tdb2: feature support.
[ccan] / ccan / tdb2 / lock.c
index 14f9c0ae307b348b1a7401b6001bbd419a8e0c69..b1799a7e1b9aa0ca9ec5652f573ba74ac7019761 100644 (file)
@@ -1,4 +1,4 @@
- /* 
+ /*
    Unix SMB/CIFS implementation.
 
    trivial database library
@@ -40,10 +40,13 @@ static int fcntl_lock(struct tdb_context *tdb,
        fl.l_len = len;
        fl.l_pid = 0;
 
+       add_stat(tdb, lock_lowlevel, 1);
        if (waitflag)
                return fcntl(tdb->fd, F_SETLKW, &fl);
-       else
+       else {
+               add_stat(tdb, lock_nonblock, 1);
                return fcntl(tdb->fd, F_SETLK, &fl);
+       }
 }
 
 static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
@@ -99,7 +102,7 @@ static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
        }
 
        if (!found) {
-               fprintf(stderr, "Unlock on %u@%u not found!\n",
+               fprintf(stderr, "Unlock on %u@%u not found!",
                        (int)off, (int)len);
                abort();
        }
@@ -117,33 +120,31 @@ static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
 }
 
 /* a byte range locking function - return 0 on success
-   this functions locks/unlocks 1 byte at the specified offset.
+   this functions locks len bytes at the specified offset.
 
    note that a len of zero means lock to end of file
 */
-static int tdb_brlock(struct tdb_context *tdb,
-                     int rw_type, tdb_off_t offset, tdb_off_t len,
-                     enum tdb_lock_flags flags)
+static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
+                                int rw_type, tdb_off_t offset, tdb_off_t len,
+                                enum tdb_lock_flags flags)
 {
        int ret;
 
        if (tdb->flags & TDB_NOLOCK) {
-               return 0;
+               return TDB_SUCCESS;
        }
 
        if (rw_type == F_WRLCK && tdb->read_only) {
-               tdb->ecode = TDB_ERR_RDONLY;
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
+                                 "Write lock attempted on read-only database");
        }
 
        /* A 32 bit system cannot open a 64-bit file, but it could have
         * expanded since then: check here. */
        if ((size_t)(offset + len) != offset + len) {
-               tdb->ecode = TDB_ERR_IO;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_brlock: lock on giant offset %llu\n",
-                        (long long)(offset + len));
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+                                 "tdb_brlock: lock on giant offset %llu",
+                                 (long long)(offset + len));
        }
 
        do {
@@ -152,29 +153,29 @@ static int tdb_brlock(struct tdb_context *tdb,
        } while (ret == -1 && errno == EINTR);
 
        if (ret == -1) {
-               tdb->ecode = TDB_ERR_LOCK;
                /* Generic lock error. errno set by fcntl.
                 * EAGAIN is an expected return from non-blocking
                 * locks. */
                if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
-                       tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                                "tdb_brlock failed (fd=%d) at"
-                                " offset %llu rw_type=%d flags=%d len=%llu\n",
-                                tdb->fd, (long long)offset, rw_type,
-                                flags, (long long)len);
+                       tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                  "tdb_brlock failed (fd=%d) at"
+                                  " offset %zu rw_type=%d flags=%d len=%zu:"
+                                  " %s",
+                                  tdb->fd, (size_t)offset, rw_type,
+                                  flags, (size_t)len, strerror(errno));
                }
-               return -1;
+               return TDB_ERR_LOCK;
        }
-       return 0;
+       return TDB_SUCCESS;
 }
 
