1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/signal.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
14 #include <linux/slab.h>
15 #include <linux/export.h>
16 #include <linux/init.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/sched/cputime.h>
23 #include <linux/file.h>
25 #include <linux/proc_fs.h>
26 #include <linux/tty.h>
27 #include <linux/binfmts.h>
28 #include <linux/coredump.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/ptrace.h>
32 #include <linux/signal.h>
33 #include <linux/signalfd.h>
34 #include <linux/ratelimit.h>
35 #include <linux/tracehook.h>
36 #include <linux/capability.h>
37 #include <linux/freezer.h>
38 #include <linux/pid_namespace.h>
39 #include <linux/nsproxy.h>
40 #include <linux/user_namespace.h>
41 #include <linux/uprobes.h>
42 #include <linux/compat.h>
43 #include <linux/cn_proc.h>
44 #include <linux/compiler.h>
45 #include <linux/posix-timers.h>
46 #include <linux/cgroup.h>
47 #include <linux/audit.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/signal.h>
52 #include <asm/param.h>
53 #include <linux/uaccess.h>
54 #include <asm/unistd.h>
55 #include <asm/siginfo.h>
56 #include <asm/cacheflush.h>
57 #include <asm/syscall.h> /* for syscall_get_* */
60 * SLAB caches for signal bits.
63 static struct kmem_cache *sigqueue_cachep;
65 int print_fatal_signals __read_mostly;
67 static void __user *sig_handler(struct task_struct *t, int sig)
69 return t->sighand->action[sig - 1].sa.sa_handler;
72 static inline bool sig_handler_ignored(void __user *handler, int sig)
74 /* Is it explicitly or implicitly ignored? */
75 return handler == SIG_IGN ||
76 (handler == SIG_DFL && sig_kernel_ignore(sig));
79 static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
83 handler = sig_handler(t, sig);
85 /* SIGKILL and SIGSTOP may not be sent to the global init */
86 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
89 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
90 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
93 /* Only allow kernel generated signals to this kthread */
94 if (unlikely((t->flags & PF_KTHREAD) &&
95 (handler == SIG_KTHREAD_KERNEL) && !force))
98 return sig_handler_ignored(handler, sig);
101 static bool sig_ignored(struct task_struct *t, int sig, bool force)
104 * Blocked signals are never ignored, since the
105 * signal handler may change by the time it is
108 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
112 * Tracers may want to know about even ignored signal unless it
113 * is SIGKILL which can't be reported anyway but can be ignored
114 * by SIGNAL_UNKILLABLE task.
116 if (t->ptrace && sig != SIGKILL)
119 return sig_task_ignored(t, sig, force);
123 * Re-calculate pending state from the set of locally pending
124 * signals, globally pending signals, and blocked signals.
126 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
131 switch (_NSIG_WORDS) {
133 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
134 ready |= signal->sig[i] &~ blocked->sig[i];
137 case 4: ready = signal->sig[3] &~ blocked->sig[3];
138 ready |= signal->sig[2] &~ blocked->sig[2];
139 ready |= signal->sig[1] &~ blocked->sig[1];
140 ready |= signal->sig[0] &~ blocked->sig[0];
143 case 2: ready = signal->sig[1] &~ blocked->sig[1];
144 ready |= signal->sig[0] &~ blocked->sig[0];
147 case 1: ready = signal->sig[0] &~ blocked->sig[0];
152 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
154 static bool recalc_sigpending_tsk(struct task_struct *t)
156 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
157 PENDING(&t->pending, &t->blocked) ||
158 PENDING(&t->signal->shared_pending, &t->blocked) ||
159 cgroup_task_frozen(t)) {
160 set_tsk_thread_flag(t, TIF_SIGPENDING);
165 * We must never clear the flag in another thread, or in current
166 * when it's possible the current syscall is returning -ERESTART*.
167 * So we don't clear it here, and only callers who know they should do.
173 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
174 * This is superfluous when called on current, the wakeup is a harmless no-op.
176 void recalc_sigpending_and_wake(struct task_struct *t)
178 if (recalc_sigpending_tsk(t))
179 signal_wake_up(t, 0);
182 void recalc_sigpending(void)
184 if (!recalc_sigpending_tsk(current) && !freezing(current))
185 clear_thread_flag(TIF_SIGPENDING);
188 EXPORT_SYMBOL(recalc_sigpending);
190 void calculate_sigpending(void)
192 /* Have any signals or users of TIF_SIGPENDING been delayed
195 spin_lock_irq(¤t->sighand->siglock);
196 set_tsk_thread_flag(current, TIF_SIGPENDING);
198 spin_unlock_irq(¤t->sighand->siglock);
201 /* Given the mask, find the first available signal that should be serviced. */
203 #define SYNCHRONOUS_MASK \
204 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
205 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
207 int next_signal(struct sigpending *pending, sigset_t *mask)
209 unsigned long i, *s, *m, x;
212 s = pending->signal.sig;
216 * Handle the first word specially: it contains the
217 * synchronous signals that need to be dequeued first.
221 if (x & SYNCHRONOUS_MASK)
222 x &= SYNCHRONOUS_MASK;
227 switch (_NSIG_WORDS) {
229 for (i = 1; i < _NSIG_WORDS; ++i) {
233 sig = ffz(~x) + i*_NSIG_BPW + 1;
242 sig = ffz(~x) + _NSIG_BPW + 1;
253 static inline void print_dropped_signal(int sig)
255 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
257 if (!print_fatal_signals)
260 if (!__ratelimit(&ratelimit_state))
263 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
264 current->comm, current->pid, sig);
268 * task_set_jobctl_pending - set jobctl pending bits
270 * @mask: pending bits to set
272 * Clear @mask from @task->jobctl. @mask must be subset of
273 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
274 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
275 * cleared. If @task is already being killed or exiting, this function
279 * Must be called with @task->sighand->siglock held.
282 * %true if @mask is set, %false if made noop because @task was dying.
284 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
286 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
287 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
288 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
290 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
293 if (mask & JOBCTL_STOP_SIGMASK)
294 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
296 task->jobctl |= mask;
301 * task_clear_jobctl_trapping - clear jobctl trapping bit
304 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
305 * Clear it and wake up the ptracer. Note that we don't need any further
306 * locking. @task->siglock guarantees that @task->parent points to the
310 * Must be called with @task->sighand->siglock held.
312 void task_clear_jobctl_trapping(struct task_struct *task)
314 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
315 task->jobctl &= ~JOBCTL_TRAPPING;
316 smp_mb(); /* advised by wake_up_bit() */
317 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
322 * task_clear_jobctl_pending - clear jobctl pending bits
324 * @mask: pending bits to clear
326 * Clear @mask from @task->jobctl. @mask must be subset of
327 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
328 * STOP bits are cleared together.
330 * If clearing of @mask leaves no stop or trap pending, this function calls
331 * task_clear_jobctl_trapping().
334 * Must be called with @task->sighand->siglock held.
336 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
338 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
340 if (mask & JOBCTL_STOP_PENDING)
341 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
343 task->jobctl &= ~mask;
345 if (!(task->jobctl & JOBCTL_PENDING_MASK))
346 task_clear_jobctl_trapping(task);
350 * task_participate_group_stop - participate in a group stop
351 * @task: task participating in a group stop
353 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
354 * Group stop states are cleared and the group stop count is consumed if
355 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
356 * stop, the appropriate `SIGNAL_*` flags are set.
359 * Must be called with @task->sighand->siglock held.
362 * %true if group stop completion should be notified to the parent, %false
365 static bool task_participate_group_stop(struct task_struct *task)
367 struct signal_struct *sig = task->signal;
368 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
370 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
372 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
377 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
378 sig->group_stop_count--;
381 * Tell the caller to notify completion iff we are entering into a
382 * fresh group stop. Read comment in do_signal_stop() for details.
384 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
385 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
391 void task_join_group_stop(struct task_struct *task)
393 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
394 struct signal_struct *sig = current->signal;
396 if (sig->group_stop_count) {
397 sig->group_stop_count++;
398 mask |= JOBCTL_STOP_CONSUME;
399 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
402 /* Have the new thread join an on-going signal group stop */
403 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
407 * allocate a new signal queue record
408 * - this may be called without locks if and only if t == current, otherwise an
409 * appropriate lock must be held to stop the target task from exiting
411 static struct sigqueue *
412 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
413 int override_rlimit, const unsigned int sigqueue_flags)
415 struct sigqueue *q = NULL;
416 struct ucounts *ucounts = NULL;
420 * Protect access to @t credentials. This can go away when all
421 * callers hold rcu read lock.
423 * NOTE! A pending signal will hold on to the user refcount,
424 * and we get/put the refcount only when the sigpending count
425 * changes from/to zero.
428 ucounts = task_ucounts(t);
429 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
434 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
435 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
437 print_dropped_signal(sig);
440 if (unlikely(q == NULL)) {
441 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
443 INIT_LIST_HEAD(&q->list);
444 q->flags = sigqueue_flags;
445 q->ucounts = ucounts;
450 static void __sigqueue_free(struct sigqueue *q)
452 if (q->flags & SIGQUEUE_PREALLOC)
455 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
458 kmem_cache_free(sigqueue_cachep, q);
461 void flush_sigqueue(struct sigpending *queue)
465 sigemptyset(&queue->signal);
466 while (!list_empty(&queue->list)) {
467 q = list_entry(queue->list.next, struct sigqueue , list);
468 list_del_init(&q->list);
474 * Flush all pending signals for this kthread.
476 void flush_signals(struct task_struct *t)
480 spin_lock_irqsave(&t->sighand->siglock, flags);
481 clear_tsk_thread_flag(t, TIF_SIGPENDING);
482 flush_sigqueue(&t->pending);
483 flush_sigqueue(&t->signal->shared_pending);
484 spin_unlock_irqrestore(&t->sighand->siglock, flags);
486 EXPORT_SYMBOL(flush_signals);
488 #ifdef CONFIG_POSIX_TIMERS
489 static void __flush_itimer_signals(struct sigpending *pending)
491 sigset_t signal, retain;
492 struct sigqueue *q, *n;
494 signal = pending->signal;
495 sigemptyset(&retain);
497 list_for_each_entry_safe(q, n, &pending->list, list) {
498 int sig = q->info.si_signo;
500 if (likely(q->info.si_code != SI_TIMER)) {
501 sigaddset(&retain, sig);
503 sigdelset(&signal, sig);
504 list_del_init(&q->list);
509 sigorsets(&pending->signal, &signal, &retain);
512 void flush_itimer_signals(void)
514 struct task_struct *tsk = current;
517 spin_lock_irqsave(&tsk->sighand->siglock, flags);
518 __flush_itimer_signals(&tsk->pending);
519 __flush_itimer_signals(&tsk->signal->shared_pending);
520 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
524 void ignore_signals(struct task_struct *t)
528 for (i = 0; i < _NSIG; ++i)
529 t->sighand->action[i].sa.sa_handler = SIG_IGN;
535 * Flush all handlers for a task.
539 flush_signal_handlers(struct task_struct *t, int force_default)
542 struct k_sigaction *ka = &t->sighand->action[0];
543 for (i = _NSIG ; i != 0 ; i--) {
544 if (force_default || ka->sa.sa_handler != SIG_IGN)
545 ka->sa.sa_handler = SIG_DFL;
547 #ifdef __ARCH_HAS_SA_RESTORER
548 ka->sa.sa_restorer = NULL;
550 sigemptyset(&ka->sa.sa_mask);
555 bool unhandled_signal(struct task_struct *tsk, int sig)
557 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
558 if (is_global_init(tsk))
561 if (handler != SIG_IGN && handler != SIG_DFL)
564 /* if ptraced, let the tracer determine */
568 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
571 struct sigqueue *q, *first = NULL;
574 * Collect the siginfo appropriate to this signal. Check if
575 * there is another siginfo for the same signal.
577 list_for_each_entry(q, &list->list, list) {
578 if (q->info.si_signo == sig) {
585 sigdelset(&list->signal, sig);
589 list_del_init(&first->list);
590 copy_siginfo(info, &first->info);
593 (first->flags & SIGQUEUE_PREALLOC) &&
594 (info->si_code == SI_TIMER) &&
595 (info->si_sys_private);
597 __sigqueue_free(first);
600 * Ok, it wasn't in the queue. This must be
601 * a fast-pathed signal or we must have been
602 * out of queue space. So zero out the info.
605 info->si_signo = sig;
607 info->si_code = SI_USER;
613 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
614 kernel_siginfo_t *info, bool *resched_timer)
616 int sig = next_signal(pending, mask);
619 collect_signal(sig, pending, info, resched_timer);
624 * Dequeue a signal and return the element to the caller, which is
625 * expected to free it.
627 * All callers have to hold the siglock.
