2 Unix SMB/CIFS implementation.
3 Samba database functions
4 Copyright (C) Andrew Tridgell 1999-2000
5 Copyright (C) Luke Kenneth Casson Leighton 2000
6 Copyright (C) Paul `Rusty' Russell 2000
7 Copyright (C) Jeremy Allison 2000-2003
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 2 of the License, or
12 (at your option) any later version.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 /* NOTE: If you use tdbs under valgrind, and in particular if you run
26 * tdbtorture, you may get spurious "uninitialized value" warnings. I
27 * think this is because valgrind doesn't understand that the mmap'd
28 * area may be written to by other processes. Memory can, from the
29 * point of view of the grinded process, spontaneously become
32 * I can think of a few solutions. [mbp 20030311]
34 * 1 - Write suppressions for Valgrind so that it doesn't complain
35 * about this. Probably the most reasonable but people need to
36 * remember to use them.
38 * 2 - Use IO not mmap when running under valgrind. Not so nice.
40 * 3 - Use the special valgrind macros to mark memory as valid at the
41 * right time. Probably too hard -- the process just doesn't know.
55 #define _KLIBC_HAS_ARCH_SIG_ATOMIC_T
72 #define TDB_MAGIC_FOOD "TDB file\n"
73 #define TDB_VERSION (0x26011967 + 6)
74 #define TDB_MAGIC (0x26011999U)
75 #define TDB_FREE_MAGIC (~TDB_MAGIC)
76 #define TDB_DEAD_MAGIC (0xFEE1DEAD)
77 #define TDB_ALIGNMENT 4
78 #define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
79 #define DEFAULT_HASH_SIZE 131
80 #define TDB_PAGE_SIZE 0x2000
81 #define FREELIST_TOP (sizeof(struct tdb_header))
82 #define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
83 #define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
84 #define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
85 #define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
86 #define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off))
88 /* NB assumes there is a local variable called "tdb" that is the
89 * current context, also takes doubly-parenthesized print-style
91 #define TDB_LOG(x) (tdb->log_fn?((tdb->log_fn x),0) : 0)
102 #define MAP_FAILED ((void *)-1)
105 /* free memory if the pointer is valid and zero the pointer */
107 #define SAFE_FREE(x) do { if ((x) != NULL) {free((x)); (x)=NULL;} } while(0)
110 #define BUCKET(hash) ((hash) % tdb->header.hash_size)
113 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
114 static TDB_CONTEXT *tdbs = NULL;
116 static int tdb_munmap(TDB_CONTEXT *tdb)
118 if (tdb->flags & TDB_INTERNAL)
123 int ret = munmap(tdb->map_ptr, tdb->map_size);
132 static void tdb_mmap(TDB_CONTEXT *tdb)
134 if (tdb->flags & TDB_INTERNAL)
138 if (!(tdb->flags & TDB_NOMMAP)) {
139 tdb->map_ptr = mmap(NULL, tdb->map_size,
140 PROT_READ|(tdb->read_only? 0:PROT_WRITE),
141 MAP_SHARED|MAP_FILE, tdb->fd, 0);
144 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
147 if (tdb->map_ptr == MAP_FAILED) {
149 TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n",
150 tdb->map_size, strerror(errno)));
160 /* Endian conversion: we only ever deal with 4 byte quantities */
161 static void *convert(void *buf, u32 size)
164 for (i = 0; i < size / 4; i++)
165 p[i] = TDB_BYTEREV(p[i]);
168 #define DOCONV() (tdb->flags & TDB_CONVERT)
169 #define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x)
171 /* the body of the database is made of one list_struct for the free space
172 plus a separate data list for each hash value */
174 tdb_off next; /* offset of the next record in the list */
175 tdb_len rec_len; /* total byte length of record */
176 tdb_len key_len; /* byte length of key */
177 tdb_len data_len; /* byte length of data */
178 u32 full_hash; /* the full 32 bit hash of the key */
179 u32 magic; /* try to catch errors */
180 /* the following union is implied:
182 char record[rec_len];
187 u32 totalsize; (tailer)
192 /***************************************************************
193 Allow a caller to set a "alarm" flag that tdb can check to abort
194 a blocking lock on SIGALRM.
195 ***************************************************************/
197 static sig_atomic_t *palarm_fired;
199 void tdb_set_lock_alarm(sig_atomic_t *palarm)
201 palarm_fired = palarm;
204 /* a byte range locking function - return 0 on success
205 this functions locks/unlocks 1 byte at the specified offset.
207 On error, errno is also set so that errors are passed back properly
208 through tdb_open(). */
209 static int tdb_brlock(TDB_CONTEXT *tdb, tdb_off offset,
210 int rw_type, int lck_type, int probe)
215 if (tdb->flags & TDB_NOLOCK)
217 if ((rw_type == F_WRLCK) && (tdb->read_only)) {
223 fl.l_whence = SEEK_SET;
229 ret = fcntl(tdb->fd,lck_type,&fl);
230 if (ret == -1 && errno == EINTR && palarm_fired && *palarm_fired)
232 } while (ret == -1 && errno == EINTR);
235 if (!probe && lck_type != F_SETLK) {
236 /* Ensure error code is set for log fun to examine. */
237 if (errno == EINTR && palarm_fired && *palarm_fired)
238 tdb->ecode = TDB_ERR_LOCK_TIMEOUT;
240 tdb->ecode = TDB_ERR_LOCK;
241 TDB_LOG((tdb, 5,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d\n",
242 tdb->fd, offset, rw_type, lck_type));
244 /* Was it an alarm timeout ? */
245 if (errno == EINTR && palarm_fired && *palarm_fired)
246 return TDB_ERRCODE(TDB_ERR_LOCK_TIMEOUT, -1);
247 /* Otherwise - generic lock error. */
248 /* errno set by fcntl */
249 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
254 /* lock a list in the database. list -1 is the alloc list */
255 static int tdb_lock(TDB_CONTEXT *tdb, int list, int ltype)
257 if (list < -1 || list >= (int)tdb->header.hash_size) {
258 TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n",
262 if (tdb->flags & TDB_NOLOCK)
265 /* Since fcntl locks don't nest, we do a lock for the first one,
266 and simply bump the count for future ones */
267 if (tdb->locked[list+1].count == 0) {
268 if (!tdb->read_only && tdb->header.rwlocks) {
269 if (tdb_spinlock(tdb, list, ltype)) {
270 TDB_LOG((tdb, 0, "tdb_lock spinlock failed on list ltype=%d\n",
274 } else if (tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
275 TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d (%s)\n",
276 list, ltype, strerror(errno)));
279 tdb->locked[list+1].ltype = ltype;
281 tdb->locked[list+1].count++;
285 /* unlock the database: returns void because it's too late for errors. */