-static int tdb_brunlock(struct tdb_context *tdb,
-                       int rw_type, tdb_off_t offset, size_t len)
+static enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
+                                  int rw_type, tdb_off_t offset, size_t len)
 {
        int ret;
 
        if (tdb->flags & TDB_NOLOCK) {
-               return 0;
+               return TDB_SUCCESS;
        }
 
        do {
@@ -182,47 +183,46 @@ static int tdb_brunlock(struct tdb_context *tdb,
        } while (ret == -1 && errno == EINTR);
 
        if (ret == -1) {
-               tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
-                        "tdb_brunlock failed (fd=%d) at offset %llu"
-                        " rw_type=%d len=%llu\n",
-                        tdb->fd, (long long)offset, rw_type, (long long)len);
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_brunlock failed (fd=%d) at offset %zu"
+                                 " rw_type=%d len=%zu",
+                                 tdb->fd, (size_t)offset, rw_type,
+                                 (size_t)len);
        }
-       return ret;
+       return TDB_SUCCESS;
 }
 
-#if 0
 /*
   upgrade a read lock to a write lock. This needs to be handled in a
   special way as some OSes (such as solaris) have too conservative
   deadlock detection and claim a deadlock when progress can be
-  made. For those OSes we may loop for a while.  
+  made. For those OSes we may loop for a while.
 */
-int tdb_allrecord_upgrade(struct tdb_context *tdb)
+enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
 {
        int count = 1000;
 
        if (tdb->allrecord_lock.count != 1) {
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_upgrade failed: count %u too high\n",
-                        tdb->allrecord_lock.count);
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_allrecord_upgrade failed:"
+                                 " count %u too high",
+                                 tdb->allrecord_lock.count);
        }
 
        if (tdb->allrecord_lock.off != 1) {
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_upgrade failed: already upgraded?\n");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_allrecord_upgrade failed:"
+                                 " already upgraded?");
        }
 
        while (count--) {
                struct timeval tv;
                if (tdb_brlock(tdb, F_WRLCK,
-                              TDB_HASH_LOCK_START
-                              + (1ULL << tdb->header.v.hash_bits), 0,
-                              TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
+                              TDB_HASH_LOCK_START, 0,
+                              TDB_LOCK_WAIT|TDB_LOCK_PROBE) == TDB_SUCCESS) {
                        tdb->allrecord_lock.ltype = F_WRLCK;
                        tdb->allrecord_lock.off = 0;
-                       return 0;
+                       return TDB_SUCCESS;
                }
                if (errno != EDEADLK) {
                        break;
@@ -232,11 +232,9 @@ int tdb_allrecord_upgrade(struct tdb_context *tdb)
                tv.tv_usec = 1;
                select(0, NULL, NULL, NULL, &tv);
        }
-       tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
-                "tdb_allrecord_upgrade failed\n");
-       return -1;
+       return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                         "tdb_allrecord_upgrade failed");
 }
-#endif
 
 static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
                                           tdb_off_t offset)
@@ -251,50 +249,103 @@ static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
        return NULL;
 }
 
+enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb)
+{
+       enum TDB_ERROR ecode;
+
+       ecode = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
+                                  false);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
+
+       ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
+       if (ecode != TDB_SUCCESS) {
+               tdb_allrecord_unlock(tdb, F_WRLCK);
+               return ecode;
+       }
+       ecode = tdb_transaction_recover(tdb);
+       tdb_unlock_open(tdb);
+       tdb_allrecord_unlock(tdb, F_WRLCK);
+
+       return ecode;
+}
+
 /* lock an offset in the database. */
-static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
-                        enum tdb_lock_flags flags)
+static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
+                                   tdb_off_t offset, int ltype,
+                                   enum tdb_lock_flags flags)
 {
        struct tdb_lock_type *new_lck;
+       enum TDB_ERROR ecode;
 
