2 Unix SMB/Netbios implementation.
4 Samba database functions
5 Copyright (C) Andrew Tridgell 1999-2000
6 Copyright (C) Luke Kenneth Casson Leighton 2000
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2 of the License, or
13 (at your option) any later version.
15 This program is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with this program; if not, write to the Free Software
22 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
44 #define TDB_MAGIC_FOOD "TDB file\n"
45 #define TDB_VERSION (0x26011967 + 6)
46 #define TDB_MAGIC (0x26011999U)
47 #define TDB_FREE_MAGIC (~TDB_MAGIC)
48 #define TDB_DEAD_MAGIC (0xFEE1DEAD)
49 #define TDB_ALIGNMENT 4
50 #define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
51 #define DEFAULT_HASH_SIZE 131
52 #define TDB_PAGE_SIZE 0x2000
53 #define FREELIST_TOP (sizeof(struct tdb_header))
54 #define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
55 #define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
56 #define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
57 #define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
58 #define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off))
59 #define TDB_LOG(x) (tdb->log_fn?((tdb->log_fn x),0) : 0)
70 #define MAP_FAILED ((void *)-1)
73 #define BUCKET(hash) ((hash) % tdb->header.hash_size)
76 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
77 static TDB_CONTEXT *tdbs = NULL;
79 static void tdb_munmap(TDB_CONTEXT *tdb)
81 if (tdb->flags & TDB_INTERNAL)
86 munmap(tdb->map_ptr, tdb->map_size);
91 static void tdb_mmap(TDB_CONTEXT *tdb)
93 if (tdb->flags & TDB_INTERNAL)
97 if (!(tdb->flags & TDB_NOMMAP)) {
98 tdb->map_ptr = mmap(NULL, tdb->map_size,
99 PROT_READ|(tdb->read_only? 0:PROT_WRITE),
100 MAP_SHARED|MAP_FILE, tdb->fd, 0);
103 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
106 if (tdb->map_ptr == MAP_FAILED) {
108 TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n",
109 tdb->map_size, strerror(errno)));
119 /* Endian conversion: we only ever deal with 4 byte quantities */
120 static void *convert(void *buf, u32 size)
123 for (i = 0; i < size / 4; i++)
124 p[i] = TDB_BYTEREV(p[i]);
127 #define DOCONV() (tdb->flags & TDB_CONVERT)
128 #define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x)
130 /* the body of the database is made of one list_struct for the free space
131 plus a separate data list for each hash value */
133 tdb_off next; /* offset of the next record in the list */
134 tdb_len rec_len; /* total byte length of record */
135 tdb_len key_len; /* byte length of key */
136 tdb_len data_len; /* byte length of data */
137 u32 full_hash; /* the full 32 bit hash of the key */
138 u32 magic; /* try to catch errors */
139 /* the following union is implied:
141 char record[rec_len];
146 u32 totalsize; (tailer)
151 /* a byte range locking function - return 0 on success
152 this functions locks/unlocks 1 byte at the specified offset */
153 static int tdb_brlock(TDB_CONTEXT *tdb, tdb_off offset,
154 int rw_type, int lck_type, int probe)
158 if (tdb->flags & TDB_NOLOCK)
164 fl.l_whence = SEEK_SET;
169 if (fcntl(tdb->fd,lck_type,&fl)) {
171 TDB_LOG((tdb, 5,"tdb_brlock failed at offset %d rw_type=%d lck_type=%d\n",
172 offset, rw_type, lck_type));
174 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
179 /* lock a list in the database. list -1 is the alloc list */
180 static int tdb_lock(TDB_CONTEXT *tdb, int list, int ltype)
182 if (list < -1 || list >= (int)tdb->header.hash_size) {
183 TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n",
187 if (tdb->flags & TDB_NOLOCK)
190 /* Since fcntl locks don't nest, we do a lock for the first one,
191 and simply bump the count for future ones */
192 if (tdb->locked[list+1].count == 0) {
193 if (tdb->header.rwlocks) {
194 if (tdb_spinlock(tdb, list, ltype)) {
195 TDB_LOG((tdb, 0, "tdb_lock spinlock on list ltype=%d\n",
199 } else if (tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
200 TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d (%s)\n",
201 list, ltype, strerror(errno)));
204 tdb->locked[list+1].ltype = ltype;
206 tdb->locked[list+1].count++;
210 /* unlock the database: returns void because it's too late for errors. */
211 static void tdb_unlock(TDB_CONTEXT *tdb, int list, int ltype)
213 if (tdb->flags & TDB_NOLOCK)
217 if (list < -1 || list >= (int)tdb->header.hash_size)
219 if (tdb->locked[list+1].count==0)
222 if (tdb->locked[list+1].count == 1) {
223 /* Down to last nested lock: unlock underneath */
224 if (tdb->header.rwlocks)
225 tdb_spinunlock(tdb, list, ltype);
227 tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0);
229 tdb->locked[list+1].count--;
232 /* This is based on the hash agorithm from gdbm */
233 static u32 tdb_hash(TDB_DATA *key)
235 u32 value; /* Used to compute the hash value. */
236 u32 i; /* Used to cycle through random values. */
238 /* Set the initial value from the key size. */
239 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
240 value = (value + (key->dptr[i] << (i*5 % 24)));
242 return (1103515243 * value + 12345);
245 /* check for an out of bounds access - if it is out of bounds then
