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 if (st.st_size < (size_t)off + len) {
68 /* Ensure ecode is set for log fn. */
69 tdb->ecode = TDB_ERR_IO;
70 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %u beyond eof at %u\n",
71 (int)(off + len), (int)st.st_size));
76 /* Beware >4G files! */
77 if ((tdb_off_t)st.st_size != st.st_size) {
78 /* Ensure ecode is set for log fn. */
79 tdb->ecode = TDB_ERR_IO;
80 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_oob len %llu too large!\n",
81 (long long)st.st_size));
85 /* Unmap, update size, remap */
86 if (tdb_munmap(tdb) == -1) {
87 tdb->ecode = TDB_ERR_IO;
90 tdb->map_size = st.st_size;
95 /* write a lump of data at a specified offset */
96 static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
97 const void *buf, tdb_len_t len)
103 if (tdb->read_only || tdb->traverse_read) {
104 tdb->ecode = TDB_ERR_RDONLY;
108 if (tdb->methods->tdb_oob(tdb, off, len, 0) != 0)
112 memcpy(off + (char *)tdb->map_ptr, buf, len);
114 ssize_t written = pwrite(tdb->fd, buf, len, off);
115 if ((written != (ssize_t)len) && (written != -1)) {
117 tdb->ecode = TDB_ERR_IO;
118 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: wrote only "
119 "%d of %d bytes at %d, trying once more\n",
120 (int)written, len, off));
121 written = pwrite(tdb->fd, (const char *)buf+written,
126 /* Ensure ecode is set for log fn. */
127 tdb->ecode = TDB_ERR_IO;
128 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_write failed at %d "
129 "len=%d (%s)\n", off, len, strerror(errno)));
131 } else if (written != (ssize_t)len) {
132 tdb->ecode = TDB_ERR_IO;
133 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: failed to "
134 "write %d bytes at %d in two attempts\n",
142 /* Endian conversion: we only ever deal with 4 byte quantities */
143 void *tdb_convert(void *buf, uint32_t size)
145 uint32_t i, *p = (uint32_t *)buf;
146 for (i = 0; i < size / 4; i++)
147 p[i] = TDB_BYTEREV(p[i]);
152 /* read a lump of data at a specified offset, maybe convert */
153 static int tdb_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
154 tdb_len_t len, int cv)
156 if (tdb->methods->tdb_oob(tdb, off, len, 0) != 0) {
161 memcpy(buf, off + (char *)tdb->map_ptr, len);
163 ssize_t ret = pread(tdb->fd, buf, len, off);
164 if (ret != (ssize_t)len) {
165 /* Ensure ecode is set for log fn. */
166 tdb->ecode = TDB_ERR_IO;
167 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_read failed at %d "
168 "len=%d ret=%d (%s) map_size=%d\n",
169 (int)off, (int)len, (int)ret, strerror(errno),
170 (int)tdb->map_size));
175 tdb_convert(buf, len);
183 do an unlocked scan of the hash table heads to find the next non-zero head. The value
184 will then be confirmed with the lock held
186 static void tdb_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
190 for (;h < tdb->header.hash_size;h++) {
191 if (0 != *(uint32_t *)(TDB_HASH_TOP(h) + (unsigned char *)tdb->map_ptr)) {
197 for (;h < tdb->header.hash_size;h++) {
198 if (tdb_ofs_read(tdb, TDB_HASH_TOP(h), &off) != 0 || off != 0) {
207 int tdb_munmap(struct tdb_context *tdb)
209 if (tdb->flags & TDB_INTERNAL)
216 ret = munmap(tdb->map_ptr, tdb->map_size);
225 void tdb_mmap(struct tdb_context *tdb)
227 if (tdb->flags & TDB_INTERNAL)
231 if (!(tdb->flags & TDB_NOMMAP)) {
232 tdb->map_ptr = mmap(NULL, tdb->map_size,
233 PROT_READ|(tdb->read_only? 0:PROT_WRITE),
234 MAP_SHARED|MAP_FILE, tdb->fd, 0);
237 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
240 if (tdb->map_ptr == MAP_FAILED) {
242 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_mmap failed for size %d (%s)\n",
243 tdb->map_size, strerror(errno)));
253 /* expand a file. we prefer to use ftruncate, as that is what posix
254 says to use for mmap expansion */
255 static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t addition)
259 if (tdb->read_only || tdb->traverse_read) {
260 tdb->ecode = TDB_ERR_RDONLY;
264 if (ftruncate(tdb->fd, size+addition) == -1) {
266 ssize_t written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
268 /* try once more, potentially revealing errno */
269 written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
272 /* again - give up, guessing errno */
276 TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file to %d failed (%s)\n",
277 size+addition, strerror(errno)));
282 /* now fill the file with something. This ensures that the
283 file isn't sparse, which would be very bad if we ran out of
284 disk. This must be done with write, not via mmap */
285 memset(buf, TDB_PAD_BYTE, sizeof(buf));
287 size_t n = addition>sizeof(buf)?sizeof(buf):addition;
288 ssize_t written = pwrite(tdb->fd, buf, n, size);
290 /* prevent infinite loops: try _once_ more */
291 written = pwrite(tdb->fd, buf, n, size);
294 /* give up, trying to provide a useful errno */