629 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
631 bool resched_timer = false;
634 /* We only dequeue private signals from ourselves, we don't let
635 * signalfd steal them
637 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
639 signr = __dequeue_signal(&tsk->signal->shared_pending,
640 mask, info, &resched_timer);
641 #ifdef CONFIG_POSIX_TIMERS
645 * itimers are process shared and we restart periodic
646 * itimers in the signal delivery path to prevent DoS
647 * attacks in the high resolution timer case. This is
648 * compliant with the old way of self-restarting
649 * itimers, as the SIGALRM is a legacy signal and only
650 * queued once. Changing the restart behaviour to
651 * restart the timer in the signal dequeue path is
652 * reducing the timer noise on heavy loaded !highres
655 if (unlikely(signr == SIGALRM)) {
656 struct hrtimer *tmr = &tsk->signal->real_timer;
658 if (!hrtimer_is_queued(tmr) &&
659 tsk->signal->it_real_incr != 0) {
660 hrtimer_forward(tmr, tmr->base->get_time(),
661 tsk->signal->it_real_incr);
662 hrtimer_restart(tmr);
672 if (unlikely(sig_kernel_stop(signr))) {
674 * Set a marker that we have dequeued a stop signal. Our
675 * caller might release the siglock and then the pending
676 * stop signal it is about to process is no longer in the
677 * pending bitmasks, but must still be cleared by a SIGCONT
678 * (and overruled by a SIGKILL). So those cases clear this
679 * shared flag after we've set it. Note that this flag may
680 * remain set after the signal we return is ignored or
681 * handled. That doesn't matter because its only purpose
682 * is to alert stop-signal processing code when another
683 * processor has come along and cleared the flag.
685 current->jobctl |= JOBCTL_STOP_DEQUEUED;
687 #ifdef CONFIG_POSIX_TIMERS
690 * Release the siglock to ensure proper locking order
691 * of timer locks outside of siglocks. Note, we leave
692 * irqs disabled here, since the posix-timers code is
693 * about to disable them again anyway.
695 spin_unlock(&tsk->sighand->siglock);
696 posixtimer_rearm(info);
697 spin_lock(&tsk->sighand->siglock);
699 /* Don't expose the si_sys_private value to userspace */
700 info->si_sys_private = 0;
705 EXPORT_SYMBOL_GPL(dequeue_signal);
707 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
709 struct task_struct *tsk = current;
710 struct sigpending *pending = &tsk->pending;
711 struct sigqueue *q, *sync = NULL;
714 * Might a synchronous signal be in the queue?
716 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
720 * Return the first synchronous signal in the queue.
722 list_for_each_entry(q, &pending->list, list) {
723 /* Synchronous signals have a positive si_code */
724 if ((q->info.si_code > SI_USER) &&
725 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
733 * Check if there is another siginfo for the same signal.
735 list_for_each_entry_continue(q, &pending->list, list) {
736 if (q->info.si_signo == sync->info.si_signo)
740 sigdelset(&pending->signal, sync->info.si_signo);
743 list_del_init(&sync->list);
744 copy_siginfo(info, &sync->info);
745 __sigqueue_free(sync);
746 return info->si_signo;
750 * Tell a process that it has a new active signal..
752 * NOTE! we rely on the previous spin_lock to
753 * lock interrupts for us! We can only be called with
754 * "siglock" held, and the local interrupt must
755 * have been disabled when that got acquired!
757 * No need to set need_resched since signal event passing
758 * goes through ->blocked
760 void signal_wake_up_state(struct task_struct *t, unsigned int state)
762 set_tsk_thread_flag(t, TIF_SIGPENDING);
764 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
765 * case. We don't check t->state here because there is a race with it
766 * executing another processor and just now entering stopped state.
767 * By using wake_up_state, we ensure the process will wake up and
768 * handle its death signal.
770 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
775 * Remove signals in mask from the pending set and queue.
776 * Returns 1 if any signals were found.
778 * All callers must be holding the siglock.
780 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
782 struct sigqueue *q, *n;
785 sigandsets(&m, mask, &s->signal);
786 if (sigisemptyset(&m))
789 sigandnsets(&s->signal, &s->signal, mask);
790 list_for_each_entry_safe(q, n, &s->list, list) {
791 if (sigismember(mask, q->info.si_signo)) {
792 list_del_init(&q->list);
798 static inline int is_si_special(const struct kernel_siginfo *info)
800 return info <= SEND_SIG_PRIV;
803 static inline bool si_fromuser(const struct kernel_siginfo *info)
805 return info == SEND_SIG_NOINFO ||
806 (!is_si_special(info) && SI_FROMUSER(info));
810 * called with RCU read lock from check_kill_permission()
812 static bool kill_ok_by_cred(struct task_struct *t)
814 const struct cred *cred = current_cred();
815 const struct cred *tcred = __task_cred(t);
817 return uid_eq(cred->euid, tcred->suid) ||
818 uid_eq(cred->euid, tcred->uid) ||
819 uid_eq(cred->uid, tcred->suid) ||
820 uid_eq(cred->uid, tcred->uid) ||
821 ns_capable(tcred->user_ns, CAP_KILL);
825 * Bad permissions for sending the signal
826 * - the caller must hold the RCU read lock
828 static int check_kill_permission(int sig, struct kernel_siginfo *info,
829 struct task_struct *t)
834 if (!valid_signal(sig))
837 if (!si_fromuser(info))
840 error = audit_signal_info(sig, t); /* Let audit system see the signal */
844 if (!same_thread_group(current, t) &&
845 !kill_ok_by_cred(t)) {
848 sid = task_session(t);
850 * We don't return the error if sid == NULL. The
851 * task was unhashed, the caller must notice this.
853 if (!sid || sid == task_session(current))
861 return security_task_kill(t, info, sig, NULL);
865 * ptrace_trap_notify - schedule trap to notify ptracer
866 * @t: tracee wanting to notify tracer
868 * This function schedules sticky ptrace trap which is cleared on the next
869 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
872 * If @t is running, STOP trap will be taken. If trapped for STOP and
873 * ptracer is listening for events, tracee is woken up so that it can
874 * re-trap for the new event. If trapped otherwise, STOP trap will be
875 * eventually taken without returning to userland after the existing traps
876 * are finished by PTRACE_CONT.
879 * Must be called with @task->sighand->siglock held.
881 static void ptrace_trap_notify(struct task_struct *t)
883 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
884 assert_spin_locked(&t->sighand->siglock);
886 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
887 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
891 * Handle magic process-wide effects of stop/continue signals. Unlike
892 * the signal actions, these happen immediately at signal-generation
893 * time regardless of blocking, ignoring, or handling. This does the
894 * actual continuing for SIGCONT, but not the actual stopping for stop
895 * signals. The process stop is done as a signal action for SIG_DFL.
897 * Returns true if the signal should be actually delivered, otherwise
898 * it should be dropped.
900 static bool prepare_signal(int sig, struct task_struct *p, bool force)
902 struct signal_struct *signal = p->signal;
903 struct task_struct *t;
906 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
907 if (!(signal->flags & SIGNAL_GROUP_EXIT))
908 return sig == SIGKILL;
910 * The process is in the middle of dying, nothing to do.
912 } else if (sig_kernel_stop(sig)) {
914 * This is a stop signal. Remove SIGCONT from all queues.
916 siginitset(&flush, sigmask(SIGCONT));
917 flush_sigqueue_mask(&flush, &signal->shared_pending);
918 for_each_thread(p, t)
919 flush_sigqueue_mask(&flush, &t->pending);
920 } else if (sig == SIGCONT) {
923 * Remove all stop signals from all queues, wake all threads.
925 siginitset(&flush, SIG_KERNEL_STOP_MASK);
926 flush_sigqueue_mask(&flush, &signal->shared_pending);
927 for_each_thread(p, t) {
928 flush_sigqueue_mask(&flush, &t->pending);
929 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
930 if (likely(!(t->ptrace & PT_SEIZED)))
931 wake_up_state(t, __TASK_STOPPED);
933 ptrace_trap_notify(t);
937 * Notify the parent with CLD_CONTINUED if we were stopped.
939 * If we were in the middle of a group stop, we pretend it
940 * was already finished, and then continued. Since SIGCHLD
941 * doesn't queue we report only CLD_STOPPED, as if the next
942 * CLD_CONTINUED was dropped.
945 if (signal->flags & SIGNAL_STOP_STOPPED)
946 why |= SIGNAL_CLD_CONTINUED;
947 else if (signal->group_stop_count)
948 why |= SIGNAL_CLD_STOPPED;
952 * The first thread which returns from do_signal_stop()
953 * will take ->siglock, notice SIGNAL_CLD_MASK, and
954 * notify its parent. See get_signal().
956 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
957 signal->group_stop_count = 0;
958 signal->group_exit_code = 0;
962 return !sig_ignored(p, sig, force);
966 * Test if P wants to take SIG. After we've checked all threads with this,
967 * it's equivalent to finding no threads not blocking SIG. Any threads not
968 * blocking SIG were ruled out because they are not running and already
969 * have pending signals. Such threads will dequeue from the shared queue
970 * as soon as they're available, so putting the signal on the shared queue
971 * will be equivalent to sending it to one such thread.
973 static inline bool wants_signal(int sig, struct task_struct *p)
975 if (sigismember(&p->blocked, sig))
978 if (p->flags & PF_EXITING)
984 if (task_is_stopped_or_traced(p))
987 return task_curr(p) || !task_sigpending(p);
990 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
992 struct signal_struct *signal = p->signal;
993 struct task_struct *t;
996 * Now find a thread we can wake up to take the signal off the queue.
998 * If the main thread wants the signal, it gets first crack.
999 * Probably the least surprising to the average bear.
1001 if (wants_signal(sig, p))
1003 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1005 * There is just one thread and it does not need to be woken.
1006 * It will dequeue unblocked signals before it runs again.
1011 * Otherwise try to find a suitable thread.
1013 t = signal->curr_target;
1014 while (!wants_signal(sig, t)) {
1016 if (t == signal->curr_target)
1018 * No thread needs to be woken.
1019 * Any eligible threads will see
1020 * the signal in the queue soon.
1024 signal->curr_target = t;
1028 * Found a killable thread. If the signal will be fatal,
1029 * then start taking the whole group down immediately.
1031 if (sig_fatal(p, sig) &&
1032 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1033 !sigismember(&t->real_blocked, sig) &&
1034 (sig == SIGKILL || !p->ptrace)) {
1036 * This signal will be fatal to the whole group.
1038 if (!sig_kernel_coredump(sig)) {
1040 * Start a group exit and wake everybody up.
1041 * This way we don't have other threads
1042 * running and doing things after a slower
1043 * thread has the fatal signal pending.
1045 signal->flags = SIGNAL_GROUP_EXIT;
1046 signal->group_exit_code = sig;
1047 signal->group_stop_count = 0;
1050 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1051 sigaddset(&t->pending.signal, SIGKILL);
1052 signal_wake_up(t, 1);
1053 } while_each_thread(p, t);
1059 * The signal is already in the shared-pending queue.
1060 * Tell the chosen thread to wake up and dequeue it.
1062 signal_wake_up(t, sig == SIGKILL);
1066 static inline bool legacy_queue(struct sigpending *signals, int sig)
1068 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1071 static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1072 enum pid_type type, bool force)
1074 struct sigpending *pending;
1076 int override_rlimit;
1077 int ret = 0, result;
1079 assert_spin_locked(&t->sighand->siglock);
1081 result = TRACE_SIGNAL_IGNORED;
1082 if (!prepare_signal(sig, t, force))
1085 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1087 * Short-circuit ignored signals and support queuing
1088 * exactly one non-rt signal, so that we can get more
1089 * detailed information about the cause of the signal.
1091 result = TRACE_SIGNAL_ALREADY_PENDING;
1092 if (legacy_queue(pending, sig))
1095 result = TRACE_SIGNAL_DELIVERED;
1097 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1099 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1103 * Real-time signals must be queued if sent by sigqueue, or
1104 * some other real-time mechanism. It is implementation
1105 * defined whether kill() does so. We attempt to do so, on
1106 * the principle of least surprise, but since kill is not
1107 * allowed to fail with EAGAIN when low on memory we just
1108 * make sure at least one signal gets delivered and don't
1109 * pass on the info struct.
1112 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1114 override_rlimit = 0;
1116 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1119 list_add_tail(&q->list, &pending->list);
1120 switch ((unsigned long) info) {
1121 case (unsigned long) SEND_SIG_NOINFO:
1122 clear_siginfo(&q->info);
1123 q->info.si_signo = sig;
1124 q->info.si_errno = 0;
1125 q->info.si_code = SI_USER;
1126 q->info.si_pid = task_tgid_nr_ns(current,
1127 task_active_pid_ns(t));
1130 from_kuid_munged(task_cred_xxx(t, user_ns),
1134 case (unsigned long) SEND_SIG_PRIV:
1135 clear_siginfo(&q->info);
1136 q->info.si_signo = sig;
1137 q->info.si_errno = 0;
1138 q->info.si_code = SI_KERNEL;
1143 copy_siginfo(&q->info, info);
1146 } else if (!is_si_special(info) &&
1147 sig >= SIGRTMIN && info->si_code != SI_USER) {
1149 * Queue overflow, abort. We may abort if the
1150 * signal was rt and sent by user using something
1151 * other than kill().
1153 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1158 * This is a silent loss of information. We still
1159 * send the signal, but the *info bits are lost.
