[PATCH net v2] mptcp: fix soft lockup in mptcp_recvmsg()

[PATCH net v2] mptcp: fix soft lockup in mptcp_recvmsg()

[Li Xiasong]

syzbot reported a soft lockup in mptcp_recvmsg() [0].

When receiving data with MSG_PEEK | MSG_WAITALL flags, the skb is not
removed from the sk_receive_queue. This causes sk_wait_data() to always
find available data and never perform actual waiting, leading to a soft
lockup.

Fix this by adding a 'last' parameter to track the last peeked skb.
This allows sk_wait_data() to make informed waiting decisions and prevent
infinite loops when MSG_PEEK is used.

[0]:
watchdog: BUG: soft lockup - CPU#2 stuck for 156s! [server:1963]
Modules linked in:
CPU: 2 UID: 0 PID: 1963 Comm: server Not tainted 6.19.0-rc8 #61 PREEMPT(none)
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
RIP: 0010:sk_wait_data+0x15/0x190
Code: 80 00 00 00 00 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 f3 0f 1e fa 41 56 41 55 41 54 49 89 f4 55 48 89 d5 53 48 89 fb <48> 83 ec 30 65 48 8b 05 17 a4 6b 01 48 89 44 24 28 31 c0 65 48 8b
RSP: 0018:ffffc90000603ca0 EFLAGS: 00000246
RAX: 0000000000000000 RBX: ffff888102bf0800 RCX: 0000000000000001
RDX: 0000000000000000 RSI: ffffc90000603d18 RDI: ffff888102bf0800
RBP: 0000000000000000 R08: 0000000000000002 R09: 0000000000000101
R10: 0000000000000000 R11: 0000000000000075 R12: ffffc90000603d18
R13: ffff888102bf0800 R14: ffff888102bf0800 R15: 0000000000000000
FS:  00007f6e38b8c4c0(0000) GS:ffff8881b877e000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000055aa7bff1680 CR3: 0000000105cbe000 CR4: 00000000000006f0
Call Trace:
 <TASK>
 mptcp_recvmsg+0x547/0x8c0 net/mptcp/protocol.c:2329
 inet_recvmsg+0x11f/0x130 net/ipv4/af_inet.c:891
 sock_recvmsg+0x94/0xc0 net/socket.c:1100
 __sys_recvfrom+0xb2/0x130 net/socket.c:2256
 __x64_sys_recvfrom+0x1f/0x30 net/socket.c:2267
 do_syscall_64+0x59/0x2d0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x76/0x7e arch/x86/entry/entry_64.S:131
RIP: 0033:0x7f6e386a4a1d
Code: 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 48 8d 05 f1 de 2c 00 41 89 ca 8b 00 85 c0 75 20 45 31 c9 45 31 c0 b8 2d 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 6b f3 c3 66 0f 1f 84 00 00 00 00 00 41 56 41
RSP: 002b:00007ffc3c4bb078 EFLAGS: 00000246 ORIG_RAX: 000000000000002d
RAX: ffffffffffffffda RBX: 000000000000861e RCX: 00007f6e386a4a1d
RDX: 00000000000003ff RSI: 00007ffc3c4bb150 RDI: 0000000000000004
RBP: 00007ffc3c4bb570 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000103 R11: 0000000000000246 R12: 00005605dbc00be0
R13: 00007ffc3c4bb650 R14: 0000000000000000 R15: 0000000000000000
 </TASK>

Fixes: 612f71d7328c ("mptcp: fix possible stall on recvmsg()")
Signed-off-by: Li Xiasong <lixiasong1@huawei.com>
---

v1->v2:
Thanks to Matthieu Baerts for the suggestion and review.
- Move 'last' variable definition into the while loop in mptcp_recvmsg() to
  narrow its scope.
- In __mptcp_recvmsg_mskq(), assign 'last' right after 'copied' is updated,
  regardless of MSG_PEEK flag.

v1:
https://lore.kernel.org/netdev/20260302052651.1466983-1-lixiasong1@huawei.com/

