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/export.h>
15 #include <linux/init.h>
16 #include <linux/sched.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 #include <linux/user_namespace.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/signal.h>
35 #include <asm/param.h>
36 #include <asm/uaccess.h>
37 #include <asm/unistd.h>
38 #include <asm/siginfo.h>
39 #include <asm/cacheflush.h>
40 #include "audit.h" /* audit_signal_info() */
43 * SLAB caches for signal bits.
46 static struct kmem_cache *sigqueue_cachep;
48 int print_fatal_signals __read_mostly;
50 static void __user *sig_handler(struct task_struct *t, int sig)
52 return t->sighand->action[sig - 1].sa.sa_handler;
55 static int sig_handler_ignored(void __user *handler, int sig)
57 /* Is it explicitly or implicitly ignored? */
58 return handler == SIG_IGN ||
59 (handler == SIG_DFL && sig_kernel_ignore(sig));
62 static int sig_task_ignored(struct task_struct *t, int sig, bool force)
66 handler = sig_handler(t, sig);
68 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
69 handler == SIG_DFL && !force)
72 return sig_handler_ignored(handler, sig);
75 static int sig_ignored(struct task_struct *t, int sig, bool force)
78 * Blocked signals are never ignored, since the
79 * signal handler may change by the time it is
82 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
85 if (!sig_task_ignored(t, sig, force))
89 * Tracers may want to know about even ignored signals.
95 * Re-calculate pending state from the set of locally pending
96 * signals, globally pending signals, and blocked signals.
98 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
103 switch (_NSIG_WORDS) {
105 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
106 ready |= signal->sig[i] &~ blocked->sig[i];
109 case 4: ready = signal->sig[3] &~ blocked->sig[3];
110 ready |= signal->sig[2] &~ blocked->sig[2];
111 ready |= signal->sig[1] &~ blocked->sig[1];
112 ready |= signal->sig[0] &~ blocked->sig[0];
115 case 2: ready = signal->sig[1] &~ blocked->sig[1];
116 ready |= signal->sig[0] &~ blocked->sig[0];
119 case 1: ready = signal->sig[0] &~ blocked->sig[0];
124 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
126 static int recalc_sigpending_tsk(struct task_struct *t)
128 if ((t->jobctl & JOBCTL_PENDING_MASK) ||
129 PENDING(&t->pending, &t->blocked) ||
130 PENDING(&t->signal->shared_pending, &t->blocked)) {
131 set_tsk_thread_flag(t, TIF_SIGPENDING);
135 * We must never clear the flag in another thread, or in current
136 * when it's possible the current syscall is returning -ERESTART*.
137 * So we don't clear it here, and only callers who know they should do.
143 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
144 * This is superfluous when called on current, the wakeup is a harmless no-op.
146 void recalc_sigpending_and_wake(struct task_struct *t)
148 if (recalc_sigpending_tsk(t))
149 signal_wake_up(t, 0);
152 void recalc_sigpending(void)
154 if (!recalc_sigpending_tsk(current) && !freezing(current))
155 clear_thread_flag(TIF_SIGPENDING);
159 /* Given the mask, find the first available signal that should be serviced. */
161 #define SYNCHRONOUS_MASK \
162 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
163 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
165 int next_signal(struct sigpending *pending, sigset_t *mask)
167 unsigned long i, *s, *m, x;
170 s = pending->signal.sig;
174 * Handle the first word specially: it contains the
175 * synchronous signals that need to be dequeued first.
179 if (x & SYNCHRONOUS_MASK)
180 x &= SYNCHRONOUS_MASK;
185 switch (_NSIG_WORDS) {
187 for (i = 1; i < _NSIG_WORDS; ++i) {
191 sig = ffz(~x) + i*_NSIG_BPW + 1;
200 sig = ffz(~x) + _NSIG_BPW + 1;
211 static inline void print_dropped_signal(int sig)
213 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
215 if (!print_fatal_signals)
218 if (!__ratelimit(&ratelimit_state))
221 printk(KERN_INFO "%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
222 current->comm, current->pid, sig);
226 * task_set_jobctl_pending - set jobctl pending bits
228 * @mask: pending bits to set
230 * Clear @mask from @task->jobctl. @mask must be subset of
231 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
232 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
233 * cleared. If @task is already being killed or exiting, this function
237 * Must be called with @task->sighand->siglock held.
240 * %true if @mask is set, %false if made noop because @task was dying.
242 bool task_set_jobctl_pending(struct task_struct *task, unsigned int mask)
244 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
245 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
246 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
248 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
251 if (mask & JOBCTL_STOP_SIGMASK)
252 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
254 task->jobctl |= mask;
259 * task_clear_jobctl_trapping - clear jobctl trapping bit
262 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
263 * Clear it and wake up the ptracer. Note that we don't need any further
264 * locking. @task->siglock guarantees that @task->parent points to the
268 * Must be called with @task->sighand->siglock held.
270 void task_clear_jobctl_trapping(struct task_struct *task)
272 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
273 task->jobctl &= ~JOBCTL_TRAPPING;
274 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
279 * task_clear_jobctl_pending - clear jobctl pending bits
281 * @mask: pending bits to clear
283 * Clear @mask from @task->jobctl. @mask must be subset of
284 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
285 * STOP bits are cleared together.
287 * If clearing of @mask leaves no stop or trap pending, this function calls
288 * task_clear_jobctl_trapping().
291 * Must be called with @task->sighand->siglock held.
293 void task_clear_jobctl_pending(struct task_struct *task, unsigned int mask)
295 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
297 if (mask & JOBCTL_STOP_PENDING)
298 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
300 task->jobctl &= ~mask;
302 if (!(task->jobctl & JOBCTL_PENDING_MASK))
303 task_clear_jobctl_trapping(task);
307 * task_participate_group_stop - participate in a group stop
308 * @task: task participating in a group stop
310 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
311 * Group stop states are cleared and the group stop count is consumed if
312 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
313 * stop, the appropriate %SIGNAL_* flags are set.
316 * Must be called with @task->sighand->siglock held.
319 * %true if group stop completion should be notified to the parent, %false
322 static bool task_participate_group_stop(struct task_struct *task)
324 struct signal_struct *sig = task->signal;
325 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
327 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
329 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
334 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
335 sig->group_stop_count--;
338 * Tell the caller to notify completion iff we are entering into a
339 * fresh group stop. Read comment in do_signal_stop() for details.
341 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
342 sig->flags = SIGNAL_STOP_STOPPED;
349 * allocate a new signal queue record
350 * - this may be called without locks if and only if t == current, otherwise an
351 * appropriate lock must be held to stop the target task from exiting
353 static struct sigqueue *
354 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
356 struct sigqueue *q = NULL;
357 struct user_struct *user;
360 * Protect access to @t credentials. This can go away when all
361 * callers hold rcu read lock.
364 user = get_uid(__task_cred(t)->user);
365 atomic_inc(&user->sigpending);
368 if (override_rlimit ||
369 atomic_read(&user->sigpending) <=
370 task_rlimit(t, RLIMIT_SIGPENDING)) {
371 q = kmem_cache_alloc(sigqueue_cachep, flags);
373 print_dropped_signal(sig);
376 if (unlikely(q == NULL)) {
377 atomic_dec(&user->sigpending);
380 INIT_LIST_HEAD(&q->list);
388 static void __sigqueue_free(struct sigqueue *q)
390 if (q->flags & SIGQUEUE_PREALLOC)
392 atomic_dec(&q->user->sigpending);
394 kmem_cache_free(sigqueue_cachep, q);
397 void flush_sigqueue(struct sigpending *queue)
401 sigemptyset(&queue->signal);
402 while (!list_empty(&queue->list)) {
403 q = list_entry(queue->list.next, struct sigqueue , list);
404 list_del_init(&q->list);
410 * Flush all pending signals for a task.
412 void __flush_signals(struct task_struct *t)
414 clear_tsk_thread_flag(t, TIF_SIGPENDING);
415 flush_sigqueue(&t->pending);
416 flush_sigqueue(&t->signal->shared_pending);
419 void flush_signals(struct task_struct *t)
423 spin_lock_irqsave(&t->sighand->siglock, flags);
425 spin_unlock_irqrestore(&t->sighand->siglock, flags);
428 static void __flush_itimer_signals(struct sigpending *pending)
430 sigset_t signal, retain;
431 struct sigqueue *q, *n;
433 signal = pending->signal;
434 sigemptyset(&retain);
436 list_for_each_entry_safe(q, n, &pending->list, list) {
437 int sig = q->info.si_signo;
439 if (likely(q->info.si_code != SI_TIMER)) {
440 sigaddset(&retain, sig);
442 sigdelset(&signal, sig);
443 list_del_init(&q->list);
448 sigorsets(&pending->signal, &signal, &retain);
451 void flush_itimer_signals(void)
453 struct task_struct *tsk = current;
456 spin_lock_irqsave(&tsk->sighand->siglock, flags);
457 __flush_itimer_signals(&tsk->pending);
458 __flush_itimer_signals(&tsk->signal->shared_pending);
459 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
462 void ignore_signals(struct task_struct *t)
466 for (i = 0; i < _NSIG; ++i)
467 t->sighand->action[i].sa.sa_handler = SIG_IGN;
473 * Flush all handlers for a task.