1161 result = TRACE_SIGNAL_LOSE_INFO;
1165 signalfd_notify(t, sig);
1166 sigaddset(&pending->signal, sig);
1168 /* Let multiprocess signals appear after on-going forks */
1169 if (type > PIDTYPE_TGID) {
1170 struct multiprocess_signals *delayed;
1171 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1172 sigset_t *signal = &delayed->signal;
1173 /* Can't queue both a stop and a continue signal */
1175 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1176 else if (sig_kernel_stop(sig))
1177 sigdelset(signal, SIGCONT);
1178 sigaddset(signal, sig);
1182 complete_signal(sig, t, type);
1184 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1188 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1191 switch (siginfo_layout(info->si_signo, info->si_code)) {
1200 case SIL_FAULT_TRAPNO:
1201 case SIL_FAULT_MCEERR:
1202 case SIL_FAULT_BNDERR:
1203 case SIL_FAULT_PKUERR:
1204 case SIL_FAULT_PERF_EVENT:
1212 static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1215 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1218 if (info == SEND_SIG_NOINFO) {
1219 /* Force if sent from an ancestor pid namespace */
1220 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1221 } else if (info == SEND_SIG_PRIV) {
1222 /* Don't ignore kernel generated signals */
1224 } else if (has_si_pid_and_uid(info)) {
1225 /* SIGKILL and SIGSTOP is special or has ids */
1226 struct user_namespace *t_user_ns;
1229 t_user_ns = task_cred_xxx(t, user_ns);
1230 if (current_user_ns() != t_user_ns) {
1231 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1232 info->si_uid = from_kuid_munged(t_user_ns, uid);
1236 /* A kernel generated signal? */
1237 force = (info->si_code == SI_KERNEL);
1239 /* From an ancestor pid namespace? */
1240 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1245 return __send_signal(sig, info, t, type, force);
1248 static void print_fatal_signal(int signr)
1250 struct pt_regs *regs = signal_pt_regs();
1251 pr_info("potentially unexpected fatal signal %d.\n", signr);
1253 #if defined(__i386__) && !defined(__arch_um__)
1254 pr_info("code at %08lx: ", regs->ip);
1257 for (i = 0; i < 16; i++) {
1260 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1262 pr_cont("%02x ", insn);
1272 static int __init setup_print_fatal_signals(char *str)
1274 get_option (&str, &print_fatal_signals);
1279 __setup("print-fatal-signals=", setup_print_fatal_signals);
1282 __group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1284 return send_signal(sig, info, p, PIDTYPE_TGID);
1287 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1290 unsigned long flags;
1293 if (lock_task_sighand(p, &flags)) {
1294 ret = send_signal(sig, info, p, type);
1295 unlock_task_sighand(p, &flags);
1302 * Force a signal that the process can't ignore: if necessary
1303 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1305 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1306 * since we do not want to have a signal handler that was blocked
1307 * be invoked when user space had explicitly blocked it.
1309 * We don't want to have recursive SIGSEGV's etc, for example,
1310 * that is why we also clear SIGNAL_UNKILLABLE.
1313 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t, bool sigdfl)
1315 unsigned long int flags;
1316 int ret, blocked, ignored;
1317 struct k_sigaction *action;
1318 int sig = info->si_signo;
1320 spin_lock_irqsave(&t->sighand->siglock, flags);
1321 action = &t->sighand->action[sig-1];
1322 ignored = action->sa.sa_handler == SIG_IGN;
1323 blocked = sigismember(&t->blocked, sig);
1324 if (blocked || ignored || sigdfl) {
1325 action->sa.sa_handler = SIG_DFL;
1327 sigdelset(&t->blocked, sig);
1328 recalc_sigpending_and_wake(t);
1332 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1333 * debugging to leave init killable.
1335 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1336 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1337 ret = send_signal(sig, info, t, PIDTYPE_PID);
1338 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1343 int force_sig_info(struct kernel_siginfo *info)
1345 return force_sig_info_to_task(info, current, false);
1349 * Nuke all other threads in the group.
1351 int zap_other_threads(struct task_struct *p)
1353 struct task_struct *t = p;
1356 p->signal->group_stop_count = 0;
1358 while_each_thread(p, t) {
1359 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1362 /* Don't bother with already dead threads */
1365 sigaddset(&t->pending.signal, SIGKILL);
1366 signal_wake_up(t, 1);
1372 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1373 unsigned long *flags)
1375 struct sighand_struct *sighand;
1379 sighand = rcu_dereference(tsk->sighand);
1380 if (unlikely(sighand == NULL))
1384 * This sighand can be already freed and even reused, but
1385 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1386 * initializes ->siglock: this slab can't go away, it has
1387 * the same object type, ->siglock can't be reinitialized.
1389 * We need to ensure that tsk->sighand is still the same
1390 * after we take the lock, we can race with de_thread() or
1391 * __exit_signal(). In the latter case the next iteration
1392 * must see ->sighand == NULL.
1394 spin_lock_irqsave(&sighand->siglock, *flags);
1395 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1397 spin_unlock_irqrestore(&sighand->siglock, *flags);
1404 #ifdef CONFIG_LOCKDEP
1405 void lockdep_assert_task_sighand_held(struct task_struct *task)
1407 struct sighand_struct *sighand;
1410 sighand = rcu_dereference(task->sighand);
1412 lockdep_assert_held(&sighand->siglock);
1420 * send signal info to all the members of a group
1422 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1423 struct task_struct *p, enum pid_type type)
1428 ret = check_kill_permission(sig, info, p);
1432 ret = do_send_sig_info(sig, info, p, type);
1438 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1439 * control characters do (^C, ^Z etc)
1440 * - the caller must hold at least a readlock on tasklist_lock
1442 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1444 struct task_struct *p = NULL;
1445 int retval, success;
1449 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1450 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1453 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1454 return success ? 0 : retval;
1457 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1460 struct task_struct *p;
1464 p = pid_task(pid, PIDTYPE_PID);
1466 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1468 if (likely(!p || error != -ESRCH))
1472 * The task was unhashed in between, try again. If it
1473 * is dead, pid_task() will return NULL, if we race with
1474 * de_thread() it will find the new leader.
1479 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1483 error = kill_pid_info(sig, info, find_vpid(pid));
1488 static inline bool kill_as_cred_perm(const struct cred *cred,
1489 struct task_struct *target)
1491 const struct cred *pcred = __task_cred(target);
1493 return uid_eq(cred->euid, pcred->suid) ||
1494 uid_eq(cred->euid, pcred->uid) ||
1495 uid_eq(cred->uid, pcred->suid) ||
1496 uid_eq(cred->uid, pcred->uid);
1500 * The usb asyncio usage of siginfo is wrong. The glibc support
1501 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1502 * AKA after the generic fields:
1503 * kernel_pid_t si_pid;
1504 * kernel_uid32_t si_uid;
1505 * sigval_t si_value;
1507 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1508 * after the generic fields is:
1509 * void __user *si_addr;
1511 * This is a practical problem when there is a 64bit big endian kernel
1512 * and a 32bit userspace. As the 32bit address will encoded in the low
1513 * 32bits of the pointer. Those low 32bits will be stored at higher
1514 * address than appear in a 32 bit pointer. So userspace will not
1515 * see the address it was expecting for it's completions.
1517 * There is nothing in the encoding that can allow
1518 * copy_siginfo_to_user32 to detect this confusion of formats, so
1519 * handle this by requiring the caller of kill_pid_usb_asyncio to
1520 * notice when this situration takes place and to store the 32bit
1521 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1524 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1525 struct pid *pid, const struct cred *cred)
1527 struct kernel_siginfo info;
1528 struct task_struct *p;
1529 unsigned long flags;
1532 if (!valid_signal(sig))
1535 clear_siginfo(&info);
1536 info.si_signo = sig;
1537 info.si_errno = errno;
1538 info.si_code = SI_ASYNCIO;
1539 *((sigval_t *)&info.si_pid) = addr;
1542 p = pid_task(pid, PIDTYPE_PID);
1547 if (!kill_as_cred_perm(cred, p)) {
1551 ret = security_task_kill(p, &info, sig, cred);
1556 if (lock_task_sighand(p, &flags)) {
1557 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1558 unlock_task_sighand(p, &flags);
1566 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1569 * kill_something_info() interprets pid in interesting ways just like kill(2).
1571 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1572 * is probably wrong. Should make it like BSD or SYSV.
1575 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1580 return kill_proc_info(sig, info, pid);
1582 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1586 read_lock(&tasklist_lock);
1588 ret = __kill_pgrp_info(sig, info,
1589 pid ? find_vpid(-pid) : task_pgrp(current));
1591 int retval = 0, count = 0;
1592 struct task_struct * p;
1594 for_each_process(p) {
1595 if (task_pid_vnr(p) > 1 &&
1596 !same_thread_group(p, current)) {
1597 int err = group_send_sig_info(sig, info, p,
1604 ret = count ? retval : -ESRCH;
1606 read_unlock(&tasklist_lock);
1612 * These are for backward compatibility with the rest of the kernel source.
1615 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1618 * Make sure legacy kernel users don't send in bad values
1619 * (normal paths check this in check_kill_permission).
1621 if (!valid_signal(sig))
1624 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1626 EXPORT_SYMBOL(send_sig_info);
1628 #define __si_special(priv) \
1629 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1632 send_sig(int sig, struct task_struct *p, int priv)
1634 return send_sig_info(sig, __si_special(priv), p);
1636 EXPORT_SYMBOL(send_sig);
1638 void force_sig(int sig)
1640 struct kernel_siginfo info;
1642 clear_siginfo(&info);
1643 info.si_signo = sig;
1645 info.si_code = SI_KERNEL;
1648 force_sig_info(&info);
1650 EXPORT_SYMBOL(force_sig);
1653 * When things go south during signal handling, we
1654 * will force a SIGSEGV. And if the signal that caused
1655 * the problem was already a SIGSEGV, we'll want to
1656 * make sure we don't even try to deliver the signal..
1658 void force_sigsegv(int sig)
1660 struct task_struct *p = current;
1662 if (sig == SIGSEGV) {
1663 unsigned long flags;
1664 spin_lock_irqsave(&p->sighand->siglock, flags);
1665 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1666 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1671 int force_sig_fault_to_task(int sig, int code, void __user *addr
1672 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1673 , struct task_struct *t)
1675 struct kernel_siginfo info;
1677 clear_siginfo(&info);
1678 info.si_signo = sig;
1680 info.si_code = code;
1681 info.si_addr = addr;
1684 info.si_flags = flags;
1687 return force_sig_info_to_task(&info, t, false);
1690 int force_sig_fault(int sig, int code, void __user *addr
1691 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1693 return force_sig_fault_to_task(sig, code, addr
1694 ___ARCH_SI_IA64(imm, flags, isr), current);
1697 int send_sig_fault(int sig, int code, void __user *addr
1698 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1699 , struct task_struct *t)
1701 struct kernel_siginfo info;
1703 clear_siginfo(&info);
1704 info.si_signo = sig;
1706 info.si_code = code;
1707 info.si_addr = addr;
1710 info.si_flags = flags;
1713 return send_sig_info(info.si_signo, &info, t);
1716 int force_sig_mceerr(int code, void __user *addr, short lsb)
1718 struct kernel_siginfo info;
1720 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1721 clear_siginfo(&info);
1722 info.si_signo = SIGBUS;
1724 info.si_code = code;
1725 info.si_addr = addr;
1726 info.si_addr_lsb = lsb;
1727 return force_sig_info(&info);
1730 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1732 struct kernel_siginfo info;
1734 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1735 clear_siginfo(&info);
1736 info.si_signo = SIGBUS;
1738 info.si_code = code;
1739 info.si_addr = addr;
1740 info.si_addr_lsb = lsb;
1741 return send_sig_info(info.si_signo, &info, t);
1743 EXPORT_SYMBOL(send_sig_mceerr);
1745 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1747 struct kernel_siginfo info;
1749 clear_siginfo(&info);
1750 info.si_signo = SIGSEGV;
1752 info.si_code = SEGV_BNDERR;
1753 info.si_addr = addr;
1754 info.si_lower = lower;
1755 info.si_upper = upper;
1756 return force_sig_info(&info);
1760 int force_sig_pkuerr(void __user *addr, u32 pkey)
1762 struct kernel_siginfo info;
1764 clear_siginfo(&info);
1765 info.si_signo = SIGSEGV;
1767 info.si_code = SEGV_PKUERR;
1768 info.si_addr = addr;
1769 info.si_pkey = pkey;
1770 return force_sig_info(&info);
1774 int force_sig_perf(void __user *addr, u32 type, u64 sig_data)
1776 struct kernel_siginfo info;
1778 clear_siginfo(&info);
1779 info.si_signo = SIGTRAP;
1781 info.si_code = TRAP_PERF;
1782 info.si_addr = addr;
1783 info.si_perf_data = sig_data;
1784 info.si_perf_type = type;
1786 return force_sig_info(&info);
1790 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1791 * @syscall: syscall number to send to userland
1792 * @reason: filter-supplied reason code to send to userland (via si_errno)
1794 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1796 int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1798 struct kernel_siginfo info;
1800 clear_siginfo(&info);
1801 info.si_signo = SIGSYS;
1802 info.si_code = SYS_SECCOMP;
1803 info.si_call_addr = (void __user *)KSTK_EIP(current);
1804 info.si_errno = reason;
1805 info.si_arch = syscall_get_arch(current);
1806 info.si_syscall = syscall;
1807 return force_sig_info_to_task(&info, current, force_coredump);
1810 /* For the crazy architectures that include trap information in
1811 * the errno field, instead of an actual errno value.
