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.
65 #define TDB_MAGIC_FOOD "TDB file\n"
66 #define TDB_VERSION (0x26011967 + 6)
67 #define TDB_MAGIC (0x26011999U)
68 #define TDB_FREE_MAGIC (~TDB_MAGIC)
69 #define TDB_DEAD_MAGIC (0xFEE1DEAD)
70 #define TDB_ALIGNMENT 4
71 #define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
72 #define DEFAULT_HASH_SIZE 131
73 #define TDB_PAGE_SIZE 0x2000
74 #define FREELIST_TOP (sizeof(struct tdb_header))
75 #define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
76 #define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
77 #define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
78 #define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
79 #define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off))
80 #define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + TDB_SPINLOCK_SIZE(hash_size))
83 /* NB assumes there is a local variable called "tdb" that is the
84 * current context, also takes doubly-parenthesized print-style
86 #define TDB_LOG(x) (tdb->log_fn?((tdb->log_fn x),0) : 0)
97 #define MAP_FAILED ((void *)-1)
100 /* free memory if the pointer is valid and zero the pointer */
102 #define SAFE_FREE(x) do { if ((x) != NULL) {free((x)); (x)=NULL;} } while(0)
105 #define BUCKET(hash) ((hash) % tdb->header.hash_size)
108 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
109 static TDB_CONTEXT
*tdbs
= NULL
;
111 static int tdb_munmap(TDB_CONTEXT
*tdb
)
113 if (tdb
->flags
& TDB_INTERNAL
)
118 int ret
= munmap(tdb
->map_ptr
, tdb
->map_size
);
127 static void tdb_mmap(TDB_CONTEXT
*tdb
)
129 if (tdb
->flags
& TDB_INTERNAL
)
133 if (!(tdb
->flags
& TDB_NOMMAP
)) {
134 tdb
->map_ptr
= mmap(NULL
, tdb
->map_size
,
135 PROT_READ
|(tdb
->read_only
? 0:PROT_WRITE
),
136 MAP_SHARED
|MAP_FILE
, tdb
->fd
, 0);
139 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
142 if (tdb
->map_ptr
== MAP_FAILED
) {
144 TDB_LOG((tdb
, 2, "tdb_mmap failed for size %d (%s)\n",
145 tdb
->map_size
, strerror(errno
)));
155 /* Endian conversion: we only ever deal with 4 byte quantities */
156 static void *convert(void *buf
, u32 size
)
159 for (i
= 0; i
< size
/ 4; i
++)
160 p
[i
] = TDB_BYTEREV(p
[i
]);
163 #define DOCONV() (tdb->flags & TDB_CONVERT)
164 #define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x)
166 /* the body of the database is made of one list_struct for the free space
167 plus a separate data list for each hash value */
169 tdb_off next
; /* offset of the next record in the list */
170 tdb_len rec_len
; /* total byte length of record */
171 tdb_len key_len
; /* byte length of key */
172 tdb_len data_len
; /* byte length of data */
173 u32 full_hash
; /* the full 32 bit hash of the key */
174 u32 magic
; /* try to catch errors */
175 /* the following union is implied:
177 char record[rec_len];
182 u32 totalsize; (tailer)
187 /***************************************************************
188 Allow a caller to set a "alarm" flag that tdb can check to abort
189 a blocking lock on SIGALRM.
190 ***************************************************************/
192 static sig_atomic_t *palarm_fired
;
194 void tdb_set_lock_alarm(sig_atomic_t *palarm
)
196 palarm_fired
= palarm
;
199 /* a byte range locking function - return 0 on success
200 this functions locks/unlocks 1 byte at the specified offset.
202 On error, errno is also set so that errors are passed back properly
203 through tdb_open(). */
204 static int tdb_brlock(TDB_CONTEXT
*tdb
, tdb_off offset
,
205 int rw_type
, int lck_type
, int probe
)
210 if (tdb
->flags
& TDB_NOLOCK
)
212 if ((rw_type
== F_WRLCK
) && (tdb
->read_only
)) {
218 fl
.l_whence
= SEEK_SET
;
224 ret
= fcntl(tdb
->fd
,lck_type
,&fl
);
225 if (ret
== -1 && errno
== EINTR
&& palarm_fired
&& *palarm_fired
)
227 } while (ret
== -1 && errno
== EINTR
);
230 if (!probe
&& lck_type
!= F_SETLK
) {
231 /* Ensure error code is set for log fun to examine. */
232 if (errno
== EINTR
&& palarm_fired
&& *palarm_fired
)
233 tdb
->ecode
= TDB_ERR_LOCK_TIMEOUT
;
235 tdb
->ecode
= TDB_ERR_LOCK
;
236 TDB_LOG((tdb
, 5,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d\n",
237 tdb
->fd
, offset
, rw_type
, lck_type
));
239 /* Was it an alarm timeout ? */
240 if (errno
== EINTR
&& palarm_fired
&& *palarm_fired
)
241 return TDB_ERRCODE(TDB_ERR_LOCK_TIMEOUT
, -1);
242 /* Otherwise - generic lock error. */
243 /* errno set by fcntl */
244 return TDB_ERRCODE(TDB_ERR_LOCK
, -1);
249 /* lock a list in the database. list -1 is the alloc list */
250 static int tdb_lock(TDB_CONTEXT
*tdb
, int list
, int ltype
)
252 if (list
< -1 || list
>= (int)tdb
->header
.hash_size
) {
253 TDB_LOG((tdb
, 0,"tdb_lock: invalid list %d for ltype=%d\n",
257 if (tdb
->flags
& TDB_NOLOCK
)
260 /* Since fcntl locks don't nest, we do a lock for the first one,
261 and simply bump the count for future ones */
262 if (tdb
->locked
[list
+1].count
== 0) {
263 if (!tdb
->read_only
&& tdb
->header
.rwlocks
) {
264 if (tdb_spinlock(tdb
, list
, ltype
)) {
265 TDB_LOG((tdb
, 0, "tdb_lock spinlock failed on list ltype=%d\n",
269 } else if (tdb_brlock(tdb
,FREELIST_TOP
+4*list
,ltype
,F_SETLKW
, 0)) {
270 TDB_LOG((tdb
, 0,"tdb_lock failed on list %d ltype=%d (%s)\n",
271 list
, ltype
, strerror(errno
)));
274 tdb
->locked
[list
+1].ltype
= ltype
;
276 tdb
->locked
[list
+1].count
++;
280 /* unlock the database: returns void because it's too late for errors. */
281 /* changed to return int it may be interesting to know there
282 has been an error --simo */
283 static int tdb_unlock(TDB_CONTEXT
*tdb
, int list
, int ltype
)
287 if (tdb
->flags
& TDB_NOLOCK
)
291 if (list
< -1 || list
>= (int)tdb
->header
.hash_size
) {
292 TDB_LOG((tdb
, 0, "tdb_unlock: list %d invalid (%d)\n", list
, tdb
->header
.hash_size
));
296 if (tdb
->locked
[list
+1].count
==0) {
297 TDB_LOG((tdb
, 0, "tdb_unlock: count is 0\n"));
301 if (tdb
->locked
[list
+1].count
== 1) {
302 /* Down to last nested lock: unlock underneath */
303 if (!tdb
->read_only
&& tdb
->header
.rwlocks
) {
304 ret
= tdb_spinunlock(tdb
, list
, ltype
);
306 ret
= tdb_brlock(tdb
, FREELIST_TOP
+4*list
, F_UNLCK
, F_SETLKW
, 0);
311 tdb
->locked
[list
+1].count
--;
314 TDB_LOG((tdb
, 0,"tdb_unlock: An error occurred unlocking!\n"));
318 /* This is based on the hash algorithm from gdbm */
319 static u32
tdb_hash(TDB_DATA
*key
)
321 u32 value
; /* Used to compute the hash value. */
