X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Ftransaction.c;h=68ae234d2f90ccbe4fb993dd27476df7c89afba2;hp=604861b738e75e9c7efab447f7d26d19780a5ffc;hb=dc9da1e34fe6a9d113fd57e116ebbc6d5bd54819;hpb=3835bb95d716cb6448913792b0d8d539877d93e9

diff --git a/ccan/tdb2/transaction.c b/ccan/tdb2/transaction.c
index 604861b7..68ae234d 100644
--- a/ccan/tdb2/transaction.c
+++ b/ccan/tdb2/transaction.c
@@ -53,13 +53,13 @@
 
   - don't allow any locks to be held when a transaction starts,
     otherwise we can end up with deadlock (plus lack of lock nesting
-    in posix locks would mean the lock is lost)
+    in POSIX locks would mean the lock is lost)
 
   - if the caller gains a lock during the transaction but doesn't
     release it then fail the commit
 
   - allow for nested calls to tdb_transaction_start(), re-using the
-    existing transaction record. If the inner transaction is cancelled
+    existing transaction record. If the inner transaction is canceled
     then a subsequent commit will fail
 
   - keep a mirrored copy of the tdb hash chain heads to allow for the
@@ -68,7 +68,7 @@
 
   - allow callers to mix transaction and non-transaction use of tdb,
     although once a transaction is started then an exclusive lock is
-    gained until the transaction is committed or cancelled
+    gained until the transaction is committed or canceled
 
   - the commit stategy involves first saving away all modified data
     into a linearised buffer in the transaction recovery area, then
@@ -120,6 +120,8 @@ struct tdb_transaction {
 	tdb_len_t old_map_size;
 };
 
+/* This doesn't really need to be pagesize, but we use it for similar reasons. */
+#define PAGESIZE 4096
 
 /*
   read while in a transaction. We need to check first if the data is in our list
@@ -132,8 +134,8 @@ static enum TDB_ERROR transaction_read(struct tdb_context *tdb, tdb_off_t off,
 	enum TDB_ERROR ecode;
 
 	/* break it down into block sized ops */
-	while (len + (off % getpagesize()) > getpagesize()) {
-		tdb_len_t len2 = getpagesize() - (off % getpagesize());
+	while (len + (off % PAGESIZE) > PAGESIZE) {
+		tdb_len_t len2 = PAGESIZE - (off % PAGESIZE);
 		ecode = transaction_read(tdb, off, buf, len2);
 		if (ecode != TDB_SUCCESS) {
 			return ecode;
@@ -147,7 +149,7 @@ static enum TDB_ERROR transaction_read(struct tdb_context *tdb, tdb_off_t off,
 		return TDB_SUCCESS;
 	}
 
-	blk = off / getpagesize();
+	blk = off / PAGESIZE;
 
 	/* see if we have it in the block list */
 	if (tdb->transaction->num_blocks <= blk ||
@@ -169,7 +171,7 @@ static enum TDB_ERROR transaction_read(struct tdb_context *tdb, tdb_off_t off,
 	}
 
 	/* now copy it out of this block */
-	memcpy(buf, tdb->transaction->blocks[blk] + (off % getpagesize()), len);
+	memcpy(buf, tdb->transaction->blocks[blk] + (off % PAGESIZE), len);
 	return TDB_SUCCESS;
 
 fail:
@@ -198,8 +200,8 @@ static enum TDB_ERROR transaction_write(struct tdb_context *tdb, tdb_off_t off,
 	}
 
 	/* break it up into block sized chunks */
-	while (len + (off % getpagesize()) > getpagesize()) {
-		tdb_len_t len2 = getpagesize() - (off % getpagesize());
+	while (len + (off % PAGESIZE) > PAGESIZE) {
+		tdb_len_t len2 = PAGESIZE - (off % PAGESIZE);
 		ecode = transaction_write(tdb, off, buf, len2);
 		if (ecode != TDB_SUCCESS) {
 			return -1;
@@ -215,8 +217,8 @@ static enum TDB_ERROR transaction_write(struct tdb_context *tdb, tdb_off_t off,
 		return TDB_SUCCESS;
 	}
 
-	blk = off / getpagesize();
-	off = off % getpagesize();
+	blk = off / PAGESIZE;
+	off = off % PAGESIZE;
 
 	if (tdb->transaction->num_blocks <= blk) {
 		uint8_t **new_blocks;
@@ -244,20 +246,20 @@ static enum TDB_ERROR transaction_write(struct tdb_context *tdb, tdb_off_t off,
 
