blk-wbt: account flush requests correctly

Mikulas reported a workload that saw bad performance, and figured
out what it was due to various other types of requests being
accounted as reads. Flush requests, for instance. Due to the
high latency of those, we heavily throttle the writes to keep
the latencies in balance. But they really should be accounted
as writes.

Fix this by checking the exact type of the request. If it's a
read, account as a read, if it's a write or a flush, account
as a write. Any other request we disregard. Previously everything
would have been mistakenly accounted as reads.

Reported-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/block/blk-wbt.c b/block/blk-wbt.c
index ae8de9780085ae7b8e99237ed16fc9cd02b233a5..f92fc84b5e2c497b77578408c10c22bf37929384 100644 (file)
--- a/block/blk-wbt.c
+++ b/block/blk-wbt.c
@@ -697,7 +697,15 @@ u64 wbt_default_latency_nsec(struct request_queue *q)
 
 static int wbt_data_dir(const struct request *rq)
 {
-       return rq_data_dir(rq);
+       const int op = req_op(rq);
+
+       if (op == REQ_OP_READ)
+               return READ;
+       else if (op == REQ_OP_WRITE || op == REQ_OP_FLUSH)
+               return WRITE;
+
+       /* don't account */
+       return -1;
 }
 
 int wbt_init(struct request_queue *q)