string: factorize skip_spaces and export it to be generally available
[safe/jmp/linux-2.6] / kernel / signal.c
index 0a8b0ae..6b982f2 100644 (file)
 #include <linux/ptrace.h>
 #include <linux/signal.h>
 #include <linux/signalfd.h>
+#include <linux/ratelimit.h>
+#include <linux/tracehook.h>
 #include <linux/capability.h>
 #include <linux/freezer.h>
 #include <linux/pid_namespace.h>
 #include <linux/nsproxy.h>
+#define CREATE_TRACE_POINTS
+#include <trace/events/signal.h>
 
 #include <asm/param.h>
 #include <asm/uaccess.h>
 
 static struct kmem_cache *sigqueue_cachep;
 
-static int __sig_ignored(struct task_struct *t, int sig)
+int print_fatal_signals __read_mostly;
+
+static void __user *sig_handler(struct task_struct *t, int sig)
 {
-       void __user *handler;
+       return t->sighand->action[sig - 1].sa.sa_handler;
+}
 
+static int sig_handler_ignored(void __user *handler, int sig)
+{
        /* Is it explicitly or implicitly ignored? */
-
-       handler = t->sighand->action[sig - 1].sa.sa_handler;
        return handler == SIG_IGN ||
                (handler == SIG_DFL && sig_kernel_ignore(sig));
 }
 
-static int sig_ignored(struct task_struct *t, int sig)
+static int sig_task_ignored(struct task_struct *t, int sig,
+               int from_ancestor_ns)
 {
-       /*
-        * Tracers always want to know about signals..
-        */
-       if (t->ptrace & PT_PTRACED)
-               return 0;
+       void __user *handler;
 
+       handler = sig_handler(t, sig);
+
+       if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
+                       handler == SIG_DFL && !from_ancestor_ns)
+               return 1;
+
+       return sig_handler_ignored(handler, sig);
+}
+
+static int sig_ignored(struct task_struct *t, int sig, int from_ancestor_ns)
+{
        /*
         * Blocked signals are never ignored, since the
         * signal handler may change by the time it is
@@ -66,7 +81,13 @@ static int sig_ignored(struct task_struct *t, int sig)
        if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
                return 0;
 
-       return __sig_ignored(t, sig);
+       if (!sig_task_ignored(t, sig, from_ancestor_ns))
+               return 0;
+
+       /*
+        * Tracers may want to know about even ignored signals.
+        */
+       return !tracehook_consider_ignored_signal(t, sig);
 }
 
 /*
@@ -129,7 +150,9 @@ void recalc_sigpending_and_wake(struct task_struct *t)
 
 void recalc_sigpending(void)
 {
-       if (!recalc_sigpending_tsk(current) && !freezing(current))
+       if (unlikely(tracehook_force_sigpending()))
+               set_thread_flag(TIF_SIGPENDING);
+       else if (!recalc_sigpending_tsk(current) && !freezing(current))
                clear_thread_flag(TIF_SIGPENDING);
 
 }
@@ -140,7 +163,7 @@ int next_signal(struct sigpending *pending, sigset_t *mask)
 {
        unsigned long i, *s, *m, x;
        int sig = 0;
-       
+
        s = pending->signal.sig;
        m = mask->sig;
        switch (_NSIG_WORDS) {
@@ -165,35 +188,62 @@ int next_signal(struct sigpending *pending, sigset_t *mask)
                        sig = ffz(~x) + 1;
                break;
        }
-       
+
        return sig;
 }
 
-static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
-                                        int override_rlimit)
+static inline void print_dropped_signal(int sig)
+{
+       static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
+
+       if (!print_fatal_signals)
+               return;
+
+       if (!__ratelimit(&ratelimit_state))
+               return;
+
+       printk(KERN_INFO "%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
+                               current->comm, current->pid, sig);
+}
+
+/*
+ * allocate a new signal queue record
+ * - this may be called without locks if and only if t == current, otherwise an
+ *   appopriate lock must be held to stop the target task from exiting
+ */
+static struct sigqueue *
+__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
 {
        struct sigqueue *q = NULL;
        struct user_struct *user;
 
        /*
-        * In order to avoid problems with "switch_user()", we want to make
-        * sure that the compiler doesn't re-load "t->user"
+        * We won't get problems with the target's UID changing under us
+        * because changing it requires RCU be used, and if t != current, the
+        * caller must be holding the RCU readlock (by way of a spinlock) and
+        * we use RCU protection here
         */
-       user = t->user;
-       barrier();
+       user = get_uid(__task_cred(t)->user);
        atomic_inc(&user->sigpending);
+
        if (override_rlimit ||
            atomic_read(&user->sigpending) <=
-                       t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
+                       t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur) {
                q = kmem_cache_alloc(sigqueue_cachep, flags);
+       } else {
+               print_dropped_signal(sig);
+       }
+
        if (unlikely(q == NULL)) {
                atomic_dec(&user->sigpending);
+               free_uid(user);
        } else {
                INIT_LIST_HEAD(&q->list);
                q->flags = 0;
-               q->user = get_uid(user);
+               q->user = user;
        }
-       return(q);
+
+       return q;
 }
 
 static void __sigqueue_free(struct sigqueue *q)
@@ -220,17 +270,56 @@ void flush_sigqueue(struct sigpending *queue)
 /*
  * Flush all pending signals for a task.
  */
+void __flush_signals(struct task_struct *t)
+{
+       clear_tsk_thread_flag(t, TIF_SIGPENDING);
+       flush_sigqueue(&t->pending);
+       flush_sigqueue(&t->signal->shared_pending);
+}
+
 void flush_signals(struct task_struct *t)
 {
        unsigned long flags;
 
        spin_lock_irqsave(&t->sighand->siglock, flags);
-       clear_tsk_thread_flag(t, TIF_SIGPENDING);
-       flush_sigqueue(&t->pending);
-       flush_sigqueue(&t->signal->shared_pending);
+       __flush_signals(t);
        spin_unlock_irqrestore(&t->sighand->siglock, flags);
 }
 
+static void __flush_itimer_signals(struct sigpending *pending)
+{
+       sigset_t signal, retain;
+       struct sigqueue *q, *n;
+
+       signal = pending->signal;
+       sigemptyset(&retain);
+
+       list_for_each_entry_safe(q, n, &pending->list, list) {
+               int sig = q->info.si_signo;
+
+               if (likely(q->info.si_code != SI_TIMER)) {
+                       sigaddset(&retain, sig);
+               } else {
+                       sigdelset(&signal, sig);
+                       list_del_init(&q->list);
+                       __sigqueue_free(q);
+               }
+       }
+
+       sigorsets(&pending->signal, &signal, &retain);
+}
+
+void flush_itimer_signals(void)
+{
+       struct task_struct *tsk = current;
+       unsigned long flags;
+
+       spin_lock_irqsave(&tsk->sighand->siglock, flags);
+       __flush_itimer_signals(&tsk->pending);
+       __flush_itimer_signals(&tsk->signal->shared_pending);
+       spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
+}
+
 void ignore_signals(struct task_struct *t)
 {
        int i;
@@ -261,12 +350,12 @@ flush_signal_handlers(struct task_struct *t, int force_default)
 
 int unhandled_signal(struct task_struct *tsk, int sig)
 {
+       void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
        if (is_global_init(tsk))
                return 1;
-       if (tsk->ptrace & PT_PTRACED)
+       if (handler != SIG_IGN && handler != SIG_DFL)
                return 0;
-       return (tsk->sighand->action[sig-1].sa.sa_handler == SIG_IGN) ||
-               (tsk->sighand->action[sig-1].sa.sa_handler == SIG_DFL);
+       return !tracehook_consider_fatal_signal(tsk, sig);
 }
 
 
@@ -304,13 +393,9 @@ unblock_all_signals(void)
        spin_unlock_irqrestore(&current->sighand->siglock, flags);
 }
 
-static int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
+static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
 {
        struct sigqueue *q, *first = NULL;
-       int still_pending = 0;
-
-       if (unlikely(!sigismember(&list->signal, sig)))
-               return 0;
 
