lkml.org 
[lkml]   [2008]   [Jan]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH 002 of 6] md: Fix use-after-free bug when dropping an rdev from an md array.
On Monday January 14, viro@ZenIV.linux.org.uk wrote:
> On Mon, Jan 14, 2008 at 12:45:31PM +1100, NeilBrown wrote:
> >
> > Due to possible deadlock issues we need to use a schedule work to
> > kobject_del an 'rdev' object from a different thread.
> >
> > A recent change means that kobject_add no longer gets a refernce, and
> > kobject_del doesn't put a reference. Consequently, we need to
> > explicitly hold a reference to ensure that the last reference isn't
> > dropped before the scheduled work get a chance to call kobject_del.
> >
> > Also, rename delayed_delete to md_delayed_delete to that it is more
> > obvious in a stack trace which code is to blame.
>
> I don't know... You still get kobject_del() and export_rdev()
> in unpredictable order; sure, it won't be freed under you, but...

I cannot see that that would matter.
kobject_del deletes the object from the kobj tree and free sysfs.
export_rdev disconnects the objects from md structures and releases
the connection with the device. They are quite independent.

>
> What is that deadlock problem, anyway? I don't see anything that
> would look like an obvious candidate in the stuff you are delaying...

Maybe it isn't there any more....

Once upon a time, when I
echo remove > /sys/block/mdX/md/dev-YYY/state

sysfs_write_file would hold buffer->sem while calling my store
handler.
When my store handler tried to delete the relevant kobject, it would
eventually call orphan_all_buffers which would try to take buf->sem
and deadlock.

orphan_all_buffers doesn't exist any more, so maybe the deadlock is
gone too.
However the comment at the top of sysfs_schedule_callback in
sysfs/file.c says:

*
* sysfs attribute methods must not unregister themselves or their parent
* kobject (which would amount to the same thing). Attempts to do so will
* deadlock, since unregistration is mutually exclusive with driver
* callbacks.
*

so I'm included to leave the code as it is.... ofcourse the comment
could be well out of date.

NeilBrown


\
 
 \ /
  Last update: 2008-01-14 04:25    [W:0.138 / U:0.272 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site