2 #include <ccan/asprintf/asprintf.h>
3 #include <ccan/tdb2/tdb2.h>
8 struct tdb_data tdb_null = { .dptr = NULL, .dsize = 0 };
10 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
11 static struct tdb_context *tdbs = NULL;
13 static bool tdb_already_open(dev_t device, ino_t ino)
15 struct tdb_context *i;
17 for (i = tdbs; i; i = i->next) {
18 if (i->device == device && i->inode == ino) {
26 static bool read_all(int fd, void *buf, size_t len)
30 ret = read(fd, buf, len);
38 buf = (char *)buf + ret;
44 static uint64_t random_number(struct tdb_context *tdb)
50 fd = open("/dev/urandom", O_RDONLY);
52 if (read_all(fd, &ret, sizeof(ret))) {
58 /* FIXME: Untested! Based on Wikipedia protocol description! */
59 fd = open("/dev/egd-pool", O_RDWR);
61 /* Command is 1, next byte is size we want to read. */
62 char cmd[2] = { 1, sizeof(uint64_t) };
63 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
64 char reply[1 + sizeof(uint64_t)];
65 int r = read(fd, reply, sizeof(reply));
67 /* Copy at least some bytes. */
68 memcpy(&ret, reply+1, r - 1);
69 if (reply[0] == sizeof(uint64_t)
70 && r == sizeof(reply)) {
79 /* Fallback: pid and time. */
80 gettimeofday(&now, NULL);
81 ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
82 tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
83 "tdb_open: random from getpid and time");
88 struct tdb_header hdr;
89 struct tdb_freetable ftable;
92 /* initialise a new database */
93 static enum TDB_ERROR tdb_new_database(struct tdb_context *tdb,
94 struct tdb_attribute_seed *seed,
95 struct tdb_header *hdr)
97 /* We make it up in memory, then write it out if not internal */
98 struct new_database newdb;
99 unsigned int magic_len;
101 enum TDB_ERROR ecode;
103 /* Fill in the header */
104 newdb.hdr.version = TDB_VERSION;
106 newdb.hdr.hash_seed = seed->seed;
108 newdb.hdr.hash_seed = random_number(tdb);
109 newdb.hdr.hash_test = TDB_HASH_MAGIC;
110 newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
111 sizeof(newdb.hdr.hash_test),
114 newdb.hdr.recovery = 0;
115 memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
116 /* Initial hashes are empty. */
117 memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
120 newdb.hdr.free_table = offsetof(struct new_database, ftable);
121 memset(&newdb.ftable, 0, sizeof(newdb.ftable));
122 ecode = set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
123 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
124 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
126 if (ecode != TDB_SUCCESS) {
131 memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
132 strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
134 /* This creates an endian-converted database, as if read from disk */
135 magic_len = sizeof(newdb.hdr.magic_food);
137 (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
141 if (tdb->flags & TDB_INTERNAL) {
142 tdb->map_size = sizeof(newdb);
143 tdb->map_ptr = malloc(tdb->map_size);
145 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
147 " failed to allocate");
149 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
152 if (lseek(tdb->fd, 0, SEEK_SET) == -1) {
153 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
155 " failed to seek: %s", strerror(errno));
158 if (ftruncate(tdb->fd, 0) == -1) {
159 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
161 " failed to truncate: %s", strerror(errno));
164 rlen = write(tdb->fd, &newdb, sizeof(newdb));
165 if (rlen != sizeof(newdb)) {
168 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
169 "tdb_new_database: %zi writing header: %s",
170 rlen, strerror(errno));
175 struct tdb_context *tdb_open(const char *name, int tdb_flags,
