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 (tdb_flags & ~(TDB_INTERNAL | TDB_NOLOCK | TDB_NOMMAP | TDB_CONVERT
242 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
243 "tdb_open: unknown flags %u", tdb_flags);
247 if ((open_flags & O_ACCMODE) == O_WRONLY) {
248 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
249 "tdb_open: can't open tdb %s write-only",
254 if ((open_flags & O_ACCMODE) == O_RDONLY) {
255 tdb->read_only = true;
256 tdb->mmap_flags = PROT_READ;
258 tdb->read_only = false;
259 tdb->mmap_flags = PROT_READ | PROT_WRITE;
262 /* internal databases don't need any of the rest. */
263 if (tdb->flags & TDB_INTERNAL) {
264 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
265 ecode = tdb_new_database(tdb, seed, &hdr);
266 if (ecode != TDB_SUCCESS) {
269 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
270 tdb->hash_seed = hdr.hash_seed;
271 tdb_ftable_init(tdb);
275 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
276 /* errno set by open(2) */
278 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
279 "tdb_open: could not open file %s: %s",
280 name, strerror(errno));
284 /* on exec, don't inherit the fd */
285 v = fcntl(tdb->fd, F_GETFD, 0);
286 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
288 /* ensure there is only one process initialising at once */
289 ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
290 if (ecode != TDB_SUCCESS) {
294 /* If they used O_TRUNC, read will return 0. */
295 rlen = read(tdb->fd, &hdr, sizeof(hdr));
296 if (rlen == 0 && (open_flags & O_CREAT)) {
297 ecode = tdb_new_database(tdb, seed, &hdr);
298 if (ecode != TDB_SUCCESS) {
301 } else if (rlen < 0) {
302 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
303 "tdb_open: error %s reading %s",
304 strerror(errno), name);
306 } else if (rlen < sizeof(hdr)
307 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
308 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
309 "tdb_open: %s is not a tdb file", name);
313 if (hdr.version != TDB_VERSION) {
314 if (hdr.version == bswap_64(TDB_VERSION))
315 tdb->flags |= TDB_CONVERT;
318 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
320 " %s is unknown version 0x%llx",
321 name, (long long)hdr.version);
326 tdb_convert(tdb, &hdr, sizeof(hdr));
327 tdb->hash_seed = hdr.hash_seed;
328 hash_test = TDB_HASH_MAGIC;
329 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
330 if (hdr.hash_test != hash_test) {
331 /* wrong hash variant */
332 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
334 " %s uses a different hash function",
339 if (fstat(tdb->fd, &st) == -1) {
341 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
342 "tdb_open: could not stat open %s: %s",
343 name, strerror(errno));
347 /* Is it already in the open list? If so, fail. */
348 if (tdb_already_open(st.st_dev, st.st_ino)) {
350 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
351 "tdb_open: %s (%d,%d) is already open"
353 name, (int)st.st_dev, (int)st.st_ino);
357 tdb->name = strdup(name);
359 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
360 "tdb_open: failed to allocate name");
364 tdb->device = st.st_dev;
365 tdb->inode = st.st_ino;
366 tdb_unlock_open(tdb);
368 /* This make sure we have current map_size and mmap. */
369 tdb->methods->oob(tdb, tdb->map_size + 1, true);
371 /* Now it's fully formed, recover if necessary. */
372 berr = tdb_needs_recovery(tdb);
373 if (unlikely(berr != false)) {
378 ecode = tdb_lock_and_recover(tdb);
379 if (ecode != TDB_SUCCESS) {
384 ecode = tdb_ftable_init(tdb);
385 if (ecode != TDB_SUCCESS) {
394 /* Map ecode to some logical errno. */
397 case TDB_ERR_CORRUPT:
402 saved_errno = EWOULDBLOCK;
405 saved_errno = ENOMEM;
408 saved_errno = EINVAL;
411 saved_errno = EINVAL;
420 if (tdb->flags & TDB_INTERNAL) {
426 free((char *)tdb->name);
428 if (close(tdb->fd) != 0)
429 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
430 "tdb_open: failed to close tdb->fd"
431 " on error: %s", strerror(errno));
437 static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
441 struct tdb_used_record *rec,
444 uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
445 enum TDB_ERROR ecode;
447 ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
448 keylen + dataroom, h);
449 if (ecode == TDB_SUCCESS) {
450 ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
455 static enum TDB_ERROR replace_data(struct tdb_context *tdb,
457 struct tdb_data key, struct tdb_data dbuf,
458 tdb_off_t old_off, tdb_len_t old_room,
462 enum TDB_ERROR ecode;
464 /* Allocate a new record. */
465 new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
467 if (TDB_OFF_IS_ERR(new_off)) {
471 /* We didn't like the existing one: remove it. */
473 add_stat(tdb, frees, 1);
474 ecode = add_free_record(tdb, old_off,
475 sizeof(struct tdb_used_record)
476 + key.dsize + old_room);
477 if (ecode == TDB_SUCCESS)
478 ecode = replace_in_hash(tdb, h, new_off);
480 ecode = add_to_hash(tdb, h, new_off);
482 if (ecode != TDB_SUCCESS) {
486 new_off += sizeof(struct tdb_used_record);
487 ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
488 if (ecode != TDB_SUCCESS) {
