ufs: fix truncated values handling 64 bit metadata
[linux-2.6.git] / kernel / signal.c
blob86c32b884f8efbb8f071b6343b3c37667b4e8fa3
1 /*
2 * linux/kernel/signal.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
9 * Changes to use preallocated sigqueue structures
10 * to allow signals to be sent reliably.
13 #include <linux/slab.h>
14 #include <linux/module.h>
15 #include <linux/init.h>
16 #include <linux/sched.h>
17 #include <linux/fs.h>
18 #include <linux/tty.h>
19 #include <linux/binfmts.h>
20 #include <linux/security.h>
21 #include <linux/syscalls.h>
22 #include <linux/ptrace.h>
23 #include <linux/signal.h>
24 #include <linux/signalfd.h>
25 #include <linux/ratelimit.h>
26 #include <linux/tracehook.h>
27 #include <linux/capability.h>
28 #include <linux/freezer.h>
29 #include <linux/pid_namespace.h>
30 #include <linux/nsproxy.h>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/signal.h>
34 #include <asm/param.h>
35 #include <asm/uaccess.h>
36 #include <asm/unistd.h>
37 #include <asm/siginfo.h>
38 #include "audit.h" /* audit_signal_info() */
41 * SLAB caches for signal bits.
44 static struct kmem_cache *sigqueue_cachep;
46 int print_fatal_signals __read_mostly;
48 static void __user *sig_handler(struct task_struct *t, int sig)
50 return t->sighand->action[sig - 1].sa.sa_handler;
53 static int sig_handler_ignored(void __user *handler, int sig)
55 /* Is it explicitly or implicitly ignored? */
56 return handler == SIG_IGN ||
57 (handler == SIG_DFL && sig_kernel_ignore(sig));
60 static int sig_task_ignored(struct task_struct *t, int sig,
61 int from_ancestor_ns)
63 void __user *handler;
65 handler = sig_handler(t, sig);
67 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
68 handler == SIG_DFL && !from_ancestor_ns)
69 return 1;
71 return sig_handler_ignored(handler, sig);
74 static int sig_ignored(struct task_struct *t, int sig, int from_ancestor_ns)
77 * Blocked signals are never ignored, since the
78 * signal handler may change by the time it is
79 * unblocked.
81 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
82 return 0;
84 if (!sig_task_ignored(t, sig, from_ancestor_ns))
85 return 0;
88 * Tracers may want to know about even ignored signals.
90 return !tracehook_consider_ignored_signal(t, sig);
94 * Re-calculate pending state from the set of locally pending
95 * signals, globally pending signals, and blocked signals.
97 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
99 unsigned long ready;
100 long i;
102 switch (_NSIG_WORDS) {
103 default:
104 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
105 ready |= signal->sig[i] &~ blocked->sig[i];
106 break;
108 case 4: ready = signal->sig[3] &~ blocked->sig[3];
109 ready |= signal->sig[2] &~ blocked->sig[2];
110 ready |= signal->sig[1] &~ blocked->sig[1];
111 ready |= signal->sig[0] &~ blocked->sig[0];
112 break;
114 case 2: ready = signal->sig[1] &~ blocked->sig[1];
115 ready |= signal->sig[0] &~ blocked->sig[0];
116 break;
118 case 1: ready = signal->sig[0] &~ blocked->sig[0];
120 return ready != 0;
123 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
125 static int recalc_sigpending_tsk(struct task_struct *t)
127 if ((t->group_stop & GROUP_STOP_PENDING) ||
128 PENDING(&t->pending, &t->blocked) ||
129 PENDING(&t->signal->shared_pending, &t->blocked)) {
130 set_tsk_thread_flag(t, TIF_SIGPENDING);
131 return 1;
134 * We must never clear the flag in another thread, or in current
135 * when it's possible the current syscall is returning -ERESTART*.
136 * So we don't clear it here, and only callers who know they should do.
138 return 0;
142 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
143 * This is superfluous when called on current, the wakeup is a harmless no-op.
145 void recalc_sigpending_and_wake(struct task_struct *t)
147 if (recalc_sigpending_tsk(t))
148 signal_wake_up(t, 0);
151 void recalc_sigpending(void)
153 if (unlikely(tracehook_force_sigpending()))
154 set_thread_flag(TIF_SIGPENDING);
155 else if (!recalc_sigpending_tsk(current) && !freezing(current))
156 clear_thread_flag(TIF_SIGPENDING);
160 /* Given the mask, find the first available signal that should be serviced. */
162 #define SYNCHRONOUS_MASK \
163 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
164 sigmask(SIGTRAP) | sigmask(SIGFPE))
166 int next_signal(struct sigpending *pending, sigset_t *mask)
168 unsigned long i, *s, *m, x;
169 int sig = 0;
171 s = pending->signal.sig;
172 m = mask->sig;
175 * Handle the first word specially: it contains the
176 * synchronous signals that need to be dequeued first.
178 x = *s &~ *m;
179 if (x) {
180 if (x & SYNCHRONOUS_MASK)
181 x &= SYNCHRONOUS_MASK;
182 sig = ffz(~x) + 1;
183 return sig;
186 switch (_NSIG_WORDS) {
187 default:
188 for (i = 1; i < _NSIG_WORDS; ++i) {
189 x = *++s &~ *++m;
190 if (!x)
191 continue;
192 sig = ffz(~x) + i*_NSIG_BPW + 1;
193 break;
195 break;
197 case 2:
198 x = s[1] &~ m[1];
199 if (!x)
200 break;
201 sig = ffz(~x) + _NSIG_BPW + 1;
202 break;
204 case 1:
205 /* Nothing to do */
206 break;
209 return sig;
212 static inline void print_dropped_signal(int sig)
214 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
216 if (!print_fatal_signals)
217 return;
219 if (!__ratelimit(&ratelimit_state))
220 return;
222 printk(KERN_INFO "%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
223 current->comm, current->pid, sig);
227 * task_clear_group_stop_trapping - clear group stop trapping bit
228 * @task: target task
230 * If GROUP_STOP_TRAPPING is set, a ptracer is waiting for us. Clear it
231 * and wake up the ptracer. Note that we don't need any further locking.
232 * @task->siglock guarantees that @task->parent points to the ptracer.
234 * CONTEXT:
235 * Must be called with @task->sighand->siglock held.
237 static void task_clear_group_stop_trapping(struct task_struct *task)
239 if (unlikely(task->group_stop & GROUP_STOP_TRAPPING)) {
240 task->group_stop &= ~GROUP_STOP_TRAPPING;
241 __wake_up_sync_key(&task->parent->signal->wait_chldexit,
242 TASK_UNINTERRUPTIBLE, 1, task);
247 * task_clear_group_stop_pending - clear pending group stop
248 * @task: target task
250 * Clear group stop states for @task.
252 * CONTEXT:
253 * Must be called with @task->sighand->siglock held.
255 void task_clear_group_stop_pending(struct task_struct *task)
257 task->group_stop &= ~(GROUP_STOP_PENDING | GROUP_STOP_CONSUME |
258 GROUP_STOP_DEQUEUED);
262 * task_participate_group_stop - participate in a group stop
263 * @task: task participating in a group stop
265 * @task has GROUP_STOP_PENDING set and is participating in a group stop.
266 * Group stop states are cleared and the group stop count is consumed if
267 * %GROUP_STOP_CONSUME was set. If the consumption completes the group
268 * stop, the appropriate %SIGNAL_* flags are set.
270 * CONTEXT:
271 * Must be called with @task->sighand->siglock held.
273 * RETURNS:
274 * %true if group stop completion should be notified to the parent, %false
275 * otherwise.
277 static bool task_participate_group_stop(struct task_struct *task)
279 struct signal_struct *sig = task->signal;
280 bool consume = task->group_stop & GROUP_STOP_CONSUME;
282 WARN_ON_ONCE(!(task->group_stop & GROUP_STOP_PENDING));
284 task_clear_group_stop_pending(task);
286 if (!consume)
287 return false;
289 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
290 sig->group_stop_count--;
293 * Tell the caller to notify completion iff we are entering into a
294 * fresh group stop. Read comment in do_signal_stop() for details.
296 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
297 sig->flags = SIGNAL_STOP_STOPPED;
298 return true;
300 return false;
304 * allocate a new signal queue record
305 * - this may be called without locks if and only if t == current, otherwise an
306 * appropriate lock must be held to stop the target task from exiting
308 static struct sigqueue *
309 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
311 struct sigqueue *q = NULL;
312 struct user_struct *user;
315 * Protect access to @t credentials. This can go away when all
316 * callers hold rcu read lock.
318 rcu_read_lock();
319 user = get_uid(__task_cred(t)->user);
320 atomic_inc(&user->sigpending);
321 rcu_read_unlock();
323 if (override_rlimit ||
324 atomic_read(&user->sigpending) <=
325 task_rlimit(t, RLIMIT_SIGPENDING)) {
326 q = kmem_cache_alloc(sigqueue_cachep, flags);
327 } else {
328 print_dropped_signal(sig);
331 if (unlikely(q == NULL)) {
332 atomic_dec(&user->sigpending);
333 free_uid(user);
334 } else {
335 INIT_LIST_HEAD(&q->list);
336 q->flags = 0;
337 q->user = user;
340 return q;
343 static void __sigqueue_free(struct sigqueue *q)
345 if (q->flags & SIGQUEUE_PREALLOC)
346 return;
347 atomic_dec(&q->user->sigpending);
348 free_uid(q->user);
349 kmem_cache_free(sigqueue_cachep, q);
352 void flush_sigqueue(struct sigpending *queue)
354 struct sigqueue *q;
356 sigemptyset(&queue->signal);
357 while (!list_empty(&queue->list)) {
358 q = list_entry(queue->list.next, struct sigqueue , list);
359 list_del_init(&q->list);
360 __sigqueue_free(q);
365 * Flush all pending signals for a task.
367 void __flush_signals(struct task_struct *t)
369 clear_tsk_thread_flag(t, TIF_SIGPENDING);
370 flush_sigqueue(&t->pending);
371 flush_sigqueue(&t->signal->shared_pending);
374 void flush_signals(struct task_struct *t)
376 unsigned long flags;
378 spin_lock_irqsave(&t->sighand->siglock, flags);
379 __flush_signals(t);
380 spin_unlock_irqrestore(&t->sighand->siglock, flags);
383 static void __flush_itimer_signals(struct sigpending *pending)
385 sigset_t signal, retain;
386 struct sigqueue *q, *n;
388 signal = pending->signal;
389 sigemptyset(&retain);
391 list_for_each_entry_safe(q, n, &pending->list, list) {
392 int sig = q->info.si_signo;
394 if (likely(q->info.si_code != SI_TIMER)) {
395 sigaddset(&retain, sig);
396 } else {
397 sigdelset(&signal, sig);
398 list_del_init(&q->list);
399 __sigqueue_free(q);
403 sigorsets(&pending->signal, &signal, &retain);
406 void flush_itimer_signals(void)
408 struct task_struct *tsk = current;
409 unsigned long flags;
411 spin_lock_irqsave(&tsk->sighand->siglock, flags);
412 __flush_itimer_signals(&tsk->pending);
413 __flush_itimer_signals(&tsk->signal->shared_pending);
414 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
417 void ignore_signals(struct task_struct *t)
419 int i;
421 for (i = 0; i < _NSIG; ++i)
422 t->sighand->action[i].sa.sa_handler = SIG_IGN;
424 flush_signals(t);
428 * Flush all handlers for a task.