 	/* allocate and fill a block? */
 	if (tdb->transaction->blocks[blk] == NULL) {
-		tdb->transaction->blocks[blk] = (uint8_t *)calloc(getpagesize(), 1);
+		tdb->transaction->blocks[blk] = (uint8_t *)calloc(PAGESIZE, 1);
 		if (tdb->transaction->blocks[blk] == NULL) {
 			ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 					   "transaction_write:"
 					   " failed to allocate");
 			goto fail;
 		}
-		if (tdb->transaction->old_map_size > blk * getpagesize()) {
-			tdb_len_t len2 = getpagesize();
-			if (len2 + (blk * getpagesize()) > tdb->transaction->old_map_size) {
-				len2 = tdb->transaction->old_map_size - (blk * getpagesize());
+		if (tdb->transaction->old_map_size > blk * PAGESIZE) {
+			tdb_len_t len2 = PAGESIZE;
+			if (len2 + (blk * PAGESIZE) > tdb->transaction->old_map_size) {
+				len2 = tdb->transaction->old_map_size - (blk * PAGESIZE);
 			}
 			ecode = tdb->transaction->io_methods->tread(tdb,
-					blk * getpagesize(),
+					blk * PAGESIZE,
 					tdb->transaction->blocks[blk],
 					len2);
 			if (ecode != TDB_SUCCESS) {
@@ -297,7 +299,7 @@ fail:
 
 
 /*
-  write while in a transaction - this varient never expands the transaction blocks, it only
+  write while in a transaction - this variant never expands the transaction blocks, it only
   updates existing blocks. This means it cannot change the recovery size
 */
 static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
@@ -306,8 +308,8 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
 	size_t blk;
 
 	/* break it up into block sized chunks */
-	while (len + (off % getpagesize()) > getpagesize()) {
-		tdb_len_t len2 = getpagesize() - (off % getpagesize());
+	while (len + (off % PAGESIZE) > PAGESIZE) {
+		tdb_len_t len2 = PAGESIZE - (off % PAGESIZE);
 		transaction_write_existing(tdb, off, buf, len2);
 		len -= len2;
 		off += len2;
@@ -320,8 +322,8 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
 		return;
 	}
 
-	blk = off / getpagesize();
-	off = off % getpagesize();
+	blk = off / PAGESIZE;
+	off = off % PAGESIZE;
 
 	if (tdb->transaction->num_blocks <= blk ||
 	    tdb->transaction->blocks[blk] == NULL) {
@@ -347,14 +349,14 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
 static enum TDB_ERROR transaction_oob(struct tdb_context *tdb, tdb_off_t len,
 				      bool probe)
 {
-	if (len <= tdb->map_size) {
+	if (len <= tdb->file->map_size) {
 		return TDB_SUCCESS;
 	}
 	if (!probe) {
 		tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 			   "tdb_oob len %lld beyond transaction size %lld",
 			   (long long)len,
-			   (long long)tdb->map_size);
+			   (long long)tdb->file->map_size);
 	}
 	return TDB_ERR_IO;
 }
@@ -369,22 +371,20 @@ static enum TDB_ERROR transaction_expand_file(struct tdb_context *tdb,
 
 	/* add a write to the transaction elements, so subsequent
 	   reads see the zero data */
-	ecode = transaction_write(tdb, tdb->map_size, NULL, addition);
-	if (ecode != TDB_SUCCESS) {
-		tdb->ecode = ecode;
-		return ecode;
+	ecode = transaction_write(tdb, tdb->file->map_size, NULL, addition);
+	if (ecode == TDB_SUCCESS) {
+		tdb->file->map_size += addition;
 	}
-	tdb->map_size += addition;
 	return ecode;
 }
 
 static void *transaction_direct(struct tdb_context *tdb, tdb_off_t off,
 				size_t len, bool write_mode)
 {
-	size_t blk = off / getpagesize(), end_blk;
+	size_t blk = off / PAGESIZE, end_blk;
 
 	/* This is wrong for zero-length blocks, but will fail gracefully */
-	end_blk = (off + len - 1) / getpagesize();
+	end_blk = (off + len - 1) / PAGESIZE;
 
 	/* Can only do direct if in single block and we've already copied. */
 	if (write_mode) {
@@ -394,17 +394,17 @@ static void *transaction_direct(struct tdb_context *tdb, tdb_off_t off,
 			return NULL;
 		if (tdb->transaction->blocks[blk] == NULL)
 			return NULL;
-		return tdb->transaction->blocks[blk] + off % getpagesize();
+		return tdb->transaction->blocks[blk] + off % PAGESIZE;
 	}
 
