2 Unix SMB/CIFS implementation.
4 trivial database library
6 Copyright (C) Andrew Tridgell 1999-2004
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 2 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, write to the Free Software
26 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
48 #include "lib/tdb/include/tdb.h"
49 #include "system/time.h"
50 #include "system/shmem.h"
51 #include "system/filesys.h"
54 #define TDB_MAGIC_FOOD "TDB file\n"
55 #define TDB_VERSION (0x26011967 + 6)
56 #define TDB_MAGIC (0x26011999U)
57 #define TDB_FREE_MAGIC (~TDB_MAGIC)
58 #define TDB_DEAD_MAGIC (0xFEE1DEAD)
59 #define TDB_ALIGNMENT 4
60 #define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
61 #define DEFAULT_HASH_SIZE 131
62 #define TDB_PAGE_SIZE 0x2000
63 #define FREELIST_TOP (sizeof(struct tdb_header))
64 #define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
65 #define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
66 #define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
67 #define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
68 #define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off))
69 #define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + TDB_SPINLOCK_SIZE(hash_size))
72 /* NB assumes there is a local variable called "tdb" that is the
73 * current context, also takes doubly-parenthesized print-style
75 #define TDB_LOG(x) tdb->log_fn x
86 #define MAP_FAILED ((void *)-1)
89 #ifndef discard_const_p
90 # if defined(__intptr_t_defined) || defined(HAVE_INTPTR_T)
91 # define discard_const(ptr) ((void *)((intptr_t)(ptr)))
93 # define discard_const(ptr) ((void *)(ptr))
95 # define discard_const_p(type, ptr) ((type *)discard_const(ptr))
98 /* free memory if the pointer is valid and zero the pointer */
100 #define SAFE_FREE(x) do { if ((x) != NULL) {free(discard_const_p(void *, (x))); (x)=NULL;} } while(0)
103 #define BUCKET(hash) ((hash) % tdb->header.hash_size)
106 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
107 static TDB_CONTEXT *tdbs = NULL;
109 static int tdb_munmap(TDB_CONTEXT *tdb)
111 if (tdb->flags & TDB_INTERNAL)
116 int ret = munmap(tdb->map_ptr, tdb->map_size);
125 static void tdb_mmap(TDB_CONTEXT *tdb)
127 if (tdb->flags & TDB_INTERNAL)
131 if (!(tdb->flags & TDB_NOMMAP)) {
132 tdb->map_ptr = mmap(NULL, tdb->map_size,
133 PROT_READ|(tdb->read_only? 0:PROT_WRITE),
134 MAP_SHARED|MAP_FILE, tdb->fd, 0);
137 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
140 if (tdb->map_ptr == MAP_FAILED) {
142 TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n",
143 tdb->map_size, strerror(errno)));
153 /* Endian conversion: we only ever deal with 4 byte quantities */
154 static void *convert(void *buf, u32 size)
157 for (i = 0; i < size / 4; i++)
158 p[i] = TDB_BYTEREV(p[i]);
161 #define DOCONV() (tdb->flags & TDB_CONVERT)
162 #define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x)
164 /* the body of the database is made of one list_struct for the free space
165 plus a separate data list for each hash value */
167 tdb_off next; /* offset of the next record in the list */
168 tdb_len rec_len; /* total byte length of record */
169 tdb_len key_len; /* byte length of key */
170 tdb_len data_len; /* byte length of data */
171 u32 full_hash; /* the full 32 bit hash of the key */
172 u32 magic; /* try to catch errors */
173 /* the following union is implied:
175 char record[rec_len];
180 u32 totalsize; (tailer)
185 /* a byte range locking function - return 0 on success
186 this functions locks/unlocks 1 byte at the specified offset.
188 On error, errno is also set so that errors are passed back properly
189 through tdb_open(). */
190 static int tdb_brlock(TDB_CONTEXT *tdb, tdb_off offset,
191 int rw_type, int lck_type, int probe)
196 if (tdb->flags & TDB_NOLOCK)
198 if ((rw_type == F_WRLCK) && (tdb->read_only)) {
204 fl.l_whence = SEEK_SET;
210 ret = fcntl(tdb->fd,lck_type,&fl);
211 } while (ret == -1 && errno == EINTR);
214 if (!probe && lck_type != F_SETLK) {
215 /* Ensure error code is set for log fun to examine. */
216 tdb->ecode = TDB_ERR_LOCK;
217 TDB_LOG((tdb, 5,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d\n",
218 tdb->fd, offset, rw_type, lck_type));
220 /* Generic lock error. errno set by fcntl.
