X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Ftransaction.c;h=604861b738e75e9c7efab447f7d26d19780a5ffc;hp=53bcc21c62f45ae687f6a9c6876e015f9e297cf1;hb=3835bb95d716cb6448913792b0d8d539877d93e9;hpb=5e8b9af5e7fe5f1ccac407873a3b782b8a629782

diff --git a/ccan/tdb2/transaction.c b/ccan/tdb2/transaction.c
index 53bcc21c..604861b7 100644
--- a/ccan/tdb2/transaction.c
+++ b/ccan/tdb2/transaction.c
@@ -125,16 +125,18 @@ struct tdb_transaction {
   read while in a transaction. We need to check first if the data is in our list
   of transaction elements, then if not do a real read
 */
-static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
-			    tdb_len_t len)
+static enum TDB_ERROR transaction_read(struct tdb_context *tdb, tdb_off_t off,
+				       void *buf, tdb_len_t len)
 {
 	size_t blk;
+	enum TDB_ERROR ecode;
 
 	/* break it down into block sized ops */
 	while (len + (off % getpagesize()) > getpagesize()) {
 		tdb_len_t len2 = getpagesize() - (off % getpagesize());
-		if (transaction_read(tdb, off, buf, len2) != 0) {
-			return -1;
+		ecode = transaction_read(tdb, off, buf, len2);
+		if (ecode != TDB_SUCCESS) {
+			return ecode;
 		}
 		len -= len2;
 		off += len2;
@@ -142,7 +144,7 @@ static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
 	}
 
 	if (len == 0) {
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	blk = off / getpagesize();
@@ -151,7 +153,8 @@ static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
 	if (tdb->transaction->num_blocks <= blk ||
 	    tdb->transaction->blocks[blk] == NULL) {
 		/* nope, do a real read */
-		if (tdb->transaction->io_methods->read(tdb, off, buf, len) != 0) {
+		ecode = tdb->transaction->io_methods->tread(tdb, off, buf, len);
+		if (ecode != TDB_SUCCESS) {
 			goto fail;
 		}
 		return 0;
@@ -160,46 +163,45 @@ static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
 	/* 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) {
+			ecode = TDB_ERR_IO;
 			goto fail;
 		}
 	}
 
 	/* now copy it out of this block */
 	memcpy(buf, tdb->transaction->blocks[blk] + (off % getpagesize()), len);
-	return 0;
+	return TDB_SUCCESS;
 
 fail:
-	tdb->ecode = TDB_ERR_IO;
-	tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-		 "transaction_read: failed at off=%llu len=%llu\n",
-		 (long long)off, (long long)len);
 	tdb->transaction->transaction_error = 1;
-	return -1;
+	return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+			  "transaction_read: failed at off=%zu len=%zu",
+			  (size_t)off, (size_t)len);
 }
 
 
 /*
   write while in a transaction
 */
-static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
-			     const void *buf, tdb_len_t len)
+static enum TDB_ERROR transaction_write(struct tdb_context *tdb, tdb_off_t off,
+					const void *buf, tdb_len_t len)
 {
 	size_t blk;
+	enum TDB_ERROR ecode;
 
 	/* Only a commit is allowed on a prepared transaction */
 	if (tdb->transaction->prepared) {
-		tdb->ecode = TDB_ERR_EINVAL;
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "transaction_write: transaction already prepared,"
-			 " write not allowed\n");
-		tdb->transaction->transaction_error = 1;
-		return -1;
+		ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_ERROR,
+				   "transaction_write: transaction already"
+				   " prepared, write not allowed");
+		goto fail;
 	}
 