 	/* Single which we have copied? */
 	if (blk == end_blk
 	    && blk < tdb->transaction->num_blocks
 	    && tdb->transaction->blocks[blk])
-		return tdb->transaction->blocks[blk] + off % getpagesize();
+		return tdb->transaction->blocks[blk] + off % PAGESIZE;
 
 	/* Otherwise must be all not copied. */
-	while (blk < end_blk) {
+	while (blk <= end_blk) {
 		if (blk >= tdb->transaction->num_blocks)
 			break;
 		if (tdb->transaction->blocks[blk])
@@ -432,15 +432,15 @@ static enum TDB_ERROR transaction_sync(struct tdb_context *tdb,
 		return TDB_SUCCESS;
 	}
 
-	if (fsync(tdb->fd) != 0) {
+	if (fsync(tdb->file->fd) != 0) {
 		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 				  "tdb_transaction: fsync failed: %s",
 				  strerror(errno));
 	}
 #ifdef MS_SYNC
-	if (tdb->map_ptr) {
-		tdb_off_t moffset = offset & ~(getpagesize()-1);
-		if (msync(moffset + (char *)tdb->map_ptr,
+	if (tdb->file->map_ptr) {
+		tdb_off_t moffset = offset & ~(PAGESIZE-1);
+		if (msync(moffset + (char *)tdb->file->map_ptr,
 			  length + (offset - moffset), MS_SYNC) != 0) {
 			return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 					  "tdb_transaction: msync failed: %s",
@@ -469,7 +469,7 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
 		return;
 	}
 
-	tdb->map_size = tdb->transaction->old_map_size;
+	tdb->file->map_size = tdb->transaction->old_map_size;
 
 	/* free all the transaction blocks */
 	for (i=0;i<tdb->transaction->num_blocks;i++) {
@@ -497,8 +497,8 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
 		}
 	}
 
-	if (tdb->allrecord_lock.count)
-		tdb_allrecord_unlock(tdb, tdb->allrecord_lock.ltype);
+	if (tdb->file->allrecord_lock.count)
+		tdb_allrecord_unlock(tdb, tdb->file->allrecord_lock.ltype);
 
 	/* restore the normal io methods */
 	tdb->methods = tdb->transaction->io_methods;
@@ -515,42 +515,48 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
   start a tdb transaction. No token is returned, as only a single
   transaction is allowed to be pending per tdb_context
 */
-int tdb_transaction_start(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_start(struct tdb_context *tdb)
 {
 	enum TDB_ERROR ecode;
 
 	/* some sanity checks */
 	if (tdb->read_only || (tdb->flags & TDB_INTERNAL)) {
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
-			   "tdb_transaction_start: cannot start a transaction"
-			   " on a read-only or internal db");
-		return -1;
+		return tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+						    TDB_LOG_USE_ERROR,
+						    "tdb_transaction_start:"
+						    " cannot start a"
+						    " transaction on a "
+						    "read-only or internal db");
 	}
 
 	/* cope with nested tdb_transaction_start() calls */
 	if (tdb->transaction != NULL) {
-		tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
-			   "tdb_transaction_start:"
-			   " already inside transaction");
-		return -1;
+		return tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO,
+						    TDB_LOG_USE_ERROR,
+						    "tdb_transaction_start:"
+						    " already inside"
+						    " transaction");
 	}
 
 	if (tdb_has_hash_locks(tdb)) {
 		/* 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_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
-			   "tdb_transaction_start: cannot start a transaction"
-			   " with locks held");
-		return -1;
+		   of nested locks in POSIX */
+		return tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK,
+						    TDB_LOG_USE_ERROR,
+						    "tdb_transaction_start:"
+						    " cannot start a"
+						    " transaction with locks"
+						    " held");
 	}
 
 	tdb->transaction = (struct tdb_transaction *)
 		calloc(sizeof(struct tdb_transaction), 1);
 	if (tdb->transaction == NULL) {
-		tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
-			   "tdb_transaction_start: cannot allocate");
-		return -1;
+		return tdb->last_error = tdb_logerr(tdb, TDB_ERR_OOM,
+						    TDB_LOG_ERROR,
+						    "tdb_transaction_start:"
+						    " cannot allocate");
 	}
 
 	/* get the transaction write lock. This is a blocking lock. As
@@ -558,36 +564,34 @@ int tdb_transaction_start(struct tdb_context *tdb)
 	   make this async, which we will probably do in the future */
 	ecode = tdb_transaction_lock(tdb, F_WRLCK);
 	if (ecode != TDB_SUCCESS) {
-		tdb->ecode = ecode;
 		SAFE_FREE(tdb->transaction->blocks);
 		SAFE_FREE(tdb->transaction);
-		return -1;
+		return tdb->last_error = ecode;
 	}
 
