Re: [syzbot] [net?] WARNING in inet_sock_destruct (4)

[Muhammad Usama Anjum <usama.anjum@collabora.com>]

Hi Eric,

On 8/29/23 8:19 PM, Eric Dumazet wrote:
> On Tue, Aug 29, 2023 at 2:44 PM Muhammad Usama Anjum
> <usama.anjum@collabora.com> wrote:
>>
>> On 6/23/23 7:36 PM, syzbot wrote:
>>> Hello,
>>>
>>> syzbot found the following issue on:
>>>
>>> HEAD commit: 45a3e24f65e9 Linux 6.4-rc7
>>> git tree: upstream
>>> console output: https://syzkaller.appspot.com/x/log.txt?x=160cc82f280000
>>> kernel config: https://syzkaller.appspot.com/x/.config?x=2cbd298d0aff1140
>>> dashboard link: https://syzkaller.appspot.com/bug?extid=de6565462ab540f50e47
>>> compiler: gcc (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils for Debian) 2.35.2
>>> syz repro: https://syzkaller.appspot.com/x/repro.syz?x=160aacb7280000
>>> C reproducer: https://syzkaller.appspot.com/x/repro.c?x=17c115d3280000
>>>
>>> Downloadable assets:
>>> disk image: https://storage.googleapis.com/syzbot-assets/c09bcd4ec365/disk-45a3e24f.raw.xz
>>> vmlinux: https://storage.googleapis.com/syzbot-assets/03549b639718/vmlinux-45a3e24f.xz
>>> kernel image: https://storage.googleapis.com/syzbot-assets/91f203e5f63e/bzImage-45a3e24f.xz
>>>
>>> The issue was bisected to:
>>>
>>> commit 565b4824c39fa335cba2028a09d7beb7112f3c9a
>>> Author: Jiri Pirko <jiri@nvidia.com>
>>> Date: Mon Feb 6 09:41:51 2023 +0000
>>>
>>> devlink: change port event netdev notifier from per-net to global
>>>
>>> bisection log: https://syzkaller.appspot.com/x/bisect.txt?x=110a1a5b280000
>>> final oops: https://syzkaller.appspot.com/x/report.txt?x=130a1a5b280000
>>> console output: https://syzkaller.appspot.com/x/log.txt?x=150a1a5b280000
>>>
>>> IMPORTANT: if you fix the issue, please add the following tag to the commit:
>>> Reported-by: syzbot+de6565462ab540f50e47@syzkaller.appspotmail.com
>>> Fixes: 565b4824c39f ("devlink: change port event netdev notifier from per-net to global")
>>>
>>> ------------[ cut here ]------------
>>> WARNING: CPU: 0 PID: 5025 at net/ipv4/af_inet.c:154 inet_sock_destruct+0x6df/0x8a0 net/ipv4/af_inet.c:154
>> This same warning has been spotted and reported:
>> https://bugzilla.kernel.org/show_bug.cgi?id=217555
>>
>> Syzbot has found the same warning on 4.14, 5.15, 6.1, 6.5-rc and latest
>> mainline (1c59d383390f9) kernels. The provided reproducers (such as
>> https://syzkaller.appspot.com/text?tag=ReproC&x=15a10e8aa80000) are
>> reproducing the same warnings on multicore (at least 2 CPUs) qemu instance.
> 
> Can you test the following fix ?
Just tested the fix on 1c59d383390f9, it didn't fix the warning.

Please let me know if you need help in testing more.

> Thanks.
> 
> diff --git a/net/dccp/ipv6.c b/net/dccp/ipv6.c
> index 25816e790527dbd6ff55ffb94762b5974e8144aa..1085357b30c9a0d4bf7a578cebf3eeddec953632
> 100644
> --- a/net/dccp/ipv6.c
> +++ b/net/dccp/ipv6.c
> @@ -377,8 +377,13 @@ static int dccp_v6_conn_request(struct sock *sk,
> struct sk_buff *skb)
>         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
>             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
>             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
> +               /* Only initialize ireq->pktops once.
> +                * We must take a refcount on skb because ireq->pktops
> +                * could be consumed immediately.
> +                */
>                 refcount_inc(&skb->users);
> -               ireq->pktopts = skb;
> +               if (cmpxchg(&ireq->pktopts, NULL, skb))
> +                       refcount_dec(&skb->users);
>         }
>         ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
> 
> diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
> index 6e86721e1cdbb8d47b754a2675f6ab1643c7342c..d45aa267473c4ab817cfda06966a536718b50a53
> 100644
> --- a/net/ipv6/tcp_ipv6.c
> +++ b/net/ipv6/tcp_ipv6.c
> @@ -798,8 +798,13 @@ static void tcp_v6_init_req(struct request_sock *req,
>              np->rxopt.bits.rxinfo ||
>              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
>              np->rxopt.bits.rxohlim || np->repflow)) {
> +               /* Only initialize ireq->pktops once.
> +                * We must take a refcount on skb because ireq->pktops
> +                * could be consumed immediately.
> +                */
>                 refcount_inc(&skb->users);
> -               ireq->pktopts = skb;
> +               if (cmpxchg(&ireq->pktopts, NULL, skb))
> +                       refcount_dec(&skb->users);
>         }
>  }

-- 
BR,
Muhammad Usama Anjum