Hello,
this is the 2.4-test11-pre7 version of the ext2 COMPAT flag cleanup. A
series of issues during ext3 testing brought this out, but it is equally
a bug under 2.2 and 2.4 kernels. Basically, the ext2 COMPAT flags should
prevent a kernel from mounting a filesystem for which it doesn't understand
any INCOMPAT flags, and read-only mount a filesystem for which it doesn't
understand any RO_COMPAT flags.
On 2.2 and 2.4 kernels this isn't really true because we always set the
flags (e.g. LARGE_FILE or for ext3 RECOVERY) in the ext2 code, but these
flags are only checked at mount time if we have a rev 1 filesystem, and
are not checked at all on a remount read-write.
The patch does a few things:
- always check the compat flags, even if we have a rev 0 filesystem,
because we previously set these flags on rev 0 filesystems (BUGFIX)
- when setting a flag on a rev 0 filesystem, update it to be a rev 1
filesystem (fill in a few fields in the superblock) (BUGFIX)
- check the RO_COMPAT flags at remount time to ensure we don't read-write
mount a filesystem we don't understand (BUGFIX)
- print a warning at mount time if flags are already set on a rev 0
filesystem (e2fsck will verify if the flags are valid separately)
- removed duplicate s_compat fields from ext2_sb_info, to avoid problems
with keeping on-disk and in-memory flags in sync (this was the original
cause of the ext3 bug that started this)
- moves the SPARSE_SUPER flag checking into a helper routine that also
tells you if the group has a superblock (cleaner code)
- move s_num_db_per_group handling to its own helper routine because of
general cleanup, and future changes (per Ted and Stephen) that will
happen with group descriptor layout
The patch has been OK'd by Stephen for ext3, and no complaints at least
from Ted (he is aware of the issues).
Cheers, Andreas
--
Andreas Dilger \ "If a man ate a pound of pasta and a pound of antipasto,
\ would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/ -- Dogbert