fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / kernel / signal.c
index 83cd433..c0afecc 100644 (file)
@@ -10,7 +10,6 @@
  *             to allow signals to be sent reliably.
  */
 
-#include <linux/config.h>
 #include <linux/slab.h>
 #include <linux/module.h>
 #include <linux/smp_lock.h>
 #include <linux/binfmts.h>
 #include <linux/security.h>
 #include <linux/syscalls.h>
-#include <linux/ptrace.h>
-#include <linux/posix-timers.h>
+#include <linux/tracehook.h>
 #include <linux/signal.h>
-#include <linux/audit.h>
 #include <linux/capability.h>
-#include <linux/vs_cvirt.h>
+#include <linux/freezer.h>
+#include <linux/pid_namespace.h>
+#include <linux/nsproxy.h>
+#include <linux/vs_context.h>
+#include <linux/vs_pid.h>
+
 #include <asm/param.h>
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
 #include <asm/siginfo.h>
+#include "audit.h"     /* audit_signal_info() */
 
 /*
  * SLAB caches for signal bits.
  */
 
-static kmem_cache_t *sigqueue_cachep;
+static struct kmem_cache *sigqueue_cachep;
 
 /*
  * In POSIX a signal is sent either to a specific thread (Linux task)
@@ -148,6 +151,8 @@ static kmem_cache_t *sigqueue_cachep;
 #define sig_kernel_stop(sig) \
                (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_STOP_MASK))
 
+#define sig_needs_tasklist(sig)        ((sig) == SIGCONT)
+
 #define sig_user_defined(t, signr) \
        (((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_DFL) &&  \
         ((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_IGN))
@@ -160,12 +165,6 @@ static int sig_ignored(struct task_struct *t, int sig)
 {
        void __user * handler;
 
-       /*
-        * Tracers always want to know about signals..
-        */
-       if (t->ptrace & PT_PTRACED)
-               return 0;
-
        /*
         * Blocked signals are never ignored, since the
         * signal handler may change by the time it is
@@ -176,8 +175,12 @@ static int sig_ignored(struct task_struct *t, 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));
+       if (handler != SIG_IGN &&
+           (handler != SIG_DFL || !sig_kernel_ignore(sig)))
+               return 0;
+
+       /* It's ignored, we can short-circuit unless a debugger wants it.  */
+       return !tracehook_consider_ignored_signal(t, sig, handler);
 }
 
 /*
@@ -217,7 +220,8 @@ fastcall void recalc_sigpending_tsk(struct task_struct *t)
        if (t->signal->group_stop_count > 0 ||
            (freezing(t)) ||
            PENDING(&t->pending, &t->blocked) ||
-           PENDING(&t->signal->shared_pending, &t->blocked))
+           PENDING(&t->signal->shared_pending, &t->blocked) ||
+           tracehook_induce_sigpending(t))
                set_tsk_thread_flag(t, TIF_SIGPENDING);
        else
                clear_tsk_thread_flag(t, TIF_SIGPENDING);
@@ -268,18 +272,25 @@ static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
                                         int override_rlimit)
 {
        struct sigqueue *q = NULL;
+       struct user_struct *user;
 
-       atomic_inc(&t->user->sigpending);
+       /*
+        * In order to avoid problems with "switch_user()", we want to make
+        * sure that the compiler doesn't re-load "t->user"
+        */
+       user = t->user;
+       barrier();
+       atomic_inc(&user->sigpending);
        if (override_rlimit ||
-           atomic_read(&t->user->sigpending) <=
+           atomic_read(&user->sigpending) <=
                        t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
                q = kmem_cache_alloc(sigqueue_cachep, flags);
        if (unlikely(q == NULL)) {
-               atomic_dec(&t->user->sigpending);
+               atomic_dec(&user->sigpending);
        } else {
                INIT_LIST_HEAD(&q->list);
                q->flags = 0;
-               q->user = get_uid(t->user);
+               q->user = get_uid(user);
        }
        return(q);
 }
@@ -293,7 +304,7 @@ static void __sigqueue_free(struct sigqueue *q)
        kmem_cache_free(sigqueue_cachep, q);
 }
 
-static void flush_sigqueue(struct sigpending *queue)
+void flush_sigqueue(struct sigpending *queue)
 {
        struct sigqueue *q;
 
@@ -308,9 +319,7 @@ static void flush_sigqueue(struct sigpending *queue)
 /*
  * Flush all pending signals for a task.
  */
-
-void
-flush_signals(struct task_struct *t)
+void flush_signals(struct task_struct *t)
 {
        unsigned long flags;
 
@@ -321,109 +330,6 @@ flush_signals(struct task_struct *t)
        spin_unlock_irqrestore(&t->sighand->siglock, flags);
 }
 
