lkml.org 
[lkml]   [2011]   [Jan]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: Important for fs devs: rcu-walk merged upstream
From
Date
On Thu, 13 Jan 2011, Miklos Szeredi wrote:
> On Sat, 8 Jan 2011, Nick Piggin wrote:
> > The vfs-scale branch is now upstream. If you haven't
> > looked yet, your filesystem is likely to have been
> > touched, so check it out.
> >
> > Also look at Documentation/filesystems/porting and
> > path-lookup.txt.
> >
> > The dcache_lock stuff should have been all done for you
> > (for in-tree filesystems, I can help out of tree fses with
> > conversions there if you ping me offline).
> >
> > The rcu-walk stuff can be more tricky for your filesystem
> > to take advantage of.
> >
> > If you supply a .d_revalidate, .permission, or .check_acl,
> > then path walking is going to be slow and unscalable for
> > you.
> >
> > Out of tree filesystems: you _have_ to at least add a line
> > of code to the above functions in order to specify that
> > you don't want to participate in rcu-walk.
> >
> > Otherwise, you don't have to care about rcu-walk if you
> > have a legacy or special filesystem like configfs then I'd
> > advise against anything fancy. But if you have a
> > userbase and you expect them to actually do any path
> > lookups into your filesystem, please take a look.
>
> One other thing: I know ECHILD is safe since no sane filesystem will
> return it in its permission or revalidate callbacks, and even if it
> does that's just a loss of optimization.

And it's not entirely safe either. A fuse filesystem returning ECHILD
would make nameidata_dentry_drop_rcu() to BUG. So some sort of errno
filter is necessary in the fuse kernel module.

Thanks,
Miklos


\
 
 \ /
  Last update: 2011-01-13 23:11    [W:0.213 / U:0.260 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site