-       if (offset >= TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-                        "tdb_nest_lock: invalid offset %llu ltype=%d\n",
-                        (long long)offset, ltype);
-               return -1;
+       if (offset > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_nest_lock: invalid offset %zu ltype=%d",
+                                 (size_t)offset, ltype);
        }
 
        if (tdb->flags & TDB_NOLOCK)
-               return 0;
+               return TDB_SUCCESS;
+
+       add_stat(tdb, locks, 1);
 
        new_lck = find_nestlock(tdb, offset);
        if (new_lck) {
-               /*
-                * Just increment the in-memory struct, posix locks
-                * don't stack.
-                */
+               if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
+                       return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                         "tdb_nest_lock:"
+                                         " offset %zu has read lock",
+                                         (size_t)offset);
+               }
+               /* Just increment the struct, posix locks don't stack. */
                new_lck->count++;
-               return 0;
+               return TDB_SUCCESS;
+       }
+
+       if (tdb->num_lockrecs
+           && offset >= TDB_HASH_LOCK_START
+           && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_nest_lock: already have a hash lock?");
        }
 
        new_lck = (struct tdb_lock_type *)realloc(
                tdb->lockrecs,
                sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
        if (new_lck == NULL) {
-               tdb->ecode = TDB_ERR_OOM;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_nest_lock: unable to allocate %llu lock struct",
-                        (long long)(tdb->num_lockrecs + 1));
-               errno = ENOMEM;
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+                                 "tdb_nest_lock:"
+                                 " unable to allocate %zu lock struct",
+                                 tdb->num_lockrecs + 1);
        }
        tdb->lockrecs = new_lck;
 
        /* Since fcntl locks don't nest, we do a lock for the first one,
           and simply bump the count for future ones */
-       if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
-               return -1;
+       ecode = tdb_brlock(tdb, ltype, offset, 1, flags);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
+
+       /* First time we grab a lock, perhaps someone died in commit? */
+       if (!(flags & TDB_LOCK_NOCHECK)
+           && tdb->num_lockrecs == 0) {
+               tdb_bool_err berr = tdb_needs_recovery(tdb);
+               if (berr != false) {
+                       tdb_brunlock(tdb, ltype, offset, 1);
+
+                       if (berr < 0)
+                               return berr;
+                       ecode = tdb_lock_and_recover(tdb);
+                       if (ecode == TDB_SUCCESS) {
+                               ecode = tdb_brlock(tdb, ltype, offset, 1,
+                                                  flags);
+                       }
+                       if (ecode != TDB_SUCCESS) {
+                               return ecode;
+                       }
+               }
        }
 
        tdb->lockrecs[tdb->num_lockrecs].off = offset;
@@ -302,62 +353,28 @@ static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
        tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
        tdb->num_lockrecs++;
 
-       return 0;
+       return TDB_SUCCESS;
 }
 
-static int tdb_lock_and_recover(struct tdb_context *tdb)
+static enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
+                                     tdb_off_t off, int ltype)
 {
-#if 0 /* FIXME */
-
-       int ret;
-
-       /* We need to match locking order in transaction commit. */
-       if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT)) {
-               return -1;
-       }
-
-       if (tdb_brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT)) {
-               tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
-               return -1;
-       }
-
-       ret = tdb_transaction_recover(tdb);
-
-       tdb_brunlock(tdb, F_WRLCK, OPEN_LOCK, 1);
-       tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
-
-       return ret;
-#else
-       abort();
-       return -1;
-#endif
-}
-
-static bool tdb_needs_recovery(struct tdb_context *tdb)
-{
-       /* FIXME */
-       return false;
-}
-
-static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
-{
-       int ret = -1;
        struct tdb_lock_type *lck;
+       enum TDB_ERROR ecode;
 
        if (tdb->flags & TDB_NOLOCK)
-               return 0;
+               return TDB_SUCCESS;
 
        lck = find_nestlock(tdb, off);
        if ((lck == NULL) || (lck->count == 0)) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_nest_unlock: no lock for %llu\n", (long long)off);
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_nest_unlock: no lock for %zu",
+                                 (size_t)off);
        }
 
        if (lck->count > 1) {
                lck->count--;
-               return 0;
+               return TDB_SUCCESS;
        }
 
