This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
- version 2 of the License, or (at your option) any later version.
+ version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
- License along with this library; if not, write to the Free Software
- Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
by the header. This removes the need for extra journal files as
used by some other databases
- - dymacially allocated the transaction recover record, re-using it
+ - dynamically allocated the transaction recover record, re-using it
for subsequent transactions. If a larger record is needed then
tdb_free() the old record to place it on the normal tdb freelist
before allocating the new record
struct tdb_transaction {
/* we keep a mirrored copy of the tdb hash heads here so
tdb_next_hash_chain() can operate efficiently */
- u32 *hash_heads;
+ uint32_t *hash_heads;
/* the original io methods - used to do IOs to the real db */
const struct tdb_methods *io_methods;
- /* the list of transaction elements. We use a doubly linked
- list with a last pointer to allow us to keep the list
- ordered, with first element at the front of the list. It
- needs to be doubly linked as the read/write traversals need
- to be backwards, while the commit needs to be forwards */
- struct tdb_transaction_el {
- struct tdb_transaction_el *next, *prev;
- tdb_off_t offset;
- tdb_len_t length;
- unsigned char *data;
- } *elements, *elements_last;
+ /* the list of transaction blocks. When a block is first
+ written to, it gets created in this list */
+ uint8_t **blocks;
+ uint32_t num_blocks;
+ uint32_t block_size; /* bytes in each block */
+ uint32_t last_block_size; /* number of valid bytes in the last block */
/* non-zero when an internal transaction error has
occurred. All write operations will then fail until the
static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
tdb_len_t len, int cv)
{
- struct tdb_transaction_el *el;
-
- /* we need to walk the list backwards to get the most recent data */
- for (el=tdb->transaction->elements_last;el;el=el->prev) {
- tdb_len_t partial;
+ uint32_t blk;
- if (off+len <= el->offset) {
- continue;
- }
- if (off >= el->offset + el->length) {
- continue;
+ /* break it down into block sized ops */
+ while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+ if (transaction_read(tdb, off, buf, len2, cv) != 0) {
+ return -1;
}
+ len -= len2;
+ off += len2;
+ buf = (void *)(len2 + (char *)buf);
+ }
- /* an overlapping read - needs to be split into up to
- 2 reads and a memcpy */
- if (off < el->offset) {
- partial = el->offset - off;
- if (transaction_read(tdb, off, buf, partial, cv) != 0) {
- goto fail;
- }
- len -= partial;
- off += partial;
- buf = (void *)(partial + (char *)buf);
- }
- if (off + len <= el->offset + el->length) {
- partial = len;
- } else {
- partial = el->offset + el->length - off;
- }
- memcpy(buf, el->data + (off - el->offset), partial);
- if (cv) {
- tdb_convert(buf, len);
- }
- len -= partial;
- off += partial;
- buf = (void *)(partial + (char *)buf);
-
- if (len != 0 && transaction_read(tdb, off, buf, len, cv) != 0) {
+ if (len == 0) {
+ return 0;
+ }
+
+ blk = off / tdb->transaction->block_size;
+
+ /* see if we have it in the block list */
+ if (tdb->transaction->num_blocks <= blk ||
+ tdb->transaction->blocks[blk] == NULL) {
+ /* nope, do a real read */
+ if (tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv) != 0) {
goto fail;
}
-
return 0;
}
- /* its not in the transaction elements - do a real read */
- return tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv);
+ /* it is in the block list. Now check for the last block */
+ if (blk == tdb->transaction->num_blocks-1) {
+ if (len > tdb->transaction->last_block_size) {
+ goto fail;
+ }
+ }
+
+ /* now copy it out of this block */
+ memcpy(buf, tdb->transaction->blocks[blk] + (off % tdb->transaction->block_size), len);
+ if (cv) {
+ tdb_convert(buf, len);
+ }
+ return 0;
fail:
- TDB_LOG((tdb, 0, "transaction_read: failed at off=%d len=%d\n", off, len));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_read: failed at off=%d len=%d\n", off, len));
tdb->ecode = TDB_ERR_IO;
tdb->transaction->transaction_error = 1;
return -1;
static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
const void *buf, tdb_len_t len)
{
- struct tdb_transaction_el *el;
+ uint32_t blk;
- if (len == 0) {
- return 0;
