header cleaning: don't include smp_lock.h when not used
[linux-3.10.git] / kernel / signal.c
index 92025b1..1368e67 100644 (file)
  *             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/init.h>
 #include <linux/sched.h>
 #include <linux/fs.h>
 #include <linux/syscalls.h>
 #include <linux/ptrace.h>
 #include <linux/signal.h>
-#include <linux/audit.h>
 #include <linux/capability.h>
+#include <linux/freezer.h>
+#include <linux/pid_namespace.h>
+#include <linux/nsproxy.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)
@@ -268,18 +270,25 @@ 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);
 }
@@ -418,9 +427,8 @@ 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)) {
@@ -433,9 +441,7 @@ static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
 
                if (!collect_signal(sig, pending, info))
                        sig = 0;
-                               
        }
-       recalc_sigpending();
 
        return sig;
 }
@@ -449,25 +455,50 @@ static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
 {
        int signr = __dequeue_signal(&tsk->pending, mask, info);
-       if (!signr)
+       if (!signr) {
                signr = __dequeue_signal(&tsk->signal->shared_pending,
                                         mask, info);
-       if (signr && unlikely(sig_kernel_stop(signr))) {
-               /*
-                * Set a marker that we have dequeued a stop signal.  Our
-                * caller might release the siglock and then the pending
-                * stop signal it is about to process is no longer in the
-                * pending bitmasks, but must still be cleared by a SIGCONT
-                * (and overruled by a SIGKILL).  So those cases clear this
-                * shared flag after we've set it.  Note that this flag may
-                * remain set after the signal we return is ignored or
-                * handled.  That doesn't matter because its only purpose
-                * 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;
-       }
+               /*
+                * itimer signal ?
+                *
+                * itimers are process shared and we restart periodic
+                * itimers in the signal delivery path to prevent DoS
+                * attacks in the high resolution timer case. This is
+                * compliant with the old way of self restarting
+                * itimers, as the SIGALRM is a legacy signal and only
+                * queued once. Changing the restart behaviour to
+                * restart the timer in the signal dequeue path is
+                * reducing the timer noise on heavy loaded !highres
+                * systems too.
+                */
+               if (unlikely(signr == SIGALRM)) {
+                       struct hrtimer *tmr = &tsk->signal->real_timer;
+
+                       if (!hrtimer_is_queued(tmr) &&
+                           tsk->signal->it_real_incr.tv64 != 0) {
+                               hrtimer_forward(tmr, tmr->base->get_time(),
+                                               tsk->signal->it_real_incr);
+                               hrtimer_restart(tmr);
+                       }
+               }
+       }
+       recalc_sigpending_tsk(tsk);
+       if (signr && unlikely(sig_kernel_stop(signr))) {
+               /*
+                * Set a marker that we have dequeued a stop signal.  Our
+                * caller might release the siglock and then the pending
+                * stop signal it is about to process is no longer in the
+                * pending bitmasks, but must still be cleared by a SIGCONT
+                * (and overruled by a SIGKILL).  So those cases clear this
+                * shared flag after we've set it.  Note that this flag may
+                * remain set after the signal we return is ignored or
+                * handled.  That doesn't matter because its only purpose
+                * 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;
+       }
        if ( signr &&
             ((info->si_code & __SI_MASK) == __SI_TIMER) &&
             info->si_sys_private){
@@ -578,13 +609,13 @@ static int check_kill_permission(int sig, struct siginfo *info,
        error = -EPERM;
        if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
            && ((sig != SIGCONT) ||
-               (current->signal->session != t->signal->session))
+               (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 = security_task_kill(t, info, sig);
+       error = security_task_kill(t, info, sig, 0);
        if (!error)
                audit_signal_info(sig, t); /* Let audit system see the signal */
        return error;
@@ -769,8 +800,7 @@ 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.  */
@@ -793,22 +823,31 @@ 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);
 
@@ -869,7 +908,6 @@ __group_complete_signal(int sig, struct task_struct *p)
                if (t == NULL)
                        /* restart balancing at this thread */
                        t = p->signal->curr_target = p;
-               BUG_ON(t->tgid != p->tgid);
 
                while (!wants_signal(sig, t)) {
                        t = next_thread(t);
@@ -1051,65 +1089,69 @@ int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
 }
 
 /*
- * 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_pg_info(int sig, struct siginfo *info, pid_t pgrp)
+int kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
 {
        int retval;
 
        read_lock(&tasklist_lock);
-       retval = __kill_pg_info(sig, info, pgrp);
+       retval = __kill_pgrp_info(sig, info, pgrp);
        read_unlock(&tasklist_lock);
 