431 void
432 flush_signal_handlers(struct task_struct *t, int force_default)
434 int i;
435 struct k_sigaction *ka = &t->sighand->action[0];
436 for (i = _NSIG ; i != 0 ; i--) {
437 if (force_default || ka->sa.sa_handler != SIG_IGN)
438 ka->sa.sa_handler = SIG_DFL;
439 ka->sa.sa_flags = 0;
440 sigemptyset(&ka->sa.sa_mask);
441 ka++;
445 int unhandled_signal(struct task_struct *tsk, int sig)
447 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
448 if (is_global_init(tsk))
449 return 1;
450 if (handler != SIG_IGN && handler != SIG_DFL)
451 return 0;
452 return !tracehook_consider_fatal_signal(tsk, sig);
456 * Notify the system that a driver wants to block all signals for this
457 * process, and wants to be notified if any signals at all were to be
458 * sent/acted upon. If the notifier routine returns non-zero, then the
459 * signal will be acted upon after all. If the notifier routine returns 0,
460 * then then signal will be blocked. Only one block per process is
461 * allowed. priv is a pointer to private data that the notifier routine
462 * can use to determine if the signal should be blocked or not.
464 void
465 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
467 unsigned long flags;
469 spin_lock_irqsave(&current->sighand->siglock, flags);
470 current->notifier_mask = mask;
471 current->notifier_data = priv;
472 current->notifier = notifier;
473 spin_unlock_irqrestore(&current->sighand->siglock, flags);
476 /* Notify the system that blocking has ended. */
478 void
479 unblock_all_signals(void)
481 unsigned long flags;
483 spin_lock_irqsave(&current->sighand->siglock, flags);
484 current->notifier = NULL;
485 current->notifier_data = NULL;
486 recalc_sigpending();
487 spin_unlock_irqrestore(&current->sighand->siglock, flags);
490 static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
492 struct sigqueue *q, *first = NULL;
495 * Collect the siginfo appropriate to this signal. Check if
496 * there is another siginfo for the same signal.
498 list_for_each_entry(q, &list->list, list) {
499 if (q->info.si_signo == sig) {
500 if (first)
501 goto still_pending;
502 first = q;
506 sigdelset(&list->signal, sig);
508 if (first) {
509 still_pending:
510 list_del_init(&first->list);
511 copy_siginfo(info, &first->info);
512 __sigqueue_free(first);
513 } else {
515 * Ok, it wasn't in the queue. This must be
516 * a fast-pathed signal or we must have been
517 * out of queue space. So zero out the info.
519 info->si_signo = sig;
520 info->si_errno = 0;
521 info->si_code = SI_USER;
522 info->si_pid = 0;
523 info->si_uid = 0;
527 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
528 siginfo_t *info)
530 int sig = next_signal(pending, mask);
532 if (sig) {
533 if (current->notifier) {
534 if (sigismember(current->notifier_mask, sig)) {
535 if (!(current->notifier)(current->notifier_data)) {
536 clear_thread_flag(TIF_SIGPENDING);
537 return 0;
542 collect_signal(sig, pending, info);
545 return sig;
549 * Dequeue a signal and return the element to the caller, which is
550 * expected to free it.
552 * All callers have to hold the siglock.
554 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
556 int signr;
558 /* We only dequeue private signals from ourselves, we don't let
559 * signalfd steal them
561 signr = __dequeue_signal(&tsk->pending, mask, info);
562 if (!signr) {
563 signr = __dequeue_signal(&tsk->signal->shared_pending,
564 mask, info);
566 * itimer signal ?
568 * itimers are process shared and we restart periodic
569 * itimers in the signal delivery path to prevent DoS
570 * attacks in the high resolution timer case. This is
571 * compliant with the old way of self-restarting
572 * itimers, as the SIGALRM is a legacy signal and only
573 * queued once. Changing the restart behaviour to
574 * restart the timer in the signal dequeue path is
575 * reducing the timer noise on heavy loaded !highres
576 * systems too.
578 if (unlikely(signr == SIGALRM)) {
579 struct hrtimer *tmr = &tsk->signal->real_timer;
581 if (!hrtimer_is_queued(tmr) &&
582 tsk->signal->it_real_incr.tv64 != 0) {
583 hrtimer_forward(tmr, tmr->base->get_time(),
584 tsk->signal->it_real_incr);
585 hrtimer_restart(tmr);
590 recalc_sigpending();
591 if (!signr)
592 return 0;
594 if (unlikely(sig_kernel_stop(signr))) {
596 * Set a marker that we have dequeued a stop signal. Our
597 * caller might release the siglock and then the pending
598 * stop signal it is about to process is no longer in the
599 * pending bitmasks, but must still be cleared by a SIGCONT
600 * (and overruled by a SIGKILL). So those cases clear this
601 * shared flag after we've set it. Note that this flag may
602 * remain set after the signal we return is ignored or
603 * handled. That doesn't matter because its only purpose
604 * is to alert stop-signal processing code when another
605 * processor has come along and cleared the flag.
607 current->group_stop |= GROUP_STOP_DEQUEUED;
609 if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
611 * Release the siglock to ensure proper locking order
612 * of timer locks outside of siglocks. Note, we leave
613 * irqs disabled here, since the posix-timers code is
614 * about to disable them again anyway.
616 spin_unlock(&tsk->sighand->siglock);
617 do_schedule_next_timer(info);
618 spin_lock(&tsk->sighand->siglock);
620 return signr;
624 * Tell a process that it has a new active signal..
626 * NOTE! we rely on the previous spin_lock to
627 * lock interrupts for us! We can only be called with
628 * "siglock" held, and the local interrupt must
629 * have been disabled when that got acquired!
631 * No need to set need_resched since signal event passing
632 * goes through ->blocked
634 void signal_wake_up(struct task_struct *t, int resume)
636 unsigned int mask;
638 set_tsk_thread_flag(t, TIF_SIGPENDING);
641 * For SIGKILL, we want to wake it up in the stopped/traced/killable
642 * case. We don't check t->state here because there is a race with it
643 * executing another processor and just now entering stopped state.
644 * By using wake_up_state, we ensure the process will wake up and
645 * handle its death signal.
647 mask = TASK_INTERRUPTIBLE;
648 if (resume)
649 mask |= TASK_WAKEKILL;
650 if (!wake_up_state(t, mask))
651 kick_process(t);
655 * Remove signals in mask from the pending set and queue.
656 * Returns 1 if any signals were found.
658 * All callers must be holding the siglock.
660 * This version takes a sigset mask and looks at all signals,
661 * not just those in the first mask word.
663 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
665 struct sigqueue *q, *n;
666 sigset_t m;
668 sigandsets(&m, mask, &s->signal);
669 if (sigisemptyset(&m))
670 return 0;
672 sigandnsets(&s->signal, &s->signal, mask);
673 list_for_each_entry_safe(q, n, &s->list, list) {
674 if (sigismember(mask, q->info.si_signo)) {
675 list_del_init(&q->list);
676 __sigqueue_free(q);
679 return 1;
682 * Remove signals in mask from the pending set and queue.
683 * Returns 1 if any signals were found.
685 * All callers must be holding the siglock.
687 static int rm_from_queue(unsigned long mask, struct sigpending *s)
689 struct sigqueue *q, *n;
691 if (!sigtestsetmask(&s->signal, mask))
692 return 0;
694 sigdelsetmask(&s->signal, mask);
695 list_for_each_entry_safe(q, n, &s->list, list) {
696 if (q->info.si_signo < SIGRTMIN &&
697 (mask & sigmask(q->info.si_signo))) {
698 list_del_init(&q->list);
699 __sigqueue_free(q);
702 return 1;
705 static inline int is_si_special(const struct siginfo *info)
707 return info <= SEND_SIG_FORCED;
710 static inline bool si_fromuser(const struct siginfo *info)
712 return info == SEND_SIG_NOINFO ||
713 (!is_si_special(info) && SI_FROMUSER(info));
717 * called with RCU read lock from check_kill_permission()
719 static int kill_ok_by_cred(struct task_struct *t)
721 const struct cred *cred = current_cred();
722 const struct cred *tcred = __task_cred(t);
724 if (cred->user->user_ns == tcred->user->user_ns &&
725 (cred->euid == tcred->suid ||
726 cred->euid == tcred->uid ||
727 cred->uid == tcred->suid ||
728 cred->uid == tcred->uid))
729 return 1;
731 if (ns_capable(tcred->user->user_ns, CAP_KILL))
732 return 1;
734 return 0;
738 * Bad permissions for sending the signal
739 * - the caller must hold the RCU read lock
741 static int check_kill_permission(int sig, struct siginfo *info,
742 struct task_struct *t)
744 struct pid *sid;
745 int error;
747 if (!valid_signal(sig))
748 return -EINVAL;
750 if (!si_fromuser(info))
751 return 0;
753 error = audit_signal_info(sig, t); /* Let audit system see the signal */
754 if (error)
755 return error;
757 if (!same_thread_group(current, t) &&
758 !kill_ok_by_cred(t)) {
759 switch (sig) {
760 case SIGCONT:
761 sid = task_session(t);
763 * We don't return the error if sid == NULL. The
764 * task was unhashed, the caller must notice this.
766 if (!sid || sid == task_session(current))
767 break;
768 default:
769 return -EPERM;
773 return security_task_kill(t, info, sig, 0);
777 * Handle magic process-wide effects of stop/continue signals. Unlike
778 * the signal actions, these happen immediately at signal-generation
779 * time regardless of blocking, ignoring, or handling. This does the
780 * actual continuing for SIGCONT, but not the actual stopping for stop
781 * signals. The process stop is done as a signal action for SIG_DFL.
783 * Returns true if the signal should be actually delivered, otherwise
784 * it should be dropped.
786 static int prepare_signal(int sig, struct task_struct *p, int from_ancestor_ns)
788 struct signal_struct *signal = p->signal;
789 struct task_struct *t;
791 if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
793 * The process is in the middle of dying, nothing to do.
795 } else if (sig_kernel_stop(sig)) {
797 * This is a stop signal. Remove SIGCONT from all queues.
799 rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
800 t = p;
801 do {
802 rm_from_queue(sigmask(SIGCONT), &t->pending);
803 } while_each_thread(p, t);
804 } else if (sig == SIGCONT) {
805 unsigned int why;
807 * Remove all stop signals from all queues, wake all threads.
809 rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
810 t = p;
811 do {
812 task_clear_group_stop_pending(t);
813 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
814 wake_up_state(t, __TASK_STOPPED);
815 } while_each_thread(p, t);
818 * Notify the parent with CLD_CONTINUED if we were stopped.