1813 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1815 struct kernel_siginfo info;
1817 clear_siginfo(&info);
1818 info.si_signo = SIGTRAP;
1819 info.si_errno = errno;
1820 info.si_code = TRAP_HWBKPT;
1821 info.si_addr = addr;
1822 return force_sig_info(&info);
1825 /* For the rare architectures that include trap information using
1828 int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1830 struct kernel_siginfo info;
1832 clear_siginfo(&info);
1833 info.si_signo = sig;
1835 info.si_code = code;
1836 info.si_addr = addr;
1837 info.si_trapno = trapno;
1838 return force_sig_info(&info);
1841 /* For the rare architectures that include trap information using
1844 int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1845 struct task_struct *t)
1847 struct kernel_siginfo info;
1849 clear_siginfo(&info);
1850 info.si_signo = sig;
1852 info.si_code = code;
1853 info.si_addr = addr;
1854 info.si_trapno = trapno;
1855 return send_sig_info(info.si_signo, &info, t);
1858 int kill_pgrp(struct pid *pid, int sig, int priv)
1862 read_lock(&tasklist_lock);
1863 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1864 read_unlock(&tasklist_lock);
1868 EXPORT_SYMBOL(kill_pgrp);
1870 int kill_pid(struct pid *pid, int sig, int priv)
1872 return kill_pid_info(sig, __si_special(priv), pid);
1874 EXPORT_SYMBOL(kill_pid);
1877 * These functions support sending signals using preallocated sigqueue
1878 * structures. This is needed "because realtime applications cannot
1879 * afford to lose notifications of asynchronous events, like timer
1880 * expirations or I/O completions". In the case of POSIX Timers
1881 * we allocate the sigqueue structure from the timer_create. If this
1882 * allocation fails we are able to report the failure to the application
1883 * with an EAGAIN error.
1885 struct sigqueue *sigqueue_alloc(void)
1887 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1890 void sigqueue_free(struct sigqueue *q)
1892 unsigned long flags;
1893 spinlock_t *lock = ¤t->sighand->siglock;
1895 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1897 * We must hold ->siglock while testing q->list
1898 * to serialize with collect_signal() or with
1899 * __exit_signal()->flush_sigqueue().
1901 spin_lock_irqsave(lock, flags);
1902 q->flags &= ~SIGQUEUE_PREALLOC;
1904 * If it is queued it will be freed when dequeued,
1905 * like the "regular" sigqueue.
1907 if (!list_empty(&q->list))
1909 spin_unlock_irqrestore(lock, flags);
1915 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1917 int sig = q->info.si_signo;
1918 struct sigpending *pending;
1919 struct task_struct *t;
1920 unsigned long flags;
1923 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1927 t = pid_task(pid, type);
1928 if (!t || !likely(lock_task_sighand(t, &flags)))
1931 ret = 1; /* the signal is ignored */
1932 result = TRACE_SIGNAL_IGNORED;
1933 if (!prepare_signal(sig, t, false))
1937 if (unlikely(!list_empty(&q->list))) {
1939 * If an SI_TIMER entry is already queue just increment
1940 * the overrun count.
1942 BUG_ON(q->info.si_code != SI_TIMER);
1943 q->info.si_overrun++;
1944 result = TRACE_SIGNAL_ALREADY_PENDING;
1947 q->info.si_overrun = 0;
1949 signalfd_notify(t, sig);
1950 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1951 list_add_tail(&q->list, &pending->list);
1952 sigaddset(&pending->signal, sig);
1953 complete_signal(sig, t, type);
1954 result = TRACE_SIGNAL_DELIVERED;
1956 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1957 unlock_task_sighand(t, &flags);
1963 static void do_notify_pidfd(struct task_struct *task)
1967 WARN_ON(task->exit_state == 0);
1968 pid = task_pid(task);
1969 wake_up_all(&pid->wait_pidfd);
1973 * Let a parent know about the death of a child.
1974 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1976 * Returns true if our parent ignored us and so we've switched to
1979 bool do_notify_parent(struct task_struct *tsk, int sig)
1981 struct kernel_siginfo info;
1982 unsigned long flags;
1983 struct sighand_struct *psig;
1984 bool autoreap = false;
1989 /* do_notify_parent_cldstop should have been called instead. */
1990 BUG_ON(task_is_stopped_or_traced(tsk));
1992 BUG_ON(!tsk->ptrace &&
1993 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1995 /* Wake up all pidfd waiters */
1996 do_notify_pidfd(tsk);
1998 if (sig != SIGCHLD) {
2000 * This is only possible if parent == real_parent.
2001 * Check if it has changed security domain.
2003 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2007 clear_siginfo(&info);
2008 info.si_signo = sig;
2011 * We are under tasklist_lock here so our parent is tied to
2012 * us and cannot change.
2014 * task_active_pid_ns will always return the same pid namespace
2015 * until a task passes through release_task.
2017 * write_lock() currently calls preempt_disable() which is the
2018 * same as rcu_read_lock(), but according to Oleg, this is not
2019 * correct to rely on this
2022 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2023 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2027 task_cputime(tsk, &utime, &stime);
2028 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2029 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2031 info.si_status = tsk->exit_code & 0x7f;
2032 if (tsk->exit_code & 0x80)
2033 info.si_code = CLD_DUMPED;
2034 else if (tsk->exit_code & 0x7f)
2035 info.si_code = CLD_KILLED;
2037 info.si_code = CLD_EXITED;
2038 info.si_status = tsk->exit_code >> 8;
2041 psig = tsk->parent->sighand;
2042 spin_lock_irqsave(&psig->siglock, flags);
2043 if (!tsk->ptrace && sig == SIGCHLD &&
2044 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2045 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2047 * We are exiting and our parent doesn't care. POSIX.1
2048 * defines special semantics for setting SIGCHLD to SIG_IGN
2049 * or setting the SA_NOCLDWAIT flag: we should be reaped
2050 * automatically and not left for our parent's wait4 call.
2051 * Rather than having the parent do it as a magic kind of
2052 * signal handler, we just set this to tell do_exit that we
2053 * can be cleaned up without becoming a zombie. Note that
2054 * we still call __wake_up_parent in this case, because a
2055 * blocked sys_wait4 might now return -ECHILD.
2057 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2058 * is implementation-defined: we do (if you don't want
2059 * it, just use SIG_IGN instead).
2062 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2066 * Send with __send_signal as si_pid and si_uid are in the
2067 * parent's namespaces.
2069 if (valid_signal(sig) && sig)
2070 __send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2071 __wake_up_parent(tsk, tsk->parent);
2072 spin_unlock_irqrestore(&psig->siglock, flags);
2078 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2079 * @tsk: task reporting the state change
2080 * @for_ptracer: the notification is for ptracer
2081 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2083 * Notify @tsk's parent that the stopped/continued state has changed. If
2084 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2085 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2088 * Must be called with tasklist_lock at least read locked.
2090 static void do_notify_parent_cldstop(struct task_struct *tsk,
2091 bool for_ptracer, int why)
2093 struct kernel_siginfo info;
2094 unsigned long flags;
2095 struct task_struct *parent;
2096 struct sighand_struct *sighand;
2100 parent = tsk->parent;
2102 tsk = tsk->group_leader;
2103 parent = tsk->real_parent;
2106 clear_siginfo(&info);
2107 info.si_signo = SIGCHLD;
2110 * see comment in do_notify_parent() about the following 4 lines
2113 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2114 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2117 task_cputime(tsk, &utime, &stime);
2118 info.si_utime = nsec_to_clock_t(utime);
2119 info.si_stime = nsec_to_clock_t(stime);
2124 info.si_status = SIGCONT;
2127 info.si_status = tsk->signal->group_exit_code & 0x7f;
2130 info.si_status = tsk->exit_code & 0x7f;
2136 sighand = parent->sighand;
2137 spin_lock_irqsave(&sighand->siglock, flags);
2138 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2139 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2140 __group_send_sig_info(SIGCHLD, &info, parent);
2142 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2144 __wake_up_parent(tsk, parent);
2145 spin_unlock_irqrestore(&sighand->siglock, flags);
2148 static inline bool may_ptrace_stop(void)
2150 if (!likely(current->ptrace))
2153 * Are we in the middle of do_coredump?
2154 * If so and our tracer is also part of the coredump stopping
2155 * is a deadlock situation, and pointless because our tracer
2156 * is dead so don't allow us to stop.
2157 * If SIGKILL was already sent before the caller unlocked
2158 * ->siglock we must see ->core_state != NULL. Otherwise it
2159 * is safe to enter schedule().
2161 * This is almost outdated, a task with the pending SIGKILL can't
2162 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2163 * after SIGKILL was already dequeued.
2165 if (unlikely(current->mm->core_state) &&
2166 unlikely(current->mm == current->parent->mm))
2174 * This must be called with current->sighand->siglock held.
2176 * This should be the path for all ptrace stops.
2177 * We always set current->last_siginfo while stopped here.
2178 * That makes it a way to test a stopped process for
2179 * being ptrace-stopped vs being job-control-stopped.
2181 * If we actually decide not to stop at all because the tracer
2182 * is gone, we keep current->exit_code unless clear_code.
2184 static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2185 __releases(¤t->sighand->siglock)
2186 __acquires(¤t->sighand->siglock)
2188 bool gstop_done = false;
2190 if (arch_ptrace_stop_needed(exit_code, info)) {
2192 * The arch code has something special to do before a
2193 * ptrace stop. This is allowed to block, e.g. for faults
2194 * on user stack pages. We can't keep the siglock while
2195 * calling arch_ptrace_stop, so we must release it now.
2196 * To preserve proper semantics, we must do this before
2197 * any signal bookkeeping like checking group_stop_count.
2199 spin_unlock_irq(¤t->sighand->siglock);
2200 arch_ptrace_stop(exit_code, info);
2201 spin_lock_irq(¤t->sighand->siglock);
2205 * schedule() will not sleep if there is a pending signal that
2206 * can awaken the task.
2208 set_special_state(TASK_TRACED);
2211 * We're committing to trapping. TRACED should be visible before
2212 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2213 * Also, transition to TRACED and updates to ->jobctl should be
2214 * atomic with respect to siglock and should be done after the arch
2215 * hook as siglock is released and regrabbed across it.
2220 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2222 * set_current_state() smp_wmb();
2224 * wait_task_stopped()
2225 * task_stopped_code()
2226 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2230 current->last_siginfo = info;
2231 current->exit_code = exit_code;
2234 * If @why is CLD_STOPPED, we're trapping to participate in a group
2235 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2236 * across siglock relocks since INTERRUPT was scheduled, PENDING
2237 * could be clear now. We act as if SIGCONT is received after
2238 * TASK_TRACED is entered - ignore it.
2240 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2241 gstop_done = task_participate_group_stop(current);
2243 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2244 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2245 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2246 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2248 /* entering a trap, clear TRAPPING */
2249 task_clear_jobctl_trapping(current);
2251 spin_unlock_irq(¤t->sighand->siglock);
2252 read_lock(&tasklist_lock);
2253 if (may_ptrace_stop()) {
2255 * Notify parents of the stop.
2257 * While ptraced, there are two parents - the ptracer and
2258 * the real_parent of the group_leader. The ptracer should
2259 * know about every stop while the real parent is only
2260 * interested in the completion of group stop. The states
2261 * for the two don't interact with each other. Notify
2262 * separately unless they're gonna be duplicates.
2264 do_notify_parent_cldstop(current, true, why);
2265 if (gstop_done && ptrace_reparented(current))
2266 do_notify_parent_cldstop(current, false, why);
2269 * Don't want to allow preemption here, because
2270 * sys_ptrace() needs this task to be inactive.
2272 * XXX: implement read_unlock_no_resched().
2275 read_unlock(&tasklist_lock);
2276 cgroup_enter_frozen();
2277 preempt_enable_no_resched();
2278 freezable_schedule();
2279 cgroup_leave_frozen(true);
2282 * By the time we got the lock, our tracer went away.
2283 * Don't drop the lock yet, another tracer may come.
2285 * If @gstop_done, the ptracer went away between group stop
2286 * completion and here. During detach, it would have set
2287 * JOBCTL_STOP_PENDING on us and we'll re-enter
2288 * TASK_STOPPED in do_signal_stop() on return, so notifying
2289 * the real parent of the group stop completion is enough.
2292 do_notify_parent_cldstop(current, false, why);
2294 /* tasklist protects us from ptrace_freeze_traced() */
2295 __set_current_state(TASK_RUNNING);
2297 current->exit_code = 0;
2298 read_unlock(&tasklist_lock);
2302 * We are back. Now reacquire the siglock before touching
2303 * last_siginfo, so that we are sure to have synchronized with
2304 * any signal-sending on another CPU that wants to examine it.
2306 spin_lock_irq(¤t->sighand->siglock);
2307 current->last_siginfo = NULL;
2309 /* LISTENING can be set only during STOP traps, clear it */
2310 current->jobctl &= ~JOBCTL_LISTENING;
2313 * Queued signals ignored us while we were stopped for tracing.
2314 * So check for any that we should take before resuming user mode.
2315 * This sets TIF_SIGPENDING, but never clears it.
