exec: rework the group exit and fix the race with kill
authorOleg Nesterov <oleg@tv-sign.ru>
Tue, 5 Feb 2008 06:27:24 +0000 (22:27 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Tue, 5 Feb 2008 17:44:07 +0000 (09:44 -0800)
As Roland pointed out, we have the very old problem with exec.  de_thread()
sets SIGNAL_GROUP_EXIT, kills other threads, changes ->group_leader and then
clears signal->flags.  All signals (even fatal ones) sent in this window
(which is not too small) will be lost.

With this patch exec doesn't abuse SIGNAL_GROUP_EXIT.  signal_group_exit(),
the new helper, should be used to detect exit_group() or exec() in progress.
It can have more users, but this patch does only strictly necessary changes.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Robin Holt <holt@sgi.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/exec.c
include/linux/sched.h
kernel/exit.c
kernel/signal.c

index 966c5c5b6741a25af7c3708fdd3e50e5434d710b..be923e4bc38972954325aee09363cb38bff60884 100644 (file)
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -760,7 +760,7 @@ static int de_thread(struct task_struct *tsk)
         */
        read_lock(&tasklist_lock);
        spin_lock_irq(lock);
-       if (sig->flags & SIGNAL_GROUP_EXIT) {
+       if (signal_group_exit(sig)) {
                /*
                 * Another group action in progress, just
                 * return so that the signal is processed.
@@ -778,6 +778,7 @@ static int de_thread(struct task_struct *tsk)
        if (unlikely(tsk->group_leader == task_child_reaper(tsk)))
                task_active_pid_ns(tsk)->child_reaper = tsk;
 
+       sig->group_exit_task = tsk;
        zap_other_threads(tsk);
        read_unlock(&tasklist_lock);
 
@@ -802,7 +803,6 @@ static int de_thread(struct task_struct *tsk)
        }
 
        sig->notify_count = count;
-       sig->group_exit_task = tsk;
        while (atomic_read(&sig->count) > count) {
                __set_current_state(TASK_UNINTERRUPTIBLE);
                spin_unlock_irq(lock);
@@ -871,15 +871,10 @@ static int de_thread(struct task_struct *tsk)
                leader->exit_state = EXIT_DEAD;
 
                write_unlock_irq(&tasklist_lock);
-        }
+       }
 
        sig->group_exit_task = NULL;
        sig->notify_count = 0;
-       /*
-        * There may be one thread left which is just exiting,
-        * but it's safe to stop telling the group to kill themselves.
-        */
-       sig->flags = 0;
 
 no_thread_group:
        exit_itimers(sig);
@@ -1549,7 +1544,7 @@ static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
        int err = -EAGAIN;
 
        spin_lock_irq(&tsk->sighand->siglock);
-       if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
+       if (!signal_group_exit(tsk->signal)) {
                tsk->signal->group_exit_code = exit_code;
                zap_process(tsk);
                err = 0;
index 680bb03a4b9098d07f03afad76ef3b797f0487a2..483ea4e1accf3f4ffba55c76c225966491b5db53 100644 (file)
@@ -555,6 +555,13 @@ struct signal_struct {
 #define SIGNAL_STOP_CONTINUED  0x00000004 /* SIGCONT since WCONTINUED reap */
 #define SIGNAL_GROUP_EXIT      0x00000008 /* group exit in progress */
 
+/* If true, all threads except ->group_exit_task have pending SIGKILL */
+static inline int signal_group_exit(const struct signal_struct *sig)
+{
+       return  (sig->flags & SIGNAL_GROUP_EXIT) ||
+               (sig->group_exit_task != NULL);
+}
+
 /*
  * Some day this will be a full-fledged user tracking system..
  */
index 9e459fefda77191218466814a9355c85d4b09ea7..9d3d0f0b27d9a6165384d6af55013488aacfdc72 100644 (file)
@@ -1083,11 +1083,12 @@ do_group_exit(int exit_code)
                struct signal_struct *const sig = current->signal;
                struct sighand_struct *const sighand = current->sighand;
                spin_lock_irq(&sighand->siglock);
-               if (sig->flags & SIGNAL_GROUP_EXIT)
+               if (signal_group_exit(sig))
                        /* Another thread got here before we took the lock.  */
                        exit_code = sig->group_exit_code;
                else {
                        sig->group_exit_code = exit_code;
+                       sig->flags = SIGNAL_GROUP_EXIT;
                        zap_other_threads(current);
                }
                spin_unlock_irq(&sighand->siglock);
index 1117b28488c20056ae04565a667838d0baed913e..6a5f97cd337a3bc980428ea1c186efedbe991a25 100644 (file)
@@ -957,7 +957,6 @@ void zap_other_threads(struct task_struct *p)
 {
        struct task_struct *t;
 
-       p->signal->flags = SIGNAL_GROUP_EXIT;
        p->signal->group_stop_count = 0;
 
        for (t = next_thread(p); t != p; t = next_thread(t)) {
@@ -1697,7 +1696,8 @@ static int do_signal_stop(int signr)
        } else {
                struct task_struct *t;
 
-               if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
+               if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
+                   unlikely(sig->group_exit_task))
                        return 0;
                /*
                 * There is no group stop already in progress.