On 2/11/21 6:17 PM, Darrick J. Wong wrote: > On Thu, Feb 11, 2021 at 05:29:05PM -0600, Eric Sandeen wrote: >> On 2/11/21 4:59 PM, Darrick J. Wong wrote: >>> From: Darrick J. Wong <djwong@xxxxxxxxxx> >>> >>> Quietly set up the ability to tell xfs_repair to set NEEDSREPAIR at >>> program start and (presumably) clear it by the end of the run. This >>> code isn't terribly useful to users; it's mainly here so that fstests >>> can exercise the functionality. We don't document this flag in the >>> manual pages at all because repair clears needsrepair at exit, which >>> means the knobs only exist for fstests to exercise the functionality. >>> >>> Note that we can't do any of these upgrades until we've at least done a >>> preliminary scan of the primary super and the log. >>> >>> Signed-off-by: Darrick J. Wong <djwong@xxxxxxxxxx> >>> Reviewed-by: Christoph Hellwig <hch@xxxxxx> >>> Reviewed-by: Brian Foster <bfoster@xxxxxxxxxx> >> >> >> I'm still a little on the fence about the cmdline option for crashing >> repair at a certain point from the POV that Brian kind of pointed out >> that this doesn't exactly scale as we need more hooks. > > (That's in the next patch.) I. Am. Awesome. ... > Probably yes, but ... uh I don't want this to drag on into building a > generic error injection framework for userspace. > > I would /really/ like to get inobtcount/bigtime tests into the kernel > without a giant detour they have nearly zero test coverage from the > wider community. Yeah, I dont' want that either. this (er, next patch) is s3kr1t and if we have something better later we can change it. I'll just merge stuff as-is and move forward. -Eric