286 /* changed to return int it may be interesting to know there
287 has been an error --simo */
288 static int tdb_unlock(TDB_CONTEXT *tdb, int list, int ltype)
292 if (tdb->flags & TDB_NOLOCK)
296 if (list < -1 || list >= (int)tdb->header.hash_size) {
297 TDB_LOG((tdb, 0, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
301 if (tdb->locked[list+1].count==0) {
302 TDB_LOG((tdb, 0, "tdb_unlock: count is 0\n"));
306 if (tdb->locked[list+1].count == 1) {
307 /* Down to last nested lock: unlock underneath */
308 if (!tdb->read_only && tdb->header.rwlocks) {
309 ret = tdb_spinunlock(tdb, list, ltype);
311 ret = tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0);
316 tdb->locked[list+1].count--;
319 TDB_LOG((tdb, 0,"tdb_unlock: An error occurred unlocking!\n"));
323 /* This is based on the hash algorithm from gdbm */
324 static u32 tdb_hash(TDB_DATA *key)
326 u32 value; /* Used to compute the hash value. */
327 u32 i; /* Used to cycle through random values. */
329 /* Set the initial value from the key size. */
330 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
331 value = (value + (key->dptr[i] << (i*5 % 24)));
333 return (1103515243 * value + 12345);
336 /* check for an out of bounds access - if it is out of bounds then
337 see if the database has been expanded by someone else and expand
339 note that "len" is the minimum length needed for the db
341 static int tdb_oob(TDB_CONTEXT *tdb, tdb_off len, int probe)
344 if (len <= tdb->map_size)
346 if (tdb->flags & TDB_INTERNAL) {
348 /* Ensure ecode is set for log fn. */
349 tdb->ecode = TDB_ERR_IO;
350 TDB_LOG((tdb, 0,"tdb_oob len %d beyond internal malloc size %d\n",
351 (int)len, (int)tdb->map_size));
353 return TDB_ERRCODE(TDB_ERR_IO, -1);
356 if (fstat(tdb->fd, &st) == -1)
357 return TDB_ERRCODE(TDB_ERR_IO, -1);
359 if (st.st_size < (size_t)len) {
361 /* Ensure ecode is set for log fn. */
362 tdb->ecode = TDB_ERR_IO;
363 TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n",
364 (int)len, (int)st.st_size));
366 return TDB_ERRCODE(TDB_ERR_IO, -1);
369 /* Unmap, update size, remap */
370 if (tdb_munmap(tdb) == -1)
371 return TDB_ERRCODE(TDB_ERR_IO, -1);
372 tdb->map_size = st.st_size;
377 /* write a lump of data at a specified offset */
378 static int tdb_write(TDB_CONTEXT *tdb, tdb_off off, void *buf, tdb_len len)
380 if (tdb_oob(tdb, off + len, 0) != 0)
384 memcpy(off + (char *)tdb->map_ptr, buf, len);
386 else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) {
388 else if (lseek(tdb->fd, off, SEEK_SET) != off
389 || write(tdb->fd, buf, len) != (ssize_t)len) {
391 /* Ensure ecode is set for log fn. */
392 tdb->ecode = TDB_ERR_IO;
393 TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n",
394 off, len, strerror(errno)));
395 return TDB_ERRCODE(TDB_ERR_IO, -1);
400 /* read a lump of data at a specified offset, maybe convert */
401 static int tdb_read(TDB_CONTEXT *tdb,tdb_off off,void *buf,tdb_len len,int cv)
403 if (tdb_oob(tdb, off + len, 0) != 0)
407 memcpy(buf, off + (char *)tdb->map_ptr, len);
409 else if (pread(tdb->fd, buf, len, off) != (ssize_t)len) {
411 else if (lseek(tdb->fd, off, SEEK_SET) != off
412 || read(tdb->fd, buf, len) != (ssize_t)len) {
414 /* Ensure ecode is set for log fn. */
415 tdb->ecode = TDB_ERR_IO;
416 TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d (%s)\n",
417 off, len, strerror(errno)));
418 return TDB_ERRCODE(TDB_ERR_IO, -1);
425 /* read a lump of data, allocating the space for it */
426 static char *tdb_alloc_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_len len)
430 if (!(buf = malloc(len))) {
431 /* Ensure ecode is set for log fn. */
432 tdb->ecode = TDB_ERR_OOM;
433 TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
434 len, strerror(errno)));
435 return TDB_ERRCODE(TDB_ERR_OOM, buf);
437 if (tdb_read(tdb, offset, buf, len, 0) == -1) {
444 /* read/write a tdb_off */
445 static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
447 return tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
449 static int ofs_write(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
452 return tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
455 /* read/write a record */
456 static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
458 if (tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
460 if (TDB_BAD_MAGIC(rec)) {
461 /* Ensure ecode is set for log fn. */
462 tdb->ecode = TDB_ERR_CORRUPT;
463 TDB_LOG((tdb, 0,"rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
464 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
466 return tdb_oob(tdb, rec->next+sizeof(*rec), 0);
468 static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
470 struct list_struct r = *rec;
471 return tdb_write(tdb, offset, CONVERT(r), sizeof(r));
474 /* read a freelist record and check for simple errors */
475 static int rec_free_read(TDB_CONTEXT *tdb, tdb_off off, struct list_struct *rec)
477 if (tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
480 if (rec->magic == TDB_MAGIC) {
481 /* this happens when a app is showdown while deleting a record - we should
482 not completely fail when this happens */
483 TDB_LOG((tdb, 0,"rec_free_read non-free magic at offset=%d - fixing\n",
485 rec->magic = TDB_FREE_MAGIC;
486 if (tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
490 if (rec->magic != TDB_FREE_MAGIC) {
491 /* Ensure ecode is set for log fn. */
492 tdb->ecode = TDB_ERR_CORRUPT;
493 TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n",
495 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
497 if (tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
502 /* update a record tailer (must hold allocation lock) */
503 static int update_tailer(TDB_CONTEXT *tdb, tdb_off offset,
504 const struct list_struct *rec)
508 /* Offset of tailer from record header */
509 totalsize = sizeof(*rec) + rec->rec_len;
510 return ofs_write(tdb, offset + totalsize - sizeof(tdb_off),
514 static tdb_off tdb_dump_record(TDB_CONTEXT *tdb, tdb_off offset)
516 struct list_struct rec;
517 tdb_off tailer_ofs, tailer;
519 if (tdb_read(tdb, offset, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
520 printf("ERROR: failed to read record at %u\n", offset);
524 printf(" rec: offset=%u next=%d rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
525 offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, rec.full_hash, rec.magic);
527 tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off);
