2 Unix SMB/CIFS implementation.
4 trivial database library
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
10 ** NOTE! The following LGPL license applies to the tdb
11 ** library. This does NOT imply that all of Samba is released
14 This library is free software; you can redistribute it and/or
15 modify it under the terms of the GNU Lesser General Public
16 License as published by the Free Software Foundation; either
17 version 3 of the License, or (at your option) any later version.
19 This library is distributed in the hope that it will be useful,
20 but WITHOUT ANY WARRANTY; without even the implied warranty of
21 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 Lesser General Public License for more details.
24 You should have received a copy of the GNU Lesser General Public
25 License along with this library; if not, see <http://www.gnu.org/licenses/>.
29 #include "tdb_private.h"
31 /* check for an out of bounds access - if it is out of bounds then
32 see if the database has been expanded by someone else and expand
35 static int tdb_oob(struct tdb_context
*tdb
, tdb_off_t off
, tdb_len_t len
,
39 if (len
+ off
< len
) {
41 /* Ensure ecode is set for log fn. */
42 tdb
->ecode
= TDB_ERR_IO
;
43 TDB_LOG((tdb
, TDB_DEBUG_FATAL
,"tdb_oob off %d len %d wrap\n",
49 if (off
+ len
<= tdb
->map_size
)
51 if (tdb
->flags
& TDB_INTERNAL
) {
53 /* Ensure ecode is set for log fn. */
54 tdb
->ecode
= TDB_ERR_IO
;
55 TDB_LOG((tdb
, TDB_DEBUG_FATAL
,"tdb_oob len %u beyond internal malloc size %u\n",
56 (int)(off
+ len
), (int)tdb
->map_size
));
61 if (fstat(tdb
->fd
, &st
) == -1) {
62 tdb
->ecode
= TDB_ERR_IO
;
66 /* Beware >4G files! */
67 if ((tdb_off_t
)st
.st_size
!= st
.st_size
) {
68 /* Ensure ecode is set for log fn. */
69 tdb
->ecode
= TDB_ERR_IO
;
70 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_oob len %llu too large!\n",
71 (long long)st
.st_size
));
75 /* Unmap, update size, remap. We do this unconditionally, to handle
76 * the unusual case where the db is truncated.
78 * This can happen to a child using tdb_reopen_all(true) on a
79 * TDB_CLEAR_IF_FIRST tdb whose parent crashes: the next
80 * opener will truncate the database. */
81 if (tdb_munmap(tdb
) == -1) {
82 tdb
->ecode
= TDB_ERR_IO
;
85 tdb
->map_size
= st
.st_size
;
86 if (tdb_mmap(tdb
) != 0) {
90 if (st
.st_size
< (size_t)off
+ len
) {
92 /* Ensure ecode is set for log fn. */
93 tdb
->ecode
= TDB_ERR_IO
;
94 TDB_LOG((tdb
, TDB_DEBUG_FATAL
,"tdb_oob len %u beyond eof at %u\n",
95 (int)(off
+ len
), (int)st
.st_size
));
102 /* write a lump of data at a specified offset */
103 static int tdb_write(struct tdb_context
*tdb
, tdb_off_t off
,
104 const void *buf
, tdb_len_t len
)
110 if (tdb
->read_only
|| tdb
->traverse_read
) {
111 tdb
->ecode
= TDB_ERR_RDONLY
;
115 if (tdb
->methods
->tdb_oob(tdb
, off
, len
, 0) != 0)
119 memcpy(off
+ (char *)tdb
->map_ptr
, buf
, len
);
121 #ifdef HAVE_INCOHERENT_MMAP
122 tdb
->ecode
= TDB_ERR_IO
;
125 ssize_t written
= pwrite(tdb
->fd
, buf
, len
, off
);
126 if ((written
!= (ssize_t
)len