492 new_off += key.dsize;
493 ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
494 if (ecode != TDB_SUCCESS) {
498 /* FIXME: tdb_increment_seqnum(tdb); */
502 enum TDB_ERROR tdb_store(struct tdb_context *tdb,
503 struct tdb_data key, struct tdb_data dbuf, int flag)
507 tdb_len_t old_room = 0;
508 struct tdb_used_record rec;
509 enum TDB_ERROR ecode;
511 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
512 if (TDB_OFF_IS_ERR(off)) {
516 /* Now we have lock on this hash bucket. */
517 if (flag == TDB_INSERT) {
519 ecode = TDB_ERR_EXISTS;
524 old_room = rec_data_length(&rec)
525 + rec_extra_padding(&rec);
526 if (old_room >= dbuf.dsize) {
527 /* Can modify in-place. Easy! */
528 ecode = update_rec_hdr(tdb, off,
529 key.dsize, dbuf.dsize,
531 if (ecode != TDB_SUCCESS) {
534 ecode = tdb->methods->twrite(tdb,
539 if (ecode != TDB_SUCCESS) {
542 tdb_unlock_hashes(tdb, h.hlock_start,
543 h.hlock_range, F_WRLCK);
547 if (flag == TDB_MODIFY) {
548 /* if the record doesn't exist and we
549 are in TDB_MODIFY mode then we should fail
551 ecode = TDB_ERR_NOEXIST;
557 /* If we didn't use the old record, this implies we're growing. */
558 ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
560 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
564 enum TDB_ERROR tdb_append(struct tdb_context *tdb,
565 struct tdb_data key, struct tdb_data dbuf)
569 struct tdb_used_record rec;
570 tdb_len_t old_room = 0, old_dlen;
571 unsigned char *newdata;
572 struct tdb_data new_dbuf;
573 enum TDB_ERROR ecode;
575 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
576 if (TDB_OFF_IS_ERR(off)) {
581 old_dlen = rec_data_length(&rec);
582 old_room = old_dlen + rec_extra_padding(&rec);
584 /* Fast path: can append in place. */
585 if (rec_extra_padding(&rec) >= dbuf.dsize) {
586 ecode = update_rec_hdr(tdb, off, key.dsize,
587 old_dlen + dbuf.dsize, &rec,
589 if (ecode != TDB_SUCCESS) {
593 off += sizeof(rec) + key.dsize + old_dlen;
594 ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
600 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
602 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
604 " failed to allocate %zu bytes",
605 (size_t)(key.dsize + old_dlen
609 ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
611 if (ecode != TDB_SUCCESS) {
612 goto out_free_newdata;
614 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
615 new_dbuf.dptr = newdata;
616 new_dbuf.dsize = old_dlen + dbuf.dsize;
622 /* If they're using tdb_append(), it implies they're growing record. */
623 ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
628 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
632 enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
633 struct tdb_data *data)
636 struct tdb_used_record rec;
638 enum TDB_ERROR ecode;
640 off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
641 if (TDB_OFF_IS_ERR(off)) {
646 ecode = TDB_ERR_NOEXIST;
648 data->dsize = rec_data_length(&rec);
649 data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
651 if (TDB_PTR_IS_ERR(data->dptr)) {
652 ecode = TDB_PTR_ERR(data->dptr);
657 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
661 enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
664 struct tdb_used_record rec;
666 enum TDB_ERROR ecode;
668 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
669 if (TDB_OFF_IS_ERR(off)) {
674 ecode = TDB_ERR_NOEXIST;
678 ecode = delete_from_hash(tdb, &h);
679 if (ecode != TDB_SUCCESS) {
683 /* Free the deleted entry. */
684 add_stat(tdb, frees, 1);
685 ecode = add_free_record(tdb, off,
686 sizeof(struct tdb_used_record)
687 + rec_key_length(&rec)
688 + rec_data_length(&rec)
689 + rec_extra_padding(&rec));
692 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
696 int tdb_close(struct tdb_context *tdb)
698 struct tdb_context **i;
701 tdb_trace(tdb, "tdb_close");
703 if (tdb->transaction) {
704 tdb_transaction_cancel(tdb);
708 if (tdb->flags & TDB_INTERNAL)
713 free((char *)tdb->name);
715 ret = close(tdb->fd);
720 /* Remove from contexts list */
721 for (i = &tdbs; *i; i = &(*i)->next) {
736 const char *tdb_errorstr(enum TDB_ERROR ecode)
738 /* Gcc warns if you miss a case in the switch, so use that. */
740 case TDB_SUCCESS: return "Success";
741 case TDB_ERR_CORRUPT: return "Corrupt database";
742 case TDB_ERR_IO: return "IO Error";
743 case TDB_ERR_LOCK: return "Locking error";
744 case TDB_ERR_OOM: return "Out of memory";
745 case TDB_ERR_EXISTS: return "Record exists";
746 case TDB_ERR_EINVAL: return "Invalid parameter";
747 case TDB_ERR_NOEXIST: return "Record does not exist";
748 case TDB_ERR_RDONLY: return "write not permitted";
750 return "Invalid error code";
753 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
754 enum TDB_ERROR ecode,
755 enum tdb_log_level level,
756 const char *fmt, ...)
761 /* tdb_open paths care about errno, so save it. */
762 int saved_errno = errno;
768 len = vasprintf(&message, fmt, ap);
772 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
773 "out of memory formatting message:");
774 tdb->logfn(tdb, level, tdb->log_private, fmt);
776 tdb->logfn(tdb, level, tdb->log_private, message);