tdb2: unify tdb1_store into tdb_store
[ccan] / ccan / tdb2 / tdb.c
index 61791c9d5b5ca8af9ef22a1f88e07ce8f8bbafc7..a7aa4572382b0f8b10ef03c4adc08599fd7e7293 100644 (file)
+ /*
+   Trivial Database 2: fetch, store and misc routines.
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
 #include "private.h"
-#include <ccan/tdb2/tdb2.h>
-#include <ccan/hash/hash.h>
-#include <ccan/build_assert/build_assert.h>
-#include <ccan/likely/likely.h>
-#include <assert.h>
-
-/* The null return. */
-struct tdb_data tdb_null = { .dptr = NULL, .dsize = 0 };
-
-/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
-static struct tdb_context *tdbs = NULL;
-
-PRINTF_ATTRIBUTE(4, 5) static void
-null_log_fn(struct tdb_context *tdb,
-           enum tdb_debug_level level, void *priv,
-           const char *fmt, ...)
+#include <ccan/asprintf/asprintf.h>
+#include <stdarg.h>
+
+static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
+                                    tdb_off_t off,
+                                    tdb_len_t keylen,
+                                    tdb_len_t datalen,
+                                    struct tdb_used_record *rec,
+                                    uint64_t h)
 {
+       uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
+       enum TDB_ERROR ecode;
+
+       ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
+                          keylen + dataroom, h);
+       if (ecode == TDB_SUCCESS) {
+               ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
+       }
+       return ecode;
 }
 
-/* We do a lot of work assuming our copy of the header volatile area
- * is uptodate, and usually it is.  However, once we grab a lock, we have to
- * re-check it. */
-bool header_changed(struct tdb_context *tdb)
+static enum TDB_ERROR replace_data(struct tdb_context *tdb,
+                                  struct hash_info *h,
+                                  struct tdb_data key, struct tdb_data dbuf,
+                                  tdb_off_t old_off, tdb_len_t old_room,
+                                  bool growing)
 {
-       uint64_t gen;
+       tdb_off_t new_off;
+       enum TDB_ERROR ecode;
 
-       if (!(tdb->flags & TDB_NOLOCK) && tdb->header_uptodate) {
-               tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
-                        "warning: header uptodate already\n");
+       /* Allocate a new record. */
+       new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
+                       growing);
+       if (TDB_OFF_IS_ERR(new_off)) {
+               return new_off;
        }
 
-       /* We could get a partial update if we're not holding any locks. */
-       assert((tdb->flags & TDB_NOLOCK) || tdb_has_locks(tdb));
+       /* We didn't like the existing one: remove it. */
+       if (old_off) {
+               tdb->stats.frees++;
+               ecode = add_free_record(tdb, old_off,
+                                       sizeof(struct tdb_used_record)
+                                       + key.dsize + old_room,
+                                       TDB_LOCK_WAIT, true);
+               if (ecode == TDB_SUCCESS)
+                       ecode = replace_in_hash(tdb, h, new_off);
+       } else {
+               ecode = add_to_hash(tdb, h, new_off);
+       }
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
 
-       tdb->header_uptodate = true;
-       gen = tdb_read_off(tdb, offsetof(struct tdb_header, v.generation));
-       if (unlikely(gen != tdb->header.v.generation)) {
-               tdb_read_convert(tdb, offsetof(struct tdb_header, v),
-                                &tdb->header.v, sizeof(tdb->header.v));
-               return true;
+       new_off += sizeof(struct tdb_used_record);
+       ecode = tdb->tdb2.io->twrite(tdb, new_off, key.dptr, key.dsize);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
        }
-       return false;
-}
 
-int write_header(struct tdb_context *tdb)
-{
-       assert(tdb_read_off(tdb, offsetof(struct tdb_header, v.generation))
-              == tdb->header.v.generation);
-       tdb->header.v.generation++;
-       return tdb_write_convert(tdb, offsetof(struct tdb_header, v),
-                                &tdb->header.v, sizeof(tdb->header.v));
-}
+       new_off += key.dsize;
+       ecode = tdb->tdb2.io->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
 
-static uint64_t jenkins_hash(const void *key, size_t length, uint64_t seed,
-                            void *arg)
-{
-       return hash64_stable((const unsigned char *)key, length, seed);
-}
+       if (tdb->flags & TDB_SEQNUM)
+               tdb_inc_seqnum(tdb);
 
-uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len)
-{
-       return tdb->khash(ptr, len, tdb->header.hash_seed, tdb->hash_priv);
+       return TDB_SUCCESS;
 }
 
-static bool tdb_already_open(dev_t device, ino_t ino)
+static enum TDB_ERROR update_data(struct tdb_context *tdb,
+                                 tdb_off_t off,
+                                 struct tdb_data dbuf,
+                                 tdb_len_t extra)
 {
-       struct tdb_context *i;
-       
-       for (i = tdbs; i; i = i->next) {
-               if (i->device == device && i->inode == ino) {
-                       return true;
-               }
-       }
+       enum TDB_ERROR ecode;
 