322 u32 i
; /* Used to cycle through random values. */
324 /* Set the initial value from the key size. */
325 for (value
= 0x238F13AF * key
->dsize
, i
=0; i
< key
->dsize
; i
++)
326 value
= (value
+ (key
->dptr
[i
] << (i
*5 % 24)));
328 return (1103515243 * value
+ 12345);
331 /* check for an out of bounds access - if it is out of bounds then
332 see if the database has been expanded by someone else and expand
334 note that "len" is the minimum length needed for the db
336 static int tdb_oob(TDB_CONTEXT
*tdb
, tdb_off len
, int probe
)
339 if (len
<= tdb
->map_size
)
341 if (tdb
->flags
& TDB_INTERNAL
) {
343 /* Ensure ecode is set for log fn. */
344 tdb
->ecode
= TDB_ERR_IO
;
345 TDB_LOG((tdb
, 0,"tdb_oob len %d beyond internal malloc size %d\n",
346 (int)len
, (int)tdb
->map_size
));
348 return TDB_ERRCODE(TDB_ERR_IO
, -1);
351 if (fstat(tdb
->fd
, &st
) == -1)
352 return TDB_ERRCODE(TDB_ERR_IO
, -1);
354 if (st
.st_size
< (size_t)len
) {
356 /* Ensure ecode is set for log fn. */
357 tdb
->ecode
= TDB_ERR_IO
;
358 TDB_LOG((tdb
, 0,"tdb_oob len %d beyond eof at %d\n",
359 (int)len
, (int)st
.st_size
));
361 return TDB_ERRCODE(TDB_ERR_IO
, -1);
364 /* Unmap, update size, remap */
365 if (tdb_munmap(tdb
) == -1)
366 return TDB_ERRCODE(TDB_ERR_IO
, -1);
367 tdb
->map_size
= st
.st_size
;
372 /* write a lump of data at a specified offset */
373 static int tdb_write(TDB_CONTEXT
*tdb
, tdb_off off
, void *buf
, tdb_len len
)
375 if (tdb_oob(tdb
, off
+ len
, 0) != 0)
379 memcpy(off
+ (char *)tdb
->map_ptr
, buf
, len
);
381 else if (pwrite(tdb
->fd
, buf
, len
, off
) != (ssize_t
)len
) {
383 else if (lseek(tdb
->fd
, off
, SEEK_SET
) != off
384 || write(tdb
->fd
, buf
, len
) != (ssize_t
)len
) {
386 /* Ensure ecode is set for log fn. */
387 tdb
->ecode
= TDB_ERR_IO
;
388 TDB_LOG((tdb
, 0,"tdb_write failed at %d len=%d (%s)\n",
389 off
, len
, strerror(errno
)));
390 return TDB_ERRCODE(TDB_ERR_IO
, -1);
395 /* read a lump of data at a specified offset, maybe convert */
396 static int tdb_read(TDB_CONTEXT
*tdb
,tdb_off off
,void *buf
,tdb_len len
,int cv
)
398 if (tdb_oob(tdb
, off
+ len
, 0) != 0)
402 memcpy(buf
, off
+ (char *)tdb
->map_ptr
, len
);
404 else if (pread(tdb
->fd
, buf
, len
, off
) != (ssize_t
)len
) {
406 else if (lseek(tdb
->fd
, off
, SEEK_SET
) != off
407 || read(tdb
->fd
, buf
, len
) != (ssize_t
)len
) {
409 /* Ensure ecode is set for log fn. */
410 tdb
->ecode
= TDB_ERR_IO
;
411 TDB_LOG((tdb
, 0,"tdb_read failed at %d len=%d (%s)\n",
412 off
, len
, strerror(errno
)));
413 return TDB_ERRCODE(TDB_ERR_IO
, -1);
420 /* read a lump of data, allocating the space for it */
421 static char *tdb_alloc_read(TDB_CONTEXT
*tdb
, tdb_off offset
, tdb_len len
)
425 if (!(buf
= malloc(len
))) {
426 /* Ensure ecode is set for log fn. */
427 tdb
->ecode
= TDB_ERR_OOM
;
428 TDB_LOG((tdb
, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
429 len
, strerror(errno
)));
430 return TDB_ERRCODE(TDB_ERR_OOM
, buf
);
432 if (tdb_read(tdb
, offset
, buf
, len
, 0) == -1) {
439 /* read/write a tdb_off */
440 static int ofs_read(TDB_CONTEXT
*tdb
, tdb_off offset
, tdb_off
*d
)
442 return tdb_read(tdb
, offset
, (char*)d
, sizeof(*d
), DOCONV());
444 static int ofs_write(TDB_CONTEXT
*tdb
, tdb_off offset
, tdb_off
*d
)
447 return tdb_write(tdb
, offset
, CONVERT(off
), sizeof(*d
));
450 /* read/write a record */
451 static int rec_read(TDB_CONTEXT
*tdb
, tdb_off offset
, struct list_struct
*rec
)
453 if (tdb_read(tdb
, offset
, rec
, sizeof(*rec
),DOCONV()) == -1)
455 if (TDB_BAD_MAGIC(rec
)) {
456 /* Ensure ecode is set for log fn. */
457 tdb
->ecode
= TDB_ERR_CORRUPT
;
458 TDB_LOG((tdb
, 0,"rec_read bad magic 0x%x at offset=%d\n", rec
->magic
, offset
));
459 return TDB_ERRCODE(TDB_ERR_CORRUPT
, -1);
461 return tdb_oob(tdb
, rec
->next
+sizeof(*rec
), 0);
463 static int rec_write(TDB_CONTEXT
*tdb
, tdb_off offset
, struct list_struct
*rec
)
465 struct list_struct r
= *rec
;
466 return tdb_write(tdb
, offset
, CONVERT(r
), sizeof(r
));
469 /* read a freelist record and check for simple errors */
470 static int rec_free_read(TDB_CONTEXT
*tdb
, tdb_off off
, struct list_struct
*rec
)
472 if (tdb_read(tdb
, off
, rec
, sizeof(*rec
),DOCONV()) == -1)
475 if (rec
->magic
== TDB_MAGIC
) {
476 /* this happens when a app is showdown while deleting a record - we should
477 not completely fail when this happens */
478 TDB_LOG((tdb
, 0,"rec_free_read non-free magic at offset=%d - fixing\n",
480 rec
->magic
= TDB_FREE_MAGIC
;
481 if (tdb_write(tdb
, off
, rec
, sizeof(*rec
)) == -1)
485 if (rec
->magic
!= TDB_FREE_MAGIC
) {
486 /* Ensure ecode is set for log fn. */
487 tdb
->ecode
= TDB_ERR_CORRUPT
;
488 TDB_LOG((tdb
, 0,"rec_free_read bad magic 0x%x at offset=%d\n",
490 return TDB_ERRCODE(TDB_ERR_CORRUPT
, -1);
492 if (tdb_oob(tdb
, rec
->next
+sizeof(*rec
), 0) != 0)
497 /* update a record tailer (must hold allocation lock) */
498 static int update_tailer(TDB_CONTEXT
*tdb
, tdb_off offset
,
499 const struct list_struct
*rec
)
503 /* Offset of tailer from record header */
504 totalsize
= sizeof(*rec
) + rec
->rec_len
;
505 return ofs_write(tdb
, offset
+ totalsize
- sizeof(tdb_off
),
509 static tdb_off
tdb_dump_record(TDB_CONTEXT
*tdb
, tdb_off offset
)
511 struct list_struct rec
;
512 tdb_off tailer_ofs
, tailer
;
514 if (tdb_read(tdb
, offset
, (char *)&rec
, sizeof(rec
), DOCONV()) == -1) {
515 printf("ERROR: failed to read record at %u\n", offset
);
519 printf(" rec: offset=%u next=%d rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
520 offset
, rec
.next
, rec
.rec_len
, rec
.key_len
, rec
.data_len
, rec
.full_hash
, rec
.magic
);
522 tailer_ofs
= offset
+ sizeof(rec
) + rec
.rec_len
- sizeof(tdb_off
);
523 if (ofs_read(tdb
, tailer_ofs
, &tailer
) == -1) {
524 printf("ERROR: failed to read tailer at %u\n", tailer_ofs
);
528 if (tailer
!= rec
.rec_len
+ sizeof(rec
)) {
529 printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
530 (unsigned)tailer
, (unsigned)(rec
.rec_len
+ sizeof(rec
)));
535 static int tdb_dump_chain(TDB_CONTEXT
*tdb
, int i
)
537 tdb_off rec_ptr
, top
;
539 top
= TDB_HASH_TOP(i
);
541 if (tdb_lock(tdb
, i
, F_WRLCK
) != 0)
544 if (ofs_read(tdb
, top
, &rec_ptr
) == -1)
545 return tdb_unlock(tdb
, i
, F_WRLCK
);
548 printf("hash=%d\n", i
);
551 rec_ptr
= tdb_dump_record(tdb
, rec_ptr
);
554 return tdb_unlock(tdb
