Sync tdb with the tdb changes in ctdb. Spoke to tridge about

this. Fixes insidious problem with order n^2 freelist merging.
Jeremy.
(This used to be commit c6609c042b128e7d63eb64cfdfb0f6b65cb59d76)

10 files changed, 358 insertions, 218 deletions

diff --git a/source3/lib/tdb/common/freelist.c b/source3/lib/tdb/common/freelist.c
index b109643f23..c086c151fa 100644
--- a/source3/lib/tdb/common/freelist.c
+++ b/source3/lib/tdb/common/freelist.c
@@ -27,6 +27,12 @@
 
 #include "tdb_private.h"
 
+/* 'right' merges can involve O(n^2) cost when combined with a
+   traverse, so they are disabled until we find a way to do them in 
+   O(1) time
+*/
+#define USE_RIGHT_MERGES 0
+
 /* read a freelist record and check for simple errors */
 int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off, struct list_struct *rec)
 {
@@ -56,7 +62,7 @@ int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off, struct list_struct
 }
 
 
-
+#if USE_RIGHT_MERGES
 /* Remove an element from the freelist.  Must have alloc lock. */
 static int remove_from_freelist(struct tdb_context *tdb, tdb_off_t off, tdb_off_t next)
 {
@@ -75,6 +81,7 @@ static int remove_from_freelist(struct tdb_context *tdb, tdb_off_t off, tdb_off_
 	TDB_LOG((tdb, TDB_DEBUG_FATAL,"remove_from_freelist: not on list at off=%d\n", off));
 	return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
 }
+#endif
 
 
 /* update a record tailer (must hold allocation lock) */
@@ -93,8 +100,6 @@ static int update_tailer(struct tdb_context *tdb, tdb_off_t offset,
    neccessary. */
 int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec)
 {
-	tdb_off_t right, left;
-
 	/* Allocation and tailer lock */
 	if (tdb_lock(tdb, -1, F_WRLCK) != 0)
 		return -1;
@@ -105,9 +110,10 @@ int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec)
 		goto fail;
 	}
 
+#if USE_RIGHT_MERGES
 	/* Look right first (I'm an Australian, dammit) */
-	right = offset + sizeof(*rec) + rec->rec_len;
-	if (right + sizeof(*rec) <= tdb->map_size) {
+	if (offset + sizeof(*rec) + rec->rec_len + sizeof(*rec) <= tdb->map_size) {
+		tdb_off_t right = offset + sizeof(*rec) + rec->rec_len;
 		struct list_struct r;
 
 		if (tdb->methods->tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
@@ -122,13 +128,18 @@ int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct list_struct *rec)
 				goto left;
 			}
 			rec->rec_len += sizeof(r) + r.rec_len;
+			if (update_tailer(tdb, offset, rec) == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed at %u\n", offset));
+				goto fail;
+			}
 		}
 	}
-
 left:
+#endif
+
 	/* Look left */
-	left = offset - sizeof(tdb_off_t);
-	if (left > TDB_DATA_START(tdb->header.hash_size)) {
+	if (offset - sizeof(tdb_off_t) > TDB_DATA_START(tdb->header.hash_size)) {
+		tdb_off_t left = offset - sizeof(tdb_off_t);
 		struct list_struct l;
 		tdb_off_t leftsize;
 		
@@ -145,7 +156,12 @@ left:
 
 		left = offset - leftsize;
 
-		/* Now read in record */
+		if (leftsize > offset ||
+		    left < TDB_DATA_START(tdb->header.hash_size)) {
+			goto update;
+		}
+
+		/* Now read in the left record */
 		if (tdb->methods->tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
 			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
 			goto update;
@@ -153,21 +169,24 @@ left:
 
 		/* If it's free, expand to include it. */
 		if (l.magic == TDB_FREE_MAGIC) {
-			if (remove_from_freelist(tdb, left, l.next) == -1) {
-				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: left free failed at %u\n", left));
-				goto update;
-			} else {
-				offset = left;
-				rec->rec_len += leftsize;
+			/* we now merge the new record into the left record, rather than the other 
+			   way around. This makes the operation O(1) instead of O(n). This change
+			   prevents traverse from being O(n^2) after a lot of deletes */
+			l.rec_len += sizeof(*rec) + rec->rec_len;
+			if (tdb_rec_write(tdb, left, &l) == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_left failed at %u\n", left));
+				goto fail;
 			}
+			if (update_tailer(tdb, left, &l) == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed at %u\n", offset));
+				goto fail;
+			}
+			tdb_unlock(tdb, -1, F_WRLCK);
+			return 0;
 		}
 	}
 
 update:
-	if (update_tailer(tdb, offset, rec) == -1) {
-		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed at %u\n", offset));
-		goto fail;
-	}
 
 	/* Now, prepend to free list */
 	rec->magic = TDB_FREE_MAGIC;
@@ -261,6 +280,7 @@ tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct list_st
 		tdb_off_t rec_ptr, last_ptr;
 		tdb_len_t rec_len;
 	} bestfit;
+	float multiplier = 1.0;
 
 	if (tdb_lock(tdb, -1, F_WRLCK) == -1)
 		return 0;
@@ -295,18 +315,27 @@ tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct list_st
 				bestfit.rec_len = rec->rec_len;
 				bestfit.rec_ptr = rec_ptr;
 				bestfit.last_ptr = last_ptr;
-				/* consider a fit to be good enough if
-				   we aren't wasting more than half
-				   the space */
-				if (bestfit.rec_len < 2*length) {
-					break;
-				}
 			}
 		}
 
 		/* move to the next record */
 		last_ptr = rec_ptr;
 		rec_ptr = rec->next;
+
+		/* if we've found a record that is big enough, then
+		   stop searching if its also not too big. The
+		   definition of 'too big' changes as we scan
+		   through */
+		if (bestfit.rec_len > 0 &&
+		    bestfit.rec_len < length * multiplier) {
+			break;
+		}
+		
+		/* this multiplier means we only extremely rarely
+		   search more than 50 or so records. At 50 records we
+		   accept records up to 11 times larger than what we
+		   want */
+		multiplier *= 1.05;
 	}
 
 	if (bestfit.rec_ptr != 0) {
@@ -328,3 +357,25 @@ tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct list_st
 	return 0;
 }
 
+
+
+/* 
+   return the size of the freelist - used to decide if we should repack 
+*/
+int tdb_freelist_size(struct tdb_context *tdb)
+{
+	tdb_off_t ptr;
+	int count=0;
+
+	if (tdb_lock(tdb, -1, F_RDLCK) == -1) {
+		return -1;
+	}
+
+	ptr = FREELIST_TOP;
+	while (tdb_ofs_read(tdb, ptr, &ptr) == 0 && ptr != 0) {
+		count++;
+	}
+
+	tdb_unlock(tdb, -1, F_RDLCK);
+	return count;
+}
diff --git a/source3/lib/tdb/common/io.c b/source3/lib/tdb/common/io.c
index 8ab0768883..172ab69d8c 100644
--- a/source3/lib/tdb/common/io.c
+++ b/source3/lib/tdb/common/io.c
@@ -101,8 +101,8 @@ static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
 					 off+written);
 		}
 		if (written == -1) {
-			/* Ensure ecode is set for log fn. */
-			tdb->ecode = TDB_ERR_IO;
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_IO;
 			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_write failed at %d "
 				 "len=%d (%s)\n", off, len, strerror(errno)));
 			return TDB_ERRCODE(TDB_ERR_IO, -1);
@@ -111,8 +111,8 @@ static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
 				 "write %d bytes at %d in two attempts\n",
 				 len, off));
 			errno = ENOSPC;
-			return TDB_ERRCODE(TDB_ERR_IO, -1);
-		}
+		return TDB_ERRCODE(TDB_ERR_IO, -1);
+	}
 	}
 	return 0;
 }
@@ -230,7 +230,7 @@ void tdb_mmap(struct tdb_context *tdb)
   says to use for mmap expansion */
 static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t addition)
 {
-	char buf[1024];
+	char buf[8192];
 
 	if (tdb->read_only || tdb->traverse_read) {
 		tdb->ecode = TDB_ERR_RDONLY;
@@ -294,7 +294,7 @@ static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t ad
 int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
 {
 	struct list_struct rec;
-	tdb_off_t offset;
+	tdb_off_t offset, new_size;	
 
 	if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
@@ -304,9 +304,11 @@ int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
 	/* must know about any previous expansions by another process */
 	tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
 
-	/* always make room for at least 10 more records, and round
-           the database up to a multiple of the page size */
-	size = TDB_ALIGN(tdb->map_size + size*10, tdb->page_size) - tdb->map_size;
+	/* always make room for at least 100 more records, and at
+           least 25% more space. Round the database up to a multiple
+           of the page size */
+	new_size = MAX(tdb->map_size + size*100, tdb->map_size * 1.25);
+	size = TDB_ALIGN(new_size, tdb->page_size) - tdb->map_size;
 
 	if (!(tdb->flags & TDB_INTERNAL))
 		tdb_munmap(tdb);
diff --git a/source3/lib/tdb/common/lock.c b/source3/lib/tdb/common/lock.c
index e3fe888c46..f156c0fa7b 100644
--- a/source3/lib/tdb/common/lock.c
+++ b/source3/lib/tdb/common/lock.c
@@ -505,6 +505,9 @@ int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
 /* record lock stops delete underneath */
 int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
 {
+	if (tdb->global_lock.count) {
+		return 0;
+	}
 	return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0, 1) : 0;
 }
 
@@ -537,6 +540,10 @@ int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
 	struct tdb_traverse_lock *i;
 	uint32_t count = 0;
 
