2022-02-11 15:50:01

by Stefano Garzarella

[permalink] [raw]
Subject: [PATCH] block: clear iocb->private in blkdev_bio_end_io_async()

iocb_bio_iopoll() expects iocb->private to be cleared before
releasing the bio.

We already do this in blkdev_bio_end_io(), but we forgot in the
recently added blkdev_bio_end_io_async().

Fixes: 54a88eb838d3 ("block: add single bio async direct IO helper")
Cc: [email protected]
Signed-off-by: Stefano Garzarella <[email protected]>
---
I haven't seen a failure, I was just reading the code to understand iopoll,
so IIUC we should clean iocb->private in blkdev_bio_end_io_async().

Thanks,
Stefano
---
block/fops.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/block/fops.c b/block/fops.c
index 4f59e0f5bf30..a18e7fbd97b8 100644
--- a/block/fops.c
+++ b/block/fops.c
@@ -289,6 +289,8 @@ static void blkdev_bio_end_io_async(struct bio *bio)
struct kiocb *iocb = dio->iocb;
ssize_t ret;

+ WRITE_ONCE(iocb->private, NULL);
+
if (likely(!bio->bi_status)) {
ret = dio->size;
iocb->ki_pos += ret;
--
2.34.1


2022-02-22 04:33:33

by Ming Lei

[permalink] [raw]
Subject: Re: [PATCH] block: clear iocb->private in blkdev_bio_end_io_async()

On Fri, Feb 11, 2022 at 10:01:36AM +0100, Stefano Garzarella wrote:
> iocb_bio_iopoll() expects iocb->private to be cleared before
> releasing the bio.
>
> We already do this in blkdev_bio_end_io(), but we forgot in the

But also iomap_dio_bio_end_io().

> recently added blkdev_bio_end_io_async().
>
> Fixes: 54a88eb838d3 ("block: add single bio async direct IO helper")
> Cc: [email protected]
> Signed-off-by: Stefano Garzarella <[email protected]>
> ---
> I haven't seen a failure, I was just reading the code to understand iopoll,
> so IIUC we should clean iocb->private in blkdev_bio_end_io_async().
>
> Thanks,
> Stefano
> ---
> block/fops.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/block/fops.c b/block/fops.c
> index 4f59e0f5bf30..a18e7fbd97b8 100644
> --- a/block/fops.c
> +++ b/block/fops.c
> @@ -289,6 +289,8 @@ static void blkdev_bio_end_io_async(struct bio *bio)
> struct kiocb *iocb = dio->iocb;
> ssize_t ret;
>
> + WRITE_ONCE(iocb->private, NULL);

It might cause race in case of concurrent polling on same queue without
clearing iocb->private, so looks fine:

Reviewed-by: Ming Lei <[email protected]>

Thanks,
Ming

2022-02-22 05:30:52

by Stefano Garzarella

[permalink] [raw]
Subject: Re: [PATCH] block: clear iocb->private in blkdev_bio_end_io_async()

Gentle ping :-)


On Fri, Feb 11, 2022 at 10:01 AM Stefano Garzarella <[email protected]> wrote:
>
> iocb_bio_iopoll() expects iocb->private to be cleared before
> releasing the bio.
>
> We already do this in blkdev_bio_end_io(), but we forgot in the
> recently added blkdev_bio_end_io_async().
>
> Fixes: 54a88eb838d3 ("block: add single bio async direct IO helper")
> Cc: [email protected]
> Signed-off-by: Stefano Garzarella <[email protected]>
> ---
> I haven't seen a failure, I was just reading the code to understand iopoll,
> so IIUC we should clean iocb->private in blkdev_bio_end_io_async().
>
> Thanks,
> Stefano
> ---
> block/fops.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/block/fops.c b/block/fops.c
> index 4f59e0f5bf30..a18e7fbd97b8 100644
> --- a/block/fops.c
> +++ b/block/fops.c
> @@ -289,6 +289,8 @@ static void blkdev_bio_end_io_async(struct bio *bio)
> struct kiocb *iocb = dio->iocb;
> ssize_t ret;
>
> + WRITE_ONCE(iocb->private, NULL);
> +
> if (likely(!bio->bi_status)) {
> ret = dio->size;
> iocb->ki_pos += ret;
> --
> 2.34.1
>

2022-02-22 08:34:30

by Christoph Hellwig

[permalink] [raw]

2022-02-22 15:01:40

by Jens Axboe

[permalink] [raw]
Subject: Re: [PATCH] block: clear iocb->private in blkdev_bio_end_io_async()

On Fri, 11 Feb 2022 10:01:36 +0100, Stefano Garzarella wrote:
> iocb_bio_iopoll() expects iocb->private to be cleared before
> releasing the bio.
>
> We already do this in blkdev_bio_end_io(), but we forgot in the
> recently added blkdev_bio_end_io_async().
>
>
> [...]

Applied, thanks!

[1/1] block: clear iocb->private in blkdev_bio_end_io_async()
commit: bb49c6fa8b845591b317b0d7afea4ae60ec7f3aa

Best regards,
--
Jens Axboe