477 flush_signal_handlers(struct task_struct *t, int force_default)
480 struct k_sigaction *ka = &t->sighand->action[0];
481 for (i = _NSIG ; i != 0 ; i--) {
482 if (force_default || ka->sa.sa_handler != SIG_IGN)
483 ka->sa.sa_handler = SIG_DFL;
485 sigemptyset(&ka->sa.sa_mask);
490 int unhandled_signal(struct task_struct *tsk, int sig)
492 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
493 if (is_global_init(tsk))
495 if (handler != SIG_IGN && handler != SIG_DFL)
497 /* if ptraced, let the tracer determine */
502 * Notify the system that a driver wants to block all signals for this
503 * process, and wants to be notified if any signals at all were to be
504 * sent/acted upon. If the notifier routine returns non-zero, then the
505 * signal will be acted upon after all. If the notifier routine returns 0,
506 * then then signal will be blocked. Only one block per process is
507 * allowed. priv is a pointer to private data that the notifier routine
508 * can use to determine if the signal should be blocked or not.
511 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
515 spin_lock_irqsave(¤t->sighand->siglock, flags);
516 current->notifier_mask = mask;
517 current->notifier_data = priv;
518 current->notifier = notifier;
519 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
522 /* Notify the system that blocking has ended. */
525 unblock_all_signals(void)
529 spin_lock_irqsave(¤t->sighand->siglock, flags);
530 current->notifier = NULL;
531 current->notifier_data = NULL;
533 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
536 static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
538 struct sigqueue *q, *first = NULL;
541 * Collect the siginfo appropriate to this signal. Check if
542 * there is another siginfo for the same signal.
544 list_for_each_entry(q, &list->list, list) {
545 if (q->info.si_signo == sig) {
552 sigdelset(&list->signal, sig);
556 list_del_init(&first->list);
557 copy_siginfo(info, &first->info);
558 __sigqueue_free(first);
561 * Ok, it wasn't in the queue. This must be
562 * a fast-pathed signal or we must have been
563 * out of queue space. So zero out the info.
565 info->si_signo = sig;
567 info->si_code = SI_USER;
573 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
576 int sig = next_signal(pending, mask);
579 if (current->notifier) {
580 if (sigismember(current->notifier_mask, sig)) {
581 if (!(current->notifier)(current->notifier_data)) {
582 clear_thread_flag(TIF_SIGPENDING);
588 collect_signal(sig, pending, info);
595 * Dequeue a signal and return the element to the caller, which is
596 * expected to free it.
598 * All callers have to hold the siglock.
600 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
604 /* We only dequeue private signals from ourselves, we don't let
605 * signalfd steal them
607 signr = __dequeue_signal(&tsk->pending, mask, info);
609 signr = __dequeue_signal(&tsk->signal->shared_pending,
614 * itimers are process shared and we restart periodic
615 * itimers in the signal delivery path to prevent DoS
616 * attacks in the high resolution timer case. This is
617 * compliant with the old way of self-restarting
618 * itimers, as the SIGALRM is a legacy signal and only
619 * queued once. Changing the restart behaviour to
620 * restart the timer in the signal dequeue path is
621 * reducing the timer noise on heavy loaded !highres
624 if (unlikely(signr == SIGALRM)) {
625 struct hrtimer *tmr = &tsk->signal->real_timer;
627 if (!hrtimer_is_queued(tmr) &&
628 tsk->signal->it_real_incr.tv64 != 0) {
629 hrtimer_forward(tmr, tmr->base->get_time(),
630 tsk->signal->it_real_incr);
631 hrtimer_restart(tmr);
640 if (unlikely(sig_kernel_stop(signr))) {
642 * Set a marker that we have dequeued a stop signal. Our
643 * caller might release the siglock and then the pending
644 * stop signal it is about to process is no longer in the
645 * pending bitmasks, but must still be cleared by a SIGCONT
646 * (and overruled by a SIGKILL). So those cases clear this
647 * shared flag after we've set it. Note that this flag may
648 * remain set after the signal we return is ignored or
649 * handled. That doesn't matter because its only purpose
650 * is to alert stop-signal processing code when another
651 * processor has come along and cleared the flag.
653 current->jobctl |= JOBCTL_STOP_DEQUEUED;
655 if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
657 * Release the siglock to ensure proper locking order
658 * of timer locks outside of siglocks. Note, we leave
659 * irqs disabled here, since the posix-timers code is
660 * about to disable them again anyway.
662 spin_unlock(&tsk->sighand->siglock);
663 do_schedule_next_timer(info);
664 spin_lock(&tsk->sighand->siglock);
670 * Tell a process that it has a new active signal..
672 * NOTE! we rely on the previous spin_lock to
673 * lock interrupts for us! We can only be called with
674 * "siglock" held, and the local interrupt must
675 * have been disabled when that got acquired!
677 * No need to set need_resched since signal event passing
678 * goes through ->blocked
680 void signal_wake_up(struct task_struct *t, int resume)
684 set_tsk_thread_flag(t, TIF_SIGPENDING);
687 * For SIGKILL, we want to wake it up in the stopped/traced/killable
688 * case. We don't check t->state here because there is a race with it
689 * executing another processor and just now entering stopped state.
690 * By using wake_up_state, we ensure the process will wake up and
691 * handle its death signal.
693 mask = TASK_INTERRUPTIBLE;
695 mask |= TASK_WAKEKILL;
696 if (!wake_up_state(t, mask))
701 * Remove signals in mask from the pending set and queue.
702 * Returns 1 if any signals were found.
704 * All callers must be holding the siglock.
706 * This version takes a sigset mask and looks at all signals,
707 * not just those in the first mask word.
709 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
711 struct sigqueue *q, *n;
714 sigandsets(&m, mask, &s->signal);
715 if (sigisemptyset(&m))
718 sigandnsets(&s->signal, &s->signal, mask);
719 list_for_each_entry_safe(q, n, &s->list, list) {
720 if (sigismember(mask, q->info.si_signo)) {
721 list_del_init(&q->list);
728 * Remove signals in mask from the pending set and queue.
729 * Returns 1 if any signals were found.
731 * All callers must be holding the siglock.
733 static int rm_from_queue(unsigned long mask, struct sigpending *s)
735 struct sigqueue *q, *n;
737 if (!sigtestsetmask(&s->signal, mask))
740 sigdelsetmask(&s->signal, mask);
741 list_for_each_entry_safe(q, n, &s->list, list) {
742 if (q->info.si_signo < SIGRTMIN &&
743 (mask & sigmask(q->info.si_signo))) {
744 list_del_init(&q->list);
751 static inline int is_si_special(const struct siginfo *info)
753 return info <= SEND_SIG_FORCED;
756 static inline bool si_fromuser(const struct siginfo *info)
758 return info == SEND_SIG_NOINFO ||
759 (!is_si_special(info) && SI_FROMUSER(info));
763 * called with RCU read lock from check_kill_permission()
765 static int kill_ok_by_cred(struct task_struct *t)
767 const struct cred *cred = current_cred();
768 const struct cred *tcred = __task_cred(t);
770 if (uid_eq(cred->euid, tcred->suid) ||
771 uid_eq(cred->euid, tcred->uid) ||
772 uid_eq(cred->uid, tcred->suid) ||
773 uid_eq(cred->uid, tcred->uid))
776 if (ns_capable(tcred->user_ns, CAP_KILL))
783 * Bad permissions for sending the signal
784 * - the caller must hold the RCU read lock
786 static int check_kill_permission(int sig, struct siginfo *info,
787 struct task_struct *t)
792 if (!valid_signal(sig))
795 if (!si_fromuser(info))
798 error = audit_signal_info(sig, t); /* Let audit system see the signal */
802 if (!same_thread_group(current, t) &&
803 !kill_ok_by_cred(t)) {
806 sid = task_session(t);
808 * We don't return the error if sid == NULL. The
809 * task was unhashed, the caller must notice this.
811 if (!sid || sid == task_session(current))
818 return security_task_kill(t, info, sig, 0);
822 * ptrace_trap_notify - schedule trap to notify ptracer
823 * @t: tracee wanting to notify tracer
825 * This function schedules sticky ptrace trap which is cleared on the next
826 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
829 * If @t is running, STOP trap will be taken. If trapped for STOP and
830 * ptracer is listening for events, tracee is woken up so that it can
831 * re-trap for the new event. If trapped otherwise, STOP trap will be
832 * eventually taken without returning to userland after the existing traps
833 * are finished by PTRACE_CONT.
836 * Must be called with @task->sighand->siglock held.
838 static void ptrace_trap_notify(struct task_struct *t)
840 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
841 assert_spin_locked(&t->sighand->siglock);
843 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
844 signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
848 * Handle magic process-wide effects of stop/continue signals. Unlike
849 * the signal actions, these happen immediately at signal-generation
850 * time regardless of blocking, ignoring, or handling. This does the
851 * actual continuing for SIGCONT, but not the actual stopping for stop
852 * signals. The process stop is done as a signal action for SIG_DFL.
854 * Returns true if the signal should be actually delivered, otherwise
855 * it should be dropped.
857 static int prepare_signal(int sig, struct task_struct *p, bool force)
859 struct signal_struct *signal = p->signal;
860 struct task_struct *t;
862 if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
864 * The process is in the middle of dying, nothing to do.
866 } else if (sig_kernel_stop(sig)) {
868 * This is a stop signal. Remove SIGCONT from all queues.
870 rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
873 rm_from_queue(sigmask(SIGCONT), &t->pending);
874 } while_each_thread(p, t);
875 } else if (sig == SIGCONT) {
878 * Remove all stop signals from all queues, wake all threads.
880 rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
883 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
884 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
885 if (likely(!(t->ptrace & PT_SEIZED)))
886 wake_up_state(t, __TASK_STOPPED);
888 ptrace_trap_notify(t);
889 } while_each_thread(p, t);
892 * Notify the parent with CLD_CONTINUED if we were stopped.
894 * If we were in the middle of a group stop, we pretend it
895 * was already finished, and then continued. Since SIGCHLD
896 * doesn't queue we report only CLD_STOPPED, as if the next
897 * CLD_CONTINUED was dropped.