+	if (tdb->global_lock.count) {
+		return 0;
+	}
+
 	if (off == 0)
 		return 0;
 	for (i = &tdb->travlocks; i; i = i->next)
diff --git a/source3/lib/tdb/common/open.c b/source3/lib/tdb/common/open.c
index 6efa482ac2..6bd8fda2bf 100644
--- a/source3/lib/tdb/common/open.c
+++ b/source3/lib/tdb/common/open.c
@@ -35,7 +35,7 @@ static struct tdb_context *tdbs = NULL;
 static unsigned int default_tdb_hash(TDB_DATA *key)
 {
 	uint32_t value;	/* Used to compute the hash value.  */
-	uint32_t i;	/* Used to cycle through random values. */
+	uint32_t   i;	/* Used to cycle through random values. */
 
 	/* Set the initial value from the key size. */
 	for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
@@ -90,7 +90,7 @@ static int tdb_new_database(struct tdb_context *tdb, int hash_size)
 		size -= written;
 		written = write(tdb->fd, newdb+written, size);
 		if (written == size) {
-			ret = 0;
+		ret = 0;
 		} else if (written >= 0) {
 			/* a second incomplete write - we give up.
 			 * guessing the errno... */
@@ -152,6 +152,7 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 	int rev = 0, locked = 0;
 	unsigned char *vp;
 	uint32_t vertest;
+	unsigned v;
 
 	if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
 		/* Can't log this */
@@ -178,7 +179,9 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 		tdb->page_size = 0x2000;
 	}
 
-	tdb->max_dead_records = (open_flags & TDB_VOLATILE) ? 5 : 0;
+	if (open_flags & TDB_VOLATILE) {
+		tdb->max_dead_records = 5;
+	}
 
 	if ((open_flags & O_ACCMODE) == O_WRONLY) {
 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
@@ -213,6 +216,10 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 		goto fail;	/* errno set by open(2) */
 	}
 
+	/* on exec, don't inherit the fd */
+	v = fcntl(tdb->fd, F_GETFD, 0);
+        fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
+
 	/* ensure there is only one process initialising at once */
 	if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
@@ -240,7 +247,7 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 		/* its not a valid database - possibly initialise it */
 		if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
 			if (errno == 0) {
-				errno = EIO; /* ie bad format or something */
+			errno = EIO; /* ie bad format or something */
 			}
 			goto fail;
 		}
@@ -281,6 +288,7 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 	tdb->map_size = st.st_size;
 	tdb->device = st.st_dev;
 	tdb->inode = st.st_ino;
+	tdb->max_dead_records = 0;
 	tdb_mmap(tdb);
 	if (locked) {
 		if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
diff --git a/source3/lib/tdb/common/tdb.c b/source3/lib/tdb/common/tdb.c
index 0e9d1dbd74..bf3abb71ac 100644
--- a/source3/lib/tdb/common/tdb.c
+++ b/source3/lib/tdb/common/tdb.c
@@ -102,8 +102,7 @@ static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, uint32_t hash,
 }
 
 /* As tdb_find, but if you succeed, keep the lock */
-tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, 
-							 uint32_t hash, int locktype,
+tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
 			   struct list_struct *rec)
 {
 	uint32_t rec_ptr;
@@ -237,14 +236,15 @@ int tdb_exists(struct tdb_context *tdb, TDB_DATA key)
 }
 
 /* actually delete an entry in the database given the offset */
-int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct list_struct*rec)
+int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct list_struct *rec)
 {
 	tdb_off_t last_ptr, i;
 	struct list_struct lastrec;
 
 	if (tdb->read_only || tdb->traverse_read) return -1;
 
-	if (tdb_write_lock_record(tdb, rec_ptr) == -1) {
+	if (tdb->traverse_write != 0 || 
+	    tdb_write_lock_record(tdb, rec_ptr) == -1) {
 		/* Someone traversing here: mark it as dead */
 		rec->magic = TDB_DEAD_MAGIC;
 		return tdb_rec_write(tdb, rec_ptr, rec);
@@ -666,6 +666,16 @@ int tdb_get_flags(struct tdb_context *tdb)
 	return tdb->flags;
 }
 
+void tdb_add_flags(struct tdb_context *tdb, unsigned flags)
+{
+	tdb->flags |= flags;
+}
+
+void tdb_remove_flags(struct tdb_context *tdb, unsigned flags)
+{
+	tdb->flags &= ~flags;
+}
+
 
