Re: [PATCH v2] proc: proc_pid_ns takes super_block as an argument

[Eric W. Biederman]

Alexey Gladkov <gladkov.alexey@xxxxxxxxx> writes:

> On Mon, May 18, 2020 at 07:08:57AM -0500, Eric W. Biederman wrote:
>> Alexey Gladkov <gladkov.alexey@xxxxxxxxx> writes:
>> 
>> > The proc_pid_ns() can be used for both inode and dentry. To avoid making
>> > two identical functions, change the argument type of the proc_pid_ns().
>> >
>> > Link: https://lore.kernel.org/lkml/c3461e26-1407-2262-c709-dac0df3da2d0@xxxxxxxxxxxxxxxxxxx/
>> > Reported-by: syzbot+c1af344512918c61362c@xxxxxxxxxxxxxxxxxxxxxxxxx
>> > Signed-off-by: Alexey Gladkov <gladkov.alexey@xxxxxxxxx>
>> 
>> So overall this looks good.
>> 
>> However, the description leaves a little bit to be desired as it does
>> not describe why it is bad to use dentry->d_sb.  A fixes tag would be
>> nice if for no other reason than to help anyone who decides to backport
>> this.
>
> OK. I will add it.

Thank you.  It really helps to have the full description of why in
the commit comments.

>> And can you please compile test this?
>> 
>> There is a very silly typo in proc that keeps this from compiling.
>
> I compiled the kernel with this patch and ran the kernel, but accidentally
> did not check children_seq_show(). Sorry.

Yes, children_seq_show is behind a sneaky CONFIG option.

Eric