[PATCH v2 1/2] vhost: fix possible FDs leak

[Maxime Coquelin]

On 1/30/23 15:25, David Marchand wrote:
> On Mon, Jan 30, 2023 at 10:46 AM Maxime Coquelin
> <maxime.coquelin at redhat.com> wrote:
>>
>>
>>
>> On 1/29/23 10:25, David Marchand wrote:
>>> On Fri, Jan 27, 2023 at 5:55 PM Maxime Coquelin
>>> <maxime.coquelin at redhat.com> wrote:
>>>>
>>>> On failure, read_vhost_message() only closed the message
>>>> FDs if the header size was unexpected, but there are other
>>>> cases where it is required. For exemple in the case the
>>>> payload size read from the header is greater than the
>>>> expected maximum payload size.
>>>>
>>>> This patch fixes this by closing all messages FDs in all
>>>> error cases.
>>>>
>>>> Fixes: bf472259dde6 ("vhost: fix possible denial of service by leaking FDs")
>>>> Cc: stable at dpdk.org
>>>>
>>>> Signed-off-by: Maxime Coquelin <maxime.coquelin at redhat.com>
>>>
>>> Reviewed-by: David Marchand <david.marchand at redhat.com>
>>>
>>> We mentionned offlist that the request type can be logged to help with debug.
>>> Do you intend to add this as a follow up patch?
>>>
>>>
>>
>> Thinking about it, that's not that trivial because read_vhost_message()
>> is called for both master and slave channels, so we would need to
>> differentiate between both to print proper request name.
>>
>> It is doable by comparing the file descriptor passed as parameter with
>> the slave one stored in struct virtio_net, but that's not super clean.
> 
>  From the 3 different callers of read_vhost_message, I think the only
> one that could gain from this debug is vhost_user_msg_handler().
> 
> And here, we could look at the message content on read_vhost_message return.
> Like this untested and ugly snippet:
> 
> diff --git a/lib/vhost/vhost_user.c b/lib/vhost/vhost_user.c
> index 943058725e..1556f21a58 100644
> --- a/lib/vhost/vhost_user.c
> +++ b/lib/vhost/vhost_user.c
> @@ -2994,13 +2994,10 @@ vhost_user_msg_handler(int vid, int fd)
>                  }
>          }
> 
> +       ctx.msg.request.master = VHOST_USER_NONE;
>          ret = read_vhost_message(dev, fd, &ctx);
> -       if (ret <= 0) {
> -               if (ret < 0)
> -                       VHOST_LOG_CONFIG(dev->ifname, ERR, "vhost read
> message failed\n");
> -               else
> -                       VHOST_LOG_CONFIG(dev->ifname, INFO, "vhost
> peer closed\n");
> -
> +       if (ret == 0) {
> +               VHOST_LOG_CONFIG(dev->ifname, INFO, "vhost peer closed\n");
>                  return -1;
>          }
> 
> @@ -3010,6 +3007,14 @@ vhost_user_msg_handler(int vid, int fd)
>          else
>                  msg_handler = NULL;
> 
> +       if (ret < 0) {
> +               VHOST_LOG_CONFIG(dev->ifname, ERR, "vhost read message
> %s%s%sfailed\n",
> +                       msg_handler != NULL ? "for " : "",
> +                       msg_handler != NULL ? msg_handler->description : "",
> +                       msg_handler != NULL ? " " : "");
> +               return -1;
> +       }
> +
>          if (msg_handler != NULL && msg_handler->description != NULL) {
>                  if (request != VHOST_USER_IOTLB_MSG)
>                          VHOST_LOG_CONFIG(dev->ifname, INFO,
> 
> 

It looks good to me. I'm squashing this with patch 1, adding you SoB.

Thanks,
Maxime