Re: [PATCH 19/47] xfs: add tracepoints and error injection for deferred extent freeing

[Brian Foster <bfoster@redhat.com>]

On Tue, Aug 02, 2016 at 01:24:25PM -0700, Darrick J. Wong wrote:
> On Tue, Aug 02, 2016 at 02:48:00PM -0400, Brian Foster wrote:
> > On Wed, Jul 20, 2016 at 09:58:10PM -0700, Darrick J. Wong wrote:
> > > Add a couple of tracepoints for the deferred extent free operation and
> > > a site for injecting errors while finishing the operation.  This makes
> > > it easier to debug deferred ops and test log redo.
> > > 
> > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > ---
> > >  fs/xfs/libxfs/xfs_alloc.c |    7 +++++++
> > >  fs/xfs/libxfs/xfs_bmap.c  |    2 ++
> > >  fs/xfs/xfs_error.h        |    4 +++-
> > >  fs/xfs/xfs_trace.h        |    5 ++++-
> > >  4 files changed, 16 insertions(+), 2 deletions(-)
> > > 
> > > 
> > > diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c
> > > index 5993f87..22ac3f1 100644
> > > --- a/fs/xfs/libxfs/xfs_alloc.c
> > > +++ b/fs/xfs/libxfs/xfs_alloc.c
> > > @@ -2702,6 +2702,13 @@ xfs_free_extent(
> > >  
> > >  	ASSERT(len != 0);
> > >  
> > > +	trace_xfs_bmap_free_deferred(mp, agno, 0, agbno, len);
> > > +
> > 
> > Hmm, but the bmap code isn't the only caller here. I was thinking that
> > maybe we'd be better served by pushing these down into xfs_defer_add()
> > and friends, but I guess we don't necessarily have the extent
> > information at that layer.
> 
> Correct, we don't.
> 
> > How about we just rename these tracepoints to match the function names
> > so they don't confuse me in the future? :)
> 
> I broke with the function name convention so that the entry and exit
> tracepoints to deferred items would all have names that easily matched
> wildcards to make my life easier:
> 
> # trace-cmd record -e 'xfs_*_defer' -e 'xfs_*_deferred' -F ...
> 

The xfs_bmap_free_defer tp seems reasonable, but xfs_bmap_free_deferred
doesn't really make sense when called via other places where nothing has
been deferred.

I'm not totally sure I follow what the goal is with this tracepoint wrt
to naming that the trace_xfs_defer_*() tracepoints don't already achieve
(combined with standard infrastructure tracepoints), but if the name is
important, we should at least put it somewhere where it's only triggered
via a deferred operation (e.g., xfs_extent_free_finish_item()).

> > > +	if (XFS_TEST_ERROR(false, mp,
> > > +			XFS_ERRTAG_FREE_EXTENT,
> > > +			XFS_RANDOM_FREE_EXTENT))
> > > +		return -EIO;
> > > +
> > >  	error = xfs_free_extent_fix_freelist(tp, agno, &agbp);
> > >  	if (error)
> > >  		return error;
> > > diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c
> > > index 540a6b7..8e14ff4 100644
> > > --- a/fs/xfs/libxfs/xfs_bmap.c
> > > +++ b/fs/xfs/libxfs/xfs_bmap.c
> > > @@ -596,6 +596,8 @@ xfs_bmap_add_free(
> > >  	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
> > >  	new->xefi_startblock = bno;
> > >  	new->xefi_blockcount = (xfs_extlen_t)len;
> > > +	trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
> > > +			XFS_FSB_TO_AGBNO(mp, bno), len);
> > >  	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
> > >  }
> > >  
> > > diff --git a/fs/xfs/xfs_error.h b/fs/xfs/xfs_error.h
> > > index 2e4f67f..da6f951 100644
> > > --- a/fs/xfs/xfs_error.h
> > > +++ b/fs/xfs/xfs_error.h
> > > @@ -90,7 +90,8 @@ extern void xfs_verifier_error(struct xfs_buf *bp);
> > >  #define XFS_ERRTAG_STRATCMPL_IOERR			19
> > >  #define XFS_ERRTAG_DIOWRITE_IOERR			20
> > >  #define XFS_ERRTAG_BMAPIFORMAT				21
> > > -#define XFS_ERRTAG_MAX					22
> > > +#define XFS_ERRTAG_FREE_EXTENT				22
> > > +#define XFS_ERRTAG_MAX					23
> > >  
> > >  /*
> > >   * Random factors for above tags, 1 means always, 2 means 1/2 time, etc.
> > > @@ -117,6 +118,7 @@ extern void xfs_verifier_error(struct xfs_buf *bp);
> > >  #define XFS_RANDOM_STRATCMPL_IOERR			(XFS_RANDOM_DEFAULT/10)
> > >  #define XFS_RANDOM_DIOWRITE_IOERR			(XFS_RANDOM_DEFAULT/10)
> > >  #define	XFS_RANDOM_BMAPIFORMAT				XFS_RANDOM_DEFAULT
> > > +#define XFS_RANDOM_FREE_EXTENT				1
> > >  
> > 
> > Why not XFS_RANDOM_DEFAULT?
> 
> I don't want the injected redo errors going off at random -- I want to set
> one and have it go off the very next time we hit it.  This way I can write
> the xfstests like so:
> 
> mkfs/mount
> write to a file; sync
> set a logic bomb for the next time we free an extent
> truncate the file
> <watch bomb go off>
> unmount
> check that xfs_repair grumbles about the dirty log
> mount
> check the file
> 
> Were the free-extent bomb set to go off at XFS_RANDOM_DEFAULT intervals, it'd
> only have a 1/100 chance of triggering, which makes it harder to test.
> 

Ok.

Brian

> --D
> 
> > 
> > Brian
> > 
> > >  #ifdef DEBUG
> > >  extern int xfs_error_test_active;
> > > diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h
> > > index a45b030..939caf5 100644
> > > --- a/fs/xfs/xfs_trace.h
> > > +++ b/fs/xfs/xfs_trace.h
> > > @@ -2417,9 +2417,12 @@ DEFINE_DEFER_PENDING_EVENT(xfs_defer_pending_cancel);
> > >  DEFINE_DEFER_PENDING_EVENT(xfs_defer_pending_finish);
> > >  DEFINE_DEFER_PENDING_EVENT(xfs_defer_pending_abort);
> > >  
> > > -DEFINE_PHYS_EXTENT_DEFERRED_EVENT(xfs_defer_phys_extent);
> > >  DEFINE_MAP_EXTENT_DEFERRED_EVENT(xfs_defer_map_extent);
> > >  
> > > +#define DEFINE_BMAP_FREE_DEFERRED_EVENT DEFINE_PHYS_EXTENT_DEFERRED_EVENT
> > > +DEFINE_BMAP_FREE_DEFERRED_EVENT(xfs_bmap_free_defer);
> > > +DEFINE_BMAP_FREE_DEFERRED_EVENT(xfs_bmap_free_deferred);
> > > +
> > >  #endif /* _TRACE_XFS_H */
> > >  
> > >  #undef TRACE_INCLUDE_PATH
> > > 
> > > _______________________________________________
> > > xfs mailing list
> > > xfs@oss.sgi.com
> > > http://oss.sgi.com/mailman/listinfo/xfs