-       return false;
-}
-
-static uint64_t random_number(struct tdb_context *tdb)
-{
-       int fd;
-       uint64_t ret = 0;
-       struct timeval now;
-
-       fd = open("/dev/urandom", O_RDONLY);
-       if (fd >= 0) {
-               if (tdb_read_all(fd, &ret, sizeof(ret))) {
-                       tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
-                                "tdb_open: random from /dev/urandom\n");
-                       close(fd);
-                       return ret;
-               }
-               close(fd);
-       }
-       /* FIXME: Untested!  Based on Wikipedia protocol description! */
-       fd = open("/dev/egd-pool", O_RDWR);
-       if (fd >= 0) {
-               /* Command is 1, next byte is size we want to read. */
-               char cmd[2] = { 1, sizeof(uint64_t) };
-               if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
-                       char reply[1 + sizeof(uint64_t)];
-                       int r = read(fd, reply, sizeof(reply));
-                       if (r > 1) {
-                               tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
-                                        "tdb_open: %u random bytes from"
-                                        " /dev/egd-pool\n", r-1);
-                               /* Copy at least some bytes. */
-                               memcpy(&ret, reply+1, r - 1);
-                               if (reply[0] == sizeof(uint64_t)
-                                   && r == sizeof(reply)) {
-                                       close(fd);
-                                       return ret;
-                               }
-                       }
-               }
-               close(fd);
+       ecode = tdb->tdb2.io->twrite(tdb, off, dbuf.dptr, dbuf.dsize);
+       if (ecode == TDB_SUCCESS && extra) {
+               /* Put a zero in; future versions may append other data. */
+               ecode = tdb->tdb2.io->twrite(tdb, off + dbuf.dsize, "", 1);
        }
+       if (tdb->flags & TDB_SEQNUM)
+               tdb_inc_seqnum(tdb);
 
-       /* Fallback: pid and time. */
-       gettimeofday(&now, NULL);
-       ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
-       tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
-                "tdb_open: random from getpid and time\n");
-       return ret;
+       return ecode;
 }
 
-struct new_database {
-       struct tdb_header hdr;
-       struct tdb_used_record hrec;
-       tdb_off_t hash[1ULL << INITIAL_HASH_BITS];
-       struct tdb_used_record frec;
-       tdb_off_t free[INITIAL_FREE_BUCKETS + 1]; /* One overflow bucket */
-};
-
-/* initialise a new database */
-static int tdb_new_database(struct tdb_context *tdb)
+enum TDB_ERROR tdb_store(struct tdb_context *tdb,
+                        struct tdb_data key, struct tdb_data dbuf, int flag)
 {
-       /* We make it up in memory, then write it out if not internal */
-       struct new_database newdb;
-       unsigned int magic_off = offsetof(struct tdb_header, magic_food);
-
-       /* Fill in the header */
-       newdb.hdr.version = TDB_VERSION;
-       newdb.hdr.hash_seed = random_number(tdb);
-       newdb.hdr.hash_test = TDB_HASH_MAGIC;
-       newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
-                                        sizeof(newdb.hdr.hash_test),
-                                        newdb.hdr.hash_seed,
-                                        tdb->hash_priv);
-
-       newdb.hdr.v.generation = 0;
-
-       /* The initial zone must cover the initial database size! */
-       BUILD_ASSERT((1ULL << INITIAL_ZONE_BITS) >= sizeof(newdb));
-
-       /* Free array has 1 zone, 10 buckets.  All buckets empty. */
-       newdb.hdr.v.num_zones = 1;
-       newdb.hdr.v.zone_bits = INITIAL_ZONE_BITS;
-       newdb.hdr.v.free_buckets = INITIAL_FREE_BUCKETS;
-       newdb.hdr.v.free_off = offsetof(struct new_database, free);
-       set_header(tdb, &newdb.frec, 0,
-                  sizeof(newdb.free), sizeof(newdb.free), 0);
-       memset(newdb.free, 0, sizeof(newdb.free));
-
-       /* Initial hashes are empty. */
-       newdb.hdr.v.hash_bits = INITIAL_HASH_BITS;
-       newdb.hdr.v.hash_off = offsetof(struct new_database, hash);
-       set_header(tdb, &newdb.hrec, 0,
-                  sizeof(newdb.hash), sizeof(newdb.hash), 0);
-       memset(newdb.hash, 0, sizeof(newdb.hash));
-
-       /* Magic food */
-       memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
-       strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
-
-       /* This creates an endian-converted database, as if read from disk */
-       tdb_convert(tdb,
-                   (char *)&newdb.hdr + magic_off,
-                   sizeof(newdb) - magic_off);
-
-       tdb->header = newdb.hdr;
+       struct hash_info h;
+       tdb_off_t off;
+       tdb_len_t old_room = 0;
+       struct tdb_used_record rec;
+       enum TDB_ERROR ecode;
 
