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->ecode = TDB_SUCCESS;
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 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
231 "tdb_open: unknown attribute type %u",
235 attr = attr->base.next;
238 if ((open_flags & O_ACCMODE) == O_WRONLY) {
239 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
240 "tdb_open: can't open tdb %s write-only", name);
244 if ((open_flags & O_ACCMODE) == O_RDONLY) {
245 tdb->read_only = true;
246 /* read only databases don't do locking */
247 tdb->flags |= TDB_NOLOCK;
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 if (tdb_new_database(tdb, seed, &hdr) != 0) {
260 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
261 tdb->hash_seed = hdr.hash_seed;
262 tdb_ftable_init(tdb);
266 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
267 /* errno set by open(2) */
269 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
270 "tdb_open: could not open file %s: %s",
271 name, strerror(errno));
275 /* on exec, don't inherit the fd */
276 v = fcntl(tdb->fd, F_GETFD, 0);
277 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
279 /* ensure there is only one process initialising at once */
280 tdb->ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
281 if (tdb->ecode != TDB_SUCCESS) {
285 /* If they used O_TRUNC, read will return 0. */
286 rlen = read(tdb->fd, &hdr, sizeof(hdr));
287 if (rlen == 0 && (open_flags & O_CREAT)) {
288 if (tdb_new_database(tdb, seed, &hdr) == -1) {
291 } else if (rlen < 0) {
292 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
293 "tdb_open: error %s reading %s",
294 strerror(errno), name);
296 } else if (rlen < sizeof(hdr)
297 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
298 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
299 "tdb_open: %s is not a tdb file", name);
303 if (hdr.version != TDB_VERSION) {
304 if (hdr.version == bswap_64(TDB_VERSION))
305 tdb->flags |= TDB_CONVERT;
308 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
309 "tdb_open: %s is unknown version 0x%llx",
310 name, (long long)hdr.version);
315 tdb_convert(tdb, &hdr, sizeof(hdr));
316 tdb->hash_seed = hdr.hash_seed;
317 hash_test = TDB_HASH_MAGIC;
318 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
319 if (hdr.hash_test != hash_test) {
320 /* wrong hash variant */
321 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
322 "tdb_open: %s uses a different hash function",
327 if (fstat(tdb->fd, &st) == -1) {
329 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
330 "tdb_open: could not stat open %s: %s",
331 name, strerror(errno));
335 /* Is it already in the open list? If so, fail. */
336 if (tdb_already_open(st.st_dev, st.st_ino)) {
338 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
339 "tdb_open: %s (%d,%d) is already open in this"
341 name, (int)st.st_dev, (int)st.st_ino);
345 tdb->name = strdup(name);
347 tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
348 "tdb_open: failed to allocate name");
352 tdb->device = st.st_dev;
353 tdb->inode = st.st_ino;
354 tdb_unlock_open(tdb);
356 /* This make sure we have current map_size and mmap. */
357 tdb->methods->oob(tdb, tdb->map_size + 1, true);
359 /* Now it's fully formed, recover if necessary. */
360 berr = tdb_needs_recovery(tdb);
361 if (unlikely(berr != false)) {
366 ecode = tdb_lock_and_recover(tdb);
367 if (ecode != TDB_SUCCESS) {
373 tdb->ecode = tdb_ftable_init(tdb);
374 if (tdb->ecode != TDB_SUCCESS) {
383 /* Map ecode to some logical errno. */
385 switch (tdb->ecode) {
386 case TDB_ERR_CORRUPT:
391 saved_errno = EWOULDBLOCK;
394 saved_errno = ENOMEM;
397 saved_errno = EINVAL;
400 saved_errno = EINVAL;
409 if (tdb->flags & TDB_INTERNAL) {
415 free((char *)tdb->name);
417 if (close(tdb->fd) != 0)
418 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
419 "tdb_open: failed to close tdb->fd"
420 " on error: %s", strerror(errno));
426 static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
430 struct tdb_used_record *rec,
433 uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
434 enum TDB_ERROR ecode;
436 ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
437 keylen + dataroom, h);
438 if (ecode == TDB_SUCCESS) {
439 ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
444 static enum TDB_ERROR replace_data(struct tdb_context *tdb,
446 struct tdb_data key, struct tdb_data dbuf,
447 tdb_off_t old_off, tdb_len_t old_room,
451 enum TDB_ERROR ecode;
453 /* Allocate a new record. */
454 new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
456 if (TDB_OFF_IS_ERR(new_off)) {
460 /* We didn't like the existing one: remove it. */
462 add_stat(tdb, frees, 1);
463 ecode = add_free_record(tdb, old_off,
464 sizeof(struct tdb_used_record)
465 + key.dsize + old_room);
466 if (ecode == TDB_SUCCESS)
467 ecode = replace_in_hash(tdb, h, new_off);
469 ecode = add_to_hash(tdb, h, new_off);
471 if (ecode != TDB_SUCCESS) {
475 new_off += sizeof(struct tdb_used_record);
476 ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
477 if (ecode != TDB_SUCCESS) {
481 new_off += key.dsize;
482 ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