 	/* break it up into block sized chunks */
 	while (len + (off % getpagesize()) > getpagesize()) {
 		tdb_len_t len2 = getpagesize() - (off % getpagesize());
-		if (transaction_write(tdb, off, buf, len2) != 0) {
+		ecode = transaction_write(tdb, off, buf, len2);
+		if (ecode != TDB_SUCCESS) {
 			return -1;
 		}
 		len -= len2;
@@ -210,7 +212,7 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
 	}
 
 	if (len == 0) {
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	blk = off / getpagesize();
@@ -228,7 +230,9 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
 				(blk+1)*sizeof(uint8_t *));
 		}
 		if (new_blocks == NULL) {
-			tdb->ecode = TDB_ERR_OOM;
+			ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+					   "transaction_write:"
+					   " failed to allocate");
 			goto fail;
 		}
 		memset(&new_blocks[tdb->transaction->num_blocks], 0,
@@ -242,18 +246,27 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
 	if (tdb->transaction->blocks[blk] == NULL) {
 		tdb->transaction->blocks[blk] = (uint8_t *)calloc(getpagesize(), 1);
 		if (tdb->transaction->blocks[blk] == NULL) {
-			tdb->ecode = TDB_ERR_OOM;
-			tdb->transaction->transaction_error = 1;
-			return -1;
+			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->io_methods->read(tdb, blk * getpagesize(),
-							       tdb->transaction->blocks[blk],
-							       len2) != 0) {
+			ecode = tdb->transaction->io_methods->tread(tdb,
+					blk * getpagesize(),
+					tdb->transaction->blocks[blk],
+					len2);
+			if (ecode != TDB_SUCCESS) {
+				ecode = tdb_logerr(tdb, ecode,
+						   TDB_LOG_ERROR,
+						   "transaction_write:"
+						   " failed to"
+						   " read old block: %s",
+						   strerror(errno));
 				SAFE_FREE(tdb->transaction->blocks[blk]);
 				goto fail;
 			}
@@ -275,15 +288,11 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
 		}
 	}
 
-	return 0;
+	return TDB_SUCCESS;
 
 fail:
-	tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-		 "transaction_write: failed at off=%llu len=%llu\n",
-		 (long long)((blk*getpagesize()) + off),
-		 (long long)len);
 	tdb->transaction->transaction_error = 1;
-	return -1;
+	return ecode;
 }
 
 
@@ -335,34 +344,74 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
 /*
   out of bounds check during a transaction
 */
-static int transaction_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
+static enum TDB_ERROR transaction_oob(struct tdb_context *tdb, tdb_off_t len,
+				      bool probe)
 {
 	if (len <= tdb->map_size) {
-		return 0;
+		return TDB_SUCCESS;
 	}
-	tdb->ecode = TDB_ERR_IO;
-	return -1;
+	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);
+	}
+	return TDB_ERR_IO;
 }
 
 /*
   transaction version of tdb_expand().
 */
-static int transaction_expand_file(struct tdb_context *tdb, tdb_off_t addition)
+static enum TDB_ERROR transaction_expand_file(struct tdb_context *tdb,
+					      tdb_off_t addition)
 {
+	enum TDB_ERROR ecode;
+
 	/* add a write to the transaction elements, so subsequent
 	   reads see the zero data */
-	if (transaction_write(tdb, tdb->map_size, NULL, addition) != 0) {
-		return -1;
+	ecode = transaction_write(tdb, tdb->map_size, NULL, addition);
+	if (ecode != TDB_SUCCESS) {
+		tdb->ecode = ecode;
+		return ecode;
 	}
 	tdb->map_size += addition;
-	return 0;
+	return ecode;
 }
 
 static void *transaction_direct(struct tdb_context *tdb, tdb_off_t off,
-				size_t len)
+				size_t len, bool write_mode)
 {
-	/* FIXME */
-	return NULL;
+	size_t blk = off / getpagesize(), end_blk;
+
+	/* This is wrong for zero-length blocks, but will fail gracefully */
+	end_blk = (off + len - 1) / getpagesize();
+
+	/* Can only do direct if in single block and we've already copied. */
+	if (write_mode) {
+		if (blk != end_blk)
+			return NULL;
+		if (blk >= tdb->transaction->num_blocks)
+			return NULL;
+		if (tdb->transaction->blocks[blk] == NULL)
+			return NULL;
+		return tdb->transaction->blocks[blk] + off % getpagesize();
+	}
+
+	/* 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();
+
+	/* Otherwise must be all not copied. */
+	while (blk < end_blk) {
+		if (blk >= tdb->transaction->num_blocks)
+			break;
+		if (tdb->transaction->blocks[blk])
+			return NULL;
+		blk++;
+	}
+	return tdb->transaction->io_methods->direct(tdb, off, len, false);
 }
 