528 if (ofs_read(tdb, tailer_ofs, &tailer) == -1) {
529 printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
533 if (tailer != rec.rec_len + sizeof(rec)) {
534 printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
535 (unsigned)tailer, (unsigned)(rec.rec_len + sizeof(rec)));
540 static int tdb_dump_chain(TDB_CONTEXT *tdb, int i)
542 tdb_off rec_ptr, top;
544 top = TDB_HASH_TOP(i);
546 if (tdb_lock(tdb, i, F_WRLCK) != 0)
549 if (ofs_read(tdb, top, &rec_ptr) == -1)
550 return tdb_unlock(tdb, i, F_WRLCK);
553 printf("hash=%d\n", i);
556 rec_ptr = tdb_dump_record(tdb, rec_ptr);
559 return tdb_unlock(tdb, i, F_WRLCK);
562 void tdb_dump_all(TDB_CONTEXT *tdb)
565 for (i=0;i<tdb->header.hash_size;i++) {
566 tdb_dump_chain(tdb, i);
568 printf("freelist:\n");
569 tdb_dump_chain(tdb, -1);
572 int tdb_printfreelist(TDB_CONTEXT *tdb)
576 tdb_off offset, rec_ptr;
577 struct list_struct rec;
579 if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
582 offset = FREELIST_TOP;
584 /* read in the freelist top */
585 if (ofs_read(tdb, offset, &rec_ptr) == -1) {
586 tdb_unlock(tdb, -1, F_WRLCK);
590 printf("freelist top=[0x%08x]\n", rec_ptr );
592 if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
593 tdb_unlock(tdb, -1, F_WRLCK);
597 if (rec.magic != TDB_FREE_MAGIC) {
598 printf("bad magic 0x%08x in free list\n", rec.magic);
599 tdb_unlock(tdb, -1, F_WRLCK);
603 printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)]\n", rec.next, rec.rec_len, rec.rec_len );
604 total_free += rec.rec_len;
606 /* move to the next record */
609 printf("total rec_len = [0x%08x (%d)]\n", (int)total_free,
612 return tdb_unlock(tdb, -1, F_WRLCK);
615 /* Remove an element from the freelist. Must have alloc lock. */
616 static int remove_from_freelist(TDB_CONTEXT *tdb, tdb_off off, tdb_off next)
620 /* read in the freelist top */
621 last_ptr = FREELIST_TOP;
622 while (ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
624 /* We've found it! */
625 return ofs_write(tdb, last_ptr, &next);
627 /* Follow chain (next offset is at start of record) */
630 TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off));
631 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
634 /* Add an element into the freelist. Merge adjacent records if
636 static int tdb_free(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
640 /* Allocation and tailer lock */
641 if (tdb_lock(tdb, -1, F_WRLCK) != 0)
644 /* set an initial tailer, so if we fail we don't leave a bogus record */
645 if (update_tailer(tdb, offset, rec) != 0) {
646 TDB_LOG((tdb, 0, "tdb_free: upfate_tailer failed!\n"));
650 /* Look right first (I'm an Australian, dammit) */
651 right = offset + sizeof(*rec) + rec->rec_len;
652 if (right + sizeof(*rec) <= tdb->map_size) {
653 struct list_struct r;
655 if (tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
656 TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", right));
660 /* If it's free, expand to include it. */
661 if (r.magic == TDB_FREE_MAGIC) {
662 if (remove_from_freelist(tdb, right, r.next) == -1) {
663 TDB_LOG((tdb, 0, "tdb_free: right free failed at %u\n", right));
666 rec->rec_len += sizeof(r) + r.rec_len;
672 left = offset - sizeof(tdb_off);
673 if (left > TDB_HASH_TOP(tdb->header.hash_size-1)) {
674 struct list_struct l;
677 /* Read in tailer and jump back to header */
678 if (ofs_read(tdb, left, &leftsize) == -1) {
679 TDB_LOG((tdb, 0, "tdb_free: left offset read failed at %u\n", left));
682 left = offset - leftsize;
684 /* Now read in record */
685 if (tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
686 TDB_LOG((tdb, 0, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
690 /* If it's free, expand to include it. */
691 if (l.magic == TDB_FREE_MAGIC) {
692 if (remove_from_freelist(tdb, left, l.next) == -1) {
693 TDB_LOG((tdb, 0, "tdb_free: left free failed at %u\n", left));
697 rec->rec_len += leftsize;
703 if (update_tailer(tdb, offset, rec) == -1) {
704 TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", offset));
708 /* Now, prepend to free list */
709 rec->magic = TDB_FREE_MAGIC;
711 if (ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
712 rec_write(tdb, offset, rec) == -1 ||
713 ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
714 TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", offset));
718 /* And we're done. */
719 tdb_unlock(tdb, -1, F_WRLCK);
723 tdb_unlock(tdb, -1, F_WRLCK);
728 /* expand a file. we prefer to use ftruncate, as that is what posix
729 says to use for mmap expansion */
730 static int expand_file(TDB_CONTEXT *tdb, tdb_off size, tdb_off addition)
733 #if HAVE_FTRUNCATE_EXTEND
734 if (ftruncate(tdb->fd, size+addition) != 0) {
735 TDB_LOG((tdb, 0, "expand_file ftruncate to %d failed (%s)\n",
736 size+addition, strerror(errno)));
743 if (pwrite(tdb->fd, &b, 1, (size+addition) - 1) != 1) {
745 if (lseek(tdb->fd, (size+addition) - 1, SEEK_SET) != (size+addition) - 1 ||
746 write(tdb->fd, &b, 1) != 1) {
748 TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n",
749 size+addition, strerror(errno)));
754 /* now fill the file with something. This ensures that the file isn't sparse, which would be
755 very bad if we ran out of disk. This must be done with write, not via mmap */
756 memset(buf, 0x42, sizeof(buf));
758 int n = addition>sizeof(buf)?sizeof(buf):addition;
760 int ret = pwrite(tdb->fd, buf, n, size);
763 if (lseek(tdb->fd, size, SEEK_SET) != size)
765 ret = write(tdb->fd, buf, n);
768 TDB_LOG((tdb, 0, "expand_file write of %d failed (%s)\n",
769 n, strerror(errno)));
779 /* expand the database at least size bytes by expanding the underlying
780 file and doing the mmap again if necessary */
781 static int tdb_expand(TDB_CONTEXT *tdb, tdb_off size)
783 struct list_struct rec;
786 if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
787 TDB_LOG((tdb, 0, "lock failed in tdb_expand\n"));
791 /* must know about any previous expansions by another process */
792 tdb_oob(tdb, tdb->map_size + 1, 1);
794 /* always make room for at least 10 more records, and round
795 the database up to a multiple of TDB_PAGE_SIZE */
796 size = TDB_ALIGN(tdb->map_size + size*10, TDB_PAGE_SIZE) - tdb->map_size;
798 if (!(tdb->flags & TDB_INTERNAL))
802 * We must ensure the file is unmapped before doing this
803 * to ensure consistency with systems like OpenBSD where
804 * writes and mmaps are not consistent.