221 * EAGAIN is an expected return from non-blocking
223 if (errno != EAGAIN) {
224 TDB_LOG((tdb, 5, "tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d: %s\n",
225 tdb->fd, offset, rw_type, lck_type,
228 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
233 /* lock a list in the database. list -1 is the alloc list */
234 static int tdb_lock(TDB_CONTEXT *tdb, int list, int ltype)
236 if (list < -1 || list >= (int)tdb->header.hash_size) {
237 TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n",
241 if (tdb->flags & TDB_NOLOCK)
244 /* Since fcntl locks don't nest, we do a lock for the first one,
245 and simply bump the count for future ones */
246 if (tdb->locked[list+1].count == 0) {
247 if (!tdb->read_only && tdb->header.rwlocks) {
248 if (tdb_spinlock(tdb, list, ltype)) {
249 TDB_LOG((tdb, 0, "tdb_lock spinlock failed on list %d ltype=%d\n",
253 } else if (tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
254 TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d (%s)\n",
255 list, ltype, strerror(errno)));
258 tdb->locked[list+1].ltype = ltype;
260 tdb->locked[list+1].count++;
264 /* unlock the database: returns void because it's too late for errors. */
265 /* changed to return int it may be interesting to know there
266 has been an error --simo */
267 static int tdb_unlock(TDB_CONTEXT *tdb, int list, int ltype)
271 if (tdb->flags & TDB_NOLOCK)
275 if (list < -1 || list >= (int)tdb->header.hash_size) {
276 TDB_LOG((tdb, 0, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
280 if (tdb->locked[list+1].count==0) {
281 TDB_LOG((tdb, 0, "tdb_unlock: count is 0\n"));
285 if (tdb->locked[list+1].count == 1) {
286 /* Down to last nested lock: unlock underneath */
287 if (!tdb->read_only && tdb->header.rwlocks) {
288 ret = tdb_spinunlock(tdb, list, ltype);
290 ret = tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0);
295 tdb->locked[list+1].count--;
298 TDB_LOG((tdb, 0,"tdb_unlock: An error occurred unlocking!\n"));
302 /* This is based on the hash algorithm from gdbm */
303 static u32 default_tdb_hash(TDB_DATA *key)
305 u32 value; /* Used to compute the hash value. */
306 u32 i; /* Used to cycle through random values. */
308 /* Set the initial value from the key size. */
309 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
310 value = (value + (key->dptr[i] << (i*5 % 24)));
312 return (1103515243 * value + 12345);
315 /* check for an out of bounds access - if it is out of bounds then
316 see if the database has been expanded by someone else and expand
318 note that "len" is the minimum length needed for the db
320 static int tdb_oob(TDB_CONTEXT *tdb, tdb_off len, int probe)
323 if (len <= tdb->map_size)
325 if (tdb->flags & TDB_INTERNAL) {
327 /* Ensure ecode is set for log fn. */
328 tdb->ecode = TDB_ERR_IO;
329 TDB_LOG((tdb, 0,"tdb_oob len %d beyond internal malloc size %d\n",
330 (int)len, (int)tdb->map_size));
332 return TDB_ERRCODE(TDB_ERR_IO, -1);
335 if (fstat(tdb->fd, &st) == -1)
336 return TDB_ERRCODE(TDB_ERR_IO, -1);
338 if (st.st_size < (size_t)len) {
340 /* Ensure ecode is set for log fn. */
341 tdb->ecode = TDB_ERR_IO;
342 TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n",
343 (int)len, (int)st.st_size));
345 return TDB_ERRCODE(TDB_ERR_IO, -1);
348 /* Unmap, update size, remap */
349 if (tdb_munmap(tdb) == -1)
350 return TDB_ERRCODE(TDB_ERR_IO, -1);
351 tdb->map_size = st.st_size;
356 /* write a lump of data at a specified offset */
357 static int tdb_write(TDB_CONTEXT *tdb, tdb_off off, void *buf, tdb_len len)
359 if (tdb_oob(tdb, off + len, 0) != 0)
363 memcpy(off + (char *)tdb->map_ptr, buf, len);
365 else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) {
367 else if (lseek(tdb->fd, off, SEEK_SET) != off
368 || write(tdb->fd, buf, len) != (ssize_t)len) {
370 /* Ensure ecode is set for log fn. */
371 tdb->ecode = TDB_ERR_IO;
372 TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n",
373 off, len, strerror(errno)));
374 return TDB_ERRCODE(TDB_ERR_IO, -1);
379 /* read a lump of data at a specified offset, maybe convert */
380 static int tdb_read(TDB_CONTEXT *tdb,tdb_off off,void *buf,tdb_len len,int cv)
382 if (tdb_oob(tdb, off + len, 0) != 0)
386 memcpy(buf, off + (char *)tdb->map_ptr, len);
388 else if (pread(tdb->fd, buf, len, off) != (ssize_t)len) {
390 else if (lseek(tdb->fd, off, SEEK_SET) != off
391 || read(tdb->fd, buf, len) != (ssize_t)len) {
393 /* Ensure ecode is set for log fn. */
394 tdb->ecode = TDB_ERR_IO;
395 TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d (%s)\n",
396 off, len, strerror(errno)));
397 return TDB_ERRCODE(TDB_ERR_IO, -1);
404 /* read a lump of data, allocating the space for it */
405 static char *tdb_alloc_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_len len)
409 if (!(buf = malloc(len))) {
410 /* Ensure ecode is set for log fn. */
411 tdb->ecode = TDB_ERR_OOM;
412 TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
413 len, strerror(errno)));
414 return TDB_ERRCODE(TDB_ERR_OOM, buf);
416 if (tdb_read(tdb, offset, buf, len, 0) == -1) {
423 /* read/write a tdb_off */
424 static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
426 return tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
428 static int ofs_write(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
431 return tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
434 /* read/write a record */
435 static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
437 if (tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
439 if (TDB_BAD_MAGIC(rec)) {
440 /* Ensure ecode is set for log fn. */
441 tdb->ecode = TDB_ERR_CORRUPT;
442 TDB_LOG((tdb, 0,"rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
443 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
445 return tdb_oob(tdb, rec->next+sizeof(*rec), 0);
447 static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
449 struct list_struct r = *rec;
450 return tdb_write(tdb, offset, CONVERT(r), sizeof(r));
453 /* read a freelist record and check for simple errors */
454 static int rec_free_read(TDB_CONTEXT *tdb, tdb_off off, struct list_struct *rec)
456 if (tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
459 if (rec->magic == TDB_MAGIC) {
460 /* this happens when a app is showdown while deleting a record - we should
461 not completely fail when this happens */
462 TDB_LOG((tdb, 0,"rec_free_read non-free magic 0x%x at offset=%d - fixing\n",
464 rec->magic = TDB_FREE_MAGIC;
465 if (tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
469 if (rec->magic != TDB_FREE_MAGIC) {
470 /* Ensure ecode is set for log fn. */
471 tdb->ecode = TDB_ERR_CORRUPT;
472 TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n",
474 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
476 if (tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
481 /* update a record tailer (must hold allocation lock) */
482 static int update_tailer(TDB_CONTEXT *tdb, tdb_off offset,
483 const struct list_struct *rec)
487 /* Offset of tailer from record header */
488 totalsize = sizeof(*rec) + rec->rec_len;
