tdb2: clean up locking a little bit, fix hash wrap problem.
authorRusty Russell <rusty@rustcorp.com.au>
Thu, 26 Aug 2010 12:18:28 +0000 (21:48 +0930)
committerRusty Russell <rusty@rustcorp.com.au>
Thu, 26 Aug 2010 12:18:28 +0000 (21:48 +0930)
ccan/tdb2/tdb.c

index 2a89e36a42e70ec150cda354da831bb20cf0b145..151eebe92a20286bea09b20e39e6dcac0e2ef7ef 100644 (file)
@@ -380,122 +380,150 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
        return NULL;
 }
 
-static int tdb_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
+static tdb_off_t hash_off(struct tdb_context *tdb, uint64_t list)
 {
-       return memcmp(data.dptr, key.dptr, data.dsize) == 0;
+       return tdb->header.v.hash_off
+               + ((list & ((1ULL << tdb->header.v.hash_bits) - 1))
+                  * sizeof(tdb_off_t));
 }
 
-static void unlock_lists(struct tdb_context *tdb,
-                        uint64_t start, uint64_t end, int ltype)
+/* Returns 0 if the entry is a zero (definitely not a match).
+ * Returns a valid entry offset if it's a match.  Fills in rec.
+ * Otherwise returns TDB_OFF_ERR: keep searching. */
+static tdb_off_t entry_matches(struct tdb_context *tdb,
+                              uint64_t list,
+                              uint64_t hash,
+                              const struct tdb_data *key,
+                              struct tdb_used_record *rec)
 {
-       for (;;) {
-               tdb_unlock_list(tdb, start, ltype);
-               if (start == end)
-                       return;
-               start = (start + 1) & ((1ULL << tdb->header.v.hash_bits) - 1);
-       }
-}
-
-/* FIXME: Return header copy? */
-/* Returns -1 or offset of entry (0 if not found).
- * Locks hash entried from *start to *end (where the entry was found). */
-static tdb_off_t find_bucket_and_lock(struct tdb_context *tdb,
-                                     const struct tdb_data *key,
-                                     uint64_t hash,
-                                     uint64_t *start,
-                                     uint64_t *end,
-                                     uint64_t *room,
-                                     int ltype)
-{
-       uint64_t hextra;
        tdb_off_t off;
+       uint64_t keylen;
+       const unsigned char *rkey;
 
-       /* FIXME: can we avoid locks for some fast paths? */
-       *start = tdb_lock_list(tdb, hash, ltype, TDB_LOCK_WAIT);
-       if (*start == TDB_OFF_ERR)
+       off = tdb_read_off(tdb, tdb->header.v.hash_off
+                          + list * sizeof(tdb_off_t));
+       if (off == 0 || off == TDB_OFF_ERR)
+               return off;
+
+#if 0 /* FIXME: Check other bits. */
+       unsigned int bits, bitmask, hoffextra;
+       /* Bottom three bits show how many extra hash bits. */
+       bits = (off & ((1 << TDB_EXTRA_HASHBITS_NUM) - 1)) + 1;
+       bitmask = (1 << bits)-1;
+       hoffextra = ((off >> TDB_EXTRA_HASHBITS_NUM) & bitmask);
+       uint64_t hextra = hash >> tdb->header.v.hash_bits;
+       if ((hextra & bitmask) != hoffextra) 
                return TDB_OFF_ERR;
+       off &= ~...;
+#endif
 
-       *end = *start;
-       hextra = hash >> tdb->header.v.hash_bits;
+       if (tdb_read_convert(tdb, off, rec, sizeof(*rec)) == -1)
+               return TDB_OFF_ERR;
 
-       while ((off = tdb_read_off(tdb, tdb->header.v.hash_off
-                                  + *end * sizeof(tdb_off_t)))
-              != TDB_OFF_ERR) {
-               struct tdb_used_record pad, *r;
-               uint64_t keylen;
+       /* FIXME: check extra bits in header! */
+       keylen = rec_key_length(rec);
+       if (keylen != key->dsize)
+               return TDB_OFF_ERR;
 
