Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/signal.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 *
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
12 */
13
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>
24#include <linux/fs.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/livepatch.h>
47#include <linux/cgroup.h>
48#include <linux/audit.h>
49
50#define CREATE_TRACE_POINTS
51#include <trace/events/signal.h>
52
53#include <asm/param.h>
54#include <linux/uaccess.h>
55#include <asm/unistd.h>
56#include <asm/siginfo.h>
57#include <asm/cacheflush.h>
58
59/*
60 * SLAB caches for signal bits.
61 */
62
63static struct kmem_cache *sigqueue_cachep;
64
65int print_fatal_signals __read_mostly;
66
67static void __user *sig_handler(struct task_struct *t, int sig)
68{
69 return t->sighand->action[sig - 1].sa.sa_handler;
70}
71
72static inline bool sig_handler_ignored(void __user *handler, int sig)
73{
74 /* Is it explicitly or implicitly ignored? */
75 return handler == SIG_IGN ||
76 (handler == SIG_DFL && sig_kernel_ignore(sig));
77}
78
79static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
80{
81 void __user *handler;
82
83 handler = sig_handler(t, sig);
84
85 /* SIGKILL and SIGSTOP may not be sent to the global init */
86 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
87 return true;
88
89 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
90 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
91 return true;
92
93 /* Only allow kernel generated signals to this kthread */
94 if (unlikely((t->flags & PF_KTHREAD) &&
95 (handler == SIG_KTHREAD_KERNEL) && !force))
96 return true;
97
98 return sig_handler_ignored(handler, sig);
99}
100
101static bool sig_ignored(struct task_struct *t, int sig, bool force)
102{
103 /*
104 * Blocked signals are never ignored, since the
105 * signal handler may change by the time it is
106 * unblocked.
107 */
108 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
109 return false;
110
111 /*
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.
115 */
116 if (t->ptrace && sig != SIGKILL)
117 return false;
118
119 return sig_task_ignored(t, sig, force);
120}
121
122/*
123 * Re-calculate pending state from the set of locally pending
124 * signals, globally pending signals, and blocked signals.
125 */
126static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
127{
128 unsigned long ready;
129 long i;
130
131 switch (_NSIG_WORDS) {
132 default:
133 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
134 ready |= signal->sig[i] &~ blocked->sig[i];
135 break;
136
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];
141 break;
142
143 case 2: ready = signal->sig[1] &~ blocked->sig[1];
144 ready |= signal->sig[0] &~ blocked->sig[0];
145 break;
146
147 case 1: ready = signal->sig[0] &~ blocked->sig[0];
148 }
149 return ready != 0;
150}
151
152#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
153
154static bool recalc_sigpending_tsk(struct task_struct *t)
155{
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);
161 return true;
162 }
163
164 /*
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.
168 */
169 return false;
170}
171
172/*
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.
175 */
176void recalc_sigpending_and_wake(struct task_struct *t)
177{
178 if (recalc_sigpending_tsk(t))
179 signal_wake_up(t, 0);
180}
181
182void recalc_sigpending(void)
183{
184 if (!recalc_sigpending_tsk(current) && !freezing(current) &&
185 !klp_patch_pending(current))
186 clear_thread_flag(TIF_SIGPENDING);
187
188}
189EXPORT_SYMBOL(recalc_sigpending);
190
191void calculate_sigpending(void)
192{
193 /* Have any signals or users of TIF_SIGPENDING been delayed
194 * until after fork?
195 */
196 spin_lock_irq(¤t->sighand->siglock);
197 set_tsk_thread_flag(current, TIF_SIGPENDING);
198 recalc_sigpending();
199 spin_unlock_irq(¤t->sighand->siglock);
200}
201
202/* Given the mask, find the first available signal that should be serviced. */
203
204#define SYNCHRONOUS_MASK \
205 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
206 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
207
208int next_signal(struct sigpending *pending, sigset_t *mask)
209{
210 unsigned long i, *s, *m, x;
211 int sig = 0;
212
213 s = pending->signal.sig;
214 m = mask->sig;
215
216 /*
217 * Handle the first word specially: it contains the
218 * synchronous signals that need to be dequeued first.
219 */
220 x = *s &~ *m;
221 if (x) {
222 if (x & SYNCHRONOUS_MASK)
223 x &= SYNCHRONOUS_MASK;
224 sig = ffz(~x) + 1;
225 return sig;
226 }
227
228 switch (_NSIG_WORDS) {
229 default:
230 for (i = 1; i < _NSIG_WORDS; ++i) {
231 x = *++s &~ *++m;
232 if (!x)
233 continue;
234 sig = ffz(~x) + i*_NSIG_BPW + 1;
235 break;
236 }
237 break;
238
239 case 2:
240 x = s[1] &~ m[1];
241 if (!x)
242 break;
243 sig = ffz(~x) + _NSIG_BPW + 1;
244 break;
245
246 case 1:
247 /* Nothing to do */
248 break;
249 }
250
251 return sig;
252}
253
254static inline void print_dropped_signal(int sig)
255{
256 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
257
258 if (!print_fatal_signals)
259 return;
260
261 if (!__ratelimit(&ratelimit_state))
262 return;
263
264 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
265 current->comm, current->pid, sig);
266}
267
268/**
269 * task_set_jobctl_pending - set jobctl pending bits
270 * @task: target task
271 * @mask: pending bits to set
272 *
273 * Clear @mask from @task->jobctl. @mask must be subset of
274 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
275 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
276 * cleared. If @task is already being killed or exiting, this function
277 * becomes noop.
278 *
279 * CONTEXT:
280 * Must be called with @task->sighand->siglock held.
281 *
282 * RETURNS:
283 * %true if @mask is set, %false if made noop because @task was dying.