 /*
   enable sequence number handling on an open tdb
@@ -674,3 +684,57 @@ void tdb_enable_seqnum(struct tdb_context *tdb)
 {
 	tdb->flags |= TDB_SEQNUM;
 }
+
+
+/*
+  wipe the entire database, deleting all records. This can be done
+  very fast by using a global lock. The entire data portion of the
+  file becomes a single entry in the freelist.
+ */
+int tdb_wipe_all(struct tdb_context *tdb)
+{
+	int i;
+	tdb_off_t offset = 0;
+	ssize_t data_len;
+
+	if (tdb_lockall(tdb) != 0) {
+		return -1;
+	}
+
+	/* wipe the hashes */
+	for (i=0;i<tdb->header.hash_size;i++) {
+		if (tdb_ofs_write(tdb, TDB_HASH_TOP(i), &offset) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write hash %d\n", i));
+			goto failed;
+		}
+	}
+
+	/* wipe the freelist */
+	if (tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write freelist\n"));
+		goto failed;
+	}
+
+	/* add all the rest of the file to the freelist */
+	data_len = (tdb->map_size - TDB_DATA_START(tdb->header.hash_size)) - sizeof(struct list_struct);
+	if (data_len > 0) {
+		struct list_struct rec;
+		memset(&rec,'\0',sizeof(rec));
+		rec.rec_len = data_len;
+		if (tdb_free(tdb, TDB_DATA_START(tdb->header.hash_size), &rec) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to add free record\n"));
+			goto failed;
+		}
+	}
+
+	if (tdb_unlockall(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to unlock\n"));
+		goto failed;
+	}
+
+	return 0;
+
+failed:
+	tdb_unlockall(tdb);
+	return -1;
+}
diff --git a/source3/lib/tdb/common/tdb_private.h b/source3/lib/tdb/common/tdb_private.h
index 58c30c1706..63a6d04e72 100644
--- a/source3/lib/tdb/common/tdb_private.h
+++ b/source3/lib/tdb/common/tdb_private.h
@@ -38,6 +38,10 @@
 typedef uint32_t tdb_len_t;
 typedef uint32_t tdb_off_t;
 
+#ifndef offsetof
+#define offsetof(t,f) ((unsigned int)&((t *)0)->f)
+#endif
+
 #define TDB_MAGIC_FOOD "TDB file\n"
 #define TDB_VERSION (0x26011967 + 6)
 #define TDB_MAGIC (0x26011999U)
@@ -54,7 +58,7 @@ typedef uint32_t tdb_off_t;
 #define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
 #define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off_t))
 #define TDB_HASHTABLE_SIZE(tdb) ((tdb->header.hash_size+1)*sizeof(tdb_off_t))
-#define TDB_DATA_START(hash_size) TDB_HASH_TOP(hash_size-1)
+#define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + sizeof(tdb_off_t))
 #define TDB_RECOVERY_HEAD offsetof(struct tdb_header, recovery_start)
 #define TDB_SEQNUM_OFS    offsetof(struct tdb_header, sequence_number)
 #define TDB_PAD_BYTE 0x42
@@ -144,6 +148,7 @@ struct tdb_context {
 	tdb_len_t map_size; /* how much space has been mapped */
 	int read_only; /* opened read-only */
 	int traverse_read; /* read-only traversal */
+	int traverse_write; /* read-write traversal */
 	struct tdb_lock_type global_lock;
 	int num_lockrecs;
 	struct tdb_lock_type *lockrecs; /* only real locks, all with count>0 */
@@ -173,7 +178,6 @@ struct tdb_context {
 int tdb_munmap(struct tdb_context *tdb);
 void tdb_mmap(struct tdb_context *tdb);
 int tdb_lock(struct tdb_context *tdb, int list, int ltype);
-int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
 int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
 int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset, int rw_type, int lck_type, int probe, size_t len);
 int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
diff --git a/source3/lib/tdb/common/transaction.c b/source3/lib/tdb/common/transaction.c
index 7eaacf7a16..2da09face0 100644
--- a/source3/lib/tdb/common/transaction.c
+++ b/source3/lib/tdb/common/transaction.c
@@ -87,13 +87,6 @@
 
 */
 
-struct tdb_transaction_el {
-	struct tdb_transaction_el *next, *prev;
-	tdb_off_t offset;
-	tdb_len_t length;
-	unsigned char *data;
-};
-
 /*
   hold the context of any current transaction
 */
@@ -105,12 +98,12 @@ struct tdb_transaction {
 	/* the original io methods - used to do IOs to the real db */
 	const struct tdb_methods *io_methods;
 
-	/* the list of transaction elements. We use a doubly linked
-	   list with a last pointer to allow us to keep the list
-	   ordered, with first element at the front of the list. It
-	   needs to be doubly linked as the read/write traversals need
-	   to be backwards, while the commit needs to be forwards */
-	struct tdb_transaction_el *elements, *elements_last;
+	/* the list of transaction blocks. When a block is first
+	   written to, it gets created in this list */
+	uint8_t **blocks;
+	uint32_t num_blocks;
+	uint32_t block_size;      /* bytes in each block */
+	uint32_t last_block_size; /* number of valid bytes in the last block */
 
 	/* non-zero when an internal transaction error has
 	   occurred. All write operations will then fail until the
@@ -134,52 +127,48 @@ struct tdb_transaction {
 static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf, 
 			    tdb_len_t len, int cv)
 {
-	struct tdb_transaction_el *el;
-
-	/* we need to walk the list backwards to get the most recent data */
-	for (el=tdb->transaction->elements_last;el;el=el->prev) {
-		tdb_len_t partial;
+	uint32_t blk;
 