        /*
@@ -366,7 +383,7 @@ static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
         * element, we're about to overwrite it with the last array element
         * anyway.
         */
-       ret = tdb_brunlock(tdb, ltype, off, 1);
+       ecode = tdb_brunlock(tdb, ltype, off, 1);
 
        /*
         * Shrink the array by overwriting the element just unlocked with the
@@ -374,35 +391,32 @@ static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
         */
        *lck = tdb->lockrecs[--tdb->num_lockrecs];
 
-       return ret;
+       return ecode;
 }
 
-#if 0
 /*
   get the transaction lock
  */
-int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
-                        enum tdb_lock_flags lockflags)
+enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype)
 {
-       return tdb_nest_lock(tdb, TRANSACTION_LOCK, ltype, lockflags);
+       return tdb_nest_lock(tdb, TDB_TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
 }
 
 /*
   release the transaction lock
  */
-int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
+void tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
 {
-       return tdb_nest_unlock(tdb, TRANSACTION_LOCK, ltype, false);
+       tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
 }
-#endif
 
 /* We only need to lock individual bytes, but Linux merges consecutive locks
  * so we lock in contiguous ranges. */
-static int tdb_lock_gradual(struct tdb_context *tdb,
-                           int ltype, enum tdb_lock_flags flags,
-                           tdb_off_t off, tdb_off_t len)
+static enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
+                                      int ltype, enum tdb_lock_flags flags,
+                                      tdb_off_t off, tdb_off_t len)
 {
-       int ret;
+       enum TDB_ERROR ecode;
        enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
 
        if (len <= 1) {
@@ -413,92 +427,84 @@ static int tdb_lock_gradual(struct tdb_context *tdb,
        }
 
        /* First we try non-blocking. */
-       ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
-       if (ret == 0) {
-               return 0;
+       if (tdb_brlock(tdb, ltype, off, len, nb_flags) == TDB_SUCCESS) {
+               return TDB_SUCCESS;
        }
 
        /* Try locking first half, then second. */
-       ret = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
-       if (ret == -1)
-               return -1;
+       ecode = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
+       if (ecode != TDB_SUCCESS)
+               return ecode;
 
-       ret = tdb_lock_gradual(tdb, ltype, flags,
-                                   off + len / 2, len - len / 2);
-       if (ret == -1) {
+       ecode = tdb_lock_gradual(tdb, ltype, flags,
+                                off + len / 2, len - len / 2);
+       if (ecode != TDB_SUCCESS) {
                tdb_brunlock(tdb, ltype, off, len / 2);
-               return -1;
        }
-       return 0;
+       return ecode;
 }
 
 /* lock/unlock entire database.  It can only be upgradable if you have some
  * other way of guaranteeing exclusivity (ie. transaction write lock). */
-int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
-                      enum tdb_lock_flags flags, bool upgradable)
+enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
+                                 enum tdb_lock_flags flags, bool upgradable)
 {
-       /* FIXME: There are no locks on read-only dbs */
-       if (tdb->read_only) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_lock: read-only\n");
-               return -1;
-       }
+       enum TDB_ERROR ecode;
+       tdb_bool_err berr;
 
-       if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
+       if (tdb->allrecord_lock.count
+           && (ltype == F_RDLCK || tdb->allrecord_lock.ltype == F_WRLCK)) {
                tdb->allrecord_lock.count++;
-               return 0;
+               return TDB_SUCCESS;
        }
 
        if (tdb->allrecord_lock.count) {
                /* a global lock of a different type exists */
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_lock: already have %s lock\n",
-                        tdb->allrecord_lock.ltype == F_RDLCK
-                        ? "read" : "write");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+                                 "tdb_allrecord_lock: already have %s lock",
+                                 tdb->allrecord_lock.ltype == F_RDLCK
+                                 ? "read" : "write");
        }
 