-       if (tdb->flags & TDB_INTERNAL) {
-               tdb->map_size = sizeof(newdb);
-               tdb->map_ptr = malloc(tdb->map_size);
-               if (!tdb->map_ptr) {
-                       tdb->ecode = TDB_ERR_OOM;
-                       return -1;
-               }
-               memcpy(tdb->map_ptr, &newdb, tdb->map_size);
-               return 0;
+       if (tdb->flags & TDB_VERSION1) {
+               if (tdb1_store(tdb, key, dbuf, flag) == -1)
+                       return tdb->last_error;
+               return TDB_SUCCESS;
        }
-       if (lseek(tdb->fd, 0, SEEK_SET) == -1)
-               return -1;
 
-       if (ftruncate(tdb->fd, 0) == -1)
-               return -1;
+       off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
+       if (TDB_OFF_IS_ERR(off)) {
+               return tdb->last_error = off;
+       }
 
-       if (!tdb_pwrite_all(tdb->fd, &newdb, sizeof(newdb), 0)) {
-               tdb->ecode = TDB_ERR_IO;
-               return -1;
+       /* Now we have lock on this hash bucket. */
+       if (flag == TDB_INSERT) {
+               if (off) {
+                       ecode = TDB_ERR_EXISTS;
+                       goto out;
+               }
+       } else {
+               if (off) {
+                       old_room = rec_data_length(&rec)
+                               + rec_extra_padding(&rec);
+                       if (old_room >= dbuf.dsize) {
+                               /* Can modify in-place.  Easy! */
+                               ecode = update_rec_hdr(tdb, off,
+                                                      key.dsize, dbuf.dsize,
+                                                      &rec, h.h);
+                               if (ecode != TDB_SUCCESS) {
+                                       goto out;
+                               }
+                               ecode = update_data(tdb,
+                                                   off + sizeof(rec)
+                                                   + key.dsize, dbuf,
+                                                   old_room - dbuf.dsize);
+                               if (ecode != TDB_SUCCESS) {
+                                       goto out;
+                               }
+                               tdb_unlock_hashes(tdb, h.hlock_start,
+                                                 h.hlock_range, F_WRLCK);
+                               return tdb->last_error = TDB_SUCCESS;
+                       }
+               } else {
+                       if (flag == TDB_MODIFY) {
+                               /* if the record doesn't exist and we
+                                  are in TDB_MODIFY mode then we should fail
+                                  the store */
+                               ecode = TDB_ERR_NOEXIST;
+                               goto out;
+                       }
+               }
        }
-       return 0;
+
+       /* If we didn't use the old record, this implies we're growing. */
+       ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
+out:
+       tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
+       return tdb->last_error = ecode;
 }
 
-struct tdb_context *tdb_open(const char *name, int tdb_flags,
-                            int open_flags, mode_t mode,
-                            union tdb_attribute *attr)
+enum TDB_ERROR tdb_append(struct tdb_context *tdb,
+                         struct tdb_data key, struct tdb_data dbuf)
 {
-       struct tdb_context *tdb;
-       struct stat st;
-       int save_errno;
-       uint64_t hash_test;
-       unsigned v;
-
-       tdb = malloc(sizeof(*tdb));
-       if (!tdb) {
-               /* Can't log this */
-               errno = ENOMEM;
-               goto fail;
-       }
-       tdb->name = NULL;
-       tdb->map_ptr = NULL;
-       tdb->fd = -1;
-       /* map_size will be set below. */
-       tdb->ecode = TDB_SUCCESS;
-       /* header will be read in below. */
-       tdb->header_uptodate = false;
-       tdb->flags = tdb_flags;
-       tdb->log = null_log_fn;
-       tdb->log_priv = NULL;
-       tdb->khash = jenkins_hash;
-       tdb->hash_priv = NULL;
-       tdb->transaction = NULL;
-       /* last_zone will be set below. */
-       tdb_io_init(tdb);
-       tdb_lock_init(tdb);
-
-       /* FIXME */
-       if (attr) {
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_open: attributes not yet supported\n");
-               errno = EINVAL;
-               goto fail;
-       }
+       struct hash_info h;
+       tdb_off_t off;
+       struct tdb_used_record rec;
+       tdb_len_t old_room = 0, old_dlen;
+       unsigned char *newdata;
+       struct tdb_data new_dbuf;
+       enum TDB_ERROR ecode;
 
-       if ((open_flags & O_ACCMODE) == O_WRONLY) {
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_open: can't open tdb %s write-only\n", name);
-               errno = EINVAL;
-               goto fail;
+       off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
+       if (TDB_OFF_IS_ERR(off)) {
+               return tdb->last_error = off;
        }
 
-       if ((open_flags & O_ACCMODE) == O_RDONLY) {
-               tdb->read_only = true;
-               /* read only databases don't do locking */
-               tdb->flags |= TDB_NOLOCK;
-       } else
-               tdb->read_only = false;
+       if (off) {
+               old_dlen = rec_data_length(&rec);
+               old_room = old_dlen + rec_extra_padding(&rec);
+
+               /* Fast path: can append in place. */
+               if (rec_extra_padding(&rec) >= dbuf.dsize) {
+                       ecode = update_rec_hdr(tdb, off, key.dsize,
+                                              old_dlen + dbuf.dsize, &rec,
+                                              h.h);
+                       if (ecode != TDB_SUCCESS) {
+                               goto out;
+                       }
 
