X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb%2Ftdb.c;h=7317a3aa521554c898d12a5e5b8f57c7dafea36f;hp=c44987a14e01f343c008e403eb789bd3973a33dd;hb=d9cbd7d4454ae35e4e2f6d18a9469bf26948e4b9;hpb=f26355f247709f0150028ec48df7d3f2af463423

diff --git a/ccan/tdb/tdb.c b/ccan/tdb/tdb.c
index c44987a1..7317a3aa 100644
--- a/ccan/tdb/tdb.c
+++ b/ccan/tdb/tdb.c
@@ -59,13 +59,14 @@ static void tdb_increment_seqnum(struct tdb_context *tdb)
 		return;
 	}
 
-	if (tdb_brlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, F_SETLKW, 1, 1) != 0) {
+	if (tdb_nest_lock(tdb, TDB_SEQNUM_OFS, F_WRLCK,
+			  TDB_LOCK_WAIT|TDB_LOCK_PROBE) != 0) {
 		return;
 	}
 
 	tdb_increment_seqnum_nonblock(tdb);
 
-	tdb_brlock(tdb, TDB_SEQNUM_OFS, F_UNLCK, F_SETLKW, 1, 1);
+	tdb_nest_unlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, false);
 }
 
 static int tdb_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
@@ -76,7 +77,7 @@ static int tdb_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
 /* 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, uint32_t hash,
-			struct list_struct *r)
+			struct tdb_record *r)
 {
 	tdb_off_t rec_ptr;
 	
@@ -96,14 +97,21 @@ static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, uint32_t hash,
 				      NULL) == 0) {
 			return rec_ptr;
 		}
+		/* detect tight infinite loop */
+		if (rec_ptr == r->next) {
+			tdb->ecode = TDB_ERR_CORRUPT;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_find: loop detected.\n"));
+			return 0;
+		}
 		rec_ptr = r->next;
 	}
-	return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
+	tdb->ecode = TDB_ERR_NOEXIST;
+	return 0;
 }
 
 /* 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,
-			   struct list_struct *rec)
+			   struct tdb_record *rec)
 {
 	uint32_t rec_ptr;
 
@@ -114,6 +122,7 @@ tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t has
 	return rec_ptr;
 }
 
+static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key);
 
 /* update an entry in place - this only works if the new data size
    is <= the old data size and the key exists.
@@ -121,13 +130,32 @@ tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t has
 */
 static int tdb_update_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, TDB_DATA dbuf)
 {
-	struct list_struct rec;
+	struct tdb_record rec;
 	tdb_off_t rec_ptr;
 
 	/* find entry */
 	if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
 		return -1;
 
+	/* it could be an exact duplicate of what is there - this is
+	 * surprisingly common (eg. with a ldb re-index). */
+	if (rec.key_len == key.dsize && 
+	    rec.data_len == dbuf.dsize &&
+	    rec.full_hash == hash) {
+		TDB_DATA data = _tdb_fetch(tdb, key);
+		if (data.dsize == dbuf.dsize &&
+		    memcmp(data.dptr, dbuf.dptr, data.dsize) == 0) {
+			if (data.dptr) {
+				free(data.dptr);
+			}
+			return 0;
+		}
+		if (data.dptr) {
+			free(data.dptr);
+		}
+	}
+	 
+
 	/* must be long enough key, data and tailer */
 	if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off_t)) {
 		tdb->ecode = TDB_SUCCESS; /* Not really an error */
@@ -153,18 +181,18 @@ static int tdb_update_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash,
  * then the TDB_DATA will have zero length but
  * a non-zero pointer
  */
-static TDB_DATA do_tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
+static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
 {
 	tdb_off_t rec_ptr;
-	struct list_struct rec;
+	struct tdb_record rec;
 	TDB_DATA ret;
 	uint32_t hash;
 
 	/* find which hash bucket it is in */
 	hash = tdb->hash_fn(&key);
-	if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
+	if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
 		return tdb_null;
-	}
+
 	ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
 				  rec.data_len);
 	ret.dsize = rec.data_len;
@@ -174,7 +202,7 @@ static TDB_DATA do_tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
 
 TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
 {
-	TDB_DATA ret = do_tdb_fetch(tdb, key);
+	TDB_DATA ret = _tdb_fetch(tdb, key);
 
 	tdb_trace_1rec_retrec(tdb, "tdb_fetch", key, ret);
 	return ret;
@@ -185,7 +213,7 @@ TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
  * function. The parsing function is executed under the chain read lock, so it
  * should be fast and should not block on other syscalls.
  *
- * DONT CALL OTHER TDB CALLS FROM THE PARSER, THIS MIGHT LEAD TO SEGFAULTS.
+ * DON'T CALL OTHER TDB CALLS FROM THE PARSER, THIS MIGHT LEAD TO SEGFAULTS.
  *
  * For mmapped tdb's that do not have a transaction open it points the parsing
  * function directly at the mmap area, it avoids the malloc/memcpy in this
@@ -202,7 +230,7 @@ int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
 		     void *private_data)
 {
 	tdb_off_t rec_ptr;
-	struct list_struct rec;
+	struct tdb_record rec;
 	int ret;
 	uint32_t hash;
 
@@ -210,9 +238,9 @@ int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
 	hash = tdb->hash_fn(&key);
 
 	if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
-		tdb_trace_1rec_ret(tdb, "tdb_parse_record", key,
-				   -TDB_ERR_NOEXIST);
-		return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
+		tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, -1);
+		tdb->ecode = TDB_ERR_NOEXIST;
+		return 0;
 	}
 	tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, 0);
 
@@ -232,7 +260,7 @@ int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
 */
 static int tdb_exists_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
 {
-	struct list_struct rec;
+	struct tdb_record rec;
 	
 	if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0)
 		return 0;
@@ -251,14 +279,14 @@ 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 tdb_record *rec)
 {
 	tdb_off_t last_ptr, i;
-	struct list_struct lastrec;
+	struct tdb_record lastrec;
 
 	if (tdb->read_only || tdb->traverse_read) return -1;
 
-	if (tdb->traverse_write != 0 || 
+	if (((tdb->traverse_write != 0) && (!TDB_DEAD(rec))) ||
 	    tdb_write_lock_record(tdb, rec_ptr) == -1) {
 		/* Someone traversing here: mark it as dead */
 		rec->magic = TDB_DEAD_MAGIC;
@@ -290,7 +318,7 @@ static int tdb_count_dead(struct tdb_context *tdb, uint32_t hash)
 {
 	int res = 0;
 	tdb_off_t rec_ptr;
-	struct list_struct rec;
+	struct tdb_record rec;
 	
 	/* read in the hash top */
 	if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
@@ -314,7 +342,7 @@ static int tdb_count_dead(struct tdb_context *tdb, uint32_t hash)
 static int tdb_purge_dead(struct tdb_context *tdb, uint32_t hash)
 {
 	int res = -1;
-	struct list_struct rec;
+	struct tdb_record rec;
 	tdb_off_t rec_ptr;
 
 	if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
@@ -350,7 +378,7 @@ static int tdb_purge_dead(struct tdb_context *tdb, uint32_t hash)
 static int tdb_delete_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
 {
 	tdb_off_t rec_ptr;
-	struct list_struct rec;
+	struct tdb_record rec;
 	int ret;
 
 	if (tdb->max_dead_records != 0) {
@@ -413,7 +441,7 @@ int tdb_delete(struct tdb_context *tdb, TDB_DATA key)
  * See if we have a dead record around with enough space
  */
 static tdb_off_t tdb_find_dead(struct tdb_context *tdb, uint32_t hash,
-			       struct list_struct *r, tdb_len_t length)
+			       struct tdb_record *r, tdb_len_t length)
 {
 	tdb_off_t rec_ptr;
 	
@@ -438,10 +466,10 @@ static tdb_off_t tdb_find_dead(struct tdb_context *tdb, uint32_t hash,
 	return 0;
 }
 
-static int _tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
-		      int flag, uint32_t hash)
+static int _tdb_store(struct tdb_context *tdb, TDB_DATA key,
+		       TDB_DATA dbuf, int flag, uint32_t hash)
 {
-	struct list_struct rec;
+	struct tdb_record rec;
 	tdb_off_t rec_ptr;
 	char *p = NULL;
 	int ret = -1;
@@ -565,7 +593,7 @@ static int _tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 }
 
 /* store an element in the database, replacing any existing element
-   with the same key 
+   with the same key
 
    return 0 on success, -1 on failure
 */
@@ -576,8 +604,7 @@ int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
 
 	if (tdb->read_only || tdb->traverse_read) {
 		tdb->ecode = TDB_ERR_RDONLY;
-		tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag,
-					-TDB_ERR_RDONLY);
+		tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag, -1);
 		return -1;
 	}
 
