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/>.
28 #include "tdb_private.h"
30 #define MAX(a,b) ((a) > (b) ? (a) : (b))
33 /* check for an out of bounds access - if it is out of bounds then
34 see if the database has been expanded by someone else and expand
36 note that "len" is the minimum length needed for the db
38 static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
41 if (len <= tdb->map_size)
43 if (tdb->flags & TDB_INTERNAL) {
45 /* Ensure ecode is set for log fn. */
46 tdb->ecode = TDB_ERR_IO;
47 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond internal malloc size %d\n",
48 (int)len, (int)tdb->map_size));
53 if (fstat(tdb->fd, &st) == -1) {
54 tdb->ecode = TDB_ERR_IO;
58 if (st.st_size < (size_t)len) {
60 /* Ensure ecode is set for log fn. */
61 tdb->ecode = TDB_ERR_IO;
62 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond eof at %d\n",
63 (int)len, (int)st.st_size));
68 /* Unmap, update size, remap */
69 if (tdb_munmap(tdb) == -1) {
70 tdb->ecode = TDB_ERR_IO;
73 tdb->map_size = st.st_size;
78 /* write a lump of data at a specified offset */
79 static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
80 const void *buf, tdb_len_t len)
86 if (tdb->read_only || tdb->traverse_read) {
87 tdb->ecode = TDB_ERR_RDONLY;
91 if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0)
95 memcpy(off + (char *)tdb->map_ptr, buf, len);
97 ssize_t written = pwrite(tdb->fd, buf, len, off);
98 if ((written != (ssize_t)len) && (written != -1)) {
100 tdb->ecode = TDB_ERR_IO;
101 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: wrote only "
102 "%d of %d bytes at %d, trying once more\n",
103 (int)written, len, off));
104 written = pwrite(tdb->fd, (const char *)buf+written,
109 /* Ensure ecode is set for log fn. */
110 tdb->ecode = TDB_ERR_IO;
111 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_write failed at %d "
112 "len=%d (%s)\n", off, len, strerror(errno)));
114 } else if (written != (ssize_t)len) {
115 tdb->ecode = TDB_ERR_IO;
116 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: failed to "
117 "write %d bytes at %d in two attempts\n",
125 /* Endian conversion: we only ever deal with 4 byte quantities */
126 void *tdb_convert(void *buf, uint32_t size)
129 unsigned char *p = buf, tmp;
131 for (i = 0; i < size; i += 4) {
143 /* read a lump of data at a specified offset, maybe convert */
144 static int tdb_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
145 tdb_len_t len, int cv)
147 if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0) {
152 memcpy(buf, off + (char *)tdb->map_ptr, len);
154 ssize_t ret = pread(tdb->fd, buf, len, off);
155 if (ret != (ssize_t)len) {
156 /* Ensure ecode is set for log fn. */
157 tdb->ecode = TDB_ERR_IO;
158 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_read failed at %d "
159 "len=%d ret=%d (%s) map_size=%d\n",
160 (int)off, (int)len, (int)ret, strerror(errno),
161 (int)tdb->map_size));
166 tdb_convert(buf, len);
174 do an unlocked scan of the hash table heads to find the next non-zero head. The value
175 will then be confirmed with the lock held
177 static void tdb_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
181 for (;h < tdb->header.hash_size;h++) {
182 if (0 != *(uint32_t *)(TDB_HASH_TOP(h) + (unsigned char *)tdb->map_ptr)) {
188 for (;h < tdb->header.hash_size;h++) {
189 if (tdb_ofs_read(tdb, TDB_HASH_TOP(h), &off) != 0 || off != 0) {
198 int tdb_munmap(struct tdb_context *tdb)
200 if (tdb->flags & TDB_INTERNAL)
207 ret = munmap(tdb->map_ptr, tdb->map_size);
216 void tdb_mmap(struct tdb_context *tdb)
218 if (tdb->flags & TDB_INTERNAL)
222 if (!(tdb->flags & TDB_NOMMAP)) {
223 tdb->map_ptr = mmap(NULL, tdb->map_size,
224 PROT_READ|(tdb->read_only? 0:PROT_WRITE),
225 MAP_SHARED, tdb->fd, 0);
228 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
231 if (tdb->map_ptr == MAP_FAILED) {
233 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_mmap failed for size %d (%s)\n",
234 tdb->map_size, strerror(errno)));
244 /* expand a file. we prefer to use ftruncate, as that is what posix
245 says to use for mmap expansion */
246 static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t addition)
250 if (tdb->read_only || tdb->traverse_read) {
251 tdb->ecode = TDB_ERR_RDONLY;
255 if (ftruncate(tdb->fd, size+addition) == -1) {
257 ssize_t written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
259 /* try once more, potentially revealing errno */
260 written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
263 /* again - give up, guessing errno */
267 TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file to %d failed (%s)\n",
268 size+addition, strerror(errno)));
273 /* now fill the file with something. This ensures that the