246 see if the database has been expanded by someone else and expand
248 note that "len" is the minimum length needed for the db
250 static int tdb_oob(TDB_CONTEXT *tdb, tdb_off len, int probe)
253 if (len <= tdb->map_size)
255 if (tdb->flags & TDB_INTERNAL)
258 if (fstat(tdb->fd, &st) == -1)
259 return TDB_ERRCODE(TDB_ERR_IO, -1);
261 if (st.st_size < (size_t)len) {
263 TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n",
264 (int)len, (int)st.st_size));
266 return TDB_ERRCODE(TDB_ERR_IO, -1);
269 /* Unmap, update size, remap */
271 tdb->map_size = st.st_size;
276 /* write a lump of data at a specified offset */
277 static int tdb_write(TDB_CONTEXT *tdb, tdb_off off, void *buf, tdb_len len)
279 if (tdb_oob(tdb, off + len, 0) != 0)
283 memcpy(off + (char *)tdb->map_ptr, buf, len);
285 else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) {
287 else if (lseek(tdb->fd, off, SEEK_SET) != off
288 || write(tdb->fd, buf, len) != (ssize_t)len) {
290 TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n",
291 off, len, strerror(errno)));
292 return TDB_ERRCODE(TDB_ERR_IO, -1);
297 /* read a lump of data at a specified offset, maybe convert */
298 static int tdb_read(TDB_CONTEXT *tdb,tdb_off off,void *buf,tdb_len len,int cv)
300 if (tdb_oob(tdb, off + len, 0) != 0)
304 memcpy(buf, off + (char *)tdb->map_ptr, len);
306 else if (pread(tdb->fd, buf, len, off) != (ssize_t)len) {
308 else if (lseek(tdb->fd, off, SEEK_SET) != off
309 || read(tdb->fd, buf, len) != (ssize_t)len) {
311 TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d (%s)\n",
312 off, len, strerror(errno)));
313 return TDB_ERRCODE(TDB_ERR_IO, -1);
320 /* read a lump of data, allocating the space for it */
321 static char *tdb_alloc_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_len len)
325 if (!(buf = malloc(len))) {
326 TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
327 len, strerror(errno)));
328 return TDB_ERRCODE(TDB_ERR_OOM, buf);
330 if (tdb_read(tdb, offset, buf, len, 0) == -1) {
337 /* read/write a tdb_off */
338 static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
340 return tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
342 static int ofs_write(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
345 return tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
348 /* read/write a record */
349 static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
351 if (tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
353 if (TDB_BAD_MAGIC(rec)) {
354 TDB_LOG((tdb, 0,"rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
355 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
357 return tdb_oob(tdb, rec->next+sizeof(*rec), 0);
359 static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
361 struct list_struct r = *rec;
362 return tdb_write(tdb, offset, CONVERT(r), sizeof(r));
365 /* read a freelist record and check for simple errors */
366 static int rec_free_read(TDB_CONTEXT *tdb, tdb_off off, struct list_struct *rec)
368 if (tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
370 if (rec->magic != TDB_FREE_MAGIC) {
371 TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n",
373 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
375 if (tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
380 /* update a record tailer (must hold allocation lock) */
381 static int update_tailer(TDB_CONTEXT *tdb, tdb_off offset,
382 const struct list_struct *rec)
386 /* Offset of tailer from record header */
387 totalsize = sizeof(*rec) + rec->rec_len;
388 return ofs_write(tdb, offset + totalsize - sizeof(tdb_off),
393 static tdb_off tdb_dump_record(TDB_CONTEXT *tdb, tdb_off offset)
395 struct list_struct rec;
396 tdb_off tailer_ofs, tailer;
398 if (tdb_read(tdb, offset, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
399 printf("ERROR: failed to read record at %u\n", offset);
403 printf(" rec: offset=%u next=%d rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
404 offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, rec.full_hash, rec.magic);
406 tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off);
407 if (ofs_read(tdb, tailer_ofs, &tailer) == -1) {
408 printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
412 if (tailer != rec.rec_len + sizeof(rec)) {
413 printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n", tailer, rec.rec_len + sizeof(rec));
418 static void tdb_dump_chain(TDB_CONTEXT *tdb, int i)
420 tdb_off rec_ptr, top;
422 top = TDB_HASH_TOP(i);
424 tdb_lock(tdb, i, F_WRLCK);
426 if (ofs_read(tdb, top, &rec_ptr) == -1) {
427 tdb_unlock(tdb, i, F_WRLCK);
432 printf("hash=%d\n", i);
435 rec_ptr = tdb_dump_record(tdb, rec_ptr);
437 tdb_unlock(tdb, i, F_WRLCK);
440 void tdb_dump_all(TDB_CONTEXT *tdb)
444 for (i=0;i<tdb->header.hash_size;i++) {
445 tdb_dump_chain(tdb, i);
447 printf("freelist:\n");
448 tdb_dump_chain(tdb, -1);
451 void tdb_printfreelist(TDB_CONTEXT *tdb)
454 tdb_off offset, rec_ptr, last_ptr;
455 struct list_struct rec, lastrec, newrec;
457 tdb_lock(tdb, -1, F_WRLCK);
460 offset = FREELIST_TOP;
462 /* read in the freelist top */
463 if (ofs_read(tdb, offset, &rec_ptr) == -1) {
