tdb2: allow multiple chain locks.
[ccan] / ccan / tdb2 / transaction.c
index 92658026ac9c006c5e6f9bf31b99f77bb116abc6..68ae234d2f90ccbe4fb993dd27476df7c89afba2 100644 (file)
 
   - don't allow any locks to be held when a transaction starts,
     otherwise we can end up with deadlock (plus lack of lock nesting
-    in posix locks would mean the lock is lost)
+    in POSIX locks would mean the lock is lost)
 
   - if the caller gains a lock during the transaction but doesn't
     release it then fail the commit
 
   - allow for nested calls to tdb_transaction_start(), re-using the
-    existing transaction record. If the inner transaction is cancelled
+    existing transaction record. If the inner transaction is canceled
     then a subsequent commit will fail
 
   - keep a mirrored copy of the tdb hash chain heads to allow for the
@@ -68,7 +68,7 @@
 
   - allow callers to mix transaction and non-transaction use of tdb,
     although once a transaction is started then an exclusive lock is
-    gained until the transaction is committed or cancelled
+    gained until the transaction is committed or canceled
 
   - the commit stategy involves first saving away all modified data
     into a linearised buffer in the transaction recovery area, then
@@ -120,21 +120,25 @@ struct tdb_transaction {
        tdb_len_t old_map_size;
 };
 
+/* This doesn't really need to be pagesize, but we use it for similar reasons. */
+#define PAGESIZE 4096
 
 /*
   read while in a transaction. We need to check first if the data is in our list
   of transaction elements, then if not do a real read
 */
-static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
-                           tdb_len_t len)
+static enum TDB_ERROR transaction_read(struct tdb_context *tdb, tdb_off_t off,
+                                      void *buf, tdb_len_t len)
 {
        size_t blk;
+       enum TDB_ERROR ecode;
 
        /* break it down into block sized ops */
-       while (len + (off % getpagesize()) > getpagesize()) {
-               tdb_len_t len2 = getpagesize() - (off % getpagesize());
-               if (transaction_read(tdb, off, buf, len2) != 0) {
-                       return -1;
+       while (len + (off % PAGESIZE) > PAGESIZE) {
+               tdb_len_t len2 = PAGESIZE - (off % PAGESIZE);
+               ecode = transaction_read(tdb, off, buf, len2);
+               if (ecode != TDB_SUCCESS) {
+                       return ecode;
                }
                len -= len2;
                off += len2;
@@ -142,16 +146,17 @@ static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
        }
 
        if (len == 0) {
-               return 0;
+               return TDB_SUCCESS;
        }
 
-       blk = off / getpagesize();
+       blk = off / PAGESIZE;
 
        /* see if we have it in the block list */
        if (tdb->transaction->num_blocks <= blk ||
            tdb->transaction->blocks[blk] == NULL) {
                /* nope, do a real read */
-               if (tdb->transaction->io_methods->read(tdb, off, buf, len) != 0) {
+               ecode = tdb->transaction->io_methods->tread(tdb, off, buf, len);
+               if (ecode != TDB_SUCCESS) {
                        goto fail;
                }
                return 0;
@@ -160,43 +165,45 @@ static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
        /* it is in the block list. Now check for the last block */
        if (blk == tdb->transaction->num_blocks-1) {
                if (len > tdb->transaction->last_block_size) {
+                       ecode = TDB_ERR_IO;
                        goto fail;
                }
        }
 
        /* now copy it out of this block */
-       memcpy(buf, tdb->transaction->blocks[blk] + (off % getpagesize()), len);
-       return 0;
+       memcpy(buf, tdb->transaction->blocks[blk] + (off % PAGESIZE), len);
+       return TDB_SUCCESS;
 
 fail:
-       tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
-                  "transaction_read: failed at off=%zu len=%zu",
-                  (size_t)off, (size_t)len);
        tdb->transaction->transaction_error = 1;
-       return -1;
+       return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                         "transaction_read: failed at off=%zu len=%zu",
+                         (size_t)off, (size_t)len);
 }
 
 
 /*
   write while in a transaction
 */
-static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
-                            const void *buf, tdb_len_t len)
+static enum TDB_ERROR transaction_write(struct tdb_context *tdb, tdb_off_t off,
+                                       const void *buf, tdb_len_t len)
 {
        size_t blk;
+       enum TDB_ERROR ecode;
 
        /* Only a commit is allowed on a prepared transaction */
        if (tdb->transaction->prepared) {
-               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_FATAL,
-                        "transaction_write: transaction already prepared,"
-                        " write not allowed");
+               ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_ERROR,
+                                  "transaction_write: transaction already"
+                                  " prepared, write not allowed");
                goto fail;
        }
 
        /* break it up into block sized chunks */
-       while (len + (off % getpagesize()) > getpagesize()) {
-               tdb_len_t len2 = getpagesize() - (off % getpagesize());
-               if (transaction_write(tdb, off, buf, len2) != 0) {
+       while (len + (off % PAGESIZE) > PAGESIZE) {
+               tdb_len_t len2 = PAGESIZE - (off % PAGESIZE);
+               ecode = transaction_write(tdb, off, buf, len2);
+               if (ecode != TDB_SUCCESS) {
                        return -1;
                }
                len -= len2;
@@ -207,11 +214,11 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
        }
 
        if (len == 0) {
-               return 0;
+               return TDB_SUCCESS;
        }
 
