On 2020/4/3 16:19, Ming Lei wrote:
BTW, Yu Kuai posted one patch for this issue, looks that approach
is simpler:
https://lore.kernel.org/linux-block/20200324132315.22133-1-yukuai3@xxxxxxxxxx/
I think the issue might not be fixed with the patch seires.
At first, I think there are two key points for the issure:
1. The final release of queue is delayed in a workqueue
2. The creation of 'q->debugfs_dir' might failed(only if 1 exist)
And if we can fix any of the above problem, the UAF issue will be fixed.
(BTW, I did not come up with a good idea for problem 1, and my approach
is for problem 2.)
The third patch "block: avoid deferral of blk_release_queue() work" is
not enough to fix problem 1:
a. if CONFIG_DEBUG_KOBJECT_RELEASE is enable:
static void kobject_release(struct kref *kref)
{
struct kobject *kobj = container_of(kref, struct kobject, kref);
#ifdef CONFIG_DEBUG_KOBJECT_RELEASE
unsigned long delay = HZ + HZ * (get_random_int() & 0x3);
pr_info("kobject: '%s' (%p): %s, parent %p (delayed %ld)\n",
┊kobject_name(kobj), kobj, __func__, kobj->parent, delay);
INIT_DELAYED_WORK(&kobj->release, kobject_delayed_cleanup);
schedule_delayed_work(&kobj->release, delay);
#else
kobject_cleanup(kobj);
#endif
}
b. when 'kobject_put' is called from blk_cleanup_queue, can we make sure
it is the last reference?
Futhermore, I do understand the second patch fix the UAF problem by
using 'q->debugfs_dir' instead of 'q->blk_trace->dir', but the problem 2
still exist and need to be fixed.
Thanks!
Yu Kuai