807 /* expand the file itself */
808 if (!(tdb->flags & TDB_INTERNAL)) {
809 if (expand_file(tdb, tdb->map_size, size) != 0)
813 tdb->map_size += size;
815 if (tdb->flags & TDB_INTERNAL)
816 tdb->map_ptr = realloc(tdb->map_ptr, tdb->map_size);
819 * We must ensure the file is remapped before adding the space
820 * to ensure consistency with systems like OpenBSD where
821 * writes and mmaps are not consistent.
824 /* We're ok if the mmap fails as we'll fallback to read/write */
828 /* form a new freelist record */
829 memset(&rec,'\0',sizeof(rec));
830 rec.rec_len = size - sizeof(rec);
832 /* link it into the free list */
833 offset = tdb->map_size - size;
834 if (tdb_free(tdb, offset, &rec) == -1)
837 tdb_unlock(tdb, -1, F_WRLCK);
840 tdb_unlock(tdb, -1, F_WRLCK);
844 /* allocate some space from the free list. The offset returned points
845 to a unconnected list_struct within the database with room for at
846 least length bytes of total data
848 0 is returned if the space could not be allocated
850 static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length,
851 struct list_struct *rec)
853 tdb_off rec_ptr, last_ptr, newrec_ptr;
854 struct list_struct newrec;
856 if (tdb_lock(tdb, -1, F_WRLCK) == -1)
859 /* Extra bytes required for tailer */
860 length += sizeof(tdb_off);
863 last_ptr = FREELIST_TOP;
865 /* read in the freelist top */
866 if (ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
869 /* keep looking until we find a freelist record big enough */
871 if (rec_free_read(tdb, rec_ptr, rec) == -1)
874 if (rec->rec_len >= length) {
875 /* found it - now possibly split it up */
876 if (rec->rec_len > length + MIN_REC_SIZE) {
877 /* Length of left piece */
878 length = TDB_ALIGN(length, TDB_ALIGNMENT);
880 /* Right piece to go on free list */
881 newrec.rec_len = rec->rec_len
882 - (sizeof(*rec) + length);
883 newrec_ptr = rec_ptr + sizeof(*rec) + length;
885 /* And left record is shortened */
886 rec->rec_len = length;
890 /* Remove allocated record from the free list */
891 if (ofs_write(tdb, last_ptr, &rec->next) == -1)
894 /* Update header: do this before we drop alloc
895 lock, otherwise tdb_free() might try to
896 merge with us, thinking we're free.
897 (Thanks Jeremy Allison). */
898 rec->magic = TDB_MAGIC;
899 if (rec_write(tdb, rec_ptr, rec) == -1)
902 /* Did we create new block? */
904 /* Update allocated record tailer (we
906 if (update_tailer(tdb, rec_ptr, rec) == -1)
909 /* Free new record */
910 if (tdb_free(tdb, newrec_ptr, &newrec) == -1)
914 /* all done - return the new record offset */
915 tdb_unlock(tdb, -1, F_WRLCK);
918 /* move to the next record */
922 /* we didn't find enough space. See if we can expand the
923 database and if we can then try again */
924 if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
927 tdb_unlock(tdb, -1, F_WRLCK);
931 /* initialise a new database with a specified hash size */
932 static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size)
934 struct tdb_header *newdb;
937 /* We make it up in memory, then write it out if not internal */
938 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off);
939 if (!(newdb = calloc(size, 1)))
940 return TDB_ERRCODE(TDB_ERR_OOM, -1);
942 /* Fill in the header */
943 newdb->version = TDB_VERSION;
944 newdb->hash_size = hash_size;
946 newdb->rwlocks = size;
948 if (tdb->flags & TDB_INTERNAL) {
949 tdb->map_size = size;
950 tdb->map_ptr = (char *)newdb;
951 memcpy(&tdb->header, newdb, sizeof(tdb->header));
952 /* Convert the `ondisk' version if asked. */
956 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
959 if (ftruncate(tdb->fd, 0) == -1)
962 /* This creates an endian-converted header, as if read from disk */
964 memcpy(&tdb->header, newdb, sizeof(tdb->header));
965 /* Don't endian-convert the magic food! */
966 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
967 if (write(tdb->fd, newdb, size) != size)
970 ret = tdb_create_rwlocks(tdb->fd, hash_size);
977 /* Returns 0 on fail. On success, return offset of record, and fills
979 static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash,
980 struct list_struct *r)
984 /* read in the hash top */
985 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
988 /* keep looking until we find the right record */
990 if (rec_read(tdb, rec_ptr, r) == -1)
993 if (!TDB_DEAD(r) && hash==r->full_hash && key.dsize==r->key_len) {
995 /* a very likely hit - read the key */
996 k = tdb_alloc_read(tdb, rec_ptr + sizeof(*r),
1001 if (memcmp(key.dptr, k, key.dsize) == 0) {
1009 return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
1012 /* If they do lockkeys, check that this hash is one they locked */
1013 static int tdb_keylocked(TDB_CONTEXT *tdb, u32 hash)
1016 if (!tdb->lockedkeys)
1018 for (i = 0; i < tdb->lockedkeys[0]; i++)
1019 if (tdb->lockedkeys[i+1] == hash)
1021 return TDB_ERRCODE(TDB_ERR_NOLOCK, 0);
1024 /* As tdb_find, but if you succeed, keep the lock */
1025 static tdb_off tdb_find_lock(TDB_CONTEXT *tdb, TDB_DATA key, int locktype,
1026 struct list_struct *rec)
1030 hash = tdb_hash(&key);
1031 if (!tdb_keylocked(tdb, hash))
1033 if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
1035 if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
1036 tdb_unlock(tdb, BUCKET(hash), locktype);
1040 enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb)
1045 static struct tdb_errname {
1046 enum TDB_ERROR ecode; const char *estring;
1047 } emap[] = { {TDB_SUCCESS, "Success"},
1048 {TDB_ERR_CORRUPT, "Corrupt database"},
1049 {TDB_ERR_IO, "IO Error"},
1050 {TDB_ERR_LOCK, "Locking error"},
1051 {TDB_ERR_OOM, "Out of memory"},
1052 {TDB_ERR_EXISTS, "Record exists"},
1053 {TDB_ERR_NOLOCK, "Lock exists on other keys"},
1054 {TDB_ERR_NOEXIST, "Record does not exist"} };
1056 /* Error string for the last tdb error */
1057 const char *tdb_errorstr(TDB_CONTEXT *tdb)
1060 for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
1061 if (tdb->ecode == emap[i].ecode)
1062 return emap[i].estring;
1063 return "Invalid error code";
1066 /* update an entry in place - this only works if the new data size
1067 is <= the old data size and the key exists.
