tdb: don't reduce file size on transaction recovery.

[ccan] / ccan / tdb / transaction.c

diff --git a/ccan/tdb/transaction.c b/ccan/tdb/transaction.c
index ded3fa836fc721a07932c54a6f741b89c8329066..6b539d4045ee6ef2e91f7943387e2492ffa681ad 100644 (file)
--- a/ccan/tdb/transaction.c
+++ b/ccan/tdb/transaction.c
@@ -466,8 +466,8 @@ static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t
        return 0;
 }
 
-/* ltype is F_WRLCK after prepare. */
-int _tdb_transaction_cancel(struct tdb_context *tdb, int ltype)
+
+static int _tdb_transaction_cancel(struct tdb_context *tdb)
 {
        int i, ret = 0;
 
@@ -510,7 +510,6 @@ int _tdb_transaction_cancel(struct tdb_context *tdb, int ltype)
        /* restore the normal io methods */
        tdb->methods = tdb->transaction->io_methods;
 
-       tdb_brunlock(tdb, ltype, FREELIST_TOP, 0);
        tdb_transaction_unlock(tdb, F_WRLCK);
        SAFE_FREE(tdb->transaction->hash_heads);
        SAFE_FREE(tdb->transaction);
@@ -583,10 +582,9 @@ int tdb_transaction_start(struct tdb_context *tdb)
        
        /* get a read lock from the freelist to the end of file. This
           is upgraded to a write lock during the commit */
-       if (tdb_brlock(tdb, F_RDLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT) == -1) {
+       if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true) == -1) {
                TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get hash locks\n"));
-               tdb->ecode = TDB_ERR_LOCK;
-               goto fail;
+               goto fail_allrecord_lock;
        }
 
        /* setup a copy of the hash table heads so the hash scan in
@@ -619,7 +617,8 @@ int tdb_transaction_start(struct tdb_context *tdb)
        return 0;
        
 fail:
-       tdb_brunlock(tdb, F_RDLCK, FREELIST_TOP, 0);
+       tdb_allrecord_unlock(tdb, F_RDLCK, false);
+fail_allrecord_lock:
        tdb_transaction_unlock(tdb, F_WRLCK);
        SAFE_FREE(tdb->transaction->blocks);
        SAFE_FREE(tdb->transaction->hash_heads);
@@ -633,11 +632,8 @@ fail:
 */
 int tdb_transaction_cancel(struct tdb_context *tdb)
 {
-       int ltype = F_RDLCK;
        tdb_trace(tdb, "tdb_transaction_cancel");
-       if (tdb->transaction && tdb->transaction->prepared)
-               ltype = F_WRLCK;
-       return _tdb_transaction_cancel(tdb, ltype);
+       return _tdb_transaction_cancel(tdb);
 }
 
 /*
@@ -687,10 +683,16 @@ 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);
@@ -897,14 +899,14 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
 
        if (tdb->transaction->prepared) {
                tdb->ecode = TDB_ERR_EINVAL;
-               _tdb_transaction_cancel(tdb, F_WRLCK);
+               _tdb_transaction_cancel(tdb);
                TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction already prepared\n"));
                return -1;
        }
 
        if (tdb->transaction->transaction_error) {
                tdb->ecode = TDB_ERR_IO;
-               _tdb_transaction_cancel(tdb, F_RDLCK);
+               _tdb_transaction_cancel(tdb);
                TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction error pending\n"));
                return -1;
        }
@@ -927,15 +929,14 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
        if (tdb_have_extra_locks(tdb)) {
                tdb->ecode = TDB_ERR_LOCK;
                TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: locks pending on commit\n"));
-               _tdb_transaction_cancel(tdb, F_RDLCK);
+               _tdb_transaction_cancel(tdb);
                return -1;
        }
 
        /* upgrade the main transaction lock region to a write lock */
-       if (tdb_brlock_upgrade(tdb, FREELIST_TOP, 0) == -1) {
+       if (tdb_allrecord_upgrade(tdb) == -1) {
                TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: failed to upgrade hash locks\n"));
-               tdb->ecode = TDB_ERR_LOCK;
-               _tdb_transaction_cancel(tdb, F_RDLCK);
+               _tdb_transaction_cancel(tdb);
                return -1;
        }
 
@@ -943,7 +944,7 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
           during the commit */
        if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
                TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: failed to get open lock\n"));
-               _tdb_transaction_cancel(tdb, F_WRLCK);
+               _tdb_transaction_cancel(tdb);
                return -1;
        }
 
@@ -951,7 +952,7 @@ static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
                /* 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_transaction_cancel(tdb, F_WRLCK);
+                       _tdb_transaction_cancel(tdb);
                        return -1;
                }
        }
@@ -965,7 +966,7 @@ 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_transaction_cancel(tdb, F_WRLCK);
+                       _tdb_transaction_cancel(tdb);
                        return -1;
                }
                tdb->map_size = tdb->transaction->old_map_size;
@@ -1017,7 +1018,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 
        /* check for a null transaction */
        if (tdb->transaction->blocks == NULL) {
-               _tdb_transaction_cancel(tdb, F_RDLCK);
+               _tdb_transaction_cancel(tdb);
                return 0;
        }
 
@@ -1053,7 +1054,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
                        tdb->methods = methods;
                        tdb_transaction_recover(tdb); 
 
-                       _tdb_transaction_cancel(tdb, F_WRLCK);
+                       _tdb_transaction_cancel(tdb);
 
                        TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
                        return -1;
@@ -1088,7 +1089,7 @@ int tdb_transaction_commit(struct tdb_context *tdb)
 
        /* use a transaction cancel to free memory and remove the
           transaction locks */
-       _tdb_transaction_cancel(tdb, F_WRLCK);
+       _tdb_transaction_cancel(tdb);
 
        if (need_repack) {
                return tdb_repack(tdb);
@@ -1202,16 +1203,6 @@ int tdb_transaction_recover(struct tdb_context *tdb)
                return -1;                      
        }
        
-       /* reduce the file size to the old size */
-       tdb_munmap(tdb);
-       if (ftruncate(tdb->fd, recovery_eof) != 0) {
-               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to reduce to recovery size\n"));
-               tdb->ecode = TDB_ERR_IO;
-               return -1;                      
-       }
-       tdb->map_size = recovery_eof;
-       tdb_mmap(tdb);
-
        if (transaction_sync(tdb, 0, recovery_eof) == -1) {
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync2 recovery\n"));
                tdb->ecode = TDB_ERR_IO;