820 * If we were in the middle of a group stop, we pretend it
821 * was already finished, and then continued. Since SIGCHLD
822 * doesn't queue we report only CLD_STOPPED, as if the next
823 * CLD_CONTINUED was dropped.
825 why = 0;
826 if (signal->flags & SIGNAL_STOP_STOPPED)
827 why |= SIGNAL_CLD_CONTINUED;
828 else if (signal->group_stop_count)
829 why |= SIGNAL_CLD_STOPPED;
831 if (why) {
833 * The first thread which returns from do_signal_stop()
834 * will take ->siglock, notice SIGNAL_CLD_MASK, and
835 * notify its parent. See get_signal_to_deliver().
837 signal->flags = why | SIGNAL_STOP_CONTINUED;
838 signal->group_stop_count = 0;
839 signal->group_exit_code = 0;
843 return !sig_ignored(p, sig, from_ancestor_ns);
847 * Test if P wants to take SIG. After we've checked all threads with this,
848 * it's equivalent to finding no threads not blocking SIG. Any threads not
849 * blocking SIG were ruled out because they are not running and already
850 * have pending signals. Such threads will dequeue from the shared queue
851 * as soon as they're available, so putting the signal on the shared queue
852 * will be equivalent to sending it to one such thread.
854 static inline int wants_signal(int sig, struct task_struct *p)
856 if (sigismember(&p->blocked, sig))
857 return 0;
858 if (p->flags & PF_EXITING)
859 return 0;
860 if (sig == SIGKILL)
861 return 1;
862 if (task_is_stopped_or_traced(p))
863 return 0;
864 return task_curr(p) || !signal_pending(p);
867 static void complete_signal(int sig, struct task_struct *p, int group)
869 struct signal_struct *signal = p->signal;
870 struct task_struct *t;
873 * Now find a thread we can wake up to take the signal off the queue.
875 * If the main thread wants the signal, it gets first crack.
876 * Probably the least surprising to the average bear.
878 if (wants_signal(sig, p))
879 t = p;
880 else if (!group || thread_group_empty(p))
882 * There is just one thread and it does not need to be woken.
883 * It will dequeue unblocked signals before it runs again.
885 return;
886 else {
888 * Otherwise try to find a suitable thread.
890 t = signal->curr_target;
891 while (!wants_signal(sig, t)) {
892 t = next_thread(t);
893 if (t == signal->curr_target)
895 * No thread needs to be woken.
896 * Any eligible threads will see
897 * the signal in the queue soon.
899 return;
901 signal->curr_target = t;
905 * Found a killable thread. If the signal will be fatal,
906 * then start taking the whole group down immediately.
908 if (sig_fatal(p, sig) &&
909 !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
910 !sigismember(&t->real_blocked, sig) &&
911 (sig == SIGKILL ||
912 !tracehook_consider_fatal_signal(t, sig))) {
914 * This signal will be fatal to the whole group.
916 if (!sig_kernel_coredump(sig)) {
918 * Start a group exit and wake everybody up.
919 * This way we don't have other threads
920 * running and doing things after a slower
921 * thread has the fatal signal pending.
923 signal->flags = SIGNAL_GROUP_EXIT;
924 signal->group_exit_code = sig;
925 signal->group_stop_count = 0;
926 t = p;
927 do {
928 task_clear_group_stop_pending(t);
929 sigaddset(&t->pending.signal, SIGKILL);
930 signal_wake_up(t, 1);
931 } while_each_thread(p, t);
932 return;
937 * The signal is already in the shared-pending queue.
938 * Tell the chosen thread to wake up and dequeue it.
940 signal_wake_up(t, sig == SIGKILL);
941 return;
944 static inline int legacy_queue(struct sigpending *signals, int sig)
946 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
949 static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
950 int group, int from_ancestor_ns)
952 struct sigpending *pending;
953 struct sigqueue *q;
954 int override_rlimit;
956 trace_signal_generate(sig, info, t);
958 assert_spin_locked(&t->sighand->siglock);
960 if (!prepare_signal(sig, t, from_ancestor_ns))
961 return 0;
963 pending = group ? &t->signal->shared_pending : &t->pending;
965 * Short-circuit ignored signals and support queuing
966 * exactly one non-rt signal, so that we can get more
967 * detailed information about the cause of the signal.
969 if (legacy_queue(pending, sig))
970 return 0;
972 * fast-pathed signals for kernel-internal things like SIGSTOP
973 * or SIGKILL.
975 if (info == SEND_SIG_FORCED)
976 goto out_set;
979 * Real-time signals must be queued if sent by sigqueue, or
980 * some other real-time mechanism. It is implementation
981 * defined whether kill() does so. We attempt to do so, on
982 * the principle of least surprise, but since kill is not
983 * allowed to fail with EAGAIN when low on memory we just
984 * make sure at least one signal gets delivered and don't
985 * pass on the info struct.
987 if (sig < SIGRTMIN)
988 override_rlimit = (is_si_special(info) || info->si_code >= 0);
989 else
990 override_rlimit = 0;
992 q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
993 override_rlimit);
994 if (q) {
995 list_add_tail(&q->list, &pending->list);
996 switch ((unsigned long) info) {
997 case (unsigned long) SEND_SIG_NOINFO:
998 q->info.si_signo = sig;
999 q->info.si_errno = 0;
1000 q->info.si_code = SI_USER;
1001 q->info.si_pid = task_tgid_nr_ns(current,
1002 task_active_pid_ns(t));
1003 q->info.si_uid = current_uid();
1004 break;
1005 case (unsigned long) SEND_SIG_PRIV:
1006 q->info.si_signo = sig;
1007 q->info.si_errno = 0;
1008 q->info.si_code = SI_KERNEL;
1009 q->info.si_pid = 0;
1010 q->info.si_uid = 0;
1011 break;
1012 default:
1013 copy_siginfo(&q->info, info);
1014 if (from_ancestor_ns)
1015 q->info.si_pid = 0;
1016 break;
1018 } else if (!is_si_special(info)) {
1019 if (sig >= SIGRTMIN && info->si_code != SI_USER) {
1021 * Queue overflow, abort. We may abort if the
1022 * signal was rt and sent by user using something
1023 * other than kill().
1025 trace_signal_overflow_fail(sig, group, info);
1026 return -EAGAIN;
1027 } else {
1029 * This is a silent loss of information. We still
1030 * send the signal, but the *info bits are lost.
1032 trace_signal_lose_info(sig, group, info);
1036 out_set:
1037 signalfd_notify(t, sig);
1038 sigaddset(&pending->signal, sig);
1039 complete_signal(sig, t, group);
1040 return 0;
1043 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
1044 int group)
1046 int from_ancestor_ns = 0;
1048 #ifdef CONFIG_PID_NS
1049 from_ancestor_ns = si_fromuser(info) &&
1050 !task_pid_nr_ns(current, task_active_pid_ns(t));
1051 #endif
1053 return __send_signal(sig, info, t, group, from_ancestor_ns);
1056 static void print_fatal_signal(struct pt_regs *regs, int signr)
1058 printk("%s/%d: potentially unexpected fatal signal %d.\n",
1059 current->comm, task_pid_nr(current), signr);
1061 #if defined(__i386__) && !defined(__arch_um__)
1062 printk("code at %08lx: ", regs->ip);
1064 int i;
1065 for (i = 0; i < 16; i++) {
1066 unsigned char insn;
1068 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1069 break;
1070 printk("%02x ", insn);
1073 #endif
1074 printk("\n");
1075 preempt_disable();
1076 show_regs(regs);
1077 preempt_enable();
1080 static int __init setup_print_fatal_signals(char *str)
1082 get_option (&str, &print_fatal_signals);
1084 return 1;
1087 __setup("print-fatal-signals=", setup_print_fatal_signals);
1090 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1092 return send_signal(sig, info, p, 1);
1095 static int
1096 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1098 return send_signal(sig, info, t, 0);
1101 int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
1102 bool group)
1104 unsigned long flags;
1105 int ret = -ESRCH;
1107 if (lock_task_sighand(p, &flags)) {
1108 ret = send_signal(sig, info, p, group);
1109 unlock_task_sighand(p, &flags);
1112 return ret;
1116 * Force a signal that the process can't ignore: if necessary
1117 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1119 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1120 * since we do not want to have a signal handler that was blocked
1121 * be invoked when user space had explicitly blocked it.
1123 * We don't want to have recursive SIGSEGV's etc, for example,
1124 * that is why we also clear SIGNAL_UNKILLABLE.
1127 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1129 unsigned long int flags;
1130 int ret, blocked, ignored;
1131 struct k_sigaction *action;
1133 spin_lock_irqsave(&t->sighand->siglock, flags);
1134 action = &t->sighand->action[sig-1];
1135 ignored = action->sa.sa_handler == SIG_IGN;
1136 blocked = sigismember(&t->blocked, sig);
1137 if (blocked || ignored) {
1138 action->sa.sa_handler = SIG_DFL;
1139 if (blocked) {
1140 sigdelset(&t->blocked, sig);
1141 recalc_sigpending_and_wake(t);
1144 if (action->sa.sa_handler == SIG_DFL)
1145 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1146 ret = specific_send_sig_info(sig, info, t);
1147 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1149 return ret;
1153 * Nuke all other threads in the group.
1155 int zap_other_threads(struct task_struct *p)
1157 struct task_struct *t = p;
1158 int count = 0;
1160 p->signal->group_stop_count = 0;
1162 while_each_thread(p, t) {
1163 task_clear_group_stop_pending(t);
1164 count++;
1166 /* Don't bother with already dead threads */
1167 if (t->exit_state)
1168 continue;
1169 sigaddset(&t->pending.signal, SIGKILL);
1170 signal_wake_up(t, 1);
1173 return count;
1176 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1177 unsigned long *flags)
1179 struct sighand_struct *sighand;
1181 rcu_read_lock();
1182 for (;;) {
1183 sighand = rcu_dereference(tsk->sighand);
1184 if (unlikely(sighand == NULL))
1185 break;
1187 spin_lock_irqsave(&sighand->siglock, *flags);
1188 if (likely(sighand == tsk->sighand))
1189 break;
1190 spin_unlock_irqrestore(&sighand->siglock, *flags);
1192 rcu_read_unlock();
1194 return sighand;
1198 * send signal info to all the members of a group
1200 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1202 int ret;
1204 rcu_read_lock();
1205 ret = check_kill_permission(sig, info, p);
1206 rcu_read_unlock();
1208 if (!ret && sig)
1209 ret = do_send_sig_info(sig, info, p, true);
1211 return ret;
1215 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1216 * control characters do (^C, ^Z etc)
1217 * - the caller must hold at least a readlock on tasklist_lock
1219 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1221 struct task_struct *p = NULL;
1222 int retval, success;
1224 success = 0;
1225 retval = -ESRCH;
1226 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1227 int err = group_send_sig_info(sig, info, p);
1228 success |= !err;
1229 retval = err;
1230 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1231 return success ? 0 : retval;
1234 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1236 int error = -ESRCH;
1237 struct task_struct *p;
1239 rcu_read_lock();
1240 retry:
1241 p = pid_task(pid, PIDTYPE_PID);
1242 if (p) {
1243 error = group_send_sig_info(sig, info, p);
1244 if (unlikely(error == -ESRCH))
1246 * The task was unhashed in between, try again.