) && (written
!= -1)) {
128 tdb
->ecode
= TDB_ERR_IO
;
129 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_write: wrote only "
130 "%d of %d bytes at %d, trying once more\n",
131 (int)written
, len
, off
));
132 written
= pwrite(tdb
->fd
, (const char *)buf
+written
,
137 /* Ensure ecode is set for log fn. */
138 tdb
->ecode
= TDB_ERR_IO
;
139 TDB_LOG((tdb
, TDB_DEBUG_FATAL
,"tdb_write failed at %d "
140 "len=%d (%s)\n", off
, len
, strerror(errno
)));
142 } else if (written
!= (ssize_t
)len
) {
143 tdb
->ecode
= TDB_ERR_IO
;
144 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_write: failed to "
145 "write %d bytes at %d in two attempts\n",
154 /* Endian conversion: we only ever deal with 4 byte quantities */
155 void *tdb_convert(void *buf
, uint32_t size
)
157 uint32_t i
, *p
= (uint32_t *)buf
;
158 for (i
= 0; i
< size
/ 4; i
++)
159 p
[i
] = TDB_BYTEREV(p
[i
]);
164 /* read a lump of data at a specified offset, maybe convert */
165 static int tdb_read(struct tdb_context
*tdb
, tdb_off_t off
, void *buf
,
166 tdb_len_t len
, int cv
)
168 if (tdb
->methods
->tdb_oob(tdb
, off
, len
, 0) != 0) {
173 memcpy(buf
, off
+ (char *)tdb
->map_ptr
, len
);
175 #ifdef HAVE_INCOHERENT_MMAP
176 tdb
->ecode
= TDB_ERR_IO
;
179 ssize_t ret
= pread(tdb
->fd
, buf
, len
, off
);
180 if (ret
!= (ssize_t
)len
) {
181 /* Ensure ecode is set for log fn. */
182 tdb
->ecode
= TDB_ERR_IO
;
183 TDB_LOG((tdb
, TDB_DEBUG_FATAL
,"tdb_read failed at %d "
184 "len=%d ret=%d (%s) map_size=%d\n",
185 (int)off
, (int)len
, (int)ret
, strerror(errno
),
186 (int)tdb
->map_size
));
192 tdb_convert(buf
, len
);
200 do an unlocked scan of the hash table heads to find the next non-zero head. The value
201 will then be confirmed with the lock held
203 static void tdb_next_hash_chain(struct tdb_context
*tdb
, uint32_t *chain
)
207 for (;h
< tdb
->header
.hash_size
;h
++) {
208 if (0 != *(uint32_t *)(TDB_HASH_TOP(h
) + (unsigned char *)tdb
->map_ptr
)) {
214 for (;h
< tdb
->header
.hash_size
;h
++) {
215 if (tdb_ofs_read(tdb
, TDB_HASH_TOP(h
), &off
) != 0 || off
!= 0) {
224 int tdb_munmap(struct tdb_context
*tdb
)
226 if (tdb
->flags
& TDB_INTERNAL
)
233 ret
= munmap(tdb
->map_ptr
, tdb
->map_size
);
242 /* If mmap isn't coherent, *everyone* must always mmap. */
243 static bool should_mmap(const struct tdb_context
*tdb
)
245 #ifdef HAVE_INCOHERENT_MMAP
248 return !(tdb
->flags
& TDB_NOMMAP
);
252 int tdb_mmap(struct tdb_context
*tdb
)
254 if (tdb
->flags
& TDB_INTERNAL
)
258 if (should_mmap(tdb
)) {
259 tdb
->map_ptr
= mmap(NULL
, tdb
->map_size
,
260 PROT_READ
|(tdb
->read_only
? 0:PROT_WRITE
),
261 MAP_SHARED
|MAP_FILE
, tdb
->fd
, 0);
264 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
267 if (tdb
->map_ptr
== MAP_FAILED
) {
269 TDB_LOG((tdb
, TDB_DEBUG_WARNING
, "tdb_mmap failed for size %d (%s)\n",
270 tdb
->map_size
, strerror(errno
)));
271 #ifdef HAVE_INCOHERENT_MMAP
272 tdb
->ecode
= TDB_ERR_IO
;
285 /* expand a file. we prefer to use ftruncate, as that is what posix
286 says to use for mmap expansion */
