seed;
120 newdb.hdr.hash_seed = random_number(tdb);
121 newdb.hdr.hash_test = TDB_HASH_MAGIC;
122 newdb.hdr.hash_test = tdb->hash_fn(&newdb.hdr.hash_test,
123 sizeof(newdb.hdr.hash_test),
126 newdb.hdr.recovery = 0;
127 newdb.hdr.features_used = newdb.hdr.features_offered = TDB_FEATURE_MASK;
128 newdb.hdr.seqnum = 0;
129 memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
130 /* Initial hashes are empty. */
131 memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
134 newdb.hdr.free_table = offsetof(struct new_database, ftable);
135 memset(&newdb.ftable, 0, sizeof(newdb.ftable));
136 ecode = set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
137 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
138 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
140 if (ecode != TDB_SUCCESS) {
145 memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
146 strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
148 /* This creates an endian-converted database, as if read from disk */
149 magic_len = sizeof(newdb.hdr.magic_food);
151 (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
155 if (tdb->flags & TDB_INTERNAL) {
156 tdb->file->map_size = sizeof(newdb);
157 tdb->file->map_ptr = malloc(tdb->file->map_size);
158 if (!tdb->file->map_ptr) {
159 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
161 " failed to allocate");
163 memcpy(tdb->file->map_ptr, &newdb, tdb->file->map_size);
166 if (lseek(tdb->file->fd, 0, SEEK_SET) == -1) {
167 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
169 " failed to seek: %s", strerror(errno));
172 if (ftruncate(tdb->file->fd, 0) == -1) {
173 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
175 " failed to truncate: %s", strerror(errno));
178 rlen = write(tdb->file->fd, &newdb, sizeof(newdb));
179 if (rlen != sizeof(newdb)) {
182 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
183 "tdb_new_database: %zi writing header: %s",
184 rlen, strerror(errno));
189 static enum TDB_ERROR tdb_new_file(struct tdb_context *tdb)
191 tdb->file = malloc(sizeof(*tdb->file));
193 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
194 "tdb_open: cannot alloc tdb_file structure");
195 tdb->file->num_lockrecs = 0;
196 tdb->file->lockrecs = NULL;
197 tdb->file->allrecord_lock.count = 0;
198 tdb->file->refcnt = 1;
202 enum TDB_ERROR tdb_set_attribute(struct tdb_context *tdb,
203 const union tdb_attribute *attr)
205 switch (attr->base.attr) {
206 case TDB_ATTRIBUTE_LOG:
207 tdb->log_fn = attr->log.fn;
208 tdb->log_data = attr->log.data;
210 case TDB_ATTRIBUTE_HASH:
211 case TDB_ATTRIBUTE_SEED:
212 case TDB_ATTRIBUTE_OPENHOOK:
213 return tdb->last_error
214 = tdb_logerr(tdb, TDB_ERR_EINVAL,
217 " cannot set %s after opening",
218 attr->base.attr == TDB_ATTRIBUTE_HASH
219 ? "TDB_ATTRIBUTE_HASH"
220 : attr->base.attr == TDB_ATTRIBUTE_SEED
221 ? "TDB_ATTRIBUTE_SEED"
222 : "TDB_ATTRIBUTE_OPENHOOK");
223 case TDB_ATTRIBUTE_STATS:
224 return tdb->last_error
225 = tdb_logerr(tdb, TDB_ERR_EINVAL,
228 " cannot set TDB_ATTRIBUTE_STATS");
229 case TDB_ATTRIBUTE_FLOCK:
230 tdb->lock_fn = attr->flock.lock;
231 tdb->unlock_fn = attr->flock.unlock;
232 tdb->lock_data = attr->flock.data;
235 return tdb->last_error
236 = tdb_logerr(tdb, TDB_ERR_EINVAL,
239 " unknown attribute type %u",
245 static uint64_t jenkins_hash(const void *key, size_t length, uint64_t seed,
249 /* hash64_stable assumes lower bits are more important; they are a
250 * slightly better hash. We use the upper bits first, so swap them. */
251 ret = hash64_stable((const unsigned char *)key, length, seed);
252 return (ret >> 32) | (ret << 32);
255 enum TDB_ERROR tdb_get_attribute(struct tdb_context *tdb,
256 union tdb_attribute *attr)
258 switch (attr->base.attr) {
259 case TDB_ATTRIBUTE_LOG:
261 return tdb->last_error = TDB_ERR_NOEXIST;
262 attr->log.fn = tdb->log_fn;
263 attr->log.data = tdb->log_data;
265 case TDB_ATTRIBUTE_HASH:
266 attr->hash.fn = tdb->hash_fn;
267 attr->hash.data = tdb->hash_data;
269 case TDB_ATTRIBUTE_SEED:
270 attr->seed.seed = tdb->hash_seed;
272 case TDB_ATTRIBUTE_OPENHOOK:
274 return tdb->last_error = TDB_ERR_NOEXIST;
275 attr->openhook.fn = tdb->openhook;
276 attr->openhook.data = tdb->openhook_data;
278 case TDB_ATTRIBUTE_STATS: {
279 size_t size = attr->stats.size;
280 if (size > tdb->stats.size)
281 size = tdb->stats.size;
282 memcpy(&attr->stats, &tdb->stats, size);
285 case TDB_ATTRIBUTE_FLOCK:
286 attr->flock.lock = tdb->lock_fn;
287 attr->flock.unlock = tdb->unlock_fn;
288 attr->flock.data = tdb->lock_data;
291 return tdb->last_error
292 = tdb_logerr(tdb, TDB_ERR_EINVAL,
295 " unknown attribute type %u",
298 attr->base.next = NULL;
302 void tdb_unset_attribute(struct tdb_context *tdb,
303 enum tdb_attribute_type type)
306 case TDB_ATTRIBUTE_LOG:
309 case TDB_ATTRIBUTE_OPENHOOK:
310 tdb->openhook = NULL;
312 case TDB_ATTRIBUTE_HASH:
313 case TDB_ATTRIBUTE_SEED:
314 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
315 "tdb_unset_attribute: cannot unset %s after opening",
316 type == TDB_ATTRIBUTE_HASH
317 ? "TDB_ATTRIBUTE_HASH"
318 : "TDB_ATTRIBUTE_SEED");
320 case TDB_ATTRIBUTE_STATS:
321 tdb_logerr(tdb, TDB_ERR_EINVAL,
323 "tdb_unset_attribute:"
324 "cannot unset TDB_ATTRIBUTE_STATS");
326 case TDB_ATTRIBUTE_FLOCK:
327 tdb->lock_fn = tdb_fcntl_lock;
328 tdb->unlock_fn = tdb_fcntl_unlock;
331 tdb_logerr(tdb, TDB_ERR_EINVAL,
333 "tdb_unset_attribute: unknown attribute type %u",
338 struct tdb_context *tdb_open(const char *name, int tdb_flags,
339 int open_flags, mode_t mode,
340 union tdb_attribute *attr)
342 struct tdb_context *tdb;
348 struct tdb_header hdr;
349 struct tdb_attribute_seed *seed = NULL;
351 enum TDB_ERROR ecode;
354 tdb = malloc(sizeof(*tdb) + (name ? strlen(name) + 1 : 0));
360 /* Set name immediately for logging functions. */
362 tdb->name = strcpy((char *)(tdb + 1), name);
366 tdb->direct_access = 0;
367 tdb->flags = tdb_flags;
369 tdb->transaction = NULL;
371 tdb->open_flags = open_flags;
372 tdb->last_error = TDB_SUCCESS;
374 tdb->openhook = NULL;
375 tdb->lock_fn = tdb_fcntl_lock;
376 tdb->unlock_fn = tdb_fcntl_unlock;
377 tdb->hash_fn = jenkins_hash;
378 memset(&tdb->stats, 0, sizeof(tdb->stats));
379 tdb->stats.base.attr = TDB_ATTRIBUTE_STATS;
380 tdb->stats.size = sizeof(tdb->stats);
384 switch (attr->base.attr) {
385 case TDB_ATTRIBUTE_HASH:
386 tdb->hash_fn = attr->hash.fn;
387 tdb->hash_data = attr->hash.data;
389 case TDB_ATTRIBUTE_SEED:
392 case TDB_ATTRIBUTE_OPENHOOK:
393 tdb->openhook = attr->openhook.fn;
394 tdb->openhook_data = attr->openhook.data;
397 /* These are set as normal. */
398 ecode = tdb_set_attribute(tdb, attr);
399 if (ecode != TDB_SUCCESS)
402 attr = attr->base.next;
405 if (tdb_flags & ~(TDB_INTERNAL | TDB_NOLOCK | TDB_NOMMAP | TDB_CONVERT
406 | TDB_NOSYNC | TDB_SEQNUM | TDB_ALLOW_NESTING
408 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
409 "tdb_open: unknown flags %u", tdb_flags);
413 if ((open_flags & O_ACCMODE) == O_WRONLY) {
414 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
415 "tdb_open: can't open tdb %s write-only",
420 if ((open_flags & O_ACCMODE) == O_RDONLY) {
422 tdb->flags |= TDB_RDONLY;
424 if (tdb_flags & TDB_RDONLY) {
425 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL,
427 "tdb_open: can't use TDB_RDONLY"
428 " without O_RDONLY");
434 /* internal databases don't need any of the rest. */
435 if (tdb->flags & TDB_INTERNAL) {
436 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
437 ecode = tdb_new_file(tdb);
438 if (ecode != TDB_SUCCESS) {
442 ecode = tdb_new_database(tdb, seed, &hdr);
443 if (ecode != TDB_SUCCESS) {
446 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
447 tdb->hash_seed = hdr.hash_seed;
448 tdb_ftable_init(tdb);
452 if (stat(name, &st) != -1)
453 tdb->file = find_file(st.st_dev, st.st_ino);
458 if ((fd = open(name, open_flags, mode)) == -1) {
459 /* errno set by open(2) */
461 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
462 "tdb_open: could not open file %s: %s",
463 name, strerror(errno));
467 /* on exec, don't inherit the fd */
468 v = fcntl(fd, F_GETFD, 0);
469 fcntl(fd, F_SETFD, v | FD_CLOEXEC);
471 if (fstat(fd, &st) == -1) {