-/*
- * This function expects the tasklist_lock write-locked.
- */
-void __exit_sighand(struct task_struct *tsk)
-{
-       struct sighand_struct * sighand = tsk->sighand;
-
-       /* Ok, we're done with the signal handlers */
-       tsk->sighand = NULL;
-       if (atomic_dec_and_test(&sighand->count))
-               sighand_free(sighand);
-}
-
-void exit_sighand(struct task_struct *tsk)
-{
-       write_lock_irq(&tasklist_lock);
-       rcu_read_lock();
-       if (tsk->sighand != NULL) {
-               struct sighand_struct *sighand = rcu_dereference(tsk->sighand);
-               spin_lock(&sighand->siglock);
-               __exit_sighand(tsk);
-               spin_unlock(&sighand->siglock);
-       }
-       rcu_read_unlock();
-       write_unlock_irq(&tasklist_lock);
-}
-
-/*
- * This function expects the tasklist_lock write-locked.
- */
-void __exit_signal(struct task_struct *tsk)
-{
-       struct signal_struct * sig = tsk->signal;
-       struct sighand_struct * sighand;
-
-       if (!sig)
-               BUG();
-       if (!atomic_read(&sig->count))
-               BUG();
-       rcu_read_lock();
-       sighand = rcu_dereference(tsk->sighand);
-       spin_lock(&sighand->siglock);
-       posix_cpu_timers_exit(tsk);
-       if (atomic_dec_and_test(&sig->count)) {
-               posix_cpu_timers_exit_group(tsk);
-               tsk->signal = NULL;
-               __exit_sighand(tsk);
-               spin_unlock(&sighand->siglock);
-               flush_sigqueue(&sig->shared_pending);
-       } else {
-               /*
-                * If there is any task waiting for the group exit
-                * then notify it:
-                */
-               if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
-                       wake_up_process(sig->group_exit_task);
-                       sig->group_exit_task = NULL;
-               }
-               if (tsk == sig->curr_target)
-                       sig->curr_target = next_thread(tsk);
-               tsk->signal = NULL;
-               /*
-                * Accumulate here the counters for all threads but the
-                * group leader as they die, so they can be added into
-                * the process-wide totals when those are taken.
-                * The group leader stays around as a zombie as long
-                * as there are other threads.  When it gets reaped,
-                * the exit.c code will add its counts into these totals.
-                * We won't ever get here for the group leader, since it
-                * will have been the last reference on the signal_struct.
-                */
-               sig->utime = cputime_add(sig->utime, tsk->utime);
-               sig->stime = cputime_add(sig->stime, tsk->stime);
-               sig->min_flt += tsk->min_flt;
-               sig->maj_flt += tsk->maj_flt;
-               sig->nvcsw += tsk->nvcsw;
-               sig->nivcsw += tsk->nivcsw;
-               sig->sched_time += tsk->sched_time;
-               __exit_sighand(tsk);
-               spin_unlock(&sighand->siglock);
-               sig = NULL;     /* Marker for below.  */
-       }
-       rcu_read_unlock();
-       clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
-       flush_sigqueue(&tsk->pending);
-       if (sig) {
-               /*
-                * We are cleaning up the signal_struct here.
-                */
-               exit_thread_group_keys(sig);
-               kmem_cache_free(signal_cachep, sig);
-       }
-}
-
-void exit_signal(struct task_struct *tsk)
-{
-       atomic_dec(&tsk->signal->live);
-
-       write_lock_irq(&tasklist_lock);
-       __exit_signal(tsk);
-       write_unlock_irq(&tasklist_lock);
-}
-
 /*
  * Flush all handlers for a task.
  */
@@ -523,9 +429,8 @@ static int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
                        siginfo_t *info)
 {
-       int sig = 0;
+       int sig = next_signal(pending, mask);
 
-       sig = next_signal(pending, mask);
        if (sig) {
                if (current->notifier) {
                        if (sigismember(current->notifier_mask, sig)) {
@@ -538,9 +443,7 @@ static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
 
                if (!collect_signal(sig, pending, info))
                        sig = 0;
-                               
        }
-       recalc_sigpending();
 
        return sig;
 }
@@ -557,6 +460,7 @@ int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
        if (!signr)
                signr = __dequeue_signal(&tsk->signal->shared_pending,
                                         mask, info);
+       recalc_sigpending_tsk(tsk);
        if (signr && unlikely(sig_kernel_stop(signr))) {
                /*
                 * Set a marker that we have dequeued a stop signal.  Our
@@ -677,37 +581,41 @@ static int rm_from_queue(unsigned long mask, struct sigpending *s)
 static int check_kill_permission(int sig, struct siginfo *info,
                                 struct task_struct *t)
 {
-       int user;
        int error = -EINVAL;
 
        if (!valid_signal(sig))
                return error;
 