-		if (off+len <= el->offset) {
-			continue;
-		}
-		if (off >= el->offset + el->length) {
-			continue;
+	/* break it down into block sized ops */
+	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+		if (transaction_read(tdb, off, buf, len2, cv) != 0) {
+			return -1;
 		}
+		len -= len2;
+		off += len2;
+		buf = (void *)(len2 + (char *)buf);
+	}
 
-		/* an overlapping read - needs to be split into up to
-		   2 reads and a memcpy */
-		if (off < el->offset) {
-			partial = el->offset - off;
-			if (transaction_read(tdb, off, buf, partial, cv) != 0) {
-				goto fail;
-			}
-			len -= partial;
-			off += partial;
-			buf = (void *)(partial + (char *)buf);
-		}
-		if (off + len <= el->offset + el->length) {
-			partial = len;
-		} else {
-			partial = el->offset + el->length - off;
-		}
-		memcpy(buf, el->data + (off - el->offset), partial);
-		if (cv) {
-			tdb_convert(buf, len);
-		}
-		len -= partial;
-		off += partial;
-		buf = (void *)(partial + (char *)buf);
-		
-		if (len != 0 && transaction_read(tdb, off, buf, len, cv) != 0) {
+	if (len == 0) {
+		return 0;
+	}
+
+	blk = off / tdb->transaction->block_size;
+
+	/* see if we have it in the block list */
+	if (tdb->transaction->num_blocks <= blk ||
+	    tdb->transaction->blocks[blk] == NULL) {
+		/* nope, do a real read */
+		if (tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv) != 0) {
 			goto fail;
 		}
-
 		return 0;
 	}
 
-	/* its not in the transaction elements - do a real read */
-	return tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv);
+	/* it is in the block list. Now check for the last block */
+	if (blk == tdb->transaction->num_blocks-1) {
+		if (len > tdb->transaction->last_block_size) {
+			goto fail;
+		}
+	}
+	
+	/* now copy it out of this block */
+	memcpy(buf, tdb->transaction->blocks[blk] + (off % tdb->transaction->block_size), len);
+	if (cv) {
+		tdb_convert(buf, len);
+	}
+	return 0;
 
 fail:
 	TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_read: failed at off=%d len=%d\n", off, len));
@@ -195,12 +184,8 @@ fail:
 static int transaction_write(struct tdb_context *tdb, tdb_off_t off, 
 			     const void *buf, tdb_len_t len)
 {
-	struct tdb_transaction_el *el, *best_el=NULL;
+	uint32_t blk;
 
-	if (len == 0) {
-		return 0;
-	}
-	
 	/* if the write is to a hash head, then update the transaction
 	   hash heads */
 	if (len == sizeof(tdb_off_t) && off >= FREELIST_TOP &&
@@ -209,106 +194,88 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
 		memcpy(&tdb->transaction->hash_heads[chain], buf, len);
 	}
 
-	/* first see if we can replace an existing entry */
-	for (el=tdb->transaction->elements_last;el;el=el->prev) {
-		tdb_len_t partial;
-
-		if (best_el == NULL && off == el->offset+el->length) {
-			best_el = el;
+	/* break it up into block sized chunks */
+	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+		if (transaction_write(tdb, off, buf, len2) != 0) {
+			return -1;
 		}
-
-		if (off+len <= el->offset) {
-			continue;
-		}
-		if (off >= el->offset + el->length) {
-			continue;
+		len -= len2;
+		off += len2;
+		if (buf != NULL) {
+			buf = (const void *)(len2 + (const char *)buf);
 		}
+	}
 
-		/* an overlapping write - needs to be split into up to
-		   2 writes and a memcpy */
-		if (off < el->offset) {
-			partial = el->offset - off;
-			if (transaction_write(tdb, off, buf, partial) != 0) {
-				goto fail;
-			}
-			len -= partial;
-			off += partial;
-			buf = (const void *)(partial + (const char *)buf);
-		}
-		if (off + len <= el->offset + el->length) {
-			partial = len;
+	if (len == 0) {
+		return 0;
+	}
+
+	blk = off / tdb->transaction->block_size;
+	off = off % tdb->transaction->block_size;
+
+	if (tdb->transaction->num_blocks <= blk) {
+		uint8_t **new_blocks;
+		/* expand the blocks array */
+		if (tdb->transaction->blocks == NULL) {
+			new_blocks = malloc((blk+1)*sizeof(uint8_t *));
 		} else {
-			partial = el->offset + el->length - off;
+			new_blocks = realloc(tdb->transaction->blocks, (blk+1)*sizeof(uint8_t *));
 		}
-		memcpy(el->data + (off - el->offset), buf, partial);
-		len -= partial;
-		off += partial;
-		buf = (const void *)(partial + (const char *)buf);
-		
-		if (len != 0 && transaction_write(tdb, off, buf, len) != 0) {
+		if (new_blocks == NULL) {
+			tdb->ecode = TDB_ERR_OOM;
 			goto fail;
 		}
-
-		return 0;
+		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;
 	}
 