        /*
         * Collect the siginfo appropriate to this signal.  Check if
@@ -318,33 +403,30 @@ static int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
        */
        list_for_each_entry(q, &list->list, list) {
                if (q->info.si_signo == sig) {
-                       if (first) {
-                               still_pending = 1;
-                               break;
-                       }
+                       if (first)
+                               goto still_pending;
                        first = q;
                }
        }
+
+       sigdelset(&list->signal, sig);
+
        if (first) {
+still_pending:
                list_del_init(&first->list);
                copy_siginfo(info, &first->info);
                __sigqueue_free(first);
-               if (!still_pending)
-                       sigdelset(&list->signal, sig);
        } else {
-
                /* Ok, it wasn't in the queue.  This must be
                   a fast-pathed signal or we must have been
                   out of queue space.  So zero out the info.
                 */
-               sigdelset(&list->signal, sig);
                info->si_signo = sig;
                info->si_errno = 0;
                info->si_code = 0;
                info->si_pid = 0;
                info->si_uid = 0;
        }
-       return 1;
 }
 
 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
@@ -362,8 +444,7 @@ static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
                        }
                }
 
-               if (!collect_signal(sig, pending, info))
-                       sig = 0;
+               collect_signal(sig, pending, info);
        }
 
        return sig;
@@ -428,8 +509,7 @@ int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
                 * is to alert stop-signal processing code when another
                 * processor has come along and cleared the flag.
                 */
-               if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT))
-                       tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
+               tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
        }
        if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
                /*
@@ -529,10 +609,13 @@ static int rm_from_queue(unsigned long mask, struct sigpending *s)
 
 /*
  * Bad permissions for sending the signal
+ * - the caller must hold at least the RCU read lock
  */
 static int check_kill_permission(int sig, struct siginfo *info,
                                 struct task_struct *t)
 {
+       const struct cred *cred = current_cred(), *tcred;
+       struct pid *sid;
        int error;
 
        if (!valid_signal(sig))
@@ -545,37 +628,49 @@ static int check_kill_permission(int sig, struct siginfo *info,
        if (error)
                return error;
 
-       if (((sig != SIGCONT) || (task_session_nr(current) != task_session_nr(t)))
-           && (current->euid ^ t->suid) && (current->euid ^ t->uid)
-           && (current->uid ^ t->suid) && (current->uid ^ t->uid)
-           && !capable(CAP_KILL))
-               return -EPERM;
+       tcred = __task_cred(t);
+       if ((cred->euid ^ tcred->suid) &&
+           (cred->euid ^ tcred->uid) &&
+           (cred->uid  ^ tcred->suid) &&
+           (cred->uid  ^ tcred->uid) &&
+           !capable(CAP_KILL)) {
+               switch (sig) {
+               case SIGCONT:
+                       sid = task_session(t);
+                       /*
+                        * We don't return the error if sid == NULL. The
+                        * task was unhashed, the caller must notice this.
+                        */
+                       if (!sid || sid == task_session(current))
+                               break;
+               default:
+                       return -EPERM;
+               }
+       }
 
        return security_task_kill(t, info, sig, 0);
 }
 
-/* forward decl */
-static void do_notify_parent_cldstop(struct task_struct *tsk, int why);
-
 /*
- * Handle magic process-wide effects of stop/continue signals.
- * Unlike the signal actions, these happen immediately at signal-generation
+ * Handle magic process-wide effects of stop/continue signals. Unlike
+ * the signal actions, these happen immediately at signal-generation
  * time regardless of blocking, ignoring, or handling.  This does the
  * actual continuing for SIGCONT, but not the actual stopping for stop
- * signals.  The process stop is done as a signal action for SIG_DFL.
+ * signals. The process stop is done as a signal action for SIG_DFL.
+ *
+ * Returns true if the signal should be actually delivered, otherwise
+ * it should be dropped.
  */
-static void handle_stop_signal(int sig, struct task_struct *p)
+static int prepare_signal(int sig, struct task_struct *p, int from_ancestor_ns)
 {
        struct signal_struct *signal = p->signal;
        struct task_struct *t;
 
-       if (signal->flags & SIGNAL_GROUP_EXIT)
+       if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
                /*
-                * The process is in the middle of dying already.
+                * The process is in the middle of dying, nothing to do.
                 */
-               return;
-
-       if (sig_kernel_stop(sig)) {
+       } else if (sig_kernel_stop(sig)) {
                /*
                 * This is a stop signal.  Remove SIGCONT from all queues.
                 */
@@ -632,6 +727,11 @@ static void handle_stop_signal(int sig, struct task_struct *p)
                        why |= SIGNAL_CLD_STOPPED;
 
                if (why) {
+                       /*
+                        * The first thread which returns from do_signal_stop()
+                        * will take ->siglock, notice SIGNAL_CLD_MASK, and
+                        * notify its parent. See get_signal_to_deliver().
+                        */
                        signal->flags = why | SIGNAL_STOP_CONTINUED;
                        signal->group_stop_count = 0;
                        signal->group_exit_code = 0;
@@ -643,13 +743,106 @@ static void handle_stop_signal(int sig, struct task_struct *p)
                         */
                        signal->flags &= ~SIGNAL_STOP_DEQUEUED;
                }
-       } else if (sig == SIGKILL) {
+       }
+
+       return !sig_ignored(p, sig, from_ancestor_ns);
+}
+
+/*
+ * Test if P wants to take SIG.  After we've checked all threads with this,
+ * it's equivalent to finding no threads not blocking SIG.  Any threads not
+ * blocking SIG were ruled out because they are not running and already
+ * have pending signals.  Such threads will dequeue from the shared queue
+ * as soon as they're available, so putting the signal on the shared queue
+ * will be equivalent to sending it to one such thread.
+ */
+static inline int wants_signal(int sig, struct task_struct *p)
+{
+       if (sigismember(&p->blocked, sig))
+               return 0;
+       if (p->flags & PF_EXITING)
+               return 0;
+       if (sig == SIGKILL)
+               return 1;
+       if (task_is_stopped_or_traced(p))
+               return 0;
+       return task_curr(p) || !signal_pending(p);
+}
+
+static void complete_signal(int sig, struct task_struct *p, int group)
+{
+       struct signal_struct *signal = p->signal;
+       struct task_struct *t;
+
+       /*
+        * Now find a thread we can wake up to take the signal off the queue.
+        *
+        * If the main thread wants the signal, it gets first crack.
+        * Probably the least surprising to the average bear.
+        */
+       if (wants_signal(sig, p))
+               t = p;
+       else if (!group || thread_group_empty(p))
                /*
-                * Make sure that any pending stop signal already dequeued
-                * is undone by the wakeup for SIGKILL.
+                * There is just one thread and it does not need to be woken.
+                * It will dequeue unblocked signals before it runs again.
                 */
-               signal->flags &= ~SIGNAL_STOP_DEQUEUED;
+               return;
+       else {
+               /*
+                * Otherwise try to find a suitable thread.
+                */
+               t = signal->curr_target;
+               while (!wants_signal(sig, t)) {
+                       t = next_thread(t);
+                       if (t == signal->curr_target)
+                               /*
+                                * No thread needs to be woken.
+                                * Any eligible threads will see
+                                * the signal in the queue soon.
+                                */
+                               return;
+               }
+               signal->curr_target = t;
+       }
+
+       /*
+        * Found a killable thread.  If the signal will be fatal,
+        * then start taking the whole group down immediately.
+        */
+       if (sig_fatal(p, sig) &&
+           !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
+           !sigismember(&t->real_blocked, sig) &&
+           (sig == SIGKILL ||
+            !tracehook_consider_fatal_signal(t, sig))) {
+               /*
+                * This signal will be fatal to the whole group.
+                */
+               if (!sig_kernel_coredump(sig)) {
+                       /*
+                        * Start a group exit and wake everybody up.
+                        * This way we don't have other threads
+                        * running and doing things after a slower
+                        * thread has the fatal signal pending.
+                        */
+                       signal->flags = SIGNAL_GROUP_EXIT;
+                       signal->group_exit_code = sig;
+                       signal->group_stop_count = 0;
+                       t = p;
+                       do {
+                               sigaddset(&t->pending.signal, SIGKILL);
+                               signal_wake_up(t, 1);
+                       } while_each_thread(p, t);
+                       return;
+               }
        }
+
+       /*
+        * The signal is already in the shared-pending queue.
+        * Tell the chosen thread to wake up and dequeue it.
+        */
+       signal_wake_up(t, sig == SIGKILL);
+       return;
 }
 
 static inline int legacy_queue(struct sigpending *signals, int sig)
@@ -657,25 +850,28 @@ static inline int legacy_queue(struct sigpending *signals, int sig)
        return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
 }
 