2317 recalc_sigpending_tsk(current);
2320 static void ptrace_do_notify(int signr, int exit_code, int why)
2322 kernel_siginfo_t info;
2324 clear_siginfo(&info);
2325 info.si_signo = signr;
2326 info.si_code = exit_code;
2327 info.si_pid = task_pid_vnr(current);
2328 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2330 /* Let the debugger run. */
2331 ptrace_stop(exit_code, why, 1, &info);
2334 void ptrace_notify(int exit_code)
2336 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2337 if (unlikely(current->task_works))
2340 spin_lock_irq(¤t->sighand->siglock);
2341 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2342 spin_unlock_irq(¤t->sighand->siglock);
2346 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2347 * @signr: signr causing group stop if initiating
2349 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2350 * and participate in it. If already set, participate in the existing
2351 * group stop. If participated in a group stop (and thus slept), %true is
2352 * returned with siglock released.
2354 * If ptraced, this function doesn't handle stop itself. Instead,
2355 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2356 * untouched. The caller must ensure that INTERRUPT trap handling takes
2357 * places afterwards.
2360 * Must be called with @current->sighand->siglock held, which is released
2364 * %false if group stop is already cancelled or ptrace trap is scheduled.
2365 * %true if participated in group stop.
2367 static bool do_signal_stop(int signr)
2368 __releases(¤t->sighand->siglock)
2370 struct signal_struct *sig = current->signal;
2372 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2373 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2374 struct task_struct *t;
2376 /* signr will be recorded in task->jobctl for retries */
2377 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2379 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2380 unlikely(signal_group_exit(sig)))
2383 * There is no group stop already in progress. We must
2386 * While ptraced, a task may be resumed while group stop is
2387 * still in effect and then receive a stop signal and
2388 * initiate another group stop. This deviates from the
2389 * usual behavior as two consecutive stop signals can't
2390 * cause two group stops when !ptraced. That is why we
2391 * also check !task_is_stopped(t) below.
2393 * The condition can be distinguished by testing whether
2394 * SIGNAL_STOP_STOPPED is already set. Don't generate
2395 * group_exit_code in such case.
2397 * This is not necessary for SIGNAL_STOP_CONTINUED because
2398 * an intervening stop signal is required to cause two
2399 * continued events regardless of ptrace.
2401 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2402 sig->group_exit_code = signr;
2404 sig->group_stop_count = 0;
2406 if (task_set_jobctl_pending(current, signr | gstop))
2407 sig->group_stop_count++;
2410 while_each_thread(current, t) {
2412 * Setting state to TASK_STOPPED for a group
2413 * stop is always done with the siglock held,
2414 * so this check has no races.
2416 if (!task_is_stopped(t) &&
2417 task_set_jobctl_pending(t, signr | gstop)) {
2418 sig->group_stop_count++;
2419 if (likely(!(t->ptrace & PT_SEIZED)))
2420 signal_wake_up(t, 0);
2422 ptrace_trap_notify(t);
2427 if (likely(!current->ptrace)) {
2431 * If there are no other threads in the group, or if there
2432 * is a group stop in progress and we are the last to stop,
2433 * report to the parent.
2435 if (task_participate_group_stop(current))
2436 notify = CLD_STOPPED;
2438 set_special_state(TASK_STOPPED);
2439 spin_unlock_irq(¤t->sighand->siglock);
2442 * Notify the parent of the group stop completion. Because
2443 * we're not holding either the siglock or tasklist_lock
2444 * here, ptracer may attach inbetween; however, this is for
2445 * group stop and should always be delivered to the real
2446 * parent of the group leader. The new ptracer will get
2447 * its notification when this task transitions into
2451 read_lock(&tasklist_lock);
2452 do_notify_parent_cldstop(current, false, notify);
2453 read_unlock(&tasklist_lock);
2456 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2457 cgroup_enter_frozen();
2458 freezable_schedule();
2462 * While ptraced, group stop is handled by STOP trap.
2463 * Schedule it and let the caller deal with it.
2465 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2471 * do_jobctl_trap - take care of ptrace jobctl traps
2473 * When PT_SEIZED, it's used for both group stop and explicit
2474 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2475 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2476 * the stop signal; otherwise, %SIGTRAP.
2478 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2479 * number as exit_code and no siginfo.
2482 * Must be called with @current->sighand->siglock held, which may be
2483 * released and re-acquired before returning with intervening sleep.
2485 static void do_jobctl_trap(void)
2487 struct signal_struct *signal = current->signal;
2488 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2490 if (current->ptrace & PT_SEIZED) {
2491 if (!signal->group_stop_count &&
2492 !(signal->flags & SIGNAL_STOP_STOPPED))
2494 WARN_ON_ONCE(!signr);
2495 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2498 WARN_ON_ONCE(!signr);
2499 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2500 current->exit_code = 0;
2505 * do_freezer_trap - handle the freezer jobctl trap
2507 * Puts the task into frozen state, if only the task is not about to quit.
2508 * In this case it drops JOBCTL_TRAP_FREEZE.
2511 * Must be called with @current->sighand->siglock held,
2512 * which is always released before returning.
2514 static void do_freezer_trap(void)
2515 __releases(¤t->sighand->siglock)
2518 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2519 * let's make another loop to give it a chance to be handled.
2520 * In any case, we'll return back.
2522 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2523 JOBCTL_TRAP_FREEZE) {
2524 spin_unlock_irq(¤t->sighand->siglock);
2529 * Now we're sure that there is no pending fatal signal and no
2530 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2531 * immediately (if there is a non-fatal signal pending), and
2532 * put the task into sleep.
2534 __set_current_state(TASK_INTERRUPTIBLE);
2535 clear_thread_flag(TIF_SIGPENDING);
2536 spin_unlock_irq(¤t->sighand->siglock);
2537 cgroup_enter_frozen();
2538 freezable_schedule();
2541 static int ptrace_signal(int signr, kernel_siginfo_t *info)
2544 * We do not check sig_kernel_stop(signr) but set this marker
2545 * unconditionally because we do not know whether debugger will
2546 * change signr. This flag has no meaning unless we are going
2547 * to stop after return from ptrace_stop(). In this case it will
2548 * be checked in do_signal_stop(), we should only stop if it was
2549 * not cleared by SIGCONT while we were sleeping. See also the
2550 * comment in dequeue_signal().
2552 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2553 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2555 /* We're back. Did the debugger cancel the sig? */
2556 signr = current->exit_code;
2560 current->exit_code = 0;
2563 * Update the siginfo structure if the signal has
2564 * changed. If the debugger wanted something
2565 * specific in the siginfo structure then it should
2566 * have updated *info via PTRACE_SETSIGINFO.
2568 if (signr != info->si_signo) {
2569 clear_siginfo(info);
2570 info->si_signo = signr;
2572 info->si_code = SI_USER;
2574 info->si_pid = task_pid_vnr(current->parent);
2575 info->si_uid = from_kuid_munged(current_user_ns(),
2576 task_uid(current->parent));
2580 /* If the (new) signal is now blocked, requeue it. */
2581 if (sigismember(¤t->blocked, signr)) {
2582 send_signal(signr, info, current, PIDTYPE_PID);
2589 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2591 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2593 case SIL_FAULT_TRAPNO:
2594 case SIL_FAULT_MCEERR:
2595 case SIL_FAULT_BNDERR:
2596 case SIL_FAULT_PKUERR:
2597 case SIL_FAULT_PERF_EVENT:
2598 ksig->info.si_addr = arch_untagged_si_addr(
2599 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2611 bool get_signal(struct ksignal *ksig)
2613 struct sighand_struct *sighand = current->sighand;
2614 struct signal_struct *signal = current->signal;
2617 if (unlikely(current->task_works))
2621 * For non-generic architectures, check for TIF_NOTIFY_SIGNAL so
2622 * that the arch handlers don't all have to do it. If we get here
2623 * without TIF_SIGPENDING, just exit after running signal work.
2625 if (!IS_ENABLED(CONFIG_GENERIC_ENTRY)) {
2626 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
2627 tracehook_notify_signal();
2628 if (!task_sigpending(current))
2632 if (unlikely(uprobe_deny_signal()))
2636 * Do this once, we can't return to user-mode if freezing() == T.
2637 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2638 * thus do not need another check after return.
2643 spin_lock_irq(&sighand->siglock);
2646 * Every stopped thread goes here after wakeup. Check to see if
2647 * we should notify the parent, prepare_signal(SIGCONT) encodes
2648 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2650 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2653 if (signal->flags & SIGNAL_CLD_CONTINUED)
2654 why = CLD_CONTINUED;
2658 signal->flags &= ~SIGNAL_CLD_MASK;
2660 spin_unlock_irq(&sighand->siglock);
2663 * Notify the parent that we're continuing. This event is
2664 * always per-process and doesn't make whole lot of sense
2665 * for ptracers, who shouldn't consume the state via
2666 * wait(2) either, but, for backward compatibility, notify
2667 * the ptracer of the group leader too unless it's gonna be
2670 read_lock(&tasklist_lock);
2671 do_notify_parent_cldstop(current, false, why);
2673 if (ptrace_reparented(current->group_leader))
2674 do_notify_parent_cldstop(current->group_leader,
2676 read_unlock(&tasklist_lock);
2681 /* Has this task already been marked for death? */
2682 if (signal_group_exit(signal)) {
2683 ksig->info.si_signo = signr = SIGKILL;
2684 sigdelset(¤t->pending.signal, SIGKILL);
2685 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2686 &sighand->action[SIGKILL - 1]);
2687 recalc_sigpending();
2692 struct k_sigaction *ka;
2694 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2698 if (unlikely(current->jobctl &
2699 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2700 if (current->jobctl & JOBCTL_TRAP_MASK) {
2702 spin_unlock_irq(&sighand->siglock);
2703 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2710 * If the task is leaving the frozen state, let's update
2711 * cgroup counters and reset the frozen bit.
2713 if (unlikely(cgroup_task_frozen(current))) {
2714 spin_unlock_irq(&sighand->siglock);
2715 cgroup_leave_frozen(false);
2720 * Signals generated by the execution of an instruction
2721 * need to be delivered before any other pending signals
2722 * so that the instruction pointer in the signal stack
2723 * frame points to the faulting instruction.
2725 signr = dequeue_synchronous_signal(&ksig->info);
2727 signr = dequeue_signal(current, ¤t->blocked, &ksig->info);
2730 break; /* will return 0 */
2732 if (unlikely(current->ptrace) && signr != SIGKILL) {
2733 signr = ptrace_signal(signr, &ksig->info);
2738 ka = &sighand->action[signr-1];
2740 /* Trace actually delivered signals. */
2741 trace_signal_deliver(signr, &ksig->info, ka);
2743 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2745 if (ka->sa.sa_handler != SIG_DFL) {
2746 /* Run the handler. */
2749 if (ka->sa.sa_flags & SA_ONESHOT)
2750 ka->sa.sa_handler = SIG_DFL;
2752 break; /* will return non-zero "signr" value */
2756 * Now we are doing the default action for this signal.
2758 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2762 * Global init gets no signals it doesn't want.
2763 * Container-init gets no signals it doesn't want from same
2766 * Note that if global/container-init sees a sig_kernel_only()
2767 * signal here, the signal must have been generated internally
2768 * or must have come from an ancestor namespace. In either
2769 * case, the signal cannot be dropped.
2771 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2772 !sig_kernel_only(signr))
2775 if (sig_kernel_stop(signr)) {
2777 * The default action is to stop all threads in
2778 * the thread group. The job control signals
2779 * do nothing in an orphaned pgrp, but SIGSTOP
2780 * always works. Note that siglock needs to be
2781 * dropped during the call to is_orphaned_pgrp()
2782 * because of lock ordering with tasklist_lock.
2783 * This allows an intervening SIGCONT to be posted.
2784 * We need to check for that and bail out if necessary.
2786 if (signr != SIGSTOP) {
2787 spin_unlock_irq(&sighand->siglock);
2789 /* signals can be posted during this window */
2791 if (is_current_pgrp_orphaned())
2794 spin_lock_irq(&sighand->siglock);
2797 if (likely(do_signal_stop(ksig->info.si_signo))) {
2798 /* It released the siglock. */
2803 * We didn't actually stop, due to a race
2804 * with SIGCONT or something like that.
2810 spin_unlock_irq(&sighand->siglock);
2811 if (unlikely(cgroup_task_frozen(current)))
2812 cgroup_leave_frozen(true);
2815 * Anything else is fatal, maybe with a core dump.
2817 current->flags |= PF_SIGNALED;
2819 if (sig_kernel_coredump(signr)) {
2820 if (print_fatal_signals)
2821 print_fatal_signal(ksig->info.si_signo);
2822 proc_coredump_connector(current);
2824 * If it was able to dump core, this kills all
2825 * other threads in the group and synchronizes with
2826 * their demise. If we lost the race with another
2827 * thread getting here, it set group_exit_code
2828 * first and our do_group_exit call below will use
2829 * that value and ignore the one we pass it.
2831 do_coredump(&ksig->info);
2835 * PF_IO_WORKER threads will catch and exit on fatal signals
2836 * themselves. They have cleanup that must be performed, so
2837 * we cannot call do_exit() on their behalf.
2839 if (current->flags & PF_IO_WORKER)
2843 * Death signals, no core dump.
2845 do_group_exit(ksig->info.si_signo);
2848 spin_unlock_irq(&sighand->siglock);
2852 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2853 hide_si_addr_tag_bits(ksig);
2855 return ksig->sig > 0;
2859 * signal_delivered -
2860 * @ksig: kernel signal struct
2861 * @stepping: nonzero if debugger single-step or block-step in use
2863 * This function should be called when a signal has successfully been
2864 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2865 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2866 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2868 static void signal_delivered(struct ksignal *ksig, int stepping)
2872 /* A signal was successfully delivered, and the
2873 saved sigmask was stored on the signal frame,
2874 and will be restored by sigreturn. So we can
2875 simply clear the restore sigmask flag. */
2876 clear_restore_sigmask();
2878 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2879 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2880 sigaddset(&blocked, ksig->sig);
2881 set_current_blocked(&blocked);
2882 if (current->sas_ss_flags & SS_AUTODISARM)
2883 sas_ss_reset(current);
2884 tracehook_signal_handler(stepping);
2887 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2890 force_sigsegv(ksig->sig);
2892 signal_delivered(ksig, stepping);
2896 * It could be that complete_signal() picked us to notify about the
2897 * group-wide signal. Other threads should be notified now to take
2898 * the shared signals in @which since we will not.
2900 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2903 struct task_struct *t;
2905 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2906 if (sigisemptyset(&retarget))
2910 while_each_thread(tsk, t) {
2911 if (t->flags & PF_EXITING)
2914 if (!has_pending_signals(&retarget, &t->blocked))
2916 /* Remove the signals this thread can handle. */
2917 sigandsets(&retarget, &retarget, &t->blocked);
2919 if (!task_sigpending(t))
2920 signal_wake_up(t, 0);
2922 if (sigisemptyset(&retarget))
2927 void exit_signals(struct task_struct *tsk)
2933 * @tsk is about to have PF_EXITING set - lock out users which
2934 * expect stable threadgroup.
2936 cgroup_threadgroup_change_begin(tsk);
2938 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2939 tsk->flags |= PF_EXITING;
2940 cgroup_threadgroup_change_end(tsk);
2944 spin_lock_irq(&tsk->sighand->siglock);
2946 * From now this task is not visible for group-wide signals,
2947 * see wants_signal(), do_signal_stop().
2949 tsk->flags |= PF_EXITING;
2951 cgroup_threadgroup_change_end(tsk);
2953 if (!task_sigpending(tsk))
2956 unblocked = tsk->blocked;
2957 signotset(&unblocked);
2958 retarget_shared_pending(tsk, &unblocked);
2960 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2961 task_participate_group_stop(tsk))
2962 group_stop = CLD_STOPPED;
2964 spin_unlock_irq(&tsk->sighand->siglock);
2967 * If group stop has completed, deliver the notification. This
2968 * should always go to the real parent of the group leader.
2970 if (unlikely(group_stop)) {
2971 read_lock(&tasklist_lock);
2972 do_notify_parent_cldstop(tsk, false, group_stop);
2973 read_unlock(&tasklist_lock);
2978 * System call entry points.
2982 * sys_restart_syscall - restart a system call
2984 SYSCALL_DEFINE0(restart_syscall)
2986 struct restart_block *restart = ¤t->restart_block;
2987 return restart->fn(restart);
2990 long do_no_restart_syscall(struct restart_block *param)
2995 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2997 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
2998 sigset_t newblocked;
2999 /* A set of now blocked but previously unblocked signals. */
3000 sigandnsets(&newblocked, newset, ¤t->blocked);
3001 retarget_shared_pending(tsk, &newblocked);
3003 tsk->blocked = *newset;
3004 recalc_sigpending();
3008 * set_current_blocked - change current->blocked mask
3011 * It is wrong to change ->blocked directly, this helper should be used
3012 * to ensure the process can't miss a shared signal we are going to block.
3014 void set_current_blocked(sigset_t *newset)
3016 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3017 __set_current_blocked(newset);
3020 void __set_current_blocked(const sigset_t *newset)
3022 struct task_struct *tsk = current;
3025 * In case the signal mask hasn't changed, there is nothing we need
3026 * to do. The current->blocked shouldn't be modified by other task.
3028 if (sigequalsets(&tsk->blocked, newset))
3031 spin_lock_irq(&tsk->sighand->siglock);
3032 __set_task_blocked(tsk, newset);
3033 spin_unlock_irq(&tsk->sighand->siglock);
3037 * This is also useful for kernel threads that want to temporarily
3038 * (or permanently) block certain signals.
3040 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3041 * interface happily blocks "unblockable" signals like SIGKILL
3044 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3046 struct task_struct *tsk = current;
3049 /* Lockless, only current can change ->blocked, never from irq */
3051 *oldset = tsk->blocked;
3055 sigorsets(&newset, &tsk->blocked, set);
3058 sigandnsets(&newset, &tsk->blocked, set);
3067 __set_current_blocked(&newset);
3070 EXPORT_SYMBOL(sigprocmask);
3073 * The api helps set app-provided sigmasks.
3075 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3076 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3078 * Note that it does set_restore_sigmask() in advance, so it must be always
3079 * paired with restore_saved_sigmask_unless() before return from syscall.
3081 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3087 if (sigsetsize != sizeof(sigset_t))
3089 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3092 set_restore_sigmask();
3093 current->saved_sigmask = current->blocked;
3094 set_current_blocked(&kmask);
3099 #ifdef CONFIG_COMPAT
3100 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3107 if (sigsetsize != sizeof(compat_sigset_t))
3109 if (get_compat_sigset(&kmask, umask))
3112 set_restore_sigmask();
3113 current->saved_sigmask = current->blocked;
3114 set_current_blocked(&kmask);
3121 * sys_rt_sigprocmask - change the list of currently blocked signals
3122 * @how: whether to add, remove, or set signals
3123 * @nset: stores pending signals
3124 * @oset: previous value of signal mask if non-null
3125 * @sigsetsize: size of sigset_t type
3127 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3128 sigset_t __user *, oset, size_t, sigsetsize)
3130 sigset_t old_set, new_set;
3133 /* XXX: Don't preclude handling different sized sigset_t's. */
3134 if (sigsetsize != sizeof(sigset_t))
3137 old_set = current->blocked;
3140 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3142 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3144 error = sigprocmask(how, &new_set, NULL);
3150 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3157 #ifdef CONFIG_COMPAT
3158 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3159 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3161 sigset_t old_set = current->blocked;
3163 /* XXX: Don't preclude handling different sized sigset_t's. */
3164 if (sigsetsize != sizeof(sigset_t))
3170 if (get_compat_sigset(&new_set, nset))
3172 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3174 error = sigprocmask(how, &new_set, NULL);
3178 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3182 static void do_sigpending(sigset_t *set)
3184 spin_lock_irq(¤t->sighand->siglock);
3185 sigorsets(set, ¤t->pending.signal,
3186 ¤t->signal->shared_pending.signal);
3187 spin_unlock_irq(¤t->sighand->siglock);
3189 /* Outside the lock because only this thread touches it. */
3190 sigandsets(set, ¤t->blocked, set);
3194 * sys_rt_sigpending - examine a pending signal that has been raised
3196 * @uset: stores pending signals
3197 * @sigsetsize: size of sigset_t type or larger
3199 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3203 if (sigsetsize > sizeof(*uset))
3206 do_sigpending(&set);
3208 if (copy_to_user(uset, &set, sigsetsize))
3214 #ifdef CONFIG_COMPAT
3215 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3216 compat_size_t, sigsetsize)
3220 if (sigsetsize > sizeof(*uset))
3223 do_sigpending(&set);
3225 return put_compat_sigset(uset, &set, sigsetsize);
3229 static const struct {
3230 unsigned char limit, layout;
3232 [SIGILL] = { NSIGILL, SIL_FAULT },
3233 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3234 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3235 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3236 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3238 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3240 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3241 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3242 [SIGSYS] = { NSIGSYS, SIL_SYS },
3245 static bool known_siginfo_layout(unsigned sig, int si_code)
3247 if (si_code == SI_KERNEL)
3249 else if ((si_code > SI_USER)) {
3250 if (sig_specific_sicodes(sig)) {
3251 if (si_code <= sig_sicodes[sig].limit)
3254 else if (si_code <= NSIGPOLL)
3257 else if (si_code >= SI_DETHREAD)
3259 else if (si_code == SI_ASYNCNL)
3264 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3266 enum siginfo_layout layout = SIL_KILL;
3267 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3268 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3269 (si_code <= sig_sicodes[sig].limit)) {
3270 layout = sig_sicodes[sig].layout;
3271 /* Handle the exceptions */
3272 if ((sig == SIGBUS) &&
3273 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3274 layout = SIL_FAULT_MCEERR;
3275 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3276 layout = SIL_FAULT_BNDERR;
3278 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3279 layout = SIL_FAULT_PKUERR;
3281 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3282 layout = SIL_FAULT_PERF_EVENT;
3283 else if (IS_ENABLED(CONFIG_SPARC) &&
3284 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3285 layout = SIL_FAULT_TRAPNO;
3286 else if (IS_ENABLED(CONFIG_ALPHA) &&
3288 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3289 layout = SIL_FAULT_TRAPNO;
3291 else if (si_code <= NSIGPOLL)
3294 if (si_code == SI_TIMER)
3296 else if (si_code == SI_SIGIO)
3298 else if (si_code < 0)
3304 static inline char __user *si_expansion(const siginfo_t __user *info)
3306 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3309 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3311 char __user *expansion = si_expansion(to);
3312 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3314 if (clear_user(expansion, SI_EXPANSION_SIZE))
3319 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3320 const siginfo_t __user *from)
3322 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3323 char __user *expansion = si_expansion(from);
3324 char buf[SI_EXPANSION_SIZE];
3327 * An unknown si_code might need more than
3328 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3329 * extra bytes are 0. This guarantees copy_siginfo_to_user
3330 * will return this data to userspace exactly.
3332 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3334 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3342 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3343 const siginfo_t __user *from)
3345 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3347 to->si_signo = signo;
3348 return post_copy_siginfo_from_user(to, from);
3351 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3353 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3355 return post_copy_siginfo_from_user(to, from);
3358 #ifdef CONFIG_COMPAT
3360 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3361 * @to: compat siginfo destination
3362 * @from: kernel siginfo source
3364 * Note: This function does not work properly for the SIGCHLD on x32, but
3365 * fortunately it doesn't have to. The only valid callers for this function are
3366 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3367 * The latter does not care because SIGCHLD will never cause a coredump.
3369 void copy_siginfo_to_external32(struct compat_siginfo *to,
3370 const struct kernel_siginfo *from)
3372 memset(to, 0, sizeof(*to));
3374 to->si_signo = from->si_signo;
3375 to->si_errno = from->si_errno;
3376 to->si_code = from->si_code;
3377 switch(siginfo_layout(from->si_signo, from->si_code)) {
3379 to->si_pid = from->si_pid;
3380 to->si_uid = from->si_uid;
3383 to->si_tid = from->si_tid;
3384 to->si_overrun = from->si_overrun;
3385 to->si_int = from->si_int;
3388 to->si_band = from->si_band;
3389 to->si_fd = from->si_fd;
3392 to->si_addr = ptr_to_compat(from->si_addr);
3394 case SIL_FAULT_TRAPNO:
3395 to->si_addr = ptr_to_compat(from->si_addr);
3396 to->si_trapno = from->si_trapno;
3398 case SIL_FAULT_MCEERR:
3399 to->si_addr = ptr_to_compat(from->si_addr);
3400 to->si_addr_lsb = from->si_addr_lsb;
3402 case SIL_FAULT_BNDERR:
3403 to->si_addr = ptr_to_compat(from->si_addr);
3404 to->si_lower = ptr_to_compat(from->si_lower);
3405 to->si_upper = ptr_to_compat(from->si_upper);
3407 case SIL_FAULT_PKUERR:
3408 to->si_addr = ptr_to_compat(from->si_addr);
3409 to->si_pkey = from->si_pkey;
3411 case SIL_FAULT_PERF_EVENT:
3412 to->si_addr = ptr_to_compat(from->si_addr);
3413 to->si_perf_data = from->si_perf_data;
3414 to->si_perf_type = from->si_perf_type;
3417 to->si_pid = from->si_pid;
3418 to->si_uid = from->si_uid;
3419 to->si_status = from->si_status;
3420 to->si_utime = from->si_utime;
3421 to->si_stime = from->si_stime;
3424 to->si_pid = from->si_pid;
3425 to->si_uid = from->si_uid;
3426 to->si_int = from->si_int;
3429 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3430 to->si_syscall = from->si_syscall;
3431 to->si_arch = from->si_arch;
3436 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3437 const struct kernel_siginfo *from)
3439 struct compat_siginfo new;
3441 copy_siginfo_to_external32(&new, from);
3442 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3447 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3448 const struct compat_siginfo *from)
3451 to->si_signo = from->si_signo;
3452 to->si_errno = from->si_errno;
3453 to->si_code = from->si_code;
3454 switch(siginfo_layout(from->si_signo, from->si_code)) {
3456 to->si_pid = from->si_pid;
3457 to->si_uid = from->si_uid;
3460 to->si_tid = from->si_tid;
3461 to->si_overrun = from->si_overrun;
3462 to->si_int = from->si_int;
3465 to->si_band = from->si_band;
3466 to->si_fd = from->si_fd;
3469 to->si_addr = compat_ptr(from->si_addr);
3471 case SIL_FAULT_TRAPNO:
3472 to->si_addr = compat_ptr(from->si_addr);
3473 to->si_trapno = from->si_trapno;
3475 case SIL_FAULT_MCEERR:
3476 to->si_addr = compat_ptr(from->si_addr);
3477 to->si_addr_lsb = from->si_addr_lsb;
3479 case SIL_FAULT_BNDERR:
3480 to->si_addr = compat_ptr(from->si_addr);
3481 to->si_lower = compat_ptr(from->si_lower);
3482 to->si_upper = compat_ptr(from->si_upper);
3484 case SIL_FAULT_PKUERR:
3485 to->si_addr = compat_ptr(from->si_addr);
3486 to->si_pkey = from->si_pkey;
3488 case SIL_FAULT_PERF_EVENT:
3489 to->si_addr = compat_ptr(from->si_addr);
3490 to->si_perf_data = from->si_perf_data;
3491 to->si_perf_type = from->si_perf_type;
3494 to->si_pid = from->si_pid;
3495 to->si_uid = from->si_uid;
3496 to->si_status = from->si_status;
3497 #ifdef CONFIG_X86_X32_ABI
3498 if (in_x32_syscall()) {
3499 to->si_utime = from->_sifields._sigchld_x32._utime;
3500 to->si_stime = from->_sifields._sigchld_x32._stime;
3504 to->si_utime = from->si_utime;
3505 to->si_stime = from->si_stime;
3509 to->si_pid = from->si_pid;
3510 to->si_uid = from->si_uid;
3511 to->si_int = from->si_int;
3514 to->si_call_addr = compat_ptr(from->si_call_addr);
3515 to->si_syscall = from->si_syscall;
3516 to->si_arch = from->si_arch;
3522 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3523 const struct compat_siginfo __user *ufrom)
3525 struct compat_siginfo from;
3527 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3530 from.si_signo = signo;
3531 return post_copy_siginfo_from_user32(to, &from);
3534 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3535 const struct compat_siginfo __user *ufrom)
3537 struct compat_siginfo from;
3539 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3542 return post_copy_siginfo_from_user32(to, &from);
3544 #endif /* CONFIG_COMPAT */
3547 * do_sigtimedwait - wait for queued signals specified in @which
3548 * @which: queued signals to wait for
3549 * @info: if non-null, the signal's siginfo is returned here
3550 * @ts: upper bound on process time suspension
3552 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3553 const struct timespec64 *ts)
3555 ktime_t *to = NULL, timeout = KTIME_MAX;
3556 struct task_struct *tsk = current;
3557 sigset_t mask = *which;
3561 if (!timespec64_valid(ts))
3563 timeout = timespec64_to_ktime(*ts);
3568 * Invert the set of allowed signals to get those we want to block.