176 int open_flags, mode_t mode,
177 union tdb_attribute *attr)
179 struct tdb_context *tdb;
185 struct tdb_header hdr;
186 struct tdb_attribute_seed *seed = NULL;
188 enum TDB_ERROR ecode;
190 tdb = malloc(sizeof(*tdb));
198 tdb->direct_access = 0;
200 tdb->map_size = sizeof(struct tdb_header);
201 tdb->flags = tdb_flags;
203 tdb->transaction = NULL;
211 switch (attr->base.attr) {
212 case TDB_ATTRIBUTE_LOG:
213 tdb->logfn = attr->log.log_fn;
214 tdb->log_private = attr->log.log_private;
216 case TDB_ATTRIBUTE_HASH:
217 tdb->khash = attr->hash.hash_fn;
218 tdb->hash_priv = attr->hash.hash_private;
220 case TDB_ATTRIBUTE_SEED:
223 case TDB_ATTRIBUTE_STATS:
224 tdb->stats = &attr->stats;
225 /* They have stats we don't know about? Tell them. */
226 if (tdb->stats->size > sizeof(attr->stats))
227 tdb->stats->size = sizeof(attr->stats);
230 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL,
233 " unknown attribute type %u",
237 attr = attr->base.next;
240 if ((open_flags & O_ACCMODE) == O_WRONLY) {
241 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
242 "tdb_open: can't open tdb %s write-only",
247 if ((open_flags & O_ACCMODE) == O_RDONLY) {
248 tdb->read_only = true;
249 tdb->mmap_flags = PROT_READ;
251 tdb->read_only = false;
252 tdb->mmap_flags = PROT_READ | PROT_WRITE;
255 /* internal databases don't need any of the rest. */
256 if (tdb->flags & TDB_INTERNAL) {
257 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
258 ecode = tdb_new_database(tdb, seed, &hdr);
259 if (ecode != TDB_SUCCESS) {
262 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
263 tdb->hash_seed = hdr.hash_seed;
264 tdb_ftable_init(tdb);
268 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
269 /* errno set by open(2) */
271 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
272 "tdb_open: could not open file %s: %s",
273 name, strerror(errno));
277 /* on exec, don't inherit the fd */
278 v = fcntl(tdb->fd, F_GETFD, 0);
279 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
281 /* ensure there is only one process initialising at once */
282 ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
283 if (ecode != TDB_SUCCESS) {
287 /* If they used O_TRUNC, read will return 0. */
288 rlen = read(tdb->fd, &hdr, sizeof(hdr));
289 if (rlen == 0 && (open_flags & O_CREAT)) {
290 ecode = tdb_new_database(tdb, seed, &hdr);
291 if (ecode != TDB_SUCCESS) {
294 } else if (rlen < 0) {
295 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
296 "tdb_open: error %s reading %s",
297 strerror(errno), name);
299 } else if (rlen < sizeof(hdr)
300 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
301 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
302 "tdb_open: %s is not a tdb file", name);
306 if (hdr.version != TDB_VERSION) {
307 if (hdr.version == bswap_64(TDB_VERSION))
308 tdb->flags |= TDB_CONVERT;
311 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
313 " %s is unknown version 0x%llx",
314 name, (long long)hdr.version);
319 tdb_convert(tdb, &hdr, sizeof(hdr));
320 tdb->hash_seed = hdr.hash_seed;
321 hash_test = TDB_HASH_MAGIC;
322 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
323 if (hdr.hash_test != hash_test) {
324 /* wrong hash variant */
325 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
327 " %s uses a different hash function",
332 if (fstat(tdb->fd, &st) == -1) {
334 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
335 "tdb_open: could not stat open %s: %s",
336 name, strerror(errno));
340 /* Is it already in the open list? If so, fail. */
341 if (tdb_already_open(st.st_dev, st.st_ino)) {
343 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
344 "tdb_open: %s (%d,%d) is already open"