-static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
-                       struct sigpending *signals)
+static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
+                       int group, int from_ancestor_ns)
 {
-       struct sigqueue * q = NULL;
+       struct sigpending *pending;
+       struct sigqueue *q;
+       int override_rlimit;
+
+       trace_signal_generate(sig, info, t);
+
+       assert_spin_locked(&t->sighand->siglock);
 
+       if (!prepare_signal(sig, t, from_ancestor_ns))
+               return 0;
+
+       pending = group ? &t->signal->shared_pending : &t->pending;
        /*
         * Short-circuit ignored signals and support queuing
         * exactly one non-rt signal, so that we can get more
         * detailed information about the cause of the signal.
         */
-       if (sig_ignored(t, sig) || legacy_queue(signals, sig))
+       if (legacy_queue(pending, sig))
                return 0;
-
-       /*
-        * Deliver the signal to listening signalfds. This must be called
-        * with the sighand lock held.
-        */
-       signalfd_notify(t, sig);
-
        /*
         * fast-pathed signals for kernel-internal things like SIGSTOP
         * or SIGKILL.
@@ -691,18 +887,23 @@ static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
           make sure at least one signal gets delivered and don't
           pass on the info struct.  */
 
-       q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
-                                            (is_si_special(info) ||
-                                             info->si_code >= 0)));
+       if (sig < SIGRTMIN)
+               override_rlimit = (is_si_special(info) || info->si_code >= 0);
+       else
+               override_rlimit = 0;
+
+       q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
+               override_rlimit);
        if (q) {
-               list_add_tail(&q->list, &signals->list);
+               list_add_tail(&q->list, &pending->list);
                switch ((unsigned long) info) {
                case (unsigned long) SEND_SIG_NOINFO:
                        q->info.si_signo = sig;
                        q->info.si_errno = 0;
                        q->info.si_code = SI_USER;
-                       q->info.si_pid = task_pid_vnr(current);
-                       q->info.si_uid = current->uid;
+                       q->info.si_pid = task_tgid_nr_ns(current,
+                                                       task_active_pid_ns(t));
+                       q->info.si_uid = current_uid();
                        break;
                case (unsigned long) SEND_SIG_PRIV:
                        q->info.si_signo = sig;
@@ -713,23 +914,48 @@ static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
                        break;
                default:
                        copy_siginfo(&q->info, info);
+                       if (from_ancestor_ns)
+                               q->info.si_pid = 0;
                        break;
                }
        } else if (!is_si_special(info)) {
-               if (sig >= SIGRTMIN && info->si_code != SI_USER)
-               /*
-                * Queue overflow, abort.  We may abort if the signal was rt
-                * and sent by user using something other than kill().
-                */
+               if (sig >= SIGRTMIN && info->si_code != SI_USER) {
+                       /*
+                        * Queue overflow, abort.  We may abort if the
+                        * signal was rt and sent by user using something
+                        * other than kill().
+                        */
+                       trace_signal_overflow_fail(sig, group, info);
                        return -EAGAIN;
+               } else {
+                       /*
+                        * This is a silent loss of information.  We still
+                        * send the signal, but the *info bits are lost.
+                        */
+                       trace_signal_lose_info(sig, group, info);
+               }
        }
 
 out_set:
-       sigaddset(&signals->signal, sig);
-       return 1;
+       signalfd_notify(t, sig);
+       sigaddset(&pending->signal, sig);
+       complete_signal(sig, t, group);
+       return 0;
 }
 
-int print_fatal_signals;
+static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
+                       int group)
+{
+       int from_ancestor_ns = 0;
+
+#ifdef CONFIG_PID_NS
+       if (!is_si_special(info) && SI_FROMUSER(info) &&
+                       task_pid_nr_ns(current, task_active_pid_ns(t)) <= 0)
+               from_ancestor_ns = 1;
+#endif
+
+       return __send_signal(sig, info, t, group, from_ancestor_ns);
+}
 
 static void print_fatal_signal(struct pt_regs *regs, int signr)
 {
@@ -749,7 +975,9 @@ static void print_fatal_signal(struct pt_regs *regs, int signr)
        }
 #endif
        printk("\n");
+       preempt_disable();
        show_regs(regs);
+       preempt_enable();
 }
 
 static int __init setup_print_fatal_signals(char *str)
@@ -761,181 +989,73 @@ static int __init setup_print_fatal_signals(char *str)
 
 __setup("print-fatal-signals=", setup_print_fatal_signals);
 
-static int
-specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
-{
-       int ret;
-
-       BUG_ON(!irqs_disabled());
-       assert_spin_locked(&t->sighand->siglock);
-
-       ret = send_signal(sig, info, t, &t->pending);
-       if (ret <= 0)
-               return ret;
-
-       if (!sigismember(&t->blocked, sig))
-               signal_wake_up(t, sig == SIGKILL);
-       return 0;
-}
-
-/*
- * Force a signal that the process can't ignore: if necessary
- * we unblock the signal and change any SIG_IGN to SIG_DFL.
- *
- * Note: If we unblock the signal, we always reset it to SIG_DFL,
- * since we do not want to have a signal handler that was blocked
- * be invoked when user space had explicitly blocked it.
- *
- * We don't want to have recursive SIGSEGV's etc, for example.
- */
 int
-force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
-{
-       unsigned long int flags;
-       int ret, blocked, ignored;
-       struct k_sigaction *action;
-
-       spin_lock_irqsave(&t->sighand->siglock, flags);
-       action = &t->sighand->action[sig-1];
-       ignored = action->sa.sa_handler == SIG_IGN;
-       blocked = sigismember(&t->blocked, sig);
-       if (blocked || ignored) {
-               action->sa.sa_handler = SIG_DFL;
-               if (blocked) {
-                       sigdelset(&t->blocked, sig);
-                       recalc_sigpending_and_wake(t);
-               }
-       }
-       ret = specific_send_sig_info(sig, info, t);
-       spin_unlock_irqrestore(&t->sighand->siglock, flags);
-
-       return ret;
-}
-
-void
-force_sig_specific(int sig, struct task_struct *t)
-{
-       force_sig_info(sig, SEND_SIG_FORCED, t);
-}
-
-/*
- * Test if P wants to take SIG.  After we've checked all threads with this,
- * it's equivalent to finding no threads not blocking SIG.  Any threads not
- * blocking SIG were ruled out because they are not running and already
- * have pending signals.  Such threads will dequeue from the shared queue
- * as soon as they're available, so putting the signal on the shared queue
- * will be equivalent to sending it to one such thread.
- */
-static inline int wants_signal(int sig, struct task_struct *p)
-{
-       if (sigismember(&p->blocked, sig))
-               return 0;
-       if (p->flags & PF_EXITING)
-               return 0;
-       if (sig == SIGKILL)
-               return 1;
-       if (task_is_stopped_or_traced(p))
-               return 0;
-       return task_curr(p) || !signal_pending(p);
-}
-
-static void
-__group_complete_signal(int sig, struct task_struct *p)
+__group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
 {
-       struct task_struct *t;
-
-       /*
-        * Now find a thread we can wake up to take the signal off the queue.
-        *
-        * If the main thread wants the signal, it gets first crack.
-        * Probably the least surprising to the average bear.
-        */
-       if (wants_signal(sig, p))
-               t = p;
-       else if (thread_group_empty(p))
-               /*
-                * There is just one thread and it does not need to be woken.
-                * It will dequeue unblocked signals before it runs again.
-                */
-               return;
-       else {
-               /*
-                * Otherwise try to find a suitable thread.
-                */
-               t = p->signal->curr_target;
-               if (t == NULL)
-                       /* restart balancing at this thread */
-                       t = p->signal->curr_target = p;
-
-               while (!wants_signal(sig, t)) {
-                       t = next_thread(t);
-                       if (t == p->signal->curr_target)
-                               /*
-                                * No thread needs to be woken.
-                                * Any eligible threads will see
-                                * the signal in the queue soon.
-                                */
-                               return;
-               }
-               p->signal->curr_target = t;
-       }
-
-       /*
-        * Found a killable thread.  If the signal will be fatal,
-        * then start taking the whole group down immediately.
-        */
-       if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
-           !sigismember(&t->real_blocked, sig) &&
-           (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
-               /*
-                * This signal will be fatal to the whole group.
-                */
-               if (!sig_kernel_coredump(sig)) {
-                       /*
-                        * Start a group exit and wake everybody up.
-                        * This way we don't have other threads
-                        * running and doing things after a slower
-                        * thread has the fatal signal pending.
-                        */
-                       p->signal->flags = SIGNAL_GROUP_EXIT;
-                       p->signal->group_exit_code = sig;
-                       p->signal->group_stop_count = 0;
-                       t = p;
-                       do {
-                               sigaddset(&t->pending.signal, SIGKILL);
-                               signal_wake_up(t, 1);
-                       } while_each_thread(p, t);
-                       return;
-               }
+       return send_signal(sig, info, p, 1);
+}
+
+static int
+specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
+{
+       return send_signal(sig, info, t, 0);
+}
+
+int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
+                       bool group)
+{
+       unsigned long flags;
+       int ret = -ESRCH;
+
+       if (lock_task_sighand(p, &flags)) {
+               ret = send_signal(sig, info, p, group);
+               unlock_task_sighand(p, &flags);
        }
 
