[lvc-project] [PATCH] net: kcm: use previously opened message only once

[Jakub Kicinski]

On Thu,  1 Aug 2024 16:08:33 +0300 Dmitry Antipov wrote:
> When syzkaller reproducer injects 'alloc_skb()' failure at line
> 817, 'kcm_sendmsg()' may return with partial message saved at
> 'kcm->seq_skb'. Next call of this function will try to build the
> next message starting from the saved one, but should do it only
> once. Otherwise a complete mess in skb management causes an
> undefined behavior of any kind, including UAFs reported by KASAN.
> 
> Fixes: ab7ac4eb9832 ("kcm: Kernel Connection Multiplexor module")
> Reported-by: syzbot+b72d86aa5df17ce74c60 at syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=b72d86aa5df17ce74c60
> Signed-off-by: Dmitry Antipov <dmantipov at yandex.ru>
> ---
>  net/kcm/kcmsock.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/net/kcm/kcmsock.c b/net/kcm/kcmsock.c
> index 2f191e50d4fc..fa5ce5c88045 100644
> --- a/net/kcm/kcmsock.c
> +++ b/net/kcm/kcmsock.c
> @@ -766,6 +766,8 @@ static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
>  	if (kcm->seq_skb) {
>  		/* Previously opened message */
>  		head = kcm->seq_skb;
> +		/* ...should be used only once */
> +		kcm->seq_skb = NULL;
>  		skb = kcm_tx_msg(head)->last_skb;
>  		goto start;
>  	}

Not sure how much this matters but if we clear seq_skb then handling
here:
https://elixir.bootlin.com/linux/v6.10-rc4/source/net/kcm/kcmsock.c#L940
will work differently.