 	/* get a read lock over entire file. This is upgraded to a write
 	   lock during the commit */
 	ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true);
 	if (ecode != TDB_SUCCESS) {
-		tdb->ecode = ecode;
 		goto fail_allrecord_lock;
 	}
 
 	/* make sure we know about any file expansions already done by
 	   anyone else */
-	tdb->methods->oob(tdb, tdb->map_size + 1, true);
-	tdb->transaction->old_map_size = tdb->map_size;
+	tdb->methods->oob(tdb, tdb->file->map_size + 1, true);
+	tdb->transaction->old_map_size = tdb->file->map_size;
 
 	/* finally hook the io methods, replacing them with
 	   transaction specific methods */
 	tdb->transaction->io_methods = tdb->methods;
 	tdb->methods = &transaction_methods;
-	return 0;
+	return tdb->last_error = TDB_SUCCESS;
 
 fail_allrecord_lock:
 	tdb_transaction_unlock(tdb, F_WRLCK);
 	SAFE_FREE(tdb->transaction->blocks);
 	SAFE_FREE(tdb->transaction);
-	return -1;
+	return tdb->last_error = ecode;
 }
 
 
@@ -609,7 +613,7 @@ static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
 
 	recovery_size = sizeof(tdb_len_t);
 	for (i=0;i<tdb->transaction->num_blocks;i++) {
-		if (i * getpagesize() >= tdb->transaction->old_map_size) {
+		if (i * PAGESIZE >= tdb->transaction->old_map_size) {
 			break;
 		}
 		if (tdb->transaction->blocks[i] == NULL) {
@@ -619,7 +623,7 @@ static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
 		if (i == tdb->transaction->num_blocks-1) {
 			recovery_size += tdb->transaction->last_block_size;
 		} else {
-			recovery_size += getpagesize();
+			recovery_size += PAGESIZE;
 		}
 	}
 
@@ -630,10 +634,10 @@ static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
   allocate the recovery area, or use an existing recovery area if it is
   large enough
 */
-static int tdb_recovery_allocate(struct tdb_context *tdb,
-				 tdb_len_t *recovery_size,
-				 tdb_off_t *recovery_offset,
-				 tdb_len_t *recovery_max_size)
+static enum TDB_ERROR tdb_recovery_allocate(struct tdb_context *tdb,
+					    tdb_len_t *recovery_size,
+					    tdb_off_t *recovery_offset,
+					    tdb_len_t *recovery_max_size)
 {
 	struct tdb_recovery_record rec;
 	const struct tdb_methods *methods = tdb->transaction->io_methods;
@@ -643,19 +647,17 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 
 	recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
 	if (TDB_OFF_IS_ERR(recovery_head)) {
-		tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
-			 "tdb_recovery_allocate:"
-			 " failed to read recovery head");
-		return -1;
+		return tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
+				  "tdb_recovery_allocate:"
+				  " failed to read recovery head");
 	}
 
 	if (recovery_head != 0) {
 		ecode = methods->tread(tdb, recovery_head, &rec, sizeof(rec));
 		if (ecode != TDB_SUCCESS) {
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				 "tdb_recovery_allocate:"
-				 " failed to read recovery record");
-			return -1;
+			return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+					  "tdb_recovery_allocate:"
+					  " failed to read recovery record");
 		}
 		tdb_convert(tdb, &rec, sizeof(rec));
 		/* ignore invalid recovery regions: can happen in crash */
@@ -671,7 +673,7 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 		/* it fits in the existing area */
 		*recovery_max_size = rec.max_len;
 		*recovery_offset = recovery_head;
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	/* we need to free up the old recovery area, then allocate a
@@ -680,14 +682,14 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 	   us an area that is being currently used (as of the start of
 	   the transaction) */
 	if (recovery_head != 0) {
-		add_stat(tdb, frees, 1);
+		tdb->stats.frees++;
 		ecode = add_free_record(tdb, recovery_head,
 					sizeof(rec) + rec.max_len);
 		if (ecode != TDB_SUCCESS) {
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				   "tdb_recovery_allocate:"
-				   " failed to free previous recovery area");
-			return -1;
+			return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+					  "tdb_recovery_allocate:"
+					  " failed to free previous"
+					  " recovery area");
 		}
 	}
 