284 */
285bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
286{
287 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
288 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
289 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
290
291 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
292 return false;
293
294 if (mask & JOBCTL_STOP_SIGMASK)
295 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
296
297 task->jobctl |= mask;
298 return true;
299}
300
301/**
302 * task_clear_jobctl_trapping - clear jobctl trapping bit
303 * @task: target task
304 *
305 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
306 * Clear it and wake up the ptracer. Note that we don't need any further
307 * locking. @task->siglock guarantees that @task->parent points to the
308 * ptracer.
309 *
310 * CONTEXT:
311 * Must be called with @task->sighand->siglock held.
312 */
313void task_clear_jobctl_trapping(struct task_struct *task)
314{
315 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
316 task->jobctl &= ~JOBCTL_TRAPPING;
317 smp_mb(); /* advised by wake_up_bit() */
318 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
319 }
320}
321
322/**
323 * task_clear_jobctl_pending - clear jobctl pending bits
324 * @task: target task
325 * @mask: pending bits to clear
326 *
327 * Clear @mask from @task->jobctl. @mask must be subset of
328 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
329 * STOP bits are cleared together.
330 *
331 * If clearing of @mask leaves no stop or trap pending, this function calls
332 * task_clear_jobctl_trapping().
333 *
334 * CONTEXT:
335 * Must be called with @task->sighand->siglock held.
336 */
337void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
338{
339 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
340
341 if (mask & JOBCTL_STOP_PENDING)
342 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
343
344 task->jobctl &= ~mask;
345
346 if (!(task->jobctl & JOBCTL_PENDING_MASK))
347 task_clear_jobctl_trapping(task);
348}
349
350/**
351 * task_participate_group_stop - participate in a group stop
352 * @task: task participating in a group stop
353 *
354 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
355 * Group stop states are cleared and the group stop count is consumed if
356 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
357 * stop, the appropriate `SIGNAL_*` flags are set.
358 *
359 * CONTEXT:
360 * Must be called with @task->sighand->siglock held.
361 *
362 * RETURNS:
363 * %true if group stop completion should be notified to the parent, %false
364 * otherwise.
365 */
366static bool task_participate_group_stop(struct task_struct *task)
367{
368 struct signal_struct *sig = task->signal;
369 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
370
371 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
372
373 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
374
375 if (!consume)
376 return false;
377
378 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
379 sig->group_stop_count--;
380
381 /*
382 * Tell the caller to notify completion iff we are entering into a
383 * fresh group stop. Read comment in do_signal_stop() for details.
384 */
385 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
386 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
387 return true;
388 }
389 return false;
390}
391
392void task_join_group_stop(struct task_struct *task)
393{
394 /* Have the new thread join an on-going signal group stop */
395 unsigned long jobctl = current->jobctl;
396 if (jobctl & JOBCTL_STOP_PENDING) {
397 struct signal_struct *sig = current->signal;
398 unsigned long signr = jobctl & JOBCTL_STOP_SIGMASK;
399 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
400 if (task_set_jobctl_pending(task, signr | gstop)) {
401 sig->group_stop_count++;
402 }
403 }
404}
405
406/*
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
410 */
411static struct sigqueue *
412__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
413{
414 struct sigqueue *q = NULL;
415 struct user_struct *user;
416
417 /*
418 * Protect access to @t credentials. This can go away when all
419 * callers hold rcu read lock.
420 */
421 rcu_read_lock();
422 user = get_uid(__task_cred(t)->user);
423 atomic_inc(&user->sigpending);
424 rcu_read_unlock();
425
426 if (override_rlimit ||
427 atomic_read(&user->sigpending) <=
428 task_rlimit(t, RLIMIT_SIGPENDING)) {
429 q = kmem_cache_alloc(sigqueue_cachep, flags);
430 } else {
431 print_dropped_signal(sig);
432 }
433
434 if (unlikely(q == NULL)) {
435 atomic_dec(&user->sigpending);
436 free_uid(user);
437 } else {
438 INIT_LIST_HEAD(&q->list);
439 q->flags = 0;
440 q->user = user;
441 }
442
443 return q;
444}
445
446static void __sigqueue_free(struct sigqueue *q)
447{
448 if (q->flags & SIGQUEUE_PREALLOC)
449 return;
450 atomic_dec(&q->user->sigpending);
451 free_uid(q->user);
452 kmem_cache_free(sigqueue_cachep, q);
453}
454
455void flush_sigqueue(struct sigpending *queue)
456{
457 struct sigqueue *q;
458
459 sigemptyset(&queue->signal);
460 while (!list_empty(&queue->list)) {
461 q = list_entry(queue->list.next, struct sigqueue , list);
462 list_del_init(&q->list);
463 __sigqueue_free(q);
464 }
465}
466
467/*
468 * Flush all pending signals for this kthread.
469 */
470void flush_signals(struct task_struct *t)
471{
472 unsigned long flags;
473
474 spin_lock_irqsave(&t->sighand->siglock, flags);
475 clear_tsk_thread_flag(t, TIF_SIGPENDING);
476 flush_sigqueue(&t->pending);
477 flush_sigqueue(&t->signal->shared_pending);
478 spin_unlock_irqrestore(&t->sighand->siglock, flags);
479}
480EXPORT_SYMBOL(flush_signals);
481
482#ifdef CONFIG_POSIX_TIMERS
483static void __flush_itimer_signals(struct sigpending *pending)
484{
485 sigset_t signal, retain;
486 struct sigqueue *q, *n;
487
488 signal = pending->signal;
489 sigemptyset(&retain);
490
491 list_for_each_entry_safe(q, n, &pending->list, list) {
492 int sig = q->info.si_signo;
493
494 if (likely(q->info.si_code != SI_TIMER)) {
495 sigaddset(&retain, sig);
496 } else {
497 sigdelset(&signal, sig);
498 list_del_init(&q->list);
499 __sigqueue_free(q);
500 }
501 }
502
503 sigorsets(&pending->signal, &signal, &retain);
504}
505
506void flush_itimer_signals(void)
507{
508 struct task_struct *tsk = current;
509 unsigned long flags;
510
511 spin_lock_irqsave(&tsk->sighand->siglock, flags);
512 __flush_itimer_signals(&tsk->pending);
513 __flush_itimer_signals(&tsk->signal->shared_pending);
514 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
515}
516#endif
517
518void ignore_signals(struct task_struct *t)
519{
520 int i;
521
522 for (i = 0; i < _NSIG; ++i)
523 t->sighand->action[i].sa.sa_handler = SIG_IGN;
524
525 flush_signals(t);
526}
527
528/*
529 * Flush all handlers for a task.