, i
, F_WRLCK
);
557 void tdb_dump_all(TDB_CONTEXT
*tdb
)
560 for (i
=0;i
<tdb
->header
.hash_size
;i
++) {
561 tdb_dump_chain(tdb
, i
);
563 printf("freelist:\n");
564 tdb_dump_chain(tdb
, -1);
567 int tdb_printfreelist(TDB_CONTEXT
*tdb
)
571 tdb_off offset
, rec_ptr
;
572 struct list_struct rec
;
574 if ((ret
= tdb_lock(tdb
, -1, F_WRLCK
)) != 0)
577 offset
= FREELIST_TOP
;
579 /* read in the freelist top */
580 if (ofs_read(tdb
, offset
, &rec_ptr
) == -1) {
581 tdb_unlock(tdb
, -1, F_WRLCK
);
585 printf("freelist top=[0x%08x]\n", rec_ptr
);
587 if (tdb_read(tdb
, rec_ptr
, (char *)&rec
, sizeof(rec
), DOCONV()) == -1) {
588 tdb_unlock(tdb
, -1, F_WRLCK
);
592 if (rec
.magic
!= TDB_FREE_MAGIC
) {
593 printf("bad magic 0x%08x in free list\n", rec
.magic
);
594 tdb_unlock(tdb
, -1, F_WRLCK
);
598 printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)]\n", rec
.next
, rec
.rec_len
, rec
.rec_len
);
599 total_free
+= rec
.rec_len
;
601 /* move to the next record */
604 printf("total rec_len = [0x%08x (%d)]\n", (int)total_free
,
607 return tdb_unlock(tdb
, -1, F_WRLCK
);
610 /* Remove an element from the freelist. Must have alloc lock. */
611 static int remove_from_freelist(TDB_CONTEXT
*tdb
, tdb_off off
, tdb_off next
)
615 /* read in the freelist top */
616 last_ptr
= FREELIST_TOP
;
617 while (ofs_read(tdb
, last_ptr
, &i
) != -1 && i
!= 0) {
619 /* We've found it! */
620 return ofs_write(tdb
, last_ptr
, &next
);
622 /* Follow chain (next offset is at start of record) */
625 TDB_LOG((tdb
, 0,"remove_from_freelist: not on list at off=%d\n", off
));
626 return TDB_ERRCODE(TDB_ERR_CORRUPT
, -1);
629 /* Add an element into the freelist. Merge adjacent records if
631 static int tdb_free(TDB_CONTEXT
*tdb
, tdb_off offset
, struct list_struct
*rec
)
635 /* Allocation and tailer lock */
636 if (tdb_lock(tdb
, -1, F_WRLCK
) != 0)
639 /* set an initial tailer, so if we fail we don't leave a bogus record */
640 if (update_tailer(tdb
, offset
, rec
) != 0) {
641 TDB_LOG((tdb
, 0, "tdb_free: upfate_tailer failed!\n"));
645 /* Look right first (I'm an Australian, dammit) */
646 right
= offset
+ sizeof(*rec
) + rec
->rec_len
;
647 if (right
+ sizeof(*rec
) <= tdb
->map_size
) {
648 struct list_struct r
;
650 if (tdb_read(tdb
, right
, &r
, sizeof(r
), DOCONV()) == -1) {
651 TDB_LOG((tdb
, 0, "tdb_free: right read failed at %u\n", right
));
655 /* If it's free, expand to include it. */
656 if (r
.magic
== TDB_FREE_MAGIC
) {
657 if (remove_from_freelist(tdb
, right
, r
.next
) == -1) {
658 TDB_LOG((tdb
, 0, "tdb_free: right free failed at %u\n", right
));
661 rec
->rec_len
+= sizeof(r
) + r
.rec_len
;
667 left
= offset
- sizeof(tdb_off
);
668 if (left
> TDB_DATA_START(tdb
->header
.hash_size
)) {
669 struct list_struct l
;
672 /* Read in tailer and jump back to header */
673 if (ofs_read(tdb
, left
, &leftsize
) == -1) {
674 TDB_LOG((tdb
, 0, "tdb_free: left offset read failed at %u\n", left
));
677 left
= offset
- leftsize
;
679 /* Now read in record */
680 if (tdb_read(tdb
, left
, &l
, sizeof(l
), DOCONV()) == -1) {
681 TDB_LOG((tdb
, 0, "tdb_free: left read failed at %u (%u)\n", left
, leftsize
));
685 /* If it's free, expand to include it. */
686 if (l
.magic
== TDB_FREE_MAGIC
) {
687 if (remove_from_freelist(tdb
, left
, l
.next
) == -1) {
688 TDB_LOG((tdb
, 0, "tdb_free: left free failed at %u\n", left
));
692 rec
->rec_len
+= leftsize
;
698 if (update_tailer(tdb
, offset
, rec
) == -1) {
699 TDB_LOG((tdb
, 0, "tdb_free: update_tailer failed at %u\n", offset
));
703 /* Now, prepend to free list */
704 rec
->magic
= TDB_FREE_MAGIC
;
706 if (ofs_read(tdb
, FREELIST_TOP
, &rec
->next
) == -1 ||
707 rec_write(tdb
, offset
, rec
) == -1 ||
708 ofs_write(tdb
, FREELIST_TOP
, &offset
) == -1) {
709 TDB_LOG((tdb
, 0, "tdb_free record write failed at offset=%d\n", offset
));
713 /* And we're done. */
714 tdb_unlock(tdb
, -1, F_WRLCK
);
718 tdb_unlock(tdb
, -1, F_WRLCK
);
723 /* expand a file. we prefer to use ftruncate, as that is what posix
724 says to use for mmap expansion */
725 static int expand_file(TDB_CONTEXT
*tdb
, tdb_off size
, tdb_off addition
)
728 #if HAVE_FTRUNCATE_EXTEND
729 if (ftruncate(tdb
->fd
, size
+addition
) != 0) {
730 TDB_LOG((tdb
, 0, "expand_file ftruncate to %d failed (%s)\n",
731 size
+addition
, strerror(errno
)));
738 if (pwrite(tdb
->fd
, &b
, 1, (size
+addition
) - 1) != 1) {
740 if (lseek(tdb
->fd
, (size
+addition
) - 1, SEEK_SET
) != (size
+addition
) - 1 ||
741 write(tdb
->fd
, &b
, 1) != 1) {
743 TDB_LOG((tdb
, 0, "expand_file to %d failed (%s)\n",
744 size
+addition
, strerror(errno
)));
749 /* now fill the file with something. This ensures that the file isn't sparse, which would be
750 very bad if we ran out of disk. This must be done with write, not via mmap */
751 memset(buf
, 0x42, sizeof(buf
));
753 int n
= addition
>sizeof(buf
)?sizeof(buf
):addition
;
755 int ret
= pwrite(tdb
->fd
, buf
, n
, size
);
758 if (lseek(tdb
->fd
, size
, SEEK_SET
) != size
)
760 ret
= write(tdb
->fd
, buf
, n
);
763 TDB_LOG((tdb
, 0, "expand_file write of %d failed (%s)\n",
764 n
, strerror(errno
)));
774 /* expand the database at least size bytes by expanding the underlying
775 file and doing the mmap again if necessary */
776 static int tdb_expand(TDB_CONTEXT
*tdb
, tdb_off size
)
778 struct list_struct rec
;
781 if (tdb_lock(tdb
, -1, F_WRLCK
) == -1) {
782 TDB_LOG((tdb
, 0, "lock failed in tdb_expand\n"));
786 /* must know about any previous expansions by another process */
787 tdb_oob(tdb
, tdb
->map_size
+ 1, 1);
789 /* always make room for at least 10 more records, and round
790 the database up to a multiple of TDB_PAGE_SIZE */
791 size
= TDB_ALIGN(tdb
->map_size
+ size
*10, TDB_PAGE_SIZE
) - tdb
->map_size
;
793 if (!(tdb
->flags
& TDB_INTERNAL
))
797 * We must ensure the file is unmapped before doing this
798 * to ensure consistency with systems like OpenBSD where
799 * writes and mmaps are not consistent.
802 /* expand the file itself */
803 if (!(tdb
->flags
& TDB_INTERNAL
)) {
804 if (expand_file(tdb
, tdb
->map_size
, size
) != 0)
808 tdb
->map_size
+= size
;
810 if (tdb
->flags
& TDB_INTERNAL
)
811 tdb
->map_ptr
= realloc(tdb
->map_ptr
, tdb
->map_size
);
814 * We must ensure the file is remapped before adding the space
815 * to ensure consistency with systems like OpenBSD where
816 * writes and mmaps are not consistent.