-       /*
-        * The signal is already in the shared-pending queue.
-        * Tell the chosen thread to wake up and dequeue it.
-        */
-       signal_wake_up(t, sig == SIGKILL);
-       return;
+       return ret;
 }
 
+/*
+ * Force a signal that the process can't ignore: if necessary
+ * we unblock the signal and change any SIG_IGN to SIG_DFL.
+ *
+ * Note: If we unblock the signal, we always reset it to SIG_DFL,
+ * since we do not want to have a signal handler that was blocked
+ * be invoked when user space had explicitly blocked it.
+ *
+ * We don't want to have recursive SIGSEGV's etc, for example,
+ * that is why we also clear SIGNAL_UNKILLABLE.
+ */
 int
-__group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
+force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
 {
-       int ret;
+       unsigned long int flags;
+       int ret, blocked, ignored;
+       struct k_sigaction *action;
 
-       assert_spin_locked(&p->sighand->siglock);
-       handle_stop_signal(sig, p);
+       spin_lock_irqsave(&t->sighand->siglock, flags);
+       action = &t->sighand->action[sig-1];
+       ignored = action->sa.sa_handler == SIG_IGN;
+       blocked = sigismember(&t->blocked, sig);
+       if (blocked || ignored) {
+               action->sa.sa_handler = SIG_DFL;
+               if (blocked) {
+                       sigdelset(&t->blocked, sig);
+                       recalc_sigpending_and_wake(t);
+               }
+       }
+       if (action->sa.sa_handler == SIG_DFL)
+               t->signal->flags &= ~SIGNAL_UNKILLABLE;
+       ret = specific_send_sig_info(sig, info, t);
+       spin_unlock_irqrestore(&t->sighand->siglock, flags);
 
-       /*
-        * Put this signal on the shared-pending queue, or fail with EAGAIN.
-        * We always use the shared queue for process-wide signals,
-        * to avoid several races.
-        */
-       ret = send_signal(sig, info, p, &p->signal->shared_pending);
-       if (ret <= 0)
-               return ret;
+       return ret;
+}
 
-       __group_complete_signal(sig, p);
-       return 0;
+void
+force_sig_specific(int sig, struct task_struct *t)
+{
+       force_sig_info(sig, SEND_SIG_FORCED, t);
 }
 
 /*
@@ -960,12 +1080,6 @@ void zap_other_threads(struct task_struct *p)
        }
 }
 
-int __fatal_signal_pending(struct task_struct *tsk)
-{
-       return sigismember(&tsk->pending.signal, SIGKILL);
-}
-EXPORT_SYMBOL(__fatal_signal_pending);
-
 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
 {
        struct sighand_struct *sighand;
@@ -986,20 +1100,16 @@ struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long
        return sighand;
 }
 
+/*
+ * send signal info to all the members of a group
+ * - the caller must hold the RCU read lock at least
+ */
 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
 {
-       unsigned long flags;
-       int ret;
+       int ret = check_kill_permission(sig, info, p);
 
-       ret = check_kill_permission(sig, info, p);
-
-       if (!ret && sig) {
-               ret = -ESRCH;
-               if (lock_task_sighand(p, &flags)) {
-                       ret = __group_send_sig_info(sig, info, p);
-                       unlock_task_sighand(p, &flags);
-               }
-       }
+       if (!ret && sig)
+               ret = do_send_sig_info(sig, info, p, true);
 
        return ret;
 }
@@ -1007,8 +1117,8 @@ int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
 /*
  * __kill_pgrp_info() sends a signal to a process group: this is what the tty
  * control characters do (^C, ^Z etc)
+ * - the caller must hold at least a readlock on tasklist_lock
  */
-
 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
 {
        struct task_struct *p = NULL;
@@ -1064,6 +1174,7 @@ int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
 {
        int ret = -EINVAL;
        struct task_struct *p;
+       const struct cred *pcred;
 
        if (!valid_signal(sig))
                return ret;
@@ -1074,9 +1185,11 @@ int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
                ret = -ESRCH;
                goto out_unlock;
        }
-       if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
-           && (euid != p->suid) && (euid != p->uid)
-           && (uid != p->suid) && (uid != p->uid)) {
+       pcred = __task_cred(p);
+       if ((info == SEND_SIG_NOINFO ||
+            (!is_si_special(info) && SI_FROMUSER(info))) &&
+           euid != pcred->suid && euid != pcred->uid &&
+           uid  != pcred->suid && uid  != pcred->uid) {
                ret = -EPERM;
                goto out_unlock;
        }
@@ -1086,7 +1199,7 @@ int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
        if (sig && p->sighand) {
                unsigned long flags;
                spin_lock_irqsave(&p->sighand->siglock, flags);
-               ret = __group_send_sig_info(sig, info, p);
+               ret = __send_signal(sig, info, p, 1, 0);
                spin_unlock_irqrestore(&p->sighand->siglock, flags);
        }
 out_unlock:
@@ -1102,7 +1215,7 @@ EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
  * is probably wrong.  Should make it like BSD or SYSV.
  */
 
-static int kill_something_info(int sig, struct siginfo *info, int pid)
+static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
 {
        int ret;
 
@@ -1122,7 +1235,8 @@ static int kill_something_info(int sig, struct siginfo *info, int pid)
                struct task_struct * p;
 
                for_each_process(p) {
-                       if (p->pid > 1 && !same_thread_group(p, current)) {
+                       if (task_pid_vnr(p) > 1 &&
+                                       !same_thread_group(p, current)) {
                                int err = group_send_sig_info(sig, info, p);
                                ++count;
                                if (err != -EPERM)
@@ -1140,16 +1254,9 @@ static int kill_something_info(int sig, struct siginfo *info, int pid)
  * These are for backward compatibility with the rest of the kernel source.
  */
 
-/*
- * These two are the most common entry points.  They send a signal
- * just to the specific thread.
- */
 int
 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
 {
-       int ret;
-       unsigned long flags;
-
        /*
         * Make sure legacy kernel users don't send in bad values
         * (normal paths check this in check_kill_permission).
@@ -1157,18 +1264,7 @@ send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
        if (!valid_signal(sig))
                return -EINVAL;
 
-       /*
-        * We need the tasklist lock even for the specific
-        * thread case (when we don't need to follow the group
-        * lists) in order to avoid races with "p->sighand"
-        * going away or changing from under us.
-        */
-       read_lock(&tasklist_lock);  
-       spin_lock_irqsave(&p->sighand->siglock, flags);
-       ret = specific_send_sig_info(sig, info, p);
-       spin_unlock_irqrestore(&p->sighand->siglock, flags);
-       read_unlock(&tasklist_lock);
-       return ret;
+       return do_send_sig_info(sig, info, p, false);
 }
 