1068 on failure return -1.
1071 static int tdb_update(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf)
1073 struct list_struct rec;
1077 if (!(rec_ptr = tdb_find(tdb, key, tdb_hash(&key), &rec)))
1080 /* must be long enough key, data and tailer */
1081 if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off)) {
1082 tdb->ecode = TDB_SUCCESS; /* Not really an error */
1086 if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
1087 dbuf.dptr, dbuf.dsize) == -1)
1090 if (dbuf.dsize != rec.data_len) {
1092 rec.data_len = dbuf.dsize;
1093 return rec_write(tdb, rec_ptr, &rec);
1099 /* find an entry in the database given a key */
1100 /* If an entry doesn't exist tdb_err will be set to
1101 * TDB_ERR_NOEXIST. If a key has no data attached
1102 * tdb_err will not be set. Both will return a
1103 * zero pptr and zero dsize.
1106 TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
1109 struct list_struct rec;
1112 /* find which hash bucket it is in */
1113 if (!(rec_ptr = tdb_find_lock(tdb,key,F_RDLCK,&rec)))
1117 ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
1121 ret.dsize = rec.data_len;
1122 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
1126 /* check if an entry in the database exists
1128 note that 1 is returned if the key is found and 0 is returned if not found
1129 this doesn't match the conventions in the rest of this module, but is
1130 compatible with gdbm
1132 int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key)
1134 struct list_struct rec;
1136 if (tdb_find_lock(tdb, key, F_RDLCK, &rec) == 0)
1138 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
1142 /* record lock stops delete underneath */
1143 static int lock_record(TDB_CONTEXT *tdb, tdb_off off)
1145 return off ? tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
1148 Write locks override our own fcntl readlocks, so check it here.
1149 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
1150 an error to fail to get the lock here.
1153 static int write_lock_record(TDB_CONTEXT *tdb, tdb_off off)
1155 struct tdb_traverse_lock *i;
1156 for (i = &tdb->travlocks; i; i = i->next)
1159 return tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
1163 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
1164 an error to fail to get the lock here.
1167 static int write_unlock_record(TDB_CONTEXT *tdb, tdb_off off)
1169 return tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
1171 /* fcntl locks don't stack: avoid unlocking someone else's */
1172 static int unlock_record(TDB_CONTEXT *tdb, tdb_off off)
1174 struct tdb_traverse_lock *i;
1179 for (i = &tdb->travlocks; i; i = i->next)
1182 return (count == 1 ? tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
1185 /* actually delete an entry in the database given the offset */
1186 static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec)
1188 tdb_off last_ptr, i;
1189 struct list_struct lastrec;
1191 if (tdb->read_only) return -1;
1193 if (write_lock_record(tdb, rec_ptr) == -1) {
1194 /* Someone traversing here: mark it as dead */
1195 rec->magic = TDB_DEAD_MAGIC;
1196 return rec_write(tdb, rec_ptr, rec);
1198 if (write_unlock_record(tdb, rec_ptr) != 0)
1201 /* find previous record in hash chain */
1202 if (ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
1204 for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
1205 if (rec_read(tdb, i, &lastrec) == -1)
1208 /* unlink it: next ptr is at start of record. */
1210 last_ptr = TDB_HASH_TOP(rec->full_hash);
1211 if (ofs_write(tdb, last_ptr, &rec->next) == -1)
1214 /* recover the space */
1215 if (tdb_free(tdb, rec_ptr, rec) == -1)
1220 /* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
1221 static int tdb_next_lock(TDB_CONTEXT *tdb, struct tdb_traverse_lock *tlock,
1222 struct list_struct *rec)
1224 int want_next = (tlock->off != 0);
1226 /* No traversal allows if you've called tdb_lockkeys() */
1227 if (tdb->lockedkeys)
1228 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1230 /* Lock each chain from the start one. */
1231 for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
1232 if (tdb_lock(tdb, tlock->hash, F_WRLCK) == -1)
1235 /* No previous record? Start at top of chain. */
1237 if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
1241 /* Otherwise unlock the previous record. */
1242 if (unlock_record(tdb, tlock->off) != 0)
1247 /* We have offset of old record: grab next */
1248 if (rec_read(tdb, tlock->off, rec) == -1)
1250 tlock->off = rec->next;
1253 /* Iterate through chain */
1254 while( tlock->off) {
1256 if (rec_read(tdb, tlock->off, rec) == -1)
1258 if (!TDB_DEAD(rec)) {
1259 /* Woohoo: we found one! */
1260 if (lock_record(tdb, tlock->off) != 0)
1264 /* Try to clean dead ones from old traverses */
1265 current = tlock->off;
1266 tlock->off = rec->next;
1267 if (!tdb->read_only &&
1268 do_delete(tdb, current, rec) != 0)
1271 tdb_unlock(tdb, tlock->hash, F_WRLCK);
1274 /* We finished iteration without finding anything */
1275 return TDB_ERRCODE(TDB_SUCCESS, 0);
1279 if (tdb_unlock(tdb, tlock->hash, F_WRLCK) != 0)
1280 TDB_LOG((tdb, 0, "tdb_next_lock: On error unlock failed!\n"));
1284 /* traverse the entire database - calling fn(tdb, key, data) on each element.
