sysfs: only allow one scheduled removal callback per kobj
authorAlex Chiang <achiang@hp.com>
Fri, 13 Mar 2009 18:07:36 +0000 (12:07 -0600)
committerGreg Kroah-Hartman <gregkh@suse.de>
Tue, 24 Mar 2009 23:38:26 +0000 (16:38 -0700)
commit669420644c79c207f83fdf9105ae782867e2991f
tree668491b3700bcc65e45d5ff9471f6fde5d5743af
parentffa6a7054d172a2f57248dff2de600ca795c5656
sysfs: only allow one scheduled removal callback per kobj

The only way for a sysfs attribute to remove itself (without
deadlock) is to use the sysfs_schedule_callback() interface.

Vegard Nossum discovered that a poorly written sysfs ->store
callback can repeatedly schedule remove callbacks on the same
device over and over, e.g.

$ while true ; do echo 1 > /sys/devices/.../remove ; done

If the 'remove' attribute uses the sysfs_schedule_callback API
and also does not protect itself from concurrent accesses, its
callback handler will be called multiple times, and will
eventually attempt to perform operations on a freed kobject,
leading to many problems.

Instead of requiring all callers of sysfs_schedule_callback to
implement their own synchronization, provide the protection in
the infrastructure.

Now, sysfs_schedule_callback will only allow one scheduled
callback per kobject. On subsequent calls with the same kobject,
return -EAGAIN.

This is a short term fix. The long term fix is to allow sysfs
attributes to remove themselves directly, without any of this
callback hokey pokey.

[cornelia.huck@de.ibm.com: s390 ccwgroup bits]

Reported-by: vegard.nossum@gmail.com
Signed-off-by: Alex Chiang <achiang@hp.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/s390/cio/ccwgroup.c
fs/sysfs/file.c