r14799: added a tdb_get_seqnum() call, and the TDB_SEQNUM flag. This allows

for an extremely lightweight test to see if a tdb has possibly
changed.
(This used to be commit f325ba605ccceca63712c0f2c98961e35e437b3d)

3 files changed, 63 insertions, 1 deletions

diff --git a/source4/lib/tdb/common/tdb.c b/source4/lib/tdb/common/tdb.c
index 4b0d4a31c5..b0411601eb 100644
--- a/source4/lib/tdb/common/tdb.c
+++ b/source4/lib/tdb/common/tdb.c
@@ -30,6 +30,33 @@
 
 TDB_DATA tdb_null;
 
+/*
+  increment the tdb sequence number if the tdb has been opened using
+  the TDB_SEQNUM flag
+*/
+static void tdb_increment_seqnum(struct tdb_context *tdb)
+{
+	tdb_off_t seqnum=0;
+	
+	if (!(tdb->flags & TDB_SEQNUM)) {
+		return;
+	}
+
+	if (tdb_brlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, F_SETLKW, 1) != 0) {
+		return;
+	}
+
+	/* we ignore errors from this, as we have no sane way of
+	   dealing with them.
+	*/
+	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+	seqnum++;
+	tdb_ofs_write(tdb, TDB_SEQNUM_OFS, &seqnum);
+
+	tdb_brlock(tdb, TDB_SEQNUM_OFS, F_UNLCK, F_SETLKW, 1);
+}
+
+
 /* Returns 0 on fail.  On success, return offset of record, and fills
    in rec */
 static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, u32 hash,
@@ -203,6 +230,11 @@ static int tdb_delete_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash)
 	if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK, &rec)))
 		return -1;
 	ret = tdb_do_delete(tdb, rec_ptr, &rec);
+
+	if (ret == 0) {
+		tdb_increment_seqnum(tdb);
+	}
+
 	if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
 		TDB_LOG((tdb, 0, "tdb_delete: WARNING tdb_unlock failed!\n"));
 	return ret;
@@ -295,6 +327,9 @@ int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
 		/* Need to tdb_unallocate() here */
 		goto fail;
 	}
+
+	tdb_increment_seqnum(tdb);
+
  out:
 	SAFE_FREE(p); 
 	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
@@ -369,3 +404,22 @@ tdb_log_func tdb_log_fn(struct tdb_context *tdb)
 {
 	return tdb->log_fn;
 }
+
+
+/*
+  get the tdb sequence number. Only makes sense if the writers opened
+  with TDB_SEQNUM set. Note that this sequence number will wrap quite
+  quickly, so it should only be used for a 'has something changed'
+  test, not for code that relies on the count of the number of changes
+  made. If you want a counter then use a tdb record.
+
+  The aim of this sequence number is to allow for a very lightweight
+  test of a possible tdb change.
+*/
+int tdb_get_seqnum(struct tdb_context *tdb)
+{
+	tdb_off_t seqnum=0;
+
+	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+	return seqnum;
+}
diff --git a/source4/lib/tdb/common/tdb_private.h b/source4/lib/tdb/common/tdb_private.h
index d7193b2b67..d01a88695d 100644
--- a/source4/lib/tdb/common/tdb_private.h
+++ b/source4/lib/tdb/common/tdb_private.h
@@ -94,6 +94,7 @@ typedef u32 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_RECOVERY_HEAD offsetof(struct tdb_header, recovery_start)
+#define TDB_SEQNUM_OFS    offsetof(struct tdb_header, sequence_number)
 #define TDB_PAD_BYTE 0x42
 #define TDB_PAD_U32  0x42424242
 
@@ -155,7 +156,8 @@ struct tdb_header {
 	u32 hash_size; /* number of hash entries */
 	tdb_off_t rwlocks; /* obsolete - kept to detect old formats */
 	tdb_off_t recovery_start; /* offset of transaction recovery region */
-	tdb_off_t reserved[30];
+	tdb_off_t sequence_number; /* used when TDB_SEQNUM is set */
+	tdb_off_t reserved[29];
 };
 
 struct tdb_lock_type {
diff --git a/source4/lib/tdb/common/transaction.c b/source4/lib/tdb/common/transaction.c
index ace4aa51a6..6960a02ad8 100644
--- a/source4/lib/tdb/common/transaction.c
+++ b/source4/lib/tdb/common/transaction.c
@@ -892,6 +892,12 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 
 	tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
 
+	/*
+	  TODO: maybe write to some dummy hdr field, or write to magic
+	  offset without mmap, before the last sync, instead of the
+	  utime() call
+	*/
+
 	/* on some systems (like Linux 2.6.x) changes via mmap/msync
 	   don't change the mtime of the file, this means the file may
 	   not be backed up (as tdb rounding to block sizes means that