Re: [BUG] How to crash 4.9.2 x86_64: vmscan: shrink_slab

[Michal Hocko]

On Mon 09-01-17 23:02:10, Sami Farin wrote:
> # sysctl vm.vfs_cache_pressure=-100
> 
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6640827866535449472
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6640827866535450112
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-661702561611775889
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6640827866535442432
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6562613194205300197
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6640827866535439872
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-659655090764208789
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6564660665198832072
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6562613194351275164
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6562615996648922728
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6564660665198832072
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-6562613194351264981
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-569296135781119076
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-565206492037048430
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-565212096665106188
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-569296135781119076
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-565206492037043196
> kernel: vmscan: shrink_slab: super_cache_scan+0x0/0x1a0 negative objects to delete nr=-659660388715270673
> 
> 
> Alternatively,
> # sysctl vm.vfs_cache_pressure=10000000

Both values are insane and admins do not do insane things to their
machines, do they?

I am not sure how much we want to check the input value. -100 is clearly
bogus and 

		.procname	= "vfs_cache_pressure",
		.data		= &sysctl_vfs_cache_pressure,
		.maxlen		= sizeof(sysctl_vfs_cache_pressure),
		.mode		= 0644,
		.proc_handler	= &proc_dointvec,
		.extra1		= &zero,

tries to enforce min (extra1) check except proc_dointvec doesn't care
about this... This is news to me. Only proc_dointvec_minmax does care
about extra*, it seems.
-- 
Michal Hocko
SUSE Labs