489 return ofs_write(tdb, offset + totalsize - sizeof(tdb_off),
493 static tdb_off tdb_dump_record(TDB_CONTEXT *tdb, tdb_off offset)
495 struct list_struct rec;
496 tdb_off tailer_ofs, tailer;
498 if (tdb_read(tdb, offset, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
499 printf("ERROR: failed to read record at %u\n", offset);
503 printf(" rec: offset=0x%08x next=0x%08x rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
504 offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, rec.full_hash, rec.magic);
506 tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off);
507 if (ofs_read(tdb, tailer_ofs, &tailer) == -1) {
508 printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
512 if (tailer != rec.rec_len + sizeof(rec)) {
513 printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
514 (unsigned int)tailer, (unsigned int)(rec.rec_len + sizeof(rec)));
519 static int tdb_dump_chain(TDB_CONTEXT *tdb, int i)
521 tdb_off rec_ptr, top;
523 top = TDB_HASH_TOP(i);
525 if (tdb_lock(tdb, i, F_WRLCK) != 0)
528 if (ofs_read(tdb, top, &rec_ptr) == -1)
529 return tdb_unlock(tdb, i, F_WRLCK);
532 printf("hash=%d\n", i);
535 rec_ptr = tdb_dump_record(tdb, rec_ptr);
538 return tdb_unlock(tdb, i, F_WRLCK);
541 void tdb_dump_all(TDB_CONTEXT *tdb)
544 for (i=0;i<tdb->header.hash_size;i++) {
545 tdb_dump_chain(tdb, i);
547 printf("freelist:\n");
548 tdb_dump_chain(tdb, -1);
551 int tdb_printfreelist(TDB_CONTEXT *tdb)
555 tdb_off offset, rec_ptr;
556 struct list_struct rec;
558 if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
561 offset = FREELIST_TOP;
563 /* read in the freelist top */
564 if (ofs_read(tdb, offset, &rec_ptr) == -1) {
565 tdb_unlock(tdb, -1, F_WRLCK);
569 printf("freelist top=[0x%08x]\n", rec_ptr );
571 if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
572 tdb_unlock(tdb, -1, F_WRLCK);
576 if (rec.magic != TDB_FREE_MAGIC) {
577 printf("bad magic 0x%08x in free list\n", rec.magic);
578 tdb_unlock(tdb, -1, F_WRLCK);
582 printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)] (end = 0x%08x)\n",
583 rec_ptr, rec.rec_len, rec.rec_len, rec_ptr + rec.rec_len);
584 total_free += rec.rec_len;
586 /* move to the next record */
589 printf("total rec_len = [0x%08x (%d)]\n", (int)total_free,
592 return tdb_unlock(tdb, -1, F_WRLCK);
595 /* Remove an element from the freelist. Must have alloc lock. */
596 static int remove_from_freelist(TDB_CONTEXT *tdb, tdb_off off, tdb_off next)
600 /* read in the freelist top */
601 last_ptr = FREELIST_TOP;
602 while (ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
604 /* We've found it! */
605 return ofs_write(tdb, last_ptr, &next);
607 /* Follow chain (next offset is at start of record) */
610 TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off));
611 return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
614 /* Add an element into the freelist. Merge adjacent records if
616 static int tdb_free(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
620 /* Allocation and tailer lock */
621 if (tdb_lock(tdb, -1, F_WRLCK) != 0)
624 /* set an initial tailer, so if we fail we don't leave a bogus record */
625 if (update_tailer(tdb, offset, rec) != 0) {
626 TDB_LOG((tdb, 0, "tdb_free: upfate_tailer failed!\n"));
630 /* Look right first (I'm an Australian, dammit) */
631 right = offset + sizeof(*rec) + rec->rec_len;
632 if (right + sizeof(*rec) <= tdb->map_size) {
633 struct list_struct r;
635 if (tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
636 TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", right));
640 /* If it's free, expand to include it. */
641 if (r.magic == TDB_FREE_MAGIC) {
642 if (remove_from_freelist(tdb, right, r.next) == -1) {
643 TDB_LOG((tdb, 0, "tdb_free: right free failed at %u\n", right));
646 rec->rec_len += sizeof(r) + r.rec_len;
652 left = offset - sizeof(tdb_off);
653 if (left > TDB_DATA_START(tdb->header.hash_size)) {
654 struct list_struct l;
657 /* Read in tailer and jump back to header */
658 if (ofs_read(tdb, left, &leftsize) == -1) {
659 TDB_LOG((tdb, 0, "tdb_free: left offset read failed at %u\n", left));
662 left = offset - leftsize;
664 /* Now read in record */
665 if (tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
666 TDB_LOG((tdb, 0, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
670 /* If it's free, expand to include it. */
671 if (l.magic == TDB_FREE_MAGIC) {
672 if (remove_from_freelist(tdb, left, l.next) == -1) {
673 TDB_LOG((tdb, 0, "tdb_free: left free failed at %u\n", left));
677 rec->rec_len += leftsize;
683 if (update_tailer(tdb, offset, rec) == -1) {
684 TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", offset));
688 /* Now, prepend to free list */
689 rec->magic = TDB_FREE_MAGIC;
691 if (ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
692 rec_write(tdb, offset, rec) == -1 ||
693 ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
694 TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", offset));
698 /* And we're done. */
699 tdb_unlock(tdb, -1, F_WRLCK);
703 tdb_unlock(tdb, -1, F_WRLCK);
708 /* expand a file. we prefer to use ftruncate, as that is what posix
709 says to use for mmap expansion */
710 static int expand_file(TDB_CONTEXT *tdb, tdb_off size, tdb_off addition)
713 #if HAVE_FTRUNCATE_EXTEND
714 if (ftruncate(tdb->fd, size+addition) != 0) {
715 TDB_LOG((tdb, 0, "expand_file ftruncate to %d failed (%s)\n",
716 size+addition, strerror(errno)));
723 if (pwrite(tdb->fd, &b, 1, (size+addition) - 1) != 1) {
725 if (lseek(tdb->fd, (size+addition) - 1, SEEK_SET) != (size+addition) - 1 ||
726 write(tdb->fd, &b, 1) != 1) {
728 TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n",
729 size+addition, strerror(errno)));
734 /* now fill the file with something. This ensures that the file isn't sparse, which would be
735 very bad if we ran out of disk. This must be done with write, not via mmap */
736 memset(buf, 0x42, sizeof(buf));
738 int n = addition>sizeof(buf)?sizeof(buf):addition;
740 int ret = pwrite(tdb->fd, buf, n, size);
743 if (lseek(tdb->fd, size, SEEK_SET) != size)
745 ret = write(tdb->fd, buf, n);
748 TDB_LOG((tdb, 0, "expand_file write of %d failed (%s)\n",
749 n, strerror(errno)));
759 /* expand the database at least size bytes by expanding the underlying
760 file and doing the mmap again if necessary */
761 static int tdb_expand(TDB_CONTEXT *tdb, tdb_off size)
763 struct list_struct rec;
766 if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
767 TDB_LOG((tdb, 0, "lock failed in tdb_expand\n"));
771 /* must know about any previous expansions by another process */
772 tdb_oob(tdb, tdb->map_size + 1, 1);
774 /* always make room for at least 10 more records, and round
775 the database up to a multiple of TDB_PAGE_SIZE */
776 size = TDB_ALIGN(tdb->map_size + size*10, TDB_PAGE_SIZE) - tdb->map_size;
778 if (!(tdb->flags & TDB_INTERNAL))
782 * We must ensure the file is unmapped before doing this
783 * to ensure consistency with systems like OpenBSD where
784 * writes and mmaps are not consistent.