-       user = ((info == SEND_SIG_NOINFO) ||
-               (!is_si_special(info) && SI_FROMUSER(info)));
+       if ((info != SEND_SIG_NOINFO) &&
+               (is_si_special(info) || !SI_FROMUSER(info)))
+               goto skip;
+
+       vxdprintk(VXD_CBIT(misc, 7),
+               "check_kill_permission(%d,%p,%p[#%u,%u])",
+               sig, info, t, vx_task_xid(t), t->pid);
 
        error = -EPERM;
-       if (user && ((sig != SIGCONT) ||
-               (current->signal->session != t->signal->session))
+       if (((sig != SIGCONT) ||
+               (process_session(current) != process_session(t)))
            && (current->euid ^ t->suid) && (current->euid ^ t->uid)
            && (current->uid ^ t->suid) && (current->uid ^ t->uid)
            && !capable(CAP_KILL))
                return error;
 
        error = -ESRCH;
-       if (user && !vx_check(vx_task_xid(t), VX_ADMIN|VX_IDENT))
+       if (!vx_check(vx_task_xid(t), VS_WATCH_P|VS_IDENT)) {
+               vxdprintk(current->xid || VXD_CBIT(misc, 7),
+                       "signal %d[%p] xid mismatch %p[#%u,%u] xid=#%u",
+                       sig, info, t, vx_task_xid(t), t->pid, current->xid);
                return error;
-
-       error = security_task_kill(t, info, sig);
+       }
+skip:
+       error = security_task_kill(t, info, sig, 0);
        if (!error)
                audit_signal_info(sig, t); /* Let audit system see the signal */
        return error;
 }
 
-/* forward decl */
-static void do_notify_parent_cldstop(struct task_struct *tsk,
-                                    int to_self,
-                                    int why);
 
 /*
  * Handle magic process-wide effects of stop/continue signals.
@@ -757,7 +665,7 @@ static void handle_stop_signal(int sig, struct task_struct *p)
                        p->signal->group_stop_count = 0;
                        p->signal->flags = SIGNAL_STOP_CONTINUED;
                        spin_unlock(&p->sighand->siglock);
-                       do_notify_parent_cldstop(p, (p->ptrace & PT_PTRACED), CLD_STOPPED);
+                       do_notify_parent_cldstop(p, CLD_STOPPED);
                        spin_lock(&p->sighand->siglock);
                }
                rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
@@ -798,7 +706,7 @@ static void handle_stop_signal(int sig, struct task_struct *p)
                        p->signal->flags = SIGNAL_STOP_CONTINUED;
                        p->signal->group_exit_code = 0;
                        spin_unlock(&p->sighand->siglock);
-                       do_notify_parent_cldstop(p, (p->ptrace & PT_PTRACED), CLD_CONTINUED);
+                       do_notify_parent_cldstop(p, CLD_CONTINUED);
                        spin_lock(&p->sighand->siglock);
                } else {
                        /*
@@ -879,14 +787,44 @@ out_set:
 #define LEGACY_QUEUE(sigptr, sig) \
        (((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
 
+int print_fatal_signals = 0;
+
+static void print_fatal_signal(struct pt_regs *regs, int signr)
+{
+       printk("%s/%d: potentially unexpected fatal signal %d.\n",
+               current->comm, current->pid, signr);
+
+#ifdef __i386__
+       printk("code at %08lx: ", regs->eip);
+       {
+               int i;
+               for (i = 0; i < 16; i++) {
+                       unsigned char insn;
+
+                       __get_user(insn, (unsigned char *)(regs->eip + i));
+                       printk("%02x ", insn);
+               }
+       }
+#endif
+       printk("\n");
+       show_regs(regs);
+}
+
+static int __init setup_print_fatal_signals(char *str)
+{
+       get_option (&str, &print_fatal_signals);
+
+       return 1;
+}
+
+__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 = 0;
 
-       if (!irqs_disabled())
-               BUG();
+       BUG_ON(!irqs_disabled());
        assert_spin_locked(&t->sighand->siglock);
 
        /* Short-circuit ignored signals.  */
@@ -909,22 +847,31 @@ out:
 /*
  * 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;
+       int ret, blocked, ignored;
+       struct k_sigaction *action;
 
        spin_lock_irqsave(&t->sighand->siglock, flags);
-       if (t->sighand->action[sig-1].sa.sa_handler == SIG_IGN) {
-               t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
-       }
-       if (sigismember(&t->blocked, sig)) {
-               sigdelset(&t->blocked, sig);
+       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_tsk(t);
+               }
        }
-       recalc_sigpending_tsk(t);
        ret = specific_send_sig_info(sig, info, t);
        spin_unlock_irqrestore(&t->sighand->siglock, flags);
 
@@ -1005,7 +952,7 @@ __group_complete_signal(int sig, struct task_struct *p)
         */
        if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
            !sigismember(&t->real_blocked, sig) &&
-           (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
+           (sig == SIGKILL || !tracehook_consider_fatal_signal(t, sig))) {
                /*
                 * This signal will be fatal to the whole group.
                 */
@@ -1129,55 +1076,81 @@ void zap_other_threads(struct task_struct *p)
 /*
  * Must be called under rcu_read_lock() or with tasklist_lock read-held.
  */
+struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
+{
+       struct sighand_struct *sighand;
+
+       for (;;) {
+               sighand = rcu_dereference(tsk->sighand);
+               if (unlikely(sighand == NULL))
+                       break;
+
+               spin_lock_irqsave(&sighand->siglock, *flags);
+               if (likely(sighand == tsk->sighand))
+                       break;
+               spin_unlock_irqrestore(&sighand->siglock, *flags);
+       }
+
+       return sighand;
+}
+
 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
 {
        unsigned long flags;
-       struct sighand_struct *sp;
        int ret;
 
