blktrace: ignore 0 byte writes

They cause fio to exit replay, but then retry since we didn't meet
our IO limit. When we retry, we attempt to close files that the
trace already closed, and we crash.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/blktrace.c b/blktrace.c
index 1d5403a5a0fc59cad41278ed8a368033c733fcf0..721bcd4bc422a93f7333ef54185f000c6853f571 100644 (file)
--- a/blktrace.c
+++ b/blktrace.c
@@ -310,7 +310,13 @@ static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t,
 
        rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
 
-       if (t->bytes > rw_bs[rw])
+       /*
+        * Need to figure out why 0 byte writes end up here sometimes, for
+        * now just ignore them.
+        */
+       if (!t->bytes)
+               return;
+       else if (t->bytes > rw_bs[rw])
                rw_bs[rw] = t->bytes;
 
        ios[rw]++;