-               /* Didn't find it? */
-               if (!off)
-                       return 0;
+       rkey = tdb_access_read(tdb, off + sizeof(*rec), keylen);
+       if (!rkey)
+               return TDB_OFF_ERR;
+       if (memcmp(rkey, key->dptr, keylen) != 0)
+               off = TDB_OFF_ERR;
+       tdb_access_release(tdb, rkey);
+       return off;
+}
 
-#if 0 /* FIXME: Check other bits. */
-               unsigned int bits, bitmask, hoffextra;
-               /* Bottom three bits show how many extra hash bits. */
-               bits = (off & ((1 << TDB_EXTRA_HASHBITS_NUM) - 1)) + 1;
-               bitmask = (1 << bits)-1;
-               hoffextra = ((off >> TDB_EXTRA_HASHBITS_NUM) & bitmask);
-               if ((hextra & bitmask) != hoffextra) 
-                       goto lock_next;
-#endif
+/* FIXME: Optimize? */
+static void unlock_range(struct tdb_context *tdb,
+                        tdb_off_t list, tdb_len_t num,
+                        int ltype)
+{
+       tdb_off_t i;
 
-               r = tdb_get(tdb, off, &pad, sizeof(*r));
-               if (!r)
-                       goto unlock_err;
+       for (i = list; i < list + num; i++)
+               tdb_unlock_list(tdb, i, ltype);
+}
 
-               if (rec_magic(r) != TDB_MAGIC) {
-                       tdb->ecode = TDB_ERR_CORRUPT;
-                       tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-                                "find_bucket_and_lock: bad magic 0x%llx"
-                                " at offset %llu!\n",
-                                (long long)rec_magic(r), (long long)off);
-                       goto unlock_err;
+/* FIXME: Optimize? */
+static int lock_range(struct tdb_context *tdb,
+                     tdb_off_t list, tdb_len_t num,
+                     int ltype)
+{
+       tdb_off_t i;
+
+       for (i = list; i < list + num; i++) {
+               if (tdb_lock_list(tdb, i, ltype, TDB_LOCK_WAIT) != 0) {
+                       unlock_range(tdb, list, i - list, ltype);
+                       return -1;
                }
+       }
+       return 0;
+}
 
-               /* FIXME: check extra bits in header! */
-               keylen = rec_key_length(r);
-               if (keylen != key->dsize)
-                       goto lock_next;
-
-               switch (tdb_parse_data(tdb, *key, off + sizeof(*r), key->dsize,
-                                      tdb_key_compare, NULL)) {
-               case 1:
-                       /* Match! */
-                       *room = rec_data_length(r) + rec_extra_padding(r);
-                       return off >> TDB_EXTRA_HASHBITS_NUM;
-               case 0:
-                       break;
-               default:
-                       goto unlock_err;
+/* We lock hashes up to the next empty offset.  We already hold the
+ * lock on the start bucket, but we may need to release and re-grab
+ * it.  If we fail, we hold no locks at all! */
+static tdb_len_t relock_hash_to_zero(struct tdb_context *tdb,
+                                    tdb_off_t start, int ltype)
+{
+       tdb_len_t num, len, pre_locks;
+
+again:
+       num = 1ULL << tdb->header.v.hash_bits;
+       len = tdb_find_zero_off(tdb, hash_off(tdb, start), num - start);
+       if (unlikely(len == num - start)) {
+               /* We hit the end of the hash range.  Drop lock: we have
+                  to lock start of hash first. */
+               tdb_unlock_list(tdb, start, ltype);
+               /* Grab something, so header is stable. */
+               if (tdb_lock_list(tdb, 0, ltype, TDB_LOCK_WAIT))
+                       return TDB_OFF_ERR;
+               len = tdb_find_zero_off(tdb, hash_off(tdb, 0), num);
+               if (lock_range(tdb, 1, len, ltype) == -1) {
+                       tdb_unlock_list(tdb, 0, ltype);
+                       return TDB_OFF_ERR;
                }
+               pre_locks = len;
+               len = num - start;
+       } else {
+               /* We already have lock on start. */
+               start++;
+               pre_locks = 0;
+       }
+       if (unlikely(lock_range(tdb, start, len, ltype) == -1)) {
+               if (pre_locks)
+                       unlock_range(tdb, 0, pre_locks, ltype);
+               else
+                       tdb_unlock_list(tdb, start, ltype);
+               return TDB_OFF_ERR;
+       }
 
