X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Ftransaction.c;h=1f5709bf437468ccb082085de0747952ee931149;hp=66be5bdc8e533cf67c49fddea59e2951ca10e863;hb=380372e733416c2b348d5307f536d0a0807e95df;hpb=04cf551d15ee93716aa0462adadc0a3891480813

diff --git a/ccan/tdb2/transaction.c b/ccan/tdb2/transaction.c
index 66be5bdc..1f5709bf 100644
--- a/ccan/tdb2/transaction.c
+++ b/ccan/tdb2/transaction.c
@@ -25,7 +25,7 @@
 */
 
 #include "private.h"
-#define SAFE_FREE(x) do { if ((x) != NULL) {free(x); (x)=NULL;} } while(0)
+#define SAFE_FREE(x) do { if ((x) != NULL) {free((void *)x); (x)=NULL;} } while(0)
 
 /*
   transaction design:
@@ -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
@@ -88,7 +88,6 @@
     fsync/msync calls are made.
 */
 
-
 /*
   hold the context of any current transaction
 */
@@ -110,7 +109,7 @@ struct tdb_transaction {
 	/* when inside a transaction we need to keep track of any
 	   nested tdb_transaction_start() calls, as these are allowed,
 	   but don't create a new transaction */
-	int nesting;
+	unsigned int nesting;
 
 	/* set when a prepare has already occurred */
 	bool prepared;
@@ -120,21 +119,25 @@ 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 65536
 
 /*
   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;
+	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;
 		}
 		len -= len2;
 		off += len2;
@@ -142,63 +145,65 @@ 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();
+	blk = off / PAGESIZE;
 
 	/* see if we have it in the block list */
-	if (tdb->transaction->num_blocks <= blk ||
-	    tdb->transaction->blocks[blk] == NULL) {
+	if (tdb->tdb2.transaction->num_blocks <= blk ||
+	    tdb->tdb2.transaction->blocks[blk] == NULL) {
 		/* nope, do a real read */
-		if (tdb->transaction->io_methods->tread(tdb, off, buf, len)
-		    != 0) {
+		ecode = tdb->tdb2.transaction->io_methods->tread(tdb, off, buf, len);
+		if (ecode != TDB_SUCCESS) {
 			goto fail;
 		}
 		return 0;
 	}
 
 	/* 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) {
+	if (blk == tdb->tdb2.transaction->num_blocks-1) {
+		if (len > tdb->tdb2.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;
+	memcpy(buf, tdb->tdb2.transaction->blocks[blk] + (off % PAGESIZE), len);
+	return TDB_SUCCESS;
 
 fail:
-	tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
-		   "transaction_read: failed at off=%zu len=%zu",
-		   (size_t)off, (size_t)len);
-	tdb->transaction->transaction_error = 1;
-	return -1;
+	tdb->tdb2.transaction->transaction_error = 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_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_FATAL,
-			 "transaction_write: transaction already prepared,"
-			 " write not allowed");
+	if (tdb->tdb2.transaction->prepared) {
+		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) {
-			return -1;
+	while (len + (off % PAGESIZE) > PAGESIZE) {
+		tdb_len_t len2 = PAGESIZE - (off % PAGESIZE);
+		ecode = transaction_write(tdb, off, buf, len2);
+		if (ecode != TDB_SUCCESS) {
+			return ecode;
 		}
 		len -= len2;
 		off += len2;
@@ -208,86 +213,92 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
 	}
 
 	if (len == 0) {
-		return 0;
+		return TDB_SUCCESS;
 	}
 
-	blk = off / getpagesize();
-	off = off % getpagesize();
+	blk = off / PAGESIZE;
+	off = off % PAGESIZE;
 
-	if (tdb->transaction->num_blocks <= blk) {
+	if (tdb->tdb2.transaction->num_blocks <= blk) {
 		uint8_t **new_blocks;
 		/* expand the blocks array */
-		if (tdb->transaction->blocks == NULL) {
+		if (tdb->tdb2.transaction->blocks == NULL) {
 			new_blocks = (uint8_t **)malloc(
 				(blk+1)*sizeof(uint8_t *));
 		} else {
 			new_blocks = (uint8_t **)realloc(
-				tdb->transaction->blocks,
+				tdb->tdb2.transaction->blocks,
 				(blk+1)*sizeof(uint8_t *));
 		}
 		if (new_blocks == NULL) {
-			tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-				   "transaction_write: failed to allocate");
+			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,
-		       (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;
+		memset(&new_blocks[tdb->tdb2.transaction->num_blocks], 0,
+		       (1+(blk - tdb->tdb2.transaction->num_blocks))*sizeof(uint8_t *));
+		tdb->tdb2.transaction->blocks = new_blocks;
+		tdb->tdb2.transaction->num_blocks = blk+1;
+		tdb->tdb2.transaction->last_block_size = 0;
 	}
 
 	/* allocate and fill a block? */
-	if (tdb->transaction->blocks[blk] == NULL) {
-		tdb->transaction->blocks[blk] = (uint8_t *)calloc(getpagesize(), 1);
-		if (tdb->transaction->blocks[blk] == NULL) {
-			tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-				   "transaction_write: failed to allocate");
+	if (tdb->tdb2.transaction->blocks[blk] == NULL) {
+		tdb->tdb2.transaction->blocks[blk] = (uint8_t *)calloc(PAGESIZE, 1);
+		if (tdb->tdb2.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->tdb2.transaction->old_map_size > blk * PAGESIZE) {
+			tdb_len_t len2 = PAGESIZE;
+			if (len2 + (blk * PAGESIZE) > tdb->tdb2.transaction->old_map_size) {
+				len2 = tdb->tdb2.transaction->old_map_size - (blk * PAGESIZE);
 			}
-			if (tdb->transaction->io_methods->tread(tdb, blk * getpagesize(),
-								tdb->transaction->blocks[blk],
-								len2) != 0) {
-				tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-					   "transaction_write: failed to"
-					   " read old block: %s",
-					   strerror(errno));
-				SAFE_FREE(tdb->transaction->blocks[blk]);
+			ecode = tdb->tdb2.transaction->io_methods->tread(tdb,
+					blk * PAGESIZE,
+					tdb->tdb2.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->tdb2.transaction->blocks[blk]);
 				goto fail;
 			}
-			if (blk == tdb->transaction->num_blocks-1) {
-				tdb->transaction->last_block_size = len2;
+			if (blk == tdb->tdb2.transaction->num_blocks-1) {
+				tdb->tdb2.transaction->last_block_size = len2;
 			}
 		}
 	}
 