287 static int tdb_expand_file(struct tdb_context
*tdb
, tdb_off_t size
, tdb_off_t addition
)
291 if (tdb
->read_only
|| tdb
->traverse_read
) {
292 tdb
->ecode
= TDB_ERR_RDONLY
;
296 if (ftruncate(tdb
->fd
, size
+addition
) == -1) {
298 ssize_t written
= pwrite(tdb
->fd
, &b
, 1, (size
+addition
) - 1);
300 /* try once more, potentially revealing errno */
301 written
= pwrite(tdb
->fd
, &b
, 1, (size
+addition
) - 1);
304 /* again - give up, guessing errno */
308 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "expand_file to %d failed (%s)\n",
309 size
+addition
, strerror(errno
)));
314 /* now fill the file with something. This ensures that the
315 file isn't sparse, which would be very bad if we ran out of
316 disk. This must be done with write, not via mmap */
317 memset(buf
, TDB_PAD_BYTE
, sizeof(buf
));
319 size_t n
= addition
>sizeof(buf
)?sizeof(buf
):addition
;
320 ssize_t written
= pwrite(tdb
->fd
, buf
, n
, size
);
322 /* prevent infinite loops: try _once_ more */
323 written
= pwrite(tdb
->fd
, buf
, n
, size
);
326 /* give up, trying to provide a useful errno */
327 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "expand_file write "
328 "returned 0 twice: giving up!\n"));
331 } else if (written
== -1) {
332 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "expand_file write of "
333 "%d bytes failed (%s)\n", (int)n
,
336 } else if (written
!= n
) {
337 TDB_LOG((tdb
, TDB_DEBUG_WARNING
, "expand_file: wrote "
338 "only %d of %d bytes - retrying\n", (int)written
,
348 /* You need 'size', this tells you how much you should expand by. */
349 tdb_off_t
tdb_expand_adjust(tdb_off_t map_size
, tdb_off_t size
, int page_size
)
351 tdb_off_t new_size
, top_size
;
353 /* limit size in order to avoid using up huge amounts of memory for
354 * in memory tdbs if an oddball huge record creeps in */
355 if (size
> 100 * 1024) {
356 top_size
= map_size
+ size
* 2;
358 top_size
= map_size
+ size
* 100;
361 /* always make room for at least top_size more records, and at
362 least 25% more space. if the DB is smaller than 100MiB,
363 otherwise grow it by 10% only. */
364 if (map_size
> 100 * 1024 * 1024) {
365 new_size
= map_size
* 1.10;
367 new_size
= map_size
* 1.25;
370 /* Round the database up to a multiple of the page size */
371 new_size
= MAX(top_size
, new_size
);
372 return TDB_ALIGN(new_size
, page_size
) - map_size
;
375 /* expand the database at least size bytes by expanding the underlying
376 file and doing the mmap again if necessary */
377 int tdb_expand(struct tdb_context
*tdb
, tdb_off_t size
)
379 struct tdb_record rec
;
382 if (tdb_lock(tdb
, -1, F_WRLCK
) == -1) {
383 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "lock failed in tdb_expand\n"));
387 /* must know about any previous expansions by another process */
388 tdb
->methods
->tdb_oob(tdb
, tdb
->map_size
, 1, 1);
390 size
= tdb_expand_adjust(tdb
->map_size
, size
, tdb
->page_size
);
392 /* expand the file itself */
393 if (!(tdb
->flags
& TDB_INTERNAL
)) {
394 if (tdb
->methods
->tdb_expand_file(tdb
, tdb
->map_size
, size