467 printf("freelist top=[0x%08x]\n", rec_ptr );
469 if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
473 if (rec.magic != TDB_FREE_MAGIC) {
474 printf("bad magic 0x%08x in free list\n", rec.magic);
478 printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)]\n", rec.next, rec.rec_len, rec.rec_len );
479 total_free += rec.rec_len;
481 /* move to the next record */
484 printf("total rec_len = [0x%08x (%d)]\n", total_free, total_free );
486 tdb_unlock(tdb, -1, F_WRLCK);
490 /* Remove an element from the freelist. Must have alloc lock. */
491 static int remove_from_freelist(TDB_CONTEXT *tdb, tdb_off off, tdb_off next)
495 /* read in the freelist top */
496 last_ptr = FREELIST_TOP;
497 while (ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
499 /* We've found it! */
500 return ofs_write(tdb, last_ptr, &next);
502 /* Follow chain (next offset is at start of record) */
505 TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off));
506 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
509 /* Add an element into the freelist. Merge adjacent records if
511 static int tdb_free(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
515 /* Allocation and tailer lock */
516 if (tdb_lock(tdb, -1, F_WRLCK) != 0)
519 /* set an initial tailer, so if we fail we don't leave a bogus record */
520 update_tailer(tdb, offset, rec);
522 /* Look right first (I'm an Australian, dammit) */
523 right = offset + sizeof(*rec) + rec->rec_len;
524 if (right + sizeof(*rec) <= tdb->map_size) {
525 struct list_struct r;
527 if (tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
528 TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", right));
532 /* If it's free, expand to include it. */
533 if (r.magic == TDB_FREE_MAGIC) {
534 if (remove_from_freelist(tdb, right, r.next) == -1) {
535 TDB_LOG((tdb, 0, "tdb_free: right free failed at %u\n", right));
538 rec->rec_len += sizeof(r) + r.rec_len;
544 left = offset - sizeof(tdb_off);
545 if (left > TDB_HASH_TOP(tdb->header.hash_size-1)) {
546 struct list_struct l;
549 /* Read in tailer and jump back to header */
550 if (ofs_read(tdb, left, &leftsize) == -1) {
551 TDB_LOG((tdb, 0, "tdb_free: left offset read failed at %u\n", left));
554 left = offset - leftsize;
556 /* Now read in record */
557 if (tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
558 TDB_LOG((tdb, 0, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
562 /* If it's free, expand to include it. */
563 if (l.magic == TDB_FREE_MAGIC) {
564 if (remove_from_freelist(tdb, left, l.next) == -1) {
565 TDB_LOG((tdb, 0, "tdb_free: left free failed at %u\n", left));
569 rec->rec_len += leftsize;
575 if (update_tailer(tdb, offset, rec) == -1) {
576 TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", offset));
580 /* Now, prepend to free list */
581 rec->magic = TDB_FREE_MAGIC;
583 if (ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
584 rec_write(tdb, offset, rec) == -1 ||
585 ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
586 TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", offset));
590 /* And we're done. */
591 tdb_unlock(tdb, -1, F_WRLCK);
595 tdb_unlock(tdb, -1, F_WRLCK);
600 /* expand a file. we prefer to use ftruncate, as that is what posix
601 says to use for mmap expansion */
602 static int expand_file(TDB_CONTEXT *tdb, tdb_off size, tdb_off addition)
605 #if HAVE_FTRUNCATE_EXTEND
606 if (ftruncate(tdb->fd, size+addition) != 0) {
607 TDB_LOG((tdb, 0, "expand_file ftruncate to %d failed (%s)\n",
608 size+addition, strerror(errno)));
615 if (pwrite(tdb->fd, &b, 1, (size+addition) - 1) != 1) {
617 if (lseek(tdb->fd, (size+addition) - 1, SEEK_SET) != (size+addition) - 1 ||
618 write(tdb->fd, &b, 1) != 1) {
620 TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n",
621 size+addition, strerror(errno)));
626 /* now fill the file with something. This ensures that the file isn't sparse, which would be
627 very bad if we ran out of disk. This must be done with write, not via mmap */
628 memset(buf, 0x42, sizeof(buf));
630 int n = addition>sizeof(buf)?sizeof(buf):addition;
632 int ret = pwrite(tdb->fd, buf, n, size);
635 if (lseek(tdb->fd, size, SEEK_SET) != size)
637 ret = write(tdb->fd, buf, n);
640 TDB_LOG((tdb, 0, "expand_file write of %d failed (%s)\n",
641 n, strerror(errno)));
651 /* expand the database at least size bytes by expanding the underlying
652 file and doing the mmap again if necessary */
653 static int tdb_expand(TDB_CONTEXT *tdb, tdb_off size)
655 struct list_struct rec;
658 if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
659 TDB_LOG((tdb, 0, "lock failed in tdb_expand\n"));
663 /* must know about any previous expansions by another process */
664 tdb_oob(tdb, tdb->map_size + 1, 1);
666 /* always make room for at least 10 more records, and round
667 the database up to a multiple of TDB_PAGE_SIZE */
668 size = TDB_ALIGN(tdb->map_size + size*10, TDB_PAGE_SIZE) - tdb->map_size;
670 if (!(tdb->flags & TDB_INTERNAL))
674 * We must ensure the file is unmapped before doing this
675 * to ensure consistency with systems like OpenBSD where
676 * writes and mmaps are not consistent.