@@ -696,31 +698,30 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 
 	/* round up to a multiple of page size */
 	*recovery_max_size
-		= (((sizeof(rec) + *recovery_size) + getpagesize()-1)
-		   & ~(getpagesize()-1))
+		= (((sizeof(rec) + *recovery_size) + PAGESIZE-1)
+		   & ~(PAGESIZE-1))
 		- sizeof(rec);
-	*recovery_offset = tdb->map_size;
+	*recovery_offset = tdb->file->map_size;
 	recovery_head = *recovery_offset;
 
 	/* Restore ->map_size before calling underlying expand_file.
 	   Also so that we don't try to expand the file again in the
 	   transaction commit, which would destroy the recovery
 	   area */
-	addition = (tdb->map_size - tdb->transaction->old_map_size) +
+	addition = (tdb->file->map_size - tdb->transaction->old_map_size) +
 		sizeof(rec) + *recovery_max_size;
-	tdb->map_size = tdb->transaction->old_map_size;
+	tdb->file->map_size = tdb->transaction->old_map_size;
 	ecode = methods->expand_file(tdb, addition);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			 "tdb_recovery_allocate:"
-			 " failed to create recovery area");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_recovery_allocate:"
+				  " failed to create recovery area");
 	}
 
 	/* we have to reset the old map size so that we don't try to
 	   expand the file again in the transaction commit, which
 	   would destroy the recovery area */
-	tdb->transaction->old_map_size = tdb->map_size;
+	tdb->transaction->old_map_size = tdb->file->map_size;
 
 	/* write the recovery header offset and sync - we can sync without a race here
 	   as the magic ptr in the recovery record has not been set */
@@ -728,15 +729,14 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 	ecode = methods->twrite(tdb, offsetof(struct tdb_header, recovery),
 				&recovery_head, sizeof(tdb_off_t));
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			 "tdb_recovery_allocate:"
-			 " failed to write recovery head");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_recovery_allocate:"
+				  " failed to write recovery head");
 	}
 	transaction_write_existing(tdb, offsetof(struct tdb_header, recovery),
 				   &recovery_head,
 				   sizeof(tdb_off_t));
-	return 0;
+	return TDB_SUCCESS;
 }
 
 /* Set up header for the recovery record. */
@@ -754,14 +754,15 @@ static void set_recovery_header(struct tdb_recovery_record *rec,
 /*
   setup the recovery data that will be used on a crash during commit
 */
-static int transaction_setup_recovery(struct tdb_context *tdb,
-				      tdb_off_t *magic_offset)
+static enum TDB_ERROR transaction_setup_recovery(struct tdb_context *tdb,
+						 tdb_off_t *magic_offset)
 {
-	tdb_len_t recovery_size;
+	/* Initialized for GCC's 4.4.5 overzealous uninitialized warnings. */
+	tdb_len_t recovery_size = 0;
+	tdb_off_t recovery_offset = 0, recovery_max_size = 0;
 	unsigned char *data, *p;
 	const struct tdb_methods *methods = tdb->transaction->io_methods;
 	struct tdb_recovery_record *rec;
-	tdb_off_t recovery_offset, recovery_max_size;
 	tdb_off_t old_map_size = tdb->transaction->old_map_size;
 	uint64_t magic, tailer;
 	int i;
@@ -770,16 +771,17 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	/*
 	  check that the recovery area has enough space
 	*/
-	if (tdb_recovery_allocate(tdb, &recovery_size,
-				  &recovery_offset, &recovery_max_size) == -1) {
-		return -1;
+	ecode = tdb_recovery_allocate(tdb, &recovery_size,
+				      &recovery_offset, &recovery_max_size);
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
 	}
 
 	data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
 	if (data == NULL) {
-		tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
-			   "transaction_setup_recovery: cannot allocate");
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+				  "transaction_setup_recovery:"
+				  " cannot allocate");
 	}
 
 	rec = (struct tdb_recovery_record *)data;
@@ -798,8 +800,8 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 			continue;
 		}
 
