Re: [RFC] mm: allow to decrease swap.max below actual swap usage

[Roman Gushchin]

On Tue, Mar 20, 2018 at 10:35:43PM +0000, Roman Gushchin wrote:
> Currently an attempt to set swap.max into a value lower
> than the actual swap usage fails. And a user can't do much
> with it, except turning off swap globally (using swapoff).
> 
> This patch aims to fix this issue by allowing setting swap.max
> into any value (which corresponds to cgroup v2 API design),
> and schedule a background job to fit swap size into the new limit.
> 
> The following script can be used to test the memory.swap behavior:
>   #!/bin/bash
> 
>   mkdir -p /sys/fs/cgroup/test_swap
>   echo 100M > /sys/fs/cgroup/test_swap/memory.max
>   echo max > /sys/fs/cgroup/test_swap/memory.swap.max
> 
>   mkdir -p /sys/fs/cgroup/test_swap_2
>   echo 100M > /sys/fs/cgroup/test_swap_2/memory.max
>   echo max > /sys/fs/cgroup/test_swap_2/memory.swap.max
> 
>   echo $$ > /sys/fs/cgroup/test_swap/cgroup.procs
>   allocate 200M &
> 
>   echo $$ > /sys/fs/cgroup/test_swap_2/cgroup.procs
>   allocate 200M &
> 
>   sleep 2
> 
>   cat /sys/fs/cgroup/test_swap/memory.swap.current
>   cat /sys/fs/cgroup/test_swap_2/memory.swap.current
> 
>   echo max > /sys/fs/cgroup/test_swap/memory.max
>   echo 50M > /sys/fs/cgroup/test_swap/memory.swap.max
> 
>   sleep 10
> 
>   cat /sys/fs/cgroup/test_swap/memory.swap.current
>   cat /sys/fs/cgroup/test_swap_2/memory.swap.current
> 
>   pkill allocate
> 
> Original test results:
>   106024960
>   106348544
>   ./swap.sh: line 23: echo: write error: Device or resource busy
>   106024960
>   106348544
> 
> With this patch applied:
>   106045440
>   106352640
>   52428800
>   106201088

Any comments, thoughts, feedback?

Rebased version below.

---