274 file isn't sparse, which would be very bad if we ran out of
275 disk. This must be done with write, not via mmap */
276 memset(buf, TDB_PAD_BYTE, sizeof(buf));
278 size_t n = addition>sizeof(buf)?sizeof(buf):addition;
279 ssize_t written = pwrite(tdb->fd, buf, n, size);
281 /* prevent infinite loops: try _once_ more */
282 written = pwrite(tdb->fd, buf, n, size);
285 /* give up, trying to provide a useful errno */
286 TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write "
287 "returned 0 twice: giving up!\n"));
290 } else if (written == -1) {
291 TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write of "
292 "%d bytes failed (%s)\n", (int)n,
295 } else if (written != n) {
296 TDB_LOG((tdb, TDB_DEBUG_WARNING, "expand_file: wrote "
297 "only %d of %d bytes - retrying\n", (int)written,
307 /* expand the database at least size bytes by expanding the underlying
308 file and doing the mmap again if necessary */
309 int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
311 struct tdb_record rec;
312 tdb_off_t offset, new_size;
314 if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
315 TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
319 /* must know about any previous expansions by another process */
320 tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
322 /* always make room for at least 100 more records, and at
323 least 25% more space. Round the database up to a multiple
325 new_size = MAX(tdb->map_size + size*100, tdb->map_size * 1.25);
326 size = TDB_ALIGN(new_size, tdb->page_size) - tdb->map_size;
328 if (!(tdb->flags & TDB_INTERNAL))
332 * We must ensure the file is unmapped before doing this
333 * to ensure consistency with systems like OpenBSD where
334 * writes and mmaps are not consistent.
337 /* expand the file itself */
338 if (!(tdb->flags & TDB_INTERNAL)) {
339 if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
343 tdb->map_size += size;
345 if (tdb->flags & TDB_INTERNAL) {
346 char *new_map_ptr = (char *)realloc(tdb->map_ptr,
349 tdb->map_size -= size;
352 tdb->map_ptr = new_map_ptr;
355 * We must ensure the file is remapped before adding the space
356 * to ensure consistency with systems like OpenBSD where
357 * writes and mmaps are not consistent.
360 /* We're ok if the mmap fails as we'll fallback to read/write */
364 /* form a new freelist record */
365 memset(&rec,'\0',sizeof(rec));
366 rec.rec_len = size - sizeof(rec);
368 /* link it into the free list */
369 offset = tdb->map_size - size;
370 if (tdb_free(tdb, offset, &rec) == -1)
373 tdb_unlock(tdb, -1, F_WRLCK);
376 tdb_unlock(tdb, -1, F_WRLCK);
380 /* read/write a tdb_off_t */
381 int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
383 return tdb->methods->tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
386 int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
389 return tdb->methods->tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
393 /* read a lump of data, allocating the space for it */
394 unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
398 /* some systems don't like zero length malloc */
400 if (!(buf = (unsigned char *)malloc(len ? len : 1))) {
401 /* Ensure ecode is set for log fn. */
402 tdb->ecode = TDB_ERR_OOM;
403 TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_alloc_read malloc failed len=%d (%s)\n",
404 len, strerror(errno)));
407 if (tdb->methods->tdb_read(tdb, offset, buf, len, 0) == -1) {
414 /* Give a piece of tdb data to a parser */
416 int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
417 tdb_off_t offset, tdb_len_t len,
418 int (*parser)(TDB_DATA key, TDB_DATA data,
427 if ((tdb->transaction == NULL) && (tdb->map_ptr != NULL)) {
429 * Optimize by avoiding the malloc/memcpy/free, point the
430 * parser directly at the mmap area.
432 if (tdb->methods->tdb_oob(tdb, offset+len, 0) != 0) {
435 data.dptr = offset + (unsigned char *)tdb->map_ptr;
436 return parser(key, data, private_data);
439 if (!(data.dptr = tdb_alloc_read(tdb, offset, len))) {
443 result = parser(key, data, private_data);
448 /* read/write a record */
449 int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
451 if (tdb->methods->tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
453 if (TDB_BAD_MAGIC(rec)) {
454 /* Ensure ecode is set for log fn. */
455 tdb->ecode = TDB_ERR_CORRUPT;
456 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
459 return tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0);
462 int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
464 struct tdb_record r = *rec;
465 return tdb->methods->tdb_write(tdb, offset, CONVERT(r), sizeof(r));
468 static const struct tdb_methods io_methods = {
477 initialise the default methods table
479 void tdb_io_init(struct tdb_context *tdb)
481 tdb->methods = &io_methods;