Re: [PATCH -tip] remove the BKL: Replace BKL in mount/umount syscalls with a mutex

[Al Viro <viro@ZenIV.linux.org.uk>]

On Thu, Apr 16, 2009 at 01:01:50PM -0400, Christoph Hellwig wrote:
> ->remount_fs should be easy enough to, we do have proper per-sb
> protection here, but do_remount_sb will need a bit of an audit.
> (and of course pushing lock_kernel down into the many instances and
> leave the cleanup-work to the fs maintainers).
> 
> The actual mount path is more interesting as there are quite a few cases
> there.  As a first step you can take lock_kernel from outside do_mount
> into the various do_foo calls inside it, and then work on those piece
> by piece.

The only place that might care is ->get_sb() (i.e. old ->read_super()).
And only for protection of fs-type-wide data structures inside the
fs/foo/* - anything in VFS doesn't give a damn (e.g. a realistic candidate
might be something that maintains a private list of all sb->s_fs_info for
this type and doesn't bother to do any locking, relying on BKL for all
manipulations).

->write_super() and ->put_super() are other candidates, for the same
reason.  That's where BKL is generic_shutdown is coming from.

Note that while we do have other users of do_kern_mount(), they tend to
be limited to subset of fs types, so again, do not assume that "we use
do_kern_mount() without BKL anyway" means that we are safe on that path.

I'd suggest pushing that crap down into individual filesystems again.
They *ARE* serialized for given superblock, so we really are looking for
cross-fs-instance data structures.