-       blk = off / getpagesize();
-       off = off % getpagesize();
+       blk = off / PAGESIZE;
+       off = off % PAGESIZE;
 
        if (tdb->transaction->num_blocks <= blk) {
                uint8_t **new_blocks;
@@ -225,8 +232,9 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
                                (blk+1)*sizeof(uint8_t *));
                }
                if (new_blocks == NULL) {
-                       tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-                                  "transaction_write: failed to allocate");
+                       ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+                                          "transaction_write:"
+                                          " failed to allocate");
                        goto fail;
                }
                memset(&new_blocks[tdb->transaction->num_blocks], 0,
@@ -238,24 +246,29 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
 
        /* allocate and fill a block? */
        if (tdb->transaction->blocks[blk] == NULL) {
-               tdb->transaction->blocks[blk] = (uint8_t *)calloc(getpagesize(), 1);
+               tdb->transaction->blocks[blk] = (uint8_t *)calloc(PAGESIZE, 1);
                if (tdb->transaction->blocks[blk] == NULL) {
-                       tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-                                  "transaction_write: failed to allocate");
+                       ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+                                          "transaction_write:"
+                                          " failed to allocate");
                        goto fail;
                }
-               if (tdb->transaction->old_map_size > blk * getpagesize()) {
-                       tdb_len_t len2 = getpagesize();
-                       if (len2 + (blk * getpagesize()) > tdb->transaction->old_map_size) {
-                               len2 = tdb->transaction->old_map_size - (blk * getpagesize());
+               if (tdb->transaction->old_map_size > blk * PAGESIZE) {
+                       tdb_len_t len2 = PAGESIZE;
+                       if (len2 + (blk * PAGESIZE) > tdb->transaction->old_map_size) {
+                               len2 = tdb->transaction->old_map_size - (blk * PAGESIZE);
                        }
-                       if (tdb->transaction->io_methods->read(tdb, blk * getpagesize(),
-                                                              tdb->transaction->blocks[blk],
-                                                              len2) != 0) {
-                               tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-                                          "transaction_write: failed to"
-                                          " read old block: %s",
-                                          strerror(errno));
+                       ecode = tdb->transaction->io_methods->tread(tdb,
+                                       blk * PAGESIZE,
+                                       tdb->transaction->blocks[blk],
+                                       len2);
+                       if (ecode != TDB_SUCCESS) {
+                               ecode = tdb_logerr(tdb, ecode,
+                                                  TDB_LOG_ERROR,
+                                                  "transaction_write:"
+                                                  " failed to"
+                                                  " read old block: %s",
+                                                  strerror(errno));
                                SAFE_FREE(tdb->transaction->blocks[blk]);
                                goto fail;
                        }
@@ -277,16 +290,16 @@ static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
                }
        }
 
-       return 0;
+       return TDB_SUCCESS;
 
 fail:
        tdb->transaction->transaction_error = 1;
-       return -1;
+       return ecode;
 }
 
 
 /*
-  write while in a transaction - this varient never expands the transaction blocks, it only
+  write while in a transaction - this variant never expands the transaction blocks, it only
   updates existing blocks. This means it cannot change the recovery size
 */
 static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
@@ -295,8 +308,8 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
        size_t blk;
 
        /* break it up into block sized chunks */
-       while (len + (off % getpagesize()) > getpagesize()) {
-               tdb_len_t len2 = getpagesize() - (off % getpagesize());
+       while (len + (off % PAGESIZE) > PAGESIZE) {
+               tdb_len_t len2 = PAGESIZE - (off % PAGESIZE);
                transaction_write_existing(tdb, off, buf, len2);
                len -= len2;
                off += len2;
@@ -309,8 +322,8 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
                return;
        }
 
-       blk = off / getpagesize();
-       off = off % getpagesize();
+       blk = off / PAGESIZE;
+       off = off % PAGESIZE;
 
        if (tdb->transaction->num_blocks <= blk ||
            tdb->transaction->blocks[blk] == NULL) {
@@ -333,40 +346,72 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
 /*
   out of bounds check during a transaction
 */
-static int transaction_oob(struct tdb_context *tdb, tdb_off_t len, bool probe)
+static enum TDB_ERROR transaction_oob(struct tdb_context *tdb, tdb_off_t len,
+                                     bool probe)
 {
-       if (len <= tdb->map_size) {
-               return 0;
+       if (len <= tdb->file->map_size) {
+               return TDB_SUCCESS;
        }
-       tdb->ecode = TDB_ERR_IO;
        if (!probe) {
-               tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
+               tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
                           "tdb_oob len %lld beyond transaction size %lld",
                           (long long)len,
-                          (long long)tdb->map_size);
+                          (long long)tdb->file->map_size);
        }
-       return -1;
+       return TDB_ERR_IO;
 }
 
 /*
   transaction version of tdb_expand().
 */
-static int transaction_expand_file(struct tdb_context *tdb, tdb_off_t addition)
+static enum TDB_ERROR transaction_expand_file(struct tdb_context *tdb,
+                                             tdb_off_t addition)
 {
+       enum TDB_ERROR ecode;
+
        /* add a write to the transaction elements, so subsequent
           reads see the zero data */
-       if (transaction_write(tdb, tdb->map_size, NULL, addition) != 0) {
-               return -1;
+       ecode = transaction_write(tdb, tdb->file->map_size, NULL, addition);
+       if (ecode == TDB_SUCCESS) {
+               tdb->file->map_size += addition;
        }
-       tdb->map_size += addition;
-       return 0;
+       return ecode;
 }
 