-retry:
        ret = check_kill_permission(sig, info, p);
-       if (!ret && sig && (sp = rcu_dereference(p->sighand))) {
-               spin_lock_irqsave(&sp->siglock, flags);
-               if (p->sighand != sp) {
-                       spin_unlock_irqrestore(&sp->siglock, flags);
-                       goto retry;
-               }
-               if ((atomic_read(&sp->count) == 0) ||
-                               (atomic_read(&p->usage) == 0)) {
-                       spin_unlock_irqrestore(&sp->siglock, flags);
-                       return -ESRCH;
+
+       if (!ret && sig) {
+               ret = -ESRCH;
+               if (lock_task_sighand(p, &flags)) {
+                       ret = __group_send_sig_info(sig, info, p);
+                       unlock_task_sighand(p, &flags);
                }
-               ret = __group_send_sig_info(sig, info, p);
-               spin_unlock_irqrestore(&sp->siglock, flags);
        }
 
        return ret;
 }
 
 /*
- * kill_pg_info() sends a signal to a process group: this is what the tty
+ * kill_pgrp_info() sends a signal to a process group: this is what the tty
  * control characters do (^C, ^Z etc)
  */
 
-int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
+int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
 {
        struct task_struct *p = NULL;
        int retval, success;
 
-       if (pgrp <= 0)
-               return -EINVAL;
-
        success = 0;
        retval = -ESRCH;
-       do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
+       do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
                int err = group_send_sig_info(sig, info, p);
                success |= !err;
                retval = err;
-       } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
+       } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
        return success ? 0 : retval;
 }
 
+int kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
+{
+       int retval;
+
+       read_lock(&tasklist_lock);
+       retval = __kill_pgrp_info(sig, info, pgrp);
+       read_unlock(&tasklist_lock);
+
+       return retval;
+}
+
+int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
+{
+       if (pgrp <= 0)
+               return -EINVAL;
+
+       return __kill_pgrp_info(sig, info, find_pid(pgrp));
+}
+
 int
 kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
 {
@@ -1190,21 +1163,20 @@ kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
        return retval;
 }
 
-int
-kill_proc_info(int sig, struct siginfo *info, pid_t pid)
+int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
 {
        int error;
        int acquired_tasklist_lock = 0;
        struct task_struct *p;
 
        rcu_read_lock();
-       if (unlikely(sig_kernel_stop(sig) || sig == SIGCONT)) {
+       if (unlikely(sig_needs_tasklist(sig))) {
                read_lock(&tasklist_lock);
                acquired_tasklist_lock = 1;
        }
-       p = find_task_by_pid(pid);
+       p = pid_task(pid, PIDTYPE_PID);
        error = -ESRCH;
-       if (p && vx_check(vx_task_xid(p), VX_IDENT))
+       if (p && vx_check(vx_task_xid(p), VS_IDENT))
                error = group_send_sig_info(sig, info, p);
        if (unlikely(acquired_tasklist_lock))
                read_unlock(&tasklist_lock);
@@ -1212,9 +1184,18 @@ kill_proc_info(int sig, struct siginfo *info, pid_t pid)
        return error;
 }
 
-/* like kill_proc_info(), but doesn't use uid/euid of "current" */
-int kill_proc_info_as_uid(int sig, struct siginfo *info, pid_t pid,
-                     uid_t uid, uid_t euid)
+static int kill_proc_info(int sig, struct siginfo *info, pid_t pid)
+{
+       int error;
+       rcu_read_lock();
+       error = kill_pid_info(sig, info, find_pid(vx_rmap_pid(pid)));
+       rcu_read_unlock();
+       return error;
+}
+
+/* like kill_pid_info(), but doesn't use uid/euid of "current" */
+int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
+                     uid_t uid, uid_t euid, u32 secid)
 {
        int ret = -EINVAL;
        struct task_struct *p;
@@ -1223,7 +1204,7 @@ int kill_proc_info_as_uid(int sig, struct siginfo *info, pid_t pid,
                return ret;
 
        read_lock(&tasklist_lock);
-       p = find_task_by_pid(pid);
+       p = pid_task(pid, PIDTYPE_PID);
        if (!p) {
                ret = -ESRCH;
                goto out_unlock;
@@ -1234,6 +1215,9 @@ int kill_proc_info_as_uid(int sig, struct siginfo *info, pid_t pid,
                ret = -EPERM;
                goto out_unlock;
        }
+       ret = security_task_kill(p, info, sig, secid);
+       if (ret)
+               goto out_unlock;
        if (sig && p->sighand) {
                unsigned long flags;
                spin_lock_irqsave(&p->sighand->siglock, flags);
@@ -1244,7 +1228,7 @@ out_unlock:
        read_unlock(&tasklist_lock);
        return ret;
 }
-EXPORT_SYMBOL_GPL(kill_proc_info_as_uid);
+EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
 
