tdb: Apply some const, just for clarity
[samba.git] / lib / tdb / common / transaction.c
index 16e12d647043f9df6c0387dc0189165c20232f74..ccb4d955a329e82f6d30607d74321c2af64b1bc1 100644 (file)
     still available, but no transaction recovery area is used and no
     fsync/msync calls are made.
 
+  - if TDB_ALLOW_NESTING is passed to flags in tdb open, or added using
+    tdb_add_flags() transaction nesting is enabled.
+    It resets the TDB_DISALLOW_NESTING flag, as both cannot be used together.
+    The default is that transaction nesting is allowed.
+    Note: this default may change in future versions of tdb.
+
+    Beware. when transactions are nested a transaction successfully
+    completed with tdb_transaction_commit() can be silently unrolled later.
+
+  - if TDB_DISALLOW_NESTING is passed to flags in tdb open, or added using
+    tdb_add_flags() transaction nesting is disabled.
+    It resets the TDB_ALLOW_NESTING flag, as both cannot be used together.
+    An attempt create a nested transaction will fail with TDB_ERR_NESTING.
+    The default is that transaction nesting is allowed.
+    Note: this default may change in future versions of tdb.
 */
 
 
@@ -120,6 +135,9 @@ struct tdb_transaction {
        bool prepared;
        tdb_off_t magic_offset;
 
+       /* set when the GLOBAL_LOCK has been taken */
+       bool global_lock_taken;
+
        /* old file size before transaction */
        tdb_len_t old_map_size;
 
@@ -427,6 +445,10 @@ int tdb_transaction_start(struct tdb_context *tdb)
 
        /* cope with nested tdb_transaction_start() calls */
        if (tdb->transaction != NULL) {
+               if (!(tdb->flags & TDB_ALLOW_NESTING)) {
+                       tdb->ecode = TDB_ERR_NESTING;
+                       return -1;
+               }
                tdb->transaction->nesting++;
                TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_start: nesting %d\n", 
                         tdb->transaction->nesting));
@@ -574,16 +596,21 @@ int _tdb_transaction_cancel(struct tdb_context *tdb)
 
        if (tdb->transaction->magic_offset) {
                const struct tdb_methods *methods = tdb->transaction->io_methods;
-               uint32_t zero = 0;
+               const uint32_t invalid = TDB_RECOVERY_INVALID_MAGIC;
 
                /* remove the recovery marker */
-               if (methods->tdb_write(tdb, tdb->transaction->magic_offset, &zero, 4) == -1 ||
+               if (methods->tdb_write(tdb, tdb->transaction->magic_offset, &invalid, 4) == -1 ||
                transaction_sync(tdb, tdb->transaction->magic_offset, 4) == -1) {
                        TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_cancel: failed to remove recovery magic\n"));
                        ret = -1;
                }
        }
 
+       if (tdb->transaction->global_lock_taken) {
+               tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+               tdb->transaction->global_lock_taken = false;
+       }
+
        /* remove any global lock created during the transaction */
        if (tdb->global_lock.count != 0) {
                tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 4*tdb->header.hash_size);
@@ -657,7 +684,7 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
                                 tdb_off_t *recovery_offset,
                                 tdb_len_t *recovery_max_size)
 {
-       struct list_struct rec;
+       struct tdb_record rec;
        const struct tdb_methods *methods = tdb->transaction->io_methods;
        tdb_off_t recovery_head;
 
@@ -668,10 +695,16 @@ static int tdb_recovery_allocate(struct tdb_context *tdb,
 
        rec.rec_len = 0;
 
-       if (recovery_head != 0 && 
-           methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
-               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery record\n"));
-               return -1;
+       if (recovery_head != 0) {
+               if (methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
+                       TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery record\n"));
+                       return -1;
+               }
+               /* ignore invalid recovery regions: can happen in crash */
+               if (rec.magic != TDB_RECOVERY_MAGIC &&
+                   rec.magic != TDB_RECOVERY_INVALID_MAGIC) {
+                       recovery_head = 0;
+               }
        }
 
        *recovery_size = tdb_recovery_size(tdb);
@@ -743,7 +776,7 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
        tdb_len_t recovery_size;
        unsigned char *data, *p;
        const struct tdb_methods *methods = tdb->transaction->io_methods;
-       struct list_struct *rec;
+       struct tdb_record *rec;
        tdb_off_t recovery_offset, recovery_max_size;
        tdb_off_t old_map_size = tdb->transaction->old_map_size;
        uint32_t magic, tailer;
@@ -763,10 +796,10 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
                return -1;
        }
 
-       rec = (struct list_struct *)data;
+       rec = (struct tdb_record *)data;
        memset(rec, 0, sizeof(*rec));
 
-       rec->magic    = 0;
+       rec->magic    = TDB_RECOVERY_INVALID_MAGIC;
        rec->data_len = recovery_size;
        rec->rec_len  = recovery_max_size;
        rec->key_len  = old_map_size;
@@ -846,7 +879,7 @@ static int transaction_setup_recovery(struct tdb_context *tdb,
        magic = TDB_RECOVERY_MAGIC;
        CONVERT(magic);
 
-       *magic_offset = recovery_offset + offsetof(struct list_struct, magic);
+       *magic_offset = recovery_offset + offsetof(struct tdb_record, magic);
 
        if (methods->tdb_write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery magic\n"));
@@ -928,11 +961,12 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
                return -1;
        }
 
+       tdb->transaction->global_lock_taken = true;
+
        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_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: failed to setup recovery data\n"));
-                       tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
                        _tdb_transaction_cancel(tdb);
                        return -1;
                }
@@ -947,7 +981,6 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
                                             tdb->transaction->old_map_size) == -1) {
                        tdb->ecode = TDB_ERR_IO;
                        TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: expansion failed\n"));
-                       tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
                        _tdb_transaction_cancel(tdb);
                        return -1;
                }
@@ -1037,7 +1070,6 @@ int tdb_transaction_commit(struct tdb_context *tdb)
                        tdb_transaction_recover(tdb); 
 
                        _tdb_transaction_cancel(tdb);
-                       tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
 
                        TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
                        return -1;
@@ -1053,8 +1085,6 @@ int tdb_transaction_commit(struct tdb_context *tdb)
                return -1;
        }
 
-       tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
-
        /*
          TODO: maybe write to some dummy hdr field, or write to magic
          offset without mmap, before the last sync, instead of the
@@ -1094,7 +1124,7 @@ int tdb_transaction_recover(struct tdb_context *tdb)
        tdb_off_t recovery_head, recovery_eof;
        unsigned char *data, *p;
        uint32_t zero = 0;
-       struct list_struct rec;
+       struct tdb_record rec;
 
        /* find the recovery area */
        if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
@@ -1181,7 +1211,7 @@ int tdb_transaction_recover(struct tdb_context *tdb)
        }
 
        /* remove the recovery magic */
-       if (tdb_ofs_write(tdb, recovery_head + offsetof(struct list_struct, magic), 
+       if (tdb_ofs_write(tdb, recovery_head + offsetof(struct tdb_record, magic),
                          &zero) == -1) {
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery magic\n"));
                tdb->ecode = TDB_ERR_IO;