 static const struct tdb_methods transaction_methods = {
@@ -376,43 +425,41 @@ static const struct tdb_methods transaction_methods = {
 /*
   sync to disk
 */
-static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t length)
+static enum TDB_ERROR transaction_sync(struct tdb_context *tdb,
+				       tdb_off_t offset, tdb_len_t length)
 {
 	if (tdb->flags & TDB_NOSYNC) {
-		return 0;
+		return TDB_SUCCESS;
 	}
 
 	if (fsync(tdb->fd) != 0) {
-		tdb->ecode = TDB_ERR_IO;
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_transaction: fsync failed\n");
-		return -1;
+		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,
 			  length + (offset - moffset), MS_SYNC) != 0) {
-			tdb->ecode = TDB_ERR_IO;
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-				 "tdb_transaction: msync failed - %s\n",
-				 strerror(errno));
-			return -1;
+			return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+					  "tdb_transaction: msync failed: %s",
+					  strerror(errno));
 		}
 	}
 #endif
-	return 0;
+	return TDB_SUCCESS;
 }
 
 
 static void _tdb_transaction_cancel(struct tdb_context *tdb)
 {
 	int i;
+	enum TDB_ERROR ecode;
 
 	if (tdb->transaction == NULL) {
-		tdb->ecode = TDB_ERR_EINVAL;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_cancel: no transaction\n");
+		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+			   "tdb_transaction_cancel: no transaction");
 		return;
 	}
 
@@ -437,13 +484,16 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
 		uint64_t invalid = TDB_RECOVERY_INVALID_MAGIC;
 
 		/* remove the recovery marker */
-		if (methods->write(tdb, tdb->transaction->magic_offset,
-				   &invalid, sizeof(invalid)) == -1 ||
-		    transaction_sync(tdb, tdb->transaction->magic_offset,
-				     sizeof(invalid)) == -1) {
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-				 "tdb_transaction_cancel: failed to remove"
-				 " recovery magic\n");
+		ecode = methods->twrite(tdb, tdb->transaction->magic_offset,
+					&invalid, sizeof(invalid));
+		if (ecode == TDB_SUCCESS)
+			ecode = transaction_sync(tdb,
+						 tdb->transaction->magic_offset,
+						 sizeof(invalid));
+		if (ecode != TDB_SUCCESS) {
+			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				   "tdb_transaction_cancel: failed to remove"
+				   " recovery magic");
 		}
 	}
 
@@ -454,7 +504,6 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
 	tdb->methods = tdb->transaction->io_methods;
 
 	tdb_transaction_unlock(tdb, F_WRLCK);
-	tdb_unlock_expand(tdb, F_WRLCK);
 
 	if (tdb_has_open_lock(tdb))
 		tdb_unlock_open(tdb);
@@ -468,18 +517,21 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
 */
 int tdb_transaction_start(struct tdb_context *tdb)
 {
+	enum TDB_ERROR ecode;
+
 	/* some sanity checks */
 	if (tdb->read_only || (tdb->flags & TDB_INTERNAL)) {
-		tdb->ecode = TDB_ERR_EINVAL;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_start: cannot start a transaction"
-			 " on a read-only or internal db\n");
+		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;
 	}
 
 	/* cope with nested tdb_transaction_start() calls */
 	if (tdb->transaction != NULL) {
-		tdb->ecode = TDB_ERR_NESTING;
+		tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
+			   "tdb_transaction_start:"
+			   " already inside transaction");
 		return -1;
 	}
 