1247 * If it is dead, pid_task() will return NULL,
1248 * if we race with de_thread() it will find the
1249 * new leader.
1251 goto retry;
1253 rcu_read_unlock();
1255 return error;
1258 int kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1260 int error;
1261 rcu_read_lock();
1262 error = kill_pid_info(sig, info, find_vpid(pid));
1263 rcu_read_unlock();
1264 return error;
1267 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1268 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1269 uid_t uid, uid_t euid, u32 secid)
1271 int ret = -EINVAL;
1272 struct task_struct *p;
1273 const struct cred *pcred;
1274 unsigned long flags;
1276 if (!valid_signal(sig))
1277 return ret;
1279 rcu_read_lock();
1280 p = pid_task(pid, PIDTYPE_PID);
1281 if (!p) {
1282 ret = -ESRCH;
1283 goto out_unlock;
1285 pcred = __task_cred(p);
1286 if (si_fromuser(info) &&
1287 euid != pcred->suid && euid != pcred->uid &&
1288 uid != pcred->suid && uid != pcred->uid) {
1289 ret = -EPERM;
1290 goto out_unlock;
1292 ret = security_task_kill(p, info, sig, secid);
1293 if (ret)
1294 goto out_unlock;
1296 if (sig) {
1297 if (lock_task_sighand(p, &flags)) {
1298 ret = __send_signal(sig, info, p, 1, 0);
1299 unlock_task_sighand(p, &flags);
1300 } else
1301 ret = -ESRCH;
1303 out_unlock:
1304 rcu_read_unlock();
1305 return ret;
1307 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1310 * kill_something_info() interprets pid in interesting ways just like kill(2).
1312 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1313 * is probably wrong. Should make it like BSD or SYSV.
1316 static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1318 int ret;
1320 if (pid > 0) {
1321 rcu_read_lock();
1322 ret = kill_pid_info(sig, info, find_vpid(pid));
1323 rcu_read_unlock();
1324 return ret;
1327 read_lock(&tasklist_lock);
1328 if (pid != -1) {
1329 ret = __kill_pgrp_info(sig, info,
1330 pid ? find_vpid(-pid) : task_pgrp(current));
1331 } else {
1332 int retval = 0, count = 0;
1333 struct task_struct * p;
1335 for_each_process(p) {
1336 if (task_pid_vnr(p) > 1 &&
1337 !same_thread_group(p, current)) {
1338 int err = group_send_sig_info(sig, info, p);
1339 ++count;
1340 if (err != -EPERM)
1341 retval = err;
1344 ret = count ? retval : -ESRCH;
1346 read_unlock(&tasklist_lock);
1348 return ret;
1352 * These are for backward compatibility with the rest of the kernel source.
1355 int send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1358 * Make sure legacy kernel users don't send in bad values
1359 * (normal paths check this in check_kill_permission).
1361 if (!valid_signal(sig))
1362 return -EINVAL;
1364 return do_send_sig_info(sig, info, p, false);
1367 #define __si_special(priv) \
1368 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1371 send_sig(int sig, struct task_struct *p, int priv)
1373 return send_sig_info(sig, __si_special(priv), p);
1376 void
1377 force_sig(int sig, struct task_struct *p)
1379 force_sig_info(sig, SEND_SIG_PRIV, p);
1383 * When things go south during signal handling, we
1384 * will force a SIGSEGV. And if the signal that caused
1385 * the problem was already a SIGSEGV, we'll want to
1386 * make sure we don't even try to deliver the signal..
1389 force_sigsegv(int sig, struct task_struct *p)
1391 if (sig == SIGSEGV) {
1392 unsigned long flags;
1393 spin_lock_irqsave(&p->sighand->siglock, flags);
1394 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1395 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1397 force_sig(SIGSEGV, p);
1398 return 0;
1401 int kill_pgrp(struct pid *pid, int sig, int priv)
1403 int ret;
1405 read_lock(&tasklist_lock);
1406 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1407 read_unlock(&tasklist_lock);
1409 return ret;
1411 EXPORT_SYMBOL(kill_pgrp);
1413 int kill_pid(struct pid *pid, int sig, int priv)
1415 return kill_pid_info(sig, __si_special(priv), pid);
1417 EXPORT_SYMBOL(kill_pid);
1420 * These functions support sending signals using preallocated sigqueue
1421 * structures. This is needed "because realtime applications cannot
1422 * afford to lose notifications of asynchronous events, like timer
1423 * expirations or I/O completions". In the case of POSIX Timers
1424 * we allocate the sigqueue structure from the timer_create. If this
1425 * allocation fails we are able to report the failure to the application
1426 * with an EAGAIN error.
1428 struct sigqueue *sigqueue_alloc(void)
1430 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1432 if (q)
1433 q->flags |= SIGQUEUE_PREALLOC;
1435 return q;
1438 void sigqueue_free(struct sigqueue *q)
1440 unsigned long flags;
1441 spinlock_t *lock = &current->sighand->siglock;
1443 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1445 * We must hold ->siglock while testing q->list
1446 * to serialize with collect_signal() or with
1447 * __exit_signal()->flush_sigqueue().
1449 spin_lock_irqsave(lock, flags);
1450 q->flags &= ~SIGQUEUE_PREALLOC;
1452 * If it is queued it will be freed when dequeued,
1453 * like the "regular" sigqueue.
1455 if (!list_empty(&q->list))
1456 q = NULL;
1457 spin_unlock_irqrestore(lock, flags);
1459 if (q)
1460 __sigqueue_free(q);
1463 int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1465 int sig = q->info.si_signo;
1466 struct sigpending *pending;
1467 unsigned long flags;
1468 int ret;
1470 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1472 ret = -1;
1473 if (!likely(lock_task_sighand(t, &flags)))
1474 goto ret;
1476 ret = 1; /* the signal is ignored */
1477 if (!prepare_signal(sig, t, 0))
1478 goto out;
1480 ret = 0;
1481 if (unlikely(!list_empty(&q->list))) {
1483 * If an SI_TIMER entry is already queue just increment
1484 * the overrun count.
1486 BUG_ON(q->info.si_code != SI_TIMER);
1487 q->info.si_overrun++;
1488 goto out;
1490 q->info.si_overrun = 0;
1492 signalfd_notify(t, sig);
1493 pending = group ? &t->signal->shared_pending : &t->pending;
1494 list_add_tail(&q->list, &pending->list);
1495 sigaddset(&pending->signal, sig);
1496 complete_signal(sig, t, group);
1497 out:
1498 unlock_task_sighand(t, &flags);
1499 ret:
1500 return ret;
1504 * Let a parent know about the death of a child.
1505 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1507 * Returns -1 if our parent ignored us and so we've switched to
1508 * self-reaping, or else @sig.
1510 int do_notify_parent(struct task_struct *tsk, int sig)
1512 struct siginfo info;
1513 unsigned long flags;
1514 struct sighand_struct *psig;
1515 int ret = sig;
1517 BUG_ON(sig == -1);
1519 /* do_notify_parent_cldstop should have been called instead. */
1520 BUG_ON(task_is_stopped_or_traced(tsk));
1522 BUG_ON(!task_ptrace(tsk) &&
1523 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1525 info.si_signo = sig;
1526 info.si_errno = 0;
1528 * we are under tasklist_lock here so our parent is tied to
1529 * us and cannot exit and release its namespace.
1531 * the only it can is to switch its nsproxy with sys_unshare,
1532 * bu uncharing pid namespaces is not allowed, so we'll always
1533 * see relevant namespace
1535 * write_lock() currently calls preempt_disable() which is the
1536 * same as rcu_read_lock(), but according to Oleg, this is not
1537 * correct to rely on this
1539 rcu_read_lock();
1540 info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1541 info.si_uid = __task_cred(tsk)->uid;
1542 rcu_read_unlock();
1544 info.si_utime = cputime_to_clock_t(cputime_add(tsk->utime,
1545 tsk->signal->utime));
1546 info.si_stime = cputime_to_clock_t(cputime_add(tsk->stime,
1547 tsk->signal->stime));
1549 info.si_status = tsk->exit_code & 0x7f;
1550 if (tsk->exit_code & 0x80)
1551 info.si_code = CLD_DUMPED;
1552 else if (tsk->exit_code & 0x7f)
1553 info.si_code = CLD_KILLED;
1554 else {
1555 info.si_code = CLD_EXITED;
1556 info.si_status = tsk->exit_code >> 8;
1559 psig = tsk->parent->sighand;
1560 spin_lock_irqsave(&psig->siglock, flags);
1561 if (!task_ptrace(tsk) && sig == SIGCHLD &&
1562 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1563 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1565 * We are exiting and our parent doesn't care. POSIX.1
1566 * defines special semantics for setting SIGCHLD to SIG_IGN
1567 * or setting the SA_NOCLDWAIT flag: we should be reaped
1568 * automatically and not left for our parent's wait4 call.
1569 * Rather than having the parent do it as a magic kind of
1570 * signal handler, we just set this to tell do_exit that we
1571 * can be cleaned up without becoming a zombie. Note that
1572 * we still call __wake_up_parent in this case, because a
1573 * blocked sys_wait4 might now return -ECHILD.
1575 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1576 * is implementation-defined: we do (if you don't want
1577 * it, just use SIG_IGN instead).
1579 ret = tsk->exit_signal = -1;
1580 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1581 sig = -1;
1583 if (valid_signal(sig) && sig > 0)
1584 __group_send_sig_info(sig, &info, tsk->parent);
1585 __wake_up_parent(tsk, tsk->parent);
1586 spin_unlock_irqrestore(&psig->siglock, flags);
1588 return ret;
1592 * do_notify_parent_cldstop - notify parent of stopped/continued state change
1593 * @tsk: task reporting the state change
1594 * @for_ptracer: the notification is for ptracer
1595 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
1597 * Notify @tsk's parent that the stopped/continued state has changed. If
1598 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
1599 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
1601 * CONTEXT:
1602 * Must be called with tasklist_lock at least read locked.