530 */
531
532void
533flush_signal_handlers(struct task_struct *t, int force_default)
534{
535 int i;
536 struct k_sigaction *ka = &t->sighand->action[0];
537 for (i = _NSIG ; i != 0 ; i--) {
538 if (force_default || ka->sa.sa_handler != SIG_IGN)
539 ka->sa.sa_handler = SIG_DFL;
540 ka->sa.sa_flags = 0;
541#ifdef __ARCH_HAS_SA_RESTORER
542 ka->sa.sa_restorer = NULL;
543#endif
544 sigemptyset(&ka->sa.sa_mask);
545 ka++;
546 }
547}
548
549bool unhandled_signal(struct task_struct *tsk, int sig)
550{
551 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
552 if (is_global_init(tsk))
553 return true;
554
555 if (handler != SIG_IGN && handler != SIG_DFL)
556 return false;
557
558 /* if ptraced, let the tracer determine */
559 return !tsk->ptrace;
560}
561
562static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
563 bool *resched_timer)
564{
565 struct sigqueue *q, *first = NULL;
566
567 /*
568 * Collect the siginfo appropriate to this signal. Check if
569 * there is another siginfo for the same signal.
570 */
571 list_for_each_entry(q, &list->list, list) {
572 if (q->info.si_signo == sig) {
573 if (first)
574 goto still_pending;
575 first = q;
576 }
577 }
578
579 sigdelset(&list->signal, sig);
580
581 if (first) {
582still_pending:
583 list_del_init(&first->list);
584 copy_siginfo(info, &first->info);
585
586 *resched_timer =
587 (first->flags & SIGQUEUE_PREALLOC) &&
588 (info->si_code == SI_TIMER) &&
589 (info->si_sys_private);
590
591 __sigqueue_free(first);
592 } else {
593 /*
594 * Ok, it wasn't in the queue. This must be
595 * a fast-pathed signal or we must have been
596 * out of queue space. So zero out the info.
597 */
598 clear_siginfo(info);
599 info->si_signo = sig;
600 info->si_errno = 0;
601 info->si_code = SI_USER;
602 info->si_pid = 0;
603 info->si_uid = 0;
604 }
605}
606
607static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
608 kernel_siginfo_t *info, bool *resched_timer)
609{
610 int sig = next_signal(pending, mask);
611
612 if (sig)
613 collect_signal(sig, pending, info, resched_timer);
614 return sig;
615}
616
617/*
618 * Dequeue a signal and return the element to the caller, which is
619 * expected to free it.
620 *
621 * All callers have to hold the siglock.
622 */
623int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
624{
625 bool resched_timer = false;
626 int signr;
627
628 /* We only dequeue private signals from ourselves, we don't let
629 * signalfd steal them
630 */
631 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
632 if (!signr) {
633 signr = __dequeue_signal(&tsk->signal->shared_pending,
634 mask, info, &resched_timer);
635#ifdef CONFIG_POSIX_TIMERS
636 /*
637 * itimer signal ?
638 *
639 * itimers are process shared and we restart periodic
640 * itimers in the signal delivery path to prevent DoS
641 * attacks in the high resolution timer case. This is
642 * compliant with the old way of self-restarting
643 * itimers, as the SIGALRM is a legacy signal and only
644 * queued once. Changing the restart behaviour to
645 * restart the timer in the signal dequeue path is
646 * reducing the timer noise on heavy loaded !highres
647 * systems too.
648 */
649 if (unlikely(signr == SIGALRM)) {
650 struct hrtimer *tmr = &tsk->signal->real_timer;
651
652 if (!hrtimer_is_queued(tmr) &&
653 tsk->signal->it_real_incr != 0) {
654 hrtimer_forward(tmr, tmr->base->get_time(),
655 tsk->signal->it_real_incr);
656 hrtimer_restart(tmr);
657 }
658 }
659#endif
660 }
661
662 recalc_sigpending();
663 if (!signr)
664 return 0;
665
666 if (unlikely(sig_kernel_stop(signr))) {
667 /*
668 * Set a marker that we have dequeued a stop signal. Our
669 * caller might release the siglock and then the pending
670 * stop signal it is about to process is no longer in the
671 * pending bitmasks, but must still be cleared by a SIGCONT
672 * (and overruled by a SIGKILL). So those cases clear this
673 * shared flag after we've set it. Note that this flag may
674 * remain set after the signal we return is ignored or
675 * handled. That doesn't matter because its only purpose
676 * is to alert stop-signal processing code when another
677 * processor has come along and cleared the flag.
678 */
679 current->jobctl |= JOBCTL_STOP_DEQUEUED;
680 }
681#ifdef CONFIG_POSIX_TIMERS
682 if (resched_timer) {
683 /*
684 * Release the siglock to ensure proper locking order
685 * of timer locks outside of siglocks. Note, we leave
686 * irqs disabled here, since the posix-timers code is
687 * about to disable them again anyway.