787 /* expand the file itself */
788 if (!(tdb->flags & TDB_INTERNAL)) {
789 if (expand_file(tdb, tdb->map_size, size) != 0)
793 tdb->map_size += size;
795 if (tdb->flags & TDB_INTERNAL) {
796 char *new_map_ptr = realloc(tdb->map_ptr, tdb->map_size);
798 tdb->map_size -= size;
801 tdb->map_ptr = new_map_ptr;
804 * We must ensure the file is remapped before adding the space
805 * to ensure consistency with systems like OpenBSD where
806 * writes and mmaps are not consistent.
809 /* We're ok if the mmap fails as we'll fallback to read/write */
813 /* form a new freelist record */
814 memset(&rec,'\0',sizeof(rec));
815 rec.rec_len = size - sizeof(rec);
817 /* link it into the free list */
818 offset = tdb->map_size - size;
819 if (tdb_free(tdb, offset, &rec) == -1)
822 tdb_unlock(tdb, -1, F_WRLCK);
825 tdb_unlock(tdb, -1, F_WRLCK);
831 the core of tdb_allocate - called when we have decided which
832 free list entry to use
834 static tdb_off tdb_allocate_ofs(TDB_CONTEXT *tdb, tdb_len length, tdb_off rec_ptr,
835 struct list_struct *rec, tdb_off last_ptr)
837 struct list_struct newrec;
840 memset(&newrec, '\0', sizeof(newrec));
842 /* found it - now possibly split it up */
843 if (rec->rec_len > length + MIN_REC_SIZE) {
844 /* Length of left piece */
845 length = TDB_ALIGN(length, TDB_ALIGNMENT);
847 /* Right piece to go on free list */
848 newrec.rec_len = rec->rec_len - (sizeof(*rec) + length);
849 newrec_ptr = rec_ptr + sizeof(*rec) + length;
851 /* And left record is shortened */
852 rec->rec_len = length;
857 /* Remove allocated record from the free list */
858 if (ofs_write(tdb, last_ptr, &rec->next) == -1) {
862 /* Update header: do this before we drop alloc
863 lock, otherwise tdb_free() might try to
864 merge with us, thinking we're free.
865 (Thanks Jeremy Allison). */
866 rec->magic = TDB_MAGIC;
867 if (rec_write(tdb, rec_ptr, rec) == -1) {
871 /* Did we create new block? */
873 /* Update allocated record tailer (we
875 if (update_tailer(tdb, rec_ptr, rec) == -1) {
879 /* Free new record */
880 if (tdb_free(tdb, newrec_ptr, &newrec) == -1) {
885 /* all done - return the new record offset */
889 /* allocate some space from the free list. The offset returned points
890 to a unconnected list_struct within the database with room for at
891 least length bytes of total data
893 0 is returned if the space could not be allocated
895 static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length,
896 struct list_struct *rec)
898 tdb_off rec_ptr, last_ptr, newrec_ptr;
900 tdb_off rec_ptr, last_ptr;
904 if (tdb_lock(tdb, -1, F_WRLCK) == -1)
907 /* Extra bytes required for tailer */
908 length += sizeof(tdb_off);
911 last_ptr = FREELIST_TOP;
913 /* read in the freelist top */
914 if (ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
920 this is a best fit allocation strategy. Originally we used
921 a first fit strategy, but it suffered from massive fragmentation
922 issues when faced with a slowly increasing record size.
925 if (rec_free_read(tdb, rec_ptr, rec) == -1) {
929 if (rec->rec_len >= length) {
930 if (bestfit.rec_ptr == 0 ||
931 rec->rec_len < bestfit.rec_len) {
932 bestfit.rec_len = rec->rec_len;
933 bestfit.rec_ptr = rec_ptr;
934 bestfit.last_ptr = last_ptr;
935 /* consider a fit to be good enough if we aren't wasting more than half the space */
936 if (bestfit.rec_len < 2*length) {
942 /* move to the next record */
947 if (bestfit.rec_ptr != 0) {
948 if (rec_free_read(tdb, bestfit.rec_ptr, rec) == -1) {
952 newrec_ptr = tdb_allocate_ofs(tdb, length, bestfit.rec_ptr, rec, bestfit.last_ptr);
953 tdb_unlock(tdb, -1, F_WRLCK);
957 /* we didn't find enough space. See if we can expand the
958 database and if we can then try again */
959 if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
962 tdb_unlock(tdb, -1, F_WRLCK);
966 /* initialise a new database with a specified hash size */
967 static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size)
969 struct tdb_header *newdb;
972 /* We make it up in memory, then write it out if not internal */
973 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off);
974 if (!(newdb = calloc(size, 1)))
975 return TDB_ERRCODE(TDB_ERR_OOM, -1);
977 /* Fill in the header */
978 newdb->version = TDB_VERSION;
979 newdb->hash_size = hash_size;
981 newdb->rwlocks = size;
983 if (tdb->flags & TDB_INTERNAL) {
984 tdb->map_size = size;
985 tdb->map_ptr = (char *)newdb;
986 memcpy(&tdb->header, newdb, sizeof(tdb->header));
987 /* Convert the `ondisk' version if asked. */
991 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
994 if (ftruncate(tdb->fd, 0) == -1)
997 /* This creates an endian-converted header, as if read from disk */
999 memcpy(&tdb->header, newdb, sizeof(tdb->header));
1000 /* Don't endian-convert the magic food! */
1001 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
1002 if (write(tdb->fd, newdb, size) != size)
1005 ret = tdb_create_rwlocks(tdb->fd, hash_size);
1012 /* Returns 0 on fail. On success, return offset of record, and fills
1014 static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash,
1015 struct list_struct *r)
1019 /* read in the hash top */
1020 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
1023 /* keep looking until we find the right record */
1025 if (rec_read(tdb, rec_ptr, r) == -1)
1028 if (!TDB_DEAD(r) && hash==r->full_hash && key.dsize==r->key_len) {
1030 /* a very likely hit - read the key */
1031 k = tdb_alloc_read(tdb, rec_ptr + sizeof(*r),
1036 if (memcmp(key.dptr, k, key.dsize) == 0) {
1044 return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
1047 /* As tdb_find, but if you succeed, keep the lock */
1048 static tdb_off tdb_find_lock_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, int locktype,
1049 struct list_struct *rec)
1053 if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
1055 if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
1056 tdb_unlock(tdb, BUCKET(hash), locktype);
1060 enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb)
1065 static struct tdb_errname {
1066 enum TDB_ERROR ecode; const char *estring;
1067 } emap[] = { {TDB_SUCCESS, "Success"},
1068 {TDB_ERR_CORRUPT, "Corrupt database"},
1069 {TDB_ERR_IO, "IO Error"},
1070 {TDB_ERR_LOCK, "Locking error"},
1071 {TDB_ERR_OOM, "Out of memory"},
1072 {TDB_ERR_EXISTS, "Record exists"},
1073 {TDB_ERR_NOLOCK, "Lock exists on other keys"},
1074 {TDB_ERR_NOEXIST, "Record does not exist"} };
1076 /* Error string for the last tdb error */
1077 const char *tdb_errorstr(TDB_CONTEXT *tdb)
1080 for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
1081 if (tdb->ecode == emap[i].ecode)
1082 return emap[i].estring;
1083 return "Invalid error code";
1086 /* update an entry in place - this only works if the new data size
1087 is <= the old data size and the key exists.
