Re: [PATCH v2 1/3] common: filter aiodio dmesg after fs/iomap.c to fs/iomap/ move

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Mon, Jul 29, 2019 at 05:55:06PM -0700, Darrick J. Wong wrote:
> On Sun, Jul 28, 2019 at 07:30:36PM +0800, Eryu Guan wrote:
> > On Thu, Jul 25, 2019 at 11:03:30AM -0700, Darrick J. Wong wrote:
> > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > 
> > > Since the iomap code are moving to fs/iomap/ we have to add new entries
> > > to the aiodio dmesg filter to reflect this.
> > > 
> > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > ---
> > > v2: fix all the iomap regexes
> > > ---
> > >  common/filter |    9 +++++----
> > >  1 file changed, 5 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/common/filter b/common/filter
> > > index ed082d24..2e32ab10 100644
> > > --- a/common/filter
> > > +++ b/common/filter
> > > @@ -550,10 +550,10 @@ _filter_aiodio_dmesg()
> > >  	local warn2="WARNING:.*fs/xfs/xfs_file\.c:.*xfs_file_dio_aio_read.*"
> > >  	local warn3="WARNING:.*fs/xfs/xfs_file\.c:.*xfs_file_read_iter.*"
> > >  	local warn4="WARNING:.*fs/xfs/xfs_file\.c:.*xfs_file_aio_read.*"
> > > -	local warn5="WARNING:.*fs/iomap\.c:.*iomap_dio_rw.*"
> > > +	local warn5="WARNING:.*fs/iomap.*:.*iomap_dio_rw.*"
> > >  	local warn6="WARNING:.*fs/xfs/xfs_aops\.c:.*__xfs_get_blocks.*"
> > > -	local warn7="WARNING:.*fs/iomap\.c:.*iomap_dio_actor.*"
> > > -	local warn8="WARNING:.*fs/iomap\.c:.*iomap_dio_complete.*"
> > > +	local warn7="WARNING:.*fs/iomap.*:.*iomap_dio_actor.*"
> > > +	local warn8="WARNING:.*fs/iomap.*:.*iomap_dio_complete.*"
> > 
> > I don't think we need new filters anymore, as commit 5a9d929d6e13
> > ("iomap: report collisions between directio and buffered writes to
> > userspace") replaced the WARN_ON with a pr_crit(). These filters are
> > there only for old kernels.
> 
> Aaaaahh... but I /did/ write this patch because I kept hitting a WARNING
> somewhere in the iomap directio code, and you know what?  It's one of the
> warnings about a bogus iomap type in iomap_dio_actor.
> 
> I /think/ this is what happens when a buffered write sneaks in and
> creates a delalloc reservation after the directio write has zapped the
> page cache but before it actually starts iterating extents.
> Consequently iomap_dio_actor sees the delalloc extent and WARNs.
> 
> Will have to recheck this, but maybe the kernel needs to deploy that
> helper that 5a9d929d6e13 for that case.

Aha, I found it again.  The patch fixes failures in generic/446 when a
directio write through iomap encounters a delalloc extent and triggers
the WARN_ON_ONCE at the bottom of iomap_dio_actor:

WARNING: CPU: 2 PID: 1710922 at fs/iomap/direct-io.c:383 iomap_dio_actor+0x144/0x1a0

This can happen if a buffered write and a directio write race to fill a
hole and the buffered write manages to stuff a delalloc reservation into
the data mapping after the dio write has cleared the page cache.

We don't need the dio_warn_stale_pagecache() warning here because we
fail the direct write and therefore do not write anything to disk.

--D

> 
> --D
> 
> > Thanks,
> > Eryu
> > 
> > >  	local warn9="WARNING:.*fs/direct-io\.c:.*dio_complete.*"
> > >  	sed -e "s#$warn1#Intentional warnings in xfs_file_dio_aio_write#" \
> > >  	    -e "s#$warn2#Intentional warnings in xfs_file_dio_aio_read#" \
> > > @@ -563,7 +563,8 @@ _filter_aiodio_dmesg()
> > >  	    -e "s#$warn6#Intentional warnings in __xfs_get_blocks#" \
> > >  	    -e "s#$warn7#Intentional warnings in iomap_dio_actor#" \
> > >  	    -e "s#$warn8#Intentional warnings in iomap_dio_complete#" \
> > > -	    -e "s#$warn9#Intentional warnings in dio_complete#"
> > > +	    -e "s#$warn9#Intentional warnings in dio_complete#" \
> > > +	    -e "s#$warn10#Intentional warnings in iomap_dio_actor#"
> > >  }
> > >  
> > >  # We generate assert related WARNINGs on purpose and make sure test doesn't fail