[PATCH 12/13] xfs_repair: don't clear . or .. in process_dir2_data

[Brian Foster]

On Thu, Mar 19, 2015 at 12:59:15PM -0500, Eric Sandeen wrote:
> On 3/19/15 12:54 PM, Brian Foster wrote:
> > On Thu, Mar 19, 2015 at 12:29:27PM -0500, Eric Sandeen wrote:
> >> On 3/19/15 11:47 AM, Brian Foster wrote:
> >>> On Tue, Mar 17, 2015 at 03:33:14PM -0500, Eric Sandeen wrote:
> >>>> process_dir2_data() has special . and .. processing; it is able
> >>>> to correct these inodes, so there is no reason to clear them.
> >>>>
> >>>> Signed-off-by: Eric Sandeen <sandeen at redhat.com>
> >>>> ---
> >>>>  repair/dir2.c |   12 ++++++++++++
> >>>>  1 files changed, 12 insertions(+), 0 deletions(-)
> >>>>
> >>>> diff --git a/repair/dir2.c b/repair/dir2.c
> >>>> index 9e6c67d..3acf71c 100644
> >>>> --- a/repair/dir2.c
> >>>> +++ b/repair/dir2.c
> >>>> @@ -1331,6 +1331,18 @@ _("entry at block %u offset %" PRIdPTR " in directory inode %" PRIu64
> >>>>  				dep->namelen = 1;
> >>>>  			clearino = 1;
> >>>>  		}
> >>>> +
> >>>> +		/*
> >>>> +		 * We have a special dot & dotdot fixer-upper below which can
> >>>> +		 * sort out the proper inode number, so don't clear it.
> >>>> +		 */
> >>>> +		if ((dep->namelen == 1 && dep->name[0] == '.') ||
> >>>> +		    (dep->namelen == 2 &&
> >>>> +		     dep->name[0] == '.' && dep->name[1] == '.')) {
> >>>> +			clearino = 0;
> >>>> +			clearreason = NULL;
> >>>> +		}
> >>>> +		    
> >>>
> >>> Whitespace damage on the blank line above.
> >>>
> >>> Seems Ok, but the question I have is what happens if the dot or dotdot
> >>> namelen was bogus? 
> >>
> >> If namelen is 1 and name[0] is '.', or
> >> if namelen is 2 and name[0] is '.' and name[1] is '..'
> >>
> >> then how can that the len be bogus?  The test is for the name being
> >> either precisely '.' or '..' and nothing else, right?
> >>
> > 
> > Ah, yeah I see. So it would be cleared in that case.
> > 
> > Note that just above if namelen == 0 we set it to 1. Would we have the
> > opposite problem for hidden files with bogus namelen (i.e., not clear
> > entries that we should)?
> 
> Hm, yeah.  Maybe moving my new hunk above that check makes sense.
> 

I think that makes sense. I guess we ultimately can't get around a file
that starts with dot looking like the dot entry if the entry is
corrupted just right, but at least we don't risk fabricating that
scenario ourselves.

Brian

> -Eric
> 
> _______________________________________________
> xfs mailing list
> xfs at oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs