mptcp: Check reclaim amount before reducing allocation

syzbot found a page counter underflow that was triggered by MPTCP's
reclaim code:

page_counter underflow: -4294964789 nr_pages=4294967295
WARNING: CPU: 2 PID: 3785 at mm/page_counter.c:56 page_counter_cancel+0xcf/0xe0 mm/page_counter.c:56
Modules linked in:
CPU: 2 PID: 3785 Comm: kworker/2:6 Not tainted 5.16.0-rc1-syzkaller #0
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.14.0-2 04/01/2014
Workqueue: events mptcp_worker

RIP: 0010:page_counter_cancel+0xcf/0xe0 mm/page_counter.c:56
Code: c7 04 24 00 00 00 00 45 31 f6 eb 97 e8 2a 2b b5 ff 4c 89 ea 48 89 ee 48 c7 c7 00 9e b8 89 c6 05 a0 c1 ba 0b 01 e8 95 e4 4b 07 <0f> 0b eb a8 4c 89 e7 e8 25 5a fb ff eb c7 0f 1f 00 41 56 41 55 49
RSP: 0018:ffffc90002d4f918 EFLAGS: 00010082

RAX: 0000000000000000 RBX: ffff88806a494120 RCX: 0000000000000000
RDX: ffff8880688c41c0 RSI: ffffffff815e8f28 RDI: fffff520005a9f15
RBP: ffffffff000009cb R08: 0000000000000000 R09: 0000000000000000
R10: ffffffff815e2cfe R11: 0000000000000000 R12: ffff88806a494120
R13: 00000000ffffffff R14: 0000000000000000 R15: 0000000000000001
FS:  0000000000000000(0000) GS:ffff88802cc00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000001b2de21000 CR3: 000000005ad59000 CR4: 0000000000150ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
 page_counter_uncharge+0x2e/0x60 mm/page_counter.c:160
 drain_stock+0xc1/0x180 mm/memcontrol.c:2219
 refill_stock+0x139/0x2f0 mm/memcontrol.c:2271
 __sk_mem_reduce_allocated+0x24d/0x550 net/core/sock.c:2945
 __mptcp_rmem_reclaim net/mptcp/protocol.c:167 [inline]
 __mptcp_mem_reclaim_partial+0x124/0x410 net/mptcp/protocol.c:975
 mptcp_mem_reclaim_partial net/mptcp/protocol.c:982 [inline]
 mptcp_alloc_tx_skb net/mptcp/protocol.c:1212 [inline]
 mptcp_sendmsg_frag+0x18c6/0x2190 net/mptcp/protocol.c:1279
 __mptcp_push_pending+0x232/0x720 net/mptcp/protocol.c:1545
 mptcp_release_cb+0xfe/0x200 net/mptcp/protocol.c:2975
 release_sock+0xb4/0x1b0 net/core/sock.c:3306
 mptcp_worker+0x51e/0xc10 net/mptcp/protocol.c:2443
 process_one_work+0x9b2/0x1690 kernel/workqueue.c:2298
 worker_thread+0x658/0x11f0 kernel/workqueue.c:2445
 kthread+0x405/0x4f0 kernel/kthread.c:327
 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295
 </TASK>

__mptcp_mem_reclaim_partial() could call __mptcp_rmem_reclaim() with a
negative value, which passed that negative value to
__sk_mem_reduce_allocated() and triggered the splat above.

Check for a reclaim amount that is positive and large enough for
__mptcp_rmem_reclaim() to actually adjust rmem_fwd_alloc (much like
the sk_mem_reclaim_partial() code the function is based on).

v2: Use '>' instead of '>=', since SK_MEM_QUANTUM - 1 would get
right-shifted into nothing by __mptcp_rmem_reclaim.

Fixes: 6511882cdd ("mptcp: allocate fwd memory separately on the rx and tx path")
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/252
Reported-and-tested-by: syzbot+bc9e2d2dbcb347dd215a@syzkaller.appspotmail.com
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Michal Hocko <mhocko@suse.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/mptcp/protocol.c

@@ -972,7 +972,9 @@ static void __mptcp_mem_reclaim_partial(struct sock *sk)
 
 	lockdep_assert_held_once(&sk->sk_lock.slock);
 
-	__mptcp_rmem_reclaim(sk, reclaimable - 1);
+	if (reclaimable > SK_MEM_QUANTUM)
+		__mptcp_rmem_reclaim(sk, reclaimable - 1);
+
 	sk_mem_reclaim_partial(sk);
 }