900 if (signal->flags & SIGNAL_STOP_STOPPED)
901 why |= SIGNAL_CLD_CONTINUED;
902 else if (signal->group_stop_count)
903 why |= SIGNAL_CLD_STOPPED;
907 * The first thread which returns from do_signal_stop()
908 * will take ->siglock, notice SIGNAL_CLD_MASK, and
909 * notify its parent. See get_signal_to_deliver().
911 signal->flags = why | SIGNAL_STOP_CONTINUED;
912 signal->group_stop_count = 0;
913 signal->group_exit_code = 0;
917 return !sig_ignored(p, sig, force);
921 * Test if P wants to take SIG. After we've checked all threads with this,
922 * it's equivalent to finding no threads not blocking SIG. Any threads not
923 * blocking SIG were ruled out because they are not running and already
924 * have pending signals. Such threads will dequeue from the shared queue
925 * as soon as they're available, so putting the signal on the shared queue
926 * will be equivalent to sending it to one such thread.
928 static inline int wants_signal(int sig, struct task_struct *p)
930 if (sigismember(&p->blocked, sig))
932 if (p->flags & PF_EXITING)
936 if (task_is_stopped_or_traced(p))
938 return task_curr(p) || !signal_pending(p);
941 static void complete_signal(int sig, struct task_struct *p, int group)
943 struct signal_struct *signal = p->signal;
944 struct task_struct *t;
947 * Now find a thread we can wake up to take the signal off the queue.
949 * If the main thread wants the signal, it gets first crack.
950 * Probably the least surprising to the average bear.
952 if (wants_signal(sig, p))
954 else if (!group || thread_group_empty(p))
956 * There is just one thread and it does not need to be woken.
957 * It will dequeue unblocked signals before it runs again.
962 * Otherwise try to find a suitable thread.
964 t = signal->curr_target;
965 while (!wants_signal(sig, t)) {
967 if (t == signal->curr_target)
969 * No thread needs to be woken.
970 * Any eligible threads will see
971 * the signal in the queue soon.
975 signal->curr_target = t;
979 * Found a killable thread. If the signal will be fatal,
980 * then start taking the whole group down immediately.
982 if (sig_fatal(p, sig) &&
983 !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
984 !sigismember(&t->real_blocked, sig) &&
985 (sig == SIGKILL || !t->ptrace)) {
987 * This signal will be fatal to the whole group.
989 if (!sig_kernel_coredump(sig)) {
991 * Start a group exit and wake everybody up.
992 * This way we don't have other threads
993 * running and doing things after a slower
994 * thread has the fatal signal pending.
996 signal->flags = SIGNAL_GROUP_EXIT;
997 signal->group_exit_code = sig;
998 signal->group_stop_count = 0;
1001 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1002 sigaddset(&t->pending.signal, SIGKILL);
1003 signal_wake_up(t, 1);
1004 } while_each_thread(p, t);
1010 * The signal is already in the shared-pending queue.
1011 * Tell the chosen thread to wake up and dequeue it.
1013 signal_wake_up(t, sig == SIGKILL);
1017 static inline int legacy_queue(struct sigpending *signals, int sig)
1019 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1022 #ifdef CONFIG_USER_NS
1023 static inline void userns_fixup_signal_uid(struct siginfo *info, struct task_struct *t)
1025 if (current_user_ns() == task_cred_xxx(t, user_ns))
1028 if (SI_FROMKERNEL(info))
1032 info->si_uid = from_kuid_munged(task_cred_xxx(t, user_ns),
1033 make_kuid(current_user_ns(), info->si_uid));
1037 static inline void userns_fixup_signal_uid(struct siginfo *info, struct task_struct *t)
1043 static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
1044 int group, int from_ancestor_ns)
1046 struct sigpending *pending;
1048 int override_rlimit;
1049 int ret = 0, result;
1051 assert_spin_locked(&t->sighand->siglock);
1053 result = TRACE_SIGNAL_IGNORED;
1054 if (!prepare_signal(sig, t,
1055 from_ancestor_ns || (info == SEND_SIG_FORCED)))
1058 pending = group ? &t->signal->shared_pending : &t->pending;
1060 * Short-circuit ignored signals and support queuing
1061 * exactly one non-rt signal, so that we can get more
1062 * detailed information about the cause of the signal.
1064 result = TRACE_SIGNAL_ALREADY_PENDING;
1065 if (legacy_queue(pending, sig))
1068 result = TRACE_SIGNAL_DELIVERED;
1070 * fast-pathed signals for kernel-internal things like SIGSTOP
1073 if (info == SEND_SIG_FORCED)
1077 * Real-time signals must be queued if sent by sigqueue, or
1078 * some other real-time mechanism. It is implementation
1079 * defined whether kill() does so. We attempt to do so, on
1080 * the principle of least surprise, but since kill is not
1081 * allowed to fail with EAGAIN when low on memory we just
1082 * make sure at least one signal gets delivered and don't
1083 * pass on the info struct.
1086 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1088 override_rlimit = 0;
1090 q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
1093 list_add_tail(&q->list, &pending->list);
1094 switch ((unsigned long) info) {
1095 case (unsigned long) SEND_SIG_NOINFO:
1096 q->info.si_signo = sig;
1097 q->info.si_errno = 0;
1098 q->info.si_code = SI_USER;
1099 q->info.si_pid = task_tgid_nr_ns(current,
1100 task_active_pid_ns(t));
1101 q->info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
1103 case (unsigned long) SEND_SIG_PRIV:
1104 q->info.si_signo = sig;
1105 q->info.si_errno = 0;
1106 q->info.si_code = SI_KERNEL;
1111 copy_siginfo(&q->info, info);
1112 if (from_ancestor_ns)
1117 userns_fixup_signal_uid(&q->info, t);
1119 } else if (!is_si_special(info)) {
1120 if (sig >= SIGRTMIN && info->si_code != SI_USER) {
1122 * Queue overflow, abort. We may abort if the
1123 * signal was rt and sent by user using something
1124 * other than kill().
1126 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1131 * This is a silent loss of information. We still
1132 * send the signal, but the *info bits are lost.
1134 result = TRACE_SIGNAL_LOSE_INFO;
1139 signalfd_notify(t, sig);
1140 sigaddset(&pending->signal, sig);
1141 complete_signal(sig, t, group);
1143 trace_signal_generate(sig, info, t, group, result);
1147 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
1150 int from_ancestor_ns = 0;
1152 #ifdef CONFIG_PID_NS
1153 from_ancestor_ns = si_fromuser(info) &&
1154 !task_pid_nr_ns(current, task_active_pid_ns(t));
1157 return __send_signal(sig, info, t, group, from_ancestor_ns);
1160 static void print_fatal_signal(struct pt_regs *regs, int signr)
1162 printk("%s/%d: potentially unexpected fatal signal %d.\n",
1163 current->comm, task_pid_nr(current), signr);
1165 #if defined(__i386__) && !defined(__arch_um__)
1166 printk("code at %08lx: ", regs->ip);
1169 for (i = 0; i < 16; i++) {
1172 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1174 printk("%02x ", insn);
1184 static int __init setup_print_fatal_signals(char *str)
1186 get_option (&str, &print_fatal_signals);
1191 __setup("print-fatal-signals=", setup_print_fatal_signals);
1194 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1196 return send_signal(sig, info, p, 1);
1200 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1202 return send_signal(sig, info, t, 0);
1205 int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
1208 unsigned long flags;
1211 if (lock_task_sighand(p, &flags)) {
1212 ret = send_signal(sig, info, p, group);
1213 unlock_task_sighand(p, &flags);
1220 * Force a signal that the process can't ignore: if necessary
1221 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1223 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1224 * since we do not want to have a signal handler that was blocked
1225 * be invoked when user space had explicitly blocked it.
1227 * We don't want to have recursive SIGSEGV's etc, for example,
1228 * that is why we also clear SIGNAL_UNKILLABLE.
1231 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1233 unsigned long int flags;
1234 int ret, blocked, ignored;
1235 struct k_sigaction *action;
1237 spin_lock_irqsave(&t->sighand->siglock, flags);
1238 action = &t->sighand->action[sig-1];
1239 ignored = action->sa.sa_handler == SIG_IGN;
1240 blocked = sigismember(&t->blocked, sig);
1241 if (blocked || ignored) {
1242 action->sa.sa_handler = SIG_DFL;
1244 sigdelset(&t->blocked, sig);
1245 recalc_sigpending_and_wake(t);
1248 if (action->sa.sa_handler == SIG_DFL)
1249 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1250 ret = specific_send_sig_info(sig, info, t);
1251 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1257 * Nuke all other threads in the group.
1259 int zap_other_threads(struct task_struct *p)
1261 struct task_struct *t = p;
1264 p->signal->group_stop_count = 0;
1266 while_each_thread(p, t) {
1267 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1270 /* Don't bother with already dead threads */
1273 sigaddset(&t->pending.signal, SIGKILL);
1274 signal_wake_up(t, 1);
1280 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1281 unsigned long *flags)
1283 struct sighand_struct *sighand;
1286 local_irq_save(*flags);
1288 sighand = rcu_dereference(tsk->sighand);
1289 if (unlikely(sighand == NULL)) {
1291 local_irq_restore(*flags);
1295 spin_lock(&sighand->siglock);
1296 if (likely(sighand == tsk->sighand)) {
1300 spin_unlock(&sighand->siglock);
1302 local_irq_restore(*flags);
1309 * send signal info to all the members of a group
1311 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1316 ret = check_kill_permission(sig, info, p);
1320 ret = do_send_sig_info(sig, info, p, true);
1326 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1327 * control characters do (^C, ^Z etc)
1328 * - the caller must hold at least a readlock on tasklist_lock
1330 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1332 struct task_struct *p = NULL;
1333 int retval, success;
1337 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1338 int err = group_send_sig_info(sig, info, p);
1341 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1342 return success ? 0 : retval;
1345 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1348 struct task_struct *p;
1352 p = pid_task(pid, PIDTYPE_PID);
1354 error = group_send_sig_info(sig, info, p);
1355 if (unlikely(error == -ESRCH))
1357 * The task was unhashed in between, try again.