@@ -487,24 +539,26 @@ int tdb_transaction_start(struct tdb_context *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->ecode = TDB_ERR_LOCK;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_start: cannot start a transaction"
-			 " with locks held\n");
+		tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+			   "tdb_transaction_start: cannot start a transaction"
+			   " with locks held");
 		return -1;
 	}
 
 	tdb->transaction = (struct tdb_transaction *)
 		calloc(sizeof(struct tdb_transaction), 1);
 	if (tdb->transaction == NULL) {
-		tdb->ecode = TDB_ERR_OOM;
+		tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+			   "tdb_transaction_start: cannot allocate");
 		return -1;
 	}
 
 	/* 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_transaction_lock(tdb, F_WRLCK) == -1) {
+	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;
@@ -512,14 +566,12 @@ int tdb_transaction_start(struct tdb_context *tdb)
 
 	/* get a read lock over entire file. This is upgraded to a write
 	   lock during the commit */
-	if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true) == -1) {
+	ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true);
+	if (ecode != TDB_SUCCESS) {
+		tdb->ecode = ecode;
 		goto fail_allrecord_lock;
 	}
 
-	if (tdb_lock_expand(tdb, F_WRLCK) != 0) {
-		goto fail_expand_lock;
-	}
-
 	/* make sure we know about any file expansions already done by
 	   anyone else */
 	tdb->methods->oob(tdb, tdb->map_size + 1, true);
@@ -531,8 +583,6 @@ int tdb_transaction_start(struct tdb_context *tdb)
 	tdb->methods = &transaction_methods;
 	return 0;
 
-fail_expand_lock:
-	tdb_allrecord_unlock(tdb, F_RDLCK);
 fail_allrecord_lock:
 	tdb_transaction_unlock(tdb, F_WRLCK);
 	SAFE_FREE(tdb->transaction->blocks);
@@ -589,20 +639,22 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 	const struct tdb_methods *methods = tdb->transaction->io_methods;
 	tdb_off_t recovery_head;
 	size_t addition;
+	enum TDB_ERROR ecode;
 
 	recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
-	if (recovery_head == TDB_OFF_ERR) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+	if (TDB_OFF_IS_ERR(recovery_head)) {
+		tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
 			 "tdb_recovery_allocate:"
-			 " failed to read recovery head\n");
+			 " failed to read recovery head");
 		return -1;
 	}
 
 	if (recovery_head != 0) {
-		if (methods->read(tdb, recovery_head, &rec, sizeof(rec))) {
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+		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\n");
+				 " failed to read recovery record");
 			return -1;
 		}
 		tdb_convert(tdb, &rec, sizeof(rec));
@@ -628,11 +680,13 @@ 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) {
-		if (add_free_record(tdb, recovery_head,
-				    sizeof(rec) + rec.max_len) != 0) {
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-				 "tdb_recovery_allocate:"
-				 " failed to free previous recovery area\n");
+		add_stat(tdb, frees, 1);
+		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;
 		}
 	}
@@ -655,10 +709,11 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 	addition = (tdb->map_size - tdb->transaction->old_map_size) +
 		sizeof(rec) + *recovery_max_size;
 	tdb->map_size = tdb->transaction->old_map_size;
-	if (methods->expand_file(tdb, addition) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+	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\n");
+			 " failed to create recovery area");
 		return -1;
 	}
 
@@ -670,11 +725,12 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 	/* 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 */
 	tdb_convert(tdb, &recovery_head, sizeof(recovery_head));
-	if (methods->write(tdb, offsetof(struct tdb_header, recovery),
-			   &recovery_head, sizeof(tdb_off_t)) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+	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\n");
+			 " failed to write recovery head");
 		return -1;
 	}
 	transaction_write_existing(tdb, offsetof(struct tdb_header, recovery),
@@ -709,6 +765,7 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	tdb_off_t old_map_size = tdb->transaction->old_map_size;
 	uint64_t magic, tailer;
 	int i;
+	enum TDB_ERROR ecode;
 