---
 net/mptcp/protocol.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index cf1852b99963..401fb2b17685 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -2006,7 +2006,7 @@ static void mptcp_eat_recv_skb(struct sock *sk, struct sk_buff *skb)
 static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
 				size_t len, int flags, int copied_total,
 				struct scm_timestamping_internal *tss,
-				int *cmsg_flags)
+				int *cmsg_flags, struct sk_buff **last)
 {
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	struct sk_buff *skb, *tmp;
@@ -2048,6 +2048,7 @@ static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
 		}
 
 		copied += count;
+		*last = skb;
 
 		if (!(flags & MSG_PEEK)) {
 			msk->bytes_consumed += count;
@@ -2288,10 +2289,12 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 		cmsg_flags = MPTCP_CMSG_INQ;
 
 	while (copied < len) {
+		struct sk_buff *last = skb_peek_tail(&sk->sk_receive_queue);
 		int err, bytes_read;
 
 		bytes_read = __mptcp_recvmsg_mskq(sk, msg, len - copied, flags,
-						  copied, &tss, &cmsg_flags);
+						  copied, &tss, &cmsg_flags,
+						  &last);
 		if (unlikely(bytes_read < 0)) {
 			if (!copied)
 				copied = bytes_read;
@@ -2343,7 +2346,7 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 
 		pr_debug("block timeout %ld\n", timeo);
 		mptcp_cleanup_rbuf(msk, copied);
-		err = sk_wait_data(sk, &timeo, NULL);
+		err = sk_wait_data(sk, &timeo, last);
 		if (err < 0) {
 			err = copied ? : err;
 			goto out_err;
-- 
2.34.1

Re: [PATCH net v2] mptcp: fix soft lockup in mptcp_recvmsg()

[Matthieu Baerts]

Hi Li,

On 24/03/2026 09:51, Li Xiasong wrote:
> syzbot reported a soft lockup in mptcp_recvmsg() [0].
> 
> When receiving data with MSG_PEEK | MSG_WAITALL flags, the skb is not
> removed from the sk_receive_queue. This causes sk_wait_data() to always
> find available data and never perform actual waiting, leading to a soft
> lockup.
> 
> Fix this by adding a 'last' parameter to track the last peeked skb.
> This allows sk_wait_data() to make informed waiting decisions and prevent
> infinite loops when MSG_PEEK is used.

Thank you for the v2, it looks good to me:

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

@Netdev maintainers: this patch can be applied in "net" directly.

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

Re: [PATCH net v2] mptcp: fix soft lockup in mptcp_recvmsg()

[Matthieu Baerts]

Hi Li,

On 24/03/2026 09:51, Li Xiasong wrote:
> syzbot reported a soft lockup in mptcp_recvmsg() [0].
> 
> When receiving data with MSG_PEEK | MSG_WAITALL flags, the skb is not
> removed from the sk_receive_queue. This causes sk_wait_data() to always
> find available data and never perform actual waiting, leading to a soft
> lockup.
> 
> Fix this by adding a 'last' parameter to track the last peeked skb.
> This allows sk_wait_data() to make informed waiting decisions and prevent
> infinite loops when MSG_PEEK is used.

(...)

> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index cf1852b99963..401fb2b17685 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -2006,7 +2006,7 @@ static void mptcp_eat_recv_skb(struct sock *sk, struct sk_buff *skb)
>  static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>  				size_t len, int flags, int copied_total,
>  				struct scm_timestamping_internal *tss,
> -				int *cmsg_flags)
> +				int *cmsg_flags, struct sk_buff **last)
>  {
>  	struct mptcp_sock *msk = mptcp_sk(sk);
>  	struct sk_buff *skb, *tmp;
> @@ -2048,6 +2048,7 @@ static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>  		}
>  
>  		copied += count;
> +		*last = skb;

My bad, my recommendation to move this assignment here was not a good
idea, because 'skb' can be freed at some points after mptcp_eat_recv_skb
that can be called here below.

If I'm not mistaken, when MSG_PEEK is not used, sk_wait_data() can
always be called with NULL for the last skb, because the queue is
supposed to be empty. If not, no need to wait for new packets, so NULL
can be used. Is that correct?

If yes, then I guess 'last' can be initialised to NULL before calling
__mptcp_recvmsg_mskq (limit scope), and only set to 'skb' here below,
when MSG_PEEK is not used (what you had in v1).

Would that work for you?

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

Re: [PATCH net v2] mptcp: fix soft lockup in mptcp_recvmsg()

[Li Xiasong]