483 if (ecode != TDB_SUCCESS) {
487 /* FIXME: tdb_increment_seqnum(tdb); */
491 int tdb_store(struct tdb_context *tdb,
492 struct tdb_data key, struct tdb_data dbuf, int flag)
496 tdb_len_t old_room = 0;
497 struct tdb_used_record rec;
498 enum TDB_ERROR ecode;
500 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
501 if (TDB_OFF_IS_ERR(off)) {
506 /* Now we have lock on this hash bucket. */
507 if (flag == TDB_INSERT) {
509 tdb->ecode = TDB_ERR_EXISTS;
514 old_room = rec_data_length(&rec)
515 + rec_extra_padding(&rec);
516 if (old_room >= dbuf.dsize) {
517 /* Can modify in-place. Easy! */
518 ecode = update_rec_hdr(tdb, off,
519 key.dsize, dbuf.dsize,
521 if (ecode != TDB_SUCCESS) {
525 ecode = tdb->methods->twrite(tdb,
530 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 tdb->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);
551 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
552 if (ecode != TDB_SUCCESS) {
559 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
563 int tdb_append(struct tdb_context *tdb,
564 struct tdb_data key, struct tdb_data dbuf)
568 struct tdb_used_record rec;
569 tdb_len_t old_room = 0, old_dlen;
570 unsigned char *newdata;
571 struct tdb_data new_dbuf;
572 enum TDB_ERROR ecode;
574 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
575 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) {
594 off += sizeof(rec) + key.dsize + old_dlen;
595 ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
597 if (ecode != TDB_SUCCESS) {
602 /* FIXME: tdb_increment_seqnum(tdb); */
603 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
609 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
611 tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
612 "tdb_append: failed to allocate %zu bytes",
613 (size_t)(key.dsize+old_dlen+dbuf.dsize));
616 ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
618 if (ecode != TDB_SUCCESS) {
623 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
624 new_dbuf.dptr = newdata;
625 new_dbuf.dsize = old_dlen + dbuf.dsize;
631 /* If they're using tdb_append(), it implies they're growing record. */
632 ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
633 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
636 if (ecode != TDB_SUCCESS) {
643 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
647 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
650 struct tdb_used_record rec;
654 off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
655 if (TDB_OFF_IS_ERR(off)) {
661 tdb->ecode = TDB_ERR_NOEXIST;
664 ret.dsize = rec_data_length(&rec);
665 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
667 if (TDB_PTR_IS_ERR(ret.dptr)) {
668 tdb->ecode = TDB_PTR_ERR(ret.dptr);
673 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
677 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
680 struct tdb_used_record rec;
682 enum TDB_ERROR ecode;
684 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
685 if (TDB_OFF_IS_ERR(off)) {
691 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
692 tdb->ecode = TDB_ERR_NOEXIST;
696 ecode = delete_from_hash(tdb, &h);
697 if (ecode != TDB_SUCCESS) {
702 /* Free the deleted entry. */
703 add_stat(tdb, frees, 1);
704 ecode = add_free_record(tdb, off,
705 sizeof(struct tdb_used_record)
706 + rec_key_length(&rec)
707 + rec_data_length(&rec)
708 + rec_extra_padding(&rec));
709 if (ecode != TDB_SUCCESS) {
714 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
718 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
722 int tdb_close(struct tdb_context *tdb)
724 struct tdb_context **i;
727 tdb_trace(tdb, "tdb_close");
729 if (tdb->transaction) {
730 tdb_transaction_cancel(tdb);
734 if (tdb->flags & TDB_INTERNAL)
739 free((char *)tdb->name);
741 ret = close(tdb->fd);
746 /* Remove from contexts list */
747 for (i = &tdbs; *i; i = &(*i)->next) {
762 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
767 const char *tdb_errorstr(const struct tdb_context *tdb)
769 /* Gcc warns if you miss a case in the switch, so use that. */
770 switch (tdb->ecode) {
771 case TDB_SUCCESS: return "Success";
772 case TDB_ERR_CORRUPT: return "Corrupt database";
773 case TDB_ERR_IO: return "IO Error";
774 case TDB_ERR_LOCK: return "Locking error";
775 case TDB_ERR_OOM: return "Out of memory";
776 case TDB_ERR_EXISTS: return "Record exists";
777 case TDB_ERR_EINVAL: return "Invalid parameter";
778 case TDB_ERR_NOEXIST: return "Record does not exist";
779 case TDB_ERR_RDONLY: return "write not permitted";
781 return "Invalid error code";
784 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
785 enum TDB_ERROR ecode,
786 enum tdb_log_level level,
787 const char *fmt, ...)
792 /* tdb_open paths care about errno, so save it. */
793 int saved_errno = errno;
800 /* FIXME: Doesn't assume asprintf. */
802 len = vsnprintf(NULL, 0, fmt, ap);
805 message = malloc(len + 1);
807 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
808 "out of memory formatting message:");
809 tdb->logfn(tdb, level, tdb->log_private, fmt);
813 len = vsprintf(message, fmt, ap);
815 tdb->logfn(tdb, level, tdb->log_private, message);