Re: [PATCH] percpu: fix data race in pcpu_alloc_noprof() and extend spinlock protection area

[Dennis Zhou]

Hello,

On Sun, Sep 08, 2024 at 10:52:10PM +0900, Jeongjun Park wrote:
> I got the following KCSAN report during syzbot testing:
> 
> ==================================================================
> BUG: KCSAN: data-race in pcpu_alloc_noprof / pcpu_free_area
> 
> read-write to 0xffffffff883f872c of 4 bytes by task 3378 on cpu 0:
>  pcpu_update_empty_pages mm/percpu.c:602 [inline]
>  pcpu_block_update_hint_free mm/percpu.c:1044 [inline]
>  pcpu_free_area+0x4dc/0x570 mm/percpu.c:1302
>  free_percpu+0x1c6/0xb30 mm/percpu.c:2277
>  xt_percpu_counter_free+0x63/0x80 net/netfilter/x_tables.c:1951
>  cleanup_entry+0x195/0x1c0 net/ipv6/netfilter/ip6_tables.c:671
>  __do_replace+0x470/0x580 net/ipv6/netfilter/ip6_tables.c:1099
>  do_replace net/ipv6/netfilter/ip6_tables.c:1158 [inline]
>  do_ip6t_set_ctl+0x820/0x8c0 net/ipv6/netfilter/ip6_tables.c:1644
>  nf_setsockopt+0x195/0x1b0 net/netfilter/nf_sockopt.c:101
>  ipv6_setsockopt+0x126/0x140 net/ipv6/ipv6_sockglue.c:998
>  tcp_setsockopt+0x93/0xb0 net/ipv4/tcp.c:3768
>  sock_common_setsockopt+0x64/0x80 net/core/sock.c:3735
>  do_sock_setsockopt net/socket.c:2324 [inline]
>  __sys_setsockopt+0x1d8/0x250 net/socket.c:2347
>  __do_sys_setsockopt net/socket.c:2356 [inline]
>  __se_sys_setsockopt net/socket.c:2353 [inline]
>  __x64_sys_setsockopt+0x66/0x80 net/socket.c:2353
>  x64_sys_call+0x278d/0x2d60 arch/x86/include/generated/asm/syscalls_64.h:55
>  do_syscall_x64 arch/x86/entry/common.c:52 [inline]
>  do_syscall_64+0x54/0x120 arch/x86/entry/common.c:83
>  entry_SYSCALL_64_after_hwframe+0x76/0x7e
> 
> read to 0xffffffff883f872c of 4 bytes by task 3374 on cpu 1:
>  pcpu_alloc_noprof+0x9a5/0x10c0 mm/percpu.c:1894
>  xt_percpu_counter_alloc+0x79/0x110 net/netfilter/x_tables.c:1931
>  find_check_entry net/ipv4/netfilter/ip_tables.c:526 [inline]
>  translate_table+0x921/0xf70 net/ipv4/netfilter/ip_tables.c:716
>  do_replace net/ipv4/netfilter/ip_tables.c:1137 [inline]
>  do_ipt_set_ctl+0x7bd/0x8b0 net/ipv4/netfilter/ip_tables.c:1635
>  nf_setsockopt+0x195/0x1b0 net/netfilter/nf_sockopt.c:101
>  ip_setsockopt+0xea/0x100 net/ipv4/ip_sockglue.c:1424
>  tcp_setsockopt+0x93/0xb0 net/ipv4/tcp.c:3768
>  sock_common_setsockopt+0x64/0x80 net/core/sock.c:3735
>  do_sock_setsockopt net/socket.c:2324 [inline]
>  __sys_setsockopt+0x1d8/0x250 net/socket.c:2347
>  __do_sys_setsockopt net/socket.c:2356 [inline]
>  __se_sys_setsockopt net/socket.c:2353 [inline]
>  __x64_sys_setsockopt+0x66/0x80 net/socket.c:2353
>  x64_sys_call+0x278d/0x2d60 arch/x86/include/generated/asm/syscalls_64.h:55
>  do_syscall_x64 arch/x86/entry/common.c:52 [inline]
>  do_syscall_64+0x54/0x120 arch/x86/entry/common.c:83
>  entry_SYSCALL_64_after_hwframe+0x76/0x7e
> 
> value changed: 0x00000005 -> 0x00000006
> 
> Reported by Kernel Concurrency Sanitizer on:
> CPU: 1 UID: 0 PID: 3374 Comm: syz-executor.3 Not tainted 6.11.0-rc6-syzkaller-00326-gd1f2d51b711a-dirty #0
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 08/06/2024
> ==================================================================
> 
> The global variable pcpu_nr_empty_pop_pages can be protected by pcpu_lock, 
> but since pcpu_alloc_noprof reads outside the spinlock protection section,
> a data race may occur and the branch of the conditional statement may change.
> Therefore, the reading of pcpu_nr_empty_pop_pages should be modified to be 
> performed within the spinlock protection section.
> 
> However, the for_each_clear_bitrange_from loop requires and uses a spinlock, 
> but it repeatedly locks and unlocks the spinlock unnecessarily. 
> 
> Therefore, I think it is appropriate to remove the repeated spin_lock and 
> spin_unlock in for_each_clear_bitrange_from and perform the operation of 
> reading pcpu_nr_empty_pop_pages and then perform spin_unlock to postpone 
> the point in time when the spin_unlock is performed.
> 
> Reported-by: syzbot <syzkaller@xxxxxxxxxxxxxxxx>
> Fixes: e04d320838f5 ("percpu: indent the population block in pcpu_alloc()")
> Signed-off-by: Jeongjun Park <aha310510@xxxxxxxxx>
> ---
>  mm/percpu.c |  5 ++---
>  1 files changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/percpu.c b/mm/percpu.c
> index 20d91af8c033..5c958a54da51 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -1864,7 +1864,6 @@ void __percpu *pcpu_alloc_noprof(size_t size, size_t align, bool reserved,
>  
>  area_found:
>  	pcpu_stats_area_alloc(chunk, size);
> -	spin_unlock_irqrestore(&pcpu_lock, flags);
>  
>  	/* populate if not all pages are already there */
>  	if (!is_atomic) {
> @@ -1878,14 +1877,12 @@ void __percpu *pcpu_alloc_noprof(size_t size, size_t align, bool reserved,
>  
>  			ret = pcpu_populate_chunk(chunk, rs, re, pcpu_gfp);
>  
> -			spin_lock_irqsave(&pcpu_lock, flags);
>  			if (ret) {
>  				pcpu_free_area(chunk, off);
>  				err = "failed to populate";
>  				goto fail_unlock;
>  			}
>  			pcpu_chunk_populated(chunk, rs, re);
> -			spin_unlock_irqrestore(&pcpu_lock, flags);
>  		}

We don't want to do this because pcpu_populate_chunk() calls
alloc_pages_node() which can block.
>  
>  		mutex_unlock(&pcpu_alloc_mutex);
> @@ -1894,6 +1891,8 @@ void __percpu *pcpu_alloc_noprof(size_t size, size_t align, bool reserved,
>  	if (pcpu_nr_empty_pop_pages < PCPU_EMPTY_POP_PAGES_LOW)
>  		pcpu_schedule_balance_work();
>  
> +	spin_unlock_irqrestore(&pcpu_lock, flags);
> +
>  	/* clear the areas and return address relative to base address */
>  	for_each_possible_cpu(cpu)
>  		memset((void *)pcpu_chunk_addr(chunk, cpu, 0) + off, 0, size);
> --

I sent out [1] which is a more appropriate fix. I'll merge it later
today.

Thanks,
Dennis

[1] https://lore.kernel.org/lkml/20240906031151.80719-1-dennis@xxxxxxxxxx/