Re: [PATCH 3/3] net: key: Validate address family in set_ipsecrequest()

[Sabrina Dubroca <sd@queasysnail.net>]

note: There are a few issues with the format of this patch, and the
subject prefix should be "[PATCH ipsec n/3]" for all the patches in
the series. But I'm also not sure if this is the right way to fix this
syzbot report.


2025-11-06, 21:56:58 +0800, clingfei wrote:
> From: SHAURYA RANE <ssrane_b23@ee.vjti.ac.in>


From here:

> Hi syzbot,
> 
> Please test the following patch.
> 
> #syz test: git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master
> 
> Thanks,
> Shaurya Rane
> 
> From 123c5ac9ba261681b58a6217409c94722fde4249 Mon Sep 17 00:00:00 2001
> From: Shaurya Rane <ssrane_b23@ee.vjti.ac.in>
> Date: Sun, 19 Oct 2025 23:18:30 +0530
> Subject: [PATCH] net: key: Validate address family in set_ipsecrequest()

to here should be removed.


> syzbot reported a kernel BUG in set_ipsecrequest() due to an
> skb_over_panic when processing XFRM_MSG_MIGRATE messages.
> 
> The root cause is that set_ipsecrequest() does not validate the
> address family parameter before using it to calculate buffer sizes.
> When an unsupported family value (such as 0) is passed,
> pfkey_sockaddr_len() returns 0, leading to incorrect size calculations.
> 
> In pfkey_send_migrate(), the buffer size is calculated based on
> pfkey_sockaddr_pair_size(), which uses pfkey_sockaddr_len(). When
> family=0, this returns 0, so only sizeof(struct sadb_x_ipsecrequest)
> (16 bytes) is allocated per entry. However, set_ipsecrequest() is
> called multiple times in a loop (once for old_family, once for
> new_family, for each migration bundle), repeatedly calling skb_put_zero()
> with 16 bytes each time.

So the root of the problem is a mismatch between allocation size and
the actual size needed. Unexpected families are not good, sure, but
would not cause a panic if the sizes were handled correctly.

OTOH, for this old code which is being deprecated, maybe it doesn't
matter to fix it "properly". (but see below)


> This causes the tail pointer to exceed the end pointer of the skb,
> triggering skb_over_panic:
>   tail: 0x188 (392 bytes)
>   end:  0x180 (384 bytes)
> 
> Fix this by validating that pfkey_sockaddr_len() returns a non-zero
> value before proceeding with buffer operations. This ensures proper
> size calculations and prevents buffer overflow. Checking socklen
> instead of just family==0 provides comprehensive validation for all
> unsupported address families.
> 
> Reported-by: syzbot+be97dd4da14ae88b6ba4@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=be97dd4da14ae88b6ba4
> Fixes: 08de61beab8a ("[PFKEYV2]: Extension for dynamic update of
> endpoint address(es)")
> Signed-off-by: Shaurya Rane <ssrane_b23@ee.vjti.ac.in>
> ---
>  net/key/af_key.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/net/key/af_key.c b/net/key/af_key.c
> index cfda15a5aa4d..93c20a31e03d 100644
> --- a/net/key/af_key.c
> +++ b/net/key/af_key.c
> @@ -3529,7 +3529,11 @@ static int set_ipsecrequest(struct sk_buff *skb,
>  	if (!family)
>  		return -EINVAL;
>  
> -	size_req = sizeof(struct sadb_x_ipsecrequest) +
> +    /* Reject invalid/unsupported address families */

Steffen, AFAICT the whole migrate code has no family
validation. Shouldn't we check {old,new}_family to be one of
{AF_INET,AF_INET6} in xfrm_migrate_check? This should take care of the
problems that this series tries to address, and avoid having objects
installed in the kernel with unexpected families (which would match
what validate_tmpl does).


Looking quickly at xfrm_migrate_state_find, it also seems to compare
addresses without checking that both addresses are of the same
family. That seems a bit wrong, but changing the behavior of that old
code is maybe too risky.



> +    if (!socklen)
> +        return -EINVAL;
> +
> +    size_req = sizeof(struct sadb_x_ipsecrequest) +

nit: tabs should be used, not spaces

>  		   pfkey_sockaddr_pair_size(family);
>  
>  	rq = skb_put_zero(skb, size_req);

-- 
Sabrina