 	/*
 	  check that the recovery area has enough space
@@ -720,7 +777,8 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 
 	data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
 	if (data == NULL) {
-		tdb->ecode = TDB_ERR_OOM;
+		tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+			   "transaction_setup_recovery: cannot allocate");
 		return -1;
 	}
 
@@ -750,10 +808,9 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 			continue;
 		}
 		if (offset + length > tdb->map_size) {
-			tdb->ecode = TDB_ERR_CORRUPT;
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-				 "tdb_transaction_setup_recovery:"
-				 " transaction data over new region boundary\n");
+			tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
+				   "tdb_transaction_setup_recovery:"
+				   " transaction data over new region boundary");
 			free(data);
 			return -1;
 		}
@@ -764,9 +821,11 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 		/* 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->read(tdb, offset,
-				  p + sizeof(offset) + sizeof(length),
-				  length) != 0) {
+		ecode = methods->tread(tdb, offset,
+				       p + sizeof(offset) + sizeof(length),
+				       length);
+		if (ecode != TDB_SUCCESS) {
+			tdb->ecode = ecode;
 			free(data);
 			return -1;
 		}
@@ -779,11 +838,12 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	tdb_convert(tdb, p, sizeof(tailer));
 
 	/* write the recovery data to the recovery area */
-	if (methods->write(tdb, recovery_offset, data,
-			   sizeof(*rec) + recovery_size) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+	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\n");
+			 " failed to write recovery data");
 		free(data);
 		return -1;
 	}
@@ -793,9 +853,11 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	/* as we don't have ordered writes, we have to sync the recovery
 	   data before we update the magic to indicate that the recovery
 	   data is present */
-	if (transaction_sync(tdb, recovery_offset,
-			     sizeof(*rec) + recovery_size) == -1) {
+	ecode = transaction_sync(tdb, recovery_offset,
+				 sizeof(*rec) + recovery_size);
+	if (ecode != TDB_SUCCESS) {
 		free(data);
+		tdb->ecode = ecode;
 		return -1;
 	}
 
@@ -807,16 +869,19 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	*magic_offset = recovery_offset + offsetof(struct tdb_recovery_record,
 						   magic);
 
-	if (methods->write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+	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\n");
+			 " failed to write recovery magic");
 		return -1;
 	}
 	transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic));
 
 	/* ensure the recovery magic marker is on disk */
-	if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
+	ecode = transaction_sync(tdb, *magic_offset, sizeof(magic));
+	if (ecode != TDB_SUCCESS) {
+		tdb->ecode = ecode;
 		return -1;
 	}
 
@@ -826,29 +891,27 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 {
 	const struct tdb_methods *methods;
+	enum TDB_ERROR ecode;
 
 	if (tdb->transaction == NULL) {
-		tdb->ecode = TDB_ERR_EINVAL;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_prepare_commit: no transaction\n");
+		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+			   "tdb_transaction_prepare_commit: no transaction");
 		return -1;
 	}
 
 	if (tdb->transaction->prepared) {
-		tdb->ecode = TDB_ERR_EINVAL;
 		_tdb_transaction_cancel(tdb);
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_prepare_commit:"
-			 " transaction already prepared\n");
+		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+			   "tdb_transaction_prepare_commit:"
+			   " transaction already prepared");
 		return -1;
 	}
 
 	if (tdb->transaction->transaction_error) {
-		tdb->ecode = TDB_ERR_IO;
 		_tdb_transaction_cancel(tdb);
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_prepare_commit:"
-			 " transaction error pending\n");
+		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_ERROR,
+			   "tdb_transaction_prepare_commit:"
+			   " transaction error pending");
 		return -1;
 	}
 
@@ -866,30 +929,33 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 	methods = tdb->transaction->io_methods;
 