 /*
  * kill_something_info() interprets pid in interesting ways just like kill(2).
@@ -1263,7 +1247,7 @@ static int kill_something_info(int sig, struct siginfo *info, int pid)
 
                read_lock(&tasklist_lock);
                for_each_process(p) {
-                       if (vx_check(vx_task_xid(p), VX_ADMIN|VX_IDENT)&&
+                       if (vx_check(vx_task_xid(p), VS_ADMIN_P|VS_IDENT) &&
                                p->pid > 1 && p->tgid != current->tgid) {
                                int err = group_send_sig_info(sig, info, p);
                                ++count;
@@ -1363,6 +1347,18 @@ force_sigsegv(int sig, struct task_struct *p)
        return 0;
 }
 
+int kill_pgrp(struct pid *pid, int sig, int priv)
+{
+       return kill_pgrp_info(sig, __si_special(priv), pid);
+}
+EXPORT_SYMBOL(kill_pgrp);
+
+int kill_pid(struct pid *pid, int sig, int priv)
+{
+       return kill_pid_info(sig, __si_special(priv), pid);
+}
+EXPORT_SYMBOL(kill_pid);
+
 int
 kill_pg(pid_t pgrp, int sig, int priv)
 {
@@ -1415,12 +1411,10 @@ void sigqueue_free(struct sigqueue *q)
        __sigqueue_free(q);
 }
 
-int
-send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
+int send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
 {
        unsigned long flags;
        int ret = 0;
-       struct sighand_struct *sh;
 
        BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
 
@@ -1434,48 +1428,17 @@ send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
         */
        rcu_read_lock();
 
-       if (unlikely(p->flags & PF_EXITING)) {
+       if (!likely(lock_task_sighand(p, &flags))) {
                ret = -1;
                goto out_err;
        }
 
-retry:
-       sh = rcu_dereference(p->sighand);
-
-       spin_lock_irqsave(&sh->siglock, flags);
-       if (p->sighand != sh) {
-               /* We raced with exec() in a multithreaded process... */
-               spin_unlock_irqrestore(&sh->siglock, flags);
-               goto retry;
-       }
-
-       /*
-        * We do the check here again to handle the following scenario:
-        *
-        * CPU 0                CPU 1
-        * send_sigqueue
-        * check PF_EXITING
-        * interrupt            exit code running
-        *                      __exit_signal
-        *                      lock sighand->siglock
-        *                      unlock sighand->siglock
-        * lock sh->siglock
-        * add(tsk->pending)    flush_sigqueue(tsk->pending)
-        *
-        */
-
-       if (unlikely(p->flags & PF_EXITING)) {
-               ret = -1;
-               goto out;
-       }
-
        if (unlikely(!list_empty(&q->list))) {
                /*
                 * If an SI_TIMER entry is already queue just increment
                 * the overrun count.
                 */
-               if (q->info.si_code != SI_TIMER)
-                       BUG();
+               BUG_ON(q->info.si_code != SI_TIMER);
                q->info.si_overrun++;
                goto out;
        }
@@ -1491,7 +1454,7 @@ retry:
                signal_wake_up(p, sig == SIGKILL);
 
 out:
-       spin_unlock_irqrestore(&sh->siglock, flags);
+       unlock_task_sighand(p, &flags);
 out_err:
        rcu_read_unlock();
 
@@ -1523,8 +1486,7 @@ send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
                 * the overrun count.  Other uses should not try to
                 * send the signal multiple times.
                 */
-               if (q->info.si_code != SI_TIMER)
-                       BUG();
+               BUG_ON(q->info.si_code != SI_TIMER);
                q->info.si_overrun++;
                goto out;
        } 
@@ -1569,8 +1531,7 @@ void do_notify_parent(struct task_struct *tsk, int sig)
        /* do_notify_parent_cldstop should have been called instead.  */
        BUG_ON(tsk->state & (TASK_STOPPED|TASK_TRACED));
 
-       BUG_ON(!tsk->ptrace &&
-              (tsk->group_leader != tsk || !thread_group_empty(tsk)));
+       BUG_ON(tsk->group_leader != tsk || !thread_group_empty(tsk));
 
        info.si_signo = sig;
        info.si_errno = 0;
@@ -1595,7 +1556,7 @@ 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 (sig == SIGCHLD &&
            (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
             (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
                /*
@@ -1623,20 +1584,13 @@ void do_notify_parent(struct task_struct *tsk, int sig)
        spin_unlock_irqrestore(&psig->siglock, flags);
 }
 
-static void do_notify_parent_cldstop(struct task_struct *tsk, int to_self, int why)
+void do_notify_parent_cldstop(struct task_struct *tsk, int why)
 {
        struct siginfo info;
        unsigned long flags;
        struct task_struct *parent;
        struct sighand_struct *sighand;
 
-       if (to_self)
-               parent = tsk->parent;
-       else {
-               tsk = tsk->group_leader;
-               parent = tsk->real_parent;
-       }
-
        info.si_signo = SIGCHLD;
        info.si_errno = 0;
        info.si_pid = tsk->pid;
@@ -1661,6 +1615,15 @@ static void do_notify_parent_cldstop(struct task_struct *tsk, int to_self, int w
                BUG();
        }
 