 #define __si_special(priv) \
@@ -1223,34 +1319,23 @@ int kill_pid(struct pid *pid, int sig, int priv)
 }
 EXPORT_SYMBOL(kill_pid);
 
-int
-kill_proc(pid_t pid, int sig, int priv)
-{
-       int ret;
-
-       rcu_read_lock();
-       ret = kill_pid_info(sig, __si_special(priv), find_pid(pid));
-       rcu_read_unlock();
-       return ret;
-}
-
 /*
  * These functions support sending signals using preallocated sigqueue
  * structures.  This is needed "because realtime applications cannot
  * afford to lose notifications of asynchronous events, like timer
- * expirations or I/O completions".  In the case of Posix Timers 
+ * expirations or I/O completions".  In the case of Posix Timers
  * we allocate the sigqueue structure from the timer_create.  If this
  * allocation fails we are able to report the failure to the application
  * with an EAGAIN error.
  */
 struct sigqueue *sigqueue_alloc(void)
 {
-       struct sigqueue *q;
+       struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
 
-       if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
+       if (q)
                q->flags |= SIGQUEUE_PREALLOC;
-       return(q);
+
+       return q;
 }
 
 void sigqueue_free(struct sigqueue *q)
@@ -1260,118 +1345,84 @@ void sigqueue_free(struct sigqueue *q)
 
        BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
        /*
-        * If the signal is still pending remove it from the
-        * pending queue. We must hold ->siglock while testing
-        * q->list to serialize with collect_signal().
+        * We must hold ->siglock while testing q->list
+        * to serialize with collect_signal() or with
+        * __exit_signal()->flush_sigqueue().
         */
        spin_lock_irqsave(lock, flags);
+       q->flags &= ~SIGQUEUE_PREALLOC;
+       /*
+        * If it is queued it will be freed when dequeued,
+        * like the "regular" sigqueue.
+        */
        if (!list_empty(&q->list))
-               list_del_init(&q->list);
+               q = NULL;
        spin_unlock_irqrestore(lock, flags);
 
-       q->flags &= ~SIGQUEUE_PREALLOC;
-       __sigqueue_free(q);
+       if (q)
+               __sigqueue_free(q);
 }
 
-static int do_send_sigqueue(int sig, struct sigqueue *q, struct task_struct *t,
-               struct sigpending *pending)
+int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
 {
+       int sig = q->info.si_signo;
+       struct sigpending *pending;
+       unsigned long flags;
+       int ret;
+
+       BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
+
+       ret = -1;
+       if (!likely(lock_task_sighand(t, &flags)))
+               goto ret;
+
+       ret = 1; /* the signal is ignored */
+       if (!prepare_signal(sig, t, 0))
+               goto out;
+
+       ret = 0;
        if (unlikely(!list_empty(&q->list))) {
                /*
                 * If an SI_TIMER entry is already queue just increment
                 * the overrun count.
                 */
-
                BUG_ON(q->info.si_code != SI_TIMER);
                q->info.si_overrun++;
-               return 0;
+               goto out;
        }
-
-       if (sig_ignored(t, sig))
-               return 1;
+       q->info.si_overrun = 0;
 
        signalfd_notify(t, sig);
+       pending = group ? &t->signal->shared_pending : &t->pending;
        list_add_tail(&q->list, &pending->list);
        sigaddset(&pending->signal, sig);
-       return 0;
-}
-
-int send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
-{
-       unsigned long flags;
-       int ret = -1;
-
-       BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
-
-       /*
-        * The rcu based delayed sighand destroy makes it possible to
-        * run this without tasklist lock held. The task struct itself
-        * cannot go away as create_timer did get_task_struct().
-        *
-        * We return -1, when the task is marked exiting, so
-        * posix_timer_event can redirect it to the group leader
-        */
-       if (!likely(lock_task_sighand(p, &flags)))
-               goto out_err;
-
-       ret = do_send_sigqueue(sig, q, p, &p->pending);
-
-       if (!sigismember(&p->blocked, sig))
-               signal_wake_up(p, sig == SIGKILL);
-
-       unlock_task_sighand(p, &flags);
-out_err:
-       return ret;
-}
-
-int
-send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
-{
-       unsigned long flags;
-       int ret;
-
-       BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
-
-       read_lock(&tasklist_lock);
-       /* Since it_lock is held, p->sighand cannot be NULL. */
-       spin_lock_irqsave(&p->sighand->siglock, flags);
-       handle_stop_signal(sig, p);
-
-       ret = do_send_sigqueue(sig, q, p, &p->signal->shared_pending);
-
-       __group_complete_signal(sig, p);
-
-       spin_unlock_irqrestore(&p->sighand->siglock, flags);
-       read_unlock(&tasklist_lock);
+       complete_signal(sig, t, group);
+out:
+       unlock_task_sighand(t, &flags);
+ret:
        return ret;
 }
 
 /*
- * Wake up any threads in the parent blocked in wait* syscalls.
- */
-static inline void __wake_up_parent(struct task_struct *p,
-                                   struct task_struct *parent)
-{
-       wake_up_interruptible_sync(&parent->signal->wait_chldexit);
-}
-
-/*
  * Let a parent know about the death of a child.
  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
+ *
+ * Returns -1 if our parent ignored us and so we've switched to
+ * self-reaping, or else @sig.
  */
-
-void do_notify_parent(struct task_struct *tsk, int sig)
+int do_notify_parent(struct task_struct *tsk, int sig)
 {
        struct siginfo info;
        unsigned long flags;
        struct sighand_struct *psig;
+       int ret = sig;
 
        BUG_ON(sig == -1);
 
        /* do_notify_parent_cldstop should have been called instead.  */
        BUG_ON(task_is_stopped_or_traced(tsk));
 
-       BUG_ON(!tsk->ptrace &&
+       BUG_ON(!task_ptrace(tsk) &&
               (tsk->group_leader != tsk || !thread_group_empty(tsk)));
 
        info.si_signo = sig;
@@ -1390,15 +1441,13 @@ void do_notify_parent(struct task_struct *tsk, int sig)
         */
        rcu_read_lock();
        info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
+       info.si_uid = __task_cred(tsk)->uid;
        rcu_read_unlock();
 
-       info.si_uid = tsk->uid;
-
-       /* FIXME: find out whether or not this is supposed to be c*time. */
-       info.si_utime = cputime_to_jiffies(cputime_add(tsk->utime,
-                                                      tsk->signal->utime));
-       info.si_stime = cputime_to_jiffies(cputime_add(tsk->stime,
-                                                      tsk->signal->stime));
+       info.si_utime = cputime_to_clock_t(cputime_add(tsk->utime,
+                               tsk->signal->utime));
+       info.si_stime = cputime_to_clock_t(cputime_add(tsk->stime,
+                               tsk->signal->stime));
 
        info.si_status = tsk->exit_code & 0x7f;
        if (tsk->exit_code & 0x80)
@@ -1412,7 +1461,7 @@ void do_notify_parent(struct task_struct *tsk, int sig)
 
        psig = tsk->parent->sighand;
        spin_lock_irqsave(&psig->siglock, flags);
-       if (!tsk->ptrace && sig == SIGCHLD &&
+       if (!task_ptrace(tsk) && sig == SIGCHLD &&
            (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
             (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
                /*
@@ -1430,14 +1479,16 @@ void do_notify_parent(struct task_struct *tsk, int sig)
                 * is implementation-defined: we do (if you don't want
                 * it, just use SIG_IGN instead).
                 */
-               tsk->exit_signal = -1;
+               ret = tsk->exit_signal = -1;
                if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
-                       sig = 0;
+                       sig = -1;
        }
        if (valid_signal(sig) && sig > 0)
                __group_send_sig_info(sig, &info, tsk->parent);
        __wake_up_parent(tsk, tsk->parent);
        spin_unlock_irqrestore(&psig->siglock, flags);
+
+       return ret;
 }
 
 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
@@ -1447,7 +1498,7 @@ static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
        struct task_struct *parent;
        struct sighand_struct *sighand;
 