-       lock_next:
-               /* Lock next bucket. */
-               /* FIXME: We can deadlock if this wraps! */
-               off = tdb_lock_list(tdb, ++hash, ltype, TDB_LOCK_WAIT);
-               if (off == TDB_OFF_ERR)
-                       goto unlock_err;
-               *end = off;
+       /* Now, did we lose the race, and it's not zero any more? */
+       if (unlikely(tdb_read_off(tdb, hash_off(tdb, pre_locks + len)) != 0)) {
+               unlock_range(tdb, 0, pre_locks, ltype);
+               /* Leave the start locked, as expected. */
+               unlock_range(tdb, start + 1, len - 1, ltype);
+               goto again;
        }
 
-unlock_err:
-       TEST_IT(*end < *start);
-       unlock_lists(tdb, *start, *end, ltype);
-       return TDB_OFF_ERR;
+       return pre_locks + len;
 }
 
+/* FIXME: modify, don't rewrite! */
 static int update_rec_hdr(struct tdb_context *tdb,
                          tdb_off_t off,
                          tdb_len_t keylen,
                          tdb_len_t datalen,
-                         tdb_len_t room,
+                         struct tdb_used_record *rec,
                          uint64_t h)
 {
-       struct tdb_used_record rec;
+       uint64_t room = rec_data_length(rec) + rec_extra_padding(rec);
 
-       if (set_header(tdb, &rec, keylen, datalen, room - datalen, h))
+       if (set_header(tdb, rec, keylen, datalen, room - datalen, h))
                return -1;
 
-       return tdb_write_convert(tdb, off, &rec, sizeof(rec));
+       return tdb_write_convert(tdb, off, rec, sizeof(*rec));
 }
 
 /* If we fail, others will try after us. */
@@ -569,15 +597,42 @@ unlock:
 int tdb_store(struct tdb_context *tdb,
              struct tdb_data key, struct tdb_data dbuf, int flag)
 {
-       tdb_off_t new_off, off, start, end, room;
+       tdb_off_t new_off, off, old_bucket, start, num_locks = 1;
+       struct tdb_used_record rec;
        uint64_t h;
        bool growing = false;
 
        h = tdb_hash(tdb, key.dptr, key.dsize);
-       off = find_bucket_and_lock(tdb, &key, h, &start, &end, &room, F_WRLCK);
-       if (off == TDB_OFF_ERR)
+
+       /* FIXME: can we avoid locks for some fast paths? */
+       start = tdb_lock_list(tdb, h, F_WRLCK, TDB_LOCK_WAIT);
+       if (start == TDB_OFF_ERR)
                return -1;
 