1285 return -1 on error or the record count traversed
1286 if fn is NULL then it is not called
1287 a non-zero return value from fn() indicates that the traversal should stop
1289 int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *state)
1292 struct list_struct rec;
1293 struct tdb_traverse_lock tl = { NULL, 0, 0 };
1296 /* This was in the initializaton, above, but the IRIX compiler
1297 * did not like it. crh
1299 tl.next = tdb->travlocks.next;
1301 /* fcntl locks don't stack: beware traverse inside traverse */
1302 tdb->travlocks.next = &tl;
1304 /* tdb_next_lock places locks on the record returned, and its chain */
1305 while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) {
1307 /* now read the full record */
1308 key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec),
1309 rec.key_len + rec.data_len);
1312 if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0)
1314 if (unlock_record(tdb, tl.off) != 0)
1315 TDB_LOG((tdb, 0, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
1318 key.dsize = rec.key_len;
1319 dbuf.dptr = key.dptr + rec.key_len;
1320 dbuf.dsize = rec.data_len;
1322 /* Drop chain lock, call out */
1323 if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0) {
1327 if (fn && fn(tdb, key, dbuf, state)) {
1328 /* They want us to terminate traversal */
1330 if (unlock_record(tdb, tl.off) != 0) {
1331 TDB_LOG((tdb, 0, "tdb_traverse: unlock_record failed!\n"));;
1334 tdb->travlocks.next = tl.next;
1335 SAFE_FREE(key.dptr);
1338 SAFE_FREE(key.dptr);
1341 tdb->travlocks.next = tl.next;
1348 /* find the first entry in the database and return its key */
1349 TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb)
1352 struct list_struct rec;
1354 /* release any old lock */
1355 if (unlock_record(tdb, tdb->travlocks.off) != 0)
1357 tdb->travlocks.off = tdb->travlocks.hash = 0;
1359 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
1361 /* now read the key */
1362 key.dsize = rec.key_len;
1363 key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
1364 if (tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK) != 0)
1365 TDB_LOG((tdb, 0, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
1369 /* find the next entry in the database, returning its key */
1370 TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey)
1373 TDB_DATA key = tdb_null;
1374 struct list_struct rec;
1377 /* Is locked key the old key? If so, traverse will be reliable. */
1378 if (tdb->travlocks.off) {
1379 if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
1381 if (rec_read(tdb, tdb->travlocks.off, &rec) == -1
1382 || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
1384 || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
1385 /* No, it wasn't: unlock it and start from scratch */
1386 if (unlock_record(tdb, tdb->travlocks.off) != 0)
1388 if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
1390 tdb->travlocks.off = 0;
1396 if (!tdb->travlocks.off) {
1397 /* No previous element: do normal find, and lock record */
1398 tdb->travlocks.off = tdb_find_lock(tdb, oldkey, F_WRLCK, &rec);
1399 if (!tdb->travlocks.off)
1401 tdb->travlocks.hash = BUCKET(rec.full_hash);
1402 if (lock_record(tdb, tdb->travlocks.off) != 0) {
1403 TDB_LOG((tdb, 0, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
1407 oldhash = tdb->travlocks.hash;
1409 /* Grab next record: locks chain and returned record,
1410 unlocks old record */
1411 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
1412 key.dsize = rec.key_len;
1413 key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
1415 /* Unlock the chain of this new record */
1416 if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
1417 TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
1419 /* Unlock the chain of old record */
1420 if (tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK) != 0)
1421 TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
1425 /* delete an entry in the database given a key */
1426 int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
1429 struct list_struct rec;
1432 if (!(rec_ptr = tdb_find_lock(tdb, key, F_WRLCK, &rec)))
1434 ret = do_delete(tdb, rec_ptr, &rec);
1435 if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
1436 TDB_LOG((tdb, 0, "tdb_delete: WARNING tdb_unlock failed!\n"));
1440 /* store an element in the database, replacing any existing element
1443 return 0 on success, -1 on failure
1445 int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
1447 struct list_struct rec;
1453 /* find which hash bucket it is in */
1454 hash = tdb_hash(&key);
1455 if (!tdb_keylocked(tdb, hash))
1457 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
1460 /* check for it existing, on insert. */
1461 if (flag == TDB_INSERT) {
1462 if (tdb_exists(tdb, key)) {
1463 tdb->ecode = TDB_ERR_EXISTS;
1467 /* first try in-place update, on modify or replace. */
1468 if (tdb_update(tdb, key, dbuf) == 0)
1470 if (flag == TDB_MODIFY && tdb->ecode == TDB_ERR_NOEXIST)
1473 /* reset the error code potentially set by the tdb_update() */
1474 tdb->ecode = TDB_SUCCESS;
1476 /* delete any existing record - if it doesn't exist we don't
1477 care. Doing this first reduces fragmentation, and avoids
1478 coalescing with `allocated' block before it's updated. */
1479 if (flag != TDB_INSERT)
1480 tdb_delete(tdb, key);
1482 /* Copy key+value *before* allocating free space in case malloc
1483 fails and we are left with a dead spot in the tdb. */
1485 if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) {
1486 tdb->ecode = TDB_ERR_OOM;
1490 memcpy(p, key.dptr, key.dsize);
1492 memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
1494 /* now we're into insert / modify / replace of a record which
1495 * we know could not be optimised by an in-place store (for
1496 * various reasons). */
1497 if (!(rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec)))
1500 /* Read hash top into next ptr */
1501 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
1504 rec.key_len = key.dsize;
1505 rec.data_len = dbuf.dsize;
1506 rec.full_hash = hash;
1507 rec.magic = TDB_MAGIC;
1509 /* write out and point the top of the hash chain at it */
1510 if (rec_write(tdb, rec_ptr, &rec) == -1
1511 || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
1512 || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
1513 /* Need to tdb_unallocate() here */
1518 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
1525 /* Attempt to append data to an entry in place - this only works if the new data size
1526 is <= the old data size and the key exists.
1527 on failure return -1. Record must be locked before calling.
