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 int 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;
101 /* Fill in the header */
102 newdb.hdr.version = TDB_VERSION;
104 newdb.hdr.hash_seed = seed->seed;
106 newdb.hdr.hash_seed = random_number(tdb);
107 newdb.hdr.hash_test = TDB_HASH_MAGIC;
108 newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
109 sizeof(newdb.hdr.hash_test),
112 newdb.hdr.recovery = 0;
113 memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
114 /* Initial hashes are empty. */
115 memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
118 newdb.hdr.free_table = offsetof(struct new_database, ftable);
119 memset(&newdb.ftable, 0, sizeof(newdb.ftable));
120 set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
121 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
122 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr), 0);
125 memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
126 strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
128 /* This creates an endian-converted database, as if read from disk */
129 magic_len = sizeof(newdb.hdr.magic_food);
131 (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
135 if (tdb->flags & TDB_INTERNAL) {
136 tdb->map_size = sizeof(newdb);
137 tdb->map_ptr = malloc(tdb->map_size);
139 tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
140 "tdb_new_database: failed to allocate");
143 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
146 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
149 if (ftruncate(tdb->fd, 0) == -1)
152 rlen = write(tdb->fd, &newdb, sizeof(newdb));
153 if (rlen != sizeof(newdb)) {
156 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
157 "tdb_new_database: %zi writing header: %s",
158 rlen, strerror(errno));
164 struct tdb_context *tdb_open(const char *name, int tdb_flags,
165 int open_flags, mode_t mode,
166 union tdb_attribute *attr)
168 struct tdb_context *tdb;
174 struct tdb_header hdr;
175 struct tdb_attribute_seed *seed = NULL;
177 tdb = malloc(sizeof(*tdb));
185 tdb->direct_access = 0;
187 tdb->map_size = sizeof(struct tdb_header);
188 tdb->ecode = TDB_SUCCESS;
189 tdb->flags = tdb_flags;
191 tdb->transaction = NULL;
199 switch (attr->base.attr) {
200 case TDB_ATTRIBUTE_LOG:
201 tdb->logfn = attr->log.log_fn;
202 tdb->log_private = attr->log.log_private;
204 case TDB_ATTRIBUTE_HASH:
205 tdb->khash = attr->hash.hash_fn;
206 tdb->hash_priv = attr->hash.hash_private;
208 case TDB_ATTRIBUTE_SEED:
211 case TDB_ATTRIBUTE_STATS:
212 tdb->stats = &attr->stats;
213 /* They have stats we don't know about? Tell them. */
214 if (tdb->stats->size > sizeof(attr->stats))
215 tdb->stats->size = sizeof(attr->stats);
218 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
219 "tdb_open: unknown attribute type %u",
223 attr = attr->base.next;
226 if ((open_flags & O_ACCMODE) == O_WRONLY) {
227 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
228 "tdb_open: can't open tdb %s write-only", name);
232 if ((open_flags & O_ACCMODE) == O_RDONLY) {
233 tdb->read_only = true;
234 /* read only databases don't do locking */
235 tdb->flags |= TDB_NOLOCK;
236 tdb->mmap_flags = PROT_READ;
238 tdb->read_only = false;
239 tdb->mmap_flags = PROT_READ | PROT_WRITE;
242 /* internal databases don't need any of the rest. */
243 if (tdb->flags & TDB_INTERNAL) {
244 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
245 if (tdb_new_database(tdb, seed, &hdr) != 0) {
248 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
249 tdb->hash_seed = hdr.hash_seed;
250 tdb_ftable_init(tdb);
254 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
255 /* errno set by open(2) */
257 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
258 "tdb_open: could not open file %s: %s",
259 name, strerror(errno));
263 /* on exec, don't inherit the fd */
264 v = fcntl(tdb->fd, F_GETFD, 0);
265 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
267 /* ensure there is only one process initialising at once */
268 if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
269 /* errno set by tdb_brlock */
274 /* If they used O_TRUNC, read will return 0. */
275 rlen = read(tdb->fd, &hdr, sizeof(hdr));
276 if (rlen == 0 && (open_flags & O_CREAT)) {
277 if (tdb_new_database(tdb, seed, &hdr) == -1) {
280 } else if (rlen < 0) {
281 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
282 "tdb_open: error %s reading %s",
283 strerror(errno), name);
285 } else if (rlen < sizeof(hdr)
286 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
287 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
288 "tdb_open: %s is not a tdb file", name);
292 if (hdr.version != TDB_VERSION) {
293 if (hdr.version == bswap_64(TDB_VERSION))
294 tdb->flags |= TDB_CONVERT;
297 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
298 "tdb_open: %s is unknown version 0x%llx",
299 name, (long long)hdr.version);
304 tdb_convert(tdb, &hdr, sizeof(hdr));
305 tdb->hash_seed = hdr.hash_seed;
306 hash_test = TDB_HASH_MAGIC;
307 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
308 if (hdr.hash_test != hash_test) {
309 /* wrong hash variant */
310 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
311 "tdb_open: %s uses a different hash function",
316 if (fstat(tdb->fd, &st) == -1) {
318 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
319 "tdb_open: could not stat open %s: %s",
320 name, strerror(errno));
324 /* Is it already in the open list? If so, fail. */
325 if (tdb_already_open(st.st_dev, st.st_ino)) {
327 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
328 "tdb_open: %s (%d,%d) is already open in this"
330 name, (int)st.st_dev, (int)st.st_ino);
334 tdb->name = strdup(name);
336 tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
337 "tdb_open: failed to allocate name");
341 tdb->device = st.st_dev;
342 tdb->inode = st.st_ino;
343 tdb_unlock_open(tdb);
345 /* This make sure we have current map_size and mmap. */
346 tdb->methods->oob(tdb, tdb->map_size + 1, true);
348 /* Now it's fully formed, recover if necessary. */
349 if (tdb_needs_recovery(tdb) && tdb_lock_and_recover(tdb) == -1) {
353 if (tdb_ftable_init(tdb) == -1)
361 /* Map ecode to some logical errno. */
363 switch (tdb->ecode) {
364 case TDB_ERR_CORRUPT:
369 saved_errno = EWOULDBLOCK;
372 saved_errno = ENOMEM;
375 saved_errno = EINVAL;
378 saved_errno = EINVAL;
387 if (tdb->flags & TDB_INTERNAL) {
393 free((char *)tdb->name);
395 if (close(tdb->fd) != 0)
396 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
397 "tdb_open: failed to close tdb->fd"
398 " on error: %s", strerror(errno));
404 static int update_rec_hdr(struct tdb_context *tdb,
408 struct tdb_used_record *rec,
411 uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
413 if (set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
414 keylen + dataroom, h))
417 return tdb_write_convert(tdb, off, rec, sizeof(*rec));
420 /* Returns -1 on error, 0 on OK */
421 static int replace_data(struct tdb_context *tdb,
423 struct tdb_data key, struct tdb_data dbuf,
424 tdb_off_t old_off, tdb_len_t old_room,
429 /* Allocate a new record. */
430 new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
432 if (unlikely(new_off == TDB_OFF_ERR))
435 /* We didn't like the existing one: remove it. */
437 add_stat(tdb, frees, 1);
438 add_free_record(tdb, old_off,
439 sizeof(struct tdb_used_record)
440 + key.dsize + old_room);
441 if (replace_in_hash(tdb, h, new_off) == -1)
444 if (add_to_hash(tdb, h, new_off) == -1)
448 new_off += sizeof(struct tdb_used_record);
449 if (tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize) == -1)
452 new_off += key.dsize;
453 if (tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize) == -1)
456 /* FIXME: tdb_increment_seqnum(tdb); */
460 int tdb_store(struct tdb_context *tdb,
461 struct tdb_data key, struct tdb_data dbuf, int flag)
465 tdb_len_t old_room = 0;
466 struct tdb_used_record rec;
469 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
470 if (unlikely(off == TDB_OFF_ERR))
473 /* Now we have lock on this hash bucket. */
474 if (flag == TDB_INSERT) {
476 tdb->ecode = TDB_ERR_EXISTS;
481 old_room = rec_data_length(&rec)
482 + rec_extra_padding(&rec);
483 if (old_room >= dbuf.dsize) {
484 /* Can modify in-place. Easy! */
485 if (update_rec_hdr(tdb, off,
486 key.dsize, dbuf.dsize,
489 if (tdb->methods->twrite(tdb, off + sizeof(rec)
491 dbuf.dptr, dbuf.dsize))
493 tdb_unlock_hashes(tdb, h.hlock_start,
494 h.hlock_range, F_WRLCK);
498 if (flag == TDB_MODIFY) {
499 /* if the record doesn't exist and we
500 are in TDB_MODIFY mode then we should fail
502 tdb->ecode = TDB_ERR_NOEXIST;
508 /* If we didn't use the old record, this implies we're growing. */
509 ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
510 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
514 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
518 int tdb_append(struct tdb_context *tdb,
519 struct tdb_data key, struct tdb_data dbuf)
523 struct tdb_used_record rec;
524 tdb_len_t old_room = 0, old_dlen;
525 unsigned char *newdata;
526 struct tdb_data new_dbuf;
529 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
530 if (unlikely(off == TDB_OFF_ERR))
534 old_dlen = rec_data_length(&rec);
535 old_room = old_dlen + rec_extra_padding(&rec);
537 /* Fast path: can append in place. */
538 if (rec_extra_padding(&rec) >= dbuf.dsize) {
539 if (update_rec_hdr(tdb, off, key.dsize,
540 old_dlen + dbuf.dsize, &rec, h.h))
543 off += sizeof(rec) + key.dsize + old_dlen;
544 if (tdb->methods->twrite(tdb, off, dbuf.dptr,
548 /* FIXME: tdb_increment_seqnum(tdb); */
549 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
555 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
557 tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
558 "tdb_append: failed to allocate %zu bytes",
559 (size_t)(key.dsize+old_dlen+dbuf.dsize));
562 if (tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
563 newdata, old_dlen) != 0) {
567 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
568 new_dbuf.dptr = newdata;
569 new_dbuf.dsize = old_dlen + dbuf.dsize;
575 /* If they're using tdb_append(), it implies they're growing record. */
576 ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
577 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
583 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
587 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
590 struct tdb_used_record rec;
594 off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
595 if (unlikely(off == TDB_OFF_ERR))
599 tdb->ecode = TDB_ERR_NOEXIST;
602 ret.dsize = rec_data_length(&rec);
603 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
607 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
611 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
614 struct tdb_used_record rec;
617 off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
618 if (unlikely(off == TDB_OFF_ERR))
622 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
623 tdb->ecode = TDB_ERR_NOEXIST;
627 if (delete_from_hash(tdb, &h) == -1)
630 /* Free the deleted entry. */
631 add_stat(tdb, frees, 1);
632 if (add_free_record(tdb, off,
633 sizeof(struct tdb_used_record)
634 + rec_key_length(&rec)
635 + rec_data_length(&rec)
636 + rec_extra_padding(&rec)) != 0)
639 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
643 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
647 int tdb_close(struct tdb_context *tdb)
649 struct tdb_context **i;
652 tdb_trace(tdb, "tdb_close");
654 if (tdb->transaction) {
655 tdb_transaction_cancel(tdb);
659 if (tdb->flags & TDB_INTERNAL)
664 free((char *)tdb->name);
666 ret = close(tdb->fd);
671 /* Remove from contexts list */
672 for (i = &tdbs; *i; i = &(*i)->next) {
687 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
692 const char *tdb_errorstr(const struct tdb_context *tdb)
694 /* Gcc warns if you miss a case in the switch, so use that. */
695 switch (tdb->ecode) {
696 case TDB_SUCCESS: return "Success";
697 case TDB_ERR_CORRUPT: return "Corrupt database";
698 case TDB_ERR_IO: return "IO Error";
699 case TDB_ERR_LOCK: return "Locking error";
700 case TDB_ERR_OOM: return "Out of memory";
701 case TDB_ERR_EXISTS: return "Record exists";
702 case TDB_ERR_EINVAL: return "Invalid parameter";
703 case TDB_ERR_NOEXIST: return "Record does not exist";
704 case TDB_ERR_RDONLY: return "write not permitted";
706 return "Invalid error code";
709 void COLD tdb_logerr(struct tdb_context *tdb,
710 enum TDB_ERROR ecode,
711 enum tdb_log_level level,
712 const char *fmt, ...)
717 /* tdb_open paths care about errno, so save it. */
718 int saved_errno = errno;
725 /* FIXME: Doesn't assume asprintf. */
727 len = vsnprintf(NULL, 0, fmt, ap);
730 message = malloc(len + 1);
732 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
733 "out of memory formatting message:");
734 tdb->logfn(tdb, level, tdb->log_private, fmt);
738 len = vsprintf(message, fmt, ap);
740 tdb->logfn(tdb, level, tdb->log_private, message);