-       /* internal databases don't need any of the rest. */
-       if (tdb->flags & TDB_INTERNAL) {
-               tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
-               if (tdb_new_database(tdb) != 0) {
-                       tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                                "tdb_open: tdb_new_database failed!");
-                       goto fail;
+                       off += sizeof(rec) + key.dsize + old_dlen;
+                       ecode = update_data(tdb, off, dbuf,
+                                           rec_extra_padding(&rec));
+                       goto out;
                }
-               TEST_IT(tdb->flags & TDB_CONVERT);
-               tdb_convert(tdb, &tdb->header, sizeof(tdb->header));
-               return tdb;
-       }
 
-       if ((tdb->fd = open(name, open_flags, mode)) == -1) {
-               tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
-                        "tdb_open: could not open file %s: %s\n",
-                        name, strerror(errno));
-               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_lock_open(tdb) == -1) {
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_open: failed to get open lock on %s: %s\n",
-                        name, strerror(errno));
-               goto fail;      /* errno set by tdb_brlock */
-       }
-
-       if (!tdb_pread_all(tdb->fd, &tdb->header, sizeof(tdb->header), 0)
-           || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
-               if (!(open_flags & O_CREAT) || tdb_new_database(tdb) == -1) {
-                       if (errno == 0) {
-                               errno = EIO; /* ie bad format or something */
-                       }
-                       goto fail;
+               /* Slow path. */
+               newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
+               if (!newdata) {
+                       ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+                                          "tdb_append:"
+                                          " failed to allocate %zu bytes",
+                                          (size_t)(key.dsize + old_dlen
+                                                   + dbuf.dsize));
+                       goto out;
                }
-       } else if (tdb->header.version != TDB_VERSION) {
-               if (tdb->header.version == bswap_64(TDB_VERSION))
-                       tdb->flags |= TDB_CONVERT;
-               else {
-                       /* wrong version */
-                       tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                                "tdb_open: %s is unknown version 0x%llx\n",
-                                name, (long long)tdb->header.version);
-                       errno = EIO;
-                       goto fail;
+               ecode = tdb->tdb2.io->tread(tdb, off + sizeof(rec) + key.dsize,
+                                           newdata, old_dlen);
+               if (ecode != TDB_SUCCESS) {
+                       goto out_free_newdata;
                }
+               memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
+               new_dbuf.dptr = newdata;
+               new_dbuf.dsize = old_dlen + dbuf.dsize;
+       } else {
+               newdata = NULL;
+               new_dbuf = dbuf;
        }
 
-       tdb_convert(tdb, &tdb->header, sizeof(tdb->header));
-       hash_test = TDB_HASH_MAGIC;
-       hash_test = tdb->khash(&hash_test, sizeof(hash_test),
-                              tdb->header.hash_seed, tdb->hash_priv);
-       if (tdb->header.hash_test != hash_test) {
-               /* wrong hash variant */
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_open: %s uses a different hash function\n",
-                        name);
-               errno = EIO;
-               goto fail;
-       }
+       /* If they're using tdb_append(), it implies they're growing record. */
+       ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 
-       if (fstat(tdb->fd, &st) == -1)
-               goto fail;
-
-       /* Is it already in the open list?  If so, fail. */
-       if (tdb_already_open(st.st_dev, st.st_ino)) {
-               /* FIXME */
-               tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                        "tdb_open: %s (%d,%d) is already open in this process\n",
-                        name, (int)st.st_dev, (int)st.st_ino);
-               errno = EBUSY;
-               goto fail;
-       }
+out_free_newdata:
+       free(newdata);
+out:
+       tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
+       return tdb->last_error = ecode;
+}
+
+enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
+                        struct tdb_data *data)
+{
+       tdb_off_t off;
+       struct tdb_used_record rec;
+       struct hash_info h;
+       enum TDB_ERROR ecode;
 
-       tdb->name = strdup(name);
-       if (!tdb->name) {
-               errno = ENOMEM;
-               goto fail;
+       off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
+       if (TDB_OFF_IS_ERR(off)) {
+               return tdb->last_error = off;
        }
 
-       tdb->map_size = st.st_size;
-       tdb->device = st.st_dev;
-       tdb->inode = st.st_ino;
-       tdb_mmap(tdb);
-       tdb_unlock_open(tdb);
-       tdb_zone_init(tdb);
-
-       tdb->next = tdbs;
-       tdbs = tdb;
-       return tdb;
-
- fail:
-       save_errno = errno;
-
-       if (!tdb)
-               return NULL;
-
-#ifdef TDB_TRACE
-       close(tdb->tracefd);
-#endif
-       if (tdb->map_ptr) {
-               if (tdb->flags & TDB_INTERNAL) {
-                       free(tdb->map_ptr);
+       if (!off) {
+               ecode = TDB_ERR_NOEXIST;
+       } else {
+               data->dsize = rec_data_length(&rec);
+               data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
+                                           data->dsize);
+               if (TDB_PTR_IS_ERR(data->dptr)) {
+                       ecode = TDB_PTR_ERR(data->dptr);
                } else
-                       tdb_munmap(tdb);
+                       ecode = TDB_SUCCESS;
        }
-       free((char *)tdb->name);
-       if (tdb->fd != -1)
-               if (close(tdb->fd) != 0)
-                       tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-                                "tdb_open: failed to close tdb->fd"
-                                " on error!\n");
-       free(tdb);
-       errno = save_errno;
-       return NULL;
-}
 