-       if (tsk->ptrace & PT_PTRACED)
+       if (task_ptrace(tsk))
                parent = tsk->parent;
        else {
                tsk = tsk->group_leader;
@@ -1460,14 +1511,12 @@ static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
         * see comment in do_notify_parent() abot the following 3 lines
         */
        rcu_read_lock();
-       info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
+       info.si_pid = task_pid_nr_ns(tsk, parent->nsproxy->pid_ns);
+       info.si_uid = __task_cred(tsk)->uid;
        rcu_read_unlock();
 
-       info.si_uid = tsk->uid;
-
-       /* FIXME: find out whether or not this is supposed to be c*time. */
-       info.si_utime = cputime_to_jiffies(tsk->utime);
-       info.si_stime = cputime_to_jiffies(tsk->stime);
+       info.si_utime = cputime_to_clock_t(tsk->utime);
+       info.si_stime = cputime_to_clock_t(tsk->stime);
 
        info.si_code = why;
        switch (why) {
@@ -1498,7 +1547,7 @@ static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
 
 static inline int may_ptrace_stop(void)
 {
-       if (!likely(current->ptrace & PT_PTRACED))
+       if (!likely(task_ptrace(current)))
                return 0;
        /*
         * Are we in the middle of do_coredump?
@@ -1506,10 +1555,10 @@ static inline int may_ptrace_stop(void)
         * is a deadlock situation, and pointless because our tracer
         * is dead so don't allow us to stop.
         * If SIGKILL was already sent before the caller unlocked
-        * ->siglock we must see ->core_waiters != 0. Otherwise it
+        * ->siglock we must see ->core_state != NULL. Otherwise it
         * is safe to enter schedule().
         */
-       if (unlikely(current->mm->core_waiters) &&
+       if (unlikely(current->mm->core_state) &&
            unlikely(current->mm == current->parent->mm))
                return 0;
 
@@ -1522,9 +1571,8 @@ static inline int may_ptrace_stop(void)
  */
 static int sigkill_pending(struct task_struct *tsk)
 {
-       return ((sigismember(&tsk->pending.signal, SIGKILL) ||
-                sigismember(&tsk->signal->shared_pending.signal, SIGKILL)) &&
-               !unlikely(sigismember(&tsk->blocked, SIGKILL)));
+       return  sigismember(&tsk->pending.signal, SIGKILL) ||
+               sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
 }
 
 /*
@@ -1540,8 +1588,6 @@ static int sigkill_pending(struct task_struct *tsk)
  */
 static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
 {
-       int killed = 0;
-
        if (arch_ptrace_stop_needed(exit_code, info)) {
                /*
                 * The arch code has something special to do before a
@@ -1557,7 +1603,8 @@ static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
                spin_unlock_irq(&current->sighand->siglock);
                arch_ptrace_stop(exit_code, info);
                spin_lock_irq(&current->sighand->siglock);
-               killed = sigkill_pending(current);
+               if (sigkill_pending(current))
+                       return;
        }
 
        /*
@@ -1574,9 +1621,17 @@ static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
        __set_current_state(TASK_TRACED);
        spin_unlock_irq(&current->sighand->siglock);
        read_lock(&tasklist_lock);
-       if (!unlikely(killed) && may_ptrace_stop()) {
+       if (may_ptrace_stop()) {
                do_notify_parent_cldstop(current, CLD_TRAPPED);
+               /*
+                * Don't want to allow preemption here, because
+                * sys_ptrace() needs this task to be inactive.
+                *
+                * XXX: implement read_unlock_no_resched().
+                */
+               preempt_disable();
                read_unlock(&tasklist_lock);
+               preempt_enable_no_resched();
                schedule();
        } else {
                /*
@@ -1622,7 +1677,7 @@ void ptrace_notify(int exit_code)
        info.si_signo = SIGTRAP;
        info.si_code = exit_code;
        info.si_pid = task_pid_vnr(current);
-       info.si_uid = current->uid;
+       info.si_uid = current_uid();
 
        /* Let the debugger run.  */
        spin_lock_irq(&current->sighand->siglock);
@@ -1630,29 +1685,6 @@ void ptrace_notify(int exit_code)
        spin_unlock_irq(&current->sighand->siglock);
 }
 
-static void
-finish_stop(int stop_count)
-{
-       /*
-        * If there are no other threads in the group, or if there is
-        * a group stop in progress and we are the last to stop,
-        * report to the parent.  When ptraced, every thread reports itself.
-        */
-       if (stop_count == 0 || (current->ptrace & PT_PTRACED)) {
-               read_lock(&tasklist_lock);
-               do_notify_parent_cldstop(current, CLD_STOPPED);
-               read_unlock(&tasklist_lock);
-       }
-
-       do {
-               schedule();
-       } while (try_to_freeze());
-       /*
-        * Now we don't run again until continued.
-        */
-       current->exit_code = 0;
-}
-
 /*
  * This performs the stopping for SIGSTOP and other stop signals.
  * We have to stop all threads in the thread group.
@@ -1662,15 +1694,9 @@ finish_stop(int stop_count)
 static int do_signal_stop(int signr)
 {
        struct signal_struct *sig = current->signal;
-       int stop_count;
+       int notify;
 
-       if (sig->group_stop_count > 0) {
-               /*
-                * There is a group stop in progress.  We don't need to
-                * start another one.
-                */
-               stop_count = --sig->group_stop_count;
-       } else {
+       if (!sig->group_stop_count) {
                struct task_struct *t;
 
                if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
@@ -1682,7 +1708,7 @@ static int do_signal_stop(int signr)
                 */
                sig->group_exit_code = signr;
 
-               stop_count = 0;
+               sig->group_stop_count = 1;
                for (t = next_thread(current); t != current; t = next_thread(t))
                        /*
                         * Setting state to TASK_STOPPED for a group
@@ -1691,26 +1717,51 @@ static int do_signal_stop(int signr)
                         */
                        if (!(t->flags & PF_EXITING) &&
                            !task_is_stopped_or_traced(t)) {
-                               stop_count++;
+                               sig->group_stop_count++;
                                signal_wake_up(t, 0);
                        }
-               sig->group_stop_count = stop_count;
        }
+       /*
+        * If there are no other threads in the group, or if there is
+        * a group stop in progress and we are the last to stop, report
+        * to the parent.  When ptraced, every thread reports itself.
+        */
+       notify = sig->group_stop_count == 1 ? CLD_STOPPED : 0;
+       notify = tracehook_notify_jctl(notify, CLD_STOPPED);
+       /*
+        * tracehook_notify_jctl() can drop and reacquire siglock, so
+        * we keep ->group_stop_count != 0 before the call. If SIGCONT
+        * or SIGKILL comes in between ->group_stop_count == 0.
+        */
+       if (sig->group_stop_count) {
+               if (!--sig->group_stop_count)
+                       sig->flags = SIGNAL_STOP_STOPPED;
+               current->exit_code = sig->group_exit_code;
+               __set_current_state(TASK_STOPPED);
+       }
+       spin_unlock_irq(&current->sighand->siglock);
 
-       if (stop_count == 0)
-               sig->flags = SIGNAL_STOP_STOPPED;
-       current->exit_code = sig->group_exit_code;
-       __set_current_state(TASK_STOPPED);
+       if (notify) {
+               read_lock(&tasklist_lock);
+               do_notify_parent_cldstop(current, notify);
+               read_unlock(&tasklist_lock);
+       }
+
+       /* Now we don't run again until woken by SIGCONT or SIGKILL */
+       do {
+               schedule();
+       } while (try_to_freeze());
+
+       tracehook_finish_jctl();
+       current->exit_code = 0;
 
-       spin_unlock_irq(&current->sighand->siglock);
-       finish_stop(stop_count);
        return 1;
 }
 
 static int ptrace_signal(int signr, siginfo_t *info,
                         struct pt_regs *regs, void *cookie)
 {
-       if (!(current->ptrace & PT_PTRACED))
+       if (!task_ptrace(current))
                return signr;
 
        ptrace_signal_deliver(regs, cookie);
@@ -1734,7 +1785,7 @@ static int ptrace_signal(int signr, siginfo_t *info,
                info->si_errno = 0;
                info->si_code = SI_USER;
                info->si_pid = task_pid_vnr(current->parent);
-               info->si_uid = current->parent->uid;
+               info->si_uid = task_uid(current->parent);
        }
 
