On Thu, 2022-07-21 at 20:08 +0700, Ammar Faizi wrote:
> On 7/21/22 7:05 PM, Dylan Yudaken wrote:
> > It seems to be a problem with blocking reads, buffer select and
> > READV.
> > My guess is that ext4/xfs are not blocking.
> >
> > in b66e65f41426 ("io_uring: never call io_buffer_select() for a
> > buffer
> > re-select"), this line was added in __io_iov_buffer_select
> >
> > - iov[0].iov_len = len;
> > + req->rw.len = iov[0].iov_len = len;
> >
> > Basically stashing the buffer length in rw.len. The problem is that
> > the
> > next time around that breaks at
> >
> > if (req->rw.len != 1)
> > return -EINVAL;
> >
> >
> > The below fixes it as an example, but it's not great. Maybe someone
> > can
> > figure out a better patch? Otherwise I can try tomorrow:
>
> It's 8:05 PM from my end. I'll try to play with your patch after
> dinner
> while waiting for others say something.
>
I've just sent the below actually which is a bit simpler. I reran all
the tests on btrfs and xfs and it seems to work now:
diff --git a/fs/io_uring.c b/fs/io_uring.c
index a01ea49f3017..b0180679584f 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1737,6 +1737,14 @@ static void io_kbuf_recycle(struct io_kiocb
*req, unsigned issue_flags)
(req->flags & REQ_F_PARTIAL_IO))
return;
+ /*
+ * READV uses fields in `struct io_rw` (len/addr) to stash the
selected
+ * buffer data. However if that buffer is recycled the original
request
+ * data stored in addr is lost. Therefore forbid recycling for
now.
+ */
+ if (req->opcode == IORING_OP_READV)
+ return;
+
/*
* We don't need to recycle for REQ_F_BUFFER_RING, we can just
clear
* the flag and hence ensure that bl->head doesn't get
incremented.