Hi Matt,

On 3/25/2026 3:23 AM, Matthieu Baerts wrote:
> Hi Li,
> 
> On 24/03/2026 09:51, Li Xiasong wrote:
>> syzbot reported a soft lockup in mptcp_recvmsg() [0].
>>
>> When receiving data with MSG_PEEK | MSG_WAITALL flags, the skb is not
>> removed from the sk_receive_queue. This causes sk_wait_data() to always
>> find available data and never perform actual waiting, leading to a soft
>> lockup.
>>
>> Fix this by adding a 'last' parameter to track the last peeked skb.
>> This allows sk_wait_data() to make informed waiting decisions and prevent
>> infinite loops when MSG_PEEK is used.
> 
> (...)
> 
>> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
>> index cf1852b99963..401fb2b17685 100644
>> --- a/net/mptcp/protocol.c
>> +++ b/net/mptcp/protocol.c
>> @@ -2006,7 +2006,7 @@ static void mptcp_eat_recv_skb(struct sock *sk, struct sk_buff *skb)
>>  static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>>  				size_t len, int flags, int copied_total,
>>  				struct scm_timestamping_internal *tss,
>> -				int *cmsg_flags)
>> +				int *cmsg_flags, struct sk_buff **last)
>>  {
>>  	struct mptcp_sock *msk = mptcp_sk(sk);
>>  	struct sk_buff *skb, *tmp;
>> @@ -2048,6 +2048,7 @@ static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>>  		}
>>  
>>  		copied += count;
>> +		*last = skb;
> 
> My bad, my recommendation to move this assignment here was not a good
> idea, because 'skb' can be freed at some points after mptcp_eat_recv_skb
> that can be called here below.
> 
> If I'm not mistaken, when MSG_PEEK is not used, sk_wait_data() can
> always be called with NULL for the last skb, because the queue is
> supposed to be empty. If not, no need to wait for new packets, so NULL
> can be used. Is that correct?
> 
> If yes, then I guess 'last' can be initialised to NULL before calling
> __mptcp_recvmsg_mskq (limit scope), and only set to 'skb' here below,
> when MSG_PEEK is not used (what you had in v1).
> 
> Would that work for you?
> 
> Cheers,
> Matt

Thanks for the review. I analyzed the concern about the 'last' pointer.

When writing the v2 patch, I did consider this case - in non-MSG_PEEK
scenario, the skb is freed by mptcp_eat_recv_skb() after setting *last =
skb, making 'last' point to freed memory. However, in sk_wait_data(), the
'last' parameter is only used for pointer comparison:

    skb_peek_tail(&sk->sk_receive_queue) != skb

It only compares the pointer value without dereferencing, so there's no
actual UAF issue.

However, I agree it's still risky to keep a dangling pointer around. I
appreciate your suggestion and will adopt it in v3.

Best regards,
Li Xiasong

Re: [PATCH net v2] mptcp: fix soft lockup in mptcp_recvmsg()

[Matthieu Baerts]

Hi Li,

25 Mar 2026 03:39:13 Li Xiasong <lixiasong1@huawei.com>:

> Hi Matt,
>
> On 3/25/2026 3:23 AM, Matthieu Baerts wrote:
>> Hi Li,
>>
>> On 24/03/2026 09:51, Li Xiasong wrote:
>>> syzbot reported a soft lockup in mptcp_recvmsg() [0].
>>>
>>> When receiving data with MSG_PEEK | MSG_WAITALL flags, the skb is not
>>> removed from the sk_receive_queue. This causes sk_wait_data() to always
>>> find available data and never perform actual waiting, leading to a soft
>>> lockup.
>>>
>>> Fix this by adding a 'last' parameter to track the last peeked skb.
>>> This allows sk_wait_data() to make informed waiting decisions and prevent
>>> infinite loops when MSG_PEEK is used.
>>
>> (...)
>>
>>> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
>>> index cf1852b99963..401fb2b17685 100644
>>> --- a/net/mptcp/protocol.c
>>> +++ b/net/mptcp/protocol.c
>>> @@ -2006,7 +2006,7 @@ static void mptcp_eat_recv_skb(struct sock *sk, struct sk_buff *skb)
>>> static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>>>                 size_t len, int flags, int copied_total,
>>>                 struct scm_timestamping_internal *tss,
>>> -               int *cmsg_flags)
>>> +               int *cmsg_flags, struct sk_buff **last)
>>> {
>>>     struct mptcp_sock *msk = mptcp_sk(sk);
>>>     struct sk_buff *skb, *tmp;
>>> @@ -2048,6 +2048,7 @@ static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>>>         }
>>>
>>>         copied += count;
>>> +       *last = skb;
>>
>> My bad, my recommendation to move this assignment here was not a good
>> idea, because 'skb' can be freed at some points after mptcp_eat_recv_skb
>> that can be called here below.
>>
>> If I'm not mistaken, when MSG_PEEK is not used, sk_wait_data() can
>> always be called with NULL for the last skb, because the queue is
>> supposed to be empty. If not, no need to wait for new packets, so NULL
>> can be used. Is that correct?
>>
>> If yes, then I guess 'last' can be initialised to NULL before calling
>> __mptcp_recvmsg_mskq (limit scope), and only set to 'skb' here below,
>> when MSG_PEEK is not used (what you had in v1).
>>
>> Would that work for you?
>>
>> Cheers,
>> Matt
>
> Thanks for the review. I analyzed the concern about the 'last' pointer.
>
> When writing the v2 patch, I did consider this case - in non-MSG_PEEK
> scenario, the skb is freed by mptcp_eat_recv_skb() after setting *last =
> skb, making 'last' point to freed memory. However, in sk_wait_data(), the
> 'last' parameter is only used for pointer comparison:
>
>     skb_peek_tail(&sk->sk_receive_queue) != skb
>
> It only compares the pointer value without dereferencing, so there's no
> actual UAF issue.