 	/* upgrade the main transaction lock region to a write lock */
-	if (tdb_allrecord_upgrade(tdb) == -1) {
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
+	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\n");
+			 " failed to upgrade hash locks");
 		_tdb_transaction_cancel(tdb);
 		return -1;
 	}
 
 	/* get the open lock - this prevents new users attaching to the database
 	   during the commit */
-	if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_prepare_commit:"
-			 " failed to get open lock\n");
+	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;
 	}
 
+	/* 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->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
+			tdb_logerr(tdb, tdb->ecode, TDB_LOG_ERROR,
 				 "tdb_transaction_prepare_commit:"
-				 " failed to setup recovery data\n");
+				 " failed to setup recovery data");
 			_tdb_transaction_cancel(tdb);
 			return -1;
 		}
@@ -902,10 +968,11 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 		tdb_len_t add = tdb->map_size - tdb->transaction->old_map_size;
 		/* Restore original map size for tdb_expand_file */
 		tdb->map_size = tdb->transaction->old_map_size;
-		if (methods->expand_file(tdb, add) == -1) {
-			tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
+		ecode = methods->expand_file(tdb, add);
+		if (ecode != TDB_SUCCESS) {
+			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
 				 "tdb_transaction_prepare_commit:"
-				 " expansion failed\n");
+				 " expansion failed");
 			_tdb_transaction_cancel(tdb);
 			return -1;
 		}
@@ -931,25 +998,16 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 {
 	const struct tdb_methods *methods;
 	int i;
+	enum TDB_ERROR ecode;
 
 	if (tdb->transaction == NULL) {
-		tdb->ecode = TDB_ERR_EINVAL;
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_commit: no transaction\n");
+		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+			 "tdb_transaction_commit: no transaction");
 		return -1;
 	}
 
 	tdb_trace(tdb, "tdb_transaction_commit");
 
-	if (tdb->transaction->transaction_error) {
-		tdb->ecode = TDB_ERR_IO;
-		tdb_transaction_cancel(tdb);
-		tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-			 "tdb_transaction_commit: transaction error pending\n");
-		return -1;
-	}
-
-
 	if (tdb->transaction->nesting != 0) {
 		tdb->transaction->nesting--;
 		return 0;
@@ -984,11 +1042,12 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 			length = tdb->transaction->last_block_size;
 		}
 
-		if (methods->write(tdb, offset, tdb->transaction->blocks[i],
-				   length) == -1) {
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-				 "tdb_transaction_commit:"
-				 " write failed during commit\n");
+		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
@@ -1007,7 +1066,9 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 	tdb->transaction->num_blocks = 0;
 
 	/* ensure the new data is on disk */
-	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
+	ecode = transaction_sync(tdb, 0, tdb->map_size);
+	if (ecode != TDB_SUCCESS) {
+		tdb->ecode = ecode;
 		return -1;
 	}
 
@@ -1044,13 +1105,14 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 	tdb_off_t recovery_head, recovery_eof;
 	unsigned char *data, *p;
 	struct tdb_recovery_record rec;
+	enum TDB_ERROR ecode;
 
 	/* find the recovery area */
 	recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
-	if (recovery_head == TDB_OFF_ERR) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+	if (TDB_OFF_IS_ERR(recovery_head)) {
+		tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
 			 "tdb_transaction_recover:"
-			 " failed to read recovery head\n");
+			 " failed to read recovery head");
 		return -1;
 	}
 
@@ -1060,10 +1122,11 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 	}
 
 	/* read the recovery record */
-	if (tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec)) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_transaction_recover:"
-			 " failed to read recovery record\n");
+	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;
 	}
 
@@ -1073,10 +1136,9 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 	}
 
 	if (tdb->read_only) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_transaction_recover:"
-			 " attempt to recover read only database\n");
-		tdb->ecode = TDB_ERR_CORRUPT;
+		tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
+			   "tdb_transaction_recover:"
+			   " attempt to recover read only database");
 		return -1;
 	}
 
