Message-ID: <bd417bb2-212b-4657-a509-e0ed0edb5647@kernel.org>
Date: Wed, 26 Nov 2025 17:36:39 +0100
From: Matthieu Baerts <matttbe@...nel.org>
To: Kuniyuki Iwashima <kuniyu@...gle.com>
Cc: Geliang Tang <geliang@...nel.org>, Simon Horman <horms@...nel.org>,
 Kuniyuki Iwashima <kuni1840@...il.com>, netdev@...r.kernel.org,
 syzbot+3a92d359bc2ec6255a33@...kaller.appspotmail.com,
 Mat Martineau <martineau@...nel.org>, "David S. Miller"
 <davem@...emloft.net>, Eric Dumazet <edumazet@...gle.com>,
 Jakub Kicinski <kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>,
 MPTCP Linux <mptcp@...ts.linux.dev>
Subject: Re: [PATCH v1 net-next] mptcp: Initialise rcv_mss before calling
 tcp_send_active_reset() in mptcp_do_fastclose().

Hi Kuniyuki,

(+Cc MPTCP ML)

On 25/11/2025 20:53, Kuniyuki Iwashima wrote:
> syzbot reported divide-by-zero in __tcp_select_window() by
> MPTCP socket. [0]

Thank you for having released this bug report, and even more for having
sent the fix!

> We had a similar issue for the bare TCP and fixed in commit
> 499350a5a6e7 ("tcp: initialize rcv_mss to TCP_MIN_MSS instead
> of 0").
> 
> Let's apply the same fix to mptcp_do_fastclose().
> 
> [0]:
> Oops: divide error: 0000 [#1] SMP KASAN PTI
> CPU: 0 UID: 0 PID: 6068 Comm: syz.0.17 Not tainted syzkaller #0 PREEMPT(full)
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/25/2025
> RIP: 0010:__tcp_select_window+0x824/0x1320 net/ipv4/tcp_output.c:3336
> Code: ff ff ff 44 89 f1 d3 e0 89 c1 f7 d1 41 01 cc 41 21 c4 e9 a9 00 00 00 e8 ca 49 01 f8 e9 9c 00 00 00 e8 c0 49 01 f8 44 89 e0 99 <f7> 7c 24 1c 41 29 d4 48 bb 00 00 00 00 00 fc ff df e9 80 00 00 00
> RSP: 0018:ffffc90003017640 EFLAGS: 00010293
> RAX: 0000000000000000 RBX: 0000000000000000 RCX: ffff88807b469e40
> RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
> RBP: ffffc90003017730 R08: ffff888033268143 R09: 1ffff1100664d028
> R10: dffffc0000000000 R11: ffffed100664d029 R12: 0000000000000000
> R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
> FS:  000055557faa0500(0000) GS:ffff888126135000(0000) knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00007f64a1912ff8 CR3: 0000000072122000 CR4: 00000000003526f0
> Call Trace:
>  <TASK>
>  tcp_select_window net/ipv4/tcp_output.c:281 [inline]
>  __tcp_transmit_skb+0xbc7/0x3aa0 net/ipv4/tcp_output.c:1568
>  tcp_transmit_skb net/ipv4/tcp_output.c:1649 [inline]
>  tcp_send_active_reset+0x2d1/0x5b0 net/ipv4/tcp_output.c:3836
>  mptcp_do_fastclose+0x27e/0x380 net/mptcp/protocol.c:2793
>  mptcp_disconnect+0x238/0x710 net/mptcp/protocol.c:3253
>  mptcp_sendmsg_fastopen+0x2f8/0x580 net/mptcp/protocol.c:1776

Note: arf, sorry, I just noticed my small syzkaller instances found it
too, and I should have caught it before. I thought this issue was linked
to another one [1]. It is not: this one here is specific to TFO where
mptcp_disconnect() is called directly in one error path not cover by the
selftests.

[1]
https://lore.kernel.org/20251125-net-mptcp-clear-sched-rtx-v1-1-1cea4ad2165f@kernel.org

>  mptcp_sendmsg+0x1774/0x1980 net/mptcp/protocol.c:1855
>  sock_sendmsg_nosec net/socket.c:727 [inline]
>  __sock_sendmsg+0xe5/0x270 net/socket.c:742
>  __sys_sendto+0x3bd/0x520 net/socket.c:2244
>  __do_sys_sendto net/socket.c:2251 [inline]
>  __se_sys_sendto net/socket.c:2247 [inline]
>  __x64_sys_sendto+0xde/0x100 net/socket.c:2247
>  do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
>  do_syscall_64+0xfa/0xfa0 arch/x86/entry/syscall_64.c:94
>  entry_SYSCALL_64_after_hwframe+0x77/0x7f
> RIP: 0033:0x7f66e998f749
> Code: ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 a8 ff ff ff f7 d8 64 89 01 48
> RSP: 002b:00007ffff9acedb8 EFLAGS: 00000246 ORIG_RAX: 000000000000002c
> RAX: ffffffffffffffda RBX: 00007f66e9be5fa0 RCX: 00007f66e998f749
> RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000003
> RBP: 00007ffff9acee10 R08: 0000000000000000 R09: 0000000000000000
> R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000001
> R13: 00007f66e9be5fa0 R14: 00007f66e9be5fa0 R15: 0000000000000006
>  </TASK>
> 
> Fixes: ae155060247b ("mptcp: fix duplicate reset on fastclose")

I see that this patch targets "net-next". I think it would be better to
target "net". No need to send a new version, this patch applies on top
of "net" without conflicts, and Paolo told me he can apply it on top of
"net".

The patch looks good to me, and can be applied to "net" directly:

Reviewed-by: Matthieu Baerts (NGI0) <matttbe@...nel.org>


While at it, just to help me to track the backports:

Cc: stable@...r.kernel.org


> Reported-by: syzbot+3a92d359bc2ec6255a33@...kaller.appspotmail.com
> Closes: https://lore.kernel.org/netdev/69260882.a70a0220.d98e3.00b4.GAE@google.com/
> Signed-off-by: Kuniyuki Iwashima <kuniyu@...gle.com>
> ---
>  net/mptcp/protocol.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index 75bb1199bed9..40a8bdd5422a 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -2790,6 +2790,12 @@ static void mptcp_do_fastclose(struct sock *sk)
>  			goto unlock;
>  
>  		subflow->send_fastclose = 1;
> +
> +		/* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0
> +		 * issue in __tcp_select_window(), see tcp_disconnect().
> +		 */
> +		inet_csk(ssk)->icsk_ack.rcv_mss = TCP_MIN_MSS;


I initially thought this should have been set in mptcp_sendmsg_fastopen,
before calling mptcp_disconnect(), but doing that here is probably safer
to catch other cases, and it aligns with tcp_disconnect(). So that's
perfect, no need to change anything!


> +
>  		tcp_send_active_reset(ssk, ssk->sk_allocation,
>  				      SK_RST_REASON_TCP_ABORT_ON_CLOSE);
>  unlock:

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.

Powered by blists - more mailing lists