679 /* expand the file itself */
680 if (!(tdb->flags & TDB_INTERNAL)) {
681 if (expand_file(tdb, tdb->map_size, size) != 0)
685 tdb->map_size += size;
687 if (tdb->flags & TDB_INTERNAL)
688 tdb->map_ptr = realloc(tdb->map_ptr, tdb->map_size);
691 * We must ensure the file is remapped before adding the space
692 * to ensure consistency with systems like OpenBSD where
693 * writes and mmaps are not consistent.
696 /* We're ok if the mmap fails as we'll fallback to read/write */
700 /* form a new freelist record */
701 memset(&rec,'\0',sizeof(rec));
702 rec.rec_len = size - sizeof(rec);
704 /* link it into the free list */
705 offset = tdb->map_size - size;
706 if (tdb_free(tdb, offset, &rec) == -1)
709 tdb_unlock(tdb, -1, F_WRLCK);
712 tdb_unlock(tdb, -1, F_WRLCK);
716 /* allocate some space from the free list. The offset returned points
717 to a unconnected list_struct within the database with room for at
718 least length bytes of total data
720 0 is returned if the space could not be allocated
722 static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length,
723 struct list_struct *rec)
725 tdb_off rec_ptr, last_ptr, newrec_ptr;
726 struct list_struct newrec;
728 if (tdb_lock(tdb, -1, F_WRLCK) == -1)
731 /* Extra bytes required for tailer */
732 length += sizeof(tdb_off);
735 last_ptr = FREELIST_TOP;
737 /* read in the freelist top */
738 if (ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
741 /* keep looking until we find a freelist record big enough */
743 if (rec_free_read(tdb, rec_ptr, rec) == -1)
746 if (rec->rec_len >= length) {
747 /* found it - now possibly split it up */
748 if (rec->rec_len > length + MIN_REC_SIZE) {
749 /* Length of left piece */
750 length = TDB_ALIGN(length, TDB_ALIGNMENT);
752 /* Right piece to go on free list */
753 newrec.rec_len = rec->rec_len
754 - (sizeof(*rec) + length);
755 newrec_ptr = rec_ptr + sizeof(*rec) + length;
757 /* And left record is shortened */
758 rec->rec_len = length;
762 /* Remove allocated record from the free list */
763 if (ofs_write(tdb, last_ptr, &rec->next) == -1)
766 /* Update header: do this before we drop alloc
767 lock, otherwise tdb_free() might try to
768 merge with us, thinking we're free.