819 /* We're ok if the mmap fails as we'll fallback to read/write */
823 /* form a new freelist record */
824 memset(&rec
,'\0',sizeof(rec
));
825 rec
.rec_len
= size
- sizeof(rec
);
827 /* link it into the free list */
828 offset
= tdb
->map_size
- size
;
829 if (tdb_free(tdb
, offset
, &rec
) == -1)
832 tdb_unlock(tdb
, -1, F_WRLCK
);
835 tdb_unlock(tdb
, -1, F_WRLCK
);
839 /* allocate some space from the free list. The offset returned points
840 to a unconnected list_struct within the database with room for at
841 least length bytes of total data
843 0 is returned if the space could not be allocated
845 static tdb_off
tdb_allocate(TDB_CONTEXT
*tdb
, tdb_len length
,
846 struct list_struct
*rec
)
848 tdb_off rec_ptr
, last_ptr
, newrec_ptr
;
849 struct list_struct newrec
;
851 if (tdb_lock(tdb
, -1, F_WRLCK
) == -1)
854 /* Extra bytes required for tailer */
855 length
+= sizeof(tdb_off
);
858 last_ptr
= FREELIST_TOP
;
860 /* read in the freelist top */
861 if (ofs_read(tdb
, FREELIST_TOP
, &rec_ptr
) == -1)
864 /* keep looking until we find a freelist record big enough */
866 if (rec_free_read(tdb
, rec_ptr
, rec
) == -1)
869 if (rec
->rec_len
>= length
) {
870 /* found it - now possibly split it up */
871 if (rec
->rec_len
> length
+ MIN_REC_SIZE
) {
872 /* Length of left piece */
873 length
= TDB_ALIGN(length
, TDB_ALIGNMENT
);
875 /* Right piece to go on free list */
876 newrec
.rec_len
= rec
->rec_len
877 - (sizeof(*rec
) + length
);
878 newrec_ptr
= rec_ptr
+ sizeof(*rec
) + length
;
880 /* And left record is shortened */
881 rec
->rec_len
= length
;
885 /* Remove allocated record from the free list */
886 if (ofs_write(tdb
, last_ptr
, &rec
->next
) == -1)
889 /* Update header: do this before we drop alloc
890 lock, otherwise tdb_free() might try to
891 merge with us, thinking we're free.
892 (Thanks Jeremy Allison). */
893 rec
->magic
= TDB_MAGIC
;
894 if (rec_write(tdb
, rec_ptr
, rec
) == -1)
897 /* Did we create new block? */
899 /* Update allocated record tailer (we
901 if (update_tailer(tdb
, rec_ptr
, rec
) == -1)
904 /* Free new record */
905 if (tdb_free(tdb
, newrec_ptr
, &newrec
) == -1)
909 /* all done - return the new record offset */
910 tdb_unlock(tdb
, -1, F_WRLCK
);
913 /* move to the next record */
917 /* we didn't find enough space. See if we can expand the
918 database and if we can then try again */
919 if (tdb_expand(tdb
, length
+ sizeof(*rec
)) == 0)
922 tdb_unlock(tdb
, -1, F_WRLCK
);
926 /* initialise a new database with a specified hash size */
927 static int tdb_new_database(TDB_CONTEXT
*tdb
, int hash_size
)
929 struct tdb_header
*newdb
;
932 /* We make it up in memory, then write it out if not internal */
933 size
= sizeof(struct tdb_header
) + (hash_size
+1)*sizeof(tdb_off
);
934 if (!(newdb
= calloc(size
, 1)))
935 return TDB_ERRCODE(TDB_ERR_OOM
, -1);
937 /* Fill in the header */
938 newdb
->version
= TDB_VERSION
;
939 newdb
->hash_size
= hash_size
;
941 newdb
->rwlocks
= size
;
943 if (tdb
->flags
& TDB_INTERNAL
) {
944 tdb
->map_size
= size
;
945 tdb
->map_ptr
= (char *)newdb
;
946 memcpy(&tdb
->header
, newdb
, sizeof(tdb
->header
));
947 /* Convert the `ondisk' version if asked. */
951 if (lseek(tdb
->fd
, 0, SEEK_SET
) == -1)
954 if (ftruncate(tdb
->fd
, 0) == -1)
957 /* This creates an endian-converted header, as if read from disk */
959 memcpy(&tdb
->header
, newdb
, sizeof(tdb
->header
));
960 /* Don't endian-convert the magic food! */
961 memcpy(newdb
->magic_food
, TDB_MAGIC_FOOD
, strlen(TDB_MAGIC_FOOD
)+1);
962 if (write(tdb
->fd
, newdb
, size
) != size
)
965 ret
= tdb_create_rwlocks(tdb
->fd
, hash_size
);
972 /* Returns 0 on fail. On success, return offset of record, and fills
974 static tdb_off
tdb_find(TDB_CONTEXT
*tdb
, TDB_DATA key
, u32 hash
,
975 struct list_struct
*r
)
979 /* read in the hash top */
980 if (ofs_read(tdb
, TDB_HASH_TOP(hash
), &rec_ptr
) == -1)
983 /* keep looking until we find the right record */
985 if (rec_read(tdb
, rec_ptr
, r
) == -1)
988 if (!TDB_DEAD(r
) && hash
==r
->full_hash
&& key
.dsize
==r
->key_len
) {
990 /* a very likely hit - read the key */
991 k
= tdb_alloc_read(tdb
, rec_ptr
+ sizeof(*r
),
996 if (memcmp(key
.dptr
, k
, key
.dsize
) == 0) {
1004 return TDB_ERRCODE(TDB_ERR_NOEXIST
, 0);
1007 /* If they do lockkeys, check that this hash is one they locked */
1008 static int tdb_keylocked(TDB_CONTEXT
*tdb
, u32 hash
)
1011 if (!tdb
->lockedkeys
)
1013 for (i
= 0; i
< tdb
->lockedkeys
[0]; i
++)
1014 if (tdb
->lockedkeys
[i
+1] == hash
)
1016 return TDB_ERRCODE(TDB_ERR_NOLOCK
, 0);
1019 /* As tdb_find, but if you succeed, keep the lock */
1020 static tdb_off
tdb_find_lock_hash(TDB_CONTEXT
*tdb
, TDB_DATA key
, u32 hash
, int locktype
,
1021 struct list_struct
*rec
)
1025 if (!tdb_keylocked(tdb
, hash
))
1027 if (tdb_lock(tdb
, BUCKET(hash
), locktype
) == -1)
1029 if (!(rec_ptr
= tdb_find(tdb
, key
, hash
, rec
)))
1030 tdb_unlock(tdb
, BUCKET(hash
), locktype
);
1034 enum TDB_ERROR
tdb_error(TDB_CONTEXT
*tdb
)
1039 static struct tdb_errname
{
1040 enum TDB_ERROR ecode
; const char *estring
;
1041 } emap
[] = { {TDB_SUCCESS
, "Success"},
1042 {TDB_ERR_CORRUPT
, "Corrupt database"},
1043 {TDB_ERR_IO
, "IO Error"},
1044 {TDB_ERR_LOCK
, "Locking error"},
1045 {TDB_ERR_OOM
, "Out of memory"},
1046 {TDB_ERR_EXISTS
, "Record exists"},
1047 {TDB_ERR_NOLOCK
, "Lock exists on other keys"},
1048 {TDB_ERR_NOEXIST
, "Record does not exist"} };
1050 /* Error string for the last tdb error */
1051 const char *tdb_errorstr(TDB_CONTEXT
*tdb
)
1054 for (i
= 0; i
< sizeof(emap
) / sizeof(struct tdb_errname
); i
++)
1055 if (tdb
->ecode
== emap
[i
].ecode
)
1056 return emap
[i
].estring
;
1057 return "Invalid error code";
1060 /* update an entry in place - this only works if the new data size
1061 is <= the old data size and the key exists.
1062 on failure return -1.
1065 static int tdb_update_hash(TDB_CONTEXT
*tdb
, TDB_DATA key
, u32 hash
, TDB_DATA dbuf
)
1067 struct list_struct rec
;
1071 if (!(rec_ptr
= tdb_find(tdb
, key
, hash
, &rec
)))
1074 /* must be long enough key, data and tailer */
1075 if (rec
.rec_len
< key
.dsize
+ dbuf
.dsize
+ sizeof(tdb_off
)) {
1076 tdb
->ecode
= TDB_SUCCESS
; /* Not really an error */
1080 if (tdb_write(tdb
, rec_ptr
+ sizeof(rec
) + rec
.key_len
,
1081 dbuf
.dptr
, dbuf
.dsize
) == -1)
1084 if (dbuf
.dsize
!= rec
.data_len
) {
1086 rec
.data_len
= dbuf
.dsize
;
1087 return rec_write(tdb
, rec_ptr
, &rec
);
1093 /* find an entry in the database given a key */
1094 /* If an entry doesn't exist tdb_err will be set to
1095 * TDB_ERR_NOEXIST. If a key has no data attached
1096 * tdb_err will not be set. Both will return a
1097 * zero pptr and zero dsize.