@@ -1084,19 +1146,19 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 
 	data = (unsigned char *)malloc(rec.len);
 	if (data == NULL) {
-		tdb->ecode = TDB_ERR_OOM;
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_transaction_recover:"
-			 " failed to allocate recovery data\n");
+		tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+			   "tdb_transaction_recover:"
+			   " failed to allocate recovery data");
 		return -1;
 	}
 
 	/* read the full recovery data */
-	if (tdb->methods->read(tdb, recovery_head + sizeof(rec), data,
-			       rec.len) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_transaction_recover:"
-			 " failed to read recovery data\n");
+	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;
 	}
 
@@ -1110,11 +1172,12 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 		memcpy(&len, p + sizeof(ofs), sizeof(len));
 		p += sizeof(ofs) + sizeof(len);
 
-		if (tdb->methods->write(tdb, ofs, p, len) == -1) {
+		ecode = tdb->methods->twrite(tdb, ofs, p, len);
+		if (ecode != TDB_SUCCESS) {
 			free(data);
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
 				 "tdb_transaction_recover:"
-				 " failed to recover %zu bytes at offset %zu\n",
+				 " failed to recover %zu bytes at offset %zu",
 				 (size_t)len, (size_t)ofs);
 			return -1;
 		}
@@ -1123,43 +1186,48 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 
 	free(data);
 
-	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_transaction_recover: failed to sync recovery\n");
+	ecode = transaction_sync(tdb, 0, tdb->map_size);
+	if (ecode != TDB_SUCCESS) {
+		tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+			   "tdb_transaction_recover: failed to sync recovery");
 		return -1;
 	}
 
 	/* if the recovery area is after the recovered eof then remove it */
 	if (recovery_eof <= recovery_head) {
-		if (tdb_write_off(tdb, offsetof(struct tdb_header,recovery), 0)
-		    == -1) {
-			tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+		ecode = tdb_write_off(tdb, offsetof(struct tdb_header,
+						    recovery),
+				      0);
+		if (ecode != TDB_SUCCESS) {
+			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
 				 "tdb_transaction_recover:"
-				 " failed to remove recovery head\n");
+				 " failed to remove recovery head");
 			return -1;
 		}
 	}
 
 	/* remove the recovery magic */
-	if (tdb_write_off(tdb,
-			  recovery_head
-			  + offsetof(struct tdb_recovery_record, magic),
-			  TDB_RECOVERY_INVALID_MAGIC) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+	ecode = tdb_write_off(tdb,
+			      recovery_head
+			      + 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\n");
+			 " failed to remove recovery magic");
 		return -1;
 	}
 
-	if (transaction_sync(tdb, 0, recovery_eof) == -1) {
-		tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-			 "tdb_transaction_recover: failed to sync2 recovery\n");
+	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;
 	}
 
-	tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
-		 "tdb_transaction_recover: recovered %zu byte database\n",
-		 (size_t)recovery_eof);
+	tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
+		   "tdb_transaction_recover: recovered %zu byte database",
+		   (size_t)recovery_eof);
 
 	/* all done */
 	return 0;
@@ -1170,10 +1238,12 @@ bool tdb_needs_recovery(struct tdb_context *tdb)
 {
 	tdb_off_t recovery_head;
 	struct tdb_recovery_record rec;
+	enum TDB_ERROR ecode;
 
 	/* find the recovery area */
 	recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
-	if (recovery_head == TDB_OFF_ERR) {
+	if (TDB_OFF_IS_ERR(recovery_head)) {
+		tdb->ecode = recovery_head;
 		return true;
 	}
 
@@ -1183,7 +1253,9 @@ bool tdb_needs_recovery(struct tdb_context *tdb)
 	}
 
 	/* read the recovery record */
-	if (tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec)) == -1) {
+	ecode = tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec));
+	if (ecode != TDB_SUCCESS) {
+		tdb->ecode = ecode;
 		return true;
 	}