On Thu 24-08-23 10:27:46, Baokun Li wrote:
> Hello, Jan!
>
> On 2023/8/24 1:05, Jan Kara wrote:
> > On Thu 17-08-23 16:18:28, Baokun Li wrote:
> > > After 7d79b40b ("e2fsck: adjust quota counters when clearing orphaned
> > > inodes"), we load all the quotas before we process the orphaned inodes,
> > > and when we load the quotas, we check the checsum of the bbitmap for each
> > > group. If one of the bbitmap checksums is wrong, the following error will
> > > be reported:
> > >
> > > “Error initializing quota context in support library:
> > > Block bitmap checksum does not match bitmap”
> > >
> > > But loading quotas comes before checking the current superblock for the
> > > EXT2_ERROR_FS flag, which makes it impossible to use e2fsck to repair any
> > > image that contains orphan inodes and has the wrong bbitmap checksum.
> > > So delaying quota loading until after the EXT2_ERROR_FS judgment avoids
> > > the above problem.
> > >
> > > Signed-off-by: Baokun Li <libaokun1@xxxxxxxxxx>
> > This certainly looks better but I wonder if there still isn't a problem if
> > the bitmap checksums are wrong but EXT2_ERROR_FS is not set. Shouldn't we
> > rather move the initialization of the quota files after the call to
> > e2fsck_read_bitmaps()?
> >
> > Honza
> When the bitmap checksums are wrong but EXT2_ERROR_FS is not set, we must
> have lost some data (error flag or group descriptor or bitmap), so there
> is something wrong with the kernel at this time, so I don't think we
> should fix the image directly, but rather let the user realize that
> something is wrong with the filesystem logic.
I agree it means there is a problem somewhere (the storage, the kernel, or
similar). But just ignoring bitmap checksums in release_orphan_inodes() is
exactly how e2fsck behaves on filesystems without quota feature so I see no
reason for quota feature to change that because the inconsistency has
nothing to do with quotas...
> Moreover, if we don't care how this happened, but just want to fix the
> image, we only need to run "e2fsck -a" twice. After merging in the
> current patch, we always empty the orphan list before loading the quotas,
> and EXT2_ERROR_FS is set when loading the quotas fails, so this will be
> fixed the second time you run e2fsck. It will not happen that every
> e2fsck will fail like it did before.
I see, you're right so it isn't as bad as I originally thought but still my
argument above holds - IMO e2fsck should treat wrong bitmap checksums the
same way with and without the quota feature.
Honza
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