1100 TDB_DATA
tdb_fetch(TDB_CONTEXT
*tdb
, TDB_DATA key
)
1103 struct list_struct rec
;
1107 /* find which hash bucket it is in */
1108 hash
= tdb_hash(&key
);
1109 if (!(rec_ptr
= tdb_find_lock_hash(tdb
,key
,hash
,F_RDLCK
,&rec
)))
1113 ret
.dptr
= tdb_alloc_read(tdb
, rec_ptr
+ sizeof(rec
) + rec
.key_len
,
1117 ret
.dsize
= rec
.data_len
;
1118 tdb_unlock(tdb
, BUCKET(rec
.full_hash
), F_RDLCK
);
1122 /* check if an entry in the database exists
1124 note that 1 is returned if the key is found and 0 is returned if not found
1125 this doesn't match the conventions in the rest of this module, but is
1126 compatible with gdbm
1128 static int tdb_exists_hash(TDB_CONTEXT
*tdb
, TDB_DATA key
, u32 hash
)
1130 struct list_struct rec
;
1132 if (tdb_find_lock_hash(tdb
, key
, hash
, F_RDLCK
, &rec
) == 0)
1134 tdb_unlock(tdb
, BUCKET(rec
.full_hash
), F_RDLCK
);
1138 int tdb_exists(TDB_CONTEXT
*tdb
, TDB_DATA key
)
1140 u32 hash
= tdb_hash(&key
);
1141 return tdb_exists_hash(tdb
, key
, hash
);
1144 /* record lock stops delete underneath */
1145 static int lock_record(TDB_CONTEXT
*tdb
, tdb_off off
)
1147 return off
? tdb_brlock(tdb
, off
, F_RDLCK
, F_SETLKW
, 0) : 0;
1150 Write locks override our own fcntl readlocks, so check it here.
1151 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
1152 an error to fail to get the lock here.
1155 static int write_lock_record(TDB_CONTEXT
*tdb
, tdb_off off
)
1157 struct tdb_traverse_lock
*i
;
1158 for (i
= &tdb
->travlocks
; i
; i
= i
->next
)
1161 return tdb_brlock(tdb
, off
, F_WRLCK
, F_SETLK
, 1);
1165 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
1166 an error to fail to get the lock here.
1169 static int write_unlock_record(TDB_CONTEXT
*tdb
, tdb_off off
)
1171 return tdb_brlock(tdb
, off
, F_UNLCK
, F_SETLK
, 0);
1173 /* fcntl locks don't stack: avoid unlocking someone else's */
1174 static int unlock_record(TDB_CONTEXT
*tdb
, tdb_off off
)
1176 struct tdb_traverse_lock
*i
;
1181 for (i
= &tdb
->travlocks
; i
; i
= i
->next
)
1184 return (count
== 1 ? tdb_brlock(tdb
, off
, F_UNLCK
, F_SETLKW
, 0) : 0);
1187 /* actually delete an entry in the database given the offset */
1188 static int do_delete(TDB_CONTEXT
*tdb
, tdb_off rec_ptr
, struct list_struct
*rec
)
1190 tdb_off last_ptr
, i
;
1191 struct list_struct lastrec
;
1193 if (tdb
->read_only
) return -1;
1195 if (write_lock_record(tdb
, rec_ptr
) == -1) {
1196 /* Someone traversing here: mark it as dead */
1197 rec
->magic
= TDB_DEAD_MAGIC
;
1198 return rec_write(tdb
, rec_ptr
, rec
);
1200 if (write_unlock_record(tdb
, rec_ptr
) != 0)
1203 /* find previous record in hash chain */
1204 if (ofs_read(tdb
, TDB_HASH_TOP(rec
->full_hash
), &i
) == -1)
1206 for (last_ptr
= 0; i
!= rec_ptr
; last_ptr
= i
, i
= lastrec
.next
)
1207 if (rec_read(tdb
, i
, &lastrec
) == -1)
1210 /* unlink it: next ptr is at start of record. */
1212 last_ptr
= TDB_HASH_TOP(rec
->full_hash
);
1213 if (ofs_write(tdb
, last_ptr
, &rec
->next
) == -1)
1216 /* recover the space */
1217 if (tdb_free(tdb
, rec_ptr
, rec
) == -1)
1222 /* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
1223 static int tdb_next_lock(TDB_CONTEXT
*tdb
, struct tdb_traverse_lock
*tlock
,
1224 struct list_struct
*rec
)
1226 int want_next
= (tlock
->off
!= 0);
1228 /* No traversal allows if you've called tdb_lockkeys() */
1229 if (tdb
->lockedkeys
)
1230 return TDB_ERRCODE(TDB_ERR_NOLOCK
, -1);
1232 /* Lock each chain from the start one. */
1233 for (; tlock
->hash
< tdb
->header
.hash_size
; tlock
->hash
++) {
1234 if (tdb_lock(tdb
, tlock
->hash
, F_WRLCK
) == -1)
1237 /* No previous record? Start at top of chain. */
1239 if (ofs_read(tdb
, TDB_HASH_TOP(tlock
->hash
),
1243 /* Otherwise unlock the previous record. */
1244 if (unlock_record(tdb
, tlock
->off
) != 0)
1249 /* We have offset of old record: grab next */
1250 if (rec_read(tdb
, tlock
->off
, rec
) == -1)
1252 tlock
->off
= rec
->next
;
1255 /* Iterate through chain */
1256 while( tlock
->off
) {
1258 if (rec_read(tdb
, tlock
->off
, rec
) == -1)
1260 if (!TDB_DEAD(rec
)) {
1261 /* Woohoo: we found one! */
1262 if (lock_record(tdb
, tlock
->off
) != 0)
1266 /* Try to clean dead ones from old traverses */
1267 current
= tlock
->off
;
1268 tlock
->off
= rec
->next
;
1269 if (!tdb
->read_only
&&
1270 do_delete(tdb
, current
, rec
) != 0)
1273 tdb_unlock(tdb
, tlock
->hash
, F_WRLCK
);
1276 /* We finished iteration without finding anything */
1277 return TDB_ERRCODE(TDB_SUCCESS
, 0);
1281 if (tdb_unlock(tdb
, tlock
->hash
, F_WRLCK
) != 0)
1282 TDB_LOG((tdb
, 0, "tdb_next_lock: On error unlock failed!\n"));
1286 /* traverse the entire database - calling fn(tdb, key, data) on each element.