-	/* see if we can append the new entry to an existing entry */
-	if (best_el && best_el->offset + best_el->length == off && 
-	    (off+len < tdb->transaction->old_map_size ||
-	     off > tdb->transaction->old_map_size)) {
-		unsigned char *data = best_el->data;
-		el = best_el;
-		el->data = (unsigned char *)realloc(el->data,
-						    el->length + len);
-		if (el->data == NULL) {
+	/* allocate and fill a block? */
+	if (tdb->transaction->blocks[blk] == NULL) {
+		tdb->transaction->blocks[blk] = (uint8_t *)calloc(tdb->transaction->block_size, 1);
+		if (tdb->transaction->blocks[blk] == NULL) {
 			tdb->ecode = TDB_ERR_OOM;
 			tdb->transaction->transaction_error = 1;
-			el->data = data;
-			return -1;
+			return -1;			
 		}
-		if (buf) {
-			memcpy(el->data + el->length, buf, len);
-		} else {
-			memset(el->data + el->length, TDB_PAD_BYTE, len);
+		if (tdb->transaction->old_map_size > blk * tdb->transaction->block_size) {
+			tdb_len_t len2 = tdb->transaction->block_size;
+			if (len2 + (blk * tdb->transaction->block_size) > tdb->transaction->old_map_size) {
+				len2 = tdb->transaction->old_map_size - (blk * tdb->transaction->block_size);
+			}
+			if (tdb->transaction->io_methods->tdb_read(tdb, blk * tdb->transaction->block_size, 
+								   tdb->transaction->blocks[blk], 
+								   len2, 0) != 0) {
+				SAFE_FREE(tdb->transaction->blocks[blk]);				
+				tdb->ecode = TDB_ERR_IO;
+				goto fail;
+			}
+			if (blk == tdb->transaction->num_blocks-1) {
+				tdb->transaction->last_block_size = len2;
+			}			
 		}
-		el->length += len;
-		return 0;
 	}
-
-	/* add a new entry at the end of the list */
-	el = (struct tdb_transaction_el *)malloc(sizeof(*el));
-	if (el == NULL) {
-		tdb->ecode = TDB_ERR_OOM;
-		tdb->transaction->transaction_error = 1;		
-		return -1;
-	}
-	el->next = NULL;
-	el->prev = tdb->transaction->elements_last;
-	el->offset = off;
-	el->length = len;
-	el->data = (unsigned char *)malloc(len);
-	if (el->data == NULL) {
-		free(el);
-		tdb->ecode = TDB_ERR_OOM;
-		tdb->transaction->transaction_error = 1;		
-		return -1;
-	}
-	if (buf) {
-		memcpy(el->data, buf, len);
+	
+	/* overwrite part of an existing block */
+	if (buf == NULL) {
+		memset(tdb->transaction->blocks[blk] + off, 0, len);
 	} else {
-		memset(el->data, TDB_PAD_BYTE, len);
+		memcpy(tdb->transaction->blocks[blk] + off, buf, len);
 	}
-	if (el->prev) {
-		el->prev->next = el;
-	} else {
-		tdb->transaction->elements = el;
+	if (blk == tdb->transaction->num_blocks-1) {
+		if (len + off > tdb->transaction->last_block_size) {
+			tdb->transaction->last_block_size = len + off;
+		}
 	}
-	tdb->transaction->elements_last = el;
+
 	return 0;
 
 fail:
-	TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%d len=%d\n", off, len));
-	tdb->ecode = TDB_ERR_IO;
+	TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%d len=%d\n", 
+		 (blk*tdb->transaction->block_size) + off, len));
 	tdb->transaction->transaction_error = 1;
 	return -1;
 }
@@ -419,10 +386,14 @@ int tdb_transaction_start(struct tdb_context *tdb)
 		return -1;
 	}
 
+	/* a page at a time seems like a reasonable compromise between compactness and efficiency */
+	tdb->transaction->block_size = tdb->page_size;
+
 	/* get the transaction write lock. This is a blocking lock. As
 	   discussed with Volker, there are a number of ways we could
 	   make this async, which we will probably do in the future */
 	if (tdb_transaction_lock(tdb, F_WRLCK) == -1) {
+		SAFE_FREE(tdb->transaction->blocks);
 		SAFE_FREE(tdb->transaction);
 		return -1;
 	}
@@ -460,21 +431,12 @@ int tdb_transaction_start(struct tdb_context *tdb)
 	tdb->transaction->io_methods = tdb->methods;
 	tdb->methods = &transaction_methods;
 