 static void *transaction_direct(struct tdb_context *tdb, tdb_off_t off,
-                               size_t len)
+                               size_t len, bool write_mode)
 {
-       /* FIXME */
-       return NULL;
+       size_t blk = off / PAGESIZE, end_blk;
+
+       /* This is wrong for zero-length blocks, but will fail gracefully */
+       end_blk = (off + len - 1) / PAGESIZE;
+
+       /* Can only do direct if in single block and we've already copied. */
+       if (write_mode) {
+               if (blk != end_blk)
+                       return NULL;
+               if (blk >= tdb->transaction->num_blocks)
+                       return NULL;
+               if (tdb->transaction->blocks[blk] == NULL)
+                       return NULL;
+               return tdb->transaction->blocks[blk] + off % PAGESIZE;
+       }
+
+       /* Single which we have copied? */
+       if (blk == end_blk
+           && blk < tdb->transaction->num_blocks
+           && tdb->transaction->blocks[blk])
+               return tdb->transaction->blocks[blk] + off % PAGESIZE;
+
+       /* Otherwise must be all not copied. */
+       while (blk <= end_blk) {
+               if (blk >= tdb->transaction->num_blocks)
+                       break;
+               if (tdb->transaction->blocks[blk])
+                       return NULL;
+               blk++;
+       }
+       return tdb->transaction->io_methods->direct(tdb, off, len, false);
 }
 
 static const struct tdb_methods transaction_methods = {
@@ -380,40 +425,40 @@ static const struct tdb_methods transaction_methods = {
 /*
   sync to disk
 */
-static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t length)
+static enum TDB_ERROR transaction_sync(struct tdb_context *tdb,
+                                      tdb_off_t offset, tdb_len_t length)
 {
        if (tdb->flags & TDB_NOSYNC) {
-               return 0;
+               return TDB_SUCCESS;
        }
 
-       if (fsync(tdb->fd) != 0) {
-               tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
-                          "tdb_transaction: fsync failed: %s",
-                          strerror(errno));
-               return -1;
+       if (fsync(tdb->file->fd) != 0) {
+               return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+                                 "tdb_transaction: fsync failed: %s",
+                                 strerror(errno));
        }
 #ifdef MS_SYNC
-       if (tdb->map_ptr) {
-               tdb_off_t moffset = offset & ~(getpagesize()-1);
-               if (msync(moffset + (char *)tdb->map_ptr,
+       if (tdb->file->map_ptr) {
+               tdb_off_t moffset = offset & ~(PAGESIZE-1);
+               if (msync(moffset + (char *)tdb->file->map_ptr,
                          length + (offset - moffset), MS_SYNC) != 0) {
-                       tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
-                                  "tdb_transaction: msync failed: %s",
-                                  strerror(errno));
-                       return -1;
+                       return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+                                         "tdb_transaction: msync failed: %s",
+                                         strerror(errno));
                }
        }
 #endif
-       return 0;
+       return TDB_SUCCESS;
 }
 
 
 static void _tdb_transaction_cancel(struct tdb_context *tdb)
 {
        int i;
+       enum TDB_ERROR ecode;
 
        if (tdb->transaction == NULL) {
-               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
+               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
                           "tdb_transaction_cancel: no transaction");
                return;
        }
@@ -424,7 +469,7 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
                return;
        }
 
-       tdb->map_size = tdb->transaction->old_map_size;
+       tdb->file->map_size = tdb->transaction->old_map_size;
 
        /* free all the transaction blocks */
        for (i=0;i<tdb->transaction->num_blocks;i++) {
@@ -439,18 +484,21 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
                uint64_t invalid = TDB_RECOVERY_INVALID_MAGIC;
 
                /* remove the recovery marker */
-               if (methods->write(tdb, tdb->transaction->magic_offset,
-                                  &invalid, sizeof(invalid)) == -1 ||
-                   transaction_sync(tdb, tdb->transaction->magic_offset,
-                                    sizeof(invalid)) == -1) {
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
+               ecode = methods->twrite(tdb, tdb->transaction->magic_offset,
+                                       &invalid, sizeof(invalid));
+               if (ecode == TDB_SUCCESS)
+                       ecode = transaction_sync(tdb,
+                                                tdb->transaction->magic_offset,
+                                                sizeof(invalid));
+               if (ecode != TDB_SUCCESS) {
+                       tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
                                   "tdb_transaction_cancel: failed to remove"
                                   " recovery magic");
                }
        }
 
-       if (tdb->allrecord_lock.count)
-               tdb_allrecord_unlock(tdb, tdb->allrecord_lock.ltype);
+       if (tdb->file->allrecord_lock.count)
+               tdb_allrecord_unlock(tdb, tdb->file->allrecord_lock.ltype);
 
        /* restore the normal io methods */
        tdb->methods = tdb->transaction->io_methods;
@@ -467,73 +515,83 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
   start a tdb transaction. No token is returned, as only a single
   transaction is allowed to be pending per tdb_context
 */
-int tdb_transaction_start(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_start(struct tdb_context *tdb)
 {
+       enum TDB_ERROR ecode;
+
        /* some sanity checks */
        if (tdb->read_only || (tdb->flags & TDB_INTERNAL)) {
-               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
-                          "tdb_transaction_start: cannot start a transaction"
-                          " on a read-only or internal db");
-               return -1;
+               return tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                                   TDB_LOG_USE_ERROR,
+                                                   "tdb_transaction_start:"
+                                                   " cannot start a"
+                                                   " transaction on a "
+                                                   "read-only or internal db");
        }
 