1604 static void do_notify_parent_cldstop(struct task_struct *tsk,
1605 bool for_ptracer, int why)
1607 struct siginfo info;
1608 unsigned long flags;
1609 struct task_struct *parent;
1610 struct sighand_struct *sighand;
1612 if (for_ptracer) {
1613 parent = tsk->parent;
1614 } else {
1615 tsk = tsk->group_leader;
1616 parent = tsk->real_parent;
1619 info.si_signo = SIGCHLD;
1620 info.si_errno = 0;
1622 * see comment in do_notify_parent() about the following 4 lines
1624 rcu_read_lock();
1625 info.si_pid = task_pid_nr_ns(tsk, parent->nsproxy->pid_ns);
1626 info.si_uid = __task_cred(tsk)->uid;
1627 rcu_read_unlock();
1629 info.si_utime = cputime_to_clock_t(tsk->utime);
1630 info.si_stime = cputime_to_clock_t(tsk->stime);
1632 info.si_code = why;
1633 switch (why) {
1634 case CLD_CONTINUED:
1635 info.si_status = SIGCONT;
1636 break;
1637 case CLD_STOPPED:
1638 info.si_status = tsk->signal->group_exit_code & 0x7f;
1639 break;
1640 case CLD_TRAPPED:
1641 info.si_status = tsk->exit_code & 0x7f;
1642 break;
1643 default:
1644 BUG();
1647 sighand = parent->sighand;
1648 spin_lock_irqsave(&sighand->siglock, flags);
1649 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1650 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1651 __group_send_sig_info(SIGCHLD, &info, parent);
1653 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1655 __wake_up_parent(tsk, parent);
1656 spin_unlock_irqrestore(&sighand->siglock, flags);
1659 static inline int may_ptrace_stop(void)
1661 if (!likely(task_ptrace(current)))
1662 return 0;
1664 * Are we in the middle of do_coredump?
1665 * If so and our tracer is also part of the coredump stopping
1666 * is a deadlock situation, and pointless because our tracer
1667 * is dead so don't allow us to stop.
1668 * If SIGKILL was already sent before the caller unlocked
1669 * ->siglock we must see ->core_state != NULL. Otherwise it
1670 * is safe to enter schedule().
1672 if (unlikely(current->mm->core_state) &&
1673 unlikely(current->mm == current->parent->mm))
1674 return 0;
1676 return 1;
1680 * Return non-zero if there is a SIGKILL that should be waking us up.
1681 * Called with the siglock held.
1683 static int sigkill_pending(struct task_struct *tsk)
1685 return sigismember(&tsk->pending.signal, SIGKILL) ||
1686 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1690 * Test whether the target task of the usual cldstop notification - the
1691 * real_parent of @child - is in the same group as the ptracer.
1693 static bool real_parent_is_ptracer(struct task_struct *child)
1695 return same_thread_group(child->parent, child->real_parent);
1699 * This must be called with current->sighand->siglock held.
1701 * This should be the path for all ptrace stops.
1702 * We always set current->last_siginfo while stopped here.
1703 * That makes it a way to test a stopped process for
1704 * being ptrace-stopped vs being job-control-stopped.
1706 * If we actually decide not to stop at all because the tracer
1707 * is gone, we keep current->exit_code unless clear_code.
1709 static void ptrace_stop(int exit_code, int why, int clear_code, siginfo_t *info)
1710 __releases(&current->sighand->siglock)
1711 __acquires(&current->sighand->siglock)
1713 bool gstop_done = false;
1715 if (arch_ptrace_stop_needed(exit_code, info)) {
1717 * The arch code has something special to do before a
1718 * ptrace stop. This is allowed to block, e.g. for faults
1719 * on user stack pages. We can't keep the siglock while
1720 * calling arch_ptrace_stop, so we must release it now.
1721 * To preserve proper semantics, we must do this before
1722 * any signal bookkeeping like checking group_stop_count.
1723 * Meanwhile, a SIGKILL could come in before we retake the
1724 * siglock. That must prevent us from sleeping in TASK_TRACED.
1725 * So after regaining the lock, we must check for SIGKILL.
1727 spin_unlock_irq(&current->sighand->siglock);
1728 arch_ptrace_stop(exit_code, info);
1729 spin_lock_irq(&current->sighand->siglock);
1730 if (sigkill_pending(current))
1731 return;
1735 * If @why is CLD_STOPPED, we're trapping to participate in a group
1736 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
1737 * while siglock was released for the arch hook, PENDING could be
1738 * clear now. We act as if SIGCONT is received after TASK_TRACED
1739 * is entered - ignore it.
1741 if (why == CLD_STOPPED && (current->group_stop & GROUP_STOP_PENDING))
1742 gstop_done = task_participate_group_stop(current);
1744 current->last_siginfo = info;
1745 current->exit_code = exit_code;
1748 * TRACED should be visible before TRAPPING is cleared; otherwise,
1749 * the tracer might fail do_wait().
1751 set_current_state(TASK_TRACED);
1754 * We're committing to trapping. Clearing GROUP_STOP_TRAPPING and
1755 * transition to TASK_TRACED should be atomic with respect to
1756 * siglock. This hsould be done after the arch hook as siglock is
1757 * released and regrabbed across it.
1759 task_clear_group_stop_trapping(current);
1761 spin_unlock_irq(&current->sighand->siglock);
1762 read_lock(&tasklist_lock);
1763 if (may_ptrace_stop()) {
1765 * Notify parents of the stop.
1767 * While ptraced, there are two parents - the ptracer and
1768 * the real_parent of the group_leader. The ptracer should
1769 * know about every stop while the real parent is only
1770 * interested in the completion of group stop. The states
1771 * for the two don't interact with each other. Notify
1772 * separately unless they're gonna be duplicates.
1774 do_notify_parent_cldstop(current, true, why);
1775 if (gstop_done && !real_parent_is_ptracer(current))
1776 do_notify_parent_cldstop(current, false, why);
1779 * Don't want to allow preemption here, because
1780 * sys_ptrace() needs this task to be inactive.
1782 * XXX: implement read_unlock_no_resched().
1784 preempt_disable();
1785 read_unlock(&tasklist_lock);
1786 preempt_enable_no_resched();
1787 schedule();
1788 } else {
1790 * By the time we got the lock, our tracer went away.
1791 * Don't drop the lock yet, another tracer may come.
1793 * If @gstop_done, the ptracer went away between group stop
1794 * completion and here. During detach, it would have set
1795 * GROUP_STOP_PENDING on us and we'll re-enter TASK_STOPPED
1796 * in do_signal_stop() on return, so notifying the real
1797 * parent of the group stop completion is enough.
1799 if (gstop_done)
1800 do_notify_parent_cldstop(current, false, why);
1802 __set_current_state(TASK_RUNNING);
1803 if (clear_code)
1804 current->exit_code = 0;
1805 read_unlock(&tasklist_lock);
1809 * While in TASK_TRACED, we were considered "frozen enough".
1810 * Now that we woke up, it's crucial if we're supposed to be
1811 * frozen that we freeze now before running anything substantial.
1813 try_to_freeze();
1816 * We are back. Now reacquire the siglock before touching
1817 * last_siginfo, so that we are sure to have synchronized with
1818 * any signal-sending on another CPU that wants to examine it.
1820 spin_lock_irq(&current->sighand->siglock);
1821 current->last_siginfo = NULL;
1824 * Queued signals ignored us while we were stopped for tracing.
1825 * So check for any that we should take before resuming user mode.
1826 * This sets TIF_SIGPENDING, but never clears it.
1828 recalc_sigpending_tsk(current);
1831 void ptrace_notify(int exit_code)
1833 siginfo_t info;
1835 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1837 memset(&info, 0, sizeof info);
1838 info.si_signo = SIGTRAP;
1839 info.si_code = exit_code;
1840 info.si_pid = task_pid_vnr(current);
1841 info.si_uid = current_uid();
1843 /* Let the debugger run. */
1844 spin_lock_irq(&current->sighand->siglock);
1845 ptrace_stop(exit_code, CLD_TRAPPED, 1, &info);
1846 spin_unlock_irq(&current->sighand->siglock);
1850 * This performs the stopping for SIGSTOP and other stop signals.
1851 * We have to stop all threads in the thread group.
1852 * Returns non-zero if we've actually stopped and released the siglock.
1853 * Returns zero if we didn't stop and still hold the siglock.
1855 static int do_signal_stop(int signr)
1857 struct signal_struct *sig = current->signal;
1859 if (!(current->group_stop & GROUP_STOP_PENDING)) {
1860 unsigned int gstop = GROUP_STOP_PENDING | GROUP_STOP_CONSUME;
1861 struct task_struct *t;
1863 /* signr will be recorded in task->group_stop for retries */
1864 WARN_ON_ONCE(signr & ~GROUP_STOP_SIGMASK);
1866 if (!likely(current->group_stop & GROUP_STOP_DEQUEUED) ||
1867 unlikely(signal_group_exit(sig)))
1868 return 0;
1870 * There is no group stop already in progress. We must
1871 * initiate one now.
1873 * While ptraced, a task may be resumed while group stop is
1874 * still in effect and then receive a stop signal and
1875 * initiate another group stop. This deviates from the
1876 * usual behavior as two consecutive stop signals can't
1877 * cause two group stops when !ptraced. That is why we
1878 * also check !task_is_stopped(t) below.
1880 * The condition can be distinguished by testing whether
1881 * SIGNAL_STOP_STOPPED is already set. Don't generate
1882 * group_exit_code in such case.
1884 * This is not necessary for SIGNAL_STOP_CONTINUED because
1885 * an intervening stop signal is required to cause two
1886 * continued events regardless of ptrace.
1888 if (!(sig->flags & SIGNAL_STOP_STOPPED))
1889 sig->group_exit_code = signr;
1890 else
1891 WARN_ON_ONCE(!task_ptrace(current));
1893 current->group_stop &= ~GROUP_STOP_SIGMASK;
1894 current->group_stop |= signr | gstop;
1895 sig->group_stop_count = 1;
1896 for (t = next_thread(current); t != current;
1897 t = next_thread(t)) {
1898 t->group_stop &= ~GROUP_STOP_SIGMASK;
1900 * Setting state to TASK_STOPPED for a group
1901 * stop is always done with the siglock held,
1902 * so this check has no races.
1904 if (!(t->flags & PF_EXITING) && !task_is_stopped(t)) {
1905 t->group_stop |= signr | gstop;
1906 sig->group_stop_count++;
1907 signal_wake_up(t, 0);
1911 retry:
1912 if (likely(!task_ptrace(current))) {
1913 int notify = 0;
1916 * If there are no other threads in the group, or if there
1917 * is a group stop in progress and we are the last to stop,
1918 * report to the parent.
1920 if (task_participate_group_stop(current))
1921 notify = CLD_STOPPED;
1923 __set_current_state(TASK_STOPPED);
1924 spin_unlock_irq(&current->sighand->siglock);
1927 * Notify the parent of the group stop completion. Because
1928 * we're not holding either the siglock or tasklist_lock
1929 * here, ptracer may attach inbetween; however, this is for
1930 * group stop and should always be delivered to the real
1931 * parent of the group leader. The new ptracer will get
1932 * its notification when this task transitions into
1933 * TASK_TRACED.
