[PATCH] exec: don't turn PF_KTHREAD off when command not found

[PATCH] exec: don't turn PF_KTHREAD off when command not found

[KOSAKI Motohiro]

Now, do_execve() turn PF_KTHREAD off before search_binary_handler().
It has theorical risk of PF_KTHREAD lost. When ENOEXEC case, we don't
have to turn PF_KTHREAD off.

This patch move this flag modification after founding execution file.

I guess this is only theorical issue because now any kthread don't
call do_execve() directly. but anyway fixing would be better.

Cc: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
---
 fs/exec.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index ef8b9dc..81d0d06 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -1011,7 +1011,7 @@ int flush_old_exec(struct linux_binprm * bprm)
 	bprm->mm = NULL;		/* We're using it now */
 	set_exec_mm(NULL);
 
-	current->flags &= ~PF_RANDOMIZE;
+	current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
 	flush_thread();
 	current->personality &= ~bprm->per_clear;
 
@@ -1421,7 +1421,6 @@ int do_execve(const char * filename,
 	if (retval < 0)
 		goto out;
 
-	current->flags &= ~PF_KTHREAD;
 	retval = search_binary_handler(bprm,regs);
 	if (retval < 0)
 		goto out;
-- 
1.6.5.2

Re: [PATCH] exec: don't turn PF_KTHREAD off when command not found

[Oleg Nesterov]

On 09/09, KOSAKI Motohiro wrote:
>
> I guess this is only theorical issue because now any kthread don't
> call do_execve() directly.

Yes, I think that kthread should always exit if exec fails.
Otherwise we can "trust" the thread, just suppose that exec fails
after exec_mmap() succeeds.

But,

> but anyway fixing would be better.

Agreed. It looks better if we clear PF_KTHREAD after the point-of-no-return,
and the patch saves one ->flags modification.

Oleg.

Re: [PATCH] exec: don't turn PF_KTHREAD off when command not found

[Roland McGrath]

That looks reasonable to me too.


Thanks,
Roland