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

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

On Fri, Apr 17, 2009 at 06:56:43PM +0200, Ingo Molnar wrote:

> 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)

Shared between different instances of this fs type.  E.g. if FAT had
its hashtable not per-superblock (as it is now) but system-wide (as it
used to be prior to [PATCH] FAT: the inode hash from per module to per sb),
it would be suspicious.  In this particular case, it was OK from the
very beginning (I'd used a system-wide spinlock to protect it when it
had been introduced and per-superblock splitup made that spinlock per-sb
along with the hash table).

IOW, I wouldn't particulary worry about interactions between different
fs _types_ - it's interactions between the different fs _instances_
within the same driver that are likely sources of PITA.

ObOtherQuestion: no, I don't think that bdev open and bdev ioctls situation
is related to this group of BKL users.  sb_set_blocksize() might, in principle,
want a memory barrier of some kind, but that's it.