1935 if (notify) {
1936 read_lock(&tasklist_lock);
1937 do_notify_parent_cldstop(current, false, notify);
1938 read_unlock(&tasklist_lock);
1941 /* Now we don't run again until woken by SIGCONT or SIGKILL */
1942 schedule();
1944 spin_lock_irq(&current->sighand->siglock);
1945 } else {
1946 ptrace_stop(current->group_stop & GROUP_STOP_SIGMASK,
1947 CLD_STOPPED, 0, NULL);
1948 current->exit_code = 0;
1952 * GROUP_STOP_PENDING could be set if another group stop has
1953 * started since being woken up or ptrace wants us to transit
1954 * between TASK_STOPPED and TRACED. Retry group stop.
1956 if (current->group_stop & GROUP_STOP_PENDING) {
1957 WARN_ON_ONCE(!(current->group_stop & GROUP_STOP_SIGMASK));
1958 goto retry;
1961 /* PTRACE_ATTACH might have raced with task killing, clear trapping */
1962 task_clear_group_stop_trapping(current);
1964 spin_unlock_irq(&current->sighand->siglock);
1966 tracehook_finish_jctl();
1968 return 1;
1971 static int ptrace_signal(int signr, siginfo_t *info,
1972 struct pt_regs *regs, void *cookie)
1974 if (!task_ptrace(current))
1975 return signr;
1977 ptrace_signal_deliver(regs, cookie);
1979 /* Let the debugger run. */
1980 ptrace_stop(signr, CLD_TRAPPED, 0, info);
1982 /* We're back. Did the debugger cancel the sig? */
1983 signr = current->exit_code;
1984 if (signr == 0)
1985 return signr;
1987 current->exit_code = 0;
1990 * Update the siginfo structure if the signal has
1991 * changed. If the debugger wanted something
1992 * specific in the siginfo structure then it should
1993 * have updated *info via PTRACE_SETSIGINFO.
1995 if (signr != info->si_signo) {
1996 info->si_signo = signr;
1997 info->si_errno = 0;
1998 info->si_code = SI_USER;
1999 info->si_pid = task_pid_vnr(current->parent);
2000 info->si_uid = task_uid(current->parent);
2003 /* If the (new) signal is now blocked, requeue it. */
2004 if (sigismember(&current->blocked, signr)) {
2005 specific_send_sig_info(signr, info, current);
2006 signr = 0;
2009 return signr;
2012 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
2013 struct pt_regs *regs, void *cookie)
2015 struct sighand_struct *sighand = current->sighand;
2016 struct signal_struct *signal = current->signal;
2017 int signr;
2019 relock:
2021 * We'll jump back here after any time we were stopped in TASK_STOPPED.
2022 * While in TASK_STOPPED, we were considered "frozen enough".
2023 * Now that we woke up, it's crucial if we're supposed to be
2024 * frozen that we freeze now before running anything substantial.
2026 try_to_freeze();
2028 spin_lock_irq(&sighand->siglock);
2030 * Every stopped thread goes here after wakeup. Check to see if
2031 * we should notify the parent, prepare_signal(SIGCONT) encodes
2032 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2034 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2035 struct task_struct *leader;
2036 int why;
2038 if (signal->flags & SIGNAL_CLD_CONTINUED)
2039 why = CLD_CONTINUED;
2040 else
2041 why = CLD_STOPPED;
2043 signal->flags &= ~SIGNAL_CLD_MASK;
2045 spin_unlock_irq(&sighand->siglock);
2048 * Notify the parent that we're continuing. This event is
2049 * always per-process and doesn't make whole lot of sense
2050 * for ptracers, who shouldn't consume the state via
2051 * wait(2) either, but, for backward compatibility, notify
2052 * the ptracer of the group leader too unless it's gonna be
2053 * a duplicate.
2055 read_lock(&tasklist_lock);
2057 do_notify_parent_cldstop(current, false, why);
2059 leader = current->group_leader;
2060 if (task_ptrace(leader) && !real_parent_is_ptracer(leader))
2061 do_notify_parent_cldstop(leader, true, why);
2063 read_unlock(&tasklist_lock);
2065 goto relock;
2068 for (;;) {
2069 struct k_sigaction *ka;
2071 * Tracing can induce an artificial signal and choose sigaction.
2072 * The return value in @signr determines the default action,
2073 * but @info->si_signo is the signal number we will report.
2075 signr = tracehook_get_signal(current, regs, info, return_ka);
2076 if (unlikely(signr < 0))
2077 goto relock;
2078 if (unlikely(signr != 0))
2079 ka = return_ka;
2080 else {
2081 if (unlikely(current->group_stop &
2082 GROUP_STOP_PENDING) && do_signal_stop(0))
2083 goto relock;
2085 signr = dequeue_signal(current, &current->blocked,
2086 info);
2088 if (!signr)
2089 break; /* will return 0 */
2091 if (signr != SIGKILL) {
2092 signr = ptrace_signal(signr, info,
2093 regs, cookie);
2094 if (!signr)
2095 continue;
2098 ka = &sighand->action[signr-1];
2101 /* Trace actually delivered signals. */
2102 trace_signal_deliver(signr, info, ka);
2104 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2105 continue;
2106 if (ka->sa.sa_handler != SIG_DFL) {
2107 /* Run the handler. */
2108 *return_ka = *ka;
2110 if (ka->sa.sa_flags & SA_ONESHOT)
2111 ka->sa.sa_handler = SIG_DFL;
2113 break; /* will return non-zero "signr" value */
2117 * Now we are doing the default action for this signal.
2119 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2120 continue;
2123 * Global init gets no signals it doesn't want.
2124 * Container-init gets no signals it doesn't want from same
2125 * container.
2127 * Note that if global/container-init sees a sig_kernel_only()
2128 * signal here, the signal must have been generated internally
2129 * or must have come from an ancestor namespace. In either
2130 * case, the signal cannot be dropped.
2132 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2133 !sig_kernel_only(signr))
2134 continue;
2136 if (sig_kernel_stop(signr)) {
2138 * The default action is to stop all threads in
2139 * the thread group. The job control signals
2140 * do nothing in an orphaned pgrp, but SIGSTOP
2141 * always works. Note that siglock needs to be
2142 * dropped during the call to is_orphaned_pgrp()
2143 * because of lock ordering with tasklist_lock.
2144 * This allows an intervening SIGCONT to be posted.
2145 * We need to check for that and bail out if necessary.
2147 if (signr != SIGSTOP) {
2148 spin_unlock_irq(&sighand->siglock);
2150 /* signals can be posted during this window */
2152 if (is_current_pgrp_orphaned())
2153 goto relock;
2155 spin_lock_irq(&sighand->siglock);
2158 if (likely(do_signal_stop(info->si_signo))) {
2159 /* It released the siglock. */
2160 goto relock;
2164 * We didn't actually stop, due to a race
2165 * with SIGCONT or something like that.
2167 continue;
2170 spin_unlock_irq(&sighand->siglock);
2173 * Anything else is fatal, maybe with a core dump.
2175 current->flags |= PF_SIGNALED;
2177 if (sig_kernel_coredump(signr)) {
2178 if (print_fatal_signals)
2179 print_fatal_signal(regs, info->si_signo);
2181 * If it was able to dump core, this kills all
2182 * other threads in the group and synchronizes with
2183 * their demise. If we lost the race with another
2184 * thread getting here, it set group_exit_code
2185 * first and our do_group_exit call below will use
2186 * that value and ignore the one we pass it.
2188 do_coredump(info->si_signo, info->si_signo, regs);
2192 * Death signals, no core dump.
2194 do_group_exit(info->si_signo);
2195 /* NOTREACHED */
2197 spin_unlock_irq(&sighand->siglock);
2198 return signr;
2202 * It could be that complete_signal() picked us to notify about the
2203 * group-wide signal. Other threads should be notified now to take
2204 * the shared signals in @which since we will not.
2206 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2208 sigset_t retarget;
2209 struct task_struct *t;
2211 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2212 if (sigisemptyset(&retarget))
2213 return;
2215 t = tsk;
2216 while_each_thread(tsk, t) {
2217 if (t->flags & PF_EXITING)
2218 continue;
2220 if (!has_pending_signals(&retarget, &t->blocked))
2221 continue;
2222 /* Remove the signals this thread can handle. */
2223 sigandsets(&retarget, &retarget, &t->blocked);
2225 if (!signal_pending(t))
2226 signal_wake_up(t, 0);
2228 if (sigisemptyset(&retarget))
2229 break;
2233 void exit_signals(struct task_struct *tsk)
2235 int group_stop = 0;
2236 sigset_t unblocked;
2238 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2239 tsk->flags |= PF_EXITING;
2240 return;
2243 spin_lock_irq(&tsk->sighand->siglock);
2245 * From now this task is not visible for group-wide signals,
2246 * see wants_signal(), do_signal_stop().
2248 tsk->flags |= PF_EXITING;
2249 if (!signal_pending(tsk))
2250 goto out;
2252 unblocked = tsk->blocked;
2253 signotset(&unblocked);
2254 retarget_shared_pending(tsk, &unblocked);
2256 if (unlikely(tsk->group_stop & GROUP_STOP_PENDING) &&
2257 task_participate_group_stop(tsk))
2258 group_stop = CLD_STOPPED;
2259 out:
2260 spin_unlock_irq(&tsk->sighand->siglock);
2263 * If group stop has completed, deliver the notification. This
2264 * should always go to the real parent of the group leader.
2266 if (unlikely(group_stop)) {
2267 read_lock(&tasklist_lock);
2268 do_notify_parent_cldstop(tsk, false, group_stop);
2269 read_unlock(&tasklist_lock);
2273 EXPORT_SYMBOL(recalc_sigpending);
2274 EXPORT_SYMBOL_GPL(dequeue_signal);
2275 EXPORT_SYMBOL(flush_signals);
2276 EXPORT_SYMBOL(force_sig);
2277 EXPORT_SYMBOL(send_sig);
2278 EXPORT_SYMBOL(send_sig_info);
2279 EXPORT_SYMBOL(sigprocmask);
2280 EXPORT_SYMBOL(block_all_signals);
2281 EXPORT_SYMBOL(unblock_all_signals);
2285 * System call entry points.
2289 * sys_restart_syscall - restart a system call
2291 SYSCALL_DEFINE0(restart_syscall)
2293 struct restart_block *restart = &current_thread_info()->restart_block;
2294 return restart->fn(restart);
2297 long do_no_restart_syscall(struct restart_block *param)
2299 return -EINTR;
2302 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2304 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2305 sigset_t newblocked;
2306 /* A set of now blocked but previously unblocked signals. */
2307 sigandnsets(&newblocked, newset, &current->blocked);
2308 retarget_shared_pending(tsk, &newblocked);
2310 tsk->blocked = *newset;
2311 recalc_sigpending();
2315 * set_current_blocked - change current->blocked mask
2316 * @newset: new mask
2318 * It is wrong to change ->blocked directly, this helper should be used
2319 * to ensure the process can't miss a shared signal we are going to block.
2321 void set_current_blocked(const sigset_t *newset)
2323 struct task_struct *tsk = current;
2325 spin_lock_irq(&tsk->sighand->siglock);
2326 __set_task_blocked(tsk, newset);
2327 spin_unlock_irq(&tsk->sighand->siglock);
2331 * This is also useful for kernel threads that want to temporarily
2332 * (or permanently) block certain signals.