346 name, (int)st.st_dev, (int)st.st_ino);
350 tdb->name = strdup(name);
352 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
353 "tdb_open: failed to allocate name");
357 tdb->device = st.st_dev;
358 tdb->inode = st.st_ino;
359 tdb_unlock_open(tdb);
361 /* This make sure we have current map_size and mmap. */
362 tdb->methods->oob(tdb, tdb->map_size + 1, true);
364 /* Now it's fully formed, recover if necessary. */
365 berr = tdb_needs_recovery(tdb);
366 if (unlikely(berr != false)) {
371 ecode = tdb_lock_and_recover(tdb);
372 if (ecode != TDB_SUCCESS) {
377 ecode = tdb_ftable_init(tdb);
378 if (ecode != TDB_SUCCESS) {
387 /* Map ecode to some logical errno. */
390 case TDB_ERR_CORRUPT:
395 saved_errno = EWOULDBLOCK;
398 saved_errno = ENOMEM;
401 saved_errno = EINVAL;
404 saved_errno = EINVAL;
413 if (tdb->flags & TDB_INTERNAL) {
419 free((char *)tdb->name);
421 if (close(tdb->fd) != 0)
422 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
423 "tdb_open: failed to close tdb->fd"
424 " on error: %s", strerror(errno));
430 static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
434 struct tdb_used_record *rec,
437 uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
438 enum TDB_ERROR ecode;
440 ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
441 keylen + dataroom, h);
442 if (ecode == TDB_SUCCESS) {
443 ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
448 static enum TDB_ERROR replace_data(struct tdb_context *tdb,
450 struct tdb_data key, struct tdb_data dbuf,
451 tdb_off_t old_off, tdb_len_t old_room,
455 enum TDB_ERROR ecode;
457 /* Allocate a new record. */
458 new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
460 if (TDB_OFF_IS_ERR(new_off)) {
464 /* We didn't like the existing one: remove it. */
466 add_stat(tdb, frees, 1);
467 ecode = add_free_record(tdb, old_off,
468 sizeof(struct tdb_used_record)
469 + key.dsize + old_room);
470 if (ecode == TDB_SUCCESS)
471 ecode = replace_in_hash(tdb, h, new_off);
473 ecode = add_to_hash(tdb, h, new_off);
475 if (ecode != TDB_SUCCESS) {
479 new_off += sizeof(struct tdb_used_record);
480 ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
481 if (ecode != TDB_SUCCESS) {
485 new_off += key.dsize;
486 ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
487 if (ecode != TDB_SUCCESS) {
491 /* FIXME: tdb_increment_seqnum(tdb); */
495 enum TDB_ERROR tdb_store(struct tdb_context *tdb,
496 struct tdb_data key, struct tdb_data dbuf, int flag)
500 tdb_len_t old_room = 0;
501 struct tdb_used_record rec;
502 enum TDB_ERROR ecode;
504 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
505 if (TDB_OFF_IS_ERR(off)) {
509 /* Now we have lock on this hash bucket. */
510 if (flag == TDB_INSERT) {
512 ecode = TDB_ERR_EXISTS;
517 old_room = rec_data_length(&rec)
518 + rec_extra_padding(&rec);
519 if (old_room >= dbuf.dsize) {
520 /* Can modify in-place. Easy! */
521 ecode = update_rec_hdr(tdb, off,
522 key.dsize, dbuf.dsize,
524 if (ecode != TDB_SUCCESS) {
527 ecode = tdb->methods->twrite(tdb,
532 if (ecode != TDB_SUCCESS) {
535 tdb_unlock_hashes(tdb, h.hlock_start,
536 h.hlock_range, F_WRLCK);
540 if (flag == TDB_MODIFY) {
541 /* if the record doesn't exist and we
542 are in TDB_MODIFY mode then we should fail
544 ecode = TDB_ERR_NOEXIST;
550 /* If we didn't use the old record, this implies we're growing. */
551 ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
553 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
557 enum TDB_ERROR tdb_append(struct tdb_context *tdb,
558 struct tdb_data key, struct tdb_data dbuf)
562 struct tdb_used_record rec;
563 tdb_len_t old_room = 0, old_dlen;
564 unsigned char *newdata;