769 (Thanks Jeremy Allison). */
770 rec->magic = TDB_MAGIC;
771 if (rec_write(tdb, rec_ptr, rec) == -1)
774 /* Did we create new block? */
776 /* Update allocated record tailer (we
778 if (update_tailer(tdb, rec_ptr, rec) == -1)
781 /* Free new record */
782 if (tdb_free(tdb, newrec_ptr, &newrec) == -1)
786 /* all done - return the new record offset */
787 tdb_unlock(tdb, -1, F_WRLCK);
790 /* move to the next record */
794 /* we didn't find enough space. See if we can expand the
795 database and if we can then try again */
796 if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
799 tdb_unlock(tdb, -1, F_WRLCK);
803 /* initialise a new database with a specified hash size */
804 static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size)
806 struct tdb_header *newdb;
809 /* We make it up in memory, then write it out if not internal */
810 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off);
811 if (!(newdb = calloc(size, 1)))
812 return TDB_ERRCODE(TDB_ERR_OOM, -1);
814 /* Fill in the header */
815 newdb->version = TDB_VERSION;
816 newdb->hash_size = hash_size;
818 newdb->rwlocks = size;
820 if (tdb->flags & TDB_INTERNAL) {
821 tdb->map_size = size;
822 tdb->map_ptr = (char *)newdb;
823 memcpy(&tdb->header, newdb, sizeof(tdb->header));
824 /* Convert the `ondisk' version if asked. */
828 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
831 if (ftruncate(tdb->fd, 0) == -1)
834 /* This creates an endian-converted header, as if read from disk */
836 memcpy(&tdb->header, newdb, sizeof(tdb->header));
837 /* Don't endian-convert the magic food! */
838 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
839 if (write(tdb->fd, newdb, size) != size)
842 ret = tdb_create_rwlocks(tdb->fd, hash_size);
849 /* Returns 0 on fail. On success, return offset of record, and fills
851 static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash,
852 struct list_struct *r)
856 /* read in the hash top */
857 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
860 /* keep looking until we find the right record */
862 if (rec_read(tdb, rec_ptr, r) == -1)
865 if (!TDB_DEAD(r) && hash==r->full_hash && key.dsize==r->key_len) {
867 /* a very likely hit - read the key */
868 k = tdb_alloc_read(tdb, rec_ptr + sizeof(*r),
873 if (memcmp(key.dptr, k, key.dsize) == 0) {
881 return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
884 /* If they do lockkeys, check that this hash is one they locked */
885 static int tdb_keylocked(TDB_CONTEXT *tdb, u32 hash)
888 if (!tdb->lockedkeys)
890 for (i = 0; i < tdb->lockedkeys[0]; i++)
891 if (tdb->lockedkeys[i+1] == hash)
893 return TDB_ERRCODE(TDB_ERR_NOLOCK, 0);
896 /* As tdb_find, but if you succeed, keep the lock */
897 static tdb_off tdb_find_lock(TDB_CONTEXT *tdb, TDB_DATA key, int locktype,
898 struct list_struct *rec)
902 hash = tdb_hash(&key);
903 if (!tdb_keylocked(tdb, hash))
905 if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
907 if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
908 tdb_unlock(tdb, BUCKET(hash), locktype);
912 enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb)
917 static struct tdb_errname {
918 enum TDB_ERROR ecode; const char *estring;
919 } emap[] = { {TDB_SUCCESS, "Success"},
920 {TDB_ERR_CORRUPT, "Corrupt database"},
921 {TDB_ERR_IO, "IO Error"},
922 {TDB_ERR_LOCK, "Locking error"},
923 {TDB_ERR_OOM, "Out of memory"},
924 {TDB_ERR_EXISTS, "Record exists"},
925 {TDB_ERR_NOLOCK, "Lock exists on other keys"},
926 {TDB_ERR_NOEXIST, "Record does not exist"} };
928 /* Error string for the last tdb error */
929 const char *tdb_errorstr(TDB_CONTEXT *tdb)
932 for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
933 if (tdb->ecode == emap[i].ecode)
934 return emap[i].estring;
935 return "Invalid error code";
938 /* update an entry in place - this only works if the new data size
939 is <= the old data size and the key exists.
942 static int tdb_update(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf)
944 struct list_struct rec;
949 if (!(rec_ptr = tdb_find_lock(tdb, key, F_WRLCK, &rec)))
952 /* must be long enough key, data and tailer */
953 if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off)) {
954 tdb->ecode = TDB_SUCCESS; /* Not really an error */
958 if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
959 dbuf.dptr, dbuf.dsize) == -1)
962 if (dbuf.dsize != rec.data_len) {
964 rec.data_len = dbuf.dsize;
965 ret = rec_write(tdb, rec_ptr, &rec);
969 tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK);
973 /* find an entry in the database given a key */
974 TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
977 struct list_struct rec;
980 /* find which hash bucket it is in */
981 if (!(rec_ptr = tdb_find_lock(tdb,key,F_RDLCK,&rec)))
984 ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
986 ret.dsize = rec.data_len;
987 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
991 /* check if an entry in the database exists
993 note that 1 is returned if the key is found and 0 is returned if not found