3570 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3573 spin_lock_irq(&tsk->sighand->siglock);
3574 sig = dequeue_signal(tsk, &mask, info);
3575 if (!sig && timeout) {
3577 * None ready, temporarily unblock those we're interested
3578 * while we are sleeping in so that we'll be awakened when
3579 * they arrive. Unblocking is always fine, we can avoid
3580 * set_current_blocked().
3582 tsk->real_blocked = tsk->blocked;
3583 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3584 recalc_sigpending();
3585 spin_unlock_irq(&tsk->sighand->siglock);
3587 __set_current_state(TASK_INTERRUPTIBLE);
3588 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3590 spin_lock_irq(&tsk->sighand->siglock);
3591 __set_task_blocked(tsk, &tsk->real_blocked);
3592 sigemptyset(&tsk->real_blocked);
3593 sig = dequeue_signal(tsk, &mask, info);
3595 spin_unlock_irq(&tsk->sighand->siglock);
3599 return ret ? -EINTR : -EAGAIN;
3603 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3605 * @uthese: queued signals to wait for
3606 * @uinfo: if non-null, the signal's siginfo is returned here
3607 * @uts: upper bound on process time suspension
3608 * @sigsetsize: size of sigset_t type
3610 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3611 siginfo_t __user *, uinfo,
3612 const struct __kernel_timespec __user *, uts,
3616 struct timespec64 ts;
3617 kernel_siginfo_t info;
3620 /* XXX: Don't preclude handling different sized sigset_t's. */
3621 if (sigsetsize != sizeof(sigset_t))
3624 if (copy_from_user(&these, uthese, sizeof(these)))
3628 if (get_timespec64(&ts, uts))
3632 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3634 if (ret > 0 && uinfo) {
3635 if (copy_siginfo_to_user(uinfo, &info))
3642 #ifdef CONFIG_COMPAT_32BIT_TIME
3643 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3644 siginfo_t __user *, uinfo,
3645 const struct old_timespec32 __user *, uts,
3649 struct timespec64 ts;
3650 kernel_siginfo_t info;
3653 if (sigsetsize != sizeof(sigset_t))
3656 if (copy_from_user(&these, uthese, sizeof(these)))
3660 if (get_old_timespec32(&ts, uts))
3664 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3666 if (ret > 0 && uinfo) {
3667 if (copy_siginfo_to_user(uinfo, &info))
3675 #ifdef CONFIG_COMPAT
3676 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3677 struct compat_siginfo __user *, uinfo,
3678 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3681 struct timespec64 t;
3682 kernel_siginfo_t info;
3685 if (sigsetsize != sizeof(sigset_t))
3688 if (get_compat_sigset(&s, uthese))
3692 if (get_timespec64(&t, uts))
3696 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3698 if (ret > 0 && uinfo) {
3699 if (copy_siginfo_to_user32(uinfo, &info))
3706 #ifdef CONFIG_COMPAT_32BIT_TIME
3707 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3708 struct compat_siginfo __user *, uinfo,
3709 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3712 struct timespec64 t;
3713 kernel_siginfo_t info;
3716 if (sigsetsize != sizeof(sigset_t))
3719 if (get_compat_sigset(&s, uthese))
3723 if (get_old_timespec32(&t, uts))
3727 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3729 if (ret > 0 && uinfo) {
3730 if (copy_siginfo_to_user32(uinfo, &info))
3739 static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3741 clear_siginfo(info);
3742 info->si_signo = sig;
3744 info->si_code = SI_USER;
3745 info->si_pid = task_tgid_vnr(current);
3746 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3750 * sys_kill - send a signal to a process
3751 * @pid: the PID of the process
3752 * @sig: signal to be sent
3754 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3756 struct kernel_siginfo info;
3758 prepare_kill_siginfo(sig, &info);
3760 return kill_something_info(sig, &info, pid);
3764 * Verify that the signaler and signalee either are in the same pid namespace
3765 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3768 static bool access_pidfd_pidns(struct pid *pid)
3770 struct pid_namespace *active = task_active_pid_ns(current);
3771 struct pid_namespace *p = ns_of_pid(pid);
3784 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3785 siginfo_t __user *info)
3787 #ifdef CONFIG_COMPAT
3789 * Avoid hooking up compat syscalls and instead handle necessary
3790 * conversions here. Note, this is a stop-gap measure and should not be
3791 * considered a generic solution.
3793 if (in_compat_syscall())
3794 return copy_siginfo_from_user32(
3795 kinfo, (struct compat_siginfo __user *)info);
3797 return copy_siginfo_from_user(kinfo, info);
3800 static struct pid *pidfd_to_pid(const struct file *file)
3804 pid = pidfd_pid(file);
3808 return tgid_pidfd_to_pid(file);
3812 * sys_pidfd_send_signal - Signal a process through a pidfd
3813 * @pidfd: file descriptor of the process
3814 * @sig: signal to send
3815 * @info: signal info
3816 * @flags: future flags
3818 * The syscall currently only signals via PIDTYPE_PID which covers
3819 * kill(<positive-pid>, <signal>. It does not signal threads or process
3821 * In order to extend the syscall to threads and process groups the @flags
3822 * argument should be used. In essence, the @flags argument will determine
3823 * what is signaled and not the file descriptor itself. Put in other words,
3824 * grouping is a property of the flags argument not a property of the file
3827 * Return: 0 on success, negative errno on failure
3829 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3830 siginfo_t __user *, info, unsigned int, flags)
3835 kernel_siginfo_t kinfo;
3837 /* Enforce flags be set to 0 until we add an extension. */
3845 /* Is this a pidfd? */
3846 pid = pidfd_to_pid(f.file);
3853 if (!access_pidfd_pidns(pid))
3857 ret = copy_siginfo_from_user_any(&kinfo, info);
3862 if (unlikely(sig != kinfo.si_signo))
3865 /* Only allow sending arbitrary signals to yourself. */
3867 if ((task_pid(current) != pid) &&
3868 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3871 prepare_kill_siginfo(sig, &kinfo);
3874 ret = kill_pid_info(sig, &kinfo, pid);
3882 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3884 struct task_struct *p;
3888 p = find_task_by_vpid(pid);
3889 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3890 error = check_kill_permission(sig, info, p);
3892 * The null signal is a permissions and process existence
3893 * probe. No signal is actually delivered.
3895 if (!error && sig) {
3896 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3898 * If lock_task_sighand() failed we pretend the task
3899 * dies after receiving the signal. The window is tiny,
3900 * and the signal is private anyway.
3902 if (unlikely(error == -ESRCH))
3911 static int do_tkill(pid_t tgid, pid_t pid, int sig)
3913 struct kernel_siginfo info;
3915 clear_siginfo(&info);
3916 info.si_signo = sig;
3918 info.si_code = SI_TKILL;
3919 info.si_pid = task_tgid_vnr(current);
3920 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3922 return do_send_specific(tgid, pid, sig, &info);
3926 * sys_tgkill - send signal to one specific thread
3927 * @tgid: the thread group ID of the thread
3928 * @pid: the PID of the thread
3929 * @sig: signal to be sent
3931 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3932 * exists but it's not belonging to the target process anymore. This
3933 * method solves the problem of threads exiting and PIDs getting reused.
3935 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3937 /* This is only valid for single tasks */
3938 if (pid <= 0 || tgid <= 0)
3941 return do_tkill(tgid, pid, sig);
3945 * sys_tkill - send signal to one specific task
3946 * @pid: the PID of the task
3947 * @sig: signal to be sent
3949 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3951 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3953 /* This is only valid for single tasks */
3957 return do_tkill(0, pid, sig);
3960 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3962 /* Not even root can pretend to send signals from the kernel.
3963 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3965 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3966 (task_pid_vnr(current) != pid))
3969 /* POSIX.1b doesn't mention process groups. */
3970 return kill_proc_info(sig, info, pid);
3974 * sys_rt_sigqueueinfo - send signal information to a signal
3975 * @pid: the PID of the thread
3976 * @sig: signal to be sent
3977 * @uinfo: signal info to be sent
3979 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3980 siginfo_t __user *, uinfo)
3982 kernel_siginfo_t info;
3983 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3986 return do_rt_sigqueueinfo(pid, sig, &info);
3989 #ifdef CONFIG_COMPAT
3990 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3993 struct compat_siginfo __user *, uinfo)
3995 kernel_siginfo_t info;
3996 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3999 return do_rt_sigqueueinfo(pid, sig, &info);
4003 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4005 /* This is only valid for single tasks */
4006 if (pid <= 0 || tgid <= 0)
4009 /* Not even root can pretend to send signals from the kernel.
4010 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4012 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4013 (task_pid_vnr(current) != pid))
4016 return do_send_specific(tgid, pid, sig, info);
4019 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4020 siginfo_t __user *, uinfo)
4022 kernel_siginfo_t info;
4023 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4026 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4029 #ifdef CONFIG_COMPAT
4030 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4034 struct compat_siginfo __user *, uinfo)
4036 kernel_siginfo_t info;
4037 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4040 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4045 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4047 void kernel_sigaction(int sig, __sighandler_t action)
4049 spin_lock_irq(¤t->sighand->siglock);
4050 current->sighand->action[sig - 1].sa.sa_handler = action;
4051 if (action == SIG_IGN) {
4055 sigaddset(&mask, sig);
4057 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
4058 flush_sigqueue_mask(&mask, ¤t->pending);
4059 recalc_sigpending();
4061 spin_unlock_irq(¤t->sighand->siglock);
4063 EXPORT_SYMBOL(kernel_sigaction);
4065 void __weak sigaction_compat_abi(struct k_sigaction *act,
4066 struct k_sigaction *oact)
4070 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4072 struct task_struct *p = current, *t;
4073 struct k_sigaction *k;
4076 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4079 k = &p->sighand->action[sig-1];
4081 spin_lock_irq(&p->sighand->siglock);
4086 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4087 * e.g. by having an architecture use the bit in their uapi.
4089 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4092 * Clear unknown flag bits in order to allow userspace to detect missing
4093 * support for flag bits and to allow the kernel to use non-uapi bits
4097 act->sa.sa_flags &= UAPI_SA_FLAGS;
4099 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4101 sigaction_compat_abi(act, oact);
4104 sigdelsetmask(&act->sa.sa_mask,
4105 sigmask(SIGKILL) | sigmask(SIGSTOP));
4109 * "Setting a signal action to SIG_IGN for a signal that is
4110 * pending shall cause the pending signal to be discarded,
4111 * whether or not it is blocked."
