Re: [PATCH] exec: separate MM_ANONPAGES and RLIMIT_STACK accounting

[Oleg Nesterov]

Hi Guenter,

On 11/23, Guenter Roeck wrote:
>
> On Mon, Nov 12, 2018 at 05:09:10PM +0100, Oleg Nesterov wrote:
> > get_arg_page() checks bprm->rlim_stack.rlim_cur and re-calculates the
> > "extra" size for argv/envp pointers every time, this is a bit ugly and
> > even not strictly correct: acct_arg_size() must not account this size.
> >
> > Remove all the rlimit code in get_arg_page(). Instead, add bprm->argmin
> > calculated once at the start of __do_execve_file() and change copy_strings
> > to check bprm->p >= bprm->argmin.
> >
> > The patch adds the new helper, prepare_arg_pages() which initializes
> > bprm->argc/envc and bprm->argmin.
> >
> > Signed-off-by: Oleg Nesterov <oleg@xxxxxxxxxx>
> > Acked-by: Kees Cook <keescook@xxxxxxxxxxxx>
>
> This patch results in various qemu boot failures in -next. Bisect logs
> are attached. It looks like all nommu boots are failing.
                                  ^^^^^

Ah, thanks.

Yes, I forgot about the !CONFIG_MMU version of get_arg_page() which doesn't
check RLIMIT_STACK at all.

I'll send the trivial fix. Meanwile, could you test the patch below? to ensure
this is the only problem.

Oleg.

--- a/fs/exec.c
+++ b/fs/exec.c
@@ -526,8 +526,10 @@ static int copy_strings(int argc, struct user_arg_ptr argv,
 		pos = bprm->p;
 		str += len;
 		bprm->p -= len;
+#ifdef CONFIG_MMU
 		if (bprm->p < bprm->argmin)
 			goto out;
+#endif
 
 		while (len > 0) {
 			int offset, bytes_to_copy;