994 this doesn't match the conventions in the rest of this module, but is
997 int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key)
999 struct list_struct rec;
1001 if (tdb_find_lock(tdb, key, F_RDLCK, &rec) == 0)
1003 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
1007 /* record lock stops delete underneath */
1008 static int lock_record(TDB_CONTEXT *tdb, tdb_off off)
1010 return off ? tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
1012 /* write locks override our own fcntl readlocks, so check it here */
1013 static int write_lock_record(TDB_CONTEXT *tdb, tdb_off off)
1015 struct tdb_traverse_lock *i;
1016 for (i = &tdb->travlocks; i; i = i->next)
1019 return tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
1021 static int write_unlock_record(TDB_CONTEXT *tdb, tdb_off off)
1023 return tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
1025 /* fcntl locks don't stack: avoid unlocking someone else's */
1026 static int unlock_record(TDB_CONTEXT *tdb, tdb_off off)
1028 struct tdb_traverse_lock *i;
1033 for (i = &tdb->travlocks; i; i = i->next)
1036 return (count == 1 ? tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
1039 /* actually delete an entry in the database given the offset */
1040 static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec)
1042 tdb_off last_ptr, i;
1043 struct list_struct lastrec;
1045 if (tdb->read_only) return -1;
1047 if (write_lock_record(tdb, rec_ptr) == -1) {
1048 /* Someone traversing here: mark it as dead */
1049 rec->magic = TDB_DEAD_MAGIC;
1050 return rec_write(tdb, rec_ptr, rec);
1052 write_unlock_record(tdb, rec_ptr);
1054 /* find previous record in hash chain */
1055 if (ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
1057 for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
1058 if (rec_read(tdb, i, &lastrec) == -1)
1061 /* unlink it: next ptr is at start of record. */
1063 last_ptr = TDB_HASH_TOP(rec->full_hash);
1064 if (ofs_write(tdb, last_ptr, &rec->next) == -1)
1067 /* recover the space */
1068 if (tdb_free(tdb, rec_ptr, rec) == -1)
1073 /* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
1074 static int tdb_next_lock(TDB_CONTEXT *tdb, struct tdb_traverse_lock *tlock,
1075 struct list_struct *rec)
1077 int want_next = (tlock->off != 0);
1079 /* No traversal allows if you've called tdb_lockkeys() */
1080 if (tdb->lockedkeys)
1081 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1083 /* Lock each chain from the start one. */
1084 for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
1085 if (tdb_lock(tdb, tlock->hash, F_WRLCK) == -1)
1088 /* No previous record? Start at top of chain. */
1090 if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
1094 /* Otherwise unlock the previous record. */
1095 unlock_record(tdb, tlock->off);
1099 /* We have offset of old record: grab next */
1100 if (rec_read(tdb, tlock->off, rec) == -1)
1102 tlock->off = rec->next;
1105 /* Iterate through chain */
1106 while( tlock->off) {
1108 if (rec_read(tdb, tlock->off, rec) == -1)
1110 if (!TDB_DEAD(rec)) {
1111 /* Woohoo: we found one! */
1112 lock_record(tdb, tlock->off);
1115 /* Try to clean dead ones from old traverses */
1116 current = tlock->off;
1117 tlock->off = rec->next;
1118 do_delete(tdb, current, rec);
1120 tdb_unlock(tdb, tlock->hash, F_WRLCK);
1123 /* We finished iteration without finding anything */
1124 return TDB_ERRCODE(TDB_SUCCESS, 0);
1128 tdb_unlock(tdb, tlock->hash, F_WRLCK);
1132 /* traverse the entire database - calling fn(tdb, key, data) on each element.
1133 return -1 on error or the record count traversed
1134 if fn is NULL then it is not called
1135 a non-zero return value from fn() indicates that the traversal should stop
1137 int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *state)
1140 struct list_struct rec;
1141 struct tdb_traverse_lock tl = { NULL, 0, 0 };
1144 /* This was in the initializaton, above, but the IRIX compiler
1145 * did not like it. crh
1147 tl.next = tdb->travlocks.next;
1149 /* fcntl locks don't stack: beware traverse inside traverse */
1150 tdb->travlocks.next = &tl;
1152 /* tdb_next_lock places locks on the record returned, and its chain */
1153 while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) {
1155 /* now read the full record */
1156 key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec),
1157 rec.key_len + rec.data_len);
1159 tdb_unlock(tdb, tl.hash, F_WRLCK);
1160 unlock_record(tdb, tl.off);
1161 tdb->travlocks.next = tl.next;
1164 key.dsize = rec.key_len;
1165 dbuf.dptr = key.dptr + rec.key_len;
1166 dbuf.dsize = rec.data_len;
1168 /* Drop chain lock, call out */
1169 tdb_unlock(tdb, tl.hash, F_WRLCK);
1170 if (fn && fn(tdb, key, dbuf, state)) {
1171 /* They want us to terminate traversal */
1172 unlock_record(tdb, tl.off);
1173 tdb->travlocks.next = tl.next;
1179 tdb->travlocks.next = tl.next;
1186 /* find the first entry in the database and return its key */
1187 TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb)