-static tdb_off_t hash_off(struct tdb_context *tdb, uint64_t list)
-{
-       return tdb->header.v.hash_off
-               + ((list & ((1ULL << tdb->header.v.hash_bits) - 1))
-                  * sizeof(tdb_off_t));
+       tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
+       return tdb->last_error = ecode;
 }
 
-/* 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)
+bool tdb_exists(struct tdb_context *tdb, TDB_DATA key)
 {
        tdb_off_t off;
-       uint64_t keylen;
-       const unsigned char *rkey;
-
-       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
-
-       if (tdb_read_convert(tdb, off, rec, sizeof(*rec)) == -1)
-               return TDB_OFF_ERR;
-
-       /* FIXME: check extra bits in header! */
-       keylen = rec_key_length(rec);
-       if (keylen != key->dsize)
-               return TDB_OFF_ERR;
-
-       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;
-}
+       struct tdb_used_record rec;
+       struct hash_info h;
 
-/* FIXME: Optimize? */
-static void unlock_range(struct tdb_context *tdb,
-                        tdb_off_t list, tdb_len_t num,
-                        int ltype)
-{
-       tdb_off_t i;
+       off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
+       if (TDB_OFF_IS_ERR(off)) {
+               tdb->last_error = off;
+               return false;
+       }
+       tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 
-       for (i = list; i < list + num; i++)
-               tdb_unlock_list(tdb, i, ltype);
+       tdb->last_error = TDB_SUCCESS;
+       return off ? true : false;
 }
 
-/* FIXME: Optimize? */
-static int lock_range(struct tdb_context *tdb,
-                     tdb_off_t list, tdb_len_t num,
-                     int ltype)
+enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 {
-       tdb_off_t i;
+       tdb_off_t off;
+       struct tdb_used_record rec;
+       struct hash_info h;
+       enum TDB_ERROR ecode;
 
-       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;
-               }
+       off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
+       if (TDB_OFF_IS_ERR(off)) {
+               return tdb->last_error = off;
        }
-       return 0;
-}
 
-/* 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;
+       if (!off) {
+               ecode = TDB_ERR_NOEXIST;
+               goto unlock;
        }
 
-       /* 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;
+       ecode = delete_from_hash(tdb, &h);
+       if (ecode != TDB_SUCCESS) {
+               goto unlock;
        }
 
-       return pre_locks + len;
+       /* Free the deleted entry. */
+       tdb->stats.frees++;
+       ecode = add_free_record(tdb, off,
+                               sizeof(struct tdb_used_record)
+                               + rec_key_length(&rec)
+                               + rec_data_length(&rec)
+                               + rec_extra_padding(&rec),
+                               TDB_LOCK_WAIT, true);
+
+       if (tdb->flags & TDB_SEQNUM)
+               tdb_inc_seqnum(tdb);
+
+unlock:
+       tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
+       return tdb->last_error = ecode;
 }
 
-/* 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,
-                         struct tdb_used_record *rec,
-                         uint64_t h)
+unsigned int tdb_get_flags(struct tdb_context *tdb)
 {
-       uint64_t room = rec_data_length(rec) + rec_extra_padding(rec);
-
-       if (set_header(tdb, rec, keylen, datalen, room - datalen, h))
-               return -1;
+       return tdb->flags;
+}
 
-       return tdb_write_convert(tdb, off, rec, sizeof(*rec));
+static bool inside_transaction(const struct tdb_context *tdb)
+{
+       if (tdb->flags & TDB_VERSION1)
+               return tdb->tdb1.transaction != NULL;
+       else
+               return tdb->tdb2.transaction != NULL;
 }
 
-/* If we fail, others will try after us. */
-static void enlarge_hash(struct tdb_context *tdb)
+static bool readonly_changable(struct tdb_context *tdb, const char *caller)
 {
-       tdb_off_t newoff, i;
-       uint64_t h, num = 1ULL << tdb->header.v.hash_bits;
-       struct tdb_used_record pad, *r;
+       if (inside_transaction(tdb)) {
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                            TDB_LOG_USE_ERROR,
+                                            "%s: can't change"
+                                            " TDB_RDONLY inside transaction",
+                                            caller);
+               return false;
+       }
+
+       if (tdb->file->allrecord_lock.count != 0
+           || tdb->file->num_lockrecs != 0) {
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                            TDB_LOG_USE_ERROR,
+                                            "%s: can't change"
+                                            " TDB_RDONLY holding locks",
+                                            caller);
+               return false;
+       }
+       return true;
+}
 
