2 #include <ccan/tdb2/tdb2.h>
7 struct tdb_data tdb_null = { .dptr = NULL, .dsize = 0 };
9 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
10 static struct tdb_context *tdbs = NULL;
12 static bool tdb_already_open(dev_t device, ino_t ino)
14 struct tdb_context *i;
16 for (i = tdbs; i; i = i->next) {
17 if (i->device == device && i->inode == ino) {
25 static bool read_all(int fd, void *buf, size_t len)
29 ret = read(fd, buf, len);
37 buf = (char *)buf + ret;
43 static uint64_t random_number(struct tdb_context *tdb)
49 fd = open("/dev/urandom", O_RDONLY);
51 if (read_all(fd, &ret, sizeof(ret))) {
57 /* FIXME: Untested! Based on Wikipedia protocol description! */
58 fd = open("/dev/egd-pool", O_RDWR);
60 /* Command is 1, next byte is size we want to read. */
61 char cmd[2] = { 1, sizeof(uint64_t) };
62 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
63 char reply[1 + sizeof(uint64_t)];
64 int r = read(fd, reply, sizeof(reply));
66 /* Copy at least some bytes. */
67 memcpy(&ret, reply+1, r - 1);
68 if (reply[0] == sizeof(uint64_t)
69 && r == sizeof(reply)) {
78 /* Fallback: pid and time. */
79 gettimeofday(&now, NULL);
80 ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
81 tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
82 "tdb_open: random from getpid and time");
87 struct tdb_header hdr;
88 struct tdb_freetable ftable;
91 /* initialise a new database */
92 static enum TDB_ERROR tdb_new_database(struct tdb_context *tdb,
93 struct tdb_attribute_seed *seed,
94 struct tdb_header *hdr)
96 /* We make it up in memory, then write it out if not internal */
97 struct new_database newdb;
98 unsigned int magic_len;
100 enum TDB_ERROR ecode;
102 /* Fill in the header */
103 newdb.hdr.version = TDB_VERSION;
105 newdb.hdr.hash_seed = seed->seed;
107 newdb.hdr.hash_seed = random_number(tdb);
108 newdb.hdr.hash_test = TDB_HASH_MAGIC;
109 newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
110 sizeof(newdb.hdr.hash_test),
113 newdb.hdr.recovery = 0;
114 memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
115 /* Initial hashes are empty. */
116 memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
119 newdb.hdr.free_table = offsetof(struct new_database, ftable);
120 memset(&newdb.ftable, 0, sizeof(newdb.ftable));
121 ecode = set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
122 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
123 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
125 if (ecode != TDB_SUCCESS) {
130 memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
131 strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
133 /* This creates an endian-converted database, as if read from disk */
134 magic_len = sizeof(newdb.hdr.magic_food);
136 (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
140 if (tdb->flags & TDB_INTERNAL) {
141 tdb->map_size = sizeof(newdb);
142 tdb->map_ptr = malloc(tdb->map_size);
144 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
146 " failed to allocate");
148 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
151 if (lseek(tdb->fd, 0, SEEK_SET) == -1) {
152 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
154 " failed to seek: %s", strerror(errno));
157 if (ftruncate(tdb->fd, 0) == -1) {
158 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
160 " failed to truncate: %s", strerror(errno));
163 rlen = write(tdb->fd, &newdb, sizeof(newdb));
164 if (rlen != sizeof(newdb)) {
167 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
168 "tdb_new_database: %zi writing header: %s",
169 rlen, strerror(errno));
174 struct tdb_context *tdb_open(const char *name, int tdb_flags,
175 int open_flags, mode_t mode,
176 union tdb_attribute *attr)
