On Thu 20-10-16 19:04:39, Oleg Nesterov wrote:
> On 10/20, Leon Yu wrote:
> >
> > --- a/fs/proc/base.c
> > +++ b/fs/proc/base.c
> > @@ -1014,6 +1014,9 @@ static ssize_t auxv_read(struct file *file, char
> > __user *buf,
> > {
> > struct mm_struct *mm = file->private_data;
> > unsigned int nwords = 0;
> > +
> > + if (!mm)
> > + return 0;
> > do {
> > nwords += 2;
> > } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
>
> Michal disagrees and I won't argue with his patch which makes __mem_open()
> fail if ->mm == NULL. Even if I don't really understand why should we change
> the old behaviour, this _can_ break or at least confuse something/someone.
>
> However, even if we do the change above, personally I do think we should
> fix the trivial bug first, then surprise the user-space.
I am not objecting to this particular patch as it is obviously correct.
I was just hoping to have a more generic solution. The issue was
introduced in this merge window so it is not like we would need as
minimalistic approach possible. If the __mem_open approach sounds too
dangerous I will not pursue it.
Anyway feel free to add
Acked-by: Michal Hocko <mho...@suse.com>
for this patch.
--
Michal Hocko
SUSE Labs