+       /* Fast path. */
+       old_bucket = start;
+       off = entry_matches(tdb, start, h, &key, &rec);
+       if (unlikely(off == TDB_OFF_ERR)) {
+               /* Slow path, need to grab more locks and search. */
+               tdb_off_t i;
+
+               /* Warning: this may drop the lock!  Does that on error. */
+               num_locks = relock_hash_to_zero(tdb, start, F_WRLCK);
+               if (num_locks == TDB_OFF_ERR)
+                       return -1;
+
+               for (i = start; i < start + num_locks; i++) {
+                       off = entry_matches(tdb, i, h, &key, &rec);
+                       /* Empty entry or we found it? */
+                       if (off == 0 || off != TDB_OFF_ERR) {
+                               old_bucket = i;
+                               break;
+                       }
+               }
+               if (i == start + num_locks)
+                       off = 0;
+       }
+
        /* Now we have lock on this hash bucket. */
        if (flag == TDB_INSERT) {
                if (off) {
@@ -586,11 +641,12 @@ int tdb_store(struct tdb_context *tdb,
                }
        } else {
                if (off) {
-                       if (room >= key.dsize + dbuf.dsize) {
+                       if (rec_data_length(&rec) + rec_extra_padding(&rec)
+                           >= dbuf.dsize) {
                                new_off = off;
                                if (update_rec_hdr(tdb, off,
                                                   key.dsize, dbuf.dsize,
-                                                  room, h))
+                                                  &rec, h))
                                        goto fail;
                                goto write;
                        }
@@ -611,7 +667,7 @@ int tdb_store(struct tdb_context *tdb,
        /* Allocate a new record. */
        new_off = alloc(tdb, key.dsize, dbuf.dsize, h, growing);
        if (new_off == 0) {
-               unlock_lists(tdb, start, end, F_WRLCK);
+               unlock_range(tdb, start, num_locks, F_WRLCK);
                /* Expand, then try again... */
                if (tdb_expand(tdb, key.dsize, dbuf.dsize, growing) == -1)
                        return -1;
@@ -621,13 +677,14 @@ int tdb_store(struct tdb_context *tdb,
        /* We didn't like the existing one: remove it. */
        if (off) {
                add_free_record(tdb, off, sizeof(struct tdb_used_record)
-                               + key.dsize + room);
+                               + rec_key_length(&rec)
+                               + rec_data_length(&rec)
+                               + rec_extra_padding(&rec));
        }
 
 write:
-       off = tdb->header.v.hash_off + end * sizeof(tdb_off_t);
        /* FIXME: Encode extra hash bits! */
-       if (tdb_write_off(tdb, off, new_off) == -1)
+       if (tdb_write_off(tdb, hash_off(tdb, old_bucket), new_off) == -1)
                goto fail;
 
        off = new_off + sizeof(struct tdb_used_record);
@@ -638,48 +695,65 @@ write:
                goto fail;
 
        /* FIXME: tdb_increment_seqnum(tdb); */
-       unlock_lists(tdb, start, end, F_WRLCK);
+       unlock_range(tdb, start, num_locks, F_WRLCK);
 
-       /* By simple trial and error, this roughly approximates a 60%
-        * full measure. */
-       if (unlikely(end - start > 4 * tdb->header.v.hash_bits - 32))
+       /* FIXME: by simple simulation, this approximated 60% full.
+        * Check in real case! */
+       if (unlikely(num_locks > 4 * tdb->header.v.hash_bits - 31))
                enlarge_hash(tdb);
 
        return 0;
 
 fail:
-       unlock_lists(tdb, start, end, F_WRLCK);
+       unlock_range(tdb, start, num_locks, F_WRLCK);
        return -1;
 }
 
 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 {
-       tdb_off_t off, start, end, room;
+       tdb_off_t off, start, num_locks = 1;
+       struct tdb_used_record rec;
        uint64_t h;
-       struct tdb_used_record pad, *r;
        struct tdb_data ret;
 
        h = tdb_hash(tdb, key.dptr, key.dsize);
-       off = find_bucket_and_lock(tdb, &key, h, &start, &end, &room, F_RDLCK);
-       if (off == TDB_OFF_ERR)
-               return tdb_null;
 
