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))) {
52 tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
53 "tdb_open: random from /dev/urandom");
59 /* FIXME: Untested! Based on Wikipedia protocol description! */
60 fd = open("/dev/egd-pool", O_RDWR);
62 /* Command is 1, next byte is size we want to read. */
63 char cmd[2] = { 1, sizeof(uint64_t) };
64 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
65 char reply[1 + sizeof(uint64_t)];
66 int r = read(fd, reply, sizeof(reply));
68 tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
69 "tdb_open: %u random bytes from"
70 " /dev/egd-pool", r-1);
71 /* Copy at least some bytes. */
72 memcpy(&ret, reply+1, r - 1);
73 if (reply[0] == sizeof(uint64_t)
74 && r == sizeof(reply)) {
83 /* Fallback: pid and time. */
84 gettimeofday(&now, NULL);
85 ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
86 tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
87 "tdb_open: random from getpid and time");
92 struct tdb_header hdr;
93 struct tdb_freetable ftable;
96 /* initialise a new database */
97 static int tdb_new_database(struct tdb_context *tdb,
98 struct tdb_attribute_seed *seed,
99 struct tdb_header *hdr)
101 /* We make it up in memory, then write it out if not internal */
102 struct new_database newdb;
103 unsigned int magic_len;
105 /* Fill in the header */
106 newdb.hdr.version = TDB_VERSION;
108 newdb.hdr.hash_seed = seed->seed;
110 newdb.hdr.hash_seed = random_number(tdb);
111 newdb.hdr.hash_test = TDB_HASH_MAGIC;
112 newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
113 sizeof(newdb.hdr.hash_test),
116 newdb.hdr.recovery = 0;
117 memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
118 /* Initial hashes are empty. */
119 memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
122 newdb.hdr.free_table = offsetof(struct new_database, ftable);
123 memset(&newdb.ftable, 0, sizeof(newdb.ftable));
124 set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
125 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
126 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr), 0);
129 memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
130 strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
132 /* This creates an endian-converted database, as if read from disk */
133 magic_len = sizeof(newdb.hdr.magic_food);
135 (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
139 if (tdb->flags & TDB_INTERNAL) {
140 tdb->map_size = sizeof(newdb);
141 tdb->map_ptr = malloc(tdb->map_size);
143 tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
144 "tdb_new_database: failed to allocate");
147 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
150 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
153 if (ftruncate(tdb->fd, 0) == -1)
156 if (!tdb_pwrite_all(tdb->fd, &newdb, sizeof(newdb), 0)) {
157 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
158 "tdb_new_database: failed to write: %s",
165 struct tdb_context *tdb_open(const char *name, int tdb_flags,
166 int open_flags, mode_t mode,
167 union tdb_attribute *attr)
169 struct tdb_context *tdb;
175 struct tdb_header hdr;
176 struct tdb_attribute_seed *seed = NULL;
178 tdb = malloc(sizeof(*tdb));
186 tdb->direct_access = 0;
188 tdb->map_size = sizeof(struct tdb_header);
189 tdb->ecode = TDB_SUCCESS;
190 tdb->flags = tdb_flags;
192 tdb->transaction = NULL;
200 switch (attr->base.attr) {
201 case TDB_ATTRIBUTE_LOG:
202 tdb->logfn = attr->log.log_fn;
203 tdb->log_private = attr->log.log_private;
205 case TDB_ATTRIBUTE_HASH:
206 tdb->khash = attr->hash.hash_fn;
207 tdb->hash_priv = attr->hash.hash_private;
209 case TDB_ATTRIBUTE_SEED:
212 case TDB_ATTRIBUTE_STATS:
213 tdb->stats = &attr->stats;
214 /* They have stats we don't know about? Tell them. */
215 if (tdb->stats->size > sizeof(attr->stats))
216 tdb->stats->size = sizeof(attr->stats);
219 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
220 "tdb_open: unknown attribute type %u",
224 attr = attr->base.next;
227 if ((open_flags & O_ACCMODE) == O_WRONLY) {
228 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
229 "tdb_open: can't open tdb %s write-only", name);
233 if ((open_flags & O_ACCMODE) == O_RDONLY) {
234 tdb->read_only = true;
235 /* read only databases don't do locking */
236 tdb->flags |= TDB_NOLOCK;
237 tdb->mmap_flags = PROT_READ;
239 tdb->read_only = false;
240 tdb->mmap_flags = PROT_READ | PROT_WRITE;
243 /* internal databases don't need any of the rest. */
244 if (tdb->flags & TDB_INTERNAL) {
245 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
246 if (tdb_new_database(tdb, seed, &hdr) != 0) {
249 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
250 tdb->hash_seed = hdr.hash_seed;
251 tdb_ftable_init(tdb);
255 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
256 /* errno set by open(2) */
258 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
259 "tdb_open: could not open file %s: %s",
260 name, strerror(errno));
264 /* on exec, don't inherit the fd */
265 v = fcntl(tdb->fd, F_GETFD, 0);
266 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
268 /* ensure there is only one process initialising at once */
269 if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
270 /* errno set by tdb_brlock */
275 /* If they used O_TRUNC, read will return 0. */
276 rlen = read(tdb->fd, &hdr, sizeof(hdr));
277 if (rlen == 0 && (open_flags & O_CREAT)) {
278 if (tdb_new_database(tdb, seed, &hdr) == -1) {
281 } else if (rlen < 0) {
282 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
283 "tdb_open: error %s reading %s",
284 strerror(errno), name);
286 } else if (rlen < sizeof(hdr)
287 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
288 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
289 "tdb_open: %s is not a tdb file", name);
293 if (hdr.version != TDB_VERSION) {
294 if (hdr.version == bswap_64(TDB_VERSION))
295 tdb->flags |= TDB_CONVERT;
298 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
299 "tdb_open: %s is unknown version 0x%llx",
300 name, (long long)hdr.version);
305 tdb_convert(tdb, &hdr, sizeof(hdr));
306 tdb->hash_seed = hdr.hash_seed;
307 hash_test = TDB_HASH_MAGIC;
308 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
309 if (hdr.hash_test != hash_test) {
310 /* wrong hash variant */
311 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
312 "tdb_open: %s uses a different hash function",
317 if (fstat(tdb->fd, &st) == -1) {
319 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
320 "tdb_open: could not stat open %s: %s",
321 name, strerror(errno));
325 /* Is it already in the open list? If so, fail. */
326 if (tdb_already_open(st.st_dev, st.st_ino)) {
328 tdb_logerr(tdb, TDB_ERR_NESTING, TDB_DEBUG_ERROR,
329 "tdb_open: %s (%d,%d) is already open in this"
331 name, (int)st.st_dev, (int)st.st_ino);
335 tdb->name = strdup(name);
337 tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
338 "tdb_open: failed to allocate name");
342 tdb->device = st.st_dev;
343 tdb->inode = st.st_ino;
344 tdb_unlock_open(tdb);
346 /* This make sure we have current map_size and mmap. */
347 tdb->methods->oob(tdb, tdb->map_size + 1, true);
349 /* Now it's fully formed, recover if necessary. */
350 if (tdb_needs_recovery(tdb) && tdb_lock_and_recover(tdb) == -1) {
354 if (tdb_ftable_init(tdb) == -1)
362 /* Map ecode to some logical errno. */
364 switch (tdb->ecode) {
365 case TDB_ERR_CORRUPT:
370 saved_errno = EWOULDBLOCK;
373 saved_errno = ENOMEM;
376 saved_errno = EINVAL;
378 case TDB_ERR_NESTING:
382 saved_errno = EINVAL;
391 if (tdb->flags & TDB_INTERNAL) {
396 free((char *)tdb->name);
398 if (close(tdb->fd) != 0)
399 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
400 "tdb_open: failed to close tdb->fd"
407 static int update_rec_hdr(struct tdb_context *tdb,
411 struct tdb_used_record *rec,
414 uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
416 if (set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
417 keylen + dataroom, h))
420 return tdb_write_convert(tdb, off, rec, sizeof(*rec));
423 /* Returns -1 on error, 0 on OK */
424 static int replace_data(struct tdb_context *tdb,
426 struct tdb_data key, struct tdb_data dbuf,
427 tdb_off_t old_off, tdb_len_t old_room,
432 /* Allocate a new record. */
433 new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
435 if (unlikely(new_off == TDB_OFF_ERR))
438 /* We didn't like the existing one: remove it. */
440 add_stat(tdb, frees, 1);
441 add_free_record(tdb, old_off,
442 sizeof(struct tdb_used_record)
443 + key.dsize + old_room);
444 if (replace_in_hash(tdb, h, new_off) == -1)
447 if (add_to_hash(tdb, h, new_off) == -1)
451 new_off += sizeof(struct tdb_used_record);
452 if (tdb->methods->write(tdb, new_off, key.dptr, key.dsize) == -1)
455 new_off += key.dsize;
456 if (tdb->methods->write(tdb, new_off, dbuf.dptr, dbuf.dsize) == -1)
459 /* FIXME: tdb_increment_seqnum(tdb); */
463 int tdb_store(struct tdb_context *tdb,
464 struct tdb_data key, struct tdb_data dbuf, int flag)
468 tdb_len_t old_room = 0;
469 struct tdb_used_record rec;
472 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
473 if (unlikely(off == TDB_OFF_ERR))
476 /* Now we have lock on this hash bucket. */
477 if (flag == TDB_INSERT) {
479 tdb->ecode = TDB_ERR_EXISTS;
484 old_room = rec_data_length(&rec)
485 + rec_extra_padding(&rec);
486 if (old_room >= dbuf.dsize) {
487 /* Can modify in-place. Easy! */
488 if (update_rec_hdr(tdb, off,
489 key.dsize, dbuf.dsize,
492 if (tdb->methods->write(tdb, off + sizeof(rec)
494 dbuf.dptr, dbuf.dsize))
496 tdb_unlock_hashes(tdb, h.hlock_start,
497 h.hlock_range, F_WRLCK);
501 if (flag == TDB_MODIFY) {
502 /* if the record doesn't exist and we
503 are in TDB_MODIFY mode then we should fail
505 tdb->ecode = TDB_ERR_NOEXIST;
511 /* If we didn't use the old record, this implies we're growing. */
512 ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
513 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
517 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
521 int tdb_append(struct tdb_context *tdb,
522 struct tdb_data key, struct tdb_data dbuf)
526 struct tdb_used_record rec;
527 tdb_len_t old_room = 0, old_dlen;
528 unsigned char *newdata;
529 struct tdb_data new_dbuf;
532 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
533 if (unlikely(off == TDB_OFF_ERR))
537 old_dlen = rec_data_length(&rec);
538 old_room = old_dlen + rec_extra_padding(&rec);
540 /* Fast path: can append in place. */
541 if (rec_extra_padding(&rec) >= dbuf.dsize) {
542 if (update_rec_hdr(tdb, off, key.dsize,
543 old_dlen + dbuf.dsize, &rec, h.h))
546 off += sizeof(rec) + key.dsize + old_dlen;
547 if (tdb->methods->write(tdb, off, dbuf.dptr,
551 /* FIXME: tdb_increment_seqnum(tdb); */
552 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
558 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
560 tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
561 "tdb_append: failed to allocate %zu bytes",
562 (size_t)(key.dsize+old_dlen+dbuf.dsize));
565 if (tdb->methods->read(tdb, off + sizeof(rec) + key.dsize,
566 newdata, old_dlen) != 0) {
570 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
571 new_dbuf.dptr = newdata;
572 new_dbuf.dsize = old_dlen + dbuf.dsize;
578 /* If they're using tdb_append(), it implies they're growing record. */
579 ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
580 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
586 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
590 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
593 struct tdb_used_record rec;
597 off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
598 if (unlikely(off == TDB_OFF_ERR))
602 tdb->ecode = TDB_ERR_NOEXIST;
605 ret.dsize = rec_data_length(&rec);
606 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
610 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
614 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
617 struct tdb_used_record rec;
620 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
621 if (unlikely(off == TDB_OFF_ERR))
625 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
626 tdb->ecode = TDB_ERR_NOEXIST;
630 if (delete_from_hash(tdb, &h) == -1)
633 /* Free the deleted entry. */
634 add_stat(tdb, frees, 1);
635 if (add_free_record(tdb, off,
636 sizeof(struct tdb_used_record)
637 + rec_key_length(&rec)
638 + rec_data_length(&rec)
639 + rec_extra_padding(&rec)) != 0)
642 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
646 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
650 int tdb_close(struct tdb_context *tdb)
652 struct tdb_context **i;
655 tdb_trace(tdb, "tdb_close");
657 if (tdb->transaction) {
658 tdb_transaction_cancel(tdb);
662 if (tdb->flags & TDB_INTERNAL)
667 free((char *)tdb->name);
669 ret = close(tdb->fd);
674 /* Remove from contexts list */
675 for (i = &tdbs; *i; i = &(*i)->next) {
690 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
695 const char *tdb_errorstr(const struct tdb_context *tdb)
697 /* Gcc warns if you miss a case in the switch, so use that. */
698 switch (tdb->ecode) {
699 case TDB_SUCCESS: return "Success";
700 case TDB_ERR_CORRUPT: return "Corrupt database";
701 case TDB_ERR_IO: return "IO Error";
702 case TDB_ERR_LOCK: return "Locking error";
703 case TDB_ERR_OOM: return "Out of memory";
704 case TDB_ERR_EXISTS: return "Record exists";
705 case TDB_ERR_NESTING: return "Transaction already started";
706 case TDB_ERR_EINVAL: return "Invalid parameter";
707 case TDB_ERR_NOEXIST: return "Record does not exist";
708 case TDB_ERR_RDONLY: return "write not permitted";
710 return "Invalid error code";
713 void COLD tdb_logerr(struct tdb_context *tdb,
714 enum TDB_ERROR ecode,
715 enum tdb_debug_level level,
716 const char *fmt, ...)
721 /* tdb_open paths care about errno, so save it. */
722 int saved_errno = errno;
729 /* FIXME: Doesn't assume asprintf. */
731 len = vsnprintf(NULL, 0, fmt, ap);
734 message = malloc(len + 1);
736 tdb->logfn(tdb, level, tdb->log_private,
737 "out of memory formatting message");
741 len = vsprintf(message, fmt, ap);
743 tdb->logfn(tdb, level, tdb->log_private, message);