1287 return -1 on error or the record count traversed
1288 if fn is NULL then it is not called
1289 a non-zero return value from fn() indicates that the traversal should stop
1291 int tdb_traverse(TDB_CONTEXT
*tdb
, tdb_traverse_func fn
, void *state
)
1294 struct list_struct rec
;
1295 struct tdb_traverse_lock tl
= { NULL
, 0, 0 };
1298 /* This was in the initializaton, above, but the IRIX compiler
1299 * did not like it. crh
1301 tl
.next
= tdb
->travlocks
.next
;
1303 /* fcntl locks don't stack: beware traverse inside traverse */
1304 tdb
->travlocks
.next
= &tl
;
1306 /* tdb_next_lock places locks on the record returned, and its chain */
1307 while ((ret
= tdb_next_lock(tdb
, &tl
, &rec
)) > 0) {
1309 /* now read the full record */
1310 key
.dptr
= tdb_alloc_read(tdb
, tl
.off
+ sizeof(rec
),
1311 rec
.key_len
+ rec
.data_len
);
1314 if (tdb_unlock(tdb
, tl
.hash
, F_WRLCK
) != 0)
1316 if (unlock_record(tdb
, tl
.off
) != 0)
1317 TDB_LOG((tdb
, 0, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
1320 key
.dsize
= rec
.key_len
;
1321 dbuf
.dptr
= key
.dptr
+ rec
.key_len
;
1322 dbuf
.dsize
= rec
.data_len
;
1324 /* Drop chain lock, call out */
1325 if (tdb_unlock(tdb
, tl
.hash
, F_WRLCK
) != 0) {
1329 if (fn
&& fn(tdb
, key
, dbuf
, state
)) {
1330 /* They want us to terminate traversal */
1332 if (unlock_record(tdb
, tl
.off
) != 0) {
1333 TDB_LOG((tdb
, 0, "tdb_traverse: unlock_record failed!\n"));;
1336 tdb
->travlocks
.next
= tl
.next
;
1337 SAFE_FREE(key
.dptr
);
1340 SAFE_FREE(key
.dptr
);
1343 tdb
->travlocks
.next
= tl
.next
;
1350 /* find the first entry in the database and return its key */
1351 TDB_DATA
tdb_firstkey(TDB_CONTEXT
*tdb
)
1354 struct list_struct rec
;
1356 /* release any old lock */
1357 if (unlock_record(tdb
, tdb
->travlocks
.off
) != 0)
1359 tdb
->travlocks
.off
= tdb
->travlocks
.hash
= 0;
1361 if (tdb_next_lock(tdb
, &tdb
->travlocks
, &rec
) <= 0)
1363 /* now read the key */
1364 key
.dsize
= rec
.key_len
;
1365 key
.dptr
=tdb_alloc_read(tdb
,tdb
->travlocks
.off
+sizeof(rec
),key
.dsize
);
1366 if (tdb_unlock(tdb
, BUCKET(tdb
->travlocks
.hash
), F_WRLCK
) != 0)
1367 TDB_LOG((tdb
, 0, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
1371 /* find the next entry in the database, returning its key */
1372 TDB_DATA
tdb_nextkey(TDB_CONTEXT
*tdb
, TDB_DATA oldkey
)
1375 TDB_DATA key
= tdb_null
;
1376 struct list_struct rec
;
1379 /* Is locked key the old key? If so, traverse will be reliable. */
1380 if (tdb
->travlocks
.off
) {
1381 if (tdb_lock(tdb
,tdb
->travlocks
.hash
,F_WRLCK
))
1383 if (rec_read(tdb
, tdb
->travlocks
.off
, &rec
) == -1
1384 || !(k
= tdb_alloc_read(tdb
,tdb
->travlocks
.off
+sizeof(rec
),
1386 || memcmp(k
, oldkey
.dptr
, oldkey
.dsize
) != 0) {
1387 /* No, it wasn't: unlock it and start from scratch */
1388 if (unlock_record(tdb
, tdb
->travlocks
.off
) != 0)
1390 if (tdb_unlock(tdb
, tdb
->travlocks
.hash
, F_WRLCK
) != 0)
1392 tdb
->travlocks
.off
= 0;
1398 if (!tdb
->travlocks
.off
) {
1399 /* No previous element: do normal find, and lock record */
1400 tdb
->travlocks
.off
= tdb_find_lock_hash(tdb
, oldkey
, tdb_hash(&oldkey
), F_WRLCK
, &rec
);
1401 if (!tdb
->travlocks
.off
)
1403 tdb
->travlocks
.hash
= BUCKET(rec
.full_hash
);
1404 if (lock_record(tdb
, tdb
->travlocks
.off
) != 0) {
1405 TDB_LOG((tdb
, 0, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno
)));
1409 oldhash
= tdb
->travlocks
.hash
;
1411 /* Grab next record: locks chain and returned record,
1412 unlocks old record */
1413 if (tdb_next_lock(tdb
, &tdb
->travlocks
, &rec
) > 0) {
1414 key
.dsize
= rec
.key_len
;
1415 key
.dptr
= tdb_alloc_read(tdb
, tdb
->travlocks
.off
+sizeof(rec
),
1417 /* Unlock the chain of this new record */
1418 if (tdb_unlock(tdb
, tdb
->travlocks
.hash
, F_WRLCK
) != 0)
1419 TDB_LOG((tdb
, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
1421 /* Unlock the chain of old record */
1422 if (tdb_unlock(tdb
, BUCKET(oldhash
), F_WRLCK
) != 0)
1423 TDB_LOG((tdb
, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
1427 /* delete an entry in the database given a key */
1428 static int tdb_delete_hash(TDB_CONTEXT
*tdb
, TDB_DATA key
, u32 hash
)
1431 struct list_struct rec
;
1434 if (!(rec_ptr
= tdb_find_lock_hash(tdb
, key
, hash
, F_WRLCK
, &rec
)))
1436 ret
= do_delete(tdb
, rec_ptr
, &rec
);
1437 if (tdb_unlock(tdb
, BUCKET(rec
.full_hash
), F_WRLCK
) != 0)
1438 TDB_LOG((tdb
, 0, "tdb_delete: WARNING tdb_unlock failed!\n"));
1442 int tdb_delete(TDB_CONTEXT
*tdb
, TDB_DATA key
)
1444 u32 hash
= tdb_hash(&key
);
1445 return tdb_delete_hash(tdb
, key
, hash
);
1448 /* store an element in the database, replacing any existing element
1451 return 0 on success, -1 on failure
1453 int tdb_store(TDB_CONTEXT
*tdb
, TDB_DATA key
, TDB_DATA dbuf
, int flag
)
1455 struct list_struct rec
;
1461 /* find which hash bucket it is in */
1462 hash
= tdb_hash(&key
);
1463 if (!tdb_keylocked(tdb
, hash
))
1465 if (tdb_lock(tdb
, BUCKET(hash
), F_WRLCK
) == -1)
1468 /* check for it existing, on insert. */
1469 if (flag
== TDB_INSERT
) {
1470 if (tdb_exists_hash(tdb
, key
, hash
)) {
1471 tdb
->ecode
= TDB_ERR_EXISTS
;
1475 /* first try in-place update, on modify or replace. */
1476 if (tdb_update_hash(tdb
, key
, hash
, dbuf
) == 0)
1478 if (flag
== TDB_MODIFY
&& tdb
->ecode
== TDB_ERR_NOEXIST
)
1481 /* reset the error code potentially set by the tdb_update() */
1482 tdb
->ecode
= TDB_SUCCESS
;
1484 /* delete any existing record - if it doesn't exist we don't
1485 care. Doing this first reduces fragmentation, and avoids
1486 coalescing with `allocated' block before it's updated. */
1487 if (flag
!= TDB_INSERT
)
1488 tdb_delete_hash(tdb
, key
, hash
);
1490 /* Copy key+value *before* allocating free space in case malloc
1491 fails and we are left with a dead spot in the tdb. */
1493 if (!(p
= (char *)malloc(key
.dsize
+ dbuf
.dsize
))) {
1494 tdb
->ecode
= TDB_ERR_OOM
;
1498 memcpy(p
, key
.dptr
, key
.dsize
);
1500 memcpy(p
+key
.dsize
, dbuf
.dptr
, dbuf
.dsize
);
1502 /* now we're into insert / modify / replace of a record which
1503 * we know could not be optimised by an in-place store (for
1504 * various reasons). */
1505 if (!(rec_ptr
= tdb_allocate(tdb
, key
.dsize
+ dbuf
.dsize
, &rec
)))
1508 /* Read hash top into next ptr */
1509 if (ofs_read(tdb
, TDB_HASH_TOP(hash
), &rec
.next
) == -1)
1512 rec
.key_len
= key
.dsize
;
1513 rec
.data_len
= dbuf
.dsize
;
1514 rec
.full_hash
= hash
;
1515 rec
.magic
= TDB_MAGIC
;
1517 /* write out and point the top of the hash chain at it */
1518 if (rec_write(tdb
, rec_ptr
, &rec
) == -1
1519 || tdb_write(tdb
, rec_ptr
+sizeof(rec
), p
, key
.dsize
+dbuf
.dsize
)==-1
1520 || ofs_write(tdb
, TDB_HASH_TOP(hash
), &rec_ptr
) == -1) {
1521 /* Need to tdb_unallocate() here */
1526 tdb_unlock(tdb
, BUCKET(hash
), F_WRLCK
);
1533 /* Attempt to append data to an entry in place - this only works if the new data size
1534 is <= the old data size and the key exists.
1535 on failure return -1. Record must be locked before calling.