1358 * If it is dead, pid_task() will return NULL,
1359 * if we race with de_thread() it will find the
1369 int kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1373 error = kill_pid_info(sig, info, find_vpid(pid));
1378 static int kill_as_cred_perm(const struct cred *cred,
1379 struct task_struct *target)
1381 const struct cred *pcred = __task_cred(target);
1382 if (!uid_eq(cred->euid, pcred->suid) && !uid_eq(cred->euid, pcred->uid) &&
1383 !uid_eq(cred->uid, pcred->suid) && !uid_eq(cred->uid, pcred->uid))
1388 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1389 int kill_pid_info_as_cred(int sig, struct siginfo *info, struct pid *pid,
1390 const struct cred *cred, u32 secid)
1393 struct task_struct *p;
1394 unsigned long flags;
1396 if (!valid_signal(sig))
1400 p = pid_task(pid, PIDTYPE_PID);
1405 if (si_fromuser(info) && !kill_as_cred_perm(cred, p)) {
1409 ret = security_task_kill(p, info, sig, secid);
1414 if (lock_task_sighand(p, &flags)) {
1415 ret = __send_signal(sig, info, p, 1, 0);
1416 unlock_task_sighand(p, &flags);
1424 EXPORT_SYMBOL_GPL(kill_pid_info_as_cred);
1427 * kill_something_info() interprets pid in interesting ways just like kill(2).
1429 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1430 * is probably wrong. Should make it like BSD or SYSV.
1433 static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1439 ret = kill_pid_info(sig, info, find_vpid(pid));
1444 read_lock(&tasklist_lock);
1446 ret = __kill_pgrp_info(sig, info,
1447 pid ? find_vpid(-pid) : task_pgrp(current));
1449 int retval = 0, count = 0;
1450 struct task_struct * p;
1452 for_each_process(p) {
1453 if (task_pid_vnr(p) > 1 &&
1454 !same_thread_group(p, current)) {
1455 int err = group_send_sig_info(sig, info, p);
1461 ret = count ? retval : -ESRCH;
1463 read_unlock(&tasklist_lock);
1469 * These are for backward compatibility with the rest of the kernel source.
1472 int send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1475 * Make sure legacy kernel users don't send in bad values
1476 * (normal paths check this in check_kill_permission).
1478 if (!valid_signal(sig))
1481 return do_send_sig_info(sig, info, p, false);
1484 #define __si_special(priv) \
1485 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1488 send_sig(int sig, struct task_struct *p, int priv)
1490 return send_sig_info(sig, __si_special(priv), p);
1494 force_sig(int sig, struct task_struct *p)
1496 force_sig_info(sig, SEND_SIG_PRIV, p);
1500 * When things go south during signal handling, we
1501 * will force a SIGSEGV. And if the signal that caused
1502 * the problem was already a SIGSEGV, we'll want to
1503 * make sure we don't even try to deliver the signal..
1506 force_sigsegv(int sig, struct task_struct *p)
1508 if (sig == SIGSEGV) {
1509 unsigned long flags;
1510 spin_lock_irqsave(&p->sighand->siglock, flags);
1511 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1512 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1514 force_sig(SIGSEGV, p);
1518 int kill_pgrp(struct pid *pid, int sig, int priv)
1522 read_lock(&tasklist_lock);
1523 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1524 read_unlock(&tasklist_lock);
1528 EXPORT_SYMBOL(kill_pgrp);
1530 int kill_pid(struct pid *pid, int sig, int priv)
1532 return kill_pid_info(sig, __si_special(priv), pid);
1534 EXPORT_SYMBOL(kill_pid);
1537 * These functions support sending signals using preallocated sigqueue
1538 * structures. This is needed "because realtime applications cannot
1539 * afford to lose notifications of asynchronous events, like timer
1540 * expirations or I/O completions". In the case of POSIX Timers
1541 * we allocate the sigqueue structure from the timer_create. If this
1542 * allocation fails we are able to report the failure to the application
1543 * with an EAGAIN error.
1545 struct sigqueue *sigqueue_alloc(void)
1547 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1550 q->flags |= SIGQUEUE_PREALLOC;
1555 void sigqueue_free(struct sigqueue *q)
1557 unsigned long flags;
1558 spinlock_t *lock = ¤t->sighand->siglock;
1560 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1562 * We must hold ->siglock while testing q->list
1563 * to serialize with collect_signal() or with
1564 * __exit_signal()->flush_sigqueue().
1566 spin_lock_irqsave(lock, flags);
1567 q->flags &= ~SIGQUEUE_PREALLOC;
1569 * If it is queued it will be freed when dequeued,
1570 * like the "regular" sigqueue.
1572 if (!list_empty(&q->list))
1574 spin_unlock_irqrestore(lock, flags);
1580 int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1582 int sig = q->info.si_signo;
1583 struct sigpending *pending;
1584 unsigned long flags;
1587 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1590 if (!likely(lock_task_sighand(t, &flags)))
1593 ret = 1; /* the signal is ignored */
1594 result = TRACE_SIGNAL_IGNORED;
1595 if (!prepare_signal(sig, t, false))
1599 if (unlikely(!list_empty(&q->list))) {
1601 * If an SI_TIMER entry is already queue just increment
1602 * the overrun count.
1604 BUG_ON(q->info.si_code != SI_TIMER);
1605 q->info.si_overrun++;
1606 result = TRACE_SIGNAL_ALREADY_PENDING;
1609 q->info.si_overrun = 0;
1611 signalfd_notify(t, sig);
1612 pending = group ? &t->signal->shared_pending : &t->pending;
1613 list_add_tail(&q->list, &pending->list);
1614 sigaddset(&pending->signal, sig);
1615 complete_signal(sig, t, group);
1616 result = TRACE_SIGNAL_DELIVERED;
1618 trace_signal_generate(sig, &q->info, t, group, result);
1619 unlock_task_sighand(t, &flags);
1625 * Let a parent know about the death of a child.
1626 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1628 * Returns true if our parent ignored us and so we've switched to
1631 bool do_notify_parent(struct task_struct *tsk, int sig)
1633 struct siginfo info;
1634 unsigned long flags;
1635 struct sighand_struct *psig;
1636 bool autoreap = false;
1640 /* do_notify_parent_cldstop should have been called instead. */
1641 BUG_ON(task_is_stopped_or_traced(tsk));
1643 BUG_ON(!tsk->ptrace &&
1644 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1646 if (sig != SIGCHLD) {
1648 * This is only possible if parent == real_parent.
1649 * Check if it has changed security domain.
1651 if (tsk->parent_exec_id != tsk->parent->self_exec_id)
1655 info.si_signo = sig;
1658 * we are under tasklist_lock here so our parent is tied to
1659 * us and cannot exit and release its namespace.
1661 * the only it can is to switch its nsproxy with sys_unshare,
1662 * bu uncharing pid namespaces is not allowed, so we'll always
1663 * see relevant namespace
1665 * write_lock() currently calls preempt_disable() which is the
1666 * same as rcu_read_lock(), but according to Oleg, this is not
1667 * correct to rely on this
1670 info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1671 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1675 info.si_utime = cputime_to_clock_t(tsk->utime + tsk->signal->utime);
1676 info.si_stime = cputime_to_clock_t(tsk->stime + tsk->signal->stime);
1678 info.si_status = tsk->exit_code & 0x7f;
1679 if (tsk->exit_code & 0x80)
1680 info.si_code = CLD_DUMPED;
1681 else if (tsk->exit_code & 0x7f)
1682 info.si_code = CLD_KILLED;
1684 info.si_code = CLD_EXITED;
1685 info.si_status = tsk->exit_code >> 8;
1688 psig = tsk->parent->sighand;
1689 spin_lock_irqsave(&psig->siglock, flags);
1690 if (!tsk->ptrace && sig == SIGCHLD &&
1691 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1692 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1694 * We are exiting and our parent doesn't care. POSIX.1
1695 * defines special semantics for setting SIGCHLD to SIG_IGN
1696 * or setting the SA_NOCLDWAIT flag: we should be reaped
1697 * automatically and not left for our parent's wait4 call.
1698 * Rather than having the parent do it as a magic kind of
1699 * signal handler, we just set this to tell do_exit that we
1700 * can be cleaned up without becoming a zombie. Note that
1701 * we still call __wake_up_parent in this case, because a
1702 * blocked sys_wait4 might now return -ECHILD.
1704 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1705 * is implementation-defined: we do (if you don't want
1706 * it, just use SIG_IGN instead).
1709 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1712 if (valid_signal(sig) && sig)
1713 __group_send_sig_info(sig, &info, tsk->parent);
1714 __wake_up_parent(tsk, tsk->parent);
1715 spin_unlock_irqrestore(&psig->siglock, flags);
1721 * do_notify_parent_cldstop - notify parent of stopped/continued state change
1722 * @tsk: task reporting the state change
1723 * @for_ptracer: the notification is for ptracer
1724 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
1726 * Notify @tsk's parent that the stopped/continued state has changed. If
1727 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
1728 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
1731 * Must be called with tasklist_lock at least read locked.
