Re: Re: [BUG] fs/notify/mark: A potential use after free in fsnotify_put_mark_wake

[lyl2019]

> -----原始邮件-----
> 发件人: "Jan Kara" <jack@xxxxxxx>
> 发送时间: 2021-03-29 21:57:40 (星期一)
> 收件人: lyl2019@xxxxxxxxxxxxxxxx
> 抄送: jack@xxxxxxx, amir73il@xxxxxxxxx, linux-fsdevel@xxxxxxxxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx
> 主题: Re: [BUG] fs/notify/mark: A potential use after free in fsnotify_put_mark_wake
> 
> Hello!
> 
> 
> On Sun 28-03-21 17:11:43, lyl2019@xxxxxxxxxxxxxxxx wrote:
> >     My static analyzer tool reported a use after free in fsnotify_put_mark_wake
> > of the file: fs/notify/mark.c.
> > 
> > In fsnotify_put_mark_wake, it calls fsnotify_put_mark(mark). Inside the function
> > fsnotify_put_mark(), if conn is NULL, it will call fsnotify_final_mark_destroy(mark)
> > to free mark->group by fsnotify_put_group(group) and return. I also had inspected
> > the implementation of fsnotify_put_group() and found that there is no cleanup operation
> > about group->user_waits.
> > 
> > But after fsnotify_put_mark_wake() returned, mark->group is still used by 
> > if (atomic_dec_and_test(&group->user_waits) && group->shutdown) and later.
> > 
> > Is this an issue?
> 
> I don't think this scenario is possible. fsnotify_put_mark_wake() can be
> called only for marks attached to objects and these have mark->conn !=
> NULL and we are sure that fsnotify_destroy_group() will wait for all such
> marks to be torn down and freed before dropping last group reference and
> freeing the group.
> 
> 								Honza
> -- 
> Jan Kara <jack@xxxxxxxx>
> SUSE Labs, CR


Ok, thanks for your answer.