cvs commit: src/sys/fs/hpfs hpfs_vnops.c src/sys/fs/msdosfs msdosfs_denode.c src/sys/fs/nullfs null_vnops.c src/sys/fs/smbfs smbfs_node.c src/sys/gnu/ext2fs ext2_inode.c src/sys/isofs/cd9660 cd9660_node.c src/sys/kern vfs_subr.c src/sys/sys vnode.h ...

[Poul-Henning Kamp]

In message <4224BF67.1070300 at samsco.org>, Scott Long writes:
>Ceri Davies wrote:


>> We already told everyone that we wouldn't break for 5.4.
>> 
>
>All that's been talked about so far with RE with regards to 5.4 is the 
>merging of the fdesk locking, which I believe is complete now.  No 
>mention has been made of VFS locking.  If there are plans to do that, we 
>need to talk about them right away.

I reverted the vrecycle() MFC already, if we want to make the decision,
that is not what we should base it on.

If we do give up on being KLD compatible with filesystems between
5.3 and 5.4 we will not be able to bring all of the VFS work in
-current into RELENG_5, but we could bring in a lot of interesting
stuff, from my hand:

	vnode bypass for devices
	vnode bypass for fifos
	mount/nmount kernel side migration.
	root-mount simplification.

(I'll let jeffr flog his own wares.)

I personally am pretty indifferent to which path we take.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.