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

[Ingo Molnar <mingo@elte.hu>]

* Al Viro <viro@ZenIV.linux.org.uk> wrote:


> > > Old 
> > > foo_read_super/foo_write_super/foo_put_super/foo_remount_fs 
> > > for the same foo.  IOW, per-driver (and not per-fs - that's 
> > > taken care of) data structures.  Arbitrary weird ones.
[...]
> > I fear, unless i'm misunderstanding your feedback, that you are 
> > setting the purist's irrealistically high burden to get rid of the 
> > BKL from the VFS here.
> > 
> > "Arbitrary weird ones" means all BKL using sites in the kernel - all 
> > ~800 ones - up to 800x800 == close to a million interactions to 
> > check.
> 
> Sigh...  How about dumping that lovely strawman?  I've 
> exsoddingplicitly told you that all such stuff is *within* 
> *individual* *fs* *driver*.

Ah, i misunderstood: "per-driver (and not per-fs" to mean to include 
all other BKL-using Linux drivers as well ;-)

The 'not per-fs' excluded the 'fs driver' meaning (to me).

Per fs analysis is of course a must-review if we weaken or change 
locking.

> Start with taking these guys down into the superblock methods in question.
> Drop that junk on VFS-only side of things completely (mount --move,
> mount --bind, etc.).  Then we go looking for data structures that are
> 	a) internal to fs driver
> 	b) accessed by methods in question (in that fs driver)
> 	c) are shared between different filesystems.
> 
> Analysis is on per-fs basis.  And getting rid of these turds 
> doesn't have to happen in one patch.

Stupid question regarding c): wouldnt such data structures go via 
the VFS - which you said was free of BKL constraints? Or are there 
interconnected private data structures between certain types of 
closely related filesystems that the VFS does not know about? (and 
hence might have BKL assumptions)

	Ingo