1088 on failure return -1.
1091 static int tdb_update_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA dbuf)
1093 struct list_struct rec;
1097 if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
1100 /* must be long enough key, data and tailer */
1101 if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off)) {
1102 tdb->ecode = TDB_SUCCESS; /* Not really an error */
1106 if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
1107 dbuf.dptr, dbuf.dsize) == -1)
1110 if (dbuf.dsize != rec.data_len) {
1112 rec.data_len = dbuf.dsize;
1113 return rec_write(tdb, rec_ptr, &rec);
1119 /* find an entry in the database given a key */
1120 /* If an entry doesn't exist tdb_err will be set to
1121 * TDB_ERR_NOEXIST. If a key has no data attached
1122 * then the TDB_DATA will have zero length but
1123 * a non-zero pointer
1126 TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
1129 struct list_struct rec;
1133 /* find which hash bucket it is in */
1134 hash = tdb->hash_fn(&key);
1135 if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
1138 ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
1140 ret.dsize = rec.data_len;
1141 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
1145 /* check if an entry in the database exists
1147 note that 1 is returned if the key is found and 0 is returned if not found
1148 this doesn't match the conventions in the rest of this module, but is
1149 compatible with gdbm
1151 static int tdb_exists_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash)
1153 struct list_struct rec;
1155 if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0)
1157 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
1161 int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key)
1163 u32 hash = tdb->hash_fn(&key);
1164 return tdb_exists_hash(tdb, key, hash);
1167 /* record lock stops delete underneath */
1168 static int lock_record(TDB_CONTEXT *tdb, tdb_off off)
1170 return off ? tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
1173 Write locks override our own fcntl readlocks, so check it here.
1174 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
1175 an error to fail to get the lock here.
1178 static int write_lock_record(TDB_CONTEXT *tdb, tdb_off off)
1180 struct tdb_traverse_lock *i;
1181 for (i = &tdb->travlocks; i; i = i->next)
1184 return tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
1188 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
1189 an error to fail to get the lock here.
1192 static int write_unlock_record(TDB_CONTEXT *tdb, tdb_off off)
1194 return tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
1196 /* fcntl locks don't stack: avoid unlocking someone else's */
1197 static int unlock_record(TDB_CONTEXT *tdb, tdb_off off)
1199 struct tdb_traverse_lock *i;
1204 for (i = &tdb->travlocks; i; i = i->next)
1207 return (count == 1 ? tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
1210 /* actually delete an entry in the database given the offset */
1211 static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec)
1213 tdb_off last_ptr, i;
1214 struct list_struct lastrec;
1216 if (tdb->read_only) return -1;
1218 if (write_lock_record(tdb, rec_ptr) == -1) {
1219 /* Someone traversing here: mark it as dead */
1220 rec->magic = TDB_DEAD_MAGIC;
1221 return rec_write(tdb, rec_ptr, rec);
1223 if (write_unlock_record(tdb, rec_ptr) != 0)
1226 /* find previous record in hash chain */
1227 if (ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
1229 for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
1230 if (rec_read(tdb, i, &lastrec) == -1)
1233 /* unlink it: next ptr is at start of record. */
1235 last_ptr = TDB_HASH_TOP(rec->full_hash);
1236 if (ofs_write(tdb, last_ptr, &rec->next) == -1)
1239 /* recover the space */
1240 if (tdb_free(tdb, rec_ptr, rec) == -1)
1245 /* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
1246 static int tdb_next_lock(TDB_CONTEXT *tdb, struct tdb_traverse_lock *tlock,
1247 struct list_struct *rec)
1249 int want_next = (tlock->off != 0);
1251 /* Lock each chain from the start one. */
1252 for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
1254 /* this is an optimisation for the common case where
1255 the hash chain is empty, which is particularly
1256 common for the use of tdb with ldb, where large
1257 hashes are used. In that case we spend most of our
1258 time in tdb_brlock(), locking empty hash chains.
1260 To avoid this, we do an unlocked pre-check to see
1261 if the hash chain is empty before starting to look
1262 inside it. If it is empty then we can avoid that
1263 hash chain. If it isn't empty then we can't believe
1264 the value we get back, as we read it without a
1265 lock, so instead we get the lock and re-fetch the
1268 Notice that not doing this optimisation on the
1269 first hash chain is critical. We must guarantee
1270 that we have done at least one fcntl lock at the
1271 start of a search to guarantee that memory is
1272 coherent on SMP systems. If records are added by
1273 others during the search then thats OK, and we
1274 could possibly miss those with this trick, but we
1275 could miss them anyway without this trick, so the
1276 semantics don't change.