+       /*
+        * Tracing can decide that we should not do the normal notification.
+        */
+       if (tracehook_notify_cldstop(tsk, &info))
+               return;
+
+       tsk = tsk->group_leader;
+       parent = tsk->parent;
+
        sighand = parent->sighand;
        spin_lock_irqsave(&sighand->siglock, flags);
        if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
@@ -1673,108 +1636,23 @@ static void do_notify_parent_cldstop(struct task_struct *tsk, int to_self, int w
        spin_unlock_irqrestore(&sighand->siglock, flags);
 }
 
-/*
- * This must be called with current->sighand->siglock held.
- *
- * This should be the path for all ptrace stops.
- * We always set current->last_siginfo while stopped here.
- * That makes it a way to test a stopped process for
- * being ptrace-stopped vs being job-control-stopped.
- *
- * If we actually decide not to stop at all because the tracer is gone,
- * we leave nostop_code in current->exit_code.
- */
-static void ptrace_stop(int exit_code, int nostop_code, siginfo_t *info)
-{
-       /*
-        * If there is a group stop in progress,
-        * we must participate in the bookkeeping.
-        */
-       if (current->signal->group_stop_count > 0)
-               --current->signal->group_stop_count;
-
-       current->last_siginfo = info;
-       current->exit_code = exit_code;
-
-       /* Let the debugger run.  */
-       set_current_state(TASK_TRACED);
-       spin_unlock_irq(&current->sighand->siglock);
-       try_to_freeze();
-       read_lock(&tasklist_lock);
-       if (likely(current->ptrace & PT_PTRACED) &&
-           likely(current->parent != current->real_parent ||
-                  !(current->ptrace & PT_ATTACHED)) &&
-           (likely(current->parent->signal != current->signal) ||
-            !unlikely(current->signal->flags & SIGNAL_GROUP_EXIT))) {
-               do_notify_parent_cldstop(current, 1, CLD_TRAPPED);
-               read_unlock(&tasklist_lock);
-               schedule();
-       } else {
-               /*
-                * By the time we got the lock, our tracer went away.
-                * Don't stop here.
-                */
-               read_unlock(&tasklist_lock);
-               set_current_state(TASK_RUNNING);
-               current->exit_code = nostop_code;
-       }
-
-       /*
-        * We are back.  Now reacquire the siglock before touching
-        * last_siginfo, so that we are sure to have synchronized with
-        * any signal-sending on another CPU that wants to examine it.
-        */
-       spin_lock_irq(&current->sighand->siglock);
-       current->last_siginfo = NULL;
-
-       /*
-        * Queued signals ignored us while we were stopped for tracing.
-        * So check for any that we should take before resuming user mode.
-        */
-       recalc_sigpending();
-}
-
-void ptrace_notify(int exit_code)
-{
-       siginfo_t info;
-
-       BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
-
-       memset(&info, 0, sizeof info);
-       info.si_signo = SIGTRAP;
-       info.si_code = exit_code;
-       info.si_pid = current->pid;
-       info.si_uid = current->uid;
-
-       /* Let the debugger run.  */
-       spin_lock_irq(&current->sighand->siglock);
-       ptrace_stop(exit_code, 0, &info);
-       spin_unlock_irq(&current->sighand->siglock);
-}
-
 static void
 finish_stop(int stop_count)
 {
-       int to_self;
-
        /*
         * 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))
-               to_self = 1;
-       else if (stop_count == 0)
-               to_self = 0;
-       else
-               goto out;
-
-       read_lock(&tasklist_lock);
-       do_notify_parent_cldstop(current, to_self, CLD_STOPPED);
-       read_unlock(&tasklist_lock);
+       if (!tracehook_finish_stop(stop_count <= 0) && stop_count <= 0) {
+               read_lock(&tasklist_lock);
+               do_notify_parent_cldstop(current, CLD_STOPPED);
+               read_unlock(&tasklist_lock);
+       }
 
-out:
-       schedule();
+       do {
+               schedule();
+       } while (try_to_freeze());
        /*
         * Now we don't run again until continued.
         */
@@ -1787,12 +1665,10 @@ out:
  * Returns nonzero if we've actually stopped and released the siglock.
  * Returns zero if we didn't stop and still hold the siglock.
  */
-static int
-do_signal_stop(int signr)
+static int do_signal_stop(int signr)
 {
        struct signal_struct *sig = current->signal;
-       struct sighand_struct *sighand = current->sighand;
-       int stop_count = -1;
+       int stop_count;
 