-       if (tdb_has_locks(tdb)) {
+       if (tdb_has_hash_locks(tdb)) {
                /* can't combine global and chain locks */
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_lock: already have chain lock\n");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+                                 "tdb_allrecord_lock:"
+                                 " already have chain lock");
        }
 
        if (upgradable && ltype != F_RDLCK) {
                /* tdb error: you can't upgrade a write lock! */
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_lock: can't upgrade a write lock\n");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_allrecord_lock:"
+                                 " can't upgrade a write lock");
        }
 
+       add_stat(tdb, locks, 1);
 again:
        /* Lock hashes, gradually. */
-       if (tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
-                            TDB_HASH_LOCK_RANGE)) {
+       ecode = tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
+                                TDB_HASH_LOCK_RANGE);
+       if (ecode != TDB_SUCCESS) {
                if (!(flags & TDB_LOCK_PROBE)) {
-                       tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                                "tdb_allrecord_lock hashes failed (%s)\n",
-                                strerror(errno));
+                       tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                  "tdb_allrecord_lock hashes failed");
                }
-               return -1;
+               return ecode;
        }
 
-       /* Lock free lists: there to end of file. */
-       if (tdb_brlock(tdb, ltype, TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
-                      0, flags)) {
+       /* Lock free tables: there to end of file. */
+       ecode = tdb_brlock(tdb, ltype,
+                          TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
+                          0, flags);
+       if (ecode != TDB_SUCCESS) {
                if (!(flags & TDB_LOCK_PROBE)) {
-                       tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                                "tdb_allrecord_lock freelist failed (%s)\n",
-                                strerror(errno));
+                       tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                "tdb_allrecord_lock freetables failed");
                }
-               tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 
+               tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START,
                             TDB_HASH_LOCK_RANGE);
-               return -1;
+               return ecode;
        }
 
        tdb->allrecord_lock.count = 1;
@@ -508,20 +514,26 @@ again:
        tdb->allrecord_lock.off = upgradable;
 
        /* Now check for needing recovery. */
-       if (unlikely(tdb_needs_recovery(tdb))) {
-               tdb_allrecord_unlock(tdb, ltype);
-               if (tdb_lock_and_recover(tdb) == -1) {
-                       return -1;
-               }               
-               goto again;
-       }
+       if (flags & TDB_LOCK_NOCHECK)
+               return TDB_SUCCESS;
 
-       return 0;
+       berr = tdb_needs_recovery(tdb);
+       if (likely(berr == false))
+               return TDB_SUCCESS;
+
+       tdb_allrecord_unlock(tdb, ltype);
+       if (berr < 0)
+               return berr;
+       ecode = tdb_lock_and_recover(tdb);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
+       goto again;
 }
 
-int tdb_lock_open(struct tdb_context *tdb)
+enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
 {
-       return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT);
+       return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
 }
 
 void tdb_unlock_open(struct tdb_context *tdb)
@@ -529,9 +541,16 @@ void tdb_unlock_open(struct tdb_context *tdb)
        tdb_nest_unlock(tdb, TDB_OPEN_LOCK, F_WRLCK);
 }
 
-int tdb_lock_expand(struct tdb_context *tdb, int ltype)
+bool tdb_has_open_lock(struct tdb_context *tdb)
 {
-       return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype, TDB_LOCK_WAIT);
+       return find_nestlock(tdb, TDB_OPEN_LOCK) != NULL;
+}
+
+enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype)
+{
+       /* Lock doesn't protect data, so don't check (we recurse if we do!) */
+       return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype,
+                            TDB_LOCK_WAIT | TDB_LOCK_NOCHECK);
 }
 
 void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
@@ -540,43 +559,33 @@ void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
 }
 
 /* unlock entire db */
-int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
+void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
 {
-       /* FIXME: There are no locks on read-only dbs */
-       if (tdb->read_only) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_unlock: read-only\n");
-               return -1;
-       }
-
        if (tdb->allrecord_lock.count == 0) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_unlock: not locked!\n");
-               return -1;
+               tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+                          "tdb_allrecord_unlock: not locked!");
+               return;
        }
 