        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_needs_tasklist(sig))) {
+       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)
                error = group_send_sig_info(sig, info, p);
-       if (unlikely(acquired_tasklist_lock))
+
+       if (unlikely(sig_needs_tasklist(sig)))
                read_unlock(&tasklist_lock);
        rcu_read_unlock();
        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)
+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(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;
@@ -1118,7 +1160,7 @@ 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;
@@ -1129,6 +1171,9 @@ 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);
@@ -1139,7 +1184,7 @@ 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).
@@ -1150,8 +1195,10 @@ EXPORT_SYMBOL_GPL(kill_proc_info_as_uid);
 
 static int kill_something_info(int sig, struct siginfo *info, int pid)
 {
+       int ret;
+       rcu_read_lock();
        if (!pid) {
-               return kill_pg_info(sig, info, process_group(current));
+               ret = kill_pgrp_info(sig, info, task_pgrp(current));
        } else if (pid == -1) {
                int retval = 0, count = 0;
                struct task_struct * p;
@@ -1166,12 +1213,14 @@ static int kill_something_info(int sig, struct siginfo *info, int pid)
                        }
                }
                read_unlock(&tasklist_lock);
-               return count ? retval : -ESRCH;
+               ret = count ? retval : -ESRCH;
        } else if (pid < 0) {
-               return kill_pg_info(sig, info, -pid);
+               ret = kill_pgrp_info(sig, info, find_pid(-pid));
        } else {
-               return kill_proc_info(sig, info, pid);
+               ret = kill_pid_info(sig, info, find_pid(pid));
        }
+       rcu_read_unlock();
+       return ret;
 }
 
 /*
@@ -1257,11 +1306,17 @@ force_sigsegv(int sig, struct task_struct *p)
        return 0;
 }
 
-int
-kill_pg(pid_t pgrp, int sig, int priv)
+int kill_pgrp(struct pid *pid, int sig, int priv)
 {
-       return kill_pg_info(sig, __si_special(priv), pgrp);
+       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_proc(pid_t pid, int sig, int priv)
@@ -1384,8 +1439,7 @@ 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;
        } 
@@ -1534,6 +1588,35 @@ static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
        spin_unlock_irqrestore(&sighand->siglock, flags);
 }
 
+static inline int may_ptrace_stop(void)
+{
+       if (!likely(current->ptrace & PT_PTRACED))
+               return 0;
+
+       if (unlikely(current->parent == current->real_parent &&
+                   (current->ptrace & PT_ATTACHED)))
+               return 0;
+
+       if (unlikely(current->signal == current->parent->signal) &&
+           unlikely(current->signal->flags & SIGNAL_GROUP_EXIT))
+               return 0;
+
+       /*
+        * Are we in the middle of do_coredump?
+        * If so and our tracer is also part of the coredump stopping
+        * 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
+        * is safe to enter schedule().
+        */
+       if (unlikely(current->mm->core_waiters) &&
+           unlikely(current->mm == current->parent->mm))
+               return 0;
+
+       return 1;
+}
+
 /*
  * This must be called with current->sighand->siglock held.
  *
@@ -1562,11 +1645,7 @@ static void ptrace_stop(int exit_code, int nostop_code, siginfo_t *info)
        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))) {
+       if (may_ptrace_stop()) {
                do_notify_parent_cldstop(current, CLD_TRAPPED);
                read_unlock(&tasklist_lock);
                schedule();
@@ -1627,7 +1706,9 @@ finish_stop(int stop_count)
                read_unlock(&tasklist_lock);
        }
 
-       schedule();
+       do {
+               schedule();
+       } while (try_to_freeze());
        /*
         * Now we don't run again until continued.
         */
@@ -1757,9 +1838,9 @@ relock:
                        /* Let the debugger run.  */
                        ptrace_stop(signr, signr, info);
 
-                       /* We're back.  Did the debugger cancel the sig or group_exit? */
+                       /* We're back.  Did the debugger cancel the sig?  */
                        signr = current->exit_code;
-                       if (signr == 0 || current->signal->flags & SIGNAL_GROUP_EXIT)
+                       if (signr == 0)
                                continue;
 
                        current->exit_code = 0;
@@ -1802,8 +1883,12 @@ relock:
                if (sig_kernel_ignore(signr)) /* Default is nothing. */
                        continue;
 
-               /* Init gets no signals it doesn't want.  */
-               if (current == child_reaper)
+               /*
+                * 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;
 
                if (sig_kernel_stop(signr)) {
@@ -1822,7 +1907,7 @@ relock:
 
                                /* signals can be posted during this window */
 
-                               if (is_orphaned_pgrp(process_group(current)))
+                               if (is_current_pgrp_orphaned())
                                        goto relock;
 
                                spin_lock_irq(&current->sighand->siglock);
@@ -1872,7 +1957,6 @@ EXPORT_SYMBOL(recalc_sigpending);
 EXPORT_SYMBOL_GPL(dequeue_signal);
 EXPORT_SYMBOL(flush_signals);
 EXPORT_SYMBOL(force_sig);
-EXPORT_SYMBOL(kill_pg);
 EXPORT_SYMBOL(kill_proc);
 EXPORT_SYMBOL(ptrace_notify);
 EXPORT_SYMBOL(send_sig);
@@ -2199,7 +2283,7 @@ static int do_tkill(int tgid, int pid, int sig)
  *  @pid: the PID of the thread
  *  @sig: signal to be sent
  *
- *  This syscall also checks the tgid and returns -ESRCH even if the PID
+ *  This syscall also checks the @tgid and returns -ESRCH even if the PID
  *  exists but it's not belonging to the target process anymore. This
  *  method solves the problem of threads exiting and PIDs getting reused.
  */
@@ -2544,11 +2628,12 @@ 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 =
-               kmem_cache_create("sigqueue",
-                                 sizeof(struct sigqueue),
-                                 __alignof__(struct sigqueue),
-                                 SLAB_PANIC, NULL, NULL);
+       sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
 }