-       if (!off) {
-               unlock_lists(tdb, start, end, F_RDLCK);
-               tdb->ecode = TDB_SUCCESS;
+       /* FIXME: can we avoid locks for some fast paths? */
+       start = tdb_lock_list(tdb, h, F_RDLCK, TDB_LOCK_WAIT);
+       if (start == TDB_OFF_ERR)
                return tdb_null;
+
+       /* Fast path. */
+       off = entry_matches(tdb, start, h, &key, &rec);
+       if (unlikely(off == TDB_OFF_ERR)) {
+               /* Slow path, need to grab more locks and search. */
+               tdb_off_t i;
+
+               /* Warning: this may drop the lock!  Does that on error. */
+               num_locks = relock_hash_to_zero(tdb, start, F_RDLCK);
+               if (num_locks == TDB_OFF_ERR)
+                       return tdb_null;
+
+               for (i = start; i < start + num_locks; i++) {
+                       off = entry_matches(tdb, i, h, &key, &rec);
+                       /* Empty entry or we found it? */
+                       if (off == 0 || off != TDB_OFF_ERR)
+                               break;
+               }
+               if (i == start + num_locks)
+                       off = 0;
        }
 
-       r = tdb_get(tdb, off, &pad, sizeof(*r));
-       if (!r) {
-               unlock_lists(tdb, start, end, F_RDLCK);
+       if (!off) {
+               unlock_range(tdb, start, num_locks, F_RDLCK);
+               tdb->ecode = TDB_ERR_NOEXIST;
                return tdb_null;
        }
 
-       ret.dsize = rec_data_length(r);
-       ret.dptr = tdb_alloc_read(tdb, off + sizeof(*r) + key.dsize,
+       ret.dsize = rec_data_length(&rec);
+       ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
                                  ret.dsize);
-       unlock_lists(tdb, start, end, F_RDLCK);
+       unlock_range(tdb, start, num_locks, F_RDLCK);
        return ret;
 }
 
@@ -687,157 +761,127 @@ static int hash_add(struct tdb_context *tdb, uint64_t h, tdb_off_t off)
 {
        tdb_off_t i, hoff, len, num;
 
+       /* Look for next space. */
        i = (h & ((1ULL << tdb->header.v.hash_bits) - 1));
-       hoff = tdb->header.v.hash_off + i * sizeof(tdb_off_t);
        len = (1ULL << tdb->header.v.hash_bits) - i;
+       num = tdb_find_zero_off(tdb, hash_off(tdb, i), len);
 
-       /* Look for next space. */
-       num = tdb_find_zero_off(tdb, hoff, len);
        if (unlikely(num == len)) {
-               hoff = tdb->header.v.hash_off;
+               /* We wrapped.  Look through start of hash table. */
+               hoff = hash_off(tdb, 0);
                len = (1ULL << tdb->header.v.hash_bits);
                num = tdb_find_zero_off(tdb, hoff, len);
-               if (i == len)
+               if (i == len) {
+                       tdb->ecode = TDB_ERR_CORRUPT;
+                       tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
+                                "hash_add: full hash table!\n");
                        return -1;
+               }
        }
        /* FIXME: Encode extra hash bits! */
-       return tdb_write_off(tdb, hoff + num * sizeof(tdb_off_t), off);
+       return tdb_write_off(tdb, hash_off(tdb, i + num), off);
 }
 
-static int unlink_used_record(struct tdb_context *tdb, tdb_off_t chain,
-                             uint64_t *extra_locks)
+int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 {
-       tdb_off_t num, len, i, hoff;
+       tdb_off_t i, old_bucket, off, start, num_locks = 1;
+       struct tdb_used_record rec;
+       uint64_t h;
 
