Re: [PATCH] sysctl: Fix division by zero in percpu_pagelist_fraction handler

[David Rientjes]

On Sat, 3 May 2014, Oleg Drokin wrote:

> percpu_pagelist_fraction_sysctl_handler calls proc_dointvec_minmax
> and blindly assumes that return value of 0 means success.
> In fact the other valid case is when it got a zero length input.
> 
> After that it proceeds to a division by percpu_pagelist_fraction
> value which is conveniently set to a default of zero, resulting in
> division by zero.
> 
> Other than checking the bytecount to be more than zero, perhaps
> a better default value for percpu_pagelist_fraction would help too.
> 
> [  661.985469] divide error: 0000 [#1] SMP DEBUG_PAGEALLOC
> [  661.985868] Modules linked in: binfmt_misc cfg80211 rfkill rpcsec_gss_krb5 ttm drm_kms_helper drm i2c_piix4 microcode i2c_core joydev serio_raw pcspkr virtio_blk nfsd
> [  661.986008] CPU: 1 PID: 9142 Comm: badarea_io Not tainted 3.15.0-rc2-vm-nfs+ #19
> [  661.986008] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
> [  661.986008] task: ffff8800d5aeb6e0 ti: ffff8800d87a2000 task.ti: ffff8800d87a2000
> [  661.986008] RIP: 0010:[<ffffffff81152664>]  [<ffffffff81152664>] percpu_pagelist_fraction_sysctl_handler+0x84/0x120
> [  661.988031] RSP: 0018:ffff8800d87a3e78  EFLAGS: 00010246
> [  661.988031] RAX: 0000000000000f89 RBX: ffff88011f7fd000 RCX: 0000000000000000
> [  661.988031] RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000010
> [  661.988031] RBP: ffff8800d87a3e98 R08: ffffffff81d002c8 R09: ffff8800d87a3f50
> [  661.988031] R10: 000000000000000b R11: 0000000000000246 R12: 0000000000000060
> [  661.988031] R13: ffffffff81c3c3e0 R14: ffffffff81cfddf8 R15: ffff8801193b0800
> [  661.988031] FS:  00007f614f1e9740(0000) GS:ffff88011f440000(0000) knlGS:0000000000000000
> [  661.988031] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
> [  661.988031] CR2: 00007f614f1fa000 CR3: 00000000d9291000 CR4: 00000000000006e0
> [  661.988031] Stack:
> [  661.988031]  0000000000000001 ffffffffffffffea ffffffff81c3c3e0 0000000000000000
> [  661.988031]  ffff8800d87a3ee8 ffffffff8122b163 ffff8800d87a3f50 00007fff1564969c
> [  661.988031]  0000000000000000 ffff8800d8098f00 00007fff1564969c ffff8800d87a3f50
> [  661.988031] Call Trace:
> [  661.988031]  [<ffffffff8122b163>] proc_sys_call_handler+0xb3/0xc0
> [  661.988031]  [<ffffffff8122b184>] proc_sys_write+0x14/0x20
> [  661.988031]  [<ffffffff811ba93a>] vfs_write+0xba/0x1e0
> [  661.988031]  [<ffffffff811bb486>] SyS_write+0x46/0xb0
> [  661.988031]  [<ffffffff816db7ff>] tracesys+0xe1/0xe6
> [  661.988031] Code: 1f 84 00 00 00 00 00 48 83 bb b0 06 00 00 00 0f 84 7c 00 00 00 48 63 0d 93 6a e1 00 48 8b 83 b8 06 00 00 31 d2 41 bc 60 00 00 00 <48> f7 f1 ba 01 00 00 00 49 89 c5 48 c1 e8 02 48 85 c0 48 0f 44
> 
> Signed-off-by: Oleg Drokin <green@xxxxxxxxxxxxxx>
> CC: Rohit Seth <rohitseth@xxxxxxxxxx>
> ---
>  mm/page_alloc.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 5dba293..91d0265 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -5854,7 +5854,7 @@ int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
>  	int ret;
>  
>  	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
> -	if (!write || (ret < 0))
> +	if (!write || (ret < 0) || !*length)
>  		return ret;
>  
>  	mutex_lock(&pcp_batch_high_lock);

This hasn't made it to linux-next yet (probably because you didn't cc 
Andrew Morton, the mm maintainer), but I'm wondering why it's needed.  
Shouldn't this value always be >= min_percpu_pagelist_fract?

If there's something going on in proc_dointvec_minmax() that disregards 
that minimum then we need to fix it rather than the caller.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/