io_uring: mark reissue requests with REQ_F_PARTIAL_IO
authorJens Axboe <axboe@kernel.dk>
Mon, 20 Jun 2022 12:39:27 +0000 (06:39 -0600)
committerJens Axboe <axboe@kernel.dk>
Mon, 20 Jun 2022 12:39:27 +0000 (06:39 -0600)
If we mark for reissue, we assume that the buffer will remain stable.
Hence if are using a provided buffer, we need to ensure that we stick
with it for the duration of that request.

This only affects block devices that use provided buffers, as those are
the only ones that get marked with REQ_F_REISSUE.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c

index d3ee4fc..87c65a3 100644 (file)
@@ -3437,7 +3437,7 @@ static bool __io_complete_rw_common(struct io_kiocb *req, long res)
        if (unlikely(res != req->cqe.res)) {
                if ((res == -EAGAIN || res == -EOPNOTSUPP) &&
                    io_rw_should_reissue(req)) {
-                       req->flags |= REQ_F_REISSUE;
+                       req->flags |= REQ_F_REISSUE | REQ_F_PARTIAL_IO;
                        return true;
                }
                req_set_fail(req);
@@ -3487,7 +3487,7 @@ static void io_complete_rw_iopoll(struct kiocb *kiocb, long res)
                kiocb_end_write(req);
        if (unlikely(res != req->cqe.res)) {
                if (res == -EAGAIN && io_rw_should_reissue(req)) {
-                       req->flags |= REQ_F_REISSUE;
+                       req->flags |= REQ_F_REISSUE | REQ_F_PARTIAL_IO;
                        return;
                }
                req->cqe.res = res;