        /* If the (new) signal is now blocked, requeue it.  */
@@ -1763,16 +1814,24 @@ relock:
        try_to_freeze();
 
        spin_lock_irq(&sighand->siglock);
-
+       /*
+        * Every stopped thread goes here after wakeup. Check to see if
+        * we should notify the parent, prepare_signal(SIGCONT) encodes
+        * the CLD_ si_code into SIGNAL_CLD_MASK bits.
+        */
        if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
                int why = (signal->flags & SIGNAL_STOP_CONTINUED)
                                ? CLD_CONTINUED : CLD_STOPPED;
                signal->flags &= ~SIGNAL_CLD_MASK;
+
+               why = tracehook_notify_jctl(why, CLD_CONTINUED);
                spin_unlock_irq(&sighand->siglock);
 
-               read_lock(&tasklist_lock);
-               do_notify_parent_cldstop(current->group_leader, why);
-               read_unlock(&tasklist_lock);
+               if (why) {
+                       read_lock(&tasklist_lock);
+                       do_notify_parent_cldstop(current->group_leader, why);
+                       read_unlock(&tasklist_lock);
+               }
                goto relock;
        }
 
@@ -1783,17 +1842,36 @@ relock:
                    do_signal_stop(0))
                        goto relock;
 
-               signr = dequeue_signal(current, &current->blocked, info);
-               if (!signr)
-                       break; /* will return 0 */
+               /*
+                * Tracing can induce an artifical signal and choose sigaction.
+                * The return value in @signr determines the default action,
+                * but @info->si_signo is the signal number we will report.
+                */
+               signr = tracehook_get_signal(current, regs, info, return_ka);
+               if (unlikely(signr < 0))
+                       goto relock;
+               if (unlikely(signr != 0))
+                       ka = return_ka;
+               else {
+                       signr = dequeue_signal(current, &current->blocked,
+                                              info);
 
-               if (signr != SIGKILL) {
-                       signr = ptrace_signal(signr, info, regs, cookie);
                        if (!signr)
-                               continue;
+                               break; /* will return 0 */
+
+                       if (signr != SIGKILL) {
+                               signr = ptrace_signal(signr, info,
+                                                     regs, cookie);
+                               if (!signr)
+                                       continue;
+                       }
+
+                       ka = &sighand->action[signr-1];
                }
 
-               ka = &sighand->action[signr-1];
+               /* Trace actually delivered signals. */
+               trace_signal_deliver(signr, info, ka);
+
                if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
                        continue;
                if (ka->sa.sa_handler != SIG_DFL) {
@@ -1814,8 +1892,16 @@ relock:
 
                /*
                 * Global init gets no signals it doesn't want.
+                * Container-init gets no signals it doesn't want from same
+                * container.
+                *
+                * Note that if global/container-init sees a sig_kernel_only()
+                * signal here, the signal must have been generated internally
+                * or must have come from an ancestor namespace. In either
+                * case, the signal cannot be dropped.
                 */
-               if (is_global_init(current))
+               if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
+                               !sig_kernel_only(signr))
                        continue;
 
                if (sig_kernel_stop(signr)) {
@@ -1840,7 +1926,7 @@ relock:
                                spin_lock_irq(&sighand->siglock);
                        }
 
-                       if (likely(do_signal_stop(signr))) {
+                       if (likely(do_signal_stop(info->si_signo))) {
                                /* It released the siglock.  */
                                goto relock;
                        }
@@ -1858,9 +1944,10 @@ relock:
                 * Anything else is fatal, maybe with a core dump.
                 */
                current->flags |= PF_SIGNALED;
-               if ((signr != SIGKILL) && print_fatal_signals)
-                       print_fatal_signal(regs, signr);
+
                if (sig_kernel_coredump(signr)) {
+                       if (print_fatal_signals)
+                               print_fatal_signal(regs, info->si_signo);
                        /*
                         * If it was able to dump core, this kills all
                         * other threads in the group and synchronizes with
@@ -1869,13 +1956,13 @@ relock:
                         * first and our do_group_exit call below will use
                         * that value and ignore the one we pass it.
                         */
-                       do_coredump((long)signr, signr, regs);
+                       do_coredump(info->si_signo, info->si_signo, regs);
                }
 
                /*
                 * Death signals, no core dump.
                 */
-               do_group_exit(signr);
+               do_group_exit(info->si_signo);
                /* NOTREACHED */
        }
        spin_unlock_irq(&sighand->siglock);
@@ -1912,14 +1999,14 @@ void exit_signals(struct task_struct *tsk)
        if (unlikely(tsk->signal->group_stop_count) &&
                        !--tsk->signal->group_stop_count) {
                tsk->signal->flags = SIGNAL_STOP_STOPPED;
-               group_stop = 1;
+               group_stop = tracehook_notify_jctl(CLD_STOPPED, CLD_STOPPED);
        }
 out:
        spin_unlock_irq(&tsk->sighand->siglock);
 
        if (unlikely(group_stop)) {
                read_lock(&tasklist_lock);
-               do_notify_parent_cldstop(tsk, CLD_STOPPED);
+               do_notify_parent_cldstop(tsk, group_stop);
                read_unlock(&tasklist_lock);
        }
 }
@@ -1928,8 +2015,6 @@ EXPORT_SYMBOL(recalc_sigpending);
 EXPORT_SYMBOL_GPL(dequeue_signal);
 EXPORT_SYMBOL(flush_signals);
 EXPORT_SYMBOL(force_sig);
-EXPORT_SYMBOL(kill_proc);
-EXPORT_SYMBOL(ptrace_notify);
 EXPORT_SYMBOL(send_sig);
 EXPORT_SYMBOL(send_sig_info);
 EXPORT_SYMBOL(sigprocmask);
@@ -1941,7 +2026,7 @@ EXPORT_SYMBOL(unblock_all_signals);
  * System call entry points.
  */
 
-asmlinkage long sys_restart_syscall(void)
+SYSCALL_DEFINE0(restart_syscall)
 {
        struct restart_block *restart = &current_thread_info()->restart_block;
        return restart->fn(restart);
@@ -1994,8 +2079,8 @@ int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
        return error;
 }
 
-asmlinkage long
-sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
+SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, set,
+               sigset_t __user *, oset, size_t, sigsetsize)
 {
        int error = -EINVAL;
        sigset_t old_set, new_set;
@@ -2054,8 +2139,7 @@ out:
        return error;
 }      
 
-asmlinkage long
-sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
+SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
 {
        return do_sigpending(set, sigsetsize);
 }
@@ -2126,11 +2210,9 @@ int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
 
 #endif
 
-asmlinkage long
-sys_rt_sigtimedwait(const sigset_t __user *uthese,
-                   siginfo_t __user *uinfo,
-                   const struct timespec __user *uts,
-                   size_t sigsetsize)
+SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
+               siginfo_t __user *, uinfo, const struct timespec __user *, uts,
+               size_t, sigsetsize)
 {
        int ret, sig;
        sigset_t these;
@@ -2203,8 +2285,7 @@ sys_rt_sigtimedwait(const sigset_t __user *uthese,
        return ret;
 }
 
-asmlinkage long
-sys_kill(int pid, int sig)
+SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
 {
        struct siginfo info;
 
@@ -2212,44 +2293,54 @@ sys_kill(int pid, int sig)
        info.si_errno = 0;
        info.si_code = SI_USER;
        info.si_pid = task_tgid_vnr(current);
-       info.si_uid = current->uid;
+       info.si_uid = current_uid();
 
        return kill_something_info(sig, &info, pid);
 }
 
-static int do_tkill(int tgid, int pid, int sig)
+static int
+do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
 {
-       int error;
-       struct siginfo info;
        struct task_struct *p;
+       int error = -ESRCH;
 
