在 2021/9/22 13:21, Christoph Hellwig 写道:
Hi Christoph,
On Fri, Sep 10, 2021 at 05:21:20PM +0800, Zhihao Cheng wrote:
+ else {
+ /*
+ * Don't remove blk_trace under running state, in
+ * case triggering use-after-free in function
+ * __blk_add_trace().
+ */
+ if (bt->trace_state != Blktrace_running)
+ ret = blk_trace_remove_queue(q);
+ else
+ ret = -EBUSY;
+ }
goto out_unlock_bdev;
So who is going to eventually call blk_trace_free in this case?
Agree. How about removing blk_trace from running_list and stopping it
before calling blk_trace_free()?
Also Having the check in blk_trace_remove_queue would seem a little
cleaner.
Oh right, will move the check in v2.