1278 With a non-indexed ldb search this trick gains us a
1279 factor of more than 10 in speed on a linux 2.6.x
1282 if (!tlock->off && tlock->hash != 0) {
1284 if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash), &off) == 0 &&
1290 if (tdb_lock(tdb, tlock->hash, F_WRLCK) == -1)
1293 /* No previous record? Start at top of chain. */
1295 if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
1299 /* Otherwise unlock the previous record. */
1300 if (unlock_record(tdb, tlock->off) != 0)
1305 /* We have offset of old record: grab next */
1306 if (rec_read(tdb, tlock->off, rec) == -1)
1308 tlock->off = rec->next;
1311 /* Iterate through chain */
1312 while( tlock->off) {
1314 if (rec_read(tdb, tlock->off, rec) == -1)
1317 /* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
1318 if (tlock->off == rec->next) {
1319 TDB_LOG((tdb, 0, "tdb_next_lock: loop detected.\n"));
1323 if (!TDB_DEAD(rec)) {
1324 /* Woohoo: we found one! */
1325 if (lock_record(tdb, tlock->off) != 0)
1330 /* Try to clean dead ones from old traverses */
1331 current = tlock->off;
1332 tlock->off = rec->next;
1333 if (!tdb->read_only &&
1334 do_delete(tdb, current, rec) != 0)
1337 tdb_unlock(tdb, tlock->hash, F_WRLCK);
1340 /* We finished iteration without finding anything */
1341 return TDB_ERRCODE(TDB_SUCCESS, 0);
1345 if (tdb_unlock(tdb, tlock->hash, F_WRLCK) != 0)
1346 TDB_LOG((tdb, 0, "tdb_next_lock: On error unlock failed!\n"));
1350 /* traverse the entire database - calling fn(tdb, key, data) on each element.
1351 return -1 on error or the record count traversed
1352 if fn is NULL then it is not called
1353 a non-zero return value from fn() indicates that the traversal should stop
1355 int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *private)
1358 struct list_struct rec;
1359 struct tdb_traverse_lock tl = { NULL, 0, 0 };
1362 /* This was in the initializaton, above, but the IRIX compiler
1363 * did not like it. crh
1365 tl.next = tdb->travlocks.next;
1367 /* fcntl locks don't stack: beware traverse inside traverse */
1368 tdb->travlocks.next = &tl;
1370 /* tdb_next_lock places locks on the record returned, and its chain */
1371 while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) {
1373 /* now read the full record */
1374 key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec),
1375 rec.key_len + rec.data_len);
1378 if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0)
1380 if (unlock_record(tdb, tl.off) != 0)
1381 TDB_LOG((tdb, 0, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
1384 key.dsize = rec.key_len;
1385 dbuf.dptr = key.dptr + rec.key_len;
1386 dbuf.dsize = rec.data_len;
1388 /* Drop chain lock, call out */
1389 if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0) {
1393 if (fn && fn(tdb, key, dbuf, private)) {
1394 /* They want us to terminate traversal */
1396 if (unlock_record(tdb, tl.off) != 0) {
1397 TDB_LOG((tdb, 0, "tdb_traverse: unlock_record failed!\n"));;
1400 tdb->travlocks.next = tl.next;
1401 SAFE_FREE(key.dptr);
1404 SAFE_FREE(key.dptr);
1407 tdb->travlocks.next = tl.next;
1414 /* find the first entry in the database and return its key */
1415 TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb)
1418 struct list_struct rec;
1420 /* release any old lock */
1421 if (unlock_record(tdb, tdb->travlocks.off) != 0)
1423 tdb->travlocks.off = tdb->travlocks.hash = 0;
1425 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
1427 /* now read the key */
1428 key.dsize = rec.key_len;
1429 key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
1430 if (tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK) != 0)
1431 TDB_LOG((tdb, 0, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
1435 /* find the next entry in the database, returning its key */
1436 TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey)
1439 TDB_DATA key = tdb_null;
1440 struct list_struct rec;
1443 /* Is locked key the old key? If so, traverse will be reliable. */
1444 if (tdb->travlocks.off) {
1445 if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
1447 if (rec_read(tdb, tdb->travlocks.off, &rec) == -1
1448 || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
1450 || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
1451 /* No, it wasn't: unlock it and start from scratch */
1452 if (unlock_record(tdb, tdb->travlocks.off) != 0)
1454 if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
1456 tdb->travlocks.off = 0;
1462 if (!tdb->travlocks.off) {
1463 /* No previous element: do normal find, and lock record */
1464 tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), F_WRLCK, &rec);
1465 if (!tdb->travlocks.off)
1467 tdb->travlocks.hash = BUCKET(rec.full_hash);
1468 if (lock_record(tdb, tdb->travlocks.off) != 0) {
1469 TDB_LOG((tdb, 0, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
1473 oldhash = tdb->travlocks.hash;
1475 /* Grab next record: locks chain and returned record,
1476 unlocks old record */
1477 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
1478 key.dsize = rec.key_len;
1479 key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
1481 /* Unlock the chain of this new record */
1482 if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
1483 TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
1485 /* Unlock the chain of old record */
1486 if (tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK) != 0)
1487 TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
1491 /* delete an entry in the database given a key */
1492 static int tdb_delete_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash)
1495 struct list_struct rec;
1498 if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK, &rec)))
1500 ret = do_delete(tdb, rec_ptr, &rec);
1501 if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
1502 TDB_LOG((tdb, 0, "tdb_delete: WARNING tdb_unlock failed!\n"));
1506 int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
1508 u32 hash = tdb->hash_fn(&key);
1509 return tdb_delete_hash(tdb, key, hash);
1512 /* store an element in the database, replacing any existing element
1515 return 0 on success, -1 on failure
1517 int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
1519 struct list_struct rec;
1525 /* find which hash bucket it is in */
1526 hash = tdb->hash_fn(&key);
1527 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
1530 /* check for it existing, on insert. */
1531 if (flag == TDB_INSERT) {
1532 if (tdb_exists_hash(tdb, key, hash)) {
1533 tdb->ecode = TDB_ERR_EXISTS;
1537 /* first try in-place update, on modify or replace. */
1538 if (tdb_update_hash(tdb, key, hash, dbuf) == 0)
1540 if (tdb->ecode == TDB_ERR_NOEXIST &&
1541 flag == TDB_MODIFY) {
1542 /* if the record doesn't exist and we are in TDB_MODIFY mode then
1543 we should fail the store */
1547 /* reset the error code potentially set by the tdb_update() */
1548 tdb->ecode = TDB_SUCCESS;
1550 /* delete any existing record - if it doesn't exist we don't
1551 care. Doing this first reduces fragmentation, and avoids
1552 coalescing with `allocated' block before it's updated. */
1553 if (flag != TDB_INSERT)
1554 tdb_delete_hash(tdb, key, hash);
1556 /* Copy key+value *before* allocating free space in case malloc
1557 fails and we are left with a dead spot in the tdb. */
1559 if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) {
1560 tdb->ecode = TDB_ERR_OOM;
1564 memcpy(p, key.dptr, key.dsize);
1566 memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
1568 /* we have to allocate some space */
1569 if (!(rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec)))
1572 /* Read hash top into next ptr */
1573 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
1576 rec.key_len = key.dsize;
1577 rec.data_len = dbuf.dsize;
1578 rec.full_hash = hash;
1579 rec.magic = TDB_MAGIC;
1581 /* write out and point the top of the hash chain at it */
1582 if (rec_write(tdb, rec_ptr, &rec) == -1
1583 || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
1584 || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
1585 /* Need to tdb_unallocate() here */
1590 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
1597 /* Attempt to append data to an entry in place - this only works if the new data size
1598 is <= the old data size and the key exists.