        /* cope with nested tdb_transaction_start() calls */
        if (tdb->transaction != NULL) {
-               tdb_logerr(tdb, TDB_ERR_NESTING, TDB_DEBUG_ERROR,
-                          "tdb_transaction_start:"
-                          " already inside transaction");
-               return -1;
+               return tdb->last_error = tdb_logerr(tdb, TDB_ERR_IO,
+                                                   TDB_LOG_USE_ERROR,
+                                                   "tdb_transaction_start:"
+                                                   " already inside"
+                                                   " transaction");
        }
 
        if (tdb_has_hash_locks(tdb)) {
                /* the caller must not have any locks when starting a
                   transaction as otherwise we'll be screwed by lack
-                  of nested locks in posix */
-               tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
-                          "tdb_transaction_start: cannot start a transaction"
-                          " with locks held");
-               return -1;
+                  of nested locks in POSIX */
+               return tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK,
+                                                   TDB_LOG_USE_ERROR,
+                                                   "tdb_transaction_start:"
+                                                   " cannot start a"
+                                                   " transaction with locks"
+                                                   " held");
        }
 
        tdb->transaction = (struct tdb_transaction *)
                calloc(sizeof(struct tdb_transaction), 1);
        if (tdb->transaction == NULL) {
-               tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
-                          "tdb_transaction_start: cannot allocate");
-               return -1;
+               return tdb->last_error = tdb_logerr(tdb, TDB_ERR_OOM,
+                                                   TDB_LOG_ERROR,
+                                                   "tdb_transaction_start:"
+                                                   " cannot allocate");
        }
 
        /* get the transaction write lock. This is a blocking lock. As
           discussed with Volker, there are a number of ways we could
           make this async, which we will probably do in the future */
-       if (tdb_transaction_lock(tdb, F_WRLCK) == -1) {
+       ecode = tdb_transaction_lock(tdb, F_WRLCK);
+       if (ecode != TDB_SUCCESS) {
                SAFE_FREE(tdb->transaction->blocks);
                SAFE_FREE(tdb->transaction);
-               return -1;
+               return tdb->last_error = ecode;
        }
 
        /* get a read lock over entire file. This is upgraded to a write
           lock during the commit */
-       if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true) == -1) {
+       ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true);
+       if (ecode != TDB_SUCCESS) {
                goto fail_allrecord_lock;
        }
 
        /* make sure we know about any file expansions already done by
           anyone else */
-       tdb->methods->oob(tdb, tdb->map_size + 1, true);
-       tdb->transaction->old_map_size = tdb->map_size;
+       tdb->methods->oob(tdb, tdb->file->map_size + 1, true);
+       tdb->transaction->old_map_size = tdb->file->map_size;
 
        /* finally hook the io methods, replacing them with
           transaction specific methods */
        tdb->transaction->io_methods = tdb->methods;
        tdb->methods = &transaction_methods;
-       return 0;
+       return tdb->last_error = TDB_SUCCESS;
 
 fail_allrecord_lock:
        tdb_transaction_unlock(tdb, F_WRLCK);
        SAFE_FREE(tdb->transaction->blocks);
        SAFE_FREE(tdb->transaction);
-       return -1;
+       return tdb->last_error = ecode;
 }
 
 
@@ -555,7 +613,7 @@ static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
 
        recovery_size = sizeof(tdb_len_t);
        for (i=0;i<tdb->transaction->num_blocks;i++) {
-               if (i * getpagesize() >= tdb->transaction->old_map_size) {
+               if (i * PAGESIZE >= tdb->transaction->old_map_size) {
                        break;
                }
                if (tdb->transaction->blocks[i] == NULL) {
@@ -565,7 +623,7 @@ static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
                if (i == tdb->transaction->num_blocks-1) {
                        recovery_size += tdb->transaction->last_block_size;
                } else {
-                       recovery_size += getpagesize();
+                       recovery_size += PAGESIZE;
                }
        }
 
@@ -576,30 +634,30 @@ static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
   allocate the recovery area, or use an existing recovery area if it is
   large enough
 */
-static int tdb_recovery_allocate(struct tdb_context *tdb,
-                                tdb_len_t *recovery_size,
-                                tdb_off_t *recovery_offset,
-                                tdb_len_t *recovery_max_size)
+static enum TDB_ERROR tdb_recovery_allocate(struct tdb_context *tdb,
+                                           tdb_len_t *recovery_size,
+                                           tdb_off_t *recovery_offset,
+                                           tdb_len_t *recovery_max_size)
 {
        struct tdb_recovery_record rec;
        const struct tdb_methods *methods = tdb->transaction->io_methods;
        tdb_off_t recovery_head;
        size_t addition;
+       enum TDB_ERROR ecode;
 
        recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
-       if (recovery_head == TDB_OFF_ERR) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_recovery_allocate:"
-                        " failed to read recovery head");
-               return -1;
+       if (TDB_OFF_IS_ERR(recovery_head)) {
+               return tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
+                                 "tdb_recovery_allocate:"
+                                 " failed to read recovery head");
        }
 
        if (recovery_head != 0) {
-               if (methods->read(tdb, recovery_head, &rec, sizeof(rec))) {
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                                "tdb_recovery_allocate:"
-                                " failed to read recovery record");
-                       return -1;
+               ecode = methods->tread(tdb, recovery_head, &rec, sizeof(rec));
+               if (ecode != TDB_SUCCESS) {
+                       return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                         "tdb_recovery_allocate:"
+                                         " failed to read recovery record");
                }
                tdb_convert(tdb, &rec, sizeof(rec));
                /* ignore invalid recovery regions: can happen in crash */
@@ -615,7 +673,7 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
                /* it fits in the existing area */
                *recovery_max_size = rec.max_len;
                *recovery_offset = recovery_head;
-               return 0;
+               return TDB_SUCCESS;
        }
 