1537 static int tdb_append_inplace(TDB_CONTEXT
*tdb
, TDB_DATA key
, u32 hash
, TDB_DATA new_dbuf
)
1539 struct list_struct rec
;
1543 if (!(rec_ptr
= tdb_find(tdb
, key
, hash
, &rec
)))
1546 /* Append of 0 is always ok. */
1547 if (new_dbuf
.dsize
== 0)
1550 /* must be long enough for key, old data + new data and tailer */
1551 if (rec
.rec_len
< key
.dsize
+ rec
.data_len
+ new_dbuf
.dsize
+ sizeof(tdb_off
)) {
1553 tdb
->ecode
= TDB_SUCCESS
; /* Not really an error */
1557 if (tdb_write(tdb
, rec_ptr
+ sizeof(rec
) + rec
.key_len
+ rec
.data_len
,
1558 new_dbuf
.dptr
, new_dbuf
.dsize
) == -1)
1562 rec
.data_len
+= new_dbuf
.dsize
;
1563 return rec_write(tdb
, rec_ptr
, &rec
);
1566 /* Append to an entry. Create if not exist. */
1568 int tdb_append(TDB_CONTEXT
*tdb
, TDB_DATA key
, TDB_DATA new_dbuf
)
1570 struct list_struct rec
;
1575 size_t new_data_size
= 0;
1577 /* find which hash bucket it is in */
1578 hash
= tdb_hash(&key
);
1579 if (!tdb_keylocked(tdb
, hash
))
1581 if (tdb_lock(tdb
, BUCKET(hash
), F_WRLCK
) == -1)
1584 /* first try in-place. */
1585 if (tdb_append_inplace(tdb
, key
, hash
, new_dbuf
) == 0)
1588 /* reset the error code potentially set by the tdb_append_inplace() */
1589 tdb
->ecode
= TDB_SUCCESS
;
1592 if (!(rec_ptr
= tdb_find(tdb
, key
, hash
, &rec
))) {
1593 if (tdb
->ecode
!= TDB_ERR_NOEXIST
)
1596 /* Not found - create. */
1598 ret
= tdb_store(tdb
, key
, new_dbuf
, TDB_INSERT
);
1602 new_data_size
= rec
.data_len
+ new_dbuf
.dsize
;
1604 /* Copy key+old_value+value *before* allocating free space in case malloc
1605 fails and we are left with a dead spot in the tdb. */
1607 if (!(p
= (char *)malloc(key
.dsize
+ new_data_size
))) {
1608 tdb
->ecode
= TDB_ERR_OOM
;
1612 /* Copy the key in place. */
1613 memcpy(p
, key
.dptr
, key
.dsize
);
1615 /* Now read the old data into place. */
1617 tdb_read(tdb
, rec_ptr
+ sizeof(rec
) + rec
.key_len
, p
+ key
.dsize
, rec
.data_len
, 0) == -1)
1620 /* Finally append the new data. */
1622 memcpy(p
+key
.dsize
+rec
.data_len
, new_dbuf
.dptr
, new_dbuf
.dsize
);
1624 /* delete any existing record - if it doesn't exist we don't
1625 care. Doing this first reduces fragmentation, and avoids
1626 coalescing with `allocated' block before it's updated. */
1628 tdb_delete_hash(tdb
, key
, hash
);
1630 if (!(rec_ptr
= tdb_allocate(tdb
, key
.dsize
+ new_data_size
, &rec
)))
1633 /* Read hash top into next ptr */
1634 if (ofs_read(tdb
, TDB_HASH_TOP(hash
), &rec
.next
) == -1)
1637 rec
.key_len
= key
.dsize
;
1638 rec
.data_len
= new_data_size
;
1639 rec
.full_hash
= hash
;
1640 rec
.magic
= TDB_MAGIC
;
1642 /* write out and point the top of the hash chain at it */
1643 if (rec_write(tdb
, rec_ptr
, &rec
) == -1
1644 || tdb_write(tdb
, rec_ptr
+sizeof(rec
), p
, key
.dsize
+new_data_size
)==-1
1645 || ofs_write(tdb
, TDB_HASH_TOP(hash
), &rec_ptr
) == -1) {
1646 /* Need to tdb_unallocate() here */
1652 tdb_unlock(tdb
, BUCKET(hash
), F_WRLCK
);
1660 static int tdb_already_open(dev_t device
,
1665 for (i
= tdbs
; i
; i
= i
->next
) {
1666 if (i
->device
== device
&& i
->inode
== ino
) {
1674 /* open the database, creating it if necessary
1676 The open_flags and mode are passed straight to the open call on the
1677 database file. A flags value of O_WRONLY is invalid. The hash size
1678 is advisory, use zero for a default value.
1680 Return is NULL on error, in which case errno is also set. Don't
1681 try to call tdb_error or tdb_errname, just do strerror(errno).
1683 @param name may be NULL for internal databases. */
1684 TDB_CONTEXT
*tdb_open(const char *name
, int hash_size
, int tdb_flags
,
1685 int open_flags
, mode_t mode
)
1687 return tdb_open_ex(name
, hash_size
, tdb_flags
, open_flags
, mode
, NULL
);
1691 TDB_CONTEXT
*tdb_open_ex(const char *name
, int hash_size
, int tdb_flags
,
1692 int open_flags
, mode_t mode
,
1693 tdb_log_func log_fn
)
1697 int rev
= 0, locked
;
1701 if (!(tdb
= calloc(1, sizeof *tdb
))) {
1702 /* Can't log this */
1708 tdb
->map_ptr
= NULL
;
1709 tdb
->lockedkeys
= NULL
;
1710 tdb
->flags
= tdb_flags
;
1711 tdb
->open_flags
= open_flags
;
1712 tdb
->log_fn
= log_fn
;
1714 if ((open_flags
& O_ACCMODE
) == O_WRONLY
) {
1715 TDB_LOG((tdb
, 0, "tdb_open_ex: can't open tdb %s write-only\n",
1722 hash_size
= DEFAULT_HASH_SIZE
;
1723 if ((open_flags
& O_ACCMODE
) == O_RDONLY
) {
1725 /* read only databases don't do locking or clear if first */
1726 tdb
->flags
|= TDB_NOLOCK
;
1727 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
1730 /* internal databases don't mmap or lock, and start off cleared */
1731 if (tdb
->flags
& TDB_INTERNAL
) {
1732 tdb
->flags
|= (TDB_NOLOCK
| TDB_NOMMAP
);
1733 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
1734 if (tdb_new_database(tdb
, hash_size
) != 0) {
1735 TDB_LOG((tdb
, 0, "tdb_open_ex: tdb_new_database failed!"));
1741 if ((tdb
->fd
= open(name
, open_flags
, mode
)) == -1) {
1742 TDB_LOG((tdb
, 5, "tdb_open_ex: could not open file %s: %s\n",
1743 name
, strerror(errno
)));
1744 goto fail
; /* errno set by open(2) */
1747 /* ensure there is only one process initialising at once */
1748 if (tdb_brlock(tdb
, GLOBAL_LOCK
, F_WRLCK
, F_SETLKW
, 0) == -1) {
1749 TDB_LOG((tdb
, 0, "tdb_open_ex: failed to get global lock on %s: %s\n",
1750 name
, strerror(errno
)));
1751 goto fail
; /* errno set by tdb_brlock */
1754 /* we need to zero database if we are the only one with it open */
1755 if ((locked
= (tdb_brlock(tdb
, ACTIVE_LOCK
, F_WRLCK
, F_SETLK
, 0) == 0))
1756 && (tdb_flags
& TDB_CLEAR_IF_FIRST
)) {
1757 open_flags
|= O_CREAT
;
1758 if (ftruncate(tdb
->fd
, 0) == -1) {
1759 TDB_LOG((tdb
, 0, "tdb_open_ex: "
1760 "failed to truncate %s: %s\n",
1761 name
, strerror(errno
)));
1762 goto fail
; /* errno set by ftruncate */
1766 if (read(tdb
->fd
, &tdb
->header
, sizeof(tdb
->header
)) != sizeof(tdb
->header
)
1767 || strcmp(tdb
->header
.magic_food
, TDB_MAGIC_FOOD
) != 0
1768 || (tdb
->header
.version
!= TDB_VERSION
1769 && !(rev
= (tdb
->header
.version
==TDB_BYTEREV(TDB_VERSION
))))) {
1770 /* its not a valid database - possibly initialise it */
1771 if (!(open_flags
& O_CREAT
) || tdb_new_database(tdb
, hash_size
) == -1) {
1772 errno
= EIO
; /* ie bad format or something */
1775 rev
= (tdb
->flags
& TDB_CONVERT
);
1777 vp
= (unsigned char *)&tdb
->header
.version
;
1778 vertest
= (((u32
)vp
[0]) << 24) | (((u32
)vp
[1]) << 16) |
1779 (((u32
)vp
[2]) << 8) | (u32
)vp
[3];
1780 tdb
->flags
|= (vertest
==TDB_VERSION
) ? TDB_BIGENDIAN
: 0;
1782 tdb
->flags
&= ~TDB_CONVERT
;
1784 tdb
->flags
|= TDB_CONVERT
;
1785 convert(&tdb
->header
, sizeof(tdb
->header
));
1787 if (fstat(tdb
->fd
, &st
) == -1)
1790 /* Is it already in the open list? If so, fail. */
1791 if (tdb_already_open(st
.st_dev
, st
.st_ino
)) {
1792 TDB_LOG((tdb
, 2, "tdb_open_ex: "
1793 "%s (%d,%d) is already open in this process\n",
1794 name
, st
.st_dev
, st
.st_ino
));
1799 if (!(tdb
->name
= (char *)strdup(name
))) {
1804 tdb
->map_size
= st
.st_size
;
1805 tdb
->device
= st
.st_dev
;
1806 tdb
->inode
= st
.st_ino
;
1807 tdb
->locked
= calloc(tdb
->header
.hash_size
+1, sizeof(tdb
->locked
[0]));
1809 TDB_LOG((tdb
, 2, "tdb_open_ex: "
1810 "failed to allocate lock structure for %s\n",
1817 if (!tdb
->read_only
)
1818 if (tdb_clear_spinlocks(tdb
) != 0) {
1819 TDB_LOG((tdb
, 0, "tdb_open_ex: "
1820 "failed to clear spinlock\n"));
1823 if (tdb_brlock(tdb
, ACTIVE_LOCK
, F_UNLCK
, F_SETLK
, 0) == -1) {
1824 TDB_LOG((tdb
, 0, "tdb_open_ex: "
1825 "failed to take ACTIVE_LOCK on %s: %s\n",
1826 name
, strerror(errno
)));
1830 /* leave this lock in place to indicate it's in use */
1831 if (tdb_brlock(tdb
, ACTIVE_LOCK
, F_RDLCK
, F_SETLKW
, 0) == -1)
1835 /* Internal (memory-only) databases skip all the code above to
1836 * do with disk files, and resume here by releasing their
1837 * global lock and hooking into the active list. */
1838 if (tdb_brlock(tdb
, GLOBAL_LOCK
, F_UNLCK
, F_SETLKW
, 0) == -1)
1845 { int save_errno
= errno
;
1851 if (tdb
->flags
& TDB_INTERNAL
)
1852 SAFE_FREE(tdb
->map_ptr
);
1856 SAFE_FREE(tdb
->name
);
1858 if (close(tdb
->fd
) != 0)
1859 TDB_LOG((tdb
, 5, "tdb_open_ex: failed to close tdb->fd on error!\n"));
1860 SAFE_FREE(tdb
->locked
);
1870 * @returns -1 for error; 0 for success.