-       /* FIXME: We should do this without holding locks throughout. */
-       if (tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false) == -1)
+void tdb_add_flag(struct tdb_context *tdb, unsigned flag)
+{
+       if (tdb->flags & TDB_INTERNAL) {
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                            TDB_LOG_USE_ERROR,
+                                            "tdb_add_flag: internal db");
                return;
-
-       /* Someone else enlarged for us?  Nothing to do. */
-       if ((1ULL << tdb->header.v.hash_bits) != num)
-               goto unlock;
-
-       newoff = alloc(tdb, 0, num * 2, 0, false);
-       if (unlikely(newoff == TDB_OFF_ERR))
-               goto unlock;
-       if (unlikely(newoff == 0)) {
-               if (tdb_expand(tdb, 0, num * 2, false) == -1)
-                       goto unlock;
-               newoff = alloc(tdb, 0, num * 2, 0, false);
-               if (newoff == TDB_OFF_ERR || newoff == 0)
-                       goto unlock;
        }
-
-       /* FIXME: If the space before is empty, we know this is in its ideal
-        * location.  We can steal a bit from the pointer to avoid rehash. */
-       for (i = tdb_find_nonzero_off(tdb, tdb->header.v.hash_off, num);
-            i < num;
-            i += tdb_find_nonzero_off(tdb, tdb->header.v.hash_off
-                                      + i*sizeof(tdb_off_t), num - i)) {
-               tdb_off_t off;
-               off = tdb_read_off(tdb, tdb->header.v.hash_off
-                                  + i*sizeof(tdb_off_t));
-               if (unlikely(off == TDB_OFF_ERR))
-                       goto unlock;
-               if (unlikely(!off)) {
-                       tdb->ecode = TDB_ERR_CORRUPT;
-                       tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-                                "find_bucket_and_lock: zero hash bucket!\n");
-                       goto unlock;
-               }
-               h = hash_record(tdb, off);
-               /* FIXME: Encode extra hash bits! */
-               if (tdb_write_off(tdb, newoff
-                                 + (h & ((num * 2) - 1)) * sizeof(uint64_t),
-                                 off) == -1)
-                       goto unlock;
+       switch (flag) {
+       case TDB_NOLOCK:
+               tdb->flags |= TDB_NOLOCK;
+               break;
+       case TDB_NOMMAP:
+               tdb->flags |= TDB_NOMMAP;
+               tdb_munmap(tdb->file);
+               break;
+       case TDB_NOSYNC:
+               tdb->flags |= TDB_NOSYNC;
+               break;
+       case TDB_SEQNUM:
+               tdb->flags |= TDB_SEQNUM;
+               break;
+       case TDB_ALLOW_NESTING:
+               tdb->flags |= TDB_ALLOW_NESTING;
+               break;
+       case TDB_RDONLY:
+               if (readonly_changable(tdb, "tdb_add_flag"))
+                       tdb->flags |= TDB_RDONLY;
+               break;
+       default:
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                            TDB_LOG_USE_ERROR,
+                                            "tdb_add_flag: Unknown flag %u",
+                                            flag);
        }
-
-       /* Free up old hash. */
-       r = tdb_get(tdb, tdb->header.v.hash_off, &pad, sizeof(*r));
-       if (!r)
-               goto unlock;
-       add_free_record(tdb, tdb->header.v.hash_off,
-                       rec_data_length(r) + rec_extra_padding(r));
-
-       /* Now we write the modified header. */
-       tdb->header.v.generation++;
-       tdb->header.v.hash_bits++;
-       tdb->header.v.hash_off = newoff;
-       tdb_write_convert(tdb, offsetof(struct tdb_header, v),
-                         &tdb->header.v, sizeof(tdb->header.v));
-unlock:
-       tdb_allrecord_unlock(tdb, F_WRLCK);
 }
 