 	/* overwrite part of an existing block */
 	if (buf == NULL) {
-		memset(tdb->transaction->blocks[blk] + off, 0, len);
+		memset(tdb->tdb2.transaction->blocks[blk] + off, 0, len);
 	} else {
-		memcpy(tdb->transaction->blocks[blk] + off, buf, len);
+		memcpy(tdb->tdb2.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;
+	if (blk == tdb->tdb2.transaction->num_blocks-1) {
+		if (len + off > tdb->tdb2.transaction->last_block_size) {
+			tdb->tdb2.transaction->last_block_size = len + off;
 		}
 	}
 
-	return 0;
+	return TDB_SUCCESS;
 
 fail:
-	tdb->transaction->transaction_error = 1;
-	return -1;
+	tdb->tdb2.transaction->transaction_error = 1;
+	return ecode;
 }
 
 
 /*
-  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,
@@ -296,8 +307,8 @@ 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;
@@ -310,93 +321,99 @@ 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) {
+	if (tdb->tdb2.transaction->num_blocks <= blk ||
+	    tdb->tdb2.transaction->blocks[blk] == NULL) {
 		return;
 	}
 
-	if (blk == tdb->transaction->num_blocks-1 &&
-	    off + len > tdb->transaction->last_block_size) {
-		if (off >= tdb->transaction->last_block_size) {
+	if (blk == tdb->tdb2.transaction->num_blocks-1 &&
+	    off + len > tdb->tdb2.transaction->last_block_size) {
+		if (off >= tdb->tdb2.transaction->last_block_size) {
 			return;
 		}
-		len = tdb->transaction->last_block_size - off;
+		len = tdb->tdb2.transaction->last_block_size - off;
 	}
 
 	/* overwrite part of an existing block */
-	memcpy(tdb->transaction->blocks[blk] + off, buf, len);
+	memcpy(tdb->tdb2.transaction->blocks[blk] + off, buf, len);
 }
 
 
 /*
   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;
-	}
-	tdb->ecode = TDB_ERR_IO;
-	if (!probe) {
-		tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
-			   "tdb_oob len %lld beyond transaction size %lld",
-			   (long long)len,
-			   (long long)tdb->map_size);
+	if (len <= tdb->file->map_size || probe) {
+		return TDB_SUCCESS;
 	}
-	return -1;
+
+	tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+		   "tdb_oob len %lld beyond transaction size %lld",
+		   (long long)len,
+		   (long long)tdb->file->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->file->map_size, NULL, addition);
+	if (ecode == TDB_SUCCESS) {
+		tdb->file->map_size += addition;
 	}
-	tdb->map_size += addition;
-	return 0;
+	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) {
-		if (blk != end_blk)
-			return NULL;
-		if (blk >= tdb->transaction->num_blocks)
-			return NULL;
-		if (tdb->transaction->blocks[blk] == NULL)
+		tdb->stats.transaction_write_direct++;
+		if (blk != end_blk
+		    || blk >= tdb->tdb2.transaction->num_blocks
+		    || tdb->tdb2.transaction->blocks[blk] == NULL) {
+			tdb->stats.transaction_write_direct_fail++;
 			return NULL;
-		return tdb->transaction->blocks[blk] + off % getpagesize();
+		}
+		return tdb->tdb2.transaction->blocks[blk] + off % PAGESIZE;
 	}
 
+	tdb->stats.transaction_read_direct++;
 	/* 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();
+	    && blk < tdb->tdb2.transaction->num_blocks
+	    && tdb->tdb2.transaction->blocks[blk])
+		return tdb->tdb2.transaction->blocks[blk] + off % PAGESIZE;
 
 	/* Otherwise must be all not copied. */
-	while (blk < end_blk) {
-		if (blk >= tdb->transaction->num_blocks)
+	while (blk <= end_blk) {
+		if (blk >= tdb->tdb2.transaction->num_blocks)
 			break;
-		if (tdb->transaction->blocks[blk])
+		if (tdb->tdb2.transaction->blocks[blk]) {
+			tdb->stats.transaction_read_direct_fail++;
 			return NULL;
+		}
 		blk++;
 	}
-	return tdb->transaction->io_methods->direct(tdb, off, len, false);
+	return tdb->tdb2.transaction->io_methods->direct(tdb, off, len, false);
 }
 
 static const struct tdb_methods transaction_methods = {
@@ -410,160 +427,194 @@ 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_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
-			   "tdb_transaction: fsync failed: %s",
-			   strerror(errno));
-		return -1;
+	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) {
+	if (tdb->file->map_ptr) {
 		tdb_off_t moffset = offset & ~(getpagesize()-1);
-		if (msync(moffset + (char *)tdb->map_ptr,
+		if (msync(moffset + (char *)tdb->file->map_ptr,
 			  length + (offset - moffset), MS_SYNC) != 0) {
-			tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
-				   "tdb_transaction: msync failed: %s",
-				   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_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
+	if (tdb->tdb2.transaction == NULL) {
+		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
 			   "tdb_transaction_cancel: no transaction");
 		return;
 	}
 