        /* we need to free up the old recovery area, then allocate a
@@ -624,13 +682,14 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
           us an area that is being currently used (as of the start of
           the transaction) */
        if (recovery_head != 0) {
-               add_stat(tdb, frees, 1);
-               if (add_free_record(tdb, recovery_head,
-                                   sizeof(rec) + rec.max_len) != 0) {
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                                  "tdb_recovery_allocate:"
-                                  " failed to free previous recovery area");
-                       return -1;
+               tdb->stats.frees++;
+               ecode = add_free_record(tdb, recovery_head,
+                                       sizeof(rec) + rec.max_len);
+               if (ecode != TDB_SUCCESS) {
+                       return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                         "tdb_recovery_allocate:"
+                                         " failed to free previous"
+                                         " recovery area");
                }
        }
 
@@ -639,45 +698,45 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 
        /* round up to a multiple of page size */
        *recovery_max_size
-               = (((sizeof(rec) + *recovery_size) + getpagesize()-1)
-                  & ~(getpagesize()-1))
+               = (((sizeof(rec) + *recovery_size) + PAGESIZE-1)
+                  & ~(PAGESIZE-1))
                - sizeof(rec);
-       *recovery_offset = tdb->map_size;
+       *recovery_offset = tdb->file->map_size;
        recovery_head = *recovery_offset;
 
        /* Restore ->map_size before calling underlying expand_file.
           Also so that we don't try to expand the file again in the
           transaction commit, which would destroy the recovery
           area */
-       addition = (tdb->map_size - tdb->transaction->old_map_size) +
+       addition = (tdb->file->map_size - tdb->transaction->old_map_size) +
                sizeof(rec) + *recovery_max_size;
-       tdb->map_size = tdb->transaction->old_map_size;
-       if (methods->expand_file(tdb, addition) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_recovery_allocate:"
-                        " failed to create recovery area");
-               return -1;
+       tdb->file->map_size = tdb->transaction->old_map_size;
+       ecode = methods->expand_file(tdb, addition);
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_recovery_allocate:"
+                                 " failed to create recovery area");
        }
 
        /* we have to reset the old map size so that we don't try to
           expand the file again in the transaction commit, which
           would destroy the recovery area */
-       tdb->transaction->old_map_size = tdb->map_size;
+       tdb->transaction->old_map_size = tdb->file->map_size;
 
        /* write the recovery header offset and sync - we can sync without a race here
           as the magic ptr in the recovery record has not been set */
        tdb_convert(tdb, &recovery_head, sizeof(recovery_head));
-       if (methods->write(tdb, offsetof(struct tdb_header, recovery),
-                          &recovery_head, sizeof(tdb_off_t)) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_recovery_allocate:"
-                        " failed to write recovery head");
-               return -1;
+       ecode = methods->twrite(tdb, offsetof(struct tdb_header, recovery),
+                               &recovery_head, sizeof(tdb_off_t));
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_recovery_allocate:"
+                                 " failed to write recovery head");
        }
        transaction_write_existing(tdb, offsetof(struct tdb_header, recovery),
                                   &recovery_head,
                                   sizeof(tdb_off_t));
-       return 0;
+       return TDB_SUCCESS;
 }
 
 /* Set up header for the recovery record. */
@@ -695,31 +754,34 @@ static void set_recovery_header(struct tdb_recovery_record *rec,
 /*
   setup the recovery data that will be used on a crash during commit
 */
-static int transaction_setup_recovery(struct tdb_context *tdb,
-                                     tdb_off_t *magic_offset)
+static enum TDB_ERROR transaction_setup_recovery(struct tdb_context *tdb,
+                                                tdb_off_t *magic_offset)
 {
-       tdb_len_t recovery_size;
+       /* Initialized for GCC's 4.4.5 overzealous uninitialized warnings. */
+       tdb_len_t recovery_size = 0;
+       tdb_off_t recovery_offset = 0, recovery_max_size = 0;
        unsigned char *data, *p;
        const struct tdb_methods *methods = tdb->transaction->io_methods;
        struct tdb_recovery_record *rec;
-       tdb_off_t recovery_offset, recovery_max_size;
        tdb_off_t old_map_size = tdb->transaction->old_map_size;
        uint64_t magic, tailer;
        int i;
+       enum TDB_ERROR ecode;
 
        /*
          check that the recovery area has enough space
        */
-       if (tdb_recovery_allocate(tdb, &recovery_size,
-                                 &recovery_offset, &recovery_max_size) == -1) {
-               return -1;
+       ecode = tdb_recovery_allocate(tdb, &recovery_size,
+                                     &recovery_offset, &recovery_max_size);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
        }
 
        data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
        if (data == NULL) {
-               tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-                          "transaction_setup_recovery: cannot allocate");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+                                 "transaction_setup_recovery:"
+                                 " cannot allocate");
        }
 
        rec = (struct tdb_recovery_record *)data;
@@ -738,8 +800,8 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
                        continue;
                }
 