1872 int tdb_close(TDB_CONTEXT
*tdb
)
1878 if (tdb
->flags
& TDB_INTERNAL
)
1879 SAFE_FREE(tdb
->map_ptr
);
1883 SAFE_FREE(tdb
->name
);
1885 ret
= close(tdb
->fd
);
1886 SAFE_FREE(tdb
->locked
);
1887 SAFE_FREE(tdb
->lockedkeys
);
1889 /* Remove from contexts list */
1890 for (i
= &tdbs
; *i
; i
= &(*i
)->next
) {
1897 memset(tdb
, 0, sizeof(*tdb
));
1903 /* lock/unlock entire database */
1904 int tdb_lockall(TDB_CONTEXT
*tdb
)
1908 /* There are no locks on read-only dbs */
1910 return TDB_ERRCODE(TDB_ERR_LOCK
, -1);
1911 if (tdb
->lockedkeys
)
1912 return TDB_ERRCODE(TDB_ERR_NOLOCK
, -1);
1913 for (i
= 0; i
< tdb
->header
.hash_size
; i
++)
1914 if (tdb_lock(tdb
, i
, F_WRLCK
))
1917 /* If error, release locks we have... */
1918 if (i
< tdb
->header
.hash_size
) {
1921 for ( j
= 0; j
< i
; j
++)
1922 tdb_unlock(tdb
, j
, F_WRLCK
);
1923 return TDB_ERRCODE(TDB_ERR_NOLOCK
, -1);
1928 void tdb_unlockall(TDB_CONTEXT
*tdb
)
1931 for (i
=0; i
< tdb
->header
.hash_size
; i
++)
1932 tdb_unlock(tdb
, i
, F_WRLCK
);
1935 int tdb_lockkeys(TDB_CONTEXT
*tdb
, u32 number
, TDB_DATA keys
[])
1939 /* Can't lock more keys if already locked */
1940 if (tdb
->lockedkeys
)
1941 return TDB_ERRCODE(TDB_ERR_NOLOCK
, -1);
1942 if (!(tdb
->lockedkeys
= malloc(sizeof(u32
) * (number
+1))))
1943 return TDB_ERRCODE(TDB_ERR_OOM
, -1);
1944 /* First number in array is # keys */
1945 tdb
->lockedkeys
[0] = number
;
1947 /* Insertion sort by bucket */
1948 for (i
= 0; i
< number
; i
++) {
1949 hash
= tdb_hash(&keys
[i
]);
1950 for (j
= 0; j
< i
&& BUCKET(tdb
->lockedkeys
[j
+1]) < BUCKET(hash
); j
++);
1951 memmove(&tdb
->lockedkeys
[j
+2], &tdb
->lockedkeys
[j
+1], sizeof(u32
) * (i
-j
));
1952 tdb
->lockedkeys
[j
+1] = hash
;
1954 /* Finally, lock in order */
1955 for (i
= 0; i
< number
; i
++)
1956 if (tdb_lock(tdb
, i
, F_WRLCK
))
1959 /* If error, release locks we have... */
1961 for ( j
= 0; j
< i
; j
++)
1962 tdb_unlock(tdb
, j
, F_WRLCK
);
1963 SAFE_FREE(tdb
->lockedkeys
);
1964 return TDB_ERRCODE(TDB_ERR_NOLOCK
, -1);
1969 /* Unlock the keys previously locked by tdb_lockkeys() */
1970 void tdb_unlockkeys(TDB_CONTEXT
*tdb
)
1973 if (!tdb
->lockedkeys
)
1975 for (i
= 0; i
< tdb
->lockedkeys
[0]; i
++)
1976 tdb_unlock(tdb
, tdb
->lockedkeys
[i
+1], F_WRLCK
);
1977 SAFE_FREE(tdb
->lockedkeys
);
1980 /* lock/unlock one hash chain. This is meant to be used to reduce
1981 contention - it cannot guarantee how many records will be locked */
1982 int tdb_chainlock(TDB_CONTEXT
*tdb
, TDB_DATA key
)
1984 return tdb_lock(tdb
, BUCKET(tdb_hash(&key
)), F_WRLCK
);
1987 int tdb_chainunlock(TDB_CONTEXT
*tdb
, TDB_DATA key
)
1989 return tdb_unlock(tdb
, BUCKET(tdb_hash(&key
)), F_WRLCK
);
1992 int tdb_chainlock_read(TDB_CONTEXT
*tdb
, TDB_DATA key
)
1994 return tdb_lock(tdb
, BUCKET(tdb_hash(&key
)), F_RDLCK
);
1997 int tdb_chainunlock_read(TDB_CONTEXT
*tdb
, TDB_DATA key
)
1999 return tdb_unlock(tdb
, BUCKET(tdb_hash(&key
)), F_RDLCK
);
2003 /* register a loging function */
2004 void tdb_logging_function(TDB_CONTEXT
*tdb
, void (*fn
)(TDB_CONTEXT
*, int , const char *, ...))
2010 /* reopen a tdb - this is used after a fork to ensure that we have an independent
2011 seek pointer from our parent and to re-establish locks */
2012 int tdb_reopen(TDB_CONTEXT
*tdb
)
2016 if (tdb_munmap(tdb
) != 0) {
2017 TDB_LOG((tdb
, 0, "tdb_reopen: munmap failed (%s)\n", strerror(errno
)));
2020 if (close(tdb
->fd
) != 0)
2021 TDB_LOG((tdb
, 0, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
2022 tdb
->fd
= open(tdb
->name
, tdb
->open_flags
& ~(O_CREAT
|O_TRUNC
), 0);
2023 if (tdb
->fd
== -1) {
2024 TDB_LOG((tdb
, 0, "tdb_reopen: open failed (%s)\n", strerror(errno
)));
2027 if (fstat(tdb
->fd
, &st
) != 0) {
2028 TDB_LOG((tdb
, 0, "tdb_reopen: fstat failed (%s)\n", strerror(errno
)));
2031 if (st
.st_ino
!= tdb
->inode
|| st
.st_dev
!= tdb
->device
) {
2032 TDB_LOG((tdb
, 0, "tdb_reopen: file dev/inode has changed!\n"));
2036 if (tdb_brlock(tdb
, ACTIVE_LOCK
, F_RDLCK
, F_SETLKW
, 0) == -1) {
2037 TDB_LOG((tdb
, 0, "tdb_reopen: failed to obtain active lock\n"));
2048 /* reopen all tdb's */
2049 int tdb_reopen_all(void)
2053 for (tdb
=tdbs
; tdb
; tdb
= tdb
->next
) {
2054 if (tdb_reopen(tdb
) != 0) return -1;