688 */
689 spin_unlock(&tsk->sighand->siglock);
690 posixtimer_rearm(info);
691 spin_lock(&tsk->sighand->siglock);
692
693 /* Don't expose the si_sys_private value to userspace */
694 info->si_sys_private = 0;
695 }
696#endif
697 return signr;
698}
699EXPORT_SYMBOL_GPL(dequeue_signal);
700
701static int dequeue_synchronous_signal(kernel_siginfo_t *info)
702{
703 struct task_struct *tsk = current;
704 struct sigpending *pending = &tsk->pending;
705 struct sigqueue *q, *sync = NULL;
706
707 /*
708 * Might a synchronous signal be in the queue?
709 */
710 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
711 return 0;
712
713 /*
714 * Return the first synchronous signal in the queue.
715 */
716 list_for_each_entry(q, &pending->list, list) {
717 /* Synchronous signals have a postive si_code */
718 if ((q->info.si_code > SI_USER) &&
719 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
720 sync = q;
721 goto next;
722 }
723 }
724 return 0;
725next:
726 /*
727 * Check if there is another siginfo for the same signal.
728 */
729 list_for_each_entry_continue(q, &pending->list, list) {
730 if (q->info.si_signo == sync->info.si_signo)
731 goto still_pending;
732 }
733
734 sigdelset(&pending->signal, sync->info.si_signo);
735 recalc_sigpending();
736still_pending:
737 list_del_init(&sync->list);
738 copy_siginfo(info, &sync->info);
739 __sigqueue_free(sync);
740 return info->si_signo;
741}
742
743/*
744 * Tell a process that it has a new active signal..
745 *
746 * NOTE! we rely on the previous spin_lock to
747 * lock interrupts for us! We can only be called with
748 * "siglock" held, and the local interrupt must
749 * have been disabled when that got acquired!
750 *
751 * No need to set need_resched since signal event passing
752 * goes through ->blocked
753 */
754void signal_wake_up_state(struct task_struct *t, unsigned int state)
755{
756 set_tsk_thread_flag(t, TIF_SIGPENDING);
757 /*
758 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
759 * case. We don't check t->state here because there is a race with it
760 * executing another processor and just now entering stopped state.
761 * By using wake_up_state, we ensure the process will wake up and
762 * handle its death signal.
763 */
764 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
765 kick_process(t);
766}
767
768/*
769 * Remove signals in mask from the pending set and queue.
770 * Returns 1 if any signals were found.
771 *
772 * All callers must be holding the siglock.
773 */
774static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
775{
776 struct sigqueue *q, *n;
777 sigset_t m;
778
779 sigandsets(&m, mask, &s->signal);
780 if (sigisemptyset(&m))
781 return;
782
783 sigandnsets(&s->signal, &s->signal, mask);
784 list_for_each_entry_safe(q, n, &s->list, list) {
785 if (sigismember(mask, q->info.si_signo)) {
786 list_del_init(&q->list);
787 __sigqueue_free(q);
788 }
789 }
790}
791
792static inline int is_si_special(const struct kernel_siginfo *info)
793{
794 return info <= SEND_SIG_PRIV;
795}
796
797static inline bool si_fromuser(const struct kernel_siginfo *info)
798{
799 return info == SEND_SIG_NOINFO ||
800 (!is_si_special(info) && SI_FROMUSER(info));
801}
802
803/*
804 * called with RCU read lock from check_kill_permission()
805 */
806static bool kill_ok_by_cred(struct task_struct *t)
807{
808 const struct cred *cred = current_cred();
809 const struct cred *tcred = __task_cred(t);
810
811 return uid_eq(cred->euid, tcred->suid) ||
812 uid_eq(cred->euid, tcred->uid) ||
813 uid_eq(cred->uid, tcred->suid) ||
814 uid_eq(cred->uid, tcred->uid) ||
815 ns_capable(tcred->user_ns, CAP_KILL);
816}
817
818/*
819 * Bad permissions for sending the signal
820 * - the caller must hold the RCU read lock
821 */
822static int check_kill_permission(int sig, struct kernel_siginfo *info,
823 struct task_struct *t)
824{
825 struct pid *sid;
826 int error;
827
828 if (!valid_signal(sig))
829 return -EINVAL;
830
831 if (!si_fromuser(info))
832 return 0;
833
834 error = audit_signal_info(sig, t); /* Let audit system see the signal */
835 if (error)
836 return error;
837
838 if (!same_thread_group(current, t) &&
839 !kill_ok_by_cred(t)) {
840 switch (sig) {
841 case SIGCONT:
842 sid = task_session(t);
843 /*
844 * We don't return the error if sid == NULL. The
845 * task was unhashed, the caller must notice this.
846 */
847 if (!sid || sid == task_session(current))
848 break;
849 /* fall through */
850 default:
851 return -EPERM;
852 }
853 }
854
855 return security_task_kill(t, info, sig, NULL);
856}
857
858/**
859 * ptrace_trap_notify - schedule trap to notify ptracer
860 * @t: tracee wanting to notify tracer
861 *
862 * This function schedules sticky ptrace trap which is cleared on the next
863 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
864 * ptracer.
865 *
866 * If @t is running, STOP trap will be taken. If trapped for STOP and
867 * ptracer is listening for events, tracee is woken up so that it can
868 * re-trap for the new event. If trapped otherwise, STOP trap will be
869 * eventually taken without returning to userland after the existing traps
870 * are finished by PTRACE_CONT.
871 *
872 * CONTEXT:
873 * Must be called with @task->sighand->siglock held.
874 */
875static void ptrace_trap_notify(struct task_struct *t)
876{
877 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
878 assert_spin_locked(&t->sighand->siglock);
879
880 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
881 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
882}
883
884/*
885 * Handle magic process-wide effects of stop/continue signals. Unlike
886 * the signal actions, these happen immediately at signal-generation
887 * time regardless of blocking, ignoring, or handling. This does the
888 * actual continuing for SIGCONT, but not the actual stopping for stop
889 * signals. The process stop is done as a signal action for SIG_DFL.