        /* Upgradable locks are marked as write locks. */
        if (tdb->allrecord_lock.ltype != ltype
            && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_allrecord_unlock: have %s lock\n",
-                        tdb->allrecord_lock.ltype == F_RDLCK
-                        ? "read" : "write");
-               return -1;
+               tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                          "tdb_allrecord_unlock: have %s lock",
+                          tdb->allrecord_lock.ltype == F_RDLCK
+                          ? "read" : "write");
+               return;
        }
 
        if (tdb->allrecord_lock.count > 1) {
                tdb->allrecord_lock.count--;
-               return 0;
+               return;
        }
 
        tdb->allrecord_lock.count = 0;
        tdb->allrecord_lock.ltype = 0;
 
-       return tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
+       tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
 }
 
 bool tdb_has_expansion_lock(struct tdb_context *tdb)
@@ -584,56 +593,18 @@ bool tdb_has_expansion_lock(struct tdb_context *tdb)
        return find_nestlock(tdb, TDB_EXPANSION_LOCK) != NULL;
 }
 
-bool tdb_has_locks(struct tdb_context *tdb)
+bool tdb_has_hash_locks(struct tdb_context *tdb)
 {
-       return tdb->allrecord_lock.count || tdb->num_lockrecs;
-}
-
-#if 0
-/* lock entire database with write lock */
-int tdb_lockall(struct tdb_context *tdb)
-{
-       tdb_trace(tdb, "tdb_lockall");
-       return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
-}
-
-/* lock entire database with write lock - nonblocking varient */
-int tdb_lockall_nonblock(struct tdb_context *tdb)
-{
-       int ret = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_NOWAIT, false);
-       tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
-       return ret;
-}
-
-/* unlock entire database with write lock */
-int tdb_unlockall(struct tdb_context *tdb)
-{
-       tdb_trace(tdb, "tdb_unlockall");
-       return tdb_allrecord_unlock(tdb, F_WRLCK);
-}
-
-/* lock entire database with read lock */
-int tdb_lockall_read(struct tdb_context *tdb)
-{
-       tdb_trace(tdb, "tdb_lockall_read");
-       return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
-}
-
-/* lock entire database with read lock - nonblock varient */
-int tdb_lockall_read_nonblock(struct tdb_context *tdb)
-{
-       int ret = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_NOWAIT, false);
-       tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
-       return ret;
-}
+       unsigned int i;
 
-/* unlock entire database with read lock */
-int tdb_unlockall_read(struct tdb_context *tdb)
-{
-       tdb_trace(tdb, "tdb_unlockall_read");
-       return tdb_allrecord_unlock(tdb, F_RDLCK);
+       for (i=0; i<tdb->num_lockrecs; i++) {
+               if (tdb->lockrecs[i].off >= TDB_HASH_LOCK_START
+                   && tdb->lockrecs[i].off < (TDB_HASH_LOCK_START
+                                              + TDB_HASH_LOCK_RANGE))
+                       return true;
+       }
+       return false;
 }
-#endif
 
 static bool tdb_has_free_lock(struct tdb_context *tdb)
 {
@@ -647,10 +618,10 @@ static bool tdb_has_free_lock(struct tdb_context *tdb)
        return false;
 }
 