1190 struct list_struct rec;
1192 /* release any old lock */
1193 unlock_record(tdb, tdb->travlocks.off);
1194 tdb->travlocks.off = tdb->travlocks.hash = 0;
1196 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
1198 /* now read the key */
1199 key.dsize = rec.key_len;
1200 key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
1201 tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK);
1205 /* find the next entry in the database, returning its key */
1206 TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey)
1209 TDB_DATA key = tdb_null;
1210 struct list_struct rec;
1213 /* Is locked key the old key? If so, traverse will be reliable. */
1214 if (tdb->travlocks.off) {
1215 if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
1217 if (rec_read(tdb, tdb->travlocks.off, &rec) == -1
1218 || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
1220 || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
1221 /* No, it wasn't: unlock it and start from scratch */
1222 unlock_record(tdb, tdb->travlocks.off);
1223 tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK);
1224 tdb->travlocks.off = 0;
1231 if (!tdb->travlocks.off) {
1232 /* No previous element: do normal find, and lock record */
1233 tdb->travlocks.off = tdb_find_lock(tdb, oldkey, F_WRLCK, &rec);
1234 if (!tdb->travlocks.off)
1236 tdb->travlocks.hash = BUCKET(rec.full_hash);
1237 lock_record(tdb, tdb->travlocks.off);
1239 oldhash = tdb->travlocks.hash;
1241 /* Grab next record: locks chain and returned record,
1242 unlocks old record */
1243 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
1244 key.dsize = rec.key_len;
1245 key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
1247 /* Unlock the chain of this new record */
1248 tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK);
1250 /* Unlock the chain of old record */
1251 tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK);
1255 /* delete an entry in the database given a key */
1256 int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
1259 struct list_struct rec;
1262 if (!(rec_ptr = tdb_find_lock(tdb, key, F_WRLCK, &rec)))
1264 ret = do_delete(tdb, rec_ptr, &rec);
1265 tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK);
1269 /* store an element in the database, replacing any existing element
1272 return 0 on success, -1 on failure
1274 int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
1276 struct list_struct rec;
1282 /* find which hash bucket it is in */
1283 hash = tdb_hash(&key);
1284 if (!tdb_keylocked(tdb, hash))
1286 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
1289 /* check for it existing, on insert. */
1290 if (flag == TDB_INSERT) {
1291 if (tdb_exists(tdb, key)) {
1292 tdb->ecode = TDB_ERR_EXISTS;
1296 /* first try in-place update, on modify or replace. */
1297 if (tdb_update(tdb, key, dbuf) == 0)
1299 if (flag == TDB_MODIFY && tdb->ecode == TDB_ERR_NOEXIST)
1302 /* reset the error code potentially set by the tdb_update() */
1303 tdb->ecode = TDB_SUCCESS;
1305 /* delete any existing record - if it doesn't exist we don't
1306 care. Doing this first reduces fragmentation, and avoids
1307 coalescing with `allocated' block before it's updated. */
1308 if (flag != TDB_INSERT)
1309 tdb_delete(tdb, key);
1311 /* Copy key+value *before* allocating free space in case malloc
1312 fails and we are left with a dead spot in the tdb. */
1314 if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) {
1315 tdb->ecode = TDB_ERR_OOM;
1319 memcpy(p, key.dptr, key.dsize);
1320 memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
1322 /* now we're into insert / modify / replace of a record which
1323 * we know could not be optimised by an in-place store (for
1324 * various reasons). */
1325 if (!(rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec)))
1328 /* Read hash top into next ptr */
1329 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
1332 rec.key_len = key.dsize;
1333 rec.data_len = dbuf.dsize;
1334 rec.full_hash = hash;
1335 rec.magic = TDB_MAGIC;
1337 /* write out and point the top of the hash chain at it */
1338 if (rec_write(tdb, rec_ptr, &rec) == -1
1339 || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
1340 || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
1342 /* Need to tdb_unallocate() here */
1348 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
1352 /* open the database, creating it if necessary
1354 The open_flags and mode are passed straight to the open call on the
1355 database file. A flags value of O_WRONLY is invalid. The hash size
1356 is advisory, use zero for a default value.
1358 return is NULL on error */
1359 TDB_CONTEXT *tdb_open(char *name, int hash_size, int tdb_flags,
1360 int open_flags, mode_t mode)
1362 TDB_CONTEXT tdb, *ret, *i;
1364 int rev = 0, locked;
1366 memset(&tdb, 0, sizeof(tdb));
1370 tdb.lockedkeys = NULL;
1371 tdb.flags = tdb_flags;
1373 if ((open_flags & O_ACCMODE) == O_WRONLY)
1376 hash_size = DEFAULT_HASH_SIZE;