-       /* FIXME: Maybe lock more in search?  Maybe don't lock if scan
-        * finds none? */
-again:
-       len = (1ULL << tdb->header.v.hash_bits) - (chain + 1);
-       hoff = tdb->header.v.hash_off + (chain + 1) * sizeof(tdb_off_t);
-       num = tdb_find_zero_off(tdb, hoff, len);
-
-       /* We want to lock the zero entry, too.  In the wrap case,
-        * this locks one extra.  That's harmless. */
-       num++;
-
-       for (i = chain + 1; i < chain + 1 + num; i++) {
-               if (tdb_lock_list(tdb, i, F_WRLCK, TDB_LOCK_WAIT)
-                   == TDB_OFF_ERR) {
-                       if (i != chain + 1)
-                               unlock_lists(tdb, chain + 1, i-1, F_WRLCK);
+       h = tdb_hash(tdb, key.dptr, key.dsize);
+
+       /* FIXME: can we avoid locks for some fast paths? */
+       start = tdb_lock_list(tdb, h, F_WRLCK, TDB_LOCK_WAIT);
+       if (start == TDB_OFF_ERR)
+               return -1;
+
+       /* Fast path. */
+       old_bucket = start;
+       off = entry_matches(tdb, start, h, &key, &rec);
+       if (off && off != TDB_OFF_ERR) {
+               /* We can only really fastpath delete if next bucket
+                * is 0.  Note that we haven't locked it, but our lock
+                * on this bucket stops anyone overflowing into it
+                * while we look. */
+               if (tdb_read_off(tdb, hash_off(tdb, h+1)) == 0)
+                       goto delete;
+               /* Slow path. */
+               off = TDB_OFF_ERR;
+       }
+
+       if (unlikely(off == TDB_OFF_ERR)) {
+               /* Slow path, need to grab more locks and search. */
+               tdb_off_t i;
+
+               /* Warning: this may drop the lock!  Does that on error. */
+               num_locks = relock_hash_to_zero(tdb, start, F_WRLCK);
+               if (num_locks == TDB_OFF_ERR)
                        return -1;
-               }
-       }
 
-       /* The wrap case: we need those locks out of order! */
-       if (unlikely(num == len + 1)) {
-               *extra_locks = tdb_find_zero_off(tdb, tdb->header.v.hash_off,
-                                                1ULL << tdb->header.v.hash_bits);
-               (*extra_locks)++;
-               for (i = 0; i < *extra_locks; i++) {
-                       if (tdb_lock_list(tdb, i, F_WRLCK, TDB_LOCK_NOWAIT)
-                           == TDB_OFF_ERR) {
-                               /* Failed.  Caller must lock in order. */
-                               if (i)
-                                       unlock_lists(tdb, 0, i-1, F_WRLCK);
-                               unlock_lists(tdb, chain + 1, chain + num,
-                                            F_WRLCK);
-                               return 1;
+               for (i = start; i < start + num_locks; i++) {
+                       off = entry_matches(tdb, i, h, &key, &rec);
+                       /* Empty entry or we found it? */
+                       if (off == 0 || off != TDB_OFF_ERR) {
+                               old_bucket = i;
+                               break;
                        }
                }
-               num += *extra_locks;
+               if (i == start + num_locks)
+                       off = 0;
        }
 
-       /* Now we have the locks, be certain that offset is still 0! */
-       hoff = tdb->header.v.hash_off
-               + (((chain + num) * sizeof(tdb_off_t))
-                  & ((1ULL << tdb->header.v.hash_bits) - 1));
-
-       if (unlikely(tdb_read_off(tdb, hoff) != 0)) {
-               unlock_lists(tdb, chain + 1, chain + num, F_WRLCK);
-               goto again;
+       if (!off) {
+               unlock_range(tdb, start, num_locks, F_WRLCK);
+               tdb->ecode = TDB_ERR_NOEXIST;
+               return -1;
        }
 
-       /* OK, all locked.  Unlink first one. */
-       hoff = tdb->header.v.hash_off + chain * sizeof(tdb_off_t);
-       if (tdb_write_off(tdb, hoff, 0) == -1)
+delete:
+       /* This actually unlinks it. */
+       if (tdb_write_off(tdb, hash_off(tdb, old_bucket), 0) == -1)
                goto unlock_err;
 