178 struct tdb_context *tdb;
184 struct tdb_header hdr;
185 struct tdb_attribute_seed *seed = NULL;
187 enum TDB_ERROR ecode;
189 tdb = malloc(sizeof(*tdb));
197 tdb->direct_access = 0;
199 tdb->map_size = sizeof(struct tdb_header);
200 tdb->flags = tdb_flags;
202 tdb->transaction = NULL;
210 switch (attr->base.attr) {
211 case TDB_ATTRIBUTE_LOG:
212 tdb->logfn = attr->log.log_fn;
213 tdb->log_private = attr->log.log_private;
215 case TDB_ATTRIBUTE_HASH:
216 tdb->khash = attr->hash.hash_fn;
217 tdb->hash_priv = attr->hash.hash_private;
219 case TDB_ATTRIBUTE_SEED:
222 case TDB_ATTRIBUTE_STATS:
223 tdb->stats = &attr->stats;
224 /* They have stats we don't know about? Tell them. */
225 if (tdb->stats->size > sizeof(attr->stats))
226 tdb->stats->size = sizeof(attr->stats);
229 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL,
232 " unknown attribute type %u",
236 attr = attr->base.next;
239 if ((open_flags & O_ACCMODE) == O_WRONLY) {
240 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
241 "tdb_open: can't open tdb %s write-only",
246 if ((open_flags & O_ACCMODE) == O_RDONLY) {
247 tdb->read_only = true;
248 tdb->mmap_flags = PROT_READ;
250 tdb->read_only = false;
251 tdb->mmap_flags = PROT_READ | PROT_WRITE;
254 /* internal databases don't need any of the rest. */
255 if (tdb->flags & TDB_INTERNAL) {
256 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
257 ecode = tdb_new_database(tdb, seed, &hdr);
258 if (ecode != TDB_SUCCESS) {
261 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
262 tdb->hash_seed = hdr.hash_seed;
263 tdb_ftable_init(tdb);
267 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
268 /* errno set by open(2) */
270 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
271 "tdb_open: could not open file %s: %s",
272 name, strerror(errno));
276 /* on exec, don't inherit the fd */
277 v = fcntl(tdb->fd, F_GETFD, 0);
278 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
280 /* ensure there is only one process initialising at once */
281 ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
282 if (ecode != TDB_SUCCESS) {
286 /* If they used O_TRUNC, read will return 0. */
287 rlen = read(tdb->fd, &hdr, sizeof(hdr));
288 if (rlen == 0 && (open_flags & O_CREAT)) {
289 ecode = tdb_new_database(tdb, seed, &hdr);
290 if (ecode != TDB_SUCCESS) {
293 } else if (rlen < 0) {
294 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
295 "tdb_open: error %s reading %s",
296 strerror(errno), name);
298 } else if (rlen < sizeof(hdr)
299 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
300 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
301 "tdb_open: %s is not a tdb file", name);
305 if (hdr.version != TDB_VERSION) {
306 if (hdr.version == bswap_64(TDB_VERSION))
307 tdb->flags |= TDB_CONVERT;
310 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
312 " %s is unknown version 0x%llx",
313 name, (long long)hdr.version);
318 tdb_convert(tdb, &hdr, sizeof(hdr));
319 tdb->hash_seed = hdr.hash_seed;
320 hash_test = TDB_HASH_MAGIC;
321 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
322 if (hdr.hash_test != hash_test) {
323 /* wrong hash variant */
324 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
326 " %s uses a different hash function",
331 if (fstat(tdb->fd, &st) == -1) {
333 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
334 "tdb_open: could not stat open %s: %s",
335 name, strerror(errno));
339 /* Is it already in the open list? If so, fail. */
340 if (tdb_already_open(st.st_dev, st.st_ino)) {
342 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
343 "tdb_open: %s (%d,%d) is already open"
345 name, (int)st.st_dev, (int)st.st_ino);
