[PATCH] net: tcp: avoid division by zero in __tcp_select_window

[PATCH] net: tcp: avoid division by zero in __tcp_select_window

[Evan Li]

We discovered a division-by-zero bug in __tcp_select_window() since
commit ae155060247b ("mptcp: fix duplicate reset on fastclose").

Under certain conditions during MPTCP fastclose, the mss value passed to
__tcp_select_window can be zero. The existing logic attempts to perform
rounddown(free_space, mss) without validating mss, leading to a division
operation in the helper (via do_div() or inline assembly) that triggers a
UBSAN overflow and kernel oops:

UBSAN: division-overflow in net/ipv4/tcp_output.c:3333:13
division by zero
RIP: __tcp_select_window+0x58a/0x1240
Call Trace:
 __tcp_transmit_skb+0xca3/0x38b0
 tcp_send_active_reset+0x422/0x7e0
 mptcp_do_fastclose+0x158/0x1e0
 ...

The issue occurs when tcp_send_active_reset() is called on a subflow with
an unset or zero mss, which can happen during fastclose teardown due to
earlier state transitions.

This patch adds a guard to return 0 immediately if mss == 0, preventing
the unsafe rounding operation. This is safe because a zero MSS implies
invalid or uninitialized state, and returning zero window reflects that no
reliable data transmission can proceed.

Fixes: ae155060247b ("mptcp: fix duplicate reset on fastclose")
Reported-by: kitta <kitta@linux.alibaba.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220820
Co-developed-by: kitta <kitta@linux.alibaba.com>
Signed-off-by: Evan Li <evan.li@linux.alibaba.com>
---
 net/ipv4/tcp_output.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index b94efb3050d2..e6d2851a0ae9 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -3329,9 +3329,11 @@ u32 __tcp_select_window(struct sock *sk)
 		 * We also don't do any window rounding when the free space
 		 * is too small.
 		 */
-		if (window <= free_space - mss || window > free_space)
+		if (window <= free_space - mss || window > free_space) {
+			if (unlikely(mss == 0))
+				return 0;  /* Prevent division by zero */
 			window = rounddown(free_space, mss);
-		else if (mss == full_space &&
+		} else if (mss == full_space &&
 			 free_space > window + (full_space >> 1))
 			window = free_space;
 	}
-- 
2.43.7

Re: [PATCH] net: tcp: avoid division by zero in __tcp_select_window

[Eric Dumazet]

On Mon, Dec 1, 2025 at 2:45 AM Evan Li <evan.li@linux.alibaba.com> wrote:
>
> We discovered a division-by-zero bug in __tcp_select_window() since
> commit ae155060247b ("mptcp: fix duplicate reset on fastclose").
>
> Under certain conditions during MPTCP fastclose, the mss value passed to
> __tcp_select_window can be zero. The existing logic attempts to perform
> rounddown(free_space, mss) without validating mss, leading to a division
> operation in the helper (via do_div() or inline assembly) that triggers a
> UBSAN overflow and kernel oops:
>
> UBSAN: division-overflow in net/ipv4/tcp_output.c:3333:13
> division by zero
> RIP: __tcp_select_window+0x58a/0x1240
> Call Trace:
>  __tcp_transmit_skb+0xca3/0x38b0
>  tcp_send_active_reset+0x422/0x7e0
>  mptcp_do_fastclose+0x158/0x1e0
>  ...
>
> The issue occurs when tcp_send_active_reset() is called on a subflow with
> an unset or zero mss, which can happen during fastclose teardown due to
> earlier state transitions.
>
> This patch adds a guard to return 0 immediately if mss == 0, preventing
> the unsafe rounding operation. This is safe because a zero MSS implies
> invalid or uninitialized state, and returning zero window reflects that no
> reliable data transmission can proceed.
>
> Fixes: ae155060247b ("mptcp: fix duplicate reset on fastclose")
> Reported-by: kitta <kitta@linux.alibaba.com>
> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220820
> Co-developed-by: kitta <kitta@linux.alibaba.com>
> Signed-off-by: Evan Li <evan.li@linux.alibaba.com>
> ---
>  net/ipv4/tcp_output.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> index b94efb3050d2..e6d2851a0ae9 100644
> --- a/net/ipv4/tcp_output.c
> +++ b/net/ipv4/tcp_output.c
> @@ -3329,9 +3329,11 @@ u32 __tcp_select_window(struct sock *sk)
>                  * We also don't do any window rounding when the free space
>                  * is too small.
>                  */
> -               if (window <= free_space - mss || window > free_space)
> +               if (window <= free_space - mss || window > free_space) {
> +                       if (unlikely(mss == 0))
> +                               return 0;  /* Prevent division by zero */
>                         window = rounddown(free_space, mss);
> -               else if (mss == full_space &&
> +               } else if (mss == full_space &&
>                          free_space > window + (full_space >> 1))
>                         window = free_space;
>         }

I think you are missing a fix in MPTCP.

commit f07f4ea53e22429c84b20832fa098b5ecc0d4e35
Author: Kuniyuki Iwashima <kuniyu@google.com>
Date:   Tue Nov 25 19:53:29 2025 +0000

    mptcp: Initialise rcv_mss before calling tcp_send_active_reset()
in mptcp_do_fastclose().