1733 static void do_notify_parent_cldstop(struct task_struct *tsk,
1734 bool for_ptracer, int why)
1736 struct siginfo info;
1737 unsigned long flags;
1738 struct task_struct *parent;
1739 struct sighand_struct *sighand;
1742 parent = tsk->parent;
1744 tsk = tsk->group_leader;
1745 parent = tsk->real_parent;
1748 info.si_signo = SIGCHLD;
1751 * see comment in do_notify_parent() about the following 4 lines
1754 info.si_pid = task_pid_nr_ns(tsk, parent->nsproxy->pid_ns);
1755 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
1758 info.si_utime = cputime_to_clock_t(tsk->utime);
1759 info.si_stime = cputime_to_clock_t(tsk->stime);
1764 info.si_status = SIGCONT;
1767 info.si_status = tsk->signal->group_exit_code & 0x7f;
1770 info.si_status = tsk->exit_code & 0x7f;
1776 sighand = parent->sighand;
1777 spin_lock_irqsave(&sighand->siglock, flags);
1778 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1779 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1780 __group_send_sig_info(SIGCHLD, &info, parent);
1782 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1784 __wake_up_parent(tsk, parent);
1785 spin_unlock_irqrestore(&sighand->siglock, flags);
1788 static inline int may_ptrace_stop(void)
1790 if (!likely(current->ptrace))
1793 * Are we in the middle of do_coredump?
1794 * If so and our tracer is also part of the coredump stopping
1795 * is a deadlock situation, and pointless because our tracer
1796 * is dead so don't allow us to stop.
1797 * If SIGKILL was already sent before the caller unlocked
1798 * ->siglock we must see ->core_state != NULL. Otherwise it
1799 * is safe to enter schedule().
1801 if (unlikely(current->mm->core_state) &&
1802 unlikely(current->mm == current->parent->mm))
1809 * Return non-zero if there is a SIGKILL that should be waking us up.
1810 * Called with the siglock held.
1812 static int sigkill_pending(struct task_struct *tsk)
1814 return sigismember(&tsk->pending.signal, SIGKILL) ||
1815 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1819 * This must be called with current->sighand->siglock held.
1821 * This should be the path for all ptrace stops.
1822 * We always set current->last_siginfo while stopped here.
1823 * That makes it a way to test a stopped process for
1824 * being ptrace-stopped vs being job-control-stopped.
1826 * If we actually decide not to stop at all because the tracer
1827 * is gone, we keep current->exit_code unless clear_code.
1829 static void ptrace_stop(int exit_code, int why, int clear_code, siginfo_t *info)
1830 __releases(¤t->sighand->siglock)
1831 __acquires(¤t->sighand->siglock)
1833 bool gstop_done = false;
1835 if (arch_ptrace_stop_needed(exit_code, info)) {
1837 * The arch code has something special to do before a
1838 * ptrace stop. This is allowed to block, e.g. for faults
1839 * on user stack pages. We can't keep the siglock while
1840 * calling arch_ptrace_stop, so we must release it now.
1841 * To preserve proper semantics, we must do this before
1842 * any signal bookkeeping like checking group_stop_count.
1843 * Meanwhile, a SIGKILL could come in before we retake the
1844 * siglock. That must prevent us from sleeping in TASK_TRACED.
1845 * So after regaining the lock, we must check for SIGKILL.
1847 spin_unlock_irq(¤t->sighand->siglock);
1848 arch_ptrace_stop(exit_code, info);
1849 spin_lock_irq(¤t->sighand->siglock);
1850 if (sigkill_pending(current))
1855 * We're committing to trapping. TRACED should be visible before
1856 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
1857 * Also, transition to TRACED and updates to ->jobctl should be
1858 * atomic with respect to siglock and should be done after the arch
1859 * hook as siglock is released and regrabbed across it.
1861 set_current_state(TASK_TRACED);
1863 current->last_siginfo = info;
1864 current->exit_code = exit_code;
1867 * If @why is CLD_STOPPED, we're trapping to participate in a group
1868 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
1869 * across siglock relocks since INTERRUPT was scheduled, PENDING
1870 * could be clear now. We act as if SIGCONT is received after
1871 * TASK_TRACED is entered - ignore it.
1873 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
1874 gstop_done = task_participate_group_stop(current);
1876 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
1877 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
1878 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
1879 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
1881 /* entering a trap, clear TRAPPING */
1882 task_clear_jobctl_trapping(current);
1884 spin_unlock_irq(¤t->sighand->siglock);
1885 read_lock(&tasklist_lock);
1886 if (may_ptrace_stop()) {
1888 * Notify parents of the stop.
1890 * While ptraced, there are two parents - the ptracer and
1891 * the real_parent of the group_leader. The ptracer should
1892 * know about every stop while the real parent is only
1893 * interested in the completion of group stop. The states
1894 * for the two don't interact with each other. Notify
1895 * separately unless they're gonna be duplicates.
1897 do_notify_parent_cldstop(current, true, why);
1898 if (gstop_done && ptrace_reparented(current))
1899 do_notify_parent_cldstop(current, false, why);
1902 * Don't want to allow preemption here, because
1903 * sys_ptrace() needs this task to be inactive.
1905 * XXX: implement read_unlock_no_resched().
1908 read_unlock(&tasklist_lock);
1909 preempt_enable_no_resched();
1913 * By the time we got the lock, our tracer went away.
1914 * Don't drop the lock yet, another tracer may come.
1916 * If @gstop_done, the ptracer went away between group stop
1917 * completion and here. During detach, it would have set
1918 * JOBCTL_STOP_PENDING on us and we'll re-enter
1919 * TASK_STOPPED in do_signal_stop() on return, so notifying
1920 * the real parent of the group stop completion is enough.
1923 do_notify_parent_cldstop(current, false, why);
1925 __set_current_state(TASK_RUNNING);
1927 current->exit_code = 0;
1928 read_unlock(&tasklist_lock);
1932 * While in TASK_TRACED, we were considered "frozen enough".
1933 * Now that we woke up, it's crucial if we're supposed to be
1934 * frozen that we freeze now before running anything substantial.
1939 * We are back. Now reacquire the siglock before touching
1940 * last_siginfo, so that we are sure to have synchronized with
1941 * any signal-sending on another CPU that wants to examine it.
1943 spin_lock_irq(¤t->sighand->siglock);
1944 current->last_siginfo = NULL;
1946 /* LISTENING can be set only during STOP traps, clear it */
1947 current->jobctl &= ~JOBCTL_LISTENING;
1950 * Queued signals ignored us while we were stopped for tracing.
1951 * So check for any that we should take before resuming user mode.
1952 * This sets TIF_SIGPENDING, but never clears it.
1954 recalc_sigpending_tsk(current);
1957 static void ptrace_do_notify(int signr, int exit_code, int why)
1961 memset(&info, 0, sizeof info);
1962 info.si_signo = signr;
1963 info.si_code = exit_code;
1964 info.si_pid = task_pid_vnr(current);
1965 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
1967 /* Let the debugger run. */
1968 ptrace_stop(exit_code, why, 1, &info);
1971 void ptrace_notify(int exit_code)
1973 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1975 spin_lock_irq(¤t->sighand->siglock);
1976 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
1977 spin_unlock_irq(¤t->sighand->siglock);
1981 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
1982 * @signr: signr causing group stop if initiating
1984 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
1985 * and participate in it. If already set, participate in the existing
1986 * group stop. If participated in a group stop (and thus slept), %true is
1987 * returned with siglock released.
1989 * If ptraced, this function doesn't handle stop itself. Instead,
1990 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
1991 * untouched. The caller must ensure that INTERRUPT trap handling takes
1992 * places afterwards.
1995 * Must be called with @current->sighand->siglock held, which is released
1999 * %false if group stop is already cancelled or ptrace trap is scheduled.
2000 * %true if participated in group stop.
2002 static bool do_signal_stop(int signr)
2003 __releases(¤t->sighand->siglock)
2005 struct signal_struct *sig = current->signal;
2007 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2008 unsigned int gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2009 struct task_struct *t;
2011 /* signr will be recorded in task->jobctl for retries */
2012 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2014 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2015 unlikely(signal_group_exit(sig)))
2018 * There is no group stop already in progress. We must
2021 * While ptraced, a task may be resumed while group stop is
2022 * still in effect and then receive a stop signal and
2023 * initiate another group stop. This deviates from the
2024 * usual behavior as two consecutive stop signals can't
2025 * cause two group stops when !ptraced. That is why we
2026 * also check !task_is_stopped(t) below.
2028 * The condition can be distinguished by testing whether
2029 * SIGNAL_STOP_STOPPED is already set. Don't generate
2030 * group_exit_code in such case.
2032 * This is not necessary for SIGNAL_STOP_CONTINUED because
2033 * an intervening stop signal is required to cause two
2034 * continued events regardless of ptrace.
2036 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2037 sig->group_exit_code = signr;
2039 sig->group_stop_count = 0;
2041 if (task_set_jobctl_pending(current, signr | gstop))
2042 sig->group_stop_count++;
2044 for (t = next_thread(current); t != current;
2045 t = next_thread(t)) {
2047 * Setting state to TASK_STOPPED for a group
2048 * stop is always done with the siglock held,
2049 * so this check has no races.
2051 if (!task_is_stopped(t) &&
2052 task_set_jobctl_pending(t, signr | gstop)) {
2053 sig->group_stop_count++;
2054 if (likely(!(t->ptrace & PT_SEIZED)))
2055 signal_wake_up(t, 0);
2057 ptrace_trap_notify(t);
2062 if (likely(!current->ptrace)) {
2066 * If there are no other threads in the group, or if there
2067 * is a group stop in progress and we are the last to stop,
2068 * report to the parent.
2070 if (task_participate_group_stop(current))
2071 notify = CLD_STOPPED;
2073 __set_current_state(TASK_STOPPED);
2074 spin_unlock_irq(¤t->sighand->siglock);
2077 * Notify the parent of the group stop completion. Because
2078 * we're not holding either the siglock or tasklist_lock
2079 * here, ptracer may attach inbetween; however, this is for
2080 * group stop and should always be delivered to the real
2081 * parent of the group leader. The new ptracer will get
2082 * its notification when this task transitions into
2086 read_lock(&tasklist_lock);
2087 do_notify_parent_cldstop(current, false, notify);
2088 read_unlock(&tasklist_lock);
2091 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2096 * While ptraced, group stop is handled by STOP trap.