        if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
                return 0;
@@ -1802,86 +1678,37 @@ do_signal_stop(int signr)
                 * There is a group stop in progress.  We don't need to
                 * start another one.
                 */
-               signr = sig->group_exit_code;
                stop_count = --sig->group_stop_count;
-               current->exit_code = signr;
-               set_current_state(TASK_STOPPED);
-               if (stop_count == 0)
-                       sig->flags = SIGNAL_STOP_STOPPED;
-               spin_unlock_irq(&sighand->siglock);
-       }
-       else if (thread_group_empty(current)) {
-               /*
-                * Lock must be held through transition to stopped state.
-                */
-               current->exit_code = current->signal->group_exit_code = signr;
-               set_current_state(TASK_STOPPED);
-               sig->flags = SIGNAL_STOP_STOPPED;
-               spin_unlock_irq(&sighand->siglock);
-       }
-       else {
+       } else {
                /*
                 * There is no group stop already in progress.
-                * We must initiate one now, but that requires
-                * dropping siglock to get both the tasklist lock
-                * and siglock again in the proper order.  Note that
-                * this allows an intervening SIGCONT to be posted.
-                * We need to check for that and bail out if necessary.
+                * We must initiate one now.
                 */
                struct task_struct *t;
 
-               spin_unlock_irq(&sighand->siglock);
-
-               /* signals can be posted during this window */
-
-               read_lock(&tasklist_lock);
-               spin_lock_irq(&sighand->siglock);
+               sig->group_exit_code = signr;
 
-               if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED)) {
+               stop_count = 0;
+               for (t = next_thread(current); t != current; t = next_thread(t))
                        /*
-                        * Another stop or continue happened while we
-                        * didn't have the lock.  We can just swallow this
-                        * signal now.  If we raced with a SIGCONT, that
-                        * should have just cleared it now.  If we raced
-                        * with another processor delivering a stop signal,
-                        * then the SIGCONT that wakes us up should clear it.
+                        * Setting state to TASK_STOPPED for a group
+                        * stop is always done with the siglock held,
+                        * so this check has no races.
                         */
-                       read_unlock(&tasklist_lock);
-                       return 0;
-               }
-
-               if (sig->group_stop_count == 0) {
-                       sig->group_exit_code = signr;
-                       stop_count = 0;
-                       for (t = next_thread(current); t != current;
-                            t = next_thread(t))
-                               /*
-                                * Setting state to TASK_STOPPED for a group
-                                * stop is always done with the siglock held,
-                                * so this check has no races.
-                                */
-                               if (!t->exit_state &&
-                                   !(t->state & (TASK_STOPPED|TASK_TRACED))) {
-                                       stop_count++;
-                                       signal_wake_up(t, 0);
-                               }
-                       sig->group_stop_count = stop_count;
-               }
-               else {
-                       /* A race with another thread while unlocked.  */
-                       signr = sig->group_exit_code;
-                       stop_count = --sig->group_stop_count;
-               }
-
-               current->exit_code = signr;
-               set_current_state(TASK_STOPPED);
-               if (stop_count == 0)
-                       sig->flags = SIGNAL_STOP_STOPPED;
-
-               spin_unlock_irq(&sighand->siglock);
-               read_unlock(&tasklist_lock);
+                       if (!t->exit_state &&
+                           !(t->state & (TASK_STOPPED|TASK_TRACED))) {
+                               stop_count++;
+                               signal_wake_up(t, 0);
+                       }
+               sig->group_stop_count = stop_count;
        }
 
+       if (stop_count == 0)
+               sig->flags = SIGNAL_STOP_STOPPED;
+       current->exit_code = sig->group_exit_code;
+       __set_current_state(TASK_STOPPED);
+
+       spin_unlock_irq(&current->sighand->siglock);
        finish_stop(stop_count);
        return 1;
 }
@@ -1933,6 +1760,8 @@ int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
        sigset_t *mask = &current->blocked;
        int signr = 0;
 
+       try_to_freeze();
+
 relock:
        spin_lock_irq(&current->sighand->siglock);
        for (;;) {
@@ -1942,44 +1771,24 @@ relock:
                    handle_group_stop())
                        goto relock;
 
-               signr = dequeue_signal(current, mask, info);
-
-               if (!signr)
-                       break; /* will return 0 */
-
-               if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
-                       ptrace_signal_deliver(regs, cookie);
-
-                       /* Let the debugger run.  */
-                       ptrace_stop(signr, signr, info);
-
-                       /* We're back.  Did the debugger cancel the sig?  */
-                       signr = current->exit_code;
-                       if (signr == 0)
-                               continue;
-
-                       current->exit_code = 0;
-
-                       /* Update the siginfo structure if the signal has
-                          changed.  If the debugger wanted something
-                          specific in the siginfo structure then it should
-                          have updated *info via PTRACE_SETSIGINFO.  */
-                       if (signr != info->si_signo) {
-                               info->si_signo = signr;
-                               info->si_errno = 0;
-                               info->si_code = SI_USER;
-                               info->si_pid = current->parent->pid;
-                               info->si_uid = current->parent->uid;
-                       }
+               /*
+                * 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, mask, info);
 
-                       /* If the (new) signal is now blocked, requeue it.  */
-                       if (sigismember(&current->blocked, signr)) {
-                               specific_send_sig_info(signr, info, current);
-                               continue;
-                       }
+                       if (!signr)
+                               break; /* will return 0 */
+                       ka = &current->sighand->action[signr-1];
                }
 