-	if (tdb->transaction->nesting != 0) {
-		tdb->transaction->transaction_error = 1;
-		tdb->transaction->nesting--;
+	if (tdb->tdb2.transaction->nesting != 0) {
+		tdb->tdb2.transaction->transaction_error = 1;
+		tdb->tdb2.transaction->nesting--;
 		return;
 	}
 
-	tdb->map_size = tdb->transaction->old_map_size;
+	tdb->file->map_size = tdb->tdb2.transaction->old_map_size;
 
 	/* 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]);
+	for (i=0;i<tdb->tdb2.transaction->num_blocks;i++) {
+		if (tdb->tdb2.transaction->blocks[i] != NULL) {
+			free(tdb->tdb2.transaction->blocks[i]);
 		}
 	}
-	SAFE_FREE(tdb->transaction->blocks);
+	SAFE_FREE(tdb->tdb2.transaction->blocks);
 
-	if (tdb->transaction->magic_offset) {
-		const struct tdb_methods *methods = tdb->transaction->io_methods;
+	if (tdb->tdb2.transaction->magic_offset) {
+		const struct tdb_methods *methods = tdb->tdb2.transaction->io_methods;
 		uint64_t invalid = TDB_RECOVERY_INVALID_MAGIC;
 
 		/* remove the recovery marker */
-		if (methods->twrite(tdb, tdb->transaction->magic_offset,
-				    &invalid, sizeof(invalid)) == -1 ||
-		    transaction_sync(tdb, tdb->transaction->magic_offset,
-				     sizeof(invalid)) == -1) {
-			tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
+		ecode = methods->twrite(tdb, tdb->tdb2.transaction->magic_offset,
+					&invalid, sizeof(invalid));
+		if (ecode == TDB_SUCCESS)
+			ecode = transaction_sync(tdb,
+						 tdb->tdb2.transaction->magic_offset,
+						 sizeof(invalid));
+		if (ecode != TDB_SUCCESS) {
+			tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
 				   "tdb_transaction_cancel: failed to remove"
 				   " recovery magic");
 		}
 	}
 
-	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;
+	tdb->tdb2.io = tdb->tdb2.transaction->io_methods;
 
 	tdb_transaction_unlock(tdb, F_WRLCK);
 
 	if (tdb_has_open_lock(tdb))
-		tdb_unlock_open(tdb);
+		tdb_unlock_open(tdb, F_WRLCK);
 
-	SAFE_FREE(tdb->transaction);
+	SAFE_FREE(tdb->tdb2.transaction);
 }
 
 /*
   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;
+
+	if (tdb->flags & TDB_VERSION1) {
+		if (tdb1_transaction_start(tdb) == -1)
+			return tdb->last_error;
+		return TDB_SUCCESS;
+	}
+
+	tdb->stats.transactions++;
 	/* some sanity checks */
-	if (tdb->read_only || (tdb->flags & TDB_INTERNAL)) {
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
-			   "tdb_transaction_start: cannot start a transaction"
-			   " on a read-only or internal db");
-		return -1;
+	if (tdb->flags & TDB_INTERNAL) {
+		return tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+						    TDB_LOG_USE_ERROR,
+						    "tdb_transaction_start:"
+						    " cannot start a"
+						    " transaction on an"
+						    " internal tdb");
+	}
+
+	if (tdb->flags & TDB_RDONLY) {
+		return tdb->last_error = tdb_logerr(tdb, TDB_ERR_RDONLY,
+						    TDB_LOG_USE_ERROR,
+						    "tdb_transaction_start:"
+						    " cannot start a"
+						    " transaction on a "
+						    " read-only tdb");
 	}
 
 	/* cope with nested tdb_transaction_start() calls */
-	if (tdb->transaction != NULL) {
-		tdb_logerr(tdb, TDB_ERR_NESTING, TDB_DEBUG_ERROR,
-			   "tdb_transaction_start:"
-			   " already inside transaction");
-		return -1;
+	if (tdb->tdb2.transaction != NULL) {
+		if (!(tdb->flags & TDB_ALLOW_NESTING)) {
+			return tdb->last_error
+				= tdb_logerr(tdb, TDB_ERR_IO,
+					     TDB_LOG_USE_ERROR,
+					     "tdb_transaction_start:"
+					     " already inside transaction");
+		}
+		tdb->tdb2.transaction->nesting++;
+		tdb->stats.transaction_nest++;
+		return 0;
 	}
 
 	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_DEBUG_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 *)
+	tdb->tdb2.transaction = (struct tdb_transaction *)
 		calloc(sizeof(struct tdb_transaction), 1);
-	if (tdb->transaction == NULL) {
-		tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
-			   "tdb_transaction_start: cannot allocate");
-		return -1;
+	if (tdb->tdb2.transaction == NULL) {
+		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
 	   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) {
-		SAFE_FREE(tdb->transaction->blocks);
-		SAFE_FREE(tdb->transaction);
-		return -1;
+	ecode = tdb_transaction_lock(tdb, F_WRLCK);
+	if (ecode != TDB_SUCCESS) {
+		SAFE_FREE(tdb->tdb2.transaction->blocks);
+		SAFE_FREE(tdb->tdb2.transaction);
+		return tdb->last_error = ecode;
 	}
 
 	/* 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) {
 		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->tdb2.io->oob(tdb, tdb->file->map_size + 1, true);
+	tdb->tdb2.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;
+	tdb->tdb2.transaction->io_methods = tdb->tdb2.io;
+	tdb->tdb2.io = &transaction_methods;
+	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;
+	SAFE_FREE(tdb->tdb2.transaction->blocks);
+	SAFE_FREE(tdb->tdb2.transaction);
+	return tdb->last_error = ecode;
 }
 