) != 0)
398 /* form a new freelist record */
399 offset
= tdb
->map_size
;
400 memset(&rec
,'\0',sizeof(rec
));
401 rec
.rec_len
= size
- sizeof(rec
);
403 if (tdb
->flags
& TDB_INTERNAL
) {
404 char *new_map_ptr
= (char *)realloc(tdb
->map_ptr
,
405 tdb
->map_size
+ size
);
409 tdb
->map_ptr
= new_map_ptr
;
410 tdb
->map_size
+= size
;
412 /* Explicitly remap: if we're in a transaction, this won't
413 * happen automatically! */
415 tdb
->map_size
+= size
;
416 if (tdb_mmap(tdb
) != 0) {
421 /* link it into the free list */
422 if (tdb_free(tdb
, offset
, &rec
) == -1)
425 tdb_unlock(tdb
, -1, F_WRLCK
);
428 tdb_unlock(tdb
, -1, F_WRLCK
);
432 /* read/write a tdb_off_t */
433 int tdb_ofs_read(struct tdb_context
*tdb
, tdb_off_t offset
, tdb_off_t
*d
)
435 return tdb
->methods
->tdb_read(tdb
, offset
, (char*)d
, sizeof(*d
), DOCONV());
438 int tdb_ofs_write(struct tdb_context
*tdb
, tdb_off_t offset
, tdb_off_t
*d
)
441 return tdb
->methods
->tdb_write(tdb
, offset
, CONVERT(off
), sizeof(*d
));
445 /* read a lump of data, allocating the space for it */
446 unsigned char *tdb_alloc_read(struct tdb_context
*tdb
, tdb_off_t offset
, tdb_len_t len
)
450 /* some systems don't like zero length malloc */
452 if (!(buf
= (unsigned char *)malloc(len
? len
: 1))) {
453 /* Ensure ecode is set for log fn. */
454 tdb
->ecode
= TDB_ERR_OOM
;
455 TDB_LOG((tdb
, TDB_DEBUG_ERROR
,"tdb_alloc_read malloc failed len=%d (%s)\n",
456 len
, strerror(errno
)));
459 if (tdb
->methods
->tdb_read(tdb
, offset
, buf
, len
, 0) == -1) {
466 /* Give a piece of tdb data to a parser */
468 int tdb_parse_data(struct tdb_context
*tdb
, TDB_DATA key
,
469 tdb_off_t offset
, tdb_len_t len
,
470 int (*parser
)(TDB_DATA key
, TDB_DATA data
,
479 if ((tdb
->transaction
== NULL
) && (tdb
->map_ptr
!= NULL
)) {
481 * Optimize by avoiding the malloc/memcpy/free, point the
482 * parser directly at the mmap area.
484 if (tdb
->methods
->tdb_oob(tdb
, offset
, len
, 0) != 0) {
487 data
.dptr
= offset
+ (unsigned char *)tdb
->map_ptr
;
488 return parser(key
, data
, private_data
);
491 if (!(data
.dptr
= tdb_alloc_read(tdb
, offset
, len
))) {
495 result
= parser(key
, data
, private_data
);
500 /* read/write a record */
501 int tdb_rec_read(struct tdb_context
*tdb
, tdb_off_t offset
, struct tdb_record
*rec
)
503 if (tdb
->methods
->tdb_read(tdb
, offset
, rec
, sizeof(*rec
),DOCONV()) == -1)
505 if (TDB_BAD_MAGIC(rec
)) {
506 /* Ensure ecode is set for log fn. */
507 tdb
->ecode
= TDB_ERR_CORRUPT
;
508 TDB_LOG((tdb
, TDB_DEBUG_FATAL
,"tdb_rec_read bad magic 0x%x at offset=%d\n", rec
->magic
, offset
));
511 return tdb
->methods
->tdb_oob(tdb
, rec
->next
, sizeof(*rec
), 0);
514 int tdb_rec_write(struct tdb_context
*tdb
, tdb_off_t offset
, struct tdb_record
*rec
)
516 struct tdb_record r
= *rec
;
517 return tdb
->methods
->tdb_write(tdb
, offset
, CONVERT(r
), sizeof(r
));
520 static const struct tdb_methods io_methods
= {
529 initialise the default methods table
531 void tdb_io_init(struct tdb_context
*tdb
)
533 tdb
->methods
= &io_methods
;