-       error = -ESRCH;
-       info.si_signo = sig;
-       info.si_errno = 0;
-       info.si_code = SI_TKILL;
-       info.si_pid = task_tgid_vnr(current);
-       info.si_uid = current->uid;
-
-       read_lock(&tasklist_lock);
+       rcu_read_lock();
        p = find_task_by_vpid(pid);
        if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
-               error = check_kill_permission(sig, &info, p);
+               error = check_kill_permission(sig, info, p);
                /*
                 * The null signal is a permissions and process existence
                 * probe.  No signal is actually delivered.
                 */
-               if (!error && sig && p->sighand) {
-                       spin_lock_irq(&p->sighand->siglock);
-                       handle_stop_signal(sig, p);
-                       error = specific_send_sig_info(sig, &info, p);
-                       spin_unlock_irq(&p->sighand->siglock);
+               if (!error && sig) {
+                       error = do_send_sig_info(sig, info, p, false);
+                       /*
+                        * If lock_task_sighand() failed we pretend the task
+                        * dies after receiving the signal. The window is tiny,
+                        * and the signal is private anyway.
+                        */
+                       if (unlikely(error == -ESRCH))
+                               error = 0;
                }
        }
-       read_unlock(&tasklist_lock);
+       rcu_read_unlock();
 
        return error;
 }
 
+static int do_tkill(pid_t tgid, pid_t pid, int sig)
+{
+       struct siginfo info;
+
+       info.si_signo = sig;
+       info.si_errno = 0;
+       info.si_code = SI_TKILL;
+       info.si_pid = task_tgid_vnr(current);
+       info.si_uid = current_uid();
+
+       return do_send_specific(tgid, pid, sig, &info);
+}
+
 /**
  *  sys_tgkill - send signal to one specific thread
  *  @tgid: the thread group ID of the thread
@@ -2260,7 +2351,7 @@ static int do_tkill(int tgid, int pid, int sig)
  *  exists but it's not belonging to the target process anymore. This
  *  method solves the problem of threads exiting and PIDs getting reused.
  */
-asmlinkage long sys_tgkill(int tgid, int pid, int sig)
+SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
 {
        /* This is only valid for single tasks */
        if (pid <= 0 || tgid <= 0)
@@ -2272,8 +2363,7 @@ asmlinkage long sys_tgkill(int tgid, int pid, int sig)
 /*
  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
  */
-asmlinkage long
-sys_tkill(int pid, int sig)
+SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
 {
        /* This is only valid for single tasks */
        if (pid <= 0)
@@ -2282,8 +2372,8 @@ sys_tkill(int pid, int sig)
        return do_tkill(0, pid, sig);
 }
 
-asmlinkage long
-sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
+SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
+               siginfo_t __user *, uinfo)
 {
        siginfo_t info;
 
@@ -2300,6 +2390,32 @@ sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
        return kill_proc_info(sig, &info, pid);
 }
 
+long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
+{
+       /* This is only valid for single tasks */
+       if (pid <= 0 || tgid <= 0)
+               return -EINVAL;
+
+       /* Not even root can pretend to send signals from the kernel.
+          Nor can they impersonate a kill(), which adds source info.  */
+       if (info->si_code >= 0)
+               return -EPERM;
+       info->si_signo = sig;
+
+       return do_send_specific(tgid, pid, sig, info);
+}
+
+SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
+               siginfo_t __user *, uinfo)
+{
+       siginfo_t info;
+
+       if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
+               return -EFAULT;
+
+       return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
+}
+
 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
 {
        struct task_struct *t = current;
@@ -2330,7 +2446,7 @@ int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
                 *   (for example, SIGCHLD), shall cause the pending signal to
                 *   be discarded, whether or not it is blocked"
                 */
-               if (__sig_ignored(t, sig)) {
+               if (sig_handler_ignored(sig_handler(t, sig), sig)) {
                        sigemptyset(&mask);
                        sigaddset(&mask, sig);
                        rm_from_queue_full(&mask, &t->signal->shared_pending);
@@ -2351,11 +2467,9 @@ do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long s
        stack_t oss;
        int error;
 
-       if (uoss) {
-               oss.ss_sp = (void __user *) current->sas_ss_sp;
-               oss.ss_size = current->sas_ss_size;
-               oss.ss_flags = sas_ss_flags(sp);
-       }
+       oss.ss_sp = (void __user *) current->sas_ss_sp;
+       oss.ss_size = current->sas_ss_size;
+       oss.ss_flags = sas_ss_flags(sp);
 
        if (uss) {
                void __user *ss_sp;
@@ -2363,10 +2477,12 @@ do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long s
                int ss_flags;
 
                error = -EFAULT;
-               if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
-                   || __get_user(ss_sp, &uss->ss_sp)
-                   || __get_user(ss_flags, &uss->ss_flags)
-                   || __get_user(ss_size, &uss->ss_size))
+               if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
+                       goto out;
+               error = __get_user(ss_sp, &uss->ss_sp) |
+                       __get_user(ss_flags, &uss->ss_flags) |
+                       __get_user(ss_size, &uss->ss_size);
+               if (error)
                        goto out;
 
                error = -EPERM;
@@ -2398,21 +2514,23 @@ do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long s
                current->sas_ss_size = ss_size;
        }
 
+       error = 0;
        if (uoss) {
                error = -EFAULT;
-               if (copy_to_user(uoss, &oss, sizeof(oss)))
+               if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
                        goto out;
+               error = __put_user(oss.ss_sp, &uoss->ss_sp) |
+                       __put_user(oss.ss_size, &uoss->ss_size) |
+                       __put_user(oss.ss_flags, &uoss->ss_flags);
        }
 
-       error = 0;
 out:
        return error;
 }
 
 #ifdef __ARCH_WANT_SYS_SIGPENDING
 
-asmlinkage long
-sys_sigpending(old_sigset_t __user *set)
+SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
 {
        return do_sigpending(set, sizeof(*set));
 }
@@ -2423,8 +2541,8 @@ sys_sigpending(old_sigset_t __user *set)
 /* Some platforms have their own version with special arguments others
    support only sys_rt_sigprocmask.  */
 
-asmlinkage long
-sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
+SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, set,
+               old_sigset_t __user *, oset)
 {
        int error;
        old_sigset_t old_set, new_set;
@@ -2474,11 +2592,10 @@ out:
 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
 
 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
-asmlinkage long
-sys_rt_sigaction(int sig,
-                const struct sigaction __user *act,
-                struct sigaction __user *oact,
-                size_t sigsetsize)
+SYSCALL_DEFINE4(rt_sigaction, int, sig,
+               const struct sigaction __user *, act,
+               struct sigaction __user *, oact,
+               size_t, sigsetsize)
 {
        struct k_sigaction new_sa, old_sa;
        int ret = -EINVAL;
@@ -2508,15 +2625,13 @@ out:
 /*
  * For backwards compatibility.  Functionality superseded by sigprocmask.
  */
-asmlinkage long
-sys_sgetmask(void)
+SYSCALL_DEFINE0(sgetmask)
 {
        /* SMP safe */
        return current->blocked.sig[0];
 }
 
-asmlinkage long
-sys_ssetmask(int newmask)
+SYSCALL_DEFINE1(ssetmask, int, newmask)
 {
        int old;
 
@@ -2536,8 +2651,7 @@ sys_ssetmask(int newmask)
 /*
  * For backwards compatibility.  Functionality superseded by sigaction.
  */
-asmlinkage unsigned long
-sys_signal(int sig, __sighandler_t handler)
+SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
 {
        struct k_sigaction new_sa, old_sa;
        int ret;
@@ -2554,8 +2668,7 @@ sys_signal(int sig, __sighandler_t handler)
 
 #ifdef __ARCH_WANT_SYS_PAUSE
 
-asmlinkage long
-sys_pause(void)
+SYSCALL_DEFINE0(pause)
 {
        current->state = TASK_INTERRUPTIBLE;
        schedule();
@@ -2565,7 +2678,7 @@ sys_pause(void)
 #endif
 
 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
-asmlinkage long sys_rt_sigsuspend(sigset_t __user *unewset, size_t sigsetsize)
+SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
 {
        sigset_t newset;
 
@@ -2585,7 +2698,7 @@ asmlinkage long sys_rt_sigsuspend(sigset_t __user *unewset, size_t sigsetsize)
 
        current->state = TASK_INTERRUPTIBLE;
        schedule();
-       set_thread_flag(TIF_RESTORE_SIGMASK);
+       set_restore_sigmask();
        return -ERESTARTNOHAND;
 }
 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */