4 /* all lock info, to detect double-opens (fcntl file don't nest!) */
5 static struct tdb_file *files = NULL;
7 static struct tdb_file *find_file(dev_t device, ino_t ino)
11 for (i = files; i; i = i->next) {
12 if (i->device == device && i->inode == ino) {
20 static bool read_all(int fd, void *buf, size_t len)
24 ret = read(fd, buf, len);
32 buf = (char *)buf + ret;
38 static uint64_t random_number(struct tdb_context *tdb)
44 fd = open("/dev/urandom", O_RDONLY);
46 if (read_all(fd, &ret, sizeof(ret))) {
52 /* FIXME: Untested! Based on Wikipedia protocol description! */
53 fd = open("/dev/egd-pool", O_RDWR);
55 /* Command is 1, next byte is size we want to read. */
56 char cmd[2] = { 1, sizeof(uint64_t) };
57 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
58 char reply[1 + sizeof(uint64_t)];
59 int r = read(fd, reply, sizeof(reply));
61 /* Copy at least some bytes. */
62 memcpy(&ret, reply+1, r - 1);
63 if (reply[0] == sizeof(uint64_t)
64 && r == sizeof(reply)) {
73 /* Fallback: pid and time. */
74 gettimeofday(&now, NULL);
75 ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
76 tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
77 "tdb_open: random from getpid and time");
82 struct tdb_header hdr;
83 struct tdb_freetable ftable;
86 /* initialise a new database */
87 static enum TDB_ERROR tdb_new_database(struct tdb_context *tdb,
88 struct tdb_attribute_seed *seed,
89 struct tdb_header *hdr)
91 /* We make it up in memory, then write it out if not internal */
92 struct new_database newdb;
93 unsigned int magic_len;
97 /* Fill in the header */
98 newdb.hdr.version = TDB_VERSION;
100 newdb.hdr.hash_seed = seed->seed;
102 newdb.hdr.hash_seed = random_number(tdb);
103 newdb.hdr.hash_test = TDB_HASH_MAGIC;
104 newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
105 sizeof(newdb.hdr.hash_test),
108 newdb.hdr.recovery = 0;
109 newdb.hdr.features_used = newdb.hdr.features_offered = TDB_FEATURE_MASK;
110 memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
111 /* Initial hashes are empty. */
112 memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
115 newdb.hdr.free_table = offsetof(struct new_database, ftable);
116 memset(&newdb.ftable, 0, sizeof(newdb.ftable));
117 ecode = set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
118 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
119 sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
121 if (ecode != TDB_SUCCESS) {
126 memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
127 strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
129 /* This creates an endian-converted database, as if read from disk */
130 magic_len = sizeof(newdb.hdr.magic_food);
132 (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
136 if (tdb->flags & TDB_INTERNAL) {
137 tdb->file->map_size = sizeof(newdb);
138 tdb->file->map_ptr = malloc(tdb->file->map_size);
139 if (!tdb->file->map_ptr) {
140 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
142 " failed to allocate");
144 memcpy(tdb->file->map_ptr, &newdb, tdb->file->map_size);
147 if (lseek(tdb->file->fd, 0, SEEK_SET) == -1) {
148 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
150 " failed to seek: %s", strerror(errno));
153 if (ftruncate(tdb->file->fd, 0) == -1) {
154 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
156 " failed to truncate: %s", strerror(errno));
159 rlen = write(tdb->file->fd, &newdb, sizeof(newdb));
160 if (rlen != sizeof(newdb)) {
163 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
164 "tdb_new_database: %zi writing header: %s",
165 rlen, strerror(errno));
170 static enum TDB_ERROR tdb_new_file(struct tdb_context *tdb)
172 tdb->file = malloc(sizeof(*tdb->file));
174 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
175 "tdb_open: could alloc tdb_file structure");
176 tdb->file->num_lockrecs = 0;
177 tdb->file->lockrecs = NULL;
178 tdb->file->allrecord_lock.count = 0;
179 tdb->file->refcnt = 1;
183 struct tdb_context *tdb_open(const char *name, int tdb_flags,
184 int open_flags, mode_t mode,
185 union tdb_attribute *attr)
187 struct tdb_context *tdb;
193 struct tdb_header hdr;
194 struct tdb_attribute_seed *seed = NULL;
196 enum TDB_ERROR ecode;
198 tdb = malloc(sizeof(*tdb));
205 tdb->direct_access = 0;
206 tdb->flags = tdb_flags;
208 tdb->transaction = NULL;
216 switch (attr->base.attr) {
217 case TDB_ATTRIBUTE_LOG:
218 tdb->logfn = attr->log.log_fn;
219 tdb->log_private = attr->log.log_private;
221 case TDB_ATTRIBUTE_HASH:
222 tdb->khash = attr->hash.hash_fn;
223 tdb->hash_priv = attr->hash.hash_private;
225 case TDB_ATTRIBUTE_SEED:
228 case TDB_ATTRIBUTE_STATS:
229 tdb->stats = &attr->stats;
230 /* They have stats we don't know about? Tell them. */
231 if (tdb->stats->size > sizeof(attr->stats))
232 tdb->stats->size = sizeof(attr->stats);
235 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL,
238 " unknown attribute type %u",
242 attr = attr->base.next;
245 if (tdb_flags & ~(TDB_INTERNAL | TDB_NOLOCK | TDB_NOMMAP | TDB_CONVERT
247 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
248 "tdb_open: unknown flags %u", tdb_flags);
252 if ((open_flags & O_ACCMODE) == O_WRONLY) {
253 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
254 "tdb_open: can't open tdb %s write-only",
259 if ((open_flags & O_ACCMODE) == O_RDONLY) {
260 tdb->read_only = true;
261 tdb->mmap_flags = PROT_READ;
263 tdb->read_only = false;
264 tdb->mmap_flags = PROT_READ | PROT_WRITE;
267 /* internal databases don't need any of the rest. */
268 if (tdb->flags & TDB_INTERNAL) {
269 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
270 ecode = tdb_new_file(tdb);
271 if (ecode != TDB_SUCCESS) {
275 ecode = tdb_new_database(tdb, seed, &hdr);
276 if (ecode != TDB_SUCCESS) {
279 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
280 tdb->hash_seed = hdr.hash_seed;
281 tdb_ftable_init(tdb);
285 if (stat(name, &st) != -1)
286 tdb->file = find_file(st.st_dev, st.st_ino);
291 if ((fd = open(name, open_flags, mode)) == -1) {
292 /* errno set by open(2) */
294 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
295 "tdb_open: could not open file %s: %s",
296 name, strerror(errno));
300 /* on exec, don't inherit the fd */
301 v = fcntl(fd, F_GETFD, 0);
302 fcntl(fd, F_SETFD, v | FD_CLOEXEC);
304 if (fstat(fd, &st) == -1) {
306 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
307 "tdb_open: could not stat open %s: %s",
308 name, strerror(errno));
312 ecode = tdb_new_file(tdb);
313 if (ecode != TDB_SUCCESS)
316 tdb->file->next = files;
318 tdb->file->device = st.st_dev;
319 tdb->file->inode = st.st_ino;
320 tdb->file->map_ptr = NULL;
321 tdb->file->map_size = sizeof(struct tdb_header);
324 /* ensure there is only one process initialising at once */
325 ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
326 if (ecode != TDB_SUCCESS) {
330 /* If they used O_TRUNC, read will return 0. */
331 rlen = pread(tdb->file->fd, &hdr, sizeof(hdr), 0);
332 if (rlen == 0 && (open_flags & O_CREAT)) {
333 ecode = tdb_new_database(tdb, seed, &hdr);
334 if (ecode != TDB_SUCCESS) {
337 } else if (rlen < 0) {
338 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
339 "tdb_open: error %s reading %s",
340 strerror(errno), name);
342 } else if (rlen < sizeof(hdr)
343 || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
344 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
345 "tdb_open: %s is not a tdb file", name);
349 if (hdr.version != TDB_VERSION) {
350 if (hdr.version == bswap_64(TDB_VERSION))
351 tdb->flags |= TDB_CONVERT;
354 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
356 " %s is unknown version 0x%llx",
357 name, (long long)hdr.version);
362 tdb_convert(tdb, &hdr, sizeof(hdr));
363 tdb->hash_seed = hdr.hash_seed;
364 hash_test = TDB_HASH_MAGIC;
365 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
366 if (hdr.hash_test != hash_test) {
367 /* wrong hash variant */
368 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
370 " %s uses a different hash function",
375 tdb->name = strdup(name);
377 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
378 "tdb_open: failed to allocate name");
382 /* Clear any features we don't understand. */
383 if ((open_flags & O_ACCMODE) != O_RDONLY) {
384 hdr.features_used &= TDB_FEATURE_MASK;
385 if (tdb_write_convert(tdb, offsetof(struct tdb_header,
388 sizeof(hdr.features_used)) == -1)
392 tdb_unlock_open(tdb);
394 /* This make sure we have current map_size and mmap. */
395 tdb->methods->oob(tdb, tdb->file->map_size + 1, true);
397 /* Now it's fully formed, recover if necessary. */
398 berr = tdb_needs_recovery(tdb);
399 if (unlikely(berr != false)) {
404 ecode = tdb_lock_and_recover(tdb);
405 if (ecode != TDB_SUCCESS) {
410 ecode = tdb_ftable_init(tdb);
411 if (ecode != TDB_SUCCESS) {
415 /* Add to linked list if we're new. */
416 if (tdb->file->refcnt == 1)
421 /* Map ecode to some logical errno. */
423 case TDB_ERR_CORRUPT:
428 saved_errno = EWOULDBLOCK;
431 saved_errno = ENOMEM;
434 saved_errno = EINVAL;
437 saved_errno = EINVAL;
445 free((char *)tdb->name);
448 if (--tdb->file->refcnt == 0) {
449 assert(tdb->file->num_lockrecs == 0);
450 if (tdb->file->map_ptr) {
451 if (tdb->flags & TDB_INTERNAL) {
452 free(tdb->file->map_ptr);
454 tdb_munmap(tdb->file);
456 if (close(tdb->file->fd) != 0)
457 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
458 "tdb_open: failed to close tdb fd"
459 " on error: %s", strerror(errno));
460 free(tdb->file->lockrecs);
470 int tdb_close(struct tdb_context *tdb)
474 tdb_trace(tdb, "tdb_close");
476 if (tdb->transaction) {
477 tdb_transaction_cancel(tdb);
480 if (tdb->file->map_ptr) {
481 if (tdb->flags & TDB_INTERNAL)
482 free(tdb->file->map_ptr);
484 tdb_munmap(tdb->file);
486 free((char *)tdb->name);
491 if (--tdb->file->refcnt == 0) {
492 ret = close(tdb->file->fd);
494 /* Remove from files list */
495 for (i = &files; *i; i = &(*i)->next) {
496 if (*i == tdb->file) {
497 *i = tdb->file->next;
501 free(tdb->file->lockrecs);