1529 static int tdb_append_inplace(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf)
1531 struct list_struct rec;
1535 if (!(rec_ptr = tdb_find(tdb, key, tdb_hash(&key), &rec)))
1538 /* Append of 0 is always ok. */
1539 if (new_dbuf.dsize == 0)
1542 /* must be long enough for key, old data + new data and tailer */
1543 if (rec.rec_len < key.dsize + rec.data_len + new_dbuf.dsize + sizeof(tdb_off)) {
1545 tdb->ecode = TDB_SUCCESS; /* Not really an error */
1549 if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len + rec.data_len,
1550 new_dbuf.dptr, new_dbuf.dsize) == -1)
1554 rec.data_len += new_dbuf.dsize;
1555 return rec_write(tdb, rec_ptr, &rec);
1558 /* Append to an entry. Create if not exist. */
1560 int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf)
1562 struct list_struct rec;
1567 size_t new_data_size = 0;
1569 /* find which hash bucket it is in */
1570 hash = tdb_hash(&key);
1571 if (!tdb_keylocked(tdb, hash))
1573 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
1576 /* first try in-place. */
1577 if (tdb_append_inplace(tdb, key, new_dbuf) == 0)
1580 /* reset the error code potentially set by the tdb_append_inplace() */
1581 tdb->ecode = TDB_SUCCESS;
1584 if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
1585 if (tdb->ecode != TDB_ERR_NOEXIST)
1588 /* Not found - create. */
1590 ret = tdb_store(tdb, key, new_dbuf, TDB_INSERT);
1594 new_data_size = rec.data_len + new_dbuf.dsize;
1596 /* Copy key+old_value+value *before* allocating free space in case malloc
1597 fails and we are left with a dead spot in the tdb. */
1599 if (!(p = (char *)malloc(key.dsize + new_data_size))) {
1600 tdb->ecode = TDB_ERR_OOM;
1604 /* Copy the key in place. */
1605 memcpy(p, key.dptr, key.dsize);
1607 /* Now read the old data into place. */
1609 tdb_read(tdb, rec_ptr + sizeof(rec) + rec.key_len, p + key.dsize, rec.data_len, 0) == -1)
1612 /* Finally append the new data. */
1614 memcpy(p+key.dsize+rec.data_len, new_dbuf.dptr, new_dbuf.dsize);
1616 /* delete any existing record - if it doesn't exist we don't
1617 care. Doing this first reduces fragmentation, and avoids
1618 coalescing with `allocated' block before it's updated. */
1620 tdb_delete(tdb, key);
1622 if (!(rec_ptr = tdb_allocate(tdb, key.dsize + new_data_size, &rec)))
1625 /* Read hash top into next ptr */
1626 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
1629 rec.key_len = key.dsize;
1630 rec.data_len = new_data_size;
1631 rec.full_hash = hash;
1632 rec.magic = TDB_MAGIC;
1634 /* write out and point the top of the hash chain at it */
1635 if (rec_write(tdb, rec_ptr, &rec) == -1
1636 || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+new_data_size)==-1
1637 || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
1638 /* Need to tdb_unallocate() here */
1644 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
1652 static int tdb_already_open(dev_t device,
1657 for (i = tdbs; i; i = i->next) {
1658 if (i->device == device && i->inode == ino) {
1666 /* open the database, creating it if necessary
1668 The open_flags and mode are passed straight to the open call on the
1669 database file. A flags value of O_WRONLY is invalid. The hash size
1670 is advisory, use zero for a default value.
1672 Return is NULL on error, in which case errno is also set. Don't
1673 try to call tdb_error or tdb_errname, just do strerror(errno).
1675 @param name may be NULL for internal databases. */
1676 TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
1677 int open_flags, mode_t mode)
1679 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL);
1683 TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
1684 int open_flags, mode_t mode,
1685 tdb_log_func log_fn)
1689 int rev = 0, locked;
1693 if (!(tdb = calloc(1, sizeof *tdb))) {
1694 /* Can't log this */
1700 tdb->map_ptr = NULL;
1701 tdb->lockedkeys = NULL;
1702 tdb->flags = tdb_flags;
1703 tdb->open_flags = open_flags;
1704 tdb->log_fn = log_fn;
1706 if ((open_flags & O_ACCMODE) == O_WRONLY) {
1707 TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n",
1714 hash_size = DEFAULT_HASH_SIZE;
1715 if ((open_flags & O_ACCMODE) == O_RDONLY) {
1717 /* read only databases don't do locking or clear if first */
1718 tdb->flags |= TDB_NOLOCK;
1719 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
1722 /* internal databases don't mmap or lock, and start off cleared */
1723 if (tdb->flags & TDB_INTERNAL) {
1724 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
1725 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
1726 if (tdb_new_database(tdb, hash_size) != 0) {
1727 TDB_LOG((tdb, 0, "tdb_open_ex: tdb_new_database failed!"));
1733 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
1734 TDB_LOG((tdb, 5, "tdb_open_ex: could not open file %s: %s\n",
1735 name, strerror(errno)));
1736 goto fail; /* errno set by open(2) */
1739 /* ensure there is only one process initialising at once */
1740 if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
1741 TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: %s\n",
1742 name, strerror(errno)));
1743 goto fail; /* errno set by tdb_brlock */
1746 /* we need to zero database if we are the only one with it open */
1747 if ((locked = (tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))
1748 && (tdb_flags & TDB_CLEAR_IF_FIRST)) {
1749 open_flags |= O_CREAT;
1750 if (ftruncate(tdb->fd, 0) == -1) {
1751 TDB_LOG((tdb, 0, "tdb_open_ex: "
1752 "failed to truncate %s: %s\n",
1753 name, strerror(errno)));
1754 goto fail; /* errno set by ftruncate */
1758 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
1759 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
1760 || (tdb->header.version != TDB_VERSION
1761 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
1762 /* its not a valid database - possibly initialise it */
1763 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
1764 errno = EIO; /* ie bad format or something */
1767 rev = (tdb->flags & TDB_CONVERT);
1769 vp = (unsigned char *)&tdb->header.version;
1770 vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
1771 (((u32)vp[2]) << 8) | (u32)vp[3];
1772 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
1774 tdb->flags &= ~TDB_CONVERT;
1776 tdb->flags |= TDB_CONVERT;
1777 convert(&tdb->header, sizeof(tdb->header));
1779 if (fstat(tdb->fd, &st) == -1)
1782 /* Is it already in the open list? If so, fail. */
1783 if (tdb_already_open(st.st_dev, st.st_ino)) {
1784 TDB_LOG((tdb, 2, "tdb_open_ex: "
1785 "%s (%d,%d) is already open in this process\n",
1786 name, st.st_dev, st.st_ino));
1791 if (!(tdb->name = (char *)strdup(name))) {
1796 tdb->map_size = st.st_size;
1797 tdb->device = st.st_dev;
1798 tdb->inode = st.st_ino;
1799 tdb->locked = calloc(tdb->header.hash_size+1, sizeof(tdb->locked[0]));
1801 TDB_LOG((tdb, 2, "tdb_open_ex: "
1802 "failed to allocate lock structure for %s\n",
1809 if (!tdb->read_only)
1810 if (tdb_clear_spinlocks(tdb) != 0) {
1811 TDB_LOG((tdb, 0, "tdb_open_ex: "
1812 "failed to clear spinlock\n"));
1815 if (tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
1816 TDB_LOG((tdb, 0, "tdb_open_ex: "
1817 "failed to take ACTIVE_LOCK on %s: %s\n",
1818 name, strerror(errno)));
1822 /* leave this lock in place to indicate it's in use */
1823 if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
1827 /* Internal (memory-only) databases skip all the code above to
1828 * do with disk files, and resume here by releasing their
1829 * global lock and hooking into the active list. */
1830 if (tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
1837 { int save_errno = errno;
1843 if (tdb->flags & TDB_INTERNAL)
1844 SAFE_FREE(tdb->map_ptr);
1848 SAFE_FREE(tdb->name);
1850 if (close(tdb->fd) != 0)
1851 TDB_LOG((tdb, 5, "tdb_open_ex: failed to close tdb->fd on error!\n"));
1852 SAFE_FREE(tdb->locked);
1862 * @returns -1 for error; 0 for success.