@@ -592,7 +619,6 @@ int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
 	return ret;
 }
 
-
 /* Append to an entry. Create if not exist. */
 int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
 {
@@ -605,7 +631,7 @@ int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
 	if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
 		return -1;
 
-	dbuf = do_tdb_fetch(tdb, key);
+	dbuf = _tdb_fetch(tdb, key);
 
 	if (dbuf.dptr == NULL) {
 		dbuf.dptr = (unsigned char *)malloc(new_dbuf.dsize);
@@ -685,7 +711,6 @@ int tdb_get_seqnum(struct tdb_context *tdb)
 	tdb_off_t seqnum=0;
 
 	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
-	tdb_trace_ret(tdb, "tdb_get_seqnum", seqnum);
 	return seqnum;
 }
 
@@ -706,11 +731,41 @@ int tdb_get_flags(struct tdb_context *tdb)
 
 void tdb_add_flags(struct tdb_context *tdb, unsigned flags)
 {
+	if ((flags & TDB_ALLOW_NESTING) &&
+	    (flags & TDB_DISALLOW_NESTING)) {
+		tdb->ecode = TDB_ERR_NESTING;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_add_flags: "
+			"allow_nesting and disallow_nesting are not allowed together!"));
+		return;
+	}
+
+	if (flags & TDB_ALLOW_NESTING) {
+		tdb->flags &= ~TDB_DISALLOW_NESTING;
+	}
+	if (flags & TDB_DISALLOW_NESTING) {
+		tdb->flags &= ~TDB_ALLOW_NESTING;
+	}
+
 	tdb->flags |= flags;
 }
 
 void tdb_remove_flags(struct tdb_context *tdb, unsigned flags)
 {
+	if ((flags & TDB_ALLOW_NESTING) &&
+	    (flags & TDB_DISALLOW_NESTING)) {
+		tdb->ecode = TDB_ERR_NESTING;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_remove_flags: "
+			"allow_nesting and disallow_nesting are not allowed together!"));
+		return;
+	}
+
+	if (flags & TDB_ALLOW_NESTING) {
+		tdb->flags |= TDB_DISALLOW_NESTING;
+	}
+	if (flags & TDB_DISALLOW_NESTING) {
+		tdb->flags |= TDB_ALLOW_NESTING;
+	}
+
 	tdb->flags &= ~flags;
 }
 
@@ -730,7 +785,7 @@ void tdb_enable_seqnum(struct tdb_context *tdb)
  */
 static int tdb_free_region(struct tdb_context *tdb, tdb_off_t offset, ssize_t length)
 {
-	struct list_struct rec;
+	struct tdb_record rec;
 	if (length <= sizeof(rec)) {
 		/* the region is not worth adding */
 		return 0;
@@ -750,7 +805,7 @@ static int tdb_free_region(struct tdb_context *tdb, tdb_off_t offset, ssize_t le
 
 /*
   wipe the entire database, deleting all records. This can be done
-  very fast by using a global lock. The entire data portion of the
+  very fast by using a allrecord lock. The entire data portion of the
   file becomes a single entry in the freelist.
 
   This code carefully steps around the recovery area, leaving it alone
@@ -779,7 +834,7 @@ int tdb_wipe_all(struct tdb_context *tdb)
 	}
 
 	if (recovery_head != 0) {
-		struct list_struct rec;
+		struct tdb_record rec;
 		if (tdb->methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
 			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_wipe_all: failed to read recovery record\n"));
 			return -1;
@@ -841,6 +896,97 @@ failed:
 	return -1;
 }
 