-       /* Rehash the rest. */
-       for (i = 1; i < num; i++) {
-               tdb_off_t off;
-               uint64_t h;
+       /* Rehash anything following. */
+       for (i = old_bucket+1; i < h + num_locks; i++) {
+               tdb_off_t off2;
+               uint64_t h2;
 
-               hoff = tdb->header.v.hash_off
-                       + (((chain + i) * sizeof(tdb_off_t))
-                          & ((1ULL << tdb->header.v.hash_bits) - 1));
-               off = tdb_read_off(tdb, hoff);
-               if (unlikely(off == TDB_OFF_ERR))
+               off2 = tdb_read_off(tdb, hash_off(tdb, i));
+               if (unlikely(off2 == TDB_OFF_ERR))
                        goto unlock_err;
 
                /* Maybe use a bit to indicate it is in ideal place? */
-               h = hash_record(tdb, off);
+               h2 = hash_record(tdb, off2);
                /* Is it happy where it is? */
-               if ((h & ((1ULL << tdb->header.v.hash_bits)-1)) == (chain + i))
+               if ((h2 & ((1ULL << tdb->header.v.hash_bits)-1))
+                   == (i & ((1ULL << tdb->header.v.hash_bits)-1)))
                        continue;
 
                /* Remove it. */
-               if (tdb_write_off(tdb, hoff, 0) == -1)
+               if (tdb_write_off(tdb, hash_off(tdb, i), 0) == -1)
                        goto unlock_err;
 
                /* Rehash it. */
-               if (hash_add(tdb, h, off) == -1)
+               if (hash_add(tdb, h2, off2) == -1)
                        goto unlock_err;
        }
-       unlock_lists(tdb, chain + 1, chain + num, F_WRLCK);
+
+       /* Free the deleted entry. */
+       if (add_free_record(tdb, off,
+                           sizeof(struct tdb_used_record)
+                           + rec_key_length(&rec)
+                           + rec_data_length(&rec)
+                           + rec_extra_padding(&rec)) != 0)
+               goto unlock_err;
+
+       unlock_range(tdb, start, num_locks, F_WRLCK);
        return 0;
 
 unlock_err:
-       unlock_lists(tdb, chain + 1, chain + num, F_WRLCK);
+       unlock_range(tdb, start, num_locks, F_WRLCK);
        return -1;
 }
 
-int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
-{
-       tdb_off_t off, start, end, room, extra_locks = 0;
-       uint64_t h;
-       int ret;
-
-       h = tdb_hash(tdb, key.dptr, key.dsize);
-       off = find_bucket_and_lock(tdb, &key, h, &start, &end, &room, F_WRLCK);
-       if (off == TDB_OFF_ERR)
-               return -1;
-
-       if (off == 0) {
-               unlock_lists(tdb, start, end, F_WRLCK);
-               tdb->ecode = TDB_ERR_NOEXIST;
-               return -1;
-       }
-
-       ret = unlink_used_record(tdb, end, &extra_locks);
-       if (unlikely(ret == 1)) {
-               unsigned int i;
-
-               unlock_lists(tdb, start, end, F_WRLCK);
-
-               /* We need extra locks at the start. */
-               for (i = 0; i < extra_locks; i++) {
-                       if (tdb_lock_list(tdb, i, F_WRLCK, TDB_LOCK_WAIT)
-                           == TDB_OFF_ERR) {
-                               if (i)
-                                       unlock_lists(tdb, 0, i-1, F_WRLCK);
-                               return -1;
-                       }
-               }
-               /* Try again now we're holding more locks. */
-               ret = tdb_delete(tdb, key);
-               unlock_lists(tdb, 0, i, F_WRLCK);
-               return ret;
-       }
-       unlock_lists(tdb, start, end, F_WRLCK);
-       return ret;
-}
-
 int tdb_close(struct tdb_context *tdb)
 {
        struct tdb_context **i;