Re: vmalloc with GFP_NOFS

[Steven Whitehouse <swhiteho@redhat.com>]

Hi,


On 24/04/18 17:27, Michal Hocko wrote:
> Hi,
> it seems that we still have few vmalloc users who perform GFP_NOFS
> allocation:
> drivers/mtd/ubi/io.c
> fs/ext4/xattr.c
> fs/gfs2/dir.c
> fs/gfs2/quota.c
> fs/nfs/blocklayout/extent_tree.c
> fs/ubifs/debug.c
> fs/ubifs/lprops.c
> fs/ubifs/lpt_commit.c
> fs/ubifs/orphan.c
>
> Unfortunatelly vmalloc doesn't suppoer GFP_NOFS semantinc properly
> because we do have hardocded GFP_KERNEL allocations deep inside the
> vmalloc layers. That means that if GFP_NOFS really protects from
> recursion into the fs deadlocks then the vmalloc call is broken.
>
> What to do about this? Well, there are two things. Firstly, it would be
> really great to double check whether the GFP_NOFS is really needed. I
> cannot judge that because I am not familiar with the code. It would be
> great if the respective maintainers (hopefully get_maintainer.sh pointed
> me to all relevant ones). If there is not reclaim recursion issue then
> simply use the standard vmalloc (aka GFP_KERNEL request).
For GFS2, and I suspect for other fs too, it is really needed. We don't 
want to enter reclaim while holding filesystem locks.

> If the use is really valid then we have a way to do the vmalloc
> allocation properly. We have memalloc_nofs_{save,restore} scope api. How
> does that work? You simply call memalloc_nofs_save when the reclaim
> recursion critical section starts (e.g. when you take a lock which is
> then used in the reclaim path - e.g. shrinker) and memalloc_nofs_restore
> when the critical section ends. _All_ allocations within that scope
> will get GFP_NOFS semantic automagically. If you are not sure about the
> scope itself then the easiest workaround is to wrap the vmalloc itself
> with a big fat comment that this should be revisited.
>
> Does that sound like something that can be done in a reasonable time?
> I have tried to bring this up in the past but our speed is glacial and
> there are attempts to do hacks like checking for abusers inside the
> vmalloc which is just too ugly to live.
>
> Please do not hesitate to get back to me if something is not clear.
>
> Thanks!

It would be good to fix this, and it has been known as an issue for a 
long time. We might well be able to make use of the new API though. It 
might be as simple as adding the calls when we get & release glocks, but 
I'd have to check the code to be sure,

Steve.