-int tdb_lock_hashes(struct tdb_context *tdb,
-                   tdb_off_t hash_lock,
-                   tdb_len_t hash_range,
-                   int ltype, enum tdb_lock_flags waitflag)
+enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
+                              tdb_off_t hash_lock,
+                              tdb_len_t hash_range,
+                              int ltype, enum tdb_lock_flags waitflag)
 {
        /* FIXME: Do this properly, using hlock_range */
        unsigned lock = TDB_HASH_LOCK_START
@@ -659,38 +630,34 @@ int tdb_lock_hashes(struct tdb_context *tdb,
        /* a allrecord lock allows us to avoid per chain locks */
        if (tdb->allrecord_lock.count &&
            (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
-               return 0;
+               return TDB_SUCCESS;
        }
 
        if (tdb->allrecord_lock.count) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_lock_hashes: have %s allrecordlock\n",
-                        tdb->allrecord_lock.ltype == F_RDLCK
-                        ? "read" : "write");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
+                                 "tdb_lock_hashes:"
+                                 " already have %s allrecordlock",
+                                 tdb->allrecord_lock.ltype == F_RDLCK
+                                 ? "read" : "write");
        }
 
        if (tdb_has_free_lock(tdb)) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_lock_hashes: have free lock already\n");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_lock_hashes: already have free lock");
        }
 
        if (tdb_has_expansion_lock(tdb)) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_lock_hashes: have expansion lock already\n");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_lock_hashes:"
+                                 " already have expansion lock");
        }
 
        return tdb_nest_lock(tdb, lock, ltype, waitflag);
 }
 
-int tdb_unlock_hashes(struct tdb_context *tdb,
-                     tdb_off_t hash_lock,
-                     tdb_len_t hash_range, int ltype)
+enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
+                                tdb_off_t hash_lock,
+                                tdb_len_t hash_range, int ltype)
 {
        unsigned lock = TDB_HASH_LOCK_START
                + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
@@ -699,12 +666,10 @@ int tdb_unlock_hashes(struct tdb_context *tdb,
        if (tdb->allrecord_lock.count) {
                if (tdb->allrecord_lock.ltype == F_RDLCK
                    && ltype == F_WRLCK) {
-                       tdb->ecode = TDB_ERR_LOCK;
-                       tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-                                "tdb_unlock_hashes RO allrecord!\n");
-                       return -1;
+                       return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                         "tdb_unlock_hashes RO allrecord!");
                }
-               return 0;
+               return TDB_SUCCESS;
        }
 
        return tdb_nest_unlock(tdb, lock, ltype);
@@ -721,8 +686,8 @@ static tdb_off_t free_lock_off(tdb_off_t b_off)
                + b_off / sizeof(tdb_off_t);
 }
 
-int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
-                        enum tdb_lock_flags waitflag)
+enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
+                                   enum tdb_lock_flags waitflag)
 {
        assert(b_off >= sizeof(struct tdb_header));
 
@@ -730,18 +695,16 @@ int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
        if (tdb->allrecord_lock.count) {
                if (tdb->allrecord_lock.ltype == F_WRLCK)
                        return 0;
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-                        "tdb_lock_free_bucket with RO allrecordlock!\n");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_lock_free_bucket with"
+                                 " read-only allrecordlock!");
        }
 
 #if 0 /* FIXME */
        if (tdb_has_expansion_lock(tdb)) {
-               tdb->ecode = TDB_ERR_LOCK;
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_lock_free_bucket: have expansion lock already\n");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                 "tdb_lock_free_bucket:"
+                                 " already have expansion lock");
        }
 #endif
 
@@ -756,145 +719,6 @@ void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
        tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
 }
 