890 *
891 * Returns true if the signal should be actually delivered, otherwise
892 * it should be dropped.
893 */
894static bool prepare_signal(int sig, struct task_struct *p, bool force)
895{
896 struct signal_struct *signal = p->signal;
897 struct task_struct *t;
898 sigset_t flush;
899
900 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
901 if (!(signal->flags & SIGNAL_GROUP_EXIT))
902 return sig == SIGKILL;
903 /*
904 * The process is in the middle of dying, nothing to do.
905 */
906 } else if (sig_kernel_stop(sig)) {
907 /*
908 * This is a stop signal. Remove SIGCONT from all queues.
909 */
910 siginitset(&flush, sigmask(SIGCONT));
911 flush_sigqueue_mask(&flush, &signal->shared_pending);
912 for_each_thread(p, t)
913 flush_sigqueue_mask(&flush, &t->pending);
914 } else if (sig == SIGCONT) {
915 unsigned int why;
916 /*
917 * Remove all stop signals from all queues, wake all threads.
918 */
919 siginitset(&flush, SIG_KERNEL_STOP_MASK);
920 flush_sigqueue_mask(&flush, &signal->shared_pending);
921 for_each_thread(p, t) {
922 flush_sigqueue_mask(&flush, &t->pending);
923 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
924 if (likely(!(t->ptrace & PT_SEIZED)))
925 wake_up_state(t, __TASK_STOPPED);
926 else
927 ptrace_trap_notify(t);
928 }
929
930 /*
931 * Notify the parent with CLD_CONTINUED if we were stopped.
932 *
933 * If we were in the middle of a group stop, we pretend it
934 * was already finished, and then continued. Since SIGCHLD
935 * doesn't queue we report only CLD_STOPPED, as if the next
936 * CLD_CONTINUED was dropped.
937 */
938 why = 0;
939 if (signal->flags & SIGNAL_STOP_STOPPED)
940 why |= SIGNAL_CLD_CONTINUED;
941 else if (signal->group_stop_count)
942 why |= SIGNAL_CLD_STOPPED;
943
944 if (why) {
945 /*
946 * The first thread which returns from do_signal_stop()
947 * will take ->siglock, notice SIGNAL_CLD_MASK, and
948 * notify its parent. See get_signal().
949 */
950 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
951 signal->group_stop_count = 0;
952 signal->group_exit_code = 0;
953 }
954 }
955
956 return !sig_ignored(p, sig, force);
957}
958
959/*
960 * Test if P wants to take SIG. After we've checked all threads with this,
961 * it's equivalent to finding no threads not blocking SIG. Any threads not
962 * blocking SIG were ruled out because they are not running and already
963 * have pending signals. Such threads will dequeue from the shared queue
964 * as soon as they're available, so putting the signal on the shared queue
965 * will be equivalent to sending it to one such thread.
966 */
967static inline bool wants_signal(int sig, struct task_struct *p)
968{
969 if (sigismember(&p->blocked, sig))
970 return false;
971
972 if (p->flags & PF_EXITING)
973 return false;
974
975 if (sig == SIGKILL)
976 return true;
977
978 if (task_is_stopped_or_traced(p))
979 return false;
980
981 return task_curr(p) || !signal_pending(p);
982}
983
984static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
985{
986 struct signal_struct *signal = p->signal;
987 struct task_struct *t;
988
989 /*
990 * Now find a thread we can wake up to take the signal off the queue.
991 *
992 * If the main thread wants the signal, it gets first crack.
993 * Probably the least surprising to the average bear.
994 */
995 if (wants_signal(sig, p))
996 t = p;
997 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
998 /*
999 * There is just one thread and it does not need to be woken.
1000 * It will dequeue unblocked signals before it runs again.
1001 */
1002 return;
1003 else {
1004 /*
1005 * Otherwise try to find a suitable thread.
1006 */
1007 t = signal->curr_target;
1008 while (!wants_signal(sig, t)) {
1009 t = next_thread(t);
1010 if (t == signal->curr_target)
1011 /*
1012 * No thread needs to be woken.
1013 * Any eligible threads will see
1014 * the signal in the queue soon.
1015 */
1016 return;
1017 }
1018 signal->curr_target = t;
1019 }
1020
1021 /*
1022 * Found a killable thread. If the signal will be fatal,
1023 * then start taking the whole group down immediately.
1024 */
1025 if (sig_fatal(p, sig) &&
1026 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1027 !sigismember(&t->real_blocked, sig) &&
1028 (sig == SIGKILL || !p->ptrace)) {
1029 /*
1030 * This signal will be fatal to the whole group.
1031 */
1032 if (!sig_kernel_coredump(sig)) {
1033 /*
1034 * Start a group exit and wake everybody up.
1035 * This way we don't have other threads
1036 * running and doing things after a slower
1037 * thread has the fatal signal pending.
1038 */
1039 signal->flags = SIGNAL_GROUP_EXIT;
1040 signal->group_exit_code = sig;
1041 signal->group_stop_count = 0;
1042 t = p;
1043 do {
1044 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1045 sigaddset(&t->pending.signal, SIGKILL);
1046 signal_wake_up(t, 1);
1047 } while_each_thread(p, t);
1048 return;
1049 }
1050 }
1051
1052 /*
1053 * The signal is already in the shared-pending queue.
1054 * Tell the chosen thread to wake up and dequeue it.