-               offset = i * getpagesize();
-               length = getpagesize();
+               offset = i * PAGESIZE;
+               length = PAGESIZE;
                if (i == tdb->transaction->num_blocks-1) {
                        length = tdb->transaction->last_block_size;
                }
@@ -747,12 +809,12 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
                if (offset >= old_map_size) {
                        continue;
                }
-               if (offset + length > tdb->map_size) {
-                       tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL,
-                                  "tdb_transaction_setup_recovery:"
-                                  " transaction data over new region boundary");
+               if (offset + length > tdb->file->map_size) {
                        free(data);
-                       return -1;
+                       return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
+                                         "tdb_transaction_setup_recovery:"
+                                         " transaction data over new region"
+                                         " boundary");
                }
                memcpy(p, &offset, sizeof(offset));
                memcpy(p + sizeof(offset), &length, sizeof(length));
@@ -761,11 +823,12 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
                /* the recovery area contains the old data, not the
                   new data, so we have to call the original tdb_read
                   method to get it */
-               if (methods->read(tdb, offset,
-                                 p + sizeof(offset) + sizeof(length),
-                                 length) != 0) {
+               ecode = methods->tread(tdb, offset,
+                                      p + sizeof(offset) + sizeof(length),
+                                      length);
+               if (ecode != TDB_SUCCESS) {
                        free(data);
-                       return -1;
+                       return ecode;
                }
                p += sizeof(offset) + sizeof(length) + length;
        }
@@ -776,13 +839,13 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
        tdb_convert(tdb, p, sizeof(tailer));
 
        /* write the recovery data to the recovery area */
-       if (methods->write(tdb, recovery_offset, data,
-                          sizeof(*rec) + recovery_size) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_transaction_setup_recovery:"
-                        " failed to write recovery data");
+       ecode = methods->twrite(tdb, recovery_offset, data,
+                               sizeof(*rec) + recovery_size);
+       if (ecode != TDB_SUCCESS) {
                free(data);
-               return -1;
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_transaction_setup_recovery:"
+                                 " failed to write recovery data");
        }
        transaction_write_existing(tdb, recovery_offset, data,
                                   sizeof(*rec) + recovery_size);
@@ -790,10 +853,11 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
        /* as we don't have ordered writes, we have to sync the recovery
           data before we update the magic to indicate that the recovery
           data is present */
-       if (transaction_sync(tdb, recovery_offset,
-                            sizeof(*rec) + recovery_size) == -1) {
+       ecode = transaction_sync(tdb, recovery_offset,
+                                sizeof(*rec) + recovery_size);
+       if (ecode != TDB_SUCCESS) {
                free(data);
-               return -1;
+               return ecode;
        }
 
        free(data);
@@ -804,117 +868,103 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
        *magic_offset = recovery_offset + offsetof(struct tdb_recovery_record,
                                                   magic);
 
-       if (methods->write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_transaction_setup_recovery:"
-                        " failed to write recovery magic");
-               return -1;
+       ecode = methods->twrite(tdb, *magic_offset, &magic, sizeof(magic));
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_transaction_setup_recovery:"
+                                 " failed to write recovery magic");
        }
        transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic));
 
        /* ensure the recovery magic marker is on disk */
-       if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
-               return -1;
-       }
-
-       return 0;
+       return transaction_sync(tdb, *magic_offset, sizeof(magic));
 }
 
-static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
+static enum TDB_ERROR _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 {
        const struct tdb_methods *methods;
+       enum TDB_ERROR ecode;
 
        if (tdb->transaction == NULL) {
-               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
-                          "tdb_transaction_prepare_commit: no transaction");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+                                 "tdb_transaction_prepare_commit:"
+                                 " no transaction");
        }
 
        if (tdb->transaction->prepared) {
                _tdb_transaction_cancel(tdb);
-               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
-                          "tdb_transaction_prepare_commit:"
-                          " transaction already prepared");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+                                 "tdb_transaction_prepare_commit:"
+                                 " transaction already prepared");
        }
 
        if (tdb->transaction->transaction_error) {
                _tdb_transaction_cancel(tdb);
-               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
-                          "tdb_transaction_prepare_commit:"
-                          " transaction error pending");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_ERROR,
+                                 "tdb_transaction_prepare_commit:"
+                                 " transaction error pending");
        }
 
 
        if (tdb->transaction->nesting != 0) {
                tdb->transaction->nesting--;
-               return 0;
+               return TDB_SUCCESS;
        }
 
        /* check for a null transaction */
        if (tdb->transaction->blocks == NULL) {
-               return 0;
+               return TDB_SUCCESS;
        }
 
        methods = tdb->transaction->io_methods;
 
        /* upgrade the main transaction lock region to a write lock */
-       if (tdb_allrecord_upgrade(tdb) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_ERROR,
-                        "tdb_transaction_prepare_commit:"
-                        " failed to upgrade hash locks");
-               _tdb_transaction_cancel(tdb);
-               return -1;
+       ecode = tdb_allrecord_upgrade(tdb);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
        }
 
        /* get the open lock - this prevents new users attaching to the database
           during the commit */
-       if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_ERROR,
-                        "tdb_transaction_prepare_commit:"
-                        " failed to get open lock");
-               _tdb_transaction_cancel(tdb);
-               return -1;
+       ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
        }
 
        /* Since we have whole db locked, we don't need the expansion lock. */
        if (!(tdb->flags & TDB_NOSYNC)) {
                /* write the recovery data to the end of the file */
-               if (transaction_setup_recovery(tdb, &tdb->transaction->magic_offset) == -1) {
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                                "tdb_transaction_prepare_commit:"
-                                " failed to setup recovery data");
-                       _tdb_transaction_cancel(tdb);
-                       return -1;
+               ecode = transaction_setup_recovery(tdb,
+                                                  &tdb->transaction
+                                                  ->magic_offset);
+               if (ecode != TDB_SUCCESS) {
+                       return ecode;
                }
        }
 
        tdb->transaction->prepared = true;
 
        /* expand the file to the new size if needed */
-       if (tdb->map_size != tdb->transaction->old_map_size) {
-               tdb_len_t add = tdb->map_size - tdb->transaction->old_map_size;
+       if (tdb->file->map_size != tdb->transaction->old_map_size) {
+               tdb_len_t add;
+
+               add = tdb->file->map_size - tdb->transaction->old_map_size;
                /* Restore original map size for tdb_expand_file */
-               tdb->map_size = tdb->transaction->old_map_size;
-               if (methods->expand_file(tdb, add) == -1) {
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_ERROR,
-                                "tdb_transaction_prepare_commit:"
-                                " expansion failed");
-                       _tdb_transaction_cancel(tdb);
-                       return -1;
+               tdb->file->map_size = tdb->transaction->old_map_size;
+               ecode = methods->expand_file(tdb, add);
+               if (ecode != TDB_SUCCESS) {
+                       return ecode;
                }
        }
 
        /* Keep the open lock until the actual commit */
-
-       return 0;
+       return TDB_SUCCESS;
 }
 
 /*
    prepare to commit the current transaction
 */
-int tdb_transaction_prepare_commit(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_prepare_commit(struct tdb_context *tdb)
 {
        return _tdb_transaction_prepare_commit(tdb);
 }
@@ -922,43 +972,38 @@ int tdb_transaction_prepare_commit(struct tdb_context *tdb)
 /*
   commit the current transaction
 */
-int tdb_transaction_commit(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_commit(struct tdb_context *tdb)
 {
        const struct tdb_methods *methods;
        int i;
+       enum TDB_ERROR ecode;
 
        if (tdb->transaction == NULL) {
-               tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
-                        "tdb_transaction_commit: no transaction");
-               return -1;
+               return tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
+                                                   TDB_LOG_USE_ERROR,
+                                                   "tdb_transaction_commit:"
+                                                   " no transaction");
        }
 
        tdb_trace(tdb, "tdb_transaction_commit");
 
-       if (tdb->transaction->transaction_error) {
-               tdb_transaction_cancel(tdb);
-               tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
-                          "tdb_transaction_commit:"
-                          " transaction error pending");
-               return -1;
-       }
-
-
        if (tdb->transaction->nesting != 0) {
                tdb->transaction->nesting--;
-               return 0;
+               return tdb->last_error = TDB_SUCCESS;
        }
 
        /* check for a null transaction */
        if (tdb->transaction->blocks == NULL) {
                _tdb_transaction_cancel(tdb);
-               return 0;
+               return tdb->last_error = TDB_SUCCESS;
        }
 
        if (!tdb->transaction->prepared) {
-               int ret = _tdb_transaction_prepare_commit(tdb);
-               if (ret)
-                       return ret;
+               ecode = _tdb_transaction_prepare_commit(tdb);
+               if (ecode != TDB_SUCCESS) {
+                       _tdb_transaction_cancel(tdb);
+                       return tdb->last_error = ecode;
+               }
        }
 
        methods = tdb->transaction->io_methods;
@@ -972,18 +1017,15 @@ int tdb_transaction_commit(struct tdb_context *tdb)
                        continue;
                }
 
-               offset = i * getpagesize();
-               length = getpagesize();
+               offset = i * PAGESIZE;
+               length = PAGESIZE;
                if (i == tdb->transaction->num_blocks-1) {
                        length = tdb->transaction->last_block_size;
                }
 
-               if (methods->write(tdb, offset, tdb->transaction->blocks[i],
-                                  length) == -1) {
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                                  "tdb_transaction_commit:"
-                                  " write failed during commit");
-
+               ecode = methods->twrite(tdb, offset,
+                                       tdb->transaction->blocks[i], length);
+               if (ecode != TDB_SUCCESS) {
                        /* we've overwritten part of the data and
                           possibly expanded the file, so we need to
                           run the crash recovery code */
@@ -992,7 +1034,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 
                        _tdb_transaction_cancel(tdb);
 
-                       return -1;
+                       return tdb->last_error = ecode;
                }
                SAFE_FREE(tdb->transaction->blocks[i]);
        }
@@ -1001,8 +1043,9 @@ int tdb_transaction_commit(struct tdb_context *tdb)
        tdb->transaction->num_blocks = 0;
 
        /* ensure the new data is on disk */
-       if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
-               return -1;
+       ecode = transaction_sync(tdb, 0, tdb->file->map_size);
+       if (ecode != TDB_SUCCESS) {
+               return tdb->last_error = ecode;
        }
 
        /*
@@ -1021,10 +1064,11 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 #endif
 
        /* use a transaction cancel to free memory and remove the
-          transaction locks */
+          transaction locks: it "restores" map_size, too. */
+       tdb->transaction->old_map_size = tdb->file->map_size;
        _tdb_transaction_cancel(tdb);
 
-       return 0;
+       return tdb->last_error = TDB_SUCCESS;
 }
 
 
@@ -1033,63 +1077,61 @@ int tdb_transaction_commit(struct tdb_context *tdb)
   database write access already established (including the open
   lock to prevent new processes attaching)
 */
-int tdb_transaction_recover(struct tdb_context *tdb)
+enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb)
 {
        tdb_off_t recovery_head, recovery_eof;
        unsigned char *data, *p;
        struct tdb_recovery_record rec;
+       enum TDB_ERROR ecode;
 
        /* find the recovery area */
        recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
-       if (recovery_head == TDB_OFF_ERR) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_transaction_recover:"
-                        " failed to read recovery head");
-               return -1;
+       if (TDB_OFF_IS_ERR(recovery_head)) {
+               return tdb_logerr(tdb, recovery_head, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " failed to read recovery head");
        }
 
        if (recovery_head == 0) {
                /* we have never allocated a recovery record */
-               return 0;
+               return TDB_SUCCESS;
        }
 
        /* read the recovery record */
-       if (tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec)) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                          "tdb_transaction_recover:"
-                          " failed to read recovery record");
-               return -1;
+       ecode = tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec));
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " failed to read recovery record");
        }
 
        if (rec.magic != TDB_RECOVERY_MAGIC) {
                /* there is no valid recovery data */
-               return 0;
+               return TDB_SUCCESS;
        }
 
        if (tdb->read_only) {
-               tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL,
-                          "tdb_transaction_recover:"
-                          " attempt to recover read only database");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " attempt to recover read only database");
        }
 
        recovery_eof = rec.eof;
 
        data = (unsigned char *)malloc(rec.len);
        if (data == NULL) {
-               tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
-                          "tdb_transaction_recover:"
-                          " failed to allocate recovery data");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " failed to allocate recovery data");
        }
 
        /* read the full recovery data */
-       if (tdb->methods->read(tdb, recovery_head + sizeof(rec), data,
-                              rec.len) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                          "tdb_transaction_recover:"
-                          " failed to read recovery data");
-               return -1;
+       ecode = tdb->methods->tread(tdb, recovery_head + sizeof(rec), data,
+                                   rec.len);
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " failed to read recovery data");
        }
 
        /* recover the file data */
@@ -1102,71 +1144,75 @@ int tdb_transaction_recover(struct tdb_context *tdb)
                memcpy(&len, p + sizeof(ofs), sizeof(len));
                p += sizeof(ofs) + sizeof(len);
 
-               if (tdb->methods->write(tdb, ofs, p, len) == -1) {
+               ecode = tdb->methods->twrite(tdb, ofs, p, len);
+               if (ecode != TDB_SUCCESS) {
                        free(data);
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                                "tdb_transaction_recover:"
-                                " failed to recover %zu bytes at offset %zu",
-                                (size_t)len, (size_t)ofs);
-                       return -1;
+                       return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                         "tdb_transaction_recover:"
+                                         " failed to recover %zu bytes"
+                                         " at offset %zu",
+                                         (size_t)len, (size_t)ofs);
                }
                p += len;
        }
 
        free(data);
 
-       if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                          "tdb_transaction_recover: failed to sync recovery");
-               return -1;
+       ecode = transaction_sync(tdb, 0, tdb->file->map_size);
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " failed to sync recovery");
        }
 
        /* if the recovery area is after the recovered eof then remove it */
        if (recovery_eof <= recovery_head) {
-               if (tdb_write_off(tdb, offsetof(struct tdb_header,recovery), 0)
-                   == -1) {
-                       tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                                "tdb_transaction_recover:"
-                                " failed to remove recovery head");
-                       return -1;
+               ecode = tdb_write_off(tdb, offsetof(struct tdb_header,
+                                                   recovery),
+                                     0);
+               if (ecode != TDB_SUCCESS) {
+                       return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                         "tdb_transaction_recover:"
+                                         " failed to remove recovery head");
                }
        }
 
        /* remove the recovery magic */
-       if (tdb_write_off(tdb,
-                         recovery_head
-                         + offsetof(struct tdb_recovery_record, magic),
-                         TDB_RECOVERY_INVALID_MAGIC) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_transaction_recover:"
-                        " failed to remove recovery magic");
-               return -1;
+       ecode = tdb_write_off(tdb,
+                             recovery_head
+                             + offsetof(struct tdb_recovery_record, magic),
+                             TDB_RECOVERY_INVALID_MAGIC);
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " failed to remove recovery magic");
        }
 
-       if (transaction_sync(tdb, 0, recovery_eof) == -1) {
-               tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_FATAL,
-                        "tdb_transaction_recover: failed to sync2 recovery");
-               return -1;
+       ecode = transaction_sync(tdb, 0, recovery_eof);
+       if (ecode != TDB_SUCCESS) {
+               return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
+                                 "tdb_transaction_recover:"
+                                 " failed to sync2 recovery");
        }
 
-       tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
+       tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
                   "tdb_transaction_recover: recovered %zu byte database",
                   (size_t)recovery_eof);
 
        /* all done */
-       return 0;
+       return TDB_SUCCESS;
 }
 
-/* Any I/O failures we say "needs recovery". */
-bool tdb_needs_recovery(struct tdb_context *tdb)
+tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb)
 {
        tdb_off_t recovery_head;
        struct tdb_recovery_record rec;
+       enum TDB_ERROR ecode;
 
        /* find the recovery area */
        recovery_head = tdb_read_off(tdb, offsetof(struct tdb_header,recovery));
-       if (recovery_head == TDB_OFF_ERR) {
-               return true;
+       if (TDB_OFF_IS_ERR(recovery_head)) {
+               return recovery_head;
        }
 
        if (recovery_head == 0) {
@@ -1175,8 +1221,9 @@ bool tdb_needs_recovery(struct tdb_context *tdb)
        }
 
        /* read the recovery record */
-       if (tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec)) == -1) {
-               return true;
+       ecode = tdb_read_convert(tdb, recovery_head, &rec, sizeof(rec));
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
        }
 
        return (rec.magic == TDB_RECOVERY_MAGIC);