linux_binfmt->core_dump() runs before the process does exit_aio(), this
means that we can hit the kernel thread which shares the same ->mm.
Afaics, nothing really bad can happen, but perhaps it makes sense to fix
this minor bug.
It is sad we have to iterate over all threads in system and use
GFP_ATOMIC.  Hopefully we can kill theses ugly do_each_thread()s, but this
needs some nontrivial changes in mm_struct and do_coredump.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
        rcu_read_lock();
        do_each_thread(g, p)
                if (p->mm == dump_task->mm) {
+                       if (p->flags & PF_KTHREAD)
+                               continue;
+
                        t = kzalloc(offsetof(struct elf_thread_core_info,
                                             notes[info->thread_notes]),
                                    GFP_ATOMIC);
                rcu_read_lock();
                do_each_thread(g, p)
                        if (current->mm == p->mm && current != p) {
+                               if (p->flags & PF_KTHREAD)
+                                       continue;
+
                                ets = kzalloc(sizeof(*ets), GFP_ATOMIC);
                                if (!ets) {
                                        rcu_read_unlock();
 
                rcu_read_lock();
                do_each_thread(g,p)
                        if (current->mm == p->mm && current != p) {
+                               if (p->flags & PF_KTHREAD)
+                                       continue;
+
                                tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
                                if (!tmp) {
                                        rcu_read_unlock();