+struct traverse_state {
+	bool error;
+	struct tdb_context *dest_db;
+};
+
+/*
+  traverse function for repacking
+ */
+static int repack_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private_data)
+{
+	struct traverse_state *state = (struct traverse_state *)private_data;
+	if (tdb_store(state->dest_db, key, data, TDB_INSERT) != 0) {
+		state->error = true;
+		return -1;
+	}
+	return 0;
+}
+
+/*
+  repack a tdb
+ */
+int tdb_repack(struct tdb_context *tdb)
+{
+	struct tdb_context *tmp_db;
+	struct traverse_state state;
+
+	tdb_trace(tdb, "tdb_repack");
+
+	if (tdb_transaction_start(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to start transaction\n"));
+		return -1;
+	}
+
+	tmp_db = tdb_open("tmpdb", tdb_hash_size(tdb), TDB_INTERNAL, O_RDWR|O_CREAT, 0);
+	if (tmp_db == NULL) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to create tmp_db\n"));
+		tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	state.error = false;
+	state.dest_db = tmp_db;
+
+	if (tdb_traverse_read(tdb, repack_traverse, &state) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying out\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;		
+	}
+
+	if (state.error) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during traversal\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	if (tdb_wipe_all(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to wipe database\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	state.error = false;
+	state.dest_db = tdb;
+
+	if (tdb_traverse_read(tmp_db, repack_traverse, &state) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying back\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;		
+	}
+
+	if (state.error) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during second traversal\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	tdb_close(tmp_db);
+
+	if (tdb_transaction_commit(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to commit\n"));
+		return -1;
+	}
+
+	return 0;
+}
+
 #ifdef TDB_TRACE
 static void tdb_trace_write(struct tdb_context *tdb, const char *str)
 {
@@ -853,10 +999,10 @@ static void tdb_trace_write(struct tdb_context *tdb, const char *str)
 static void tdb_trace_start(struct tdb_context *tdb)
 {
 	tdb_off_t seqnum=0;
-	char msg[sizeof(tdb_off_t) * 4];
+	char msg[sizeof(tdb_off_t) * 4 + 1];
 
 	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
-	sprintf(msg, "%u ", seqnum);
+	snprintf(msg, sizeof(msg), "%u ", seqnum);
 	tdb_trace_write(tdb, msg);
 }
 
@@ -867,8 +1013,8 @@ static void tdb_trace_end(struct tdb_context *tdb)
 
 static void tdb_trace_end_ret(struct tdb_context *tdb, int ret)
 {
-	char msg[sizeof(ret) * 4];
-	sprintf(msg, " = %i\n", ret);
+	char msg[sizeof(ret) * 4 + 4];
+	snprintf(msg, sizeof(msg), " = %i\n", ret);
 	tdb_trace_write(tdb, msg);
 }
 
@@ -883,10 +1029,11 @@ static void tdb_trace_record(struct tdb_context *tdb, TDB_DATA rec)
 		return;
 	}
 
+	/* snprintf here is purely cargo-cult programming. */
 	p = msg;
-	p += sprintf(p, " %zu:", rec.dsize);
+	p += snprintf(p, sizeof(msg), " %zu:", rec.dsize);
 	for (i = 0; i < rec.dsize; i++)
-		p += sprintf(p, "%02x", rec.dptr[i]);
+		p += snprintf(p, 2, "%02x", rec.dptr[i]);
 
 	tdb_trace_write(tdb, msg);
 }
@@ -898,12 +1045,23 @@ void tdb_trace(struct tdb_context *tdb, const char *op)
 	tdb_trace_end(tdb);
 }
 
+void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op)
+{
+	char msg[sizeof(tdb_off_t) * 4 + 1];
+
+	snprintf(msg, sizeof(msg), "%u ", seqnum);
+	tdb_trace_write(tdb, msg);
+	tdb_trace_write(tdb, op);
+	tdb_trace_end(tdb);
+}
+
 void tdb_trace_open(struct tdb_context *tdb, const char *op,
 		    unsigned hash_size, unsigned tdb_flags, unsigned open_flags)
 {
 	char msg[128];
 
-	sprintf(msg, "%s %u %#x %#x", op, hash_size, tdb_flags, open_flags);
+	snprintf(msg, sizeof(msg),
+		 "%s %u 0x%x 0x%x", op, hash_size, tdb_flags, open_flags);
 	tdb_trace_start(tdb);
 	tdb_trace_write(tdb, msg);
 	tdb_trace_end(tdb);
@@ -958,9 +1116,9 @@ void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
 			     TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
 			     int ret)
 {
-	char msg[sizeof(ret) * 4];
+	char msg[1 + sizeof(ret) * 4];
 
-	sprintf(msg, " %#x", flag); 
+	snprintf(msg, sizeof(msg), " %#x", flag);
 	tdb_trace_start(tdb);
 	tdb_trace_write(tdb, op);
 	tdb_trace_record(tdb, rec1);