4113 * "Setting a signal action to SIG_DFL for a signal that is
4114 * pending and whose default action is to ignore the signal
4115 * (for example, SIGCHLD), shall cause the pending signal to
4116 * be discarded, whether or not it is blocked"
4118 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4120 sigaddset(&mask, sig);
4121 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4122 for_each_thread(p, t)
4123 flush_sigqueue_mask(&mask, &t->pending);
4127 spin_unlock_irq(&p->sighand->siglock);
4132 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4135 struct task_struct *t = current;
4138 memset(oss, 0, sizeof(stack_t));
4139 oss->ss_sp = (void __user *) t->sas_ss_sp;
4140 oss->ss_size = t->sas_ss_size;
4141 oss->ss_flags = sas_ss_flags(sp) |
4142 (current->sas_ss_flags & SS_FLAG_BITS);
4146 void __user *ss_sp = ss->ss_sp;
4147 size_t ss_size = ss->ss_size;
4148 unsigned ss_flags = ss->ss_flags;
4151 if (unlikely(on_sig_stack(sp)))
4154 ss_mode = ss_flags & ~SS_FLAG_BITS;
4155 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4159 if (ss_mode == SS_DISABLE) {
4163 if (unlikely(ss_size < min_ss_size))
4167 t->sas_ss_sp = (unsigned long) ss_sp;
4168 t->sas_ss_size = ss_size;
4169 t->sas_ss_flags = ss_flags;
4174 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4178 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4180 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4181 current_user_stack_pointer(),
4183 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4188 int restore_altstack(const stack_t __user *uss)
4191 if (copy_from_user(&new, uss, sizeof(stack_t)))
4193 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4195 /* squash all but EFAULT for now */
4199 int __save_altstack(stack_t __user *uss, unsigned long sp)
4201 struct task_struct *t = current;
4202 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4203 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4204 __put_user(t->sas_ss_size, &uss->ss_size);
4208 #ifdef CONFIG_COMPAT
4209 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4210 compat_stack_t __user *uoss_ptr)
4216 compat_stack_t uss32;
4217 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4219 uss.ss_sp = compat_ptr(uss32.ss_sp);
4220 uss.ss_flags = uss32.ss_flags;
4221 uss.ss_size = uss32.ss_size;
4223 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4224 compat_user_stack_pointer(),
4225 COMPAT_MINSIGSTKSZ);
4226 if (ret >= 0 && uoss_ptr) {
4228 memset(&old, 0, sizeof(old));
4229 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4230 old.ss_flags = uoss.ss_flags;
4231 old.ss_size = uoss.ss_size;
4232 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4238 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4239 const compat_stack_t __user *, uss_ptr,
4240 compat_stack_t __user *, uoss_ptr)
4242 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4245 int compat_restore_altstack(const compat_stack_t __user *uss)
4247 int err = do_compat_sigaltstack(uss, NULL);
4248 /* squash all but -EFAULT for now */
4249 return err == -EFAULT ? err : 0;
4252 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4255 struct task_struct *t = current;
4256 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4258 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4259 __put_user(t->sas_ss_size, &uss->ss_size);
4264 #ifdef __ARCH_WANT_SYS_SIGPENDING
4267 * sys_sigpending - examine pending signals
4268 * @uset: where mask of pending signal is returned
4270 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4274 if (sizeof(old_sigset_t) > sizeof(*uset))
4277 do_sigpending(&set);
4279 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4285 #ifdef CONFIG_COMPAT
4286 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4290 do_sigpending(&set);
4292 return put_user(set.sig[0], set32);
4298 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4300 * sys_sigprocmask - examine and change blocked signals
4301 * @how: whether to add, remove, or set signals
4302 * @nset: signals to add or remove (if non-null)
4303 * @oset: previous value of signal mask if non-null
4305 * Some platforms have their own version with special arguments;
4306 * others support only sys_rt_sigprocmask.
4309 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4310 old_sigset_t __user *, oset)
4312 old_sigset_t old_set, new_set;
4313 sigset_t new_blocked;
4315 old_set = current->blocked.sig[0];
4318 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4321 new_blocked = current->blocked;
4325 sigaddsetmask(&new_blocked, new_set);
4328 sigdelsetmask(&new_blocked, new_set);
4331 new_blocked.sig[0] = new_set;
4337 set_current_blocked(&new_blocked);
4341 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4347 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4349 #ifndef CONFIG_ODD_RT_SIGACTION
4351 * sys_rt_sigaction - alter an action taken by a process
4352 * @sig: signal to be sent
4353 * @act: new sigaction
4354 * @oact: used to save the previous sigaction
4355 * @sigsetsize: size of sigset_t type
4357 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4358 const struct sigaction __user *, act,
4359 struct sigaction __user *, oact,
4362 struct k_sigaction new_sa, old_sa;
4365 /* XXX: Don't preclude handling different sized sigset_t's. */
4366 if (sigsetsize != sizeof(sigset_t))
4369 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4372 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4376 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4381 #ifdef CONFIG_COMPAT
4382 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4383 const struct compat_sigaction __user *, act,
4384 struct compat_sigaction __user *, oact,
4385 compat_size_t, sigsetsize)
4387 struct k_sigaction new_ka, old_ka;
4388 #ifdef __ARCH_HAS_SA_RESTORER
4389 compat_uptr_t restorer;
4393 /* XXX: Don't preclude handling different sized sigset_t's. */
4394 if (sigsetsize != sizeof(compat_sigset_t))
4398 compat_uptr_t handler;
4399 ret = get_user(handler, &act->sa_handler);
4400 new_ka.sa.sa_handler = compat_ptr(handler);
4401 #ifdef __ARCH_HAS_SA_RESTORER
4402 ret |= get_user(restorer, &act->sa_restorer);
4403 new_ka.sa.sa_restorer = compat_ptr(restorer);
4405 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4406 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4411 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4413 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4415 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4416 sizeof(oact->sa_mask));
4417 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4418 #ifdef __ARCH_HAS_SA_RESTORER
4419 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4420 &oact->sa_restorer);
4426 #endif /* !CONFIG_ODD_RT_SIGACTION */
4428 #ifdef CONFIG_OLD_SIGACTION
4429 SYSCALL_DEFINE3(sigaction, int, sig,
4430 const struct old_sigaction __user *, act,
4431 struct old_sigaction __user *, oact)
4433 struct k_sigaction new_ka, old_ka;
4438 if (!access_ok(act, sizeof(*act)) ||
4439 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4440 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4441 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4442 __get_user(mask, &act->sa_mask))
4444 #ifdef __ARCH_HAS_KA_RESTORER
4445 new_ka.ka_restorer = NULL;
4447 siginitset(&new_ka.sa.sa_mask, mask);
4450 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4453 if (!access_ok(oact, sizeof(*oact)) ||
4454 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4455 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4456 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4457 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4464 #ifdef CONFIG_COMPAT_OLD_SIGACTION
4465 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4466 const struct compat_old_sigaction __user *, act,
4467 struct compat_old_sigaction __user *, oact)
4469 struct k_sigaction new_ka, old_ka;
4471 compat_old_sigset_t mask;
4472 compat_uptr_t handler, restorer;
4475 if (!access_ok(act, sizeof(*act)) ||
4476 __get_user(handler, &act->sa_handler) ||
4477 __get_user(restorer, &act->sa_restorer) ||
4478 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4479 __get_user(mask, &act->sa_mask))
4482 #ifdef __ARCH_HAS_KA_RESTORER
4483 new_ka.ka_restorer = NULL;
4485 new_ka.sa.sa_handler = compat_ptr(handler);
4486 new_ka.sa.sa_restorer = compat_ptr(restorer);
4487 siginitset(&new_ka.sa.sa_mask, mask);
4490 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4493 if (!access_ok(oact, sizeof(*oact)) ||
4494 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4495 &oact->sa_handler) ||
4496 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4497 &oact->sa_restorer) ||
4498 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4499 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4506 #ifdef CONFIG_SGETMASK_SYSCALL
4509 * For backwards compatibility. Functionality superseded by sigprocmask.
4511 SYSCALL_DEFINE0(sgetmask)
4514 return current->blocked.sig[0];
4517 SYSCALL_DEFINE1(ssetmask, int, newmask)
4519 int old = current->blocked.sig[0];
4522 siginitset(&newset, newmask);
4523 set_current_blocked(&newset);
4527 #endif /* CONFIG_SGETMASK_SYSCALL */
4529 #ifdef __ARCH_WANT_SYS_SIGNAL
4531 * For backwards compatibility. Functionality superseded by sigaction.
4533 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4535 struct k_sigaction new_sa, old_sa;
4538 new_sa.sa.sa_handler = handler;
4539 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4540 sigemptyset(&new_sa.sa.sa_mask);
4542 ret = do_sigaction(sig, &new_sa, &old_sa);
4544 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4546 #endif /* __ARCH_WANT_SYS_SIGNAL */
4548 #ifdef __ARCH_WANT_SYS_PAUSE
4550 SYSCALL_DEFINE0(pause)
4552 while (!signal_pending(current)) {
4553 __set_current_state(TASK_INTERRUPTIBLE);
4556 return -ERESTARTNOHAND;
4561 static int sigsuspend(sigset_t *set)
4563 current->saved_sigmask = current->blocked;
4564 set_current_blocked(set);
4566 while (!signal_pending(current)) {
4567 __set_current_state(TASK_INTERRUPTIBLE);
4570 set_restore_sigmask();
4571 return -ERESTARTNOHAND;
4575 * sys_rt_sigsuspend - replace the signal mask for a value with the
4576 * @unewset value until a signal is received
4577 * @unewset: new signal mask value
4578 * @sigsetsize: size of sigset_t type
4580 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4584 /* XXX: Don't preclude handling different sized sigset_t's. */
4585 if (sigsetsize != sizeof(sigset_t))
4588 if (copy_from_user(&newset, unewset, sizeof(newset)))
4590 return sigsuspend(&newset);
4593 #ifdef CONFIG_COMPAT
4594 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4598 /* XXX: Don't preclude handling different sized sigset_t's. */
4599 if (sigsetsize != sizeof(sigset_t))
4602 if (get_compat_sigset(&newset, unewset))
4604 return sigsuspend(&newset);
4608 #ifdef CONFIG_OLD_SIGSUSPEND
4609 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4612 siginitset(&blocked, mask);
4613 return sigsuspend(&blocked);
4616 #ifdef CONFIG_OLD_SIGSUSPEND3
4617 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4620 siginitset(&blocked, mask);
4621 return sigsuspend(&blocked);
4625 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4630 static inline void siginfo_buildtime_checks(void)
4632 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4634 /* Verify the offsets in the two siginfos match */
4635 #define CHECK_OFFSET(field) \
4636 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4639 CHECK_OFFSET(si_pid);
4640 CHECK_OFFSET(si_uid);
4643 CHECK_OFFSET(si_tid);
4644 CHECK_OFFSET(si_overrun);
4645 CHECK_OFFSET(si_value);
4648 CHECK_OFFSET(si_pid);
4649 CHECK_OFFSET(si_uid);
4650 CHECK_OFFSET(si_value);
4653 CHECK_OFFSET(si_pid);
4654 CHECK_OFFSET(si_uid);
4655 CHECK_OFFSET(si_status);
4656 CHECK_OFFSET(si_utime);
4657 CHECK_OFFSET(si_stime);
4660 CHECK_OFFSET(si_addr);
4661 CHECK_OFFSET(si_trapno);
4662 CHECK_OFFSET(si_addr_lsb);
4663 CHECK_OFFSET(si_lower);
4664 CHECK_OFFSET(si_upper);
4665 CHECK_OFFSET(si_pkey);
4666 CHECK_OFFSET(si_perf_data);
4667 CHECK_OFFSET(si_perf_type);
4670 CHECK_OFFSET(si_band);
4671 CHECK_OFFSET(si_fd);
4674 CHECK_OFFSET(si_call_addr);
4675 CHECK_OFFSET(si_syscall);
4676 CHECK_OFFSET(si_arch);
4680 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4681 offsetof(struct siginfo, si_addr));
4682 if (sizeof(int) == sizeof(void __user *)) {
4683 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4684 sizeof(void __user *));
4686 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4687 sizeof_field(struct siginfo, si_uid)) !=
4688 sizeof(void __user *));
4689 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4690 offsetof(struct siginfo, si_uid));
4692 #ifdef CONFIG_COMPAT
4693 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4694 offsetof(struct compat_siginfo, si_addr));
4695 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4696 sizeof(compat_uptr_t));
4697 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4698 sizeof_field(struct siginfo, si_pid));
4702 void __init signals_init(void)
4704 siginfo_buildtime_checks();
4706 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4709 #ifdef CONFIG_KGDB_KDB
4710 #include <linux/kdb.h>
4712 * kdb_send_sig - Allows kdb to send signals without exposing
4713 * signal internals. This function checks if the required locks are
4714 * available before calling the main signal code, to avoid kdb
4717 void kdb_send_sig(struct task_struct *t, int sig)
4719 static struct task_struct *kdb_prev_t;
4721 if (!spin_trylock(&t->sighand->siglock)) {
4722 kdb_printf("Can't do kill command now.\n"
4723 "The sigmask lock is held somewhere else in "
4724 "kernel, try again later\n");
4727 new_t = kdb_prev_t != t;
4729 if (!task_is_running(t) && new_t) {
4730 spin_unlock(&t->sighand->siglock);
4731 kdb_printf("Process is not RUNNING, sending a signal from "
4732 "kdb risks deadlock\n"
4733 "on the run queue locks. "
4734 "The signal has _not_ been sent.\n"
4735 "Reissue the kill command if you want to risk "
4739 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4740 spin_unlock(&t->sighand->siglock);
4742 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4745 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4747 #endif /* CONFIG_KGDB_KDB */