1599 on failure return -1. Record must be locked before calling.
1601 static int tdb_append_inplace(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA new_dbuf)
1603 struct list_struct rec;
1607 if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
1610 /* Append of 0 is always ok. */
1611 if (new_dbuf.dsize == 0)
1614 /* must be long enough for key, old data + new data and tailer */
1615 if (rec.rec_len < key.dsize + rec.data_len + new_dbuf.dsize + sizeof(tdb_off)) {
1617 tdb->ecode = TDB_SUCCESS; /* Not really an error */
1621 if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len + rec.data_len,
1622 new_dbuf.dptr, new_dbuf.dsize) == -1)
1626 rec.data_len += new_dbuf.dsize;
1627 return rec_write(tdb, rec_ptr, &rec);
1630 /* Append to an entry. Create if not exist. */
1632 int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf)
1634 struct list_struct rec;
1639 size_t new_data_size = 0;
1641 /* find which hash bucket it is in */
1642 hash = tdb->hash_fn(&key);
1643 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
1646 /* first try in-place. */
1647 if (tdb_append_inplace(tdb, key, hash, new_dbuf) == 0)
1650 /* reset the error code potentially set by the tdb_append_inplace() */
1651 tdb->ecode = TDB_SUCCESS;
1654 if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
1655 if (tdb->ecode != TDB_ERR_NOEXIST)
1658 /* Not found - create. */
1660 ret = tdb_store(tdb, key, new_dbuf, TDB_INSERT);
1664 new_data_size = rec.data_len + new_dbuf.dsize;
1666 /* Copy key+old_value+value *before* allocating free space in case malloc
1667 fails and we are left with a dead spot in the tdb. */
1669 if (!(p = (char *)malloc(key.dsize + new_data_size))) {
1670 tdb->ecode = TDB_ERR_OOM;
1674 /* Copy the key in place. */
1675 memcpy(p, key.dptr, key.dsize);
1677 /* Now read the old data into place. */
1679 tdb_read(tdb, rec_ptr + sizeof(rec) + rec.key_len, p + key.dsize, rec.data_len, 0) == -1)
1682 /* Finally append the new data. */
1684 memcpy(p+key.dsize+rec.data_len, new_dbuf.dptr, new_dbuf.dsize);
1686 /* delete any existing record - if it doesn't exist we don't
1687 care. Doing this first reduces fragmentation, and avoids
1688 coalescing with `allocated' block before it's updated. */
1690 tdb_delete_hash(tdb, key, hash);
1692 if (!(rec_ptr = tdb_allocate(tdb, key.dsize + new_data_size, &rec)))
1695 /* Read hash top into next ptr */
1696 if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
1699 rec.key_len = key.dsize;
1700 rec.data_len = new_data_size;
1701 rec.full_hash = hash;
1702 rec.magic = TDB_MAGIC;
1704 /* write out and point the top of the hash chain at it */
1705 if (rec_write(tdb, rec_ptr, &rec) == -1
1706 || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+new_data_size)==-1
1707 || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
1708 /* Need to tdb_unallocate() here */
1714 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
1722 static int tdb_already_open(dev_t device,
1727 for (i = tdbs; i; i = i->next) {
1728 if (i->device == device && i->inode == ino) {
1736 /* open the database, creating it if necessary
1738 The open_flags and mode are passed straight to the open call on the
1739 database file. A flags value of O_WRONLY is invalid. The hash size
1740 is advisory, use zero for a default value.
1742 Return is NULL on error, in which case errno is also set. Don't
1743 try to call tdb_error or tdb_errname, just do strerror(errno).
1745 @param name may be NULL for internal databases. */
1746 TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
1747 int open_flags, mode_t mode)
1749 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
1752 /* a default logging function */
1753 static void null_log_fn(TDB_CONTEXT *tdb, int level, const char *fmt, ...)