-               ka = &current->sighand->action[signr-1];
                if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
                        continue;
                if (ka->sa.sa_handler != SIG_DFL) {
@@ -1998,8 +1807,12 @@ relock:
                if (sig_kernel_ignore(signr)) /* Default is nothing. */
                        continue;
 
-               /* Init gets no signals it doesn't want.  */
-               if (current->pid == 1)
+               /*
+                * Init of a pid space gets no signals it doesn't want from
+                * within that pid space. It can of course get signals from
+                * its parent pid space.
+                */
+               if (current == child_reaper(current))
                        continue;
 
                /* virtual init is protected against user signals */
@@ -2029,7 +1842,7 @@ relock:
                                spin_lock_irq(&current->sighand->siglock);
                        }
 
-                       if (likely(do_signal_stop(signr))) {
+                       if (likely(do_signal_stop(info->si_signo))) {
                                /* It released the siglock.  */
                                goto relock;
                        }
@@ -2047,6 +1860,8 @@ relock:
                 * Anything else is fatal, maybe with a core dump.
                 */
                current->flags |= PF_SIGNALED;
+               if (print_fatal_signals)
+                       print_fatal_signal(regs, signr);
                if (sig_kernel_coredump(signr)) {
                        /*
                         * If it was able to dump core, this kills all
@@ -2056,13 +1871,13 @@ 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(&current->sighand->siglock);
@@ -2075,7 +1890,6 @@ EXPORT_SYMBOL(flush_signals);
 EXPORT_SYMBOL(force_sig);
 EXPORT_SYMBOL(kill_pg);
 EXPORT_SYMBOL(kill_proc);
-EXPORT_SYMBOL(ptrace_notify);
 EXPORT_SYMBOL(send_sig);
 EXPORT_SYMBOL(send_sig_info);
 EXPORT_SYMBOL(sigprocmask);
@@ -2115,10 +1929,11 @@ long do_no_restart_syscall(struct restart_block *param)
 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
 {
        int error;
-       sigset_t old_block;
 
        spin_lock_irq(&current->sighand->siglock);
-       old_block = current->blocked;
+       if (oldset)
+               *oldset = current->blocked;
+
        error = 0;
        switch (how) {
        case SIG_BLOCK:
@@ -2135,8 +1950,7 @@ int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
        }
        recalc_sigpending();
        spin_unlock_irq(&current->sighand->siglock);
-       if (oldset)
-               *oldset = old_block;
+
        return error;
 }
 
@@ -2323,7 +2137,6 @@ sys_rt_sigtimedwait(const sigset_t __user *uthese,
 
                        timeout = schedule_timeout_interruptible(timeout);
 
-                       try_to_freeze();
                        spin_lock_irq(&current->sighand->siglock);
                        sig = dequeue_signal(current, &these, &info);
                        current->blocked = current->real_blocked;
@@ -2445,8 +2258,7 @@ sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
        return kill_proc_info(sig, &info, pid);
 }
 
-int
-do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
+int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
 {
        struct k_sigaction *k;
        sigset_t mask;
@@ -2472,6 +2284,7 @@ do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
        if (act) {
                sigdelsetmask(&act->sa.sa_mask,
                              sigmask(SIGKILL) | sigmask(SIGSTOP));
+               *k = *act;
                /*
                 * POSIX 3.3.1.3:
                 *  "Setting a signal action to SIG_IGN for a signal that is
@@ -2484,19 +2297,8 @@ do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
                 *   be discarded, whether or not it is blocked"
                 */
                if (act->sa.sa_handler == SIG_IGN ||
-                   (act->sa.sa_handler == SIG_DFL &&
-                    sig_kernel_ignore(sig))) {
-                       /*
-                        * This is a fairly rare case, so we only take the
-                        * tasklist_lock once we're sure we'll need it.
-                        * Now we must do this little unlock and relock
-                        * dance to maintain the lock hierarchy.
-                        */
+                  (act->sa.sa_handler == SIG_DFL && sig_kernel_ignore(sig))) {
                        struct task_struct *t = current;
-                       spin_unlock_irq(&t->sighand->siglock);
-                       read_lock(&tasklist_lock);
-                       spin_lock_irq(&t->sighand->siglock);
-                       *k = *act;
                        sigemptyset(&mask);
                        sigaddset(&mask, sig);
                        rm_from_queue_full(&mask, &t->signal->shared_pending);
@@ -2505,12 +2307,7 @@ do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
                                recalc_sigpending_tsk(t);
                                t = next_thread(t);
                        } while (t != current);
-                       spin_unlock_irq(&current->sighand->siglock);
-                       read_unlock(&tasklist_lock);
-                       return 0;
                }
-
-               *k = *act;
        }
 
        spin_unlock_irq(&current->sighand->siglock);
@@ -2762,6 +2559,11 @@ asmlinkage long sys_rt_sigsuspend(sigset_t __user *unewset, size_t sigsetsize)
 }
 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
 
+__attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
+{
+       return NULL;
+}
+
 void __init signals_init(void)
 {
        sigqueue_cachep =