-	/* by calling this transaction write here, we ensure that we don't grow the
-	   transaction linked list due to hash table updates */
-	if (transaction_write(tdb, FREELIST_TOP, tdb->transaction->hash_heads, 
-			      TDB_HASHTABLE_SIZE(tdb)) != 0) {
-		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_start: failed to prime hash table\n"));
-		tdb->ecode = TDB_ERR_IO;
-		tdb->methods = tdb->transaction->io_methods;
-		goto fail;
-	}
-
 	return 0;
 	
 fail:
 	tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
 	tdb_transaction_unlock(tdb);
+	SAFE_FREE(tdb->transaction->blocks);
 	SAFE_FREE(tdb->transaction->hash_heads);
 	SAFE_FREE(tdb->transaction);
 	return -1;
@@ -486,6 +448,8 @@ fail:
 */
 int tdb_transaction_cancel(struct tdb_context *tdb)
 {	
+	int i;
+
 	if (tdb->transaction == NULL) {
 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_cancel: no transaction\n"));
 		return -1;
@@ -499,13 +463,13 @@ int tdb_transaction_cancel(struct tdb_context *tdb)
 
 	tdb->map_size = tdb->transaction->old_map_size;
 
-	/* free all the transaction elements */
-	while (tdb->transaction->elements) {
-		struct tdb_transaction_el *el = tdb->transaction->elements;
-		tdb->transaction->elements = el->next;
-		free(el->data);
-		free(el);
+	/* free all the transaction blocks */
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		if (tdb->transaction->blocks[i] != NULL) {
+			free(tdb->transaction->blocks[i]);
+		}
 	}
+	SAFE_FREE(tdb->transaction->blocks);
 
 	/* remove any global lock created during the transaction */
 	if (tdb->global_lock.count != 0) {
@@ -515,7 +479,6 @@ int tdb_transaction_cancel(struct tdb_context *tdb)
 
 	/* remove any locks created during the transaction */
 	if (tdb->num_locks != 0) {
-		int i;
 		for (i=0;i<tdb->num_lockrecs;i++) {
 			tdb_brlock(tdb,FREELIST_TOP+4*tdb->lockrecs[i].list,
 				   F_UNLCK,F_SETLKW, 0, 1);
@@ -567,16 +530,24 @@ static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t
 */
 static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
 {
-	struct tdb_transaction_el *el;
 	tdb_len_t recovery_size = 0;
+	int i;
 
 	recovery_size = sizeof(uint32_t);
-	for (el=tdb->transaction->elements;el;el=el->next) {
-		if (el->offset >= tdb->transaction->old_map_size) {
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		if (i * tdb->transaction->block_size >= tdb->transaction->old_map_size) {
+			break;
+		}
+		if (tdb->transaction->blocks[i] == NULL) {
 			continue;
 		}
-		recovery_size += 2*sizeof(tdb_off_t) + el->length;
-	}
+		recovery_size += 2*sizeof(tdb_off_t);
+		if (i == tdb->transaction->num_blocks-1) {
+			recovery_size += tdb->transaction->last_block_size;
+		} else {
+			recovery_size += tdb->transaction->block_size;
+		}
+	}	
 
 	return recovery_size;
 }
@@ -669,7 +640,6 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 static int transaction_setup_recovery(struct tdb_context *tdb, 
 				      tdb_off_t *magic_offset)
 {
-	struct tdb_transaction_el *el;
 	tdb_len_t recovery_size;
 	unsigned char *data, *p;
 	const struct tdb_methods *methods = tdb->transaction->io_methods;
@@ -677,6 +647,7 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	tdb_off_t recovery_offset, recovery_max_size;
 	tdb_off_t old_map_size = tdb->transaction->old_map_size;
 	uint32_t magic, tailer;
+	int i;
 
 	/*
 	  check that the recovery area has enough space
@@ -704,30 +675,43 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
 	/* build the recovery data into a single blob to allow us to do a single
 	   large write, which should be more efficient */
 	p = data + sizeof(*rec);
-	for (el=tdb->transaction->elements;el;el=el->next) {
-		if (el->offset >= old_map_size) {
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		tdb_off_t offset;
+		tdb_len_t length;
+
+		if (tdb->transaction->blocks[i] == NULL) {
 			continue;
 		}
-		if (el->offset + el->length > tdb->transaction->old_map_size) {
+
+		offset = i * tdb->transaction->block_size;
+		length = tdb->transaction->block_size;
+		if (i == tdb->transaction->num_blocks-1) {
+			length = tdb->transaction->last_block_size;
+		}
+		
+		if (offset >= old_map_size) {
+			continue;
+		}
+		if (offset + length > tdb->transaction->old_map_size) {
 			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: transaction data over new region boundary\n"));
 			free(data);
 			tdb->ecode = TDB_ERR_CORRUPT;
 			return -1;
 		}
-		memcpy(p, &el->offset, 4);
-		memcpy(p+4, &el->length, 4);
+		memcpy(p, &offset, 4);
+		memcpy(p+4, &length, 4);
 		if (DOCONV()) {
 			tdb_convert(p, 8);
 		}
 		/* the recovery area contains the old data, not the
 		   new data, so we have to call the original tdb_read
 		   method to get it */
-		if (methods->tdb_read(tdb, el->offset, p + 8, el->length, 0) != 0) {
+		if (methods->tdb_read(tdb, offset, p + 8, length, 0) != 0) {
 			free(data);
 			tdb->ecode = TDB_ERR_IO;
 			return -1;
 		}
-		p += 8 + el->length;
+		p += 8 + length;
 	}
 