565 struct tdb_data new_dbuf;
566 enum TDB_ERROR ecode;
568 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
569 if (TDB_OFF_IS_ERR(off)) {
574 old_dlen = rec_data_length(&rec);
575 old_room = old_dlen + rec_extra_padding(&rec);
577 /* Fast path: can append in place. */
578 if (rec_extra_padding(&rec) >= dbuf.dsize) {
579 ecode = update_rec_hdr(tdb, off, key.dsize,
580 old_dlen + dbuf.dsize, &rec,
582 if (ecode != TDB_SUCCESS) {
586 off += sizeof(rec) + key.dsize + old_dlen;
587 ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
593 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
595 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
597 " failed to allocate %zu bytes",
598 (size_t)(key.dsize + old_dlen
602 ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
604 if (ecode != TDB_SUCCESS) {
605 goto out_free_newdata;
607 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
608 new_dbuf.dptr = newdata;
609 new_dbuf.dsize = old_dlen + dbuf.dsize;
615 /* If they're using tdb_append(), it implies they're growing record. */
616 ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
621 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
625 enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
626 struct tdb_data *data)
629 struct tdb_used_record rec;
631 enum TDB_ERROR ecode;
633 off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
634 if (TDB_OFF_IS_ERR(off)) {
639 ecode = TDB_ERR_NOEXIST;
641 data->dsize = rec_data_length(&rec);
642 data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
644 if (TDB_PTR_IS_ERR(data->dptr)) {
645 ecode = TDB_PTR_ERR(data->dptr);
650 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
654 enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
657 struct tdb_used_record rec;
659 enum TDB_ERROR ecode;
661 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
662 if (TDB_OFF_IS_ERR(off)) {
667 ecode = TDB_ERR_NOEXIST;
671 ecode = delete_from_hash(tdb, &h);
672 if (ecode != TDB_SUCCESS) {
676 /* Free the deleted entry. */
677 add_stat(tdb, frees, 1);
678 ecode = add_free_record(tdb, off,
679 sizeof(struct tdb_used_record)
680 + rec_key_length(&rec)
681 + rec_data_length(&rec)
682 + rec_extra_padding(&rec));
685 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
689 int tdb_close(struct tdb_context *tdb)
691 struct tdb_context **i;
694 tdb_trace(tdb, "tdb_close");
696 if (tdb->transaction) {
697 tdb_transaction_cancel(tdb);
701 if (tdb->flags & TDB_INTERNAL)
706 free((char *)tdb->name);
708 ret = close(tdb->fd);
713 /* Remove from contexts list */
714 for (i = &tdbs; *i; i = &(*i)->next) {
729 const char *tdb_errorstr(enum TDB_ERROR ecode)
731 /* Gcc warns if you miss a case in the switch, so use that. */
733 case TDB_SUCCESS: return "Success";
734 case TDB_ERR_CORRUPT: return "Corrupt database";
735 case TDB_ERR_IO: return "IO Error";
736 case TDB_ERR_LOCK: return "Locking error";
737 case TDB_ERR_OOM: return "Out of memory";
738 case TDB_ERR_EXISTS: return "Record exists";
739 case TDB_ERR_EINVAL: return "Invalid parameter";
740 case TDB_ERR_NOEXIST: return "Record does not exist";
741 case TDB_ERR_RDONLY: return "write not permitted";
743 return "Invalid error code";
746 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
747 enum TDB_ERROR ecode,
748 enum tdb_log_level level,
749 const char *fmt, ...)
754 /* tdb_open paths care about errno, so save it. */
755 int saved_errno = errno;
761 len = vasprintf(&message, fmt, ap);
765 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
766 "out of memory formatting message:");
767 tdb->logfn(tdb, level, tdb->log_private, fmt);
769 tdb->logfn(tdb, level, tdb->log_private, message);