- }
-
/* if the write is to a hash head, then update the transaction
hash heads */
if (len == sizeof(tdb_off_t) && off >= FREELIST_TOP &&
off < FREELIST_TOP+TDB_HASHTABLE_SIZE(tdb)) {
- u32 chain = (off-FREELIST_TOP) / sizeof(tdb_off_t);
+ uint32_t chain = (off-FREELIST_TOP) / sizeof(tdb_off_t);
memcpy(&tdb->transaction->hash_heads[chain], buf, len);
}
- /* first see if we can replace an existing entry */
- for (el=tdb->transaction->elements_last;el;el=el->prev) {
- tdb_len_t partial;
+ /* break it up into block sized chunks */
+ while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+ if (transaction_write(tdb, off, buf, len2) != 0) {
+ return -1;
+ }
+ len -= len2;
+ off += len2;
+ if (buf != NULL) {
+ buf = (const void *)(len2 + (const char *)buf);
+ }
+ }
- if (off+len <= el->offset) {
- continue;
+ if (len == 0) {
+ return 0;
+ }
+
+ blk = off / tdb->transaction->block_size;
+ off = off % tdb->transaction->block_size;
+
+ if (tdb->transaction->num_blocks <= blk) {
+ uint8_t **new_blocks;
+ /* expand the blocks array */
+ if (tdb->transaction->blocks == NULL) {
+ new_blocks = malloc((blk+1)*sizeof(uint8_t *));
+ } else {
+ new_blocks = realloc(tdb->transaction->blocks, (blk+1)*sizeof(uint8_t *));
}
- if (off >= el->offset + el->length) {
- continue;
+ if (new_blocks == NULL) {
+ tdb->ecode = TDB_ERR_OOM;
+ goto fail;
}
+ memset(&new_blocks[tdb->transaction->num_blocks], 0,
+ (1+(blk - tdb->transaction->num_blocks))*sizeof(uint8_t *));
+ tdb->transaction->blocks = new_blocks;
+ tdb->transaction->num_blocks = blk+1;
+ tdb->transaction->last_block_size = 0;
+ }
- /* an overlapping write - needs to be split into up to
- 2 writes and a memcpy */
- if (off < el->offset) {
- partial = el->offset - off;
- if (transaction_write(tdb, off, buf, partial) != 0) {
+ /* allocate and fill a block? */
+ if (tdb->transaction->blocks[blk] == NULL) {
+ tdb->transaction->blocks[blk] = (uint8_t *)calloc(tdb->transaction->block_size, 1);
+ if (tdb->transaction->blocks[blk] == NULL) {
+ tdb->ecode = TDB_ERR_OOM;
+ tdb->transaction->transaction_error = 1;
+ return -1;
+ }
+ if (tdb->transaction->old_map_size > blk * tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size;
+ if (len2 + (blk * tdb->transaction->block_size) > tdb->transaction->old_map_size) {
+ len2 = tdb->transaction->old_map_size - (blk * tdb->transaction->block_size);
+ }
+ if (tdb->transaction->io_methods->tdb_read(tdb, blk * tdb->transaction->block_size,
+ tdb->transaction->blocks[blk],
+ len2, 0) != 0) {
+ SAFE_FREE(tdb->transaction->blocks[blk]);
+ tdb->ecode = TDB_ERR_IO;
goto fail;
}
- len -= partial;
- off += partial;
- buf = (const void *)(partial + (const char *)buf);
+ if (blk == tdb->transaction->num_blocks-1) {
+ tdb->transaction->last_block_size = len2;
+ }
}
- if (off + len <= el->offset + el->length) {
- partial = len;
- } else {
- partial = el->offset + el->length - off;
+ }
+
+ /* overwrite part of an existing block */
+ if (buf == NULL) {
+ memset(tdb->transaction->blocks[blk] + off, 0, len);
+ } else {
+ memcpy(tdb->transaction->blocks[blk] + off, buf, len);
+ }
+ if (blk == tdb->transaction->num_blocks-1) {
+ if (len + off > tdb->transaction->last_block_size) {
+ tdb->transaction->last_block_size = len + off;
}
- memcpy(el->data + (off - el->offset), buf, partial);
- len -= partial;
- off += partial;
- buf = (const void *)(partial + (const char *)buf);
-
- if (len != 0 && transaction_write(tdb, off, buf, len) != 0) {
- goto fail;
+ }
+
+ return 0;
+
+fail:
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%d len=%d\n",
+ (blk*tdb->transaction->block_size) + off, len));
+ tdb->transaction->transaction_error = 1;
+ return -1;
+}
+
+
+/*
+ write while in a transaction - this varient never expands the transaction blocks, it only
+ updates existing blocks. This means it cannot change the recovery size
+*/
+static int transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
+ const void *buf, tdb_len_t len)
+{
+ uint32_t blk;
+
+ /* break it up into block sized chunks */
+ while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+ tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+ if (transaction_write_existing(tdb, off, buf, len2) != 0) {
+ return -1;
+ }
+ len -= len2;
+ off += len2;
+ if (buf != NULL) {
+ buf = (const void *)(len2 + (const char *)buf);
}
+ }
+ if (len == 0) {
return 0;
}