1864 int tdb_close(TDB_CONTEXT *tdb)
1870 if (tdb->flags & TDB_INTERNAL)
1871 SAFE_FREE(tdb->map_ptr);
1875 SAFE_FREE(tdb->name);
1877 ret = close(tdb->fd);
1878 SAFE_FREE(tdb->locked);
1879 SAFE_FREE(tdb->lockedkeys);
1881 /* Remove from contexts list */
1882 for (i = &tdbs; *i; i = &(*i)->next) {
1889 memset(tdb, 0, sizeof(*tdb));
1895 /* lock/unlock entire database */
1896 int tdb_lockall(TDB_CONTEXT *tdb)
1900 /* There are no locks on read-only dbs */
1902 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
1903 if (tdb->lockedkeys)
1904 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1905 for (i = 0; i < tdb->header.hash_size; i++)
1906 if (tdb_lock(tdb, i, F_WRLCK))
1909 /* If error, release locks we have... */
1910 if (i < tdb->header.hash_size) {
1913 for ( j = 0; j < i; j++)
1914 tdb_unlock(tdb, j, F_WRLCK);
1915 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1920 void tdb_unlockall(TDB_CONTEXT *tdb)
1923 for (i=0; i < tdb->header.hash_size; i++)
1924 tdb_unlock(tdb, i, F_WRLCK);
1927 int tdb_lockkeys(TDB_CONTEXT *tdb, u32 number, TDB_DATA keys[])
1931 /* Can't lock more keys if already locked */
1932 if (tdb->lockedkeys)
1933 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1934 if (!(tdb->lockedkeys = malloc(sizeof(u32) * (number+1))))
1935 return TDB_ERRCODE(TDB_ERR_OOM, -1);
1936 /* First number in array is # keys */
1937 tdb->lockedkeys[0] = number;
1939 /* Insertion sort by bucket */
1940 for (i = 0; i < number; i++) {
1941 hash = tdb_hash(&keys[i]);
1942 for (j = 0; j < i && BUCKET(tdb->lockedkeys[j+1]) < BUCKET(hash); j++);
1943 memmove(&tdb->lockedkeys[j+2], &tdb->lockedkeys[j+1], sizeof(u32) * (i-j));
1944 tdb->lockedkeys[j+1] = hash;
1946 /* Finally, lock in order */
1947 for (i = 0; i < number; i++)
1948 if (tdb_lock(tdb, i, F_WRLCK))
1951 /* If error, release locks we have... */
1953 for ( j = 0; j < i; j++)
1954 tdb_unlock(tdb, j, F_WRLCK);
1955 SAFE_FREE(tdb->lockedkeys);
1956 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
1961 /* Unlock the keys previously locked by tdb_lockkeys() */
1962 void tdb_unlockkeys(TDB_CONTEXT *tdb)
1965 if (!tdb->lockedkeys)
1967 for (i = 0; i < tdb->lockedkeys[0]; i++)
1968 tdb_unlock(tdb, tdb->lockedkeys[i+1], F_WRLCK);
1969 SAFE_FREE(tdb->lockedkeys);
1972 /* lock/unlock one hash chain. This is meant to be used to reduce
1973 contention - it cannot guarantee how many records will be locked */
1974 int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key)
1976 return tdb_lock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
1979 int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key)
1981 return tdb_unlock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
1984 int tdb_chainlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
1986 return tdb_lock(tdb, BUCKET(tdb_hash(&key)), F_RDLCK);
1989 int tdb_chainunlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
1991 return tdb_unlock(tdb, BUCKET(tdb_hash(&key)), F_RDLCK);
1995 /* register a loging function */
1996 void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , const char *, ...))
2002 /* reopen a tdb - this is used after a fork to ensure that we have an independent
2003 seek pointer from our parent and to re-establish locks */
2004 int tdb_reopen(TDB_CONTEXT *tdb)
2008 if (tdb_munmap(tdb) != 0) {
2009 TDB_LOG((tdb, 0, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
2012 if (close(tdb->fd) != 0)
2013 TDB_LOG((tdb, 0, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
2014 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
2015 if (tdb->fd == -1) {
2016 TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", strerror(errno)));
2019 if (fstat(tdb->fd, &st) != 0) {
2020 TDB_LOG((tdb, 0, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
2023 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
2024 TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n"));
2028 if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1) {
2029 TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n"));
2040 /* reopen all tdb's */
2041 int tdb_reopen_all(void)
2045 for (tdb=tdbs; tdb; tdb = tdb->next) {
2046 if (tdb_reopen(tdb) != 0) return -1;