295 TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write "
296 "returned 0 twice: giving up!\n"));
299 } else if (written == -1) {
300 TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write of "
301 "%d bytes failed (%s)\n", (int)n,
304 } else if (written != n) {
305 TDB_LOG((tdb, TDB_DEBUG_WARNING, "expand_file: wrote "
306 "only %d of %d bytes - retrying\n", (int)written,
316 /* expand the database at least size bytes by expanding the underlying
317 file and doing the mmap again if necessary */
318 int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
320 struct tdb_record rec;
321 tdb_off_t offset, new_size, top_size, map_size;
323 if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
324 TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
328 /* must know about any previous expansions by another process */
329 tdb->methods->tdb_oob(tdb, tdb->map_size, 1, 1);
331 /* limit size in order to avoid using up huge amounts of memory for
332 * in memory tdbs if an oddball huge record creeps in */
333 if (size > 100 * 1024) {
334 top_size = tdb->map_size + size * 2;
336 top_size = tdb->map_size + size * 100;
339 /* always make room for at least top_size more records, and at
340 least 25% more space. if the DB is smaller than 100MiB,
341 otherwise grow it by 10% only. */
342 if (tdb->map_size > 100 * 1024 * 1024) {
343 map_size = tdb->map_size * 1.10;
345 map_size = tdb->map_size * 1.25;
348 /* Round the database up to a multiple of the page size */
349 new_size = MAX(top_size, map_size);
350 size = TDB_ALIGN(new_size, tdb->page_size) - tdb->map_size;
352 if (!(tdb->flags & TDB_INTERNAL))
356 * We must ensure the file is unmapped before doing this
357 * to ensure consistency with systems like OpenBSD where
358 * writes and mmaps are not consistent.
361 /* expand the file itself */
362 if (!(tdb->flags & TDB_INTERNAL)) {
363 if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
367 tdb->map_size += size;
369 if (tdb->flags & TDB_INTERNAL) {
370 char *new_map_ptr = (char *)realloc(tdb->map_ptr,
373 tdb->map_size -= size;
376 tdb->map_ptr = new_map_ptr;
379 * We must ensure the file is remapped before adding the space
380 * to ensure consistency with systems like OpenBSD where
381 * writes and mmaps are not consistent.
384 /* We're ok if the mmap fails as we'll fallback to read/write */
388 /* form a new freelist record */
389 memset(&rec,'\0',sizeof(rec));
390 rec.rec_len = size - sizeof(rec);
392 /* link it into the free list */
393 offset = tdb->map_size - size;
394 if (tdb_free(tdb, offset, &rec) == -1)
397 tdb_unlock(tdb, -1, F_WRLCK);
400 tdb_unlock(tdb, -1, F_WRLCK);
404 /* read/write a tdb_off_t */
405 int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
407 return tdb->methods->tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
410 int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
413 return tdb->methods->tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
417 /* read a lump of data, allocating the space for it */
418 unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
422 /* some systems don't like zero length malloc */
424 if (!(buf = (unsigned char *)malloc(len ? len : 1))) {
425 /* Ensure ecode is set for log fn. */
426 tdb->ecode = TDB_ERR_OOM;
427 TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_alloc_read malloc failed len=%d (%s)\n",
428 len, strerror(errno)));
431 if (tdb->methods->tdb_read(tdb, offset, buf, len, 0) == -1) {
438 /* Give a piece of tdb data to a parser */
440 int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
441 tdb_off_t offset, tdb_len_t len,
442 int (*parser)(TDB_DATA key, TDB_DATA data,
451 if ((tdb->transaction == NULL) && (tdb->map_ptr != NULL)) {
453 * Optimize by avoiding the malloc/memcpy/free, point the
454 * parser directly at the mmap area.
456 if (tdb->methods->tdb_oob(tdb, offset, len, 0) != 0) {
459 data.dptr = offset + (unsigned char *)tdb->map_ptr;
460 return parser(key, data, private_data);
463 if (!(data.dptr = tdb_alloc_read(tdb, offset, len))) {
467 result = parser(key, data, private_data);
472 /* read/write a record */
473 int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
475 if (tdb->methods->tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
477 if (TDB_BAD_MAGIC(rec)) {
478 /* Ensure ecode is set for log fn. */
479 tdb->ecode = TDB_ERR_CORRUPT;
480 TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
483 return tdb->methods->tdb_oob(tdb, rec->next, sizeof(*rec), 0);
486 int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
488 struct tdb_record r = *rec;
489 return tdb->methods->tdb_write(tdb, offset, CONVERT(r), sizeof(r));
492 static const struct tdb_methods io_methods = {
501 initialise the default methods table
503 void tdb_io_init(struct tdb_context *tdb)
505 tdb->methods = &io_methods;