2097 * Schedule it and let the caller deal with it.
2099 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2105 * do_jobctl_trap - take care of ptrace jobctl traps
2107 * When PT_SEIZED, it's used for both group stop and explicit
2108 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2109 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2110 * the stop signal; otherwise, %SIGTRAP.
2112 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2113 * number as exit_code and no siginfo.
2116 * Must be called with @current->sighand->siglock held, which may be
2117 * released and re-acquired before returning with intervening sleep.
2119 static void do_jobctl_trap(void)
2121 struct signal_struct *signal = current->signal;
2122 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2124 if (current->ptrace & PT_SEIZED) {
2125 if (!signal->group_stop_count &&
2126 !(signal->flags & SIGNAL_STOP_STOPPED))
2128 WARN_ON_ONCE(!signr);
2129 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2132 WARN_ON_ONCE(!signr);
2133 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2134 current->exit_code = 0;
2138 static int ptrace_signal(int signr, siginfo_t *info,
2139 struct pt_regs *regs, void *cookie)
2141 ptrace_signal_deliver(regs, cookie);
2143 * We do not check sig_kernel_stop(signr) but set this marker
2144 * unconditionally because we do not know whether debugger will
2145 * change signr. This flag has no meaning unless we are going
2146 * to stop after return from ptrace_stop(). In this case it will
2147 * be checked in do_signal_stop(), we should only stop if it was
2148 * not cleared by SIGCONT while we were sleeping. See also the
2149 * comment in dequeue_signal().
2151 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2152 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2154 /* We're back. Did the debugger cancel the sig? */
2155 signr = current->exit_code;
2159 current->exit_code = 0;
2162 * Update the siginfo structure if the signal has
2163 * changed. If the debugger wanted something
2164 * specific in the siginfo structure then it should
2165 * have updated *info via PTRACE_SETSIGINFO.
2167 if (signr != info->si_signo) {
2168 info->si_signo = signr;
2170 info->si_code = SI_USER;
2172 info->si_pid = task_pid_vnr(current->parent);
2173 info->si_uid = from_kuid_munged(current_user_ns(),
2174 task_uid(current->parent));
2178 /* If the (new) signal is now blocked, requeue it. */
2179 if (sigismember(¤t->blocked, signr)) {
2180 specific_send_sig_info(signr, info, current);
2187 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
2188 struct pt_regs *regs, void *cookie)
2190 struct sighand_struct *sighand = current->sighand;
2191 struct signal_struct *signal = current->signal;
2196 * We'll jump back here after any time we were stopped in TASK_STOPPED.
2197 * While in TASK_STOPPED, we were considered "frozen enough".
2198 * Now that we woke up, it's crucial if we're supposed to be
2199 * frozen that we freeze now before running anything substantial.
2203 spin_lock_irq(&sighand->siglock);
2205 * Every stopped thread goes here after wakeup. Check to see if
2206 * we should notify the parent, prepare_signal(SIGCONT) encodes
2207 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2209 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2212 if (signal->flags & SIGNAL_CLD_CONTINUED)
2213 why = CLD_CONTINUED;
2217 signal->flags &= ~SIGNAL_CLD_MASK;
2219 spin_unlock_irq(&sighand->siglock);
2222 * Notify the parent that we're continuing. This event is
2223 * always per-process and doesn't make whole lot of sense
2224 * for ptracers, who shouldn't consume the state via
2225 * wait(2) either, but, for backward compatibility, notify
2226 * the ptracer of the group leader too unless it's gonna be
2229 read_lock(&tasklist_lock);
2230 do_notify_parent_cldstop(current, false, why);
2232 if (ptrace_reparented(current->group_leader))
2233 do_notify_parent_cldstop(current->group_leader,
2235 read_unlock(&tasklist_lock);
2241 struct k_sigaction *ka;
2243 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2247 if (unlikely(current->jobctl & JOBCTL_TRAP_MASK)) {
2249 spin_unlock_irq(&sighand->siglock);
2253 signr = dequeue_signal(current, ¤t->blocked, info);
2256 break; /* will return 0 */
2258 if (unlikely(current->ptrace) && signr != SIGKILL) {
2259 signr = ptrace_signal(signr, info,
2265 ka = &sighand->action[signr-1];
2267 /* Trace actually delivered signals. */
2268 trace_signal_deliver(signr, info, ka);
2270 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2272 if (ka->sa.sa_handler != SIG_DFL) {
2273 /* Run the handler. */
2276 if (ka->sa.sa_flags & SA_ONESHOT)
2277 ka->sa.sa_handler = SIG_DFL;
2279 break; /* will return non-zero "signr" value */
2283 * Now we are doing the default action for this signal.
2285 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2289 * Global init gets no signals it doesn't want.
2290 * Container-init gets no signals it doesn't want from same
2293 * Note that if global/container-init sees a sig_kernel_only()
2294 * signal here, the signal must have been generated internally
2295 * or must have come from an ancestor namespace. In either
2296 * case, the signal cannot be dropped.
2298 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2299 !sig_kernel_only(signr))
2302 if (sig_kernel_stop(signr)) {
2304 * The default action is to stop all threads in
2305 * the thread group. The job control signals
2306 * do nothing in an orphaned pgrp, but SIGSTOP
2307 * always works. Note that siglock needs to be
2308 * dropped during the call to is_orphaned_pgrp()
2309 * because of lock ordering with tasklist_lock.
2310 * This allows an intervening SIGCONT to be posted.
2311 * We need to check for that and bail out if necessary.
2313 if (signr != SIGSTOP) {
2314 spin_unlock_irq(&sighand->siglock);
2316 /* signals can be posted during this window */
2318 if (is_current_pgrp_orphaned())
2321 spin_lock_irq(&sighand->siglock);
2324 if (likely(do_signal_stop(info->si_signo))) {
2325 /* It released the siglock. */
2330 * We didn't actually stop, due to a race
2331 * with SIGCONT or something like that.
2336 spin_unlock_irq(&sighand->siglock);
2339 * Anything else is fatal, maybe with a core dump.
2341 current->flags |= PF_SIGNALED;
2343 if (sig_kernel_coredump(signr)) {
2344 if (print_fatal_signals)
2345 print_fatal_signal(regs, info->si_signo);
2347 * If it was able to dump core, this kills all
2348 * other threads in the group and synchronizes with
2349 * their demise. If we lost the race with another
2350 * thread getting here, it set group_exit_code
2351 * first and our do_group_exit call below will use
2352 * that value and ignore the one we pass it.
2354 do_coredump(info->si_signo, info->si_signo, regs);
2358 * Death signals, no core dump.
2360 do_group_exit(info->si_signo);
2363 spin_unlock_irq(&sighand->siglock);
2368 * block_sigmask - add @ka's signal mask to current->blocked
2369 * @ka: action for @signr
2370 * @signr: signal that has been successfully delivered
2372 * This function should be called when a signal has succesfully been
2373 * delivered. It adds the mask of signals for @ka to current->blocked
2374 * so that they are blocked during the execution of the signal
2375 * handler. In addition, @signr will be blocked unless %SA_NODEFER is
2376 * set in @ka->sa.sa_flags.
2378 void block_sigmask(struct k_sigaction *ka, int signr)
2382 sigorsets(&blocked, ¤t->blocked, &ka->sa.sa_mask);
2383 if (!(ka->sa.sa_flags & SA_NODEFER))
2384 sigaddset(&blocked, signr);
2385 set_current_blocked(&blocked);
2389 * It could be that complete_signal() picked us to notify about the
2390 * group-wide signal. Other threads should be notified now to take
2391 * the shared signals in @which since we will not.
2393 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2396 struct task_struct *t;
2398 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2399 if (sigisemptyset(&retarget))
2403 while_each_thread(tsk, t) {
2404 if (t->flags & PF_EXITING)
2407 if (!has_pending_signals(&retarget, &t->blocked))
2409 /* Remove the signals this thread can handle. */
2410 sigandsets(&retarget, &retarget, &t->blocked);
2412 if (!signal_pending(t))
2413 signal_wake_up(t, 0);
2415 if (sigisemptyset(&retarget))
2420 void exit_signals(struct task_struct *tsk)
2426 * @tsk is about to have PF_EXITING set - lock out users which
2427 * expect stable threadgroup.
2429 threadgroup_change_begin(tsk);
2431 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2432 tsk->flags |= PF_EXITING;
2433 threadgroup_change_end(tsk);
2437 spin_lock_irq(&tsk->sighand->siglock);
2439 * From now this task is not visible for group-wide signals,
2440 * see wants_signal(), do_signal_stop().
2442 tsk->flags |= PF_EXITING;
2444 threadgroup_change_end(tsk);
2446 if (!signal_pending(tsk))
2449 unblocked = tsk->blocked;
2450 signotset(&unblocked);
2451 retarget_shared_pending(tsk, &unblocked);
2453 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2454 task_participate_group_stop(tsk))
2455 group_stop = CLD_STOPPED;
2457 spin_unlock_irq(&tsk->sighand->siglock);
2460 * If group stop has completed, deliver the notification. This
2461 * should always go to the real parent of the group leader.
2463 if (unlikely(group_stop)) {
2464 read_lock(&tasklist_lock);
2465 do_notify_parent_cldstop(tsk, false, group_stop);
2466 read_unlock(&tasklist_lock);
2470 EXPORT_SYMBOL(recalc_sigpending);
2471 EXPORT_SYMBOL_GPL(dequeue_signal);
2472 EXPORT_SYMBOL(flush_signals);
2473 EXPORT_SYMBOL(force_sig);
2474 EXPORT_SYMBOL(send_sig);
2475 EXPORT_SYMBOL(send_sig_info);
2476 EXPORT_SYMBOL(sigprocmask);
2477 EXPORT_SYMBOL(block_all_signals);
2478 EXPORT_SYMBOL(unblock_all_signals);
2482 * System call entry points.
