FFS writes to read-only mount

[Bruce Evans]

On Tue, 3 Jul 2007, David Cecil wrote:

> in March I mailed the list about errors I'm seeing that look something like 
> this:
> g_vfs_done():mirror/gmroots1f[WRITE(offset=1349091328, length=16384)]error = 
> 1
> (Note that gmroots1f is not a complete mirror set, just one disk ,and was 
> never part of a complete mirror set.  Also, partition f is the root 
> partition.)
>
> The problem comes about because the underlying geom object has had its access 
> permissions changed so that g_io_check fails due to acw being 0 (as per my 
> original message at the end of this email).  This problem is most easily 
> reproducible when during boot if the root filesystem is changed from rw to ro 
> mount.  Our code is now 6.1-based.

At least -current seems to have the opposite problem -- that acw is not
changed to 0 on mount -o ro, as shown by:

boot
use the system a bit
# fsck -p /          # fails of course since / is rw
# mount -u -o ro /   # fails with EBUSY
# # I don't want to try mount -f ... /
# kill -TERM 1       # shut down to single user mode to run fsck safely
# umount -A          # old bug: this still doesn't remount / ro
                      # not so old bug: it now tries to unmount devfs; this
 		     # fails of course
# mount -u -o ro /   # finish unmounting / (works according to mount status)
# fsck -p /          # fails, apparently since acw is still nonzero

In some non-current versions of FreeBSD, I have debugging code in
ffs_update() that complains about attempts to update inodes on read-only
file systems.  Such attempts certainly occur, due to historical mistakes.
They are supposed to be handled (starting sometime in 4.x) by silently
ignoring the problem and clearing the IN_MODIFIED flag and related flags
so that the update is not retried later.  I don't know of any cases where
this doesn't work.

Bruce