[Top] [All Lists]

Re: [PATCH 1/4] xfs: PF_FSTRANS should never be set in ->writepage

To: Alex Elder <aelder@xxxxxxx>
Subject: Re: [PATCH 1/4] xfs: PF_FSTRANS should never be set in ->writepage
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Wed, 25 May 2011 03:46:37 -0400
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
In-reply-to: <1306289929.2823.120.camel@doink>
References: <20110428125546.696493391@xxxxxxxxxxxxxxxxxxxxxx> <20110428130514.146517168@xxxxxxxxxxxxxxxxxxxxxx> <1306289929.2823.120.camel@doink>
User-agent: Mutt/1.5.21 (2010-09-15)
On Tue, May 24, 2011 at 09:18:49PM -0500, Alex Elder wrote:
> On Thu, 2011-04-28 at 08:55 -0400, Christoph Hellwig wrote:
> > Now that we reject direct reclaim in addition to always using GFP_NOFS
> > allocation there's no chance we'll ever end up in ->writepage with
> > PF_FSTRANS set.  Add a WARN_ON if we hit this case, and stop checking
> > if we'd actually need to start a transaction.
> > 
> > Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> Do the radix_tree_preload(GFP_KERNEL) calls in
> xfs_iget_cache_miss() and xfs_mru_cache_insert()
> pose any risk here?  (I haven't really looked
> closely, I just noticed that these were cases we
> did not use GFP_NOFS.)

They don't, given that we don't allow reclaim to proceed into
->writepage any more.

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