-/* Even if the entry isn't in this hash bucket, you'd have to lock this
- * bucket to find it. */
-static int chainlock(struct tdb_context *tdb, const TDB_DATA *key,
-                    int ltype, enum tdb_lock_flags waitflag,
-                    const char *func)
-{
-       int ret;
-       uint64_t h = tdb_hash(tdb, key->dptr, key->dsize);
-
-       ret = tdb_lock_hashes(tdb, h, 1, ltype, waitflag);
-       tdb_trace_1rec(tdb, func, *key);
-       return ret;
-}
-
-/* lock/unlock one hash chain. This is meant to be used to reduce
-   contention - it cannot guarantee how many records will be locked */
-int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
-{
-       return chainlock(tdb, &key, F_WRLCK, TDB_LOCK_WAIT, "tdb_chainlock");
-}
-
-int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
-{
-       uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
-       tdb_trace_1rec(tdb, "tdb_chainunlock", key);
-       return tdb_unlock_hashes(tdb, h, 1, F_WRLCK);
-}
-
-#if 0
-/* lock/unlock one hash chain, non-blocking. This is meant to be used
-   to reduce contention - it cannot guarantee how many records will be
-   locked */
-int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
-{
-       return chainlock(tdb, &key, F_WRLCK, TDB_LOCK_NOWAIT,
-                        "tdb_chainlock_nonblock");
-}
-
-int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
-{
-       return chainlock(tdb, &key, F_RDLCK, TDB_LOCK_WAIT,
-                        "tdb_chainlock_read");
-}
-
-int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
-{
-       uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
-       tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
-       return tdb_unlock_list(tdb, h & ((1ULL << tdb->header.v.hash_bits)-1),
-                              F_RDLCK);
-}
-
-/* record lock stops delete underneath */
-int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-       if (tdb->allrecord_lock.count) {
-               return 0;
-       }
-       return off ? tdb_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
-}
-
-/*
-  Write locks override our own fcntl readlocks, so check it here.
-  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
-  an error to fail to get the lock here.
-*/
-int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-       struct tdb_traverse_lock *i;
-       for (i = &tdb->travlocks; i; i = i->next)
-               if (i->off == off)
-                       return -1;
-       if (tdb->allrecord_lock.count) {
-               if (tdb->allrecord_lock.ltype == F_WRLCK) {
-                       return 0;
-               }
-               return -1;
-       }
-       return tdb_brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
-}
-
-int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-       if (tdb->allrecord_lock.count) {
-               return 0;
-       }
-       return tdb_brunlock(tdb, F_WRLCK, off, 1);
-}
-
-/* fcntl locks don't stack: avoid unlocking someone else's */
-int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
-{
-       struct tdb_traverse_lock *i;
-       uint32_t count = 0;
-
-       if (tdb->allrecord_lock.count) {
-               return 0;
-       }
-
-       if (off == 0)
-               return 0;
-       for (i = &tdb->travlocks; i; i = i->next)
-               if (i->off == off)
-                       count++;
-       return (count == 1 ? tdb_brunlock(tdb, F_RDLCK, off, 1) : 0);
-}
-
-/* The transaction code uses this to remove all locks. */
-void tdb_release_transaction_locks(struct tdb_context *tdb)
-{
-       unsigned int i;
-
-       if (tdb->allrecord_lock.count != 0) {
-               tdb_off_t hash_size, free_size;
-
-               hash_size = (1ULL << tdb->header.v.hash_bits)
-                       * sizeof(tdb_off_t);
-               free_size = tdb->header.v.free_zones 
-                       * (tdb->header.v.free_buckets + 1) * sizeof(tdb_off_t);
-
-               tdb_brunlock(tdb, tdb->allrecord_lock.ltype,
-                            tdb->header.v.hash_off, hash_size);
-               tdb_brunlock(tdb, tdb->allrecord_lock.ltype,
-                            tdb->header.v.free_off, free_size);
-               tdb->allrecord_lock.count = 0;
-               tdb->allrecord_lock.ltype = 0;
-       }
-
-       for (i = 0; i<tdb->num_lockrecs; i++) {
-               struct tdb_lock_type *lck = &tdb->lockrecs[i];
-
-               tdb_brunlock(tdb, lck->ltype, lck->off, 1);
-       }
-       tdb->num_lockrecs = 0;
-       SAFE_FREE(tdb->lockrecs);
-       tdb->header_uptodate = false;
-}
-#endif
-
 void tdb_lock_init(struct tdb_context *tdb)
 {
        tdb->num_lockrecs = 0;