Indeed, but the skb could be (unlikely) reused at that stage.

For me, the main point is that mptcp_eat_recv_skb() will remove the
skb from the queue. Then NULL can be passed instead, safer.

Cheers,
Matt

Re: [PATCH net v2] mptcp: fix soft lockup in mptcp_recvmsg()

[Matthieu Baerts]

On 25/03/2026 08:16, Matthieu Baerts wrote:
> Hi Li,
> 
> 25 Mar 2026 03:39:13 Li Xiasong <lixiasong1@huawei.com>:
> 
>> Hi Matt,
>>
>> On 3/25/2026 3:23 AM, Matthieu Baerts wrote:
>>> Hi Li,
>>>
>>> On 24/03/2026 09:51, Li Xiasong wrote:
>>>> syzbot reported a soft lockup in mptcp_recvmsg() [0].
>>>>
>>>> When receiving data with MSG_PEEK | MSG_WAITALL flags, the skb is not
>>>> removed from the sk_receive_queue. This causes sk_wait_data() to always
>>>> find available data and never perform actual waiting, leading to a soft
>>>> lockup.
>>>>
>>>> Fix this by adding a 'last' parameter to track the last peeked skb.
>>>> This allows sk_wait_data() to make informed waiting decisions and prevent
>>>> infinite loops when MSG_PEEK is used.
>>>
>>> (...)
>>>
>>>> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
>>>> index cf1852b99963..401fb2b17685 100644
>>>> --- a/net/mptcp/protocol.c
>>>> +++ b/net/mptcp/protocol.c
>>>> @@ -2006,7 +2006,7 @@ static void mptcp_eat_recv_skb(struct sock *sk, struct sk_buff *skb)
>>>> static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>>>>                 size_t len, int flags, int copied_total,
>>>>                 struct scm_timestamping_internal *tss,
>>>> -               int *cmsg_flags)
>>>> +               int *cmsg_flags, struct sk_buff **last)
>>>> {
>>>>     struct mptcp_sock *msk = mptcp_sk(sk);
>>>>     struct sk_buff *skb, *tmp;
>>>> @@ -2048,6 +2048,7 @@ static int __mptcp_recvmsg_mskq(struct sock *sk, struct msghdr *msg,
>>>>         }
>>>>
>>>>         copied += count;
>>>> +       *last = skb;
>>>
>>> My bad, my recommendation to move this assignment here was not a good
>>> idea, because 'skb' can be freed at some points after mptcp_eat_recv_skb
>>> that can be called here below.
>>>
>>> If I'm not mistaken, when MSG_PEEK is not used, sk_wait_data() can
>>> always be called with NULL for the last skb, because the queue is
>>> supposed to be empty. If not, no need to wait for new packets, so NULL
>>> can be used. Is that correct?
>>>
>>> If yes, then I guess 'last' can be initialised to NULL before calling
>>> __mptcp_recvmsg_mskq (limit scope), and only set to 'skb' here below,
>>> when MSG_PEEK is not used (what you had in v1).
>>>
>>> Would that work for you?
>>>
>>> Cheers,
>>> Matt
>>
>> Thanks for the review. I analyzed the concern about the 'last' pointer.
>>
>> When writing the v2 patch, I did consider this case - in non-MSG_PEEK
>> scenario, the skb is freed by mptcp_eat_recv_skb() after setting *last =
>> skb, making 'last' point to freed memory. However, in sk_wait_data(), the
>> 'last' parameter is only used for pointer comparison:
>>
>>     skb_peek_tail(&sk->sk_receive_queue) != skb
>>
>> It only compares the pointer value without dereferencing, so there's no
>> actual UAF issue.
> 
> Indeed, but the skb could be (unlikely) reused at that stage.
> 
> For me, the main point is that mptcp_eat_recv_skb() will remove the
> skb from the queue. Then NULL can be passed instead, safer.
One last thing: as noticed by the AI review, this Fixes tag could be
used instead, as it is more tied to MSG_PEEK:

  Fixes: 8e04ce45a8db ("mptcp: fix MSG_PEEK stream corruption")

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