2334 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2335 * interface happily blocks "unblockable" signals like SIGKILL
2336 * and friends.
2338 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2340 struct task_struct *tsk = current;
2341 sigset_t newset;
2343 /* Lockless, only current can change ->blocked, never from irq */
2344 if (oldset)
2345 *oldset = tsk->blocked;
2347 switch (how) {
2348 case SIG_BLOCK:
2349 sigorsets(&newset, &tsk->blocked, set);
2350 break;
2351 case SIG_UNBLOCK:
2352 sigandnsets(&newset, &tsk->blocked, set);
2353 break;
2354 case SIG_SETMASK:
2355 newset = *set;
2356 break;
2357 default:
2358 return -EINVAL;
2361 set_current_blocked(&newset);
2362 return 0;
2366 * sys_rt_sigprocmask - change the list of currently blocked signals
2367 * @how: whether to add, remove, or set signals
2368 * @set: stores pending signals
2369 * @oset: previous value of signal mask if non-null
2370 * @sigsetsize: size of sigset_t type
2372 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
2373 sigset_t __user *, oset, size_t, sigsetsize)
2375 sigset_t old_set, new_set;
2376 int error;
2378 /* XXX: Don't preclude handling different sized sigset_t's. */
2379 if (sigsetsize != sizeof(sigset_t))
2380 return -EINVAL;
2382 old_set = current->blocked;
2384 if (nset) {
2385 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
2386 return -EFAULT;
2387 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2389 error = sigprocmask(how, &new_set, NULL);
2390 if (error)
2391 return error;
2394 if (oset) {
2395 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
2396 return -EFAULT;
2399 return 0;
2402 long do_sigpending(void __user *set, unsigned long sigsetsize)
2404 long error = -EINVAL;
2405 sigset_t pending;
2407 if (sigsetsize > sizeof(sigset_t))
2408 goto out;
2410 spin_lock_irq(&current->sighand->siglock);
2411 sigorsets(&pending, &current->pending.signal,
2412 &current->signal->shared_pending.signal);
2413 spin_unlock_irq(&current->sighand->siglock);
2415 /* Outside the lock because only this thread touches it. */
2416 sigandsets(&pending, &current->blocked, &pending);
2418 error = -EFAULT;
2419 if (!copy_to_user(set, &pending, sigsetsize))
2420 error = 0;
2422 out:
2423 return error;
2427 * sys_rt_sigpending - examine a pending signal that has been raised
2428 * while blocked
2429 * @set: stores pending signals
2430 * @sigsetsize: size of sigset_t type or larger
2432 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2434 return do_sigpending(set, sigsetsize);
2437 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2439 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2441 int err;
2443 if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2444 return -EFAULT;
2445 if (from->si_code < 0)
2446 return __copy_to_user(to, from, sizeof(siginfo_t))
2447 ? -EFAULT : 0;
2449 * If you change siginfo_t structure, please be sure
2450 * this code is fixed accordingly.
2451 * Please remember to update the signalfd_copyinfo() function
2452 * inside fs/signalfd.c too, in case siginfo_t changes.
2453 * It should never copy any pad contained in the structure
2454 * to avoid security leaks, but must copy the generic
2455 * 3 ints plus the relevant union member.
2457 err = __put_user(from->si_signo, &to->si_signo);
2458 err |= __put_user(from->si_errno, &to->si_errno);
2459 err |= __put_user((short)from->si_code, &to->si_code);
2460 switch (from->si_code & __SI_MASK) {
2461 case __SI_KILL:
2462 err |= __put_user(from->si_pid, &to->si_pid);
2463 err |= __put_user(from->si_uid, &to->si_uid);
2464 break;
2465 case __SI_TIMER:
2466 err |= __put_user(from->si_tid, &to->si_tid);
2467 err |= __put_user(from->si_overrun, &to->si_overrun);
2468 err |= __put_user(from->si_ptr, &to->si_ptr);
2469 break;
2470 case __SI_POLL:
2471 err |= __put_user(from->si_band, &to->si_band);
2472 err |= __put_user(from->si_fd, &to->si_fd);
2473 break;
2474 case __SI_FAULT:
2475 err |= __put_user(from->si_addr, &to->si_addr);
2476 #ifdef __ARCH_SI_TRAPNO
2477 err |= __put_user(from->si_trapno, &to->si_trapno);
2478 #endif
2479 #ifdef BUS_MCEERR_AO
2481 * Other callers might not initialize the si_lsb field,
2482 * so check explicitly for the right codes here.
2484 if (from->si_code == BUS_MCEERR_AR || from->si_code == BUS_MCEERR_AO)
2485 err |= __put_user(from->si_addr_lsb, &to->si_addr_lsb);
2486 #endif
2487 break;
2488 case __SI_CHLD:
2489 err |= __put_user(from->si_pid, &to->si_pid);
2490 err |= __put_user(from->si_uid, &to->si_uid);
2491 err |= __put_user(from->si_status, &to->si_status);
2492 err |= __put_user(from->si_utime, &to->si_utime);
2493 err |= __put_user(from->si_stime, &to->si_stime);
2494 break;
2495 case __SI_RT: /* This is not generated by the kernel as of now. */
2496 case __SI_MESGQ: /* But this is */
2497 err |= __put_user(from->si_pid, &to->si_pid);
2498 err |= __put_user(from->si_uid, &to->si_uid);
2499 err |= __put_user(from->si_ptr, &to->si_ptr);
2500 break;
2501 default: /* this is just in case for now ... */
2502 err |= __put_user(from->si_pid, &to->si_pid);
2503 err |= __put_user(from->si_uid, &to->si_uid);
2504 break;
2506 return err;
2509 #endif
2512 * do_sigtimedwait - wait for queued signals specified in @which
2513 * @which: queued signals to wait for
2514 * @info: if non-null, the signal's siginfo is returned here
2515 * @ts: upper bound on process time suspension
2517 int do_sigtimedwait(const sigset_t *which, siginfo_t *info,
2518 const struct timespec *ts)
2520 struct task_struct *tsk = current;
2521 long timeout = MAX_SCHEDULE_TIMEOUT;
2522 sigset_t mask = *which;
2523 int sig;
2525 if (ts) {
2526 if (!timespec_valid(ts))
2527 return -EINVAL;
2528 timeout = timespec_to_jiffies(ts);
2530 * We can be close to the next tick, add another one
2531 * to ensure we will wait at least the time asked for.
2533 if (ts->tv_sec || ts->tv_nsec)
2534 timeout++;
2538 * Invert the set of allowed signals to get those we want to block.
2540 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
2541 signotset(&mask);
2543 spin_lock_irq(&tsk->sighand->siglock);
2544 sig = dequeue_signal(tsk, &mask, info);
2545 if (!sig && timeout) {
2547 * None ready, temporarily unblock those we're interested
2548 * while we are sleeping in so that we'll be awakened when
2549 * they arrive. Unblocking is always fine, we can avoid
2550 * set_current_blocked().
2552 tsk->real_blocked = tsk->blocked;
2553 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
2554 recalc_sigpending();
2555 spin_unlock_irq(&tsk->sighand->siglock);
2557 timeout = schedule_timeout_interruptible(timeout);
2559 spin_lock_irq(&tsk->sighand->siglock);
2560 __set_task_blocked(tsk, &tsk->real_blocked);
2561 siginitset(&tsk->real_blocked, 0);
2562 sig = dequeue_signal(tsk, &mask, info);
2564 spin_unlock_irq(&tsk->sighand->siglock);
2566 if (sig)
2567 return sig;
2568 return timeout ? -EINTR : -EAGAIN;
2572 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
2573 * in @uthese
2574 * @uthese: queued signals to wait for
2575 * @uinfo: if non-null, the signal's siginfo is returned here
2576 * @uts: upper bound on process time suspension
2577 * @sigsetsize: size of sigset_t type
2579 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2580 siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2581 size_t, sigsetsize)
2583 sigset_t these;
2584 struct timespec ts;
2585 siginfo_t info;
2586 int ret;
2588 /* XXX: Don't preclude handling different sized sigset_t's. */
2589 if (sigsetsize != sizeof(sigset_t))
2590 return -EINVAL;
2592 if (copy_from_user(&these, uthese, sizeof(these)))
2593 return -EFAULT;
2595 if (uts) {
2596 if (copy_from_user(&ts, uts, sizeof(ts)))
2597 return -EFAULT;
2600 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
2602 if (ret > 0 && uinfo) {
2603 if (copy_siginfo_to_user(uinfo, &info))
2604 ret = -EFAULT;
2607 return ret;
2611 * sys_kill - send a signal to a process
2612 * @pid: the PID of the process
2613 * @sig: signal to be sent
2615 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2617 struct siginfo info;
2619 info.si_signo = sig;
2620 info.si_errno = 0;
2621 info.si_code = SI_USER;
2622 info.si_pid = task_tgid_vnr(current);
2623 info.si_uid = current_uid();
2625 return kill_something_info(sig, &info, pid);
2628 static int
2629 do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
2631 struct task_struct *p;
2632 int error = -ESRCH;
2634 rcu_read_lock();
2635 p = find_task_by_vpid(pid);
2636 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2637 error = check_kill_permission(sig, info, p);
2639 * The null signal is a permissions and process existence
2640 * probe. No signal is actually delivered.
2642 if (!error && sig) {
2643 error = do_send_sig_info(sig, info, p, false);
2645 * If lock_task_sighand() failed we pretend the task
2646 * dies after receiving the signal. The window is tiny,
2647 * and the signal is private anyway.
2649 if (unlikely(error == -ESRCH))
2650 error = 0;
2653 rcu_read_unlock();
2655 return error;
2658 static int do_tkill(pid_t tgid, pid_t pid, int sig)
2660 struct siginfo info;
2662 info.si_signo = sig;
2663 info.si_errno = 0;
2664 info.si_code = SI_TKILL;
2665 info.si_pid = task_tgid_vnr(current);
2666 info.si_uid = current_uid();
2668 return do_send_specific(tgid, pid, sig, &info);
2672 * sys_tgkill - send signal to one specific thread
2673 * @tgid: the thread group ID of the thread
2674 * @pid: the PID of the thread
2675 * @sig: signal to be sent
2677 * This syscall also checks the @tgid and returns -ESRCH even if the PID
2678 * exists but it's not belonging to the target process anymore. This
2679 * method solves the problem of threads exiting and PIDs getting reused.
2681 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2683 /* This is only valid for single tasks */
2684 if (pid <= 0 || tgid <= 0)
2685 return -EINVAL;
2687 return do_tkill(tgid, pid, sig);
2691 * sys_tkill - send signal to one specific task
2692 * @pid: the PID of the task
2693 * @sig: signal to be sent
2695 * Send a signal to only one task, even if it's a CLONE_THREAD task.