 	/* and the tailer */
@@ -780,6 +764,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 	const struct tdb_methods *methods;
 	tdb_off_t magic_offset = 0;
 	uint32_t zero = 0;
+	int i;
 
 	if (tdb->transaction == NULL) {
 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: no transaction\n"));
@@ -799,7 +784,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 	}		
 
 	/* check for a null transaction */
-	if (tdb->transaction->elements == NULL) {
+	if (tdb->transaction->blocks == NULL) {
 		tdb_transaction_cancel(tdb);
 		return 0;
 	}
@@ -858,10 +843,21 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 	}
 
 	/* perform all the writes */
-	while (tdb->transaction->elements) {
-		struct tdb_transaction_el *el = tdb->transaction->elements;
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		tdb_off_t offset;
+		tdb_len_t length;
 
-		if (methods->tdb_write(tdb, el->offset, el->data, el->length) == -1) {
+		if (tdb->transaction->blocks[i] == NULL) {
+			continue;
+		}
+
+		offset = i * tdb->transaction->block_size;
+		length = tdb->transaction->block_size;
+		if (i == tdb->transaction->num_blocks-1) {
+			length = tdb->transaction->last_block_size;
+		}
+
+		if (methods->tdb_write(tdb, offset, tdb->transaction->blocks[i], length) == -1) {
 			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed during commit\n"));
 			
 			/* we've overwritten part of the data and
@@ -876,11 +872,12 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
 			return -1;
 		}
-		tdb->transaction->elements = el->next;
-		free(el->data); 
-		free(el);
+		SAFE_FREE(tdb->transaction->blocks[i]);
 	} 
 
+	SAFE_FREE(tdb->transaction->blocks);
+	tdb->transaction->num_blocks = 0;
+
 	if (!(tdb->flags & TDB_NOSYNC)) {
 		/* ensure the new data is on disk */
 		if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
diff --git a/source3/lib/tdb/common/traverse.c b/source3/lib/tdb/common/traverse.c
index 27b2cfc54a..07b0c23858 100644
--- a/source3/lib/tdb/common/traverse.c
+++ b/source3/lib/tdb/common/traverse.c
@@ -241,7 +241,9 @@ int tdb_traverse(struct tdb_context *tdb,
 		return -1;
 	}
 
+	tdb->traverse_write++;
 	ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
+	tdb->traverse_write--;
 
 	tdb_transaction_unlock(tdb);
 
@@ -333,3 +335,4 @@ TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
 	return key;
 }
+
diff --git a/source3/lib/tdb/configure.ac b/source3/lib/tdb/configure.ac
index de8fc9c2b5..1085055bc9 100644
--- a/source3/lib/tdb/configure.ac
+++ b/source3/lib/tdb/configure.ac
@@ -2,7 +2,7 @@ AC_PREREQ(2.50)
 AC_DEFUN([SMB_MODULE_DEFAULT], [echo -n ""])
 AC_DEFUN([SMB_LIBRARY_ENABLE], [echo -n ""])
 AC_DEFUN([SMB_ENABLE], [echo -n ""])
-AC_INIT(tdb, 1.1.0)
+AC_INIT(tdb, 1.1.1)
 AC_CONFIG_SRCDIR([common/tdb.c])
 AC_CONFIG_HEADER(include/config.h)
 AC_LIBREPLACE_ALL_CHECKS
diff --git a/source3/lib/tdb/include/tdb.h b/source3/lib/tdb/include/tdb.h
index fce6628b84..0008085de5 100644
--- a/source3/lib/tdb/include/tdb.h
+++ b/source3/lib/tdb/include/tdb.h
@@ -135,6 +135,8 @@ int tdb_get_seqnum(struct tdb_context *tdb);
 int tdb_hash_size(struct tdb_context *tdb);
 size_t tdb_map_size(struct tdb_context *tdb);
 int tdb_get_flags(struct tdb_context *tdb);
+void tdb_add_flags(struct tdb_context *tdb, unsigned flag);
+void tdb_remove_flags(struct tdb_context *tdb, unsigned flag);
 void tdb_enable_seqnum(struct tdb_context *tdb);
 void tdb_increment_seqnum_nonblock(struct tdb_context *tdb);
 
@@ -153,6 +155,8 @@ void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *sigptr)
 void tdb_dump_all(struct tdb_context *tdb);
 int tdb_printfreelist(struct tdb_context *tdb);
 int tdb_validate_freelist(struct tdb_context *tdb, int *pnum_entries);
+int tdb_wipe_all(struct tdb_context *tdb);
+int tdb_freelist_size(struct tdb_context *tdb);
 
 extern TDB_DATA tdb_null;