Re: [PATCH -next] mm/vmpressure: fix data-race with memcg->socket_pressure
From: Muchun Song
Date: Mon Oct 25 2021 - 05:55:49 EST
On Mon, Oct 25, 2021 at 4:14 PM Yuanzheng Song <songyuanzheng@xxxxxxxxxx> wrote:
>
> BUG: KCSAN: data-race in __sk_mem_reduce_allocated / vmpressure
>
> write to 0xffff8881286f4938 of 8 bytes by task 24550 on cpu 3:
> vmpressure+0x218/0x230 mm/vmpressure.c:307
> shrink_node_memcgs+0x2b9/0x410 mm/vmscan.c:2658
> shrink_node+0x9d2/0x11d0 mm/vmscan.c:2769
> shrink_zones+0x29f/0x470 mm/vmscan.c:2972
> do_try_to_free_pages+0x193/0x6e0 mm/vmscan.c:3027
> try_to_free_mem_cgroup_pages+0x1c0/0x3f0 mm/vmscan.c:3345
> reclaim_high mm/memcontrol.c:2440 [inline]
> mem_cgroup_handle_over_high+0x18b/0x4d0 mm/memcontrol.c:2624
> tracehook_notify_resume include/linux/tracehook.h:197 [inline]
> exit_to_user_mode_loop kernel/entry/common.c:164 [inline]
> exit_to_user_mode_prepare+0x110/0x170 kernel/entry/common.c:191
> syscall_exit_to_user_mode+0x16/0x30 kernel/entry/common.c:266
> ret_from_fork+0x15/0x30 arch/x86/entry/entry_64.S:289
>
> read to 0xffff8881286f4938 of 8 bytes by interrupt on cpu 1:
> mem_cgroup_under_socket_pressure include/linux/memcontrol.h:1483 [inline]
> sk_under_memory_pressure include/net/sock.h:1314 [inline]
> __sk_mem_reduce_allocated+0x1d2/0x270 net/core/sock.c:2696
> __sk_mem_reclaim+0x44/0x50 net/core/sock.c:2711
> sk_mem_reclaim include/net/sock.h:1490 [inline]
> ......
> net_rx_action+0x17a/0x480 net/core/dev.c:6864
> __do_softirq+0x12c/0x2af kernel/softirq.c:298
> run_ksoftirqd+0x13/0x20 kernel/softirq.c:653
> smpboot_thread_fn+0x33f/0x510 kernel/smpboot.c:165
> kthread+0x1fc/0x220 kernel/kthread.c:292
> ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:296
>
> When reading memcg->socket_pressure in mem_cgroup_under_socket_pressure()
> and writing memcg->socket_pressure in vmpressure() at the same time,
> the data-race occurs.
>
> So fix it by using READ_ONCE() and WRITE_ONCE() to read and write
> memcg->socket_pressure.
>
> Signed-off-by: Yuanzheng Song <songyuanzheng@xxxxxxxxxx>
Reviewed-by: Muchun Song <songmuchun@xxxxxxxxxxxxx>
Thanks.