-int tdb_store(struct tdb_context *tdb,
-             struct tdb_data key, struct tdb_data dbuf, int flag)
+void tdb_remove_flag(struct tdb_context *tdb, unsigned flag)
 {
-       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);
-
-       /* 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;
+       if (tdb->flags & TDB_INTERNAL) {
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                            TDB_LOG_USE_ERROR,
+                                            "tdb_remove_flag: internal db");
+               return;
        }
-
-       /* Now we have lock on this hash bucket. */
-       if (flag == TDB_INSERT) {
-               if (off) {
-                       tdb->ecode = TDB_ERR_EXISTS;
-                       goto fail;
-               }
-       } else {
-               if (off) {
-                       if (rec_data_length(&rec) + rec_extra_padding(&rec)
-                           >= dbuf.dsize) {
-                               new_off = off;
-                               if (update_rec_hdr(tdb, off,
-                                                  key.dsize, dbuf.dsize,
-                                                  &rec, h))
-                                       goto fail;
-                               goto write;
-                       }
-                       /* FIXME: See if right record is free? */
-                       /* Hint to allocator that we've realloced. */
-                       growing = true;
-               } else {
-                       if (flag == TDB_MODIFY) {
-                               /* if the record doesn't exist and we
-                                  are in TDB_MODIFY mode then we should fail
-                                  the store */
-                               tdb->ecode = TDB_ERR_NOEXIST;
-                               goto fail;
-                       }
+       switch (flag) {
+       case TDB_NOLOCK:
+               tdb->flags &= ~TDB_NOLOCK;
+               break;
+       case TDB_NOMMAP:
+               tdb->flags &= ~TDB_NOMMAP;
+               tdb_mmap(tdb);
+               break;
+       case TDB_NOSYNC:
+               tdb->flags &= ~TDB_NOSYNC;
+               break;
+       case TDB_SEQNUM:
+               tdb->flags &= ~TDB_SEQNUM;
+               break;
+       case TDB_ALLOW_NESTING:
+               tdb->flags &= ~TDB_ALLOW_NESTING;
+               break;
+       case TDB_RDONLY:
+               if ((tdb->open_flags & O_ACCMODE) == O_RDONLY) {
+                       tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                                    TDB_LOG_USE_ERROR,
+                                                    "tdb_remove_flag: can't"
+                                                    " remove TDB_RDONLY on tdb"
+                                                    " opened with O_RDONLY");
+                       break;
                }
+               if (readonly_changable(tdb, "tdb_remove_flag"))
+                       tdb->flags &= ~TDB_RDONLY;
+               break;
+       default:
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                            TDB_LOG_USE_ERROR,
+                                            "tdb_remove_flag: Unknown flag %u",
+                                            flag);
        }
-
-       /* Allocate a new record. */
-       new_off = alloc(tdb, key.dsize, dbuf.dsize, h, growing);
-       if (new_off == 0) {
-               unlock_range(tdb, start, num_locks, F_WRLCK);
-               /* Expand, then try again... */
-               if (tdb_expand(tdb, key.dsize, dbuf.dsize, growing) == -1)
-                       return -1;
-               return tdb_store(tdb, key, dbuf, flag);
-       }
-
-       /* We didn't like the existing one: remove it. */
-       if (off) {
-               add_free_record(tdb, off, sizeof(struct tdb_used_record)
-                               + rec_key_length(&rec)
-                               + rec_data_length(&rec)
-                               + rec_extra_padding(&rec));
-       }
-
-write:
-       /* FIXME: Encode extra hash bits! */
-       if (tdb_write_off(tdb, hash_off(tdb, old_bucket), new_off) == -1)
-               goto fail;
-
-       off = new_off + sizeof(struct tdb_used_record);
-       if (tdb->methods->write(tdb, off, key.dptr, key.dsize) == -1)
-               goto fail;
-       off += key.dsize;
-       if (tdb->methods->write(tdb, off, dbuf.dptr, dbuf.dsize) == -1)
-               goto fail;
-
-       /* FIXME: tdb_increment_seqnum(tdb); */
-       unlock_range(tdb, start, num_locks, F_WRLCK);
-
-       /* 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_range(tdb, start, num_locks, F_WRLCK);
-       return -1;
 }
 
-struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
+const char *tdb_errorstr(enum TDB_ERROR ecode)
 {
-       tdb_off_t off, start, num_locks = 1;
-       struct tdb_used_record rec;
-       uint64_t h;
-       struct tdb_data ret;
-
-       h = tdb_hash(tdb, key.dptr, key.dsize);
-
-       /* 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;
-       }
-
-       if (!off) {
-               unlock_range(tdb, start, num_locks, F_RDLCK);
-               tdb->ecode = TDB_ERR_NOEXIST;
-               return tdb_null;
-       }
+       /* Gcc warns if you miss a case in the switch, so use that. */
+       switch (ecode) {
+       case TDB_SUCCESS: return "Success";
+       case TDB_ERR_CORRUPT: return "Corrupt database";
+       case TDB_ERR_IO: return "IO Error";
+       case TDB_ERR_LOCK: return "Locking error";
+       case TDB_ERR_OOM: return "Out of memory";
+       case TDB_ERR_EXISTS: return "Record exists";
+       case TDB_ERR_EINVAL: return "Invalid parameter";
+       case TDB_ERR_NOEXIST: return "Record does not exist";
+       case TDB_ERR_RDONLY: return "write not permitted";
+       }
+       return "Invalid error code";
+}
 
-       ret.dsize = rec_data_length(&rec);
-       ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
-                                 ret.dsize);
-       unlock_range(tdb, start, num_locks, F_RDLCK);
-       return ret;
+enum TDB_ERROR tdb_error(struct tdb_context *tdb)
+{
+       return tdb->last_error;
 }
 
-static int hash_add(struct tdb_context *tdb, uint64_t h, tdb_off_t off)
+enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
+                              enum TDB_ERROR ecode,
+                              enum tdb_log_level level,
+                              const char *fmt, ...)
 {
-       tdb_off_t i, hoff, len, num;
-
-       /* Look for next space. */
-       i = (h & ((1ULL << tdb->header.v.hash_bits) - 1));
-       len = (1ULL << tdb->header.v.hash_bits) - i;
-       num = tdb_find_zero_off(tdb, hash_off(tdb, i), len);
-
-       if (unlikely(num == len)) {
-               /* 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) {
-                       tdb->ecode = TDB_ERR_CORRUPT;
-                       tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-                                "hash_add: full hash table!\n");
-                       return -1;
-               }
+       char *message;
+       va_list ap;
+       size_t len;
+       /* tdb_open paths care about errno, so save it. */
+       int saved_errno = errno;
+
+       if (!tdb->log_fn)
+               return ecode;
+
+       va_start(ap, fmt);
+       len = vasprintf(&message, fmt, ap);
+       va_end(ap);
+
+       if (len < 0) {
+               tdb->log_fn(tdb, TDB_LOG_ERROR, TDB_ERR_OOM,
+                           "out of memory formatting message:", tdb->log_data);
+               tdb->log_fn(tdb, level, ecode, fmt, tdb->log_data);
+       } else {
+               tdb->log_fn(tdb, level, ecode, message, tdb->log_data);
+               free(message);
        }
-       /* FIXME: Encode extra hash bits! */
-       return tdb_write_off(tdb, hash_off(tdb, i + num), off);
+       errno = saved_errno;
+       return ecode;
 }
 
-int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
+enum TDB_ERROR tdb_parse_record_(struct tdb_context *tdb,
+                                TDB_DATA key,
+                                enum TDB_ERROR (*parse)(TDB_DATA k,
+                                                        TDB_DATA d,
+                                                        void *data),
+                                void *data)
 {
-       tdb_off_t i, old_bucket, off, start, num_locks = 1;
+       tdb_off_t off;
        struct tdb_used_record rec;
-       uint64_t h;
-
-       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;
-       }
+       struct hash_info h;
+       enum TDB_ERROR ecode;
 
-       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;
+       off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
+       if (TDB_OFF_IS_ERR(off)) {
+               return tdb->last_error = off;
        }
 
        if (!off) {
-               unlock_range(tdb, start, num_locks, F_WRLCK);
-               tdb->ecode = TDB_ERR_NOEXIST;
-               return -1;
-       }
+               ecode = TDB_ERR_NOEXIST;
+       } else {
+               const void *dptr;
+               dptr = tdb_access_read(tdb, off + sizeof(rec) + key.dsize,
+                                      rec_data_length(&rec), false);
+               if (TDB_PTR_IS_ERR(dptr)) {
+                       ecode = TDB_PTR_ERR(dptr);
+               } else {
+                       TDB_DATA d = tdb_mkdata(dptr, rec_data_length(&rec));
 
-delete:
-       /* This actually unlinks it. */
-       if (tdb_write_off(tdb, hash_off(tdb, old_bucket), 0) == -1)
-               goto unlock_err;
-
-       /* Rehash anything following. */
-       for (i = hash_off(tdb, old_bucket+1);
-            i != hash_off(tdb, h + num_locks);
-            i += sizeof(tdb_off_t)) {
-               tdb_off_t off2;
-               uint64_t h2;
-
-               off2 = tdb_read_off(tdb, i);
-               if (unlikely(off2 == TDB_OFF_ERR))
-                       goto unlock_err;
-
-               /* Maybe use a bit to indicate it is in ideal place? */
-               h2 = hash_record(tdb, off2);
-               /* Is it happy where it is? */
-               if (hash_off(tdb, h2) == i)
-                       continue;
-
-               /* Remove it. */
-               if (tdb_write_off(tdb, i, 0) == -1)
-                       goto unlock_err;
-
-               /* Rehash it. */
-               if (hash_add(tdb, h2, off2) == -1)
-                       goto unlock_err;
+                       ecode = parse(key, d, data);
+                       tdb_access_release(tdb, dptr);
+               }
        }
 
-       /* 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_range(tdb, start, num_locks, F_WRLCK);
-       return -1;
+       tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
+       return tdb->last_error = ecode;
 }
 
-int tdb_close(struct tdb_context *tdb)
+const char *tdb_name(const struct tdb_context *tdb)
 {
-       struct tdb_context **i;
-       int ret = 0;
-
-       /* FIXME:
-       if (tdb->transaction) {
-               tdb_transaction_cancel(tdb);
-       }
-       */
-       tdb_trace(tdb, "tdb_close");
-
-       if (tdb->map_ptr) {
-               if (tdb->flags & TDB_INTERNAL)
-                       free(tdb->map_ptr);
-               else
-                       tdb_munmap(tdb);
-       }
-       free((char *)tdb->name);
-       if (tdb->fd != -1) {
-               ret = close(tdb->fd);
-               tdb->fd = -1;
-       }
-       free(tdb->lockrecs);
-
-       /* Remove from contexts list */
-       for (i = &tdbs; *i; i = &(*i)->next) {
-               if (*i == tdb) {
-                       *i = tdb->next;
-                       break;
-               }
-       }
-
-#ifdef TDB_TRACE
-       close(tdb->tracefd);
-#endif
-       free(tdb);
+       return tdb->name;
+}
 
-       return ret;
+int64_t tdb_get_seqnum(struct tdb_context *tdb)
+{
+       tdb_off_t off = tdb_read_off(tdb, offsetof(struct tdb_header, seqnum));
+       if (TDB_OFF_IS_ERR(off))
+               tdb->last_error = off;
+       else
+               tdb->last_error = TDB_SUCCESS;
+       return off;
 }
+       
 
-enum TDB_ERROR tdb_error(struct tdb_context *tdb)
+int tdb_fd(const struct tdb_context *tdb)
 {
-       return tdb->ecode;
+       return tdb->file->fd;
 }