473 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
474 "tdb_open: could not stat open %s: %s",
475 name, strerror(errno));
480 ecode = tdb_new_file(tdb);
481 if (ecode != TDB_SUCCESS) {
486 tdb->file->next = files;
488 tdb->file->device = st.st_dev;
489 tdb->file->inode = st.st_ino;
490 tdb->file->map_ptr = NULL;
491 tdb->file->map_size = sizeof(struct tdb_header);
494 /* ensure there is only one process initialising at once */
495 ecode = tdb_lock_open(tdb, openlock, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
496 if (ecode != TDB_SUCCESS) {
501 /* call their open hook if they gave us one. */
503 ecode = tdb->openhook(tdb->file->fd, tdb->openhook_data);
504 if (ecode != TDB_SUCCESS) {
505 tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
506 "tdb_open: open hook failed");
509 open_flags |= O_CREAT;
512 /* If they used O_TRUNC, read will return 0. */
513 rlen = pread(tdb->file->fd, &hdr, sizeof(hdr), 0);
514 if (rlen == 0 && (open_flags & O_CREAT)) {
515 ecode = tdb_new_database(tdb, seed, &hdr);
516 if (ecode != TDB_SUCCESS) {
519 } else if (rlen < 0) {
520 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
521 "tdb_open: error %s reading %s",
522 strerror(errno), name);
524 } else if (rlen < sizeof(hdr)
525 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
526 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
527 "tdb_open: %s is not a tdb file", name);
531 if (hdr.version != TDB_VERSION) {
532 if (hdr.version == bswap_64(TDB_VERSION))
533 tdb->flags |= TDB_CONVERT;
536 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
538 " %s is unknown version 0x%llx",
539 name, (long long)hdr.version);
542 } else if (tdb->flags & TDB_CONVERT) {
543 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
545 " %s does not need TDB_CONVERT",
550 tdb_convert(tdb, &hdr, sizeof(hdr));
551 tdb->hash_seed = hdr.hash_seed;
552 hash_test = TDB_HASH_MAGIC;
553 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
554 if (hdr.hash_test != hash_test) {
555 /* wrong hash variant */
556 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
558 " %s uses a different hash function",
563 /* Clear any features we don't understand. */
564 if ((open_flags & O_ACCMODE) != O_RDONLY) {
565 hdr.features_used &= TDB_FEATURE_MASK;
566 ecode = tdb_write_convert(tdb, offsetof(struct tdb_header,
569 sizeof(hdr.features_used));
570 if (ecode != TDB_SUCCESS)
574 tdb_unlock_open(tdb, openlock);
576 /* This make sure we have current map_size and mmap. */
577 ecode = tdb->methods->oob(tdb, tdb->file->map_size + 1, true);
578 if (unlikely(ecode != TDB_SUCCESS))
581 /* Now it's fully formed, recover if necessary. */
582 berr = tdb_needs_recovery(tdb);
583 if (unlikely(berr != false)) {
588 ecode = tdb_lock_and_recover(tdb);
589 if (ecode != TDB_SUCCESS) {
594 ecode = tdb_ftable_init(tdb);
595 if (ecode != TDB_SUCCESS) {
599 /* Add to linked list if we're new. */
600 if (tdb->file->refcnt == 1)
605 /* Map ecode to some logical errno. */
607 case TDB_ERR_CORRUPT:
612 saved_errno = EWOULDBLOCK;
615 saved_errno = ENOMEM;
618 saved_errno = EINVAL;
621 saved_errno = EINVAL;
630 tdb_lock_cleanup(tdb);
631 if (--tdb->file->refcnt == 0) {
632 assert(tdb->file->num_lockrecs == 0);
633 if (tdb->file->map_ptr) {
634 if (tdb->flags & TDB_INTERNAL) {
635 free(tdb->file->map_ptr);
637 tdb_munmap(tdb->file);
639 if (close(tdb->file->fd) != 0)
640 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
641 "tdb_open: failed to close tdb fd"
642 " on error: %s", strerror(errno));
643 free(tdb->file->lockrecs);
653 int tdb_close(struct tdb_context *tdb)
657 tdb_trace(tdb, "tdb_close");
659 if (tdb->transaction) {
660 tdb_transaction_cancel(tdb);
663 if (tdb->file->map_ptr) {
664 if (tdb->flags & TDB_INTERNAL)
665 free(tdb->file->map_ptr);
667 tdb_munmap(tdb->file);
672 tdb_lock_cleanup(tdb);
673 if (--tdb->file->refcnt == 0) {
674 ret = close(tdb->file->fd);
676 /* Remove from files list */
677 for (i = &files; *i; i = &(*i)->next) {
678 if (*i == tdb->file) {
679 *i = tdb->file->next;
683 free(tdb->file->lockrecs);