]> git.ozlabs.org Git - ccan/blobdiff - ccan/tdb2/transaction.c
tdb2: move mmap into struct tdb_file
[ccan] / ccan / tdb2 / transaction.c
index 66565d67c2d762b67cba9169ee3d9f8f661b78d3..73ab1e39ff4ad0c4f260be0bf1dc6d9eca95bcd2 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
@@ -297,7 +297,7 @@ fail:
 
 
 /*
-  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,
@@ -347,14 +347,14 @@ static void transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
 static enum TDB_ERROR transaction_oob(struct tdb_context *tdb, tdb_off_t len,
                                      bool probe)
 {
-       if (len <= tdb->map_size) {
+       if (len <= tdb->file->map_size) {
                return TDB_SUCCESS;
        }
        if (!probe) {
                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 TDB_ERR_IO;
 }
@@ -369,9 +369,9 @@ static enum TDB_ERROR transaction_expand_file(struct tdb_context *tdb,
 
        /* add a write to the transaction elements, so subsequent
           reads see the zero data */
-       ecode = transaction_write(tdb, tdb->map_size, NULL, addition);
+       ecode = transaction_write(tdb, tdb->file->map_size, NULL, addition);
        if (ecode == TDB_SUCCESS) {
-               tdb->map_size += addition;
+               tdb->file->map_size += addition;
        }
        return ecode;
 }
@@ -430,15 +430,15 @@ static enum TDB_ERROR transaction_sync(struct tdb_context *tdb,
                return TDB_SUCCESS;
        }
 
-       if (fsync(tdb->fd) != 0) {
+       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) {
+       if (tdb->file->map_ptr) {
                tdb_off_t moffset = offset & ~(getpagesize()-1);
-               if (msync(moffset + (char *)tdb->map_ptr,
+               if (msync(moffset + (char *)tdb->file->map_ptr,
                          length + (offset - moffset), MS_SYNC) != 0) {
                        return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
                                          "tdb_transaction: msync failed: %s",
@@ -467,7 +467,7 @@ 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++) {
@@ -495,8 +495,8 @@ static void _tdb_transaction_cancel(struct tdb_context *tdb)
                }
        }
 
-       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;
@@ -513,42 +513,38 @@ 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_LOG_USE_ERROR,
-                          "tdb_transaction_start: cannot start a transaction"
-                          " on a read-only or internal db");
-               return -1;
+               return 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_IO, TDB_LOG_USE_ERROR,
-                          "tdb_transaction_start:"
-                          " already inside transaction");
-               return -1;
+               return 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_LOG_USE_ERROR,
-                          "tdb_transaction_start: cannot start a transaction"
-                          " with locks held");
-               return -1;
+                  of nested locks in POSIX */
+               return 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_LOG_ERROR,
-                          "tdb_transaction_start: cannot allocate");
-               return -1;
+               return 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
@@ -556,36 +552,34 @@ int tdb_transaction_start(struct tdb_context *tdb)
           make this async, which we will probably do in the future */
        ecode = tdb_transaction_lock(tdb, F_WRLCK);
        if (ecode != TDB_SUCCESS) {
-               tdb->ecode = ecode;
                SAFE_FREE(tdb->transaction->blocks);
                SAFE_FREE(tdb->transaction);
-               return -1;
+               return ecode;
        }
 
        /* get a read lock over entire file. This is upgraded to a write
           lock during the commit */
        ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true);
        if (ecode != TDB_SUCCESS) {
-               tdb->ecode = ecode;
                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_SUCCESS;
 
 fail_allrecord_lock:
        tdb_transaction_unlock(tdb, F_WRLCK);
        SAFE_FREE(tdb->transaction->blocks);
        SAFE_FREE(tdb->transaction);
-       return -1;
+       return ecode;
 }
 
 
@@ -695,16 +689,16 @@ static enum TDB_ERROR tdb_recovery_allocate(struct tdb_context *tdb,
                = (((sizeof(rec) + *recovery_size) + getpagesize()-1)
                   & ~(getpagesize()-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;
+       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,
@@ -715,7 +709,7 @@ static enum TDB_ERROR tdb_recovery_allocate(struct tdb_context *tdb,
        /* 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 */
@@ -751,11 +745,12 @@ static void set_recovery_header(struct tdb_recovery_record *rec,
 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;
@@ -802,7 +797,7 @@ static enum TDB_ERROR transaction_setup_recovery(struct tdb_context *tdb,
                if (offset >= old_map_size) {
                        continue;
                }
-               if (offset + length > tdb->map_size) {
+               if (offset + length > tdb->file->map_size) {
                        free(data);
                        return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
                                          "tdb_transaction_setup_recovery:"
@@ -950,10 +945,12 @@ static enum TDB_ERROR _tdb_transaction_prepare_commit(struct tdb_context *tdb)
        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;
+               tdb->file->map_size = tdb->transaction->old_map_size;
                ecode = methods->expand_file(tdb, add);
                if (ecode != TDB_SUCCESS) {
                        tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
@@ -971,46 +968,42 @@ static enum TDB_ERROR _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 /*
    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)
 {
-       tdb->ecode = _tdb_transaction_prepare_commit(tdb);
-       if (tdb->ecode != TDB_SUCCESS)
-               return -1;
-       return 0;
+       return _tdb_transaction_prepare_commit(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_LOG_USE_ERROR,
-                        "tdb_transaction_commit: no transaction");
-               return -1;
+               return tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+                                 "tdb_transaction_commit: no transaction");
        }
 
        tdb_trace(tdb, "tdb_transaction_commit");
 
        if (tdb->transaction->nesting != 0) {
                tdb->transaction->nesting--;
-               return 0;
+               return TDB_SUCCESS;
        }
 
        /* check for a null transaction */
        if (tdb->transaction->blocks == NULL) {
                _tdb_transaction_cancel(tdb);
-               return 0;
+               return TDB_SUCCESS;
        }
 
        if (!tdb->transaction->prepared) {
-               tdb->ecode = _tdb_transaction_prepare_commit(tdb);
-               if (tdb->ecode != TDB_SUCCESS)
-                       return -1;
+               ecode = _tdb_transaction_prepare_commit(tdb);
+               if (ecode != TDB_SUCCESS)
+                       return ecode;
        }
 
        methods = tdb->transaction->io_methods;
@@ -1045,7 +1038,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 
                        _tdb_transaction_cancel(tdb);
 
-                       return -1;
+                       return ecode;
                }
                SAFE_FREE(tdb->transaction->blocks[i]);
        }
@@ -1054,10 +1047,9 @@ int tdb_transaction_commit(struct tdb_context *tdb)
        tdb->transaction->num_blocks = 0;
 
        /* ensure the new data is on disk */
-       ecode = transaction_sync(tdb, 0, tdb->map_size);
+       ecode = transaction_sync(tdb, 0, tdb->file->map_size);
        if (ecode != TDB_SUCCESS) {
-               tdb->ecode = ecode;
-               return -1;
+               return ecode;
        }
 
        /*
@@ -1079,7 +1071,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
           transaction locks */
        _tdb_transaction_cancel(tdb);
 
-       return 0;
+       return TDB_SUCCESS;
 }
 
 
@@ -1169,7 +1161,7 @@ enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb)
 
        free(data);
 
-       ecode = transaction_sync(tdb, 0, tdb->map_size);
+       ecode = transaction_sync(tdb, 0, tdb->file->map_size);
        if (ecode != TDB_SUCCESS) {
                return tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
                                  "tdb_transaction_recover:"