proc: stop using BKL
authorAlexey Dobriyan <adobriyan@gmail.com>
Mon, 27 Oct 2008 19:48:36 +0000 (22:48 +0300)
committerAlexey Dobriyan <adobriyan@gmail.com>
Mon, 5 Jan 2009 09:27:44 +0000 (12:27 +0300)
commitb4df2b92d8461444fac429c75ba6e125c63056bc
tree15149c192776b817eeb6124978f74411490aa1ce
parentfe0bdec68b77020281dc814805edfe594ae89e0f
proc: stop using BKL

There are four BKL users in proc: de_put(), proc_lookup_de(),
proc_readdir_de(), proc_root_readdir(),

1) de_put()
-----------
de_put() is classic atomic_dec_and_test() refcount wrapper -- no BKL
needed. BKL doesn't matter to possible refcount leak as well.

2) proc_lookup_de()
-------------------
Walking PDE list is protected by proc_subdir_lock(), proc_get_inode() is
potentially blocking, all callers of proc_lookup_de() eventually end up
from ->lookup hooks which is protected by directory's ->i_mutex -- BKL
doesn't protect anything.

3) proc_readdir_de()
--------------------
"." and ".." part doesn't need BKL, walking PDE list is under
proc_subdir_lock, calling filldir callback is potentially blocking
because it writes to luserspace. All proc_readdir_de() callers
eventually come from ->readdir hook which is under directory's
->i_mutex -- BKL doesn't protect anything.

4) proc_root_readdir_de()
-------------------------
proc_root_readdir_de is ->readdir hook, see (3).

Since readdir hooks doesn't use BKL anymore, switch to
generic_file_llseek, since it also takes directory's i_mutex.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
fs/proc/generic.c
fs/proc/inode.c
fs/proc/proc_net.c
fs/proc/root.c