349 tdb->name = strdup(name);
351 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
352 "tdb_open: failed to allocate name");
356 tdb->device = st.st_dev;
357 tdb->inode = st.st_ino;
358 tdb_unlock_open(tdb);
360 /* This make sure we have current map_size and mmap. */
361 tdb->methods->oob(tdb, tdb->map_size + 1, true);
363 /* Now it's fully formed, recover if necessary. */
364 berr = tdb_needs_recovery(tdb);
365 if (unlikely(berr != false)) {
370 ecode = tdb_lock_and_recover(tdb);
371 if (ecode != TDB_SUCCESS) {
376 ecode = tdb_ftable_init(tdb);
377 if (ecode != TDB_SUCCESS) {
386 /* Map ecode to some logical errno. */
389 case TDB_ERR_CORRUPT:
394 saved_errno = EWOULDBLOCK;
397 saved_errno = ENOMEM;
400 saved_errno = EINVAL;
403 saved_errno = EINVAL;
412 if (tdb->flags & TDB_INTERNAL) {
418 free((char *)tdb->name);
420 if (close(tdb->fd) != 0)
421 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
422 "tdb_open: failed to close tdb->fd"
423 " on error: %s", strerror(errno));
429 static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
433 struct tdb_used_record *rec,
436 uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
437 enum TDB_ERROR ecode;
439 ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
440 keylen + dataroom, h);
441 if (ecode == TDB_SUCCESS) {
442 ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
447 static enum TDB_ERROR replace_data(struct tdb_context *tdb,
449 struct tdb_data key, struct tdb_data dbuf,
450 tdb_off_t old_off, tdb_len_t old_room,
454 enum TDB_ERROR ecode;
456 /* Allocate a new record. */
457 new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
459 if (TDB_OFF_IS_ERR(new_off)) {
463 /* We didn't like the existing one: remove it. */
465 add_stat(tdb, frees, 1);
466 ecode = add_free_record(tdb, old_off,
467 sizeof(struct tdb_used_record)
468 + key.dsize + old_room);
469 if (ecode == TDB_SUCCESS)
470 ecode = replace_in_hash(tdb, h, new_off);
472 ecode = add_to_hash(tdb, h, new_off);
474 if (ecode != TDB_SUCCESS) {
478 new_off += sizeof(struct tdb_used_record);
479 ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
480 if (ecode != TDB_SUCCESS) {
484 new_off += key.dsize;
485 ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
486 if (ecode != TDB_SUCCESS) {
490 /* FIXME: tdb_increment_seqnum(tdb); */
494 enum TDB_ERROR tdb_store(struct tdb_context *tdb,
495 struct tdb_data key, struct tdb_data dbuf, int flag)
499 tdb_len_t old_room = 0;
500 struct tdb_used_record rec;
501 enum TDB_ERROR ecode;
503 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
504 if (TDB_OFF_IS_ERR(off)) {
508 /* Now we have lock on this hash bucket. */
509 if (flag == TDB_INSERT) {
511 ecode = TDB_ERR_EXISTS;
516 old_room = rec_data_length(&rec)
517 + rec_extra_padding(&rec);
518 if (old_room >= dbuf.dsize) {
519 /* Can modify in-place. Easy! */
520 ecode = update_rec_hdr(tdb, off,
521 key.dsize, dbuf.dsize,
523 if (ecode != TDB_SUCCESS) {
526 ecode = tdb->methods->twrite(tdb,
531 if (ecode != TDB_SUCCESS) {
534 tdb_unlock_hashes(tdb, h.hlock_start,
535 h.hlock_range, F_WRLCK);
539 if (flag == TDB_MODIFY) {
540 /* if the record doesn't exist and we
541 are in TDB_MODIFY mode then we should fail
543 ecode = TDB_ERR_NOEXIST;
549 /* If we didn't use the old record, this implies we're growing. */
550 ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
552 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
556 enum TDB_ERROR tdb_append(struct tdb_context *tdb,
557 struct tdb_data key, struct tdb_data dbuf)
561 struct tdb_used_record rec;
562 tdb_len_t old_room = 0, old_dlen;
563 unsigned char *newdata;
564 struct tdb_data new_dbuf;
565 enum TDB_ERROR ecode;
567 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
