On Wed, Apr 30, 2025 at 08:26:00AM +0000, Hans Holmberg wrote: > On 28/04/2025 17:58, Darrick J. Wong wrote: > > On Mon, Apr 28, 2025 at 12:16:34PM +0000, Hans Holmberg wrote: > >> On 25/04/2025 17:05, Darrick J. Wong wrote: > >>> ps this test should check > >>> that a readonly log device results in a norecovery mount and that > >>> pending changes don't show up if the mount succeeds? > >>> > >>> Also, ext4 supports external log devices, should this be in > >>> tests/generic? > >> > >> Doh!, actually ext4 has a test for this already, ext4/002 > >> (also based on generic/050) > >> > >> With my fix, ext4/002 passes for xfs Should/can we turn that into a > >> generic test? > > > > Yeah, it looks like ext4/002 already does most of what you want. Though > > I'd amend it to check that SCRATCH_MNT/00-99 aren't visible in the > > norecovery mounts and only appear after recovery actually runs. > > > > So I added this check to ext4/002 and while this works for xfs - the > touched files are not visible until log recovery has completed, it does > not for ext3/4. > > For ext3/4 the files are visible after the first successful (norecovery) > mount, so even though we did a shutdown, a log recovery does not seem > required (dmesg tells me that the log recovery is done in the end after > the log device is set back to rw) > > ..and I presume this is fine - for a generic test can we really assume > that a log recovery is required to see the files? Nope. I guess that's an implementation dependent behavior. TBH I'm not even sure we can 100% rely on it for xfs, since it's theoretically possible for the log to flush and checkpoint in the very small window between the creat and the shutdown call. If hoisting ext4/002 to generic works for the three main filesystems then I'm fine with just doing that without the extra tests. --D
