Re: [PATCH v4 3/6] mm: shrinkers: provide shrinkers with names

[Roman Gushchin]

On Fri, May 27, 2022 at 07:30:34PM +1000, Dave Chinner wrote:
> On Wed, May 25, 2022 at 01:25:57PM -0700, Roman Gushchin wrote:
> > Currently shrinkers are anonymous objects. For debugging purposes they
> > can be identified by count/scan function names, but it's not always
> > useful: e.g. for superblock's shrinkers it's nice to have at least
> > an idea of to which superblock the shrinker belongs.
> > 
> > This commit adds names to shrinkers. register_shrinker() and
> > prealloc_shrinker() functions are extended to take a format and
> > arguments to master a name.
> > 
> > In some cases it's not possible to determine a good name at the time
> > when a shrinker is allocated. For such cases shrinker_debugfs_rename()
> > is provided.
> > 
> > After this change the shrinker debugfs directory looks like:
> >   $ cd /sys/kernel/debug/shrinker/
> >   $ ls
> >     dqcache-16          sb-hugetlbfs-17  sb-rootfs-2      sb-tmpfs-49
> >     kfree_rcu-0         sb-hugetlbfs-33  sb-securityfs-6  sb-tracefs-13
> >     sb-aio-20           sb-iomem-12      sb-selinuxfs-22  sb-xfs:vda1-36
> >     sb-anon_inodefs-15  sb-mqueue-21     sb-sockfs-8      sb-zsmalloc-19
> >     sb-bdev-3           sb-nsfs-4        sb-sysfs-26      shadow-18
> >     sb-bpf-32           sb-pipefs-14     sb-tmpfs-1       thp_deferred_split-10
> >     sb-btrfs:vda2-24    sb-proc-25       sb-tmpfs-27      thp_zero-9
> >     sb-cgroup2-30       sb-proc-39       sb-tmpfs-29      xfs_buf-vda1-37
> >     sb-configfs-23      sb-proc-41       sb-tmpfs-35      xfs_inodegc-vda1-38
> 
> sb-xfs:vda1-36
> xfs_buf-vda1-37
> xfs_inodegc-vda1-38
> 
> That's a parsing nightmare right there. Please use the same format
> for everything. You have <subsystem>-<type>:<instance>-<id> for
> superblock stuff, but <subsys>_<type>-<instance>-<id> for the XFS
> stuff. Make it consistent so we aren't reduced to pulling out our
> hair trying to parse this in any useful way:
> 
> sb-xfs:vda1-36
> xfs-buf:vda1-37
> xfs-inodegc:vda1-38

Ok, good point, will do in the next version.

> 
> FWIW, how we are supposed to know what actually owns these:
> 
> sb-tmpfs-1
> sb-tmpfs-27
> sb-tmpfs-29
> sb-tmpfs-35
> sb-tmpfs-49
> 
> tmpfs-27 might own all the memory - how do we link that back to a
> mount point, container, user, workload, etc?

I agree, but I've no good idea what to use as an id. We can't put the mount
point, user, group etc together in the file name - it will be too lengthy
(and mount namespaces are making it even more complicated).

Maybe we can add a symlink to the mount point from within the directory?

Do you have any ideas here?

Thanks!