fasync: re-organize fasync entry insertion to allow it under a spinlock
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 27 Oct 2010 16:38:12 +0000 (12:38 -0400)
committerArnd Bergmann <arnd@arndb.de>
Wed, 27 Oct 2010 20:06:17 +0000 (22:06 +0200)
commitf7347ce4ee7c65415f84be915c018473e7076f31
tree613ce14f088ad00bdbc77cdfb686a40a4851180f
parentc5b1f0d92c36851aca09ac6c7c0c4f9690ac14f3
fasync: re-organize fasync entry insertion to allow it under a spinlock

You currently cannot use "fasync_helper()" in an atomic environment to
insert a new fasync entry, because it will need to allocate the new
"struct fasync_struct".

Yet fcntl_setlease() wants to call this under lock_flocks(), which is in
the process of being converted from the BKL to a spinlock.

In order to fix this, this abstracts out the actual fasync list
insertion and the fasync allocations into functions of their own, and
teaches fs/locks.c to pre-allocate the fasync_struct entry.  That way
the actual list insertion can happen while holding the required
spinlock.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[bfields@redhat.com: rebase on top of my changes to Arnd's patch]
Tested-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
fs/fcntl.c
fs/locks.c
include/linux/fs.h