1055 */
1056 signal_wake_up(t, sig == SIGKILL);
1057 return;
1058}
1059
1060static inline bool legacy_queue(struct sigpending *signals, int sig)
1061{
1062 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1063}
1064
1065static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1066 enum pid_type type, bool force)
1067{
1068 struct sigpending *pending;
1069 struct sigqueue *q;
1070 int override_rlimit;
1071 int ret = 0, result;
1072
1073 assert_spin_locked(&t->sighand->siglock);
1074
1075 result = TRACE_SIGNAL_IGNORED;
1076 if (!prepare_signal(sig, t, force))
1077 goto ret;
1078
1079 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1080 /*
1081 * Short-circuit ignored signals and support queuing
1082 * exactly one non-rt signal, so that we can get more
1083 * detailed information about the cause of the signal.
1084 */
1085 result = TRACE_SIGNAL_ALREADY_PENDING;
1086 if (legacy_queue(pending, sig))
1087 goto ret;
1088
1089 result = TRACE_SIGNAL_DELIVERED;
1090 /*
1091 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1092 */
1093 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1094 goto out_set;
1095
1096 /*
1097 * Real-time signals must be queued if sent by sigqueue, or
1098 * some other real-time mechanism. It is implementation
1099 * defined whether kill() does so. We attempt to do so, on
1100 * the principle of least surprise, but since kill is not
1101 * allowed to fail with EAGAIN when low on memory we just
1102 * make sure at least one signal gets delivered and don't
1103 * pass on the info struct.
1104 */
1105 if (sig < SIGRTMIN)
1106 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1107 else
1108 override_rlimit = 0;
1109
1110 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit);
1111 if (q) {
1112 list_add_tail(&q->list, &pending->list);
1113 switch ((unsigned long) info) {
1114 case (unsigned long) SEND_SIG_NOINFO:
1115 clear_siginfo(&q->info);
1116 q->info.si_signo = sig;
1117 q->info.si_errno = 0;
1118 q->info.si_code = SI_USER;
1119 q->info.si_pid = task_tgid_nr_ns(current,
1120 task_active_pid_ns(t));
1121 rcu_read_lock();
1122 q->info.si_uid =
1123 from_kuid_munged(task_cred_xxx(t, user_ns),
1124 current_uid());
1125 rcu_read_unlock();
1126 break;
1127 case (unsigned long) SEND_SIG_PRIV:
1128 clear_siginfo(&q->info);
1129 q->info.si_signo = sig;
1130 q->info.si_errno = 0;
1131 q->info.si_code = SI_KERNEL;
1132 q->info.si_pid = 0;
1133 q->info.si_uid = 0;
1134 break;
1135 default:
1136 copy_siginfo(&q->info, info);
1137 break;
1138 }
1139 } else if (!is_si_special(info) &&
1140 sig >= SIGRTMIN && info->si_code != SI_USER) {
1141 /*
1142 * Queue overflow, abort. We may abort if the
1143 * signal was rt and sent by user using something
1144 * other than kill().
1145 */
1146 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1147 ret = -EAGAIN;
1148 goto ret;
1149 } else {
1150 /*
1151 * This is a silent loss of information. We still
1152 * send the signal, but the *info bits are lost.
1153 */
1154 result = TRACE_SIGNAL_LOSE_INFO;
1155 }
1156
1157out_set:
1158 signalfd_notify(t, sig);
1159 sigaddset(&pending->signal, sig);
1160
1161 /* Let multiprocess signals appear after on-going forks */
1162 if (type > PIDTYPE_TGID) {
1163 struct multiprocess_signals *delayed;
1164 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1165 sigset_t *signal = &delayed->signal;
1166 /* Can't queue both a stop and a continue signal */
1167 if (sig == SIGCONT)
1168 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1169 else if (sig_kernel_stop(sig))
1170 sigdelset(signal, SIGCONT);
1171 sigaddset(signal, sig);
1172 }
1173 }
1174
1175 complete_signal(sig, t, type);
1176ret:
1177 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1178 return ret;
1179}
1180
1181static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1182{
1183 bool ret = false;
1184 switch (siginfo_layout(info->si_signo, info->si_code)) {
1185 case SIL_KILL:
1186 case SIL_CHLD:
1187 case SIL_RT:
1188 ret = true;
1189 break;
1190 case SIL_TIMER:
1191 case SIL_POLL:
1192 case SIL_FAULT:
1193 case SIL_FAULT_MCEERR:
1194 case SIL_FAULT_BNDERR:
1195 case SIL_FAULT_PKUERR:
1196 case SIL_SYS:
1197 ret = false;
1198 break;
1199 }
1200 return ret;
1201}
1202
1203static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1204 enum pid_type type)
1205{
1206 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1207 bool force = false;
1208
1209 if (info == SEND_SIG_NOINFO) {
1210 /* Force if sent from an ancestor pid namespace */
1211 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1212 } else if (info == SEND_SIG_PRIV) {
1213 /* Don't ignore kernel generated signals */
1214 force = true;
1215 } else if (has_si_pid_and_uid(info)) {
1216 /* SIGKILL and SIGSTOP is special or has ids */
1217 struct user_namespace *t_user_ns;
1218
1219 rcu_read_lock();
1220 t_user_ns = task_cred_xxx(t, user_ns);
1221 if (current_user_ns() != t_user_ns) {
1222 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1223 info->si_uid = from_kuid_munged(t_user_ns, uid);
1224 }
1225 rcu_read_unlock();
1226
1227 /* A kernel generated signal? */
1228 force = (info->si_code == SI_KERNEL);
1229
1230 /* From an ancestor pid namespace? */
1231 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1232 info->si_pid = 0;
1233 force = true;
1234 }
1235 }
1236 return __send_signal(sig, info, t, type, force);
1237}
1238
1239static void print_fatal_signal(int signr)
1240{
1241 struct pt_regs *regs = signal_pt_regs();
1242 pr_info("potentially unexpected fatal signal %d.\n", signr);
1243
1244#if defined(__i386__) && !defined(__arch_um__)
1245 pr_info("code at %08lx: ", regs->ip);
1246 {
1247 int i;
1248 for (i = 0; i < 16; i++) {
1249 unsigned char insn;
1250
1251 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1252 break;
1253 pr_cont("%02x ", insn);
1254 }
1255 }
1256 pr_cont("\n");
1257#endif
1258 preempt_disable();
1259 show_regs(regs);
1260 preempt_enable();
1261}
1262
1263static int __init setup_print_fatal_signals(char *str)
1264{
1265 get_option (&str, &print_fatal_signals);
1266
1267 return 1;
1268}
1269
1270__setup("print-fatal-signals=", setup_print_fatal_signals);
1271
1272int
1273__group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1274{
1275 return send_signal(sig, info, p, PIDTYPE_TGID);
1276}
1277
1278int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1279 enum pid_type type)
1280{
1281 unsigned long flags;
1282 int ret = -ESRCH;
1283
1284 if (lock_task_sighand(p, &flags)) {
1285 ret = send_signal(sig, info, p, type);
1286 unlock_task_sighand(p, &flags);
1287 }
1288
1289 return ret;
1290}
1291
1292/*
1293 * Force a signal that the process can't ignore: if necessary
1294 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1295 *
1296 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1297 * since we do not want to have a signal handler that was blocked
1298 * be invoked when user space had explicitly blocked it.
1299 *
1300 * We don't want to have recursive SIGSEGV's etc, for example,
1301 * that is why we also clear SIGNAL_UNKILLABLE.
1302 */
1303static int
1304force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1305{
1306 unsigned long int flags;
1307 int ret, blocked, ignored;
1308 struct k_sigaction *action;
1309 int sig = info->si_signo;
1310
1311 spin_lock_irqsave(&t->sighand->siglock, flags);
1312 action = &t->sighand->action[sig-1];
1313 ignored = action->sa.sa_handler == SIG_IGN;
1314 blocked = sigismember(&t->blocked, sig);
1315 if (blocked || ignored) {
1316 action->sa.sa_handler = SIG_DFL;
1317 if (blocked) {
1318 sigdelset(&t->blocked, sig);
1319 recalc_sigpending_and_wake(t);
1320 }
1321 }
1322 /*
1323 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1324 * debugging to leave init killable.
1325 */
1326 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1327 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1328 ret = send_signal(sig, info, t, PIDTYPE_PID);
1329 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1330
1331 return ret;
1332}
1333
1334int force_sig_info(struct kernel_siginfo *info)
1335{
1336 return force_sig_info_to_task(info, current);
1337}
1338
1339/*
1340 * Nuke all other threads in the group.
1341 */
1342int zap_other_threads(struct task_struct *p)
1343{
1344 struct task_struct *t = p;
1345 int count = 0;
1346
1347 p->signal->group_stop_count = 0;
1348
1349 while_each_thread(p, t) {
1350 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1351 count++;
1352
1353 /* Don't bother with already dead threads */
1354 if (t->exit_state)
1355 continue;
1356 sigaddset(&t->pending.signal, SIGKILL);
1357 signal_wake_up(t, 1);
1358 }
1359
1360 return count;
1361}
1362
1363struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1364 unsigned long *flags)
1365{
1366 struct sighand_struct *sighand;
1367
1368 rcu_read_lock();
1369 for (;;) {
1370 sighand = rcu_dereference(tsk->sighand);
1371 if (unlikely(sighand == NULL))
1372 break;
1373
1374 /*
1375 * This sighand can be already freed and even reused, but
1376 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1377 * initializes ->siglock: this slab can't go away, it has
1378 * the same object type, ->siglock can't be reinitialized.
1379 *
1380 * We need to ensure that tsk->sighand is still the same
1381 * after we take the lock, we can race with de_thread() or
1382 * __exit_signal(). In the latter case the next iteration
1383 * must see ->sighand == NULL.
1384 */
1385 spin_lock_irqsave(&sighand->siglock, *flags);
1386 if (likely(sighand == tsk->sighand))
1387 break;
1388 spin_unlock_irqrestore(&sighand->siglock, *flags);
1389 }
1390 rcu_read_unlock();
1391
1392 return sighand;
1393}
1394
1395/*
1396 * send signal info to all the members of a group
1397 */
1398int group_send_sig_info(int sig, struct kernel_siginfo *info,
1399 struct task_struct *p, enum pid_type type)
1400{
1401 int ret;
1402
1403 rcu_read_lock();
1404 ret = check_kill_permission(sig, info, p);
1405 rcu_read_unlock();
1406
1407 if (!ret && sig)
1408 ret = do_send_sig_info(sig, info, p, type);
1409
1410 return ret;
1411}
1412
1413/*
1414 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1415 * control characters do (^C, ^Z etc)
1416 * - the caller must hold at least a readlock on tasklist_lock
1417 */
1418int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1419{
1420 struct task_struct *p = NULL;
1421 int retval, success;
1422
1423 success = 0;
1424 retval = -ESRCH;
1425 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1426 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1427 success |= !err;
1428 retval = err;
1429 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1430 return success ? 0 : retval;
1431}
1432
1433int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1434{
1435 int error = -ESRCH;
1436 struct task_struct *p;
1437
1438 for (;;) {
1439 rcu_read_lock();
1440 p = pid_task(pid, PIDTYPE_PID);
1441 if (p)
1442 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1443 rcu_read_unlock();
1444 if (likely(!p || error != -ESRCH))
1445 return error;
1446
1447 /*
1448 * The task was unhashed in between, try again. If it
1449 * is dead, pid_task() will return NULL, if we race with
1450 * de_thread() it will find the new leader.