568 if (TDB_OFF_IS_ERR(off)) {
573 old_dlen = rec_data_length(&rec);
574 old_room = old_dlen + rec_extra_padding(&rec);
576 /* Fast path: can append in place. */
577 if (rec_extra_padding(&rec) >= dbuf.dsize) {
578 ecode = update_rec_hdr(tdb, off, key.dsize,
579 old_dlen + dbuf.dsize, &rec,
581 if (ecode != TDB_SUCCESS) {
585 off += sizeof(rec) + key.dsize + old_dlen;
586 ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
592 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
594 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
596 " failed to allocate %zu bytes",
597 (size_t)(key.dsize + old_dlen
601 ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
603 if (ecode != TDB_SUCCESS) {
604 goto out_free_newdata;
606 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
607 new_dbuf.dptr = newdata;
608 new_dbuf.dsize = old_dlen + dbuf.dsize;
614 /* If they're using tdb_append(), it implies they're growing record. */
615 ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
620 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
624 enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
625 struct tdb_data *data)
628 struct tdb_used_record rec;
630 enum TDB_ERROR ecode;
632 off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
633 if (TDB_OFF_IS_ERR(off)) {
638 ecode = TDB_ERR_NOEXIST;
640 data->dsize = rec_data_length(&rec);
641 data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
643 if (TDB_PTR_IS_ERR(data->dptr)) {
644 ecode = TDB_PTR_ERR(data->dptr);
649 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
653 enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
656 struct tdb_used_record rec;
658 enum TDB_ERROR ecode;
660 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
661 if (TDB_OFF_IS_ERR(off)) {
666 ecode = TDB_ERR_NOEXIST;
670 ecode = delete_from_hash(tdb, &h);
671 if (ecode != TDB_SUCCESS) {
675 /* Free the deleted entry. */
676 add_stat(tdb, frees, 1);
677 ecode = add_free_record(tdb, off,
678 sizeof(struct tdb_used_record)
679 + rec_key_length(&rec)
680 + rec_data_length(&rec)
681 + rec_extra_padding(&rec));
684 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
688 int tdb_close(struct tdb_context *tdb)
690 struct tdb_context **i;
693 tdb_trace(tdb, "tdb_close");
695 if (tdb->transaction) {
696 tdb_transaction_cancel(tdb);
700 if (tdb->flags & TDB_INTERNAL)
705 free((char *)tdb->name);
707 ret = close(tdb->fd);
712 /* Remove from contexts list */
713 for (i = &tdbs; *i; i = &(*i)->next) {
728 const char *tdb_errorstr(enum TDB_ERROR ecode)
730 /* Gcc warns if you miss a case in the switch, so use that. */
732 case TDB_SUCCESS: return "Success";
733 case TDB_ERR_CORRUPT: return "Corrupt database";
734 case TDB_ERR_IO: return "IO Error";
735 case TDB_ERR_LOCK: return "Locking error";
736 case TDB_ERR_OOM: return "Out of memory";
737 case TDB_ERR_EXISTS: return "Record exists";
738 case TDB_ERR_EINVAL: return "Invalid parameter";
739 case TDB_ERR_NOEXIST: return "Record does not exist";
740 case TDB_ERR_RDONLY: return "write not permitted";
742 return "Invalid error code";
745 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
746 enum TDB_ERROR ecode,
747 enum tdb_log_level level,
748 const char *fmt, ...)
753 /* tdb_open paths care about errno, so save it. */
754 int saved_errno = errno;
759 /* FIXME: Doesn't assume asprintf. */
761 len = vsnprintf(NULL, 0, fmt, ap);
764 message = malloc(len + 1);
766 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
767 "out of memory formatting message:");
768 tdb->logfn(tdb, level, tdb->log_private, fmt);
772 len = vsprintf(message, fmt, ap);
774 tdb->logfn(tdb, level, tdb->log_private, message);