- /* add a new entry at the end of the list */
- el = malloc(sizeof(*el));
- if (el == NULL) {
- tdb->ecode = TDB_ERR_OOM;
- tdb->transaction->transaction_error = 1;
- return -1;
- }
- el->next = NULL;
- el->prev = tdb->transaction->elements_last;
- el->offset = off;
- el->length = len;
- el->data = malloc(len);
- if (el->data == NULL) {
- free(el);
- tdb->ecode = TDB_ERR_OOM;
- tdb->transaction->transaction_error = 1;
- return -1;
+ blk = off / tdb->transaction->block_size;
+ off = off % tdb->transaction->block_size;
+
+ if (tdb->transaction->num_blocks <= blk ||
+ tdb->transaction->blocks[blk] == NULL) {
+ return 0;
}
- if (buf) {
- memcpy(el->data, buf, len);
+
+ /* overwrite part of an existing block */
+ if (buf == NULL) {
+ memset(tdb->transaction->blocks[blk] + off, 0, len);
} else {
- memset(el->data, TDB_PAD_BYTE, len);
+ memcpy(tdb->transaction->blocks[blk] + off, buf, len);
}
- if (el->prev) {
- el->prev->next = el;
- } else {
- tdb->transaction->elements = el;
+ if (blk == tdb->transaction->num_blocks-1) {
+ if (len + off > tdb->transaction->last_block_size) {
+ tdb->transaction->last_block_size = len + off;
+ }
}
- tdb->transaction->elements_last = el;
+
return 0;
fail:
- TDB_LOG((tdb, 0, "transaction_write: failed at off=%d len=%d\n", off, len));
- tdb->ecode = TDB_ERR_IO;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%d len=%d\n",
+ (blk*tdb->transaction->block_size) + off, len));
tdb->transaction->transaction_error = 1;
return -1;
}
+
/*
accelerated hash chain head search, using the cached hash heads
*/
-static void transaction_next_hash_chain(struct tdb_context *tdb, u32 *chain)
+static void transaction_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
{
- u32 h = *chain;
+ uint32_t h = *chain;
for (;h < tdb->header.hash_size;h++) {
/* the +1 takes account of the freelist */
if (0 != tdb->transaction->hash_heads[h+1]) {
/*
brlock during a transaction - ignore them
*/
-int transaction_brlock(struct tdb_context *tdb, tdb_off_t offset,
- int rw_type, int lck_type, int probe)
+static int transaction_brlock(struct tdb_context *tdb, tdb_off_t offset,
+ int rw_type, int lck_type, int probe, size_t len)
{
return 0;
}
{
/* some sanity checks */
if (tdb->read_only || (tdb->flags & TDB_INTERNAL) || tdb->traverse_read) {
- TDB_LOG((tdb, 0, "tdb_transaction_start: cannot start a transaction on a read-only or internal db\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction on a read-only or internal db\n"));
tdb->ecode = TDB_ERR_EINVAL;
return -1;
}
/* cope with nested tdb_transaction_start() calls */
if (tdb->transaction != NULL) {
tdb->transaction->nesting++;
- TDB_LOG((tdb, 0, "tdb_transaction_start: nesting %d\n",
+ TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_start: nesting %d\n",
tdb->transaction->nesting));
return 0;
}
- if (tdb->num_locks != 0) {
+ if (tdb->num_locks != 0 || tdb->global_lock.count) {
/* the caller must not have any locks when starting a
transaction as otherwise we'll be screwed by lack
of nested locks in posix */
- TDB_LOG((tdb, 0, "tdb_transaction_start: cannot start a transaction with locks held\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction with locks held\n"));
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
/* you cannot use transactions inside a traverse (although you can use
traverse inside a transaction) as otherwise you can end up with
deadlock */
- TDB_LOG((tdb, 0, "tdb_transaction_start: cannot start a transaction within a traverse\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction within a traverse\n"));
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
- tdb->transaction = calloc(sizeof(struct tdb_transaction), 1);
+ tdb->transaction = (struct tdb_transaction *)
+ calloc(sizeof(struct tdb_transaction), 1);
if (tdb->transaction == NULL) {
tdb->ecode = TDB_ERR_OOM;
return -1;
}
+ /* a page at a time seems like a reasonable compromise between compactness and efficiency */
+ tdb->transaction->block_size = tdb->page_size;
+
/* get the transaction write lock. This is a blocking lock. As
discussed with Volker, there are a number of ways we could
make this async, which we will probably do in the future */
- if (tdb_brlock_len(tdb, TRANSACTION_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_start: failed to get transaction lock\n"));
- tdb->ecode = TDB_ERR_LOCK;
+ if (tdb_transaction_lock(tdb, F_WRLCK) == -1) {
