xfs
[Top] [All Lists]

Re: [PATCH 04/19] Make effect of PF_FSTRANS to disable __GFP_FS universa

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 04/19] Make effect of PF_FSTRANS to disable __GFP_FS universal.
From: NeilBrown <neilb@xxxxxxx>
Date: Wed, 16 Apr 2014 16:17:26 +1000
Cc: linux-mm@xxxxxxxxx, linux-nfs@xxxxxxxxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx, Ming Lei <ming.lei@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <20140416053756.GC15995@dastard>
References: <20140416033623.10604.69237.stgit@xxxxxxxxxxxxxx> <20140416040336.10604.58240.stgit@xxxxxxxxxxxxxx> <20140416053756.GC15995@dastard>
On Wed, 16 Apr 2014 15:37:56 +1000 Dave Chinner <david@xxxxxxxxxxxxx> wrote:

> On Wed, Apr 16, 2014 at 02:03:36PM +1000, NeilBrown wrote:
> > Currently both xfs and nfs will handle PF_FSTRANS by disabling
> > __GFP_FS.
> > 
> > Make this effect global by repurposing memalloc_noio_flags (which
> > does the same thing for PF_MEMALLOC_NOIO and __GFP_IO) to generally
> > impost the task flags on a gfp_t.
> > Due to this repurposing we change the name of memalloc_noio_flags
> > to gfp_from_current().
> > 
> > As PF_FSTRANS now uniformly removes __GFP_FS we can remove special
> > code for this from xfs and nfs.
> > 
> > As we can now expect other code to set PF_FSTRANS, its meaning is more
> > general, so the WARN_ON in xfs_vm_writepage() which checks PF_FSTRANS
> > is not set is no longer appropriate.  PF_FSTRANS may be set for other
> > reasons than an XFS transaction.
> 
> So PF_FSTRANS no longer means "filesystem in transaction context".
> Are you going to rename to match whatever it's meaning is now?
> I'm not exactly clear on what it means now...

I did consider renaming it to "PF_MEMALLOC_NOFS" as it is similar to
"PF_MEMALLOC_NOIO", except that it disables __GFP_FS rather than __GFP_IO.
Maybe I should go ahead with that.

> 
> 
> > As lockdep cares about __GFP_FS, we need to translate PF_FSTRANS to
> > __GFP_FS before calling lockdep_alloc_trace() in various places.
> > 
> > Signed-off-by: NeilBrown <neilb@xxxxxxx>
> ....
> > diff --git a/fs/xfs/kmem.h b/fs/xfs/kmem.h
> > index 64db0e53edea..882b86270ebe 100644
> > --- a/fs/xfs/kmem.h
> > +++ b/fs/xfs/kmem.h
> > @@ -50,8 +50,6 @@ kmem_flags_convert(xfs_km_flags_t flags)
> >             lflags = GFP_ATOMIC | __GFP_NOWARN;
> >     } else {
> >             lflags = GFP_KERNEL | __GFP_NOWARN;
> > -           if ((current->flags & PF_FSTRANS) || (flags & KM_NOFS))
> > -                   lflags &= ~__GFP_FS;
> >     }
> 
> I think KM_NOFS needs to remain here, as it has use outside of
> transaction contexts that set PF_FSTRANS....

Argh, yes of course.
I'll have to re-test the other xfs changes now to see if they are really
needed.

Thanks!


> 
> >     if (flags & KM_ZERO)
> > diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
> > index db2cfb067d0b..207a7f86d5d7 100644
> > --- a/fs/xfs/xfs_aops.c
> > +++ b/fs/xfs/xfs_aops.c
> > @@ -952,13 +952,6 @@ xfs_vm_writepage(
> >                     PF_MEMALLOC))
> >             goto redirty;
> >  
> > -   /*
> > -    * Given that we do not allow direct reclaim to call us, we should
> > -    * never be called while in a filesystem transaction.
> > -    */
> > -   if (WARN_ON(current->flags & PF_FSTRANS))
> > -           goto redirty;
> 
> We still need to ensure this rule isn't broken. If it is, the
> filesystem will silently deadlock in delayed allocation rather than
> gracefully handle the problem with a warning....

Hmm... that might be tricky.  The 'new' PF_FSTRANS can definitely be set when
xfs_vm_writepage is called and we really want the write to happen.
I don't suppose there is any other way to detect if a transaction is
happening?

Thanks,
NeilBrown

Attachment: signature.asc
Description: PGP signature

<Prev in Thread] Current Thread [Next in Thread>