1377 if ((open_flags & O_ACCMODE) == O_RDONLY) {
1379 /* read only databases don't do locking */
1380 tdb.flags |= TDB_NOLOCK;
1383 /* internal databases don't mmap or lock, and start off cleared */
1384 if (tdb.flags & TDB_INTERNAL) {
1385 tdb.flags |= (TDB_NOLOCK | TDB_NOMMAP);
1386 tdb.flags &= ~TDB_CLEAR_IF_FIRST;
1387 tdb_new_database(&tdb, hash_size);
1391 if ((tdb.fd = open(name, open_flags, mode)) == -1)
1394 /* ensure there is only one process initialising at once */
1395 if (tdb_brlock(&tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1)
1398 /* we need to zero database if we are the only one with it open */
1399 if ((locked = (tdb_brlock(&tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))
1400 && (tdb_flags & TDB_CLEAR_IF_FIRST)) {
1401 open_flags |= O_CREAT;
1402 if (ftruncate(tdb.fd, 0) == -1)
1406 if (read(tdb.fd, &tdb.header, sizeof(tdb.header)) != sizeof(tdb.header)
1407 || strcmp(tdb.header.magic_food, TDB_MAGIC_FOOD) != 0
1408 || (tdb.header.version != TDB_VERSION
1409 && !(rev = (tdb.header.version==TDB_BYTEREV(TDB_VERSION))))) {
1410 /* its not a valid database - possibly initialise it */
1411 if (!(open_flags & O_CREAT) || tdb_new_database(&tdb, hash_size) == -1)
1413 rev = (tdb.flags & TDB_CONVERT);
1416 tdb.flags &= ~TDB_CONVERT;
1418 tdb.flags |= TDB_CONVERT;
1419 convert(&tdb.header, sizeof(tdb.header));
1421 if (fstat(tdb.fd, &st) == -1)
1424 /* Is it already in the open list? If so, fail. */
1425 for (i = tdbs; i; i = i->next) {
1426 if (i->device == st.st_dev && i->inode == st.st_ino) {
1433 /* map the database and fill in the return structure */
1434 tdb.name = (char *)strdup(name);
1437 tdb.map_size = st.st_size;
1438 tdb.device = st.st_dev;
1439 tdb.inode = st.st_ino;
1440 tdb.locked = calloc(tdb.header.hash_size+1, sizeof(tdb.locked[0]));
1445 tdb_clear_spinlocks(&tdb);
1446 if (tdb_brlock(&tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1)
1449 /* leave this lock in place to indicate it's in use */
1450 if (tdb_brlock(&tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
1454 if (!(ret = malloc(sizeof(tdb))))
1457 if (tdb_brlock(&tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
1465 if (tdb.flags & TDB_INTERNAL)
1479 /* close a database */
1480 int tdb_close(TDB_CONTEXT *tdb)
1486 if (tdb->flags & TDB_INTERNAL)
1494 ret = close(tdb->fd);
1497 if (tdb->lockedkeys)
1498 free(tdb->lockedkeys);
1500 /* Remove from contexts list */
1501 for (i = &tdbs; *i; i = &(*i)->next) {
1508 memset(tdb, 0, sizeof(*tdb));
1514 /* lock/unlock entire database */
1515 int tdb_lockall(TDB_CONTEXT *tdb)
1519 /* There are no locks on read-only dbs */
1521 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
1522 if (tdb->lockedkeys)
1523 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1524 for (i = 0; i < tdb->header.hash_size; i++)
1525 if (tdb_lock(tdb, i, F_WRLCK))
1528 /* If error, release locks we have... */
1529 if (i < tdb->header.hash_size) {
1532 for ( j = 0; j < i; j++)
1533 tdb_unlock(tdb, j, F_WRLCK);
1534 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1539 void tdb_unlockall(TDB_CONTEXT *tdb)
1542 for (i=0; i < tdb->header.hash_size; i++)
1543 tdb_unlock(tdb, i, F_WRLCK);
1546 int tdb_lockkeys(TDB_CONTEXT *tdb, u32 number, TDB_DATA keys[])
1550 /* Can't lock more keys if already locked */
1551 if (tdb->lockedkeys)
1552 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1553 if (!(tdb->lockedkeys = malloc(sizeof(u32) * (number+1))))
1554 return TDB_ERRCODE(TDB_ERR_OOM, -1);
1555 /* First number in array is # keys */
1556 tdb->lockedkeys[0] = number;
1558 /* Insertion sort by bucket */
1559 for (i = 0; i < number; i++) {
1560 hash = tdb_hash(&keys[i]);
1561 for (j = 0; j < i && BUCKET(tdb->lockedkeys[j+1]) < BUCKET(hash); j++);
1562 memmove(&tdb->lockedkeys[j+2], &tdb->lockedkeys[j+1], sizeof(u32) * (i-j));
1563 tdb->lockedkeys[j+1] = hash;
1565 /* Finally, lock in order */
1566 for (i = 0; i < number; i++)
1567 if (tdb_lock(tdb, i, F_WRLCK))
1570 /* If error, release locks we have... */
1572 for ( j = 0; j < i; j++)
1573 tdb_unlock(tdb, j, F_WRLCK);
1574 free(tdb->lockedkeys);
1575 tdb->lockedkeys = NULL;
1576 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1581 /* Unlock the keys previously locked by tdb_lockkeys() */
1582 void tdb_unlockkeys(TDB_CONTEXT *tdb)
1585 for (i = 0; i < tdb->lockedkeys[0]; i++)
1586 tdb_unlock(tdb, tdb->lockedkeys[i+1], F_WRLCK);
1587 free(tdb->lockedkeys);
1588 tdb->lockedkeys = NULL;
1591 /* lock/unlock one hash chain. This is meant to be used to reduce
1592 contention - it cannot guarantee how many records will be locked */
1593 int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key)
1595 return tdb_lock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
1597 void tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key)
1599 tdb_unlock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
1603 /* register a loging function */
1604 void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , const char *, ...))