+ SAFE_FREE(tdb->transaction->blocks);
SAFE_FREE(tdb->transaction);
return -1;
}
/* get a read lock from the freelist to the end of file. This
is upgraded to a write lock during the commit */
- if (tdb_brlock_len(tdb, FREELIST_TOP, F_RDLCK, F_SETLKW, 0, 0) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_start: failed to get hash locks\n"));
+ if (tdb_brlock(tdb, FREELIST_TOP, F_RDLCK, F_SETLKW, 0, 0) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get hash locks\n"));
tdb->ecode = TDB_ERR_LOCK;
goto fail;
}
/* setup a copy of the hash table heads so the hash scan in
traverse can be fast */
- tdb->transaction->hash_heads = calloc(tdb->header.hash_size+1, sizeof(tdb_off_t));
+ tdb->transaction->hash_heads = (uint32_t *)
+ calloc(tdb->header.hash_size+1, sizeof(uint32_t));
if (tdb->transaction->hash_heads == NULL) {
tdb->ecode = TDB_ERR_OOM;
goto fail;
}
if (tdb->methods->tdb_read(tdb, FREELIST_TOP, tdb->transaction->hash_heads,
TDB_HASHTABLE_SIZE(tdb), 0) != 0) {
- TDB_LOG((tdb, 0, "tdb_transaction_start: failed to read hash heads\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_start: failed to read hash heads\n"));
tdb->ecode = TDB_ERR_IO;
goto fail;
}
return 0;
fail:
- tdb_brlock_len(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
- tdb_brlock_len(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+ tdb_transaction_unlock(tdb);
+ SAFE_FREE(tdb->transaction->blocks);
SAFE_FREE(tdb->transaction->hash_heads);
SAFE_FREE(tdb->transaction);
return -1;
*/
int tdb_transaction_cancel(struct tdb_context *tdb)
{
+ int i;
+
if (tdb->transaction == NULL) {
- TDB_LOG((tdb, 0, "tdb_transaction_cancel: no transaction\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_cancel: no transaction\n"));
return -1;
}
tdb->map_size = tdb->transaction->old_map_size;
- /* free all the transaction elements */
- while (tdb->transaction->elements) {
- struct tdb_transaction_el *el = tdb->transaction->elements;
- tdb->transaction->elements = el->next;
- free(el->data);
- free(el);
+ /* free all the transaction blocks */
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ if (tdb->transaction->blocks[i] != NULL) {
+ free(tdb->transaction->blocks[i]);
+ }
+ }
+ SAFE_FREE(tdb->transaction->blocks);
+
+ /* remove any global lock created during the transaction */
+ if (tdb->global_lock.count != 0) {
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 4*tdb->header.hash_size);
+ tdb->global_lock.count = 0;
}
/* remove any locks created during the transaction */
if (tdb->num_locks != 0) {
- int h;
- for (h=0;h<tdb->header.hash_size+1;h++) {
- if (tdb->locked[h].count != 0) {
- tdb_brlock_len(tdb,FREELIST_TOP+4*h,F_UNLCK,F_SETLKW, 0, 1);
- tdb->locked[h].count = 0;
- }
+ for (i=0;i<tdb->num_lockrecs;i++) {
+ tdb_brlock(tdb,FREELIST_TOP+4*tdb->lockrecs[i].list,
+ F_UNLCK,F_SETLKW, 0, 1);
}
tdb->num_locks = 0;
+ tdb->num_lockrecs = 0;
+ SAFE_FREE(tdb->lockrecs);
}
/* restore the normal io methods */
tdb->methods = tdb->transaction->io_methods;
- tdb_brlock_len(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
- tdb_brlock_len(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+ tdb_transaction_unlock(tdb);
SAFE_FREE(tdb->transaction->hash_heads);
SAFE_FREE(tdb->transaction);
{
if (fsync(tdb->fd) != 0) {
tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, 0, "tdb_transaction: fsync failed\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: fsync failed\n"));
return -1;
}
#ifdef MS_SYNC
if (msync(moffset + (char *)tdb->map_ptr,
length + (offset - moffset), MS_SYNC) != 0) {
tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, 0, "tdb_transaction: msync failed\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: msync failed - %s\n",
+ strerror(errno)));
return -1;
}
}
*/
static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
{
- struct tdb_transaction_el *el;
tdb_len_t recovery_size = 0;
+ int i;
- recovery_size = sizeof(u32);
- for (el=tdb->transaction->elements;el;el=el->next) {
- if (el->offset >= tdb->transaction->old_map_size) {
+ recovery_size = sizeof(uint32_t);
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ if (i * tdb->transaction->block_size >= tdb->transaction->old_map_size) {
+ break;
+ }
+ if (tdb->transaction->blocks[i] == NULL) {
continue;
}
- recovery_size += 2*sizeof(tdb_off_t) + el->length;
- }
+ recovery_size += 2*sizeof(tdb_off_t);
+ if (i == tdb->transaction->num_blocks-1) {
+ recovery_size += tdb->transaction->last_block_size;
+ } else {
+ recovery_size += tdb->transaction->block_size;
+ }
+ }
return recovery_size;
}
tdb_off_t recovery_head;
if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
- TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to read recovery head\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery head\n"));
return -1;
}
if (recovery_head != 0 &&
methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
- TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to read recovery record\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery record\n"));
return -1;
}
the transaction) */
if (recovery_head != 0) {
if (tdb_free(tdb, recovery_head, &rec) == -1) {
- TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to free previous recovery area\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to free previous recovery area\n"));
return -1;
}
}
if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size,
(tdb->map_size - tdb->transaction->old_map_size) +
sizeof(rec) + *recovery_max_size) == -1) {
- TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to create recovery area\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to create recovery area\n"));
return -1;
}
CONVERT(recovery_head);
if (methods->tdb_write(tdb, TDB_RECOVERY_HEAD,
&recovery_head, sizeof(tdb_off_t)) == -1) {
- TDB_LOG((tdb, 0, "tdb_recovery_allocate: failed to write recovery head\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
+ return -1;
+ }
+ if (transaction_write_existing(tdb, TDB_RECOVERY_HEAD, &recovery_head, sizeof(tdb_off_t)) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
return -1;
}
static int transaction_setup_recovery(struct tdb_context *tdb,
tdb_off_t *magic_offset)
{
- struct tdb_transaction_el *el;
tdb_len_t recovery_size;
unsigned char *data, *p;
const struct tdb_methods *methods = tdb->transaction->io_methods;
struct list_struct *rec;
tdb_off_t recovery_offset, recovery_max_size;
tdb_off_t old_map_size = tdb->transaction->old_map_size;
- u32 magic, tailer;
+ uint32_t magic, tailer;
+ int i;
/*
check that the recovery area has enough space
return -1;
}
- data = malloc(recovery_size + sizeof(*rec));
+ data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
if (data == NULL) {
tdb->ecode = TDB_ERR_OOM;
return -1;
/* build the recovery data into a single blob to allow us to do a single
large write, which should be more efficient */
p = data + sizeof(*rec);
- for (el=tdb->transaction->elements;el;el=el->next) {
- if (el->offset >= old_map_size) {
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ tdb_off_t offset;
+ tdb_len_t length;
+
+ if (tdb->transaction->blocks[i] == NULL) {
+ continue;
+ }
+
+ offset = i * tdb->transaction->block_size;
+ length = tdb->transaction->block_size;
+ if (i == tdb->transaction->num_blocks-1) {
+ length = tdb->transaction->last_block_size;
+ }
+
+ if (offset >= old_map_size) {
continue;
}
- if (el->offset + el->length > tdb->transaction->old_map_size) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: transaction data over new region boundary\n"));
+ if (offset + length > tdb->transaction->old_map_size) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: transaction data over new region boundary\n"));
free(data);
tdb->ecode = TDB_ERR_CORRUPT;
return -1;
}
- memcpy(p, &el->offset, 4);
- memcpy(p+4, &el->length, 4);
+ memcpy(p, &offset, 4);
+ memcpy(p+4, &length, 4);
if (DOCONV()) {
tdb_convert(p, 8);
}
/* the recovery area contains the old data, not the
new data, so we have to call the original tdb_read
method to get it */
- if (methods->tdb_read(tdb, el->offset, p + 8, el->length, 0) != 0) {
+ if (methods->tdb_read(tdb, offset, p + 8, length, 0) != 0) {
free(data);
tdb->ecode = TDB_ERR_IO;
return -1;
}
- p += 8 + el->length;
+ p += 8 + length;
}
/* and the tailer */
/* write the recovery data to the recovery area */
if (methods->tdb_write(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to write recovery data\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery data\n"));
+ free(data);
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ if (transaction_write_existing(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery data\n"));
free(data);
tdb->ecode = TDB_ERR_IO;
return -1;
*magic_offset = recovery_offset + offsetof(struct list_struct, magic);
if (methods->tdb_write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to write recovery magic\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery magic\n"));
+ tdb->ecode = TDB_ERR_IO;
+ return -1;
+ }
+ if (transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery magic\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
{
const struct tdb_methods *methods;
tdb_off_t magic_offset = 0;
- u32 zero = 0;
+ uint32_t zero = 0;
+ int i;
if (tdb->transaction == NULL) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: no transaction\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: no transaction\n"));
return -1;
}
if (tdb->transaction->transaction_error) {
tdb->ecode = TDB_ERR_IO;
tdb_transaction_cancel(tdb);
- TDB_LOG((tdb, 0, "tdb_transaction_commit: transaction error pending\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: transaction error pending\n"));
return -1;
}
+
if (tdb->transaction->nesting != 0) {
tdb->transaction->nesting--;
return 0;
}
/* check for a null transaction */
- if (tdb->transaction->elements == NULL) {
+ if (tdb->transaction->blocks == NULL) {
tdb_transaction_cancel(tdb);
return 0;
}
/* if there are any locks pending then the caller has not
nested their locks properly, so fail the transaction */
- if (tdb->num_locks) {
+ if (tdb->num_locks || tdb->global_lock.count) {
tdb->ecode = TDB_ERR_LOCK;
- TDB_LOG((tdb, 0, "tdb_transaction_commit: locks pending on commit\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: locks pending on commit\n"));
tdb_transaction_cancel(tdb);
return -1;
}
/* upgrade the main transaction lock region to a write lock */
if (tdb_brlock_upgrade(tdb, FREELIST_TOP, 0) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_start: failed to upgrade hash locks\n"));
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to upgrade hash locks\n"));
tdb->ecode = TDB_ERR_LOCK;
tdb_transaction_cancel(tdb);
return -1;
/* get the global lock - this prevents new users attaching to the database
during the commit */
- if (tdb_brlock_len(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to get global lock\n"));
+ if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: failed to get global lock\n"));
tdb->ecode = TDB_ERR_LOCK;
tdb_transaction_cancel(tdb);
return -1;
if (!(tdb->flags & TDB_NOSYNC)) {
/* write the recovery data to the end of the file */
if (transaction_setup_recovery(tdb, &magic_offset) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to setup recovery data\n"));
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: failed to setup recovery data\n"));
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
tdb_transaction_cancel(tdb);
return -1;
}
tdb->map_size -
tdb->transaction->old_map_size) == -1) {
tdb->ecode = TDB_ERR_IO;
- TDB_LOG((tdb, 0, "tdb_transaction_commit: expansion failed\n"));
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: expansion failed\n"));
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
tdb_transaction_cancel(tdb);
return -1;
}
}
/* perform all the writes */
- while (tdb->transaction->elements) {
- struct tdb_transaction_el *el = tdb->transaction->elements;
+ for (i=0;i<tdb->transaction->num_blocks;i++) {
+ tdb_off_t offset;
+ tdb_len_t length;
+
+ if (tdb->transaction->blocks[i] == NULL) {
+ continue;
+ }
+
+ offset = i * tdb->transaction->block_size;
+ length = tdb->transaction->block_size;
+ if (i == tdb->transaction->num_blocks-1) {
+ length = tdb->transaction->last_block_size;
+ }
- if (methods->tdb_write(tdb, el->offset, el->data, el->length) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: write failed during commit\n"));
+ if (methods->tdb_write(tdb, offset, tdb->transaction->blocks[i], length) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed during commit\n"));
/* we've overwritten part of the data and
possibly expanded the file, so we need to
tdb_transaction_recover(tdb);
tdb_transaction_cancel(tdb);
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
- TDB_LOG((tdb, 0, "tdb_transaction_commit: write failed\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
return -1;
}
- tdb->transaction->elements = el->next;
- free(el->data);
- free(el);
+ SAFE_FREE(tdb->transaction->blocks[i]);
}
+ SAFE_FREE(tdb->transaction->blocks);
+ tdb->transaction->num_blocks = 0;
+
if (!(tdb->flags & TDB_NOSYNC)) {
/* ensure the new data is on disk */
if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
/* remove the recovery marker */
if (methods->tdb_write(tdb, magic_offset, &zero, 4) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_commit: failed to remove recovery magic\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: failed to remove recovery magic\n"));
return -1;
}
}
}
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+
+ /*
+ TODO: maybe write to some dummy hdr field, or write to magic
+ offset without mmap, before the last sync, instead of the
+ utime() call
+ */
/* on some systems (like Linux 2.6.x) changes via mmap/msync
don't change the mtime of the file, this means the file may
/* use a transaction cancel to free memory and remove the
transaction locks */
tdb_transaction_cancel(tdb);
+
return 0;
}
{
tdb_off_t recovery_head, recovery_eof;
unsigned char *data, *p;
- u32 zero = 0;
+ uint32_t zero = 0;
struct list_struct rec;
/* find the recovery area */
if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to read recovery head\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery head\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* read the recovery record */
if (tdb->methods->tdb_read(tdb, recovery_head, &rec,
sizeof(rec), DOCONV()) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to read recovery record\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery record\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
}
if (tdb->read_only) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: attempt to recover read only database\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: attempt to recover read only database\n"));
tdb->ecode = TDB_ERR_CORRUPT;
return -1;
}
recovery_eof = rec.key_len;
- data = malloc(rec.data_len);
+ data = (unsigned char *)malloc(rec.data_len);
if (data == NULL) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to allocate recovery data\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to allocate recovery data\n"));
tdb->ecode = TDB_ERR_OOM;
return -1;
}
/* read the full recovery data */
if (tdb->methods->tdb_read(tdb, recovery_head + sizeof(rec), data,
rec.data_len, 0) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to read recovery data\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery data\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* recover the file data */
p = data;
while (p+8 < data + rec.data_len) {
- u32 ofs, len;
+ uint32_t ofs, len;
if (DOCONV()) {
tdb_convert(p, 8);
}
if (tdb->methods->tdb_write(tdb, ofs, p+8, len) == -1) {
free(data);
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to recover %d bytes at offset %d\n", len, ofs));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to recover %d bytes at offset %d\n", len, ofs));
tdb->ecode = TDB_ERR_IO;
return -1;
}
free(data);
if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to sync recovery\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync recovery\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* if the recovery area is after the recovered eof then remove it */
if (recovery_eof <= recovery_head) {
if (tdb_ofs_write(tdb, TDB_RECOVERY_HEAD, &zero) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to remove recovery head\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery head\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* remove the recovery magic */
if (tdb_ofs_write(tdb, recovery_head + offsetof(struct list_struct, magic),
&zero) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to remove recovery magic\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery magic\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* reduce the file size to the old size */
tdb_munmap(tdb);
if (ftruncate(tdb->fd, recovery_eof) != 0) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to reduce to recovery size\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to reduce to recovery size\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
tdb_mmap(tdb);
if (transaction_sync(tdb, 0, recovery_eof) == -1) {
- TDB_LOG((tdb, 0, "tdb_transaction_recover: failed to sync2 recovery\n"));
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync2 recovery\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
- TDB_LOG((tdb, 0, "tdb_transaction_recover: recovered %d byte database\n",
+ TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_recover: recovered %d byte database\n",
recovery_eof));
/* all done */