 
@@ -572,467 +623,536 @@ fail_allrecord_lock:
 */
 void tdb_transaction_cancel(struct tdb_context *tdb)
 {
+	if (tdb->flags & TDB_VERSION1) {
+		tdb1_transaction_cancel(tdb);
+		return;
+	}
+	tdb->stats.transaction_cancel++;
 	_tdb_transaction_cancel(tdb);
 }
 
 /*
-  work out how much space the linearised recovery data will consume
+  work out how much space the linearised recovery data will consume (worst case)
 */
 static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
 {
 	tdb_len_t recovery_size = 0;
 	int i;
 
-	recovery_size = sizeof(tdb_len_t);
-	for (i=0;i<tdb->transaction->num_blocks;i++) {
-		if (i * getpagesize() >= tdb->transaction->old_map_size) {
+	recovery_size = 0;
+	for (i=0;i<tdb->tdb2.transaction->num_blocks;i++) {
+		if (i * PAGESIZE >= tdb->tdb2.transaction->old_map_size) {
 			break;
 		}
-		if (tdb->transaction->blocks[i] == NULL) {
+		if (tdb->tdb2.transaction->blocks[i] == NULL) {
 			continue;
 		}
 		recovery_size += 2*sizeof(tdb_off_t);
-		if (i == tdb->transaction->num_blocks-1) {
-			recovery_size += tdb->transaction->last_block_size;
+		if (i == tdb->tdb2.transaction->num_blocks-1) {
+			recovery_size += tdb->tdb2.transaction->last_block_size;
 		} else {
-			recovery_size += getpagesize();
+			recovery_size += PAGESIZE;
 		}
 	}
 
 	return recovery_size;
 }
 
-/*
-  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_area(struct tdb_context *tdb,
+					const struct tdb_methods *methods,
+					tdb_off_t *recovery_offset,
+					struct tdb_recovery_record *rec)
 {
-	struct tdb_recovery_record rec;
-	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_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_recovery_allocate:"
-			 " failed to read recovery head");
-		return -1;
-	}
-
-	if (recovery_head != 0) {
-		if (methods->tread(tdb, recovery_head, &rec, sizeof(rec))) {
-			tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-				 "tdb_recovery_allocate:"
-				 " failed to read recovery record");
-			return -1;
-		}
-		tdb_convert(tdb, &rec, sizeof(rec));
-		/* ignore invalid recovery regions: can happen in crash */
-		if (rec.magic != TDB_RECOVERY_MAGIC &&
-		    rec.magic != TDB_RECOVERY_INVALID_MAGIC) {
-			recovery_head = 0;
+	*recovery_offset = tdb_read_off(tdb,
+					offsetof(struct tdb_header, recovery));
+	if (TDB_OFF_IS_ERR(*recovery_offset)) {
+		return *recovery_offset;
+	}
+
+	if (*recovery_offset == 0) {
+		rec->max_len = 0;
+		return TDB_SUCCESS;
+	}
+
+	ecode = methods->tread(tdb, *recovery_offset, rec, sizeof(*rec));
+	if (ecode != TDB_SUCCESS)
+		return ecode;
+
+	tdb_convert(tdb, rec, sizeof(*rec));
+	/* ignore invalid recovery regions: can happen in crash */
+	if (rec->magic != TDB_RECOVERY_MAGIC &&
+	    rec->magic != TDB_RECOVERY_INVALID_MAGIC) {
+		*recovery_offset = 0;
+		rec->max_len = 0;
+	}
+	return TDB_SUCCESS;
+}
+
+static unsigned int same(const unsigned char *new,
+			 const unsigned char *old,
+			 unsigned int length)
+{
+	unsigned int i;
+
+	for (i = 0; i < length; i++) {
+		if (new[i] != old[i])
+			break;
+	}
+	return i;
+}
+
+static unsigned int different(const unsigned char *new,
+			      const unsigned char *old,
+			      unsigned int length,
+			      unsigned int min_same,
+			      unsigned int *samelen)
+{
+	unsigned int i;
+
+	*samelen = 0;
+	for (i = 0; i < length; i++) {
+		if (new[i] == old[i]) {
+			(*samelen)++;
+		} else {
+			if (*samelen >= min_same) {
+				return i - *samelen;
+			}
+			*samelen = 0;
 		}
 	}
 
-	*recovery_size = tdb_recovery_size(tdb);
+	if (*samelen < min_same)
+		*samelen = 0;
+	return length - *samelen;
+}
 
-	if (recovery_head != 0 && *recovery_size <= rec.max_len) {
-		/* it fits in the existing area */
-		*recovery_max_size = rec.max_len;
-		*recovery_offset = recovery_head;
-		return 0;
+/* Allocates recovery blob, without tdb_recovery_record at head set up. */
+static struct tdb_recovery_record *alloc_recovery(struct tdb_context *tdb,
+						  tdb_len_t *len)
+{
+	struct tdb_recovery_record *rec;
+	size_t i;
+	enum TDB_ERROR ecode;
+	unsigned char *p;
+	const struct tdb_methods *old_methods = tdb->tdb2.io;
+
+	rec = malloc(sizeof(*rec) + tdb_recovery_size(tdb));
+	if (!rec) {
+		tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+			   "transaction_setup_recovery:"
+			   " cannot allocate");
+		return TDB_ERR_PTR(TDB_ERR_OOM);
 	}
 
-	/* we need to free up the old recovery area, then allocate a
-	   new one at the end of the file. Note that we cannot use
-	   normal allocation to allocate the new one as that might return
-	   us an area that is being currently used (as of the start of
-	   the transaction) */
-	if (recovery_head != 0) {
-		add_stat(tdb, frees, 1);
-		if (add_free_record(tdb, recovery_head,
-				    sizeof(rec) + rec.max_len) != 0) {
-			tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-				   "tdb_recovery_allocate:"
-				   " failed to free previous recovery area");
-			return -1;
+	/* We temporarily revert to the old I/O methods, so we can use
+	 * tdb_access_read */
+	tdb->tdb2.io = tdb->tdb2.transaction->io_methods;
+
+	/* build the recovery data into a single blob to allow us to do a single
+	   large write, which should be more efficient */
+	p = (unsigned char *)(rec + 1);
+	for (i=0;i<tdb->tdb2.transaction->num_blocks;i++) {
+		tdb_off_t offset;
+		tdb_len_t length;
+		unsigned int off;
+		const unsigned char *buffer;
+
+		if (tdb->tdb2.transaction->blocks[i] == NULL) {
+			continue;
+		}
+
+		offset = i * PAGESIZE;
+		length = PAGESIZE;
+		if (i == tdb->tdb2.transaction->num_blocks-1) {
+			length = tdb->tdb2.transaction->last_block_size;
+		}
+
+		if (offset >= tdb->tdb2.transaction->old_map_size) {
+			continue;
+		}
+
+		if (offset + length > tdb->file->map_size) {
+			ecode = tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
+					   "tdb_transaction_setup_recovery:"
+					   " transaction data over new region"
+					   " boundary");
+			goto fail;
+		}
+		if (offset + length > tdb->tdb2.transaction->old_map_size) {
+			/* Short read at EOF. */
+			length = tdb->tdb2.transaction->old_map_size - offset;
 		}
+		buffer = tdb_access_read(tdb, offset, length, false);
+		if (TDB_PTR_IS_ERR(buffer)) {
+			ecode = TDB_PTR_ERR(buffer);
+			goto fail;
+		}
+
+		/* Skip over anything the same at the start. */
+		off = same(tdb->tdb2.transaction->blocks[i], buffer, length);
+		offset += off;
+
+		while (off < length) {
+			tdb_len_t len;
+			unsigned int samelen;
+
+			len = different(tdb->tdb2.transaction->blocks[i] + off,
+					buffer + off, length - off,
+					sizeof(offset) + sizeof(len) + 1,
+					&samelen);
+
+			memcpy(p, &offset, sizeof(offset));
+			memcpy(p + sizeof(offset), &len, sizeof(len));
+			tdb_convert(tdb, p, sizeof(offset) + sizeof(len));
+			p += sizeof(offset) + sizeof(len);
+			memcpy(p, buffer + off, len);
+			p += len;
+			off += len + samelen;
+			offset += len + samelen;
+		}
+		tdb_access_release(tdb, buffer);
 	}
 
-	/* the tdb_free() call might have increased the recovery size */
-	*recovery_size = tdb_recovery_size(tdb);
+	*len = p - (unsigned char *)(rec + 1);
+	tdb->tdb2.io = old_methods;
+	return rec;
 
-	/* round up to a multiple of page size */
-	*recovery_max_size
-		= (((sizeof(rec) + *recovery_size) + getpagesize()-1)
-		   & ~(getpagesize()-1))
-		- sizeof(rec);
-	*recovery_offset = tdb->map_size;
-	recovery_head = *recovery_offset;
+fail:
+	free(rec);
+	tdb->tdb2.io = old_methods;
+	return TDB_ERR_PTR(ecode);
+}
+
+static tdb_off_t create_recovery_area(struct tdb_context *tdb,
+				      tdb_len_t rec_length,
+				      struct tdb_recovery_record *rec)
+{
+	tdb_off_t off, recovery_off;
+	tdb_len_t addition;
+	enum TDB_ERROR ecode;
+	const struct tdb_methods *methods = tdb->tdb2.transaction->io_methods;
+
+	/* round up to a multiple of page size. Overallocate, since each
+	 * such allocation forces us to expand the file. */
+	rec->max_len
+		= (((sizeof(*rec) + rec_length + rec_length / 2)
+		    + PAGESIZE-1) & ~(PAGESIZE-1))
+		- sizeof(*rec);
+	off = tdb->file->map_size;
 
 	/* 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) +
-		sizeof(rec) + *recovery_max_size;
-	tdb->map_size = tdb->transaction->old_map_size;
-	if (methods->expand_file(tdb, addition) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_recovery_allocate:"
-			 " failed to create recovery area");
-		return -1;
+	addition = (tdb->file->map_size - tdb->tdb2.transaction->old_map_size) +
+		sizeof(*rec) + rec->max_len;
+	tdb->file->map_size = tdb->tdb2.transaction->old_map_size;
+	tdb->stats.transaction_expand_file++;
+	ecode = methods->expand_file(tdb, addition);
+	if (ecode != TDB_SUCCESS) {
+		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->tdb2.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 */
-	tdb_convert(tdb, &recovery_head, sizeof(recovery_head));
-	if (methods->twrite(tdb, offsetof(struct tdb_header, recovery),
-			    &recovery_head, sizeof(tdb_off_t)) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_recovery_allocate:"
-			 " failed to write recovery head");
-		return -1;
+	recovery_off = off;
+	tdb_convert(tdb, &recovery_off, sizeof(recovery_off));
+	ecode = methods->twrite(tdb, offsetof(struct tdb_header, recovery),
+				&recovery_off, sizeof(tdb_off_t));
+	if (ecode != TDB_SUCCESS) {
+		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,
+				   &recovery_off,
 				   sizeof(tdb_off_t));
-	return 0;
-}
-
-/* Set up header for the recovery record. */
-static void set_recovery_header(struct tdb_recovery_record *rec,
-				uint64_t magic,
-				uint64_t datalen, uint64_t actuallen,
-				uint64_t oldsize)
-{
-	rec->magic = magic;
-	rec->max_len = actuallen;
-	rec->len = datalen;
-	rec->eof = oldsize;
+	return off;
 }
 