-		offset = i * getpagesize();
-		length = getpagesize();
+		offset = i * PAGESIZE;
+		length = PAGESIZE;
 		if (i == tdb->transaction->num_blocks-1) {
 			length = tdb->transaction->last_block_size;
 		}
@@ -807,12 +809,12 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 		if (offset >= old_map_size) {
 			continue;
 		}
-		if (offset + length > tdb->map_size) {
-			tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
-				   "tdb_transaction_setup_recovery:"
-				   " transaction data over new region boundary");
+		if (offset + length > tdb->file->map_size) {
 			free(data);
-			return -1;
+			return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
+					  "tdb_transaction_setup_recovery:"
+					  " transaction data over new region"
+					  " boundary");
 		}
 		memcpy(p, &offset, sizeof(offset));
 		memcpy(p + sizeof(offset), &length, sizeof(length));
@@ -825,9 +827,8 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 				       p + sizeof(offset) + sizeof(length),
 				       length);
 		if (ecode != TDB_SUCCESS) {
-			tdb->ecode = ecode;
 			free(data);
-			return -1;
+			return ecode;
 		}
 		p += sizeof(offset) + sizeof(length) + length;
 	}
@@ -841,11 +842,10 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	ecode = methods->twrite(tdb, recovery_offset, data,
 				sizeof(*rec) + recovery_size);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			 "tdb_transaction_setup_recovery:"
-			 " failed to write recovery data");
 		free(data);
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_setup_recovery:"
+				  " failed to write recovery data");
 	}
 	transaction_write_existing(tdb, recovery_offset, data,
 				   sizeof(*rec) + recovery_size);
@@ -857,8 +857,7 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 				 sizeof(*rec) + recovery_size);
 	if (ecode != TDB_SUCCESS) {
 		free(data);
-		tdb->ecode = ecode;
-		return -1;
+		return ecode;
 	}
 
 	free(data);
@@ -871,59 +870,50 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 
 	ecode = methods->twrite(tdb, *magic_offset, &magic, sizeof(magic));
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			 "tdb_transaction_setup_recovery:"
-			 " failed to write recovery magic");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_setup_recovery:"
+				  " failed to write recovery magic");
 	}
 	transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic));
 
 	/* ensure the recovery magic marker is on disk */
-	ecode = transaction_sync(tdb, *magic_offset, sizeof(magic));
-	if (ecode != TDB_SUCCESS) {
-		tdb->ecode = ecode;
-		return -1;
-	}
-
-	return 0;
+	return transaction_sync(tdb, *magic_offset, sizeof(magic));
 }
 
-static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
+static enum TDB_ERROR _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 {
 	const struct tdb_methods *methods;
 	enum TDB_ERROR ecode;
 
 	if (tdb->transaction == NULL) {
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
-			   "tdb_transaction_prepare_commit: no transaction");
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+				  "tdb_transaction_prepare_commit:"
+				  " no transaction");
 	}
 
 	if (tdb->transaction->prepared) {
 		_tdb_transaction_cancel(tdb);
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
-			   "tdb_transaction_prepare_commit:"
-			   " transaction already prepared");
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+				  "tdb_transaction_prepare_commit:"
+				  " transaction already prepared");
 	}
 
 	if (tdb->transaction->transaction_error) {
 		_tdb_transaction_cancel(tdb);
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_ERROR,
-			   "tdb_transaction_prepare_commit:"
-			   " transaction error pending");
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_ERROR,
+				  "tdb_transaction_prepare_commit:"
+				  " transaction error pending");
 	}
 
 
 	if (tdb->transaction->nesting != 0) {
 		tdb->transaction->nesting--;
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	/* check for a null transaction */
 	if (tdb->transaction->blocks == NULL) {
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	methods = tdb->transaction->io_methods;
@@ -931,62 +921,50 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 	/* upgrade the main transaction lock region to a write lock */
 	ecode = tdb_allrecord_upgrade(tdb);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			 "tdb_transaction_prepare_commit:"
-			 " failed to upgrade hash locks");
-		_tdb_transaction_cancel(tdb);
-		return -1;
+		return ecode;
 	}
 
 	/* get the open lock - this prevents new users attaching to the database
 	   during the commit */
 	ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			   "tdb_transaction_prepare_commit:"
-			   " failed to get open lock");
-		_tdb_transaction_cancel(tdb);
-		return -1;
+		return ecode;
 	}
 
 	/* Since we have whole db locked, we don't need the expansion lock. */
 	if (!(tdb->flags & TDB_NOSYNC)) {
 		/* write the recovery data to the end of the file */
-		if (transaction_setup_recovery(tdb, &tdb->transaction->magic_offset) == -1) {
-			tdb_logerr(tdb, tdb->ecode, TDB_LOG_ERROR,
-				 "tdb_transaction_prepare_commit:"
-				 " failed to setup recovery data");
-			_tdb_transaction_cancel(tdb);
-			return -1;
+		ecode = transaction_setup_recovery(tdb,
+						   &tdb->transaction
+						   ->magic_offset);
+		if (ecode != TDB_SUCCESS) {
+			return ecode;
 		}
 	}
 
 	tdb->transaction->prepared = true;
 
 	/* expand the file to the new size if needed */
-	if (tdb->map_size != tdb->transaction->old_map_size) {
-		tdb_len_t add = tdb->map_size - tdb->transaction->old_map_size;
+	if (tdb->file->map_size != tdb->transaction->old_map_size) {
+		tdb_len_t add;
+
+		add = tdb->file->map_size - tdb->transaction->old_map_size;
 		/* Restore original map size for tdb_expand_file */
-		tdb->map_size = tdb->transaction->old_map_size;
+		tdb->file->map_size = tdb->transaction->old_map_size;
 		ecode = methods->expand_file(tdb, add);
 		if (ecode != TDB_SUCCESS) {
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				 "tdb_transaction_prepare_commit:"
-				 " expansion failed");
-			_tdb_transaction_cancel(tdb);
-			return -1;
+			return ecode;
 		}
 	}
 
 	/* Keep the open lock until the actual commit */
-
-	return 0;
+	return TDB_SUCCESS;
 }
 
 /*
    prepare to commit the current transaction
 */
-int tdb_transaction_prepare_commit(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_prepare_commit(struct tdb_context *tdb)
 {
 	return _tdb_transaction_prepare_commit(tdb);
 }
@@ -994,35 +972,38 @@ int tdb_transaction_prepare_commit(struct tdb_context *tdb)
 /*
   commit the current transaction
 */
-int tdb_transaction_commit(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_commit(struct tdb_context *tdb)
 {
 	const struct tdb_methods *methods;
 	int i;
 	enum TDB_ERROR ecode;
 
 	if (tdb->transaction == NULL) {
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
-			 "tdb_transaction_commit: no transaction");
-		return -1;
+		return tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+						    TDB_LOG_USE_ERROR,
+						    "tdb_transaction_commit:"
+						    " no transaction");
 	}
 
 	tdb_trace(tdb, "tdb_transaction_commit");
 
 	if (tdb->transaction->nesting != 0) {
 		tdb->transaction->nesting--;
-		return 0;
+		return tdb->last_error = TDB_SUCCESS;
 	}
 
 	/* check for a null transaction */
 	if (tdb->transaction->blocks == NULL) {
 		_tdb_transaction_cancel(tdb);
-		return 0;
+		return tdb->last_error = TDB_SUCCESS;
 	}
 
 	if (!tdb->transaction->prepared) {
-		int ret = _tdb_transaction_prepare_commit(tdb);
-		if (ret)
-			return ret;
+		ecode = _tdb_transaction_prepare_commit(tdb);
+		if (ecode != TDB_SUCCESS) {
+			_tdb_transaction_cancel(tdb);
+			return tdb->last_error = ecode;
+		}
 	}
 
 	methods = tdb->transaction->io_methods;
@@ -1036,8 +1017,8 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 			continue;
 		}
 
-		offset = i * getpagesize();
-		length = getpagesize();
+		offset = i * PAGESIZE;
+		length = PAGESIZE;
 		if (i == tdb->transaction->num_blocks-1) {
 			length = tdb->transaction->last_block_size;
 		}
@@ -1045,10 +1026,6 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 		ecode = methods->twrite(tdb, offset,
 					tdb->transaction->blocks[i], length);
 		if (ecode != TDB_SUCCESS) {
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				   "tdb_transaction_commit:"
-				   " write failed during commit");
-
 			/* we've overwritten part of the data and
 			   possibly expanded the file, so we need to
 			   run the crash recovery code */
@@ -1057,7 +1034,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 
 			_tdb_transaction_cancel(tdb);
 
-			return -1;
+			return tdb->last_error = ecode;
 		}
 		SAFE_FREE(tdb->transaction->blocks[i]);
 	}
@@ -1066,10 +1043,9 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 	tdb->transaction->num_blocks = 0;
 
 	/* ensure the new data is on disk */
-	ecode = transaction_sync(tdb, 0, tdb->map_size);
+	ecode = transaction_sync(tdb, 0, tdb->file->map_size);
 	if (ecode != TDB_SUCCESS) {
-		tdb->ecode = ecode;
-		return -1;
+		return tdb->last_error = ecode;
 	}
 
 	/*
@@ -1088,10 +1064,11 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 #endif
 
 	/* use a transaction cancel to free memory and remove the
-	   transaction locks */
+	   transaction locks: it "restores" map_size, too. */
+	tdb->transaction->old_map_size = tdb->file->map_size;
 	_tdb_transaction_cancel(tdb);
 
-	return 0;
+	return tdb->last_error = TDB_SUCCESS;
 }
 
 
@@ -1100,7 +1077,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
   database write access already established (including the open
   lock to prevent new processes attaching)
 */
-int tdb_transaction_recover(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb)
 {
 	tdb_off_t recovery_head, recovery_eof;
 	unsigned char *data, *p;
@@ -1110,56 +1087,51 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 	/* find the recovery area */
 	recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
 	if (TDB_OFF_IS_ERR(recovery_head)) {
-		tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
-			 "tdb_transaction_recover:"
-			 " failed to read recovery head");
-		return -1;
+		return tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to read recovery head");
 	}
 
 	if (recovery_head == 0) {
 		/* we have never allocated a recovery record */
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	/* read the recovery record */
 	ecode = tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec));
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			   "tdb_transaction_recover:"
-			   " failed to read recovery record");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to read recovery record");
 	}
 
 	if (rec.magic != TDB_RECOVERY_MAGIC) {
 		/* there is no valid recovery data */
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	if (tdb->read_only) {
-		tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
-			   "tdb_transaction_recover:"
-			   " attempt to recover read only database");
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " attempt to recover read only database");
 	}
 
 	recovery_eof = rec.eof;
 
 	data = (unsigned char *)malloc(rec.len);
 	if (data == NULL) {
-		tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
-			   "tdb_transaction_recover:"
-			   " failed to allocate recovery data");
-		return -1;
+		return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to allocate recovery data");
 	}
 
 	/* read the full recovery data */
 	ecode = tdb->methods->tread(tdb, recovery_head + sizeof(rec), data,
 				    rec.len);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			   "tdb_transaction_recover:"
-			   " failed to read recovery data");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to read recovery data");
 	}
 
 	/* recover the file data */
@@ -1175,22 +1147,22 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 		ecode = tdb->methods->twrite(tdb, ofs, p, len);
 		if (ecode != TDB_SUCCESS) {
 			free(data);
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				 "tdb_transaction_recover:"
-				 " failed to recover %zu bytes at offset %zu",
-				 (size_t)len, (size_t)ofs);
-			return -1;
+			return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+					  "tdb_transaction_recover:"
+					  " failed to recover %zu bytes"
+					  " at offset %zu",
+					  (size_t)len, (size_t)ofs);
 		}
 		p += len;
 	}
 
 	free(data);
 
-	ecode = transaction_sync(tdb, 0, tdb->map_size);
+	ecode = transaction_sync(tdb, 0, tdb->file->map_size);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			   "tdb_transaction_recover: failed to sync recovery");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to sync recovery");
 	}
 
 	/* if the recovery area is after the recovered eof then remove it */
@@ -1199,10 +1171,9 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 						    recovery),
 				      0);
 		if (ecode != TDB_SUCCESS) {
-			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-				 "tdb_transaction_recover:"
-				 " failed to remove recovery head");
-			return -1;
+			return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+					  "tdb_transaction_recover:"
+					  " failed to remove recovery head");
 		}
 	}
 
@@ -1212,17 +1183,16 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 			      + offsetof(struct tdb_recovery_record, magic),
 			      TDB_RECOVERY_INVALID_MAGIC);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			 "tdb_transaction_recover:"
-			 " failed to remove recovery magic");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to remove recovery magic");
 	}
 
 	ecode = transaction_sync(tdb, 0, recovery_eof);
 	if (ecode != TDB_SUCCESS) {
-		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
-			 "tdb_transaction_recover: failed to sync2 recovery");
-		return -1;
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to sync2 recovery");
 	}
 
 	tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
@@ -1230,11 +1200,10 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 		   (size_t)recovery_eof);
 
 	/* all done */
-	return 0;
+	return TDB_SUCCESS;
 }
 
-/* Any I/O failures we say "needs recovery". */
-bool tdb_needs_recovery(struct tdb_context *tdb)
+tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb)
 {
 	tdb_off_t recovery_head;
 	struct tdb_recovery_record rec;
@@ -1243,8 +1212,7 @@ bool tdb_needs_recovery(struct tdb_context *tdb)
 	/* find the recovery area */
 	recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
 	if (TDB_OFF_IS_ERR(recovery_head)) {
-		tdb->ecode = recovery_head;
-		return true;
+		return recovery_head;
 	}
 
 	if (recovery_head == 0) {
@@ -1255,8 +1223,7 @@ bool tdb_needs_recovery(struct tdb_context *tdb)
 	/* read the recovery record */
 	ecode = tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec));
 	if (ecode != TDB_SUCCESS) {
-		tdb->ecode = ecode;
-		return true;
+		return ecode;
 	}
 
 	return (rec.magic == TDB_RECOVERY_MAGIC);