1451 */
1452 }
1453}
1454
1455static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1456{
1457 int error;
1458 rcu_read_lock();
1459 error = kill_pid_info(sig, info, find_vpid(pid));
1460 rcu_read_unlock();
1461 return error;
1462}
1463
1464static inline bool kill_as_cred_perm(const struct cred *cred,
1465 struct task_struct *target)
1466{
1467 const struct cred *pcred = __task_cred(target);
1468
1469 return uid_eq(cred->euid, pcred->suid) ||
1470 uid_eq(cred->euid, pcred->uid) ||
1471 uid_eq(cred->uid, pcred->suid) ||
1472 uid_eq(cred->uid, pcred->uid);
1473}
1474
1475/*
1476 * The usb asyncio usage of siginfo is wrong. The glibc support
1477 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1478 * AKA after the generic fields:
1479 * kernel_pid_t si_pid;
1480 * kernel_uid32_t si_uid;
1481 * sigval_t si_value;
1482 *
1483 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1484 * after the generic fields is:
1485 * void __user *si_addr;
1486 *
1487 * This is a practical problem when there is a 64bit big endian kernel
1488 * and a 32bit userspace. As the 32bit address will encoded in the low
1489 * 32bits of the pointer. Those low 32bits will be stored at higher
1490 * address than appear in a 32 bit pointer. So userspace will not
1491 * see the address it was expecting for it's completions.
1492 *
1493 * There is nothing in the encoding that can allow
1494 * copy_siginfo_to_user32 to detect this confusion of formats, so
1495 * handle this by requiring the caller of kill_pid_usb_asyncio to
1496 * notice when this situration takes place and to store the 32bit
1497 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1498 * parameter.
1499 */
1500int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1501 struct pid *pid, const struct cred *cred)
1502{
1503 struct kernel_siginfo info;
1504 struct task_struct *p;
1505 unsigned long flags;
1506 int ret = -EINVAL;
1507
1508 clear_siginfo(&info);
1509 info.si_signo = sig;
1510 info.si_errno = errno;
1511 info.si_code = SI_ASYNCIO;
1512 *((sigval_t *)&info.si_pid) = addr;
1513
1514 if (!valid_signal(sig))
1515 return ret;
1516
1517 rcu_read_lock();
1518 p = pid_task(pid, PIDTYPE_PID);
1519 if (!p) {
1520 ret = -ESRCH;
1521 goto out_unlock;
1522 }
1523 if (!kill_as_cred_perm(cred, p)) {
1524 ret = -EPERM;
1525 goto out_unlock;
1526 }
1527 ret = security_task_kill(p, &info, sig, cred);
1528 if (ret)
1529 goto out_unlock;
1530
1531 if (sig) {
1532 if (lock_task_sighand(p, &flags)) {
1533 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1534 unlock_task_sighand(p, &flags);
1535 } else
1536 ret = -ESRCH;
1537 }
1538out_unlock:
1539 rcu_read_unlock();
1540 return ret;
1541}
1542EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1543
1544/*
1545 * kill_something_info() interprets pid in interesting ways just like kill(2).
1546 *
1547 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1548 * is probably wrong. Should make it like BSD or SYSV.
1549 */
1550
1551static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1552{
1553 int ret;
1554
1555 if (pid > 0) {
1556 rcu_read_lock();
1557 ret = kill_pid_info(sig, info, find_vpid(pid));
1558 rcu_read_unlock();
1559 return ret;
1560 }
1561
1562 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1563 if (pid == INT_MIN)
1564 return -ESRCH;
1565
1566 read_lock(&tasklist_lock);
1567 if (pid != -1) {
1568 ret = __kill_pgrp_info(sig, info,
1569 pid ? find_vpid(-pid) : task_pgrp(current));
1570 } else {
1571 int retval = 0, count = 0;
1572 struct task_struct * p;
1573
1574 for_each_process(p) {
1575 if (task_pid_vnr(p) > 1 &&
1576 !same_thread_group(p, current)) {
1577 int err = group_send_sig_info(sig, info, p,
1578 PIDTYPE_MAX);
1579 ++count;
1580 if (err != -EPERM)
1581 retval = err;
1582 }
1583 }
1584 ret = count ? retval : -ESRCH;
1585 }
1586 read_unlock(&tasklist_lock);
1587
1588 return ret;
1589}
1590
1591/*
1592 * These are for backward compatibility with the rest of the kernel source.
1593 */
1594
1595int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1596{
1597 /*
1598 * Make sure legacy kernel users don't send in bad values
1599 * (normal paths check this in check_kill_permission).
1600 */
1601 if (!valid_signal(sig))
1602 return -EINVAL;
1603
1604 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1605}
1606EXPORT_SYMBOL(send_sig_info);
1607
1608#define __si_special(priv) \
1609 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1610
1611int
1612send_sig(int sig, struct task_struct *p, int priv)
1613{
1614 return send_sig_info(sig, __si_special(priv), p);
1615}
1616EXPORT_SYMBOL(send_sig);
1617
1618void force_sig(int sig)
1619{
1620 struct kernel_siginfo info;
1621
1622 clear_siginfo(&info);
1623 info.si_signo = sig;
1624 info.si_errno = 0;
1625 info.si_code = SI_KERNEL;
1626 info.si_pid = 0;
1627 info.si_uid = 0;
1628 force_sig_info(&info);
1629}
1630EXPORT_SYMBOL(force_sig);
1631
1632/*
1633 * When things go south during signal handling, we
1634 * will force a SIGSEGV. And if the signal that caused
1635 * the problem was already a SIGSEGV, we'll want to
1636 * make sure we don't even try to deliver the signal..
1637 */
1638void force_sigsegv(int sig)
1639{
1640 struct task_struct *p = current;
1641
1642 if (sig == SIGSEGV) {
1643 unsigned long flags;
1644 spin_lock_irqsave(&p->sighand->siglock, flags);
1645 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1646 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1647 }
1648 force_sig(SIGSEGV);
1649}
1650
1651int force_sig_fault_to_task(int sig, int code, void __user *addr
1652 ___ARCH_SI_TRAPNO(int trapno)
1653 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1654 , struