linux-kernel - Re: [PATCH] The vdpasim_net_work function handles both transmit TX completion and RX processing. A descriptor is taken from the TX VQ at the start of the loop, representing a buffer previously used for the TX path's initial read from the virtual network device. This TX descriptor must be completed back to the Rx queue at the end of the work loop.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CACGkMEtc3KdQtyxUnVeBG6FrbULGanmJxBwNhmG4DB1pO7zn=g@mail.gmail.com>
Date: Wed, 17 Dec 2025 10:51:50 +0800
From: Jason Wang <jasowang@...hat.com>
To: Angus Chen <angus.chen@...uarmicro.com>
Cc: mst@...hat.com, xuanzhuo@...ux.alibaba.com, eperezma@...hat.com, 
	linux-kernel@...r.kernel.org, virtualization@...ts.linux.dev
Subject: Re: [PATCH] The vdpasim_net_work function handles both transmit TX
 completion and RX processing. A descriptor is taken from the TX VQ at the
 start of the loop, representing a buffer previously used for the TX path's
 initial read from the virtual network device. This TX descriptor must be
 completed back to the Rx queue at the end of the work loop.

On Wed, Dec 17, 2025 at 10:30 AM Angus Chen <angus.chen@...uarmicro.com> wrote:
>

It looks like the title is too long.

> However, the error handling logic was flawed:
>
> 1. Read Failure (err <= 0): The TX VQ completion
>    (vdpasim_net_complete(txq, 0)) was performed, but the unconditional
>    break that followed prevented theprocessing of subsequent pending work
>    items (other descriptors) in the queue in the same work function call.
>
> 2. Write Failure (write <= 0): If data could not be written to the RX VQ
>    descriptor (e.g., vringh_iov_push_iotlb failed), the flow would 'break'
>    before the TX VQ descriptor was completed. This led to a leak of TX VQ
>    descriptors, as the work was never notified that the TX buffer was
>    free to use again.
>
> This commit refactors the control flow to ensure:
> a) The TX VQ descriptor is completed back to the rx in all error and
>    success paths related to the current descriptor processing.
> b) The unconditional break on read failure is removed, allowing the
>    function to continue processing remaining work items
>    until the loop naturally exits.
>
> Fixes: 0899774cb360 ("vdpa_sim_net: vendor satistics")

This is suspicious, the logic was there since introduced? Or we can
split this patch

1) fix the stats
2) fix the error handling

> ---
>
> Signed-off-by: Angus Chen <angus.chen@...uarmicro.com>
> ---
>  drivers/vdpa/vdpa_sim/vdpa_sim_net.c | 28 ++++++++++++++--------------
>  1 file changed, 14 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
> index 6caf09a1907b..298bd6e8e0a0 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
> @@ -242,22 +242,22 @@ static void vdpasim_net_work(struct vdpasim *vdpasim)
>                 if (err <= 0) {
>                         ++rx_overruns;
>                         vdpasim_net_complete(txq, 0);
> -                       break;
> -               }
> -
> -               write = vringh_iov_push_iotlb(&rxq->vring, &rxq->in_iov,
> -                                             net->buffer, read);
> -               if (write <= 0) {
> -                       ++rx_errors;
> -                       break;
> +               } else {

This else seem useless?

> +
> +                       write = vringh_iov_push_iotlb(&rxq->vring, &rxq->in_iov,
> +                                                     net->buffer, read);
> +                       if (write <= 0) {
> +                               ++rx_errors;
> +                               vdpasim_net_complete(txq, 0);

I'd move this after:

                read = vringh_iov_pull_iotlb(&txq->vring, &txq->out_iov,
                                             net->buffer, PAGE_SIZE);

So we would have a single vdpasim_net_complete()

> +                               break;
> +                       }
> +
> +                       ++rx_pkts;
> +                       rx_bytes += write;
> +                       vdpasim_net_complete(txq, 0);
> +                       vdpasim_net_complete(rxq, write);
>                 }
>
> -               ++rx_pkts;
> -               rx_bytes += write;
> -
> -               vdpasim_net_complete(txq, 0);
> -               vdpasim_net_complete(rxq, write);
> -
>                 if (tx_pkts > 4) {
>                         vdpasim_schedule_work(vdpasim);
>                         goto out;
> --
> 2.34.1
>

Thanks