1758 TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
1759 int open_flags, mode_t mode,
1760 tdb_log_func log_fn,
1761 tdb_hash_func hash_fn)
1765 int rev = 0, locked = 0;
1769 if (!(tdb = calloc(1, sizeof *tdb))) {
1770 /* Can't log this */
1776 tdb->map_ptr = NULL;
1777 tdb->flags = tdb_flags;
1778 tdb->open_flags = open_flags;
1779 tdb->log_fn = log_fn?log_fn:null_log_fn;
1780 tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
1782 if ((open_flags & O_ACCMODE) == O_WRONLY) {
1783 TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n",
1790 hash_size = DEFAULT_HASH_SIZE;
1791 if ((open_flags & O_ACCMODE) == O_RDONLY) {
1793 /* read only databases don't do locking or clear if first */
1794 tdb->flags |= TDB_NOLOCK;
1795 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
1798 /* internal databases don't mmap or lock, and start off cleared */
1799 if (tdb->flags & TDB_INTERNAL) {
1800 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
1801 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
1802 if (tdb_new_database(tdb, hash_size) != 0) {
1803 TDB_LOG((tdb, 0, "tdb_open_ex: tdb_new_database failed!"));
1809 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
1810 TDB_LOG((tdb, 5, "tdb_open_ex: could not open file %s: %s\n",
1811 name, strerror(errno)));
1812 goto fail; /* errno set by open(2) */
1815 /* ensure there is only one process initialising at once */
1816 if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
1817 TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: %s\n",
1818 name, strerror(errno)));
1819 goto fail; /* errno set by tdb_brlock */
1822 /* we need to zero database if we are the only one with it open */
1823 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
1824 (locked = (tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))) {
1825 open_flags |= O_CREAT;
1826 if (ftruncate(tdb->fd, 0) == -1) {
1827 TDB_LOG((tdb, 0, "tdb_open_ex: "
1828 "failed to truncate %s: %s\n",
1829 name, strerror(errno)));
1830 goto fail; /* errno set by ftruncate */
1834 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
1835 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
1836 || (tdb->header.version != TDB_VERSION
1837 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
1838 /* its not a valid database - possibly initialise it */
1839 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
1840 errno = EIO; /* ie bad format or something */
1843 rev = (tdb->flags & TDB_CONVERT);
1845 vp = (uint8_t *)&tdb->header.version;
1846 vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
1847 (((u32)vp[2]) << 8) | (u32)vp[3];
1848 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
1850 tdb->flags &= ~TDB_CONVERT;
1852 tdb->flags |= TDB_CONVERT;
1853 convert(&tdb->header, sizeof(tdb->header));
1855 if (fstat(tdb->fd, &st) == -1)
1858 /* Is it already in the open list? If so, fail. */
1859 if (tdb_already_open(st.st_dev, st.st_ino)) {
1860 TDB_LOG((tdb, 2, "tdb_open_ex: "
1861 "%s (%d,%d) is already open in this process\n",
1862 name, (int)st.st_dev, (int)st.st_ino));
1867 if (!(tdb->name = (char *)strdup(name))) {
1872 tdb->map_size = st.st_size;
1873 tdb->device = st.st_dev;
1874 tdb->inode = st.st_ino;
1875 tdb->locked = calloc(tdb->header.hash_size+1, sizeof(tdb->locked[0]));
1877 TDB_LOG((tdb, 2, "tdb_open_ex: "
1878 "failed to allocate lock structure for %s\n",
1885 if (!tdb->read_only)
1886 if (tdb_clear_spinlocks(tdb) != 0) {
1887 TDB_LOG((tdb, 0, "tdb_open_ex: "
1888 "failed to clear spinlock\n"));
1891 if (tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
1892 TDB_LOG((tdb, 0, "tdb_open_ex: "
1893 "failed to take ACTIVE_LOCK on %s: %s\n",
1894 name, strerror(errno)));
1900 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
1901 we didn't get the initial exclusive lock as we need to let all other
1902 users know we're using it. */
1904 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
1905 /* leave this lock in place to indicate it's in use */
1906 if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
1912 /* Internal (memory-only) databases skip all the code above to
1913 * do with disk files, and resume here by releasing their
1914 * global lock and hooking into the active list. */
1915 if (tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
1922 { int save_errno = errno;
1928 if (tdb->flags & TDB_INTERNAL)
1929 SAFE_FREE(tdb->map_ptr);
1933 SAFE_FREE(tdb->name);
1935 if (close(tdb->fd) != 0)
1936 TDB_LOG((tdb, 5, "tdb_open_ex: failed to close tdb->fd on error!\n"));
1937 SAFE_FREE(tdb->locked);
1947 * @returns -1 for error; 0 for success.
1949 int tdb_close(TDB_CONTEXT *tdb)
1955 if (tdb->flags & TDB_INTERNAL)
1956 SAFE_FREE(tdb->map_ptr);
1960 SAFE_FREE(tdb->name);
1962 ret = close(tdb->fd);
1963 SAFE_FREE(tdb->locked);
1965 /* Remove from contexts list */
1966 for (i = &tdbs; *i; i = &(*i)->next) {
1973 memset(tdb, 0, sizeof(*tdb));
1979 /* lock/unlock entire database */
1980 int tdb_lockall(TDB_CONTEXT *tdb)
1984 /* There are no locks on read-only dbs */
1986 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
1987 for (i = 0; i < tdb->header.hash_size; i++)
1988 if (tdb_lock(tdb, i, F_WRLCK))
1991 /* If error, release locks we have... */
1992 if (i < tdb->header.hash_size) {
1995 for ( j = 0; j < i; j++)
1996 tdb_unlock(tdb, j, F_WRLCK);
1997 return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
2002 void tdb_unlockall(TDB_CONTEXT *tdb)
2005 for (i=0; i < tdb->header.hash_size; i++)
2006 tdb_unlock(tdb, i, F_WRLCK);
2009 /* lock/unlock one hash chain. This is meant to be used to reduce
2010 contention - it cannot guarantee how many records will be locked */
2011 int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key)
2013 return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
2016 int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key)
2018 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
2021 int tdb_chainlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
2023 return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
2026 int tdb_chainunlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
2028 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
2032 /* register a loging function */
2033 void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , const char *, ...))
2035 tdb->log_fn = fn?fn:null_log_fn;
2039 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
2040 seek pointer from our parent and to re-establish locks */
2041 int tdb_reopen(TDB_CONTEXT *tdb)
2045 if (tdb->flags & TDB_INTERNAL)
2046 return 0; /* Nothing to do. */
2047 if (tdb_munmap(tdb) != 0) {
2048 TDB_LOG((tdb, 0, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
2051 if (close(tdb->fd) != 0)
2052 TDB_LOG((tdb, 0, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
2053 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
2054 if (tdb->fd == -1) {
2055 TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", strerror(errno)));
2058 if (fstat(tdb->fd, &st) != 0) {
2059 TDB_LOG((tdb, 0, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
2062 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
2063 TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n"));
2067 if ((tdb->flags & TDB_CLEAR_IF_FIRST) && (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)) {
2068 TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n"));
2079 /* reopen all tdb's */
2080 int tdb_reopen_all(void)
2084 for (tdb=tdbs; tdb; tdb = tdb->next) {
2085 /* Ensure no clear-if-first. */
2086 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
2087 if (tdb_reopen(tdb) != 0)