2486 * sys_restart_syscall - restart a system call
2488 SYSCALL_DEFINE0(restart_syscall)
2490 struct restart_block *restart = ¤t_thread_info()->restart_block;
2491 return restart->fn(restart);
2494 long do_no_restart_syscall(struct restart_block *param)
2499 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2501 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2502 sigset_t newblocked;
2503 /* A set of now blocked but previously unblocked signals. */
2504 sigandnsets(&newblocked, newset, ¤t->blocked);
2505 retarget_shared_pending(tsk, &newblocked);
2507 tsk->blocked = *newset;
2508 recalc_sigpending();
2512 * set_current_blocked - change current->blocked mask
2515 * It is wrong to change ->blocked directly, this helper should be used
2516 * to ensure the process can't miss a shared signal we are going to block.
2518 void set_current_blocked(const sigset_t *newset)
2520 struct task_struct *tsk = current;
2522 spin_lock_irq(&tsk->sighand->siglock);
2523 __set_task_blocked(tsk, newset);
2524 spin_unlock_irq(&tsk->sighand->siglock);
2528 * This is also useful for kernel threads that want to temporarily
2529 * (or permanently) block certain signals.
2531 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2532 * interface happily blocks "unblockable" signals like SIGKILL
2535 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2537 struct task_struct *tsk = current;
2540 /* Lockless, only current can change ->blocked, never from irq */
2542 *oldset = tsk->blocked;
2546 sigorsets(&newset, &tsk->blocked, set);
2549 sigandnsets(&newset, &tsk->blocked, set);
2558 set_current_blocked(&newset);
2563 * sys_rt_sigprocmask - change the list of currently blocked signals
2564 * @how: whether to add, remove, or set signals
2565 * @nset: stores pending signals
2566 * @oset: previous value of signal mask if non-null
2567 * @sigsetsize: size of sigset_t type
2569 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
2570 sigset_t __user *, oset, size_t, sigsetsize)
2572 sigset_t old_set, new_set;
2575 /* XXX: Don't preclude handling different sized sigset_t's. */
2576 if (sigsetsize != sizeof(sigset_t))
2579 old_set = current->blocked;
2582 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
2584 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2586 error = sigprocmask(how, &new_set, NULL);
2592 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
2599 long do_sigpending(void __user *set, unsigned long sigsetsize)
2601 long error = -EINVAL;
2604 if (sigsetsize > sizeof(sigset_t))
2607 spin_lock_irq(¤t->sighand->siglock);
2608 sigorsets(&pending, ¤t->pending.signal,
2609 ¤t->signal->shared_pending.signal);
2610 spin_unlock_irq(¤t->sighand->siglock);
2612 /* Outside the lock because only this thread touches it. */
2613 sigandsets(&pending, ¤t->blocked, &pending);
2616 if (!copy_to_user(set, &pending, sigsetsize))
2624 * sys_rt_sigpending - examine a pending signal that has been raised
2626 * @set: stores pending signals
2627 * @sigsetsize: size of sigset_t type or larger
2629 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2631 return do_sigpending(set, sigsetsize);
2634 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2636 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2640 if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2642 if (from->si_code < 0)
2643 return __copy_to_user(to, from, sizeof(siginfo_t))
2646 * If you change siginfo_t structure, please be sure
2647 * this code is fixed accordingly.
2648 * Please remember to update the signalfd_copyinfo() function
2649 * inside fs/signalfd.c too, in case siginfo_t changes.
2650 * It should never copy any pad contained in the structure
2651 * to avoid security leaks, but must copy the generic
2652 * 3 ints plus the relevant union member.
2654 err = __put_user(from->si_signo, &to->si_signo);
2655 err |= __put_user(from->si_errno, &to->si_errno);
2656 err |= __put_user((short)from->si_code, &to->si_code);
2657 switch (from->si_code & __SI_MASK) {
2659 err |= __put_user(from->si_pid, &to->si_pid);
2660 err |= __put_user(from->si_uid, &to->si_uid);
2663 err |= __put_user(from->si_tid, &to->si_tid);
2664 err |= __put_user(from->si_overrun, &to->si_overrun);
2665 err |= __put_user(from->si_ptr, &to->si_ptr);
2668 err |= __put_user(from->si_band, &to->si_band);
2669 err |= __put_user(from->si_fd, &to->si_fd);
2672 err |= __put_user(from->si_addr, &to->si_addr);
2673 #ifdef __ARCH_SI_TRAPNO
2674 err |= __put_user(from->si_trapno, &to->si_trapno);
2676 #ifdef BUS_MCEERR_AO
2678 * Other callers might not initialize the si_lsb field,
2679 * so check explicitly for the right codes here.
2681 if (from->si_code == BUS_MCEERR_AR || from->si_code == BUS_MCEERR_AO)
2682 err |= __put_user(from->si_addr_lsb, &to->si_addr_lsb);
2686 err |= __put_user(from->si_pid, &to->si_pid);
2687 err |= __put_user(from->si_uid, &to->si_uid);
2688 err |= __put_user(from->si_status, &to->si_status);
2689 err |= __put_user(from->si_utime, &to->si_utime);
2690 err |= __put_user(from->si_stime, &to->si_stime);
2692 case __SI_RT: /* This is not generated by the kernel as of now. */
2693 case __SI_MESGQ: /* But this is */
2694 err |= __put_user(from->si_pid, &to->si_pid);
2695 err |= __put_user(from->si_uid, &to->si_uid);
2696 err |= __put_user(from->si_ptr, &to->si_ptr);
2698 #ifdef __ARCH_SIGSYS
2700 err |= __put_user(from->si_call_addr, &to->si_call_addr);
2701 err |= __put_user(from->si_syscall, &to->si_syscall);
2702 err |= __put_user(from->si_arch, &to->si_arch);
2705 default: /* this is just in case for now ... */
2706 err |= __put_user(from->si_pid, &to->si_pid);
2707 err |= __put_user(from->si_uid, &to->si_uid);
2716 * do_sigtimedwait - wait for queued signals specified in @which
2717 * @which: queued signals to wait for
2718 * @info: if non-null, the signal's siginfo is returned here
2719 * @ts: upper bound on process time suspension
2721 int do_sigtimedwait(const sigset_t *which, siginfo_t *info,
2722 const struct timespec *ts)
2724 struct task_struct *tsk = current;
2725 long timeout = MAX_SCHEDULE_TIMEOUT;
2726 sigset_t mask = *which;
2730 if (!timespec_valid(ts))
2732 timeout = timespec_to_jiffies(ts);
2734 * We can be close to the next tick, add another one
2735 * to ensure we will wait at least the time asked for.
2737 if (ts->tv_sec || ts->tv_nsec)
2742 * Invert the set of allowed signals to get those we want to block.
2744 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
2747 spin_lock_irq(&tsk->sighand->siglock);
2748 sig = dequeue_signal(tsk, &mask, info);
2749 if (!sig && timeout) {
2751 * None ready, temporarily unblock those we're interested
2752 * while we are sleeping in so that we'll be awakened when
2753 * they arrive. Unblocking is always fine, we can avoid
2754 * set_current_blocked().
2756 tsk->real_blocked = tsk->blocked;
2757 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
2758 recalc_sigpending();
2759 spin_unlock_irq(&tsk->sighand->siglock);
2761 timeout = schedule_timeout_interruptible(timeout);
2763 spin_lock_irq(&tsk->sighand->siglock);
2764 __set_task_blocked(tsk, &tsk->real_blocked);
2765 siginitset(&tsk->real_blocked, 0);
2766 sig = dequeue_signal(tsk, &mask, info);
2768 spin_unlock_irq(&tsk->sighand->siglock);
2772 return timeout ? -EINTR : -EAGAIN;
2776 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
2778 * @uthese: queued signals to wait for
2779 * @uinfo: if non-null, the signal's siginfo is returned here
2780 * @uts: upper bound on process time suspension
2781 * @sigsetsize: size of sigset_t type
2783 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2784 siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2792 /* XXX: Don't preclude handling different sized sigset_t's. */
2793 if (sigsetsize != sizeof(sigset_t))
2796 if (copy_from_user(&these, uthese, sizeof(these)))
2800 if (copy_from_user(&ts, uts, sizeof(ts)))
2804 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
2806 if (ret > 0 && uinfo) {
2807 if (copy_siginfo_to_user(uinfo, &info))
2815 * sys_kill - send a signal to a process
2816 * @pid: the PID of the process
2817 * @sig: signal to be sent
2819 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2821 struct siginfo info;
2823 info.si_signo = sig;
2825 info.si_code = SI_USER;
2826 info.si_pid = task_tgid_vnr(current);
2827 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2829 return kill_something_info(sig, &info, pid);
2833 do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
2835 struct task_struct *p;
2839 p = find_task_by_vpid(pid);
2840 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2841 error = check_kill_permission(sig, info, p);
2843 * The null signal is a permissions and process existence
2844 * probe. No signal is actually delivered.
2846 if (!error && sig) {
2847 error = do_send_sig_info(sig, info, p, false);
2849 * If lock_task_sighand() failed we pretend the task
2850 * dies after receiving the signal. The window is tiny,
2851 * and the signal is private anyway.
