On Tue 21-12-21 15:57:36, Andrew Morton wrote:

On Sat, 18 Dec 2021 06:04:22 -0800 syzbot [off-list ref] wrote:

quoted

syzbot has found a reproducer for the following issue on:

HEAD commit:    fbf252e09678 Add linux-next specific files for 20211216
git tree:       linux-next
console output: https://syzkaller.appspot.com/x/log.txt?x=1797de99b00000
kernel config:  https://syzkaller.appspot.com/x/.config?x=7fcbb9aa19a433c8
dashboard link: https://syzkaller.appspot.com/bug?extid=bc9e2d2dbcb347dd215a
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=135d179db00000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=113edb6db00000

Useful to have that, thanks.

I'm suspecting that mptcp is doing something strange.

Yes.

Could I as the
developers to please take a look?

quoted

IMPORTANT: if you fix the issue, please add the following tag to the commit:
Reported-by: syzbot+bc9e2d2dbcb347dd215a@syzkaller.appspotmail.com

R13: 00007ffdeb858640 R14: 00007ffdeb858680 R15: 0000000000000004
 </TASK>
------------[ cut here ]------------
page_counter underflow: -4294966651 nr_pages=4294967295

__mptcp_mem_reclaim_partial is trying to uncharge (via
__sk_mem_reduce_allocated) negative amount. nr_pages has overflown when
converted from int to unsigned int (-1). I would say that
__mptcp_mem_reclaim_partial has evaluated
	reclaimable = mptcp_sk(sk)->rmem_fwd_alloc - sk_unused_reserved_mem(sk)
to 0 and __mptcp_rmem_reclaim(sk, reclaimable - 1) made it -1.
-- 
Michal Hocko
SUSE Labs