 /*
   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_len_t recovery_size;
-	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;
-
-	/*
-	  check that the recovery area has enough space
-	*/
-	if (tdb_recovery_allocate(tdb, &recovery_size,
-				  &recovery_offset, &recovery_max_size) == -1) {
-		return -1;
-	}
-
-	data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
-	if (data == NULL) {
-		tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-			   "transaction_setup_recovery: cannot allocate");
-		return -1;
-	}
-
-	rec = (struct tdb_recovery_record *)data;
-	set_recovery_header(rec, TDB_RECOVERY_INVALID_MAGIC,
-			    recovery_size, recovery_max_size, old_map_size);
-	tdb_convert(tdb, rec, sizeof(*rec));
-
-	/* 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 (i=0;i<tdb->transaction->num_blocks;i++) {
-		tdb_off_t offset;
-		tdb_len_t length;
-
-		if (tdb->transaction->blocks[i] == NULL) {
-			continue;
-		}
+	tdb_len_t recovery_size = 0;
+	tdb_off_t recovery_off = 0;
+	tdb_off_t old_map_size = tdb->tdb2.transaction->old_map_size;
+	struct tdb_recovery_record *recovery;
+	const struct tdb_methods *methods = tdb->tdb2.transaction->io_methods;
+	uint64_t magic;
+	enum TDB_ERROR ecode;
+
+	recovery = alloc_recovery(tdb, &recovery_size);
+	if (TDB_PTR_IS_ERR(recovery))
+		return TDB_PTR_ERR(recovery);
+
+	ecode = tdb_recovery_area(tdb, methods, &recovery_off, recovery);
+	if (ecode) {
+		free(recovery);
+		return ecode;
+	}
+
+	if (recovery->max_len < recovery_size) {
+		/* Not large enough. Free up old recovery area. */
+		if (recovery_off) {
+			tdb->stats.frees++;
+			ecode = add_free_record(tdb, recovery_off,
+						sizeof(*recovery)
+						+ recovery->max_len,
+						TDB_LOCK_WAIT, true);
+			free(recovery);
+			if (ecode != TDB_SUCCESS) {
+				return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+						  "tdb_recovery_allocate:"
+						  " failed to free previous"
+						  " recovery area");
+			}
 
-		offset = i * getpagesize();
-		length = getpagesize();
-		if (i == tdb->transaction->num_blocks-1) {
-			length = tdb->transaction->last_block_size;
+			/* Refresh recovery after add_free_record above. */
+			recovery = alloc_recovery(tdb, &recovery_size);
+			if (TDB_PTR_IS_ERR(recovery))
+				return TDB_PTR_ERR(recovery);
 		}
 
-		if (offset >= old_map_size) {
-			continue;
-		}
-		if (offset + length > tdb->map_size) {
-			tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL,
-				   "tdb_transaction_setup_recovery:"
-				   " transaction data over new region boundary");
-			free(data);
-			return -1;
-		}
-		memcpy(p, &offset, sizeof(offset));
-		memcpy(p + sizeof(offset), &length, sizeof(length));
-		tdb_convert(tdb, p, sizeof(offset) + sizeof(length));
-
-		/* 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->tread(tdb, offset,
-				   p + sizeof(offset) + sizeof(length),
-				   length) != 0) {
-			free(data);
-			return -1;
+		recovery_off = create_recovery_area(tdb, recovery_size,
+						    recovery);
+		if (TDB_OFF_IS_ERR(recovery_off)) {
+			free(recovery);
+			return recovery_off;
 		}
-		p += sizeof(offset) + sizeof(length) + length;
 	}
 
-	/* and the tailer */
-	tailer = sizeof(*rec) + recovery_max_size;
-	memcpy(p, &tailer, sizeof(tailer));
-	tdb_convert(tdb, p, sizeof(tailer));
+	/* Now we know size, convert rec header. */
+	recovery->magic = TDB_RECOVERY_INVALID_MAGIC;
+	recovery->len = recovery_size;
+	recovery->eof = old_map_size;
+	tdb_convert(tdb, recovery, sizeof(*recovery));
 
 	/* write the recovery data to the recovery area */
-	if (methods->twrite(tdb, recovery_offset, data,
-			    sizeof(*rec) + recovery_size) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_transaction_setup_recovery:"
-			 " failed to write recovery data");
-		free(data);
-		return -1;
+	ecode = methods->twrite(tdb, recovery_off, recovery, recovery_size);
+	if (ecode != TDB_SUCCESS) {
+		free(recovery);
+		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);
+	transaction_write_existing(tdb, recovery_off, recovery, recovery_size);
+
+	free(recovery);
 
 	/* 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) {
-		free(data);
-		return -1;
-	}
-
-	free(data);
+	ecode = transaction_sync(tdb, recovery_off, recovery_size);
+	if (ecode != TDB_SUCCESS)
+		return ecode;
 
 	magic = TDB_RECOVERY_MAGIC;
 	tdb_convert(tdb, &magic, sizeof(magic));
 
-	*magic_offset = recovery_offset + offsetof(struct tdb_recovery_record,
-						   magic);
+	tdb->tdb2.transaction->magic_offset
+		= recovery_off + offsetof(struct tdb_recovery_record, magic);
 
-	if (methods->twrite(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_transaction_setup_recovery:"
-			 " failed to write recovery magic");
-		return -1;
+	ecode = methods->twrite(tdb, tdb->tdb2.transaction->magic_offset,
+				&magic, sizeof(magic));
+	if (ecode != TDB_SUCCESS) {
+		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));
+	transaction_write_existing(tdb, tdb->tdb2.transaction->magic_offset,
+				   &magic, sizeof(magic));
 
 	/* ensure the recovery magic marker is on disk */
-	if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
-		return -1;
-	}
-
-	return 0;
+	return transaction_sync(tdb, tdb->tdb2.transaction->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_DEBUG_ERROR,
-			   "tdb_transaction_prepare_commit: no transaction");
-		return -1;
+	if (tdb->tdb2.transaction == NULL) {
+		return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+				  "tdb_transaction_prepare_commit:"
+				  " no transaction");
 	}
 
-	if (tdb->transaction->prepared) {
+	if (tdb->tdb2.transaction->prepared) {
 		_tdb_transaction_cancel(tdb);
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_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) {
+	if (tdb->tdb2.transaction->transaction_error) {
 		_tdb_transaction_cancel(tdb);
-		tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_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;
+	if (tdb->tdb2.transaction->nesting != 0) {
+		return TDB_SUCCESS;
 	}
 
 	/* check for a null transaction */
-	if (tdb->transaction->blocks == NULL) {
-		return 0;
+	if (tdb->tdb2.transaction->blocks == NULL) {
+		return TDB_SUCCESS;
 	}
 
-	methods = tdb->transaction->io_methods;
+	methods = tdb->tdb2.transaction->io_methods;
 
 	/* upgrade the main transaction lock region to a write lock */
-	if (tdb_allrecord_upgrade(tdb) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_ERROR,
-			 "tdb_transaction_prepare_commit:"
-			 " failed to upgrade hash locks");
-		_tdb_transaction_cancel(tdb);
-		return -1;
+	ecode = tdb_allrecord_upgrade(tdb, TDB_HASH_LOCK_START);
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
 	}
 
 	/* 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_logerr(tdb, tdb->ecode, TDB_DEBUG_ERROR,
-			 "tdb_transaction_prepare_commit:"
-			 " failed to get open lock");
-		_tdb_transaction_cancel(tdb);
-		return -1;
+	ecode = tdb_lock_open(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
+	if (ecode != TDB_SUCCESS) {
+		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_DEBUG_FATAL,
-				 "tdb_transaction_prepare_commit:"
-				 " failed to setup recovery data");
-			_tdb_transaction_cancel(tdb);
-			return -1;
+		/* Sets up tdb->tdb2.transaction->recovery and
+		 * tdb->tdb2.transaction->magic_offset. */
+		ecode = transaction_setup_recovery(tdb);
+		if (ecode != TDB_SUCCESS) {
+			return ecode;
 		}
 	}
 
-	tdb->transaction->prepared = true;
+	tdb->tdb2.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->tdb2.transaction->old_map_size) {
+		tdb_len_t add;
+
+		add = tdb->file->map_size - tdb->tdb2.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_logerr(tdb, tdb->ecode, TDB_DEBUG_ERROR,
-				 "tdb_transaction_prepare_commit:"
-				 " expansion failed");
-			_tdb_transaction_cancel(tdb);
-			return -1;
+		tdb->file->map_size = tdb->tdb2.transaction->old_map_size;
+		ecode = methods->expand_file(tdb, add);
+		if (ecode != TDB_SUCCESS) {
+			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);
+	if (tdb->flags & TDB_VERSION1) {
+		if (tdb1_transaction_prepare_commit(tdb) == -1)
+			return tdb->last_error;
+		return TDB_SUCCESS;
+	}
+	return tdb->last_error = _tdb_transaction_prepare_commit(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_DEBUG_ERROR,
-			 "tdb_transaction_commit: no transaction");
-		return -1;
+	if (tdb->flags & TDB_VERSION1) {
+		if (tdb1_transaction_commit(tdb) == -1)
+			return tdb->last_error;
+		return TDB_SUCCESS;
 	}
 
-	tdb_trace(tdb, "tdb_transaction_commit");
-
-	if (tdb->transaction->transaction_error) {
-		tdb_transaction_cancel(tdb);
-		tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
-			   "tdb_transaction_commit:"
-			   " transaction error pending");
-		return -1;
+	if (tdb->tdb2.transaction == NULL) {
+		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;
+	if (tdb->tdb2.transaction->nesting != 0) {
+		tdb->tdb2.transaction->nesting--;
+		return tdb->last_error = TDB_SUCCESS;
 	}
 
 	/* check for a null transaction */
-	if (tdb->transaction->blocks == NULL) {
+	if (tdb->tdb2.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;
+	if (!tdb->tdb2.transaction->prepared) {
+		ecode = _tdb_transaction_prepare_commit(tdb);
+		if (ecode != TDB_SUCCESS) {
+			_tdb_transaction_cancel(tdb);
+			return tdb->last_error = ecode;
+		}
 	}
 
-	methods = tdb->transaction->io_methods;
+	methods = tdb->tdb2.transaction->io_methods;
 
 	/* perform all the writes */
-	for (i=0;i<tdb->transaction->num_blocks;i++) {
+	for (i=0;i<tdb->tdb2.transaction->num_blocks;i++) {
 		tdb_off_t offset;
 		tdb_len_t length;
 
-		if (tdb->transaction->blocks[i] == NULL) {
+		if (tdb->tdb2.transaction->blocks[i] == NULL) {
 			continue;
 		}
 
-		offset = i * getpagesize();
-		length = getpagesize();
-		if (i == tdb->transaction->num_blocks-1) {
-			length = tdb->transaction->last_block_size;
+		offset = i * PAGESIZE;
+		length = PAGESIZE;
+		if (i == tdb->tdb2.transaction->num_blocks-1) {
+			length = tdb->tdb2.transaction->last_block_size;
 		}
 
-		if (methods->twrite(tdb, offset, tdb->transaction->blocks[i],
-				    length) == -1) {
-			tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-				   "tdb_transaction_commit:"
-				   " write failed during commit");
-
+		ecode = methods->twrite(tdb, offset,
+					tdb->tdb2.transaction->blocks[i], length);
+		if (ecode != TDB_SUCCESS) {
 			/* we've overwritten part of the data and
 			   possibly expanded the file, so we need to
 			   run the crash recovery code */
-			tdb->methods = methods;
+			tdb->tdb2.io = methods;
 			tdb_transaction_recover(tdb);
 
 			_tdb_transaction_cancel(tdb);
 
-			return -1;
+			return tdb->last_error = ecode;
 		}
-		SAFE_FREE(tdb->transaction->blocks[i]);
+		SAFE_FREE(tdb->tdb2.transaction->blocks[i]);
 	}
 
-	SAFE_FREE(tdb->transaction->blocks);
-	tdb->transaction->num_blocks = 0;
+	SAFE_FREE(tdb->tdb2.transaction->blocks);
+	tdb->tdb2.transaction->num_blocks = 0;
 
 	/* ensure the new data is on disk */
-	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
-		return -1;
+	ecode = transaction_sync(tdb, 0, tdb->file->map_size);
+	if (ecode != TDB_SUCCESS) {
+		return tdb->last_error = ecode;
 	}
 
 	/*
@@ -1051,10 +1171,11 @@ 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->tdb2.transaction->old_map_size = tdb->file->map_size;
 	_tdb_transaction_cancel(tdb);
 
-	return 0;
+	return tdb->last_error = TDB_SUCCESS;
 }
 
 
@@ -1063,63 +1184,61 @@ 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;
 	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_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_transaction_recover:"
-			 " failed to read recovery head");
-		return -1;
+	if (TDB_OFF_IS_ERR(recovery_head)) {
+		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 */
-	if (tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec)) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			   "tdb_transaction_recover:"
-			   " failed to read recovery record");
-		return -1;
+	ecode = tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec));
+	if (ecode != TDB_SUCCESS) {
+		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_DEBUG_FATAL,
-			   "tdb_transaction_recover:"
-			   " attempt to recover read only database");
-		return -1;
+	if (tdb->flags & TDB_RDONLY) {
+		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_DEBUG_FATAL,
-			   "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 */
-	if (tdb->methods->tread(tdb, recovery_head + sizeof(rec), data,
-				rec.len) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			   "tdb_transaction_recover:"
-			   " failed to read recovery data");
-		return -1;
+	ecode = tdb->tdb2.io->tread(tdb, recovery_head + sizeof(rec), data,
+				    rec.len);
+	if (ecode != TDB_SUCCESS) {
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to read recovery data");
 	}
 
 	/* recover the file data */
@@ -1132,71 +1251,75 @@ int tdb_transaction_recover(struct tdb_context *tdb)
 		memcpy(&len, p + sizeof(ofs), sizeof(len));
 		p += sizeof(ofs) + sizeof(len);
 
-		if (tdb->methods->twrite(tdb, ofs, p, len) == -1) {
+		ecode = tdb->tdb2.io->twrite(tdb, ofs, p, len);
+		if (ecode != TDB_SUCCESS) {
 			free(data);
-			tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-				 "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);
 
-	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			   "tdb_transaction_recover: failed to sync recovery");
-		return -1;
+	ecode = transaction_sync(tdb, 0, tdb->file->map_size);
+	if (ecode != TDB_SUCCESS) {
+		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 */
 	if (recovery_eof <= recovery_head) {
-		if (tdb_write_off(tdb, offsetof(struct tdb_header,recovery), 0)
-		    == -1) {
-			tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-				 "tdb_transaction_recover:"
-				 " failed to remove recovery head");
-			return -1;
+		ecode = tdb_write_off(tdb, offsetof(struct tdb_header,
+						    recovery),
+				      0);
+		if (ecode != TDB_SUCCESS) {
+			return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+					  "tdb_transaction_recover:"
+					  " failed to remove recovery head");
 		}
 	}
 
 	/* remove the recovery magic */
-	if (tdb_write_off(tdb,
-			  recovery_head
-			  + offsetof(struct tdb_recovery_record, magic),
-			  TDB_RECOVERY_INVALID_MAGIC) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_transaction_recover:"
-			 " failed to remove recovery magic");
-		return -1;
+	ecode = tdb_write_off(tdb,
+			      recovery_head
+			      + offsetof(struct tdb_recovery_record, magic),
+			      TDB_RECOVERY_INVALID_MAGIC);
+	if (ecode != TDB_SUCCESS) {
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to remove recovery magic");
 	}
 
-	if (transaction_sync(tdb, 0, recovery_eof) == -1) {
-		tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-			 "tdb_transaction_recover: failed to sync2 recovery");
-		return -1;
+	ecode = transaction_sync(tdb, 0, recovery_eof);
+	if (ecode != TDB_SUCCESS) {
+		return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+				  "tdb_transaction_recover:"
+				  " failed to sync2 recovery");
 	}
 
-	tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
+	tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
 		   "tdb_transaction_recover: recovered %zu byte database",
 		   (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;
+	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) {
-		return true;
+	if (TDB_OFF_IS_ERR(recovery_head)) {
+		return recovery_head;
 	}
 
 	if (recovery_head == 0) {
@@ -1205,8 +1328,9 @@ bool tdb_needs_recovery(struct tdb_context *tdb)
 	}
 
 	/* read the recovery record */
-	if (tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec)) == -1) {
-		return true;
+	ecode = tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec));
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
 	}
 
 	return (rec.magic == TDB_RECOVERY_MAGIC);