2697 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2699 /* This is only valid for single tasks */
2700 if (pid <= 0)
2701 return -EINVAL;
2703 return do_tkill(0, pid, sig);
2707 * sys_rt_sigqueueinfo - send signal information to a signal
2708 * @pid: the PID of the thread
2709 * @sig: signal to be sent
2710 * @uinfo: signal info to be sent
2712 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2713 siginfo_t __user *, uinfo)
2715 siginfo_t info;
2717 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2718 return -EFAULT;
2720 /* Not even root can pretend to send signals from the kernel.
2721 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2723 if (info.si_code >= 0 || info.si_code == SI_TKILL) {
2724 /* We used to allow any < 0 si_code */
2725 WARN_ON_ONCE(info.si_code < 0);
2726 return -EPERM;
2728 info.si_signo = sig;
2730 /* POSIX.1b doesn't mention process groups. */
2731 return kill_proc_info(sig, &info, pid);
2734 long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
2736 /* This is only valid for single tasks */
2737 if (pid <= 0 || tgid <= 0)
2738 return -EINVAL;
2740 /* Not even root can pretend to send signals from the kernel.
2741 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2743 if (info->si_code >= 0 || info->si_code == SI_TKILL) {
2744 /* We used to allow any < 0 si_code */
2745 WARN_ON_ONCE(info->si_code < 0);
2746 return -EPERM;
2748 info->si_signo = sig;
2750 return do_send_specific(tgid, pid, sig, info);
2753 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
2754 siginfo_t __user *, uinfo)
2756 siginfo_t info;
2758 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2759 return -EFAULT;
2761 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
2764 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2766 struct task_struct *t = current;
2767 struct k_sigaction *k;
2768 sigset_t mask;
2770 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2771 return -EINVAL;
2773 k = &t->sighand->action[sig-1];
2775 spin_lock_irq(&current->sighand->siglock);
2776 if (oact)
2777 *oact = *k;
2779 if (act) {
2780 sigdelsetmask(&act->sa.sa_mask,
2781 sigmask(SIGKILL) | sigmask(SIGSTOP));
2782 *k = *act;
2784 * POSIX 3.3.1.3:
2785 * "Setting a signal action to SIG_IGN for a signal that is
2786 * pending shall cause the pending signal to be discarded,
2787 * whether or not it is blocked."
2789 * "Setting a signal action to SIG_DFL for a signal that is
2790 * pending and whose default action is to ignore the signal
2791 * (for example, SIGCHLD), shall cause the pending signal to
2792 * be discarded, whether or not it is blocked"
2794 if (sig_handler_ignored(sig_handler(t, sig), sig)) {
2795 sigemptyset(&mask);
2796 sigaddset(&mask, sig);
2797 rm_from_queue_full(&mask, &t->signal->shared_pending);
2798 do {
2799 rm_from_queue_full(&mask, &t->pending);
2800 t = next_thread(t);
2801 } while (t != current);
2805 spin_unlock_irq(&current->sighand->siglock);
2806 return 0;
2809 int
2810 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2812 stack_t oss;
2813 int error;
2815 oss.ss_sp = (void __user *) current->sas_ss_sp;
2816 oss.ss_size = current->sas_ss_size;
2817 oss.ss_flags = sas_ss_flags(sp);
2819 if (uss) {
2820 void __user *ss_sp;
2821 size_t ss_size;
2822 int ss_flags;
2824 error = -EFAULT;
2825 if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
2826 goto out;
2827 error = __get_user(ss_sp, &uss->ss_sp) |
2828 __get_user(ss_flags, &uss->ss_flags) |
2829 __get_user(ss_size, &uss->ss_size);
2830 if (error)
2831 goto out;
2833 error = -EPERM;
2834 if (on_sig_stack(sp))
2835 goto out;
2837 error = -EINVAL;
2839 * Note - this code used to test ss_flags incorrectly:
2840 * old code may have been written using ss_flags==0
2841 * to mean ss_flags==SS_ONSTACK (as this was the only
2842 * way that worked) - this fix preserves that older
2843 * mechanism.
2845 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2846 goto out;
2848 if (ss_flags == SS_DISABLE) {
2849 ss_size = 0;
2850 ss_sp = NULL;
2851 } else {
2852 error = -ENOMEM;
2853 if (ss_size < MINSIGSTKSZ)
2854 goto out;
2857 current->sas_ss_sp = (unsigned long) ss_sp;
2858 current->sas_ss_size = ss_size;
2861 error = 0;
2862 if (uoss) {
2863 error = -EFAULT;
2864 if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
2865 goto out;
2866 error = __put_user(oss.ss_sp, &uoss->ss_sp) |
2867 __put_user(oss.ss_size, &uoss->ss_size) |
2868 __put_user(oss.ss_flags, &uoss->ss_flags);
2871 out:
2872 return error;
2875 #ifdef __ARCH_WANT_SYS_SIGPENDING
2878 * sys_sigpending - examine pending signals
2879 * @set: where mask of pending signal is returned
2881 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
2883 return do_sigpending(set, sizeof(*set));
2886 #endif
2888 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2890 * sys_sigprocmask - examine and change blocked signals
2891 * @how: whether to add, remove, or set signals
2892 * @nset: signals to add or remove (if non-null)
2893 * @oset: previous value of signal mask if non-null
2895 * Some platforms have their own version with special arguments;
2896 * others support only sys_rt_sigprocmask.
2899 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
2900 old_sigset_t __user *, oset)
2902 old_sigset_t old_set, new_set;
2903 sigset_t new_blocked;
2905 old_set = current->blocked.sig[0];
2907 if (nset) {
2908 if (copy_from_user(&new_set, nset, sizeof(*nset)))
2909 return -EFAULT;
2910 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2912 new_blocked = current->blocked;
2914 switch (how) {
2915 case SIG_BLOCK:
2916 sigaddsetmask(&new_blocked, new_set);
2917 break;
2918 case SIG_UNBLOCK:
2919 sigdelsetmask(&new_blocked, new_set);
2920 break;
2921 case SIG_SETMASK:
2922 new_blocked.sig[0] = new_set;
2923 break;
2924 default:
2925 return -EINVAL;
2928 set_current_blocked(&new_blocked);
2931 if (oset) {
2932 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2933 return -EFAULT;
2936 return 0;
2938 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2940 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2942 * sys_rt_sigaction - alter an action taken by a process
2943 * @sig: signal to be sent
2944 * @act: new sigaction
2945 * @oact: used to save the previous sigaction
2946 * @sigsetsize: size of sigset_t type
2948 SYSCALL_DEFINE4(rt_sigaction, int, sig,
2949 const struct sigaction __user *, act,
2950 struct sigaction __user *, oact,
2951 size_t, sigsetsize)
2953 struct k_sigaction new_sa, old_sa;
2954 int ret = -EINVAL;
2956 /* XXX: Don't preclude handling different sized sigset_t's. */
2957 if (sigsetsize != sizeof(sigset_t))
2958 goto out;
2960 if (act) {
2961 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2962 return -EFAULT;
2965 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2967 if (!ret && oact) {
2968 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2969 return -EFAULT;
2971 out:
2972 return ret;
2974 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2976 #ifdef __ARCH_WANT_SYS_SGETMASK
2979 * For backwards compatibility. Functionality superseded by sigprocmask.
2981 SYSCALL_DEFINE0(sgetmask)
2983 /* SMP safe */
2984 return current->blocked.sig[0];
2987 SYSCALL_DEFINE1(ssetmask, int, newmask)
2989 int old;
2991 spin_lock_irq(&current->sighand->siglock);
2992 old = current->blocked.sig[0];
2994 siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2995 sigmask(SIGSTOP)));
2996 recalc_sigpending();
2997 spin_unlock_irq(&current->sighand->siglock);
2999 return old;
3001 #endif /* __ARCH_WANT_SGETMASK */
3003 #ifdef __ARCH_WANT_SYS_SIGNAL
3005 * For backwards compatibility. Functionality superseded by sigaction.
3007 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
3009 struct k_sigaction new_sa, old_sa;
3010 int ret;
3012 new_sa.sa.sa_handler = handler;
3013 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
3014 sigemptyset(&new_sa.sa.sa_mask);
3016 ret = do_sigaction(sig, &new_sa, &old_sa);
3018 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
3020 #endif /* __ARCH_WANT_SYS_SIGNAL */
3022 #ifdef __ARCH_WANT_SYS_PAUSE
3024 SYSCALL_DEFINE0(pause)
3026 while (!signal_pending(current)) {
3027 current->state = TASK_INTERRUPTIBLE;
3028 schedule();
3030 return -ERESTARTNOHAND;
3033 #endif
3035 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
3037 * sys_rt_sigsuspend - replace the signal mask for a value with the
3038 * @unewset value until a signal is received
3039 * @unewset: new signal mask value
3040 * @sigsetsize: size of sigset_t type
3042 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
3044 sigset_t newset;
3046 /* XXX: Don't preclude handling different sized sigset_t's. */
3047 if (sigsetsize != sizeof(sigset_t))
3048 return -EINVAL;
3050 if (copy_from_user(&newset, unewset, sizeof(newset)))
3051 return -EFAULT;
3052 sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3054 spin_lock_irq(&current->sighand->siglock);
3055 current->saved_sigmask = current->blocked;
3056 current->blocked = newset;
3057 recalc_sigpending();
3058 spin_unlock_irq(&current->sighand->siglock);
3060 current->state = TASK_INTERRUPTIBLE;
3061 schedule();
3062 set_restore_sigmask();
3063 return -ERESTARTNOHAND;
3065 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
3067 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
3069 return NULL;
3072 void __init signals_init(void)
3074 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
3077 #ifdef CONFIG_KGDB_KDB
3078 #include <linux/kdb.h>
3080 * kdb_send_sig_info - Allows kdb to send signals without exposing
3081 * signal internals. This function checks if the required locks are
3082 * available before calling the main signal code, to avoid kdb
3083 * deadlocks.
3085 void
3086 kdb_send_sig_info(struct task_struct *t, struct siginfo *info)
3088 static struct task_struct *kdb_prev_t;
3089 int sig, new_t;
3090 if (!spin_trylock(&t->sighand->siglock)) {
3091 kdb_printf("Can't do kill command now.\n"
3092 "The sigmask lock is held somewhere else in "
3093 "kernel, try again later\n");
3094 return;
3096 spin_unlock(&t->sighand->siglock);
3097 new_t = kdb_prev_t != t;
3098 kdb_prev_t = t;
3099 if (t->state != TASK_RUNNING && new_t) {
3100 kdb_printf("Process is not RUNNING, sending a signal from "
3101 "kdb risks deadlock\n"
3102 "on the run queue locks. "
3103 "The signal has _not_ been sent.\n"
3104 "Reissue the kill command if you want to risk "
3105 "the deadlock.\n");
3106 return;
3108 sig = info->si_signo;
3109 if (send_sig_info(sig, info, t))
3110 kdb_printf("Fail to deliver Signal %d to process %d.\n",
3111 sig, t->pid);
3112 else
3113 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
3115 #endif /* CONFIG_KGDB_KDB */