2853 if (unlikely(error == -ESRCH))
2862 static int do_tkill(pid_t tgid, pid_t pid, int sig)
2864 struct siginfo info;
2866 info.si_signo = sig;
2868 info.si_code = SI_TKILL;
2869 info.si_pid = task_tgid_vnr(current);
2870 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2872 return do_send_specific(tgid, pid, sig, &info);
2876 * sys_tgkill - send signal to one specific thread
2877 * @tgid: the thread group ID of the thread
2878 * @pid: the PID of the thread
2879 * @sig: signal to be sent
2881 * This syscall also checks the @tgid and returns -ESRCH even if the PID
2882 * exists but it's not belonging to the target process anymore. This
2883 * method solves the problem of threads exiting and PIDs getting reused.
2885 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2887 /* This is only valid for single tasks */
2888 if (pid <= 0 || tgid <= 0)
2891 return do_tkill(tgid, pid, sig);
2895 * sys_tkill - send signal to one specific task
2896 * @pid: the PID of the task
2897 * @sig: signal to be sent
2899 * Send a signal to only one task, even if it's a CLONE_THREAD task.
2901 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2903 /* This is only valid for single tasks */
2907 return do_tkill(0, pid, sig);
2911 * sys_rt_sigqueueinfo - send signal information to a signal
2912 * @pid: the PID of the thread
2913 * @sig: signal to be sent
2914 * @uinfo: signal info to be sent
2916 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2917 siginfo_t __user *, uinfo)
2921 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2924 /* Not even root can pretend to send signals from the kernel.
2925 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2927 if (info.si_code >= 0 || info.si_code == SI_TKILL) {
2928 /* We used to allow any < 0 si_code */
2929 WARN_ON_ONCE(info.si_code < 0);
2932 info.si_signo = sig;
2934 /* POSIX.1b doesn't mention process groups. */
2935 return kill_proc_info(sig, &info, pid);
2938 long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
2940 /* This is only valid for single tasks */
2941 if (pid <= 0 || tgid <= 0)
2944 /* Not even root can pretend to send signals from the kernel.
2945 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2947 if (info->si_code >= 0 || info->si_code == SI_TKILL) {
2948 /* We used to allow any < 0 si_code */
2949 WARN_ON_ONCE(info->si_code < 0);
2952 info->si_signo = sig;
2954 return do_send_specific(tgid, pid, sig, info);
2957 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
2958 siginfo_t __user *, uinfo)
2962 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2965 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
2968 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2970 struct task_struct *t = current;
2971 struct k_sigaction *k;
2974 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2977 k = &t->sighand->action[sig-1];
2979 spin_lock_irq(¤t->sighand->siglock);
2984 sigdelsetmask(&act->sa.sa_mask,
2985 sigmask(SIGKILL) | sigmask(SIGSTOP));
2989 * "Setting a signal action to SIG_IGN for a signal that is
2990 * pending shall cause the pending signal to be discarded,
2991 * whether or not it is blocked."
2993 * "Setting a signal action to SIG_DFL for a signal that is
2994 * pending and whose default action is to ignore the signal
2995 * (for example, SIGCHLD), shall cause the pending signal to
2996 * be discarded, whether or not it is blocked"
2998 if (sig_handler_ignored(sig_handler(t, sig), sig)) {
3000 sigaddset(&mask, sig);
3001 rm_from_queue_full(&mask, &t->signal->shared_pending);
3003 rm_from_queue_full(&mask, &t->pending);
3005 } while (t != current);
3009 spin_unlock_irq(¤t->sighand->siglock);
3014 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
3019 oss.ss_sp = (void __user *) current->sas_ss_sp;
3020 oss.ss_size = current->sas_ss_size;
3021 oss.ss_flags = sas_ss_flags(sp);
3029 if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
3031 error = __get_user(ss_sp, &uss->ss_sp) |
3032 __get_user(ss_flags, &uss->ss_flags) |
3033 __get_user(ss_size, &uss->ss_size);
3038 if (on_sig_stack(sp))
3043 * Note - this code used to test ss_flags incorrectly:
3044 * old code may have been written using ss_flags==0
3045 * to mean ss_flags==SS_ONSTACK (as this was the only
3046 * way that worked) - this fix preserves that older
3049 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
3052 if (ss_flags == SS_DISABLE) {
3057 if (ss_size < MINSIGSTKSZ)
3061 current->sas_ss_sp = (unsigned long) ss_sp;
3062 current->sas_ss_size = ss_size;
3068 if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
3070 error = __put_user(oss.ss_sp, &uoss->ss_sp) |
3071 __put_user(oss.ss_size, &uoss->ss_size) |
3072 __put_user(oss.ss_flags, &uoss->ss_flags);
3079 #ifdef __ARCH_WANT_SYS_SIGPENDING
3082 * sys_sigpending - examine pending signals
3083 * @set: where mask of pending signal is returned
3085 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
3087 return do_sigpending(set, sizeof(*set));
3092 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
3094 * sys_sigprocmask - examine and change blocked signals
3095 * @how: whether to add, remove, or set signals
3096 * @nset: signals to add or remove (if non-null)
3097 * @oset: previous value of signal mask if non-null
3099 * Some platforms have their own version with special arguments;
3100 * others support only sys_rt_sigprocmask.
3103 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
3104 old_sigset_t __user *, oset)
3106 old_sigset_t old_set, new_set;
3107 sigset_t new_blocked;
3109 old_set = current->blocked.sig[0];
3112 if (copy_from_user(&new_set, nset, sizeof(*nset)))
3114 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
3116 new_blocked = current->blocked;
3120 sigaddsetmask(&new_blocked, new_set);
3123 sigdelsetmask(&new_blocked, new_set);
3126 new_blocked.sig[0] = new_set;
3132 set_current_blocked(&new_blocked);
3136 if (copy_to_user(oset, &old_set, sizeof(*oset)))
3142 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
3144 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
3146 * sys_rt_sigaction - alter an action taken by a process
3147 * @sig: signal to be sent
3148 * @act: new sigaction
3149 * @oact: used to save the previous sigaction
3150 * @sigsetsize: size of sigset_t type
3152 SYSCALL_DEFINE4(rt_sigaction, int, sig,
3153 const struct sigaction __user *, act,
3154 struct sigaction __user *, oact,
3157 struct k_sigaction new_sa, old_sa;
3160 /* XXX: Don't preclude handling different sized sigset_t's. */
3161 if (sigsetsize != sizeof(sigset_t))
3165 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
3169 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
3172 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
3178 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
3180 #ifdef __ARCH_WANT_SYS_SGETMASK
3183 * For backwards compatibility. Functionality superseded by sigprocmask.
3185 SYSCALL_DEFINE0(sgetmask)
3188 return current->blocked.sig[0];
3191 SYSCALL_DEFINE1(ssetmask, int, newmask)
3193 int old = current->blocked.sig[0];
3196 siginitset(&newset, newmask & ~(sigmask(SIGKILL) | sigmask(SIGSTOP)));
3197 set_current_blocked(&newset);
3201 #endif /* __ARCH_WANT_SGETMASK */
3203 #ifdef __ARCH_WANT_SYS_SIGNAL
3205 * For backwards compatibility. Functionality superseded by sigaction.
3207 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
3209 struct k_sigaction new_sa, old_sa;
3212 new_sa.sa.sa_handler = handler;
3213 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
3214 sigemptyset(&new_sa.sa.sa_mask);
3216 ret = do_sigaction(sig, &new_sa, &old_sa);
3218 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
3220 #endif /* __ARCH_WANT_SYS_SIGNAL */
3222 #ifdef __ARCH_WANT_SYS_PAUSE
3224 SYSCALL_DEFINE0(pause)
3226 while (!signal_pending(current)) {
3227 current->state = TASK_INTERRUPTIBLE;
3230 return -ERESTARTNOHAND;
3235 #ifdef HAVE_SET_RESTORE_SIGMASK
3236 int sigsuspend(sigset_t *set)
3238 sigdelsetmask(set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3240 current->saved_sigmask = current->blocked;
3241 set_current_blocked(set);
3243 current->state = TASK_INTERRUPTIBLE;
3245 set_restore_sigmask();
3246 return -ERESTARTNOHAND;
3250 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
3252 * sys_rt_sigsuspend - replace the signal mask for a value with the
3253 * @unewset value until a signal is received
3254 * @unewset: new signal mask value
3255 * @sigsetsize: size of sigset_t type
3257 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
3261 /* XXX: Don't preclude handling different sized sigset_t's. */
3262 if (sigsetsize != sizeof(sigset_t))
3265 if (copy_from_user(&newset, unewset, sizeof(newset)))
3267 return sigsuspend(&newset);
3269 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
3271 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
3276 void __init signals_init(void)
3278 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
3281 #ifdef CONFIG_KGDB_KDB
3282 #include <linux/kdb.h>
3284 * kdb_send_sig_info - Allows kdb to send signals without exposing
3285 * signal internals. This function checks if the required locks are
3286 * available before calling the main signal code, to avoid kdb
3290 kdb_send_sig_info(struct task_struct *t, struct siginfo *info)
3292 static struct task_struct *kdb_prev_t;
3294 if (!spin_trylock(&t->sighand->siglock)) {
3295 kdb_printf("Can't do kill command now.\n"
3296 "The sigmask lock is held somewhere else in "
3297 "kernel, try again later\n");
3300 spin_unlock(&t->sighand->siglock);
3301 new_t = kdb_prev_t != t;
3303 if (t->state != TASK_RUNNING && new_t) {
3304 kdb_printf("Process is not RUNNING, sending a signal from "
3305 "kdb risks deadlock\n"
3306 "on the run queue locks. "
3307 "The signal has _not_ been sent.\n"
3308 "Reissue the kill command if you want to risk "
3312 sig = info->si_signo;
3313 if (send_sig_info(sig, info, t))
3314 kdb_printf("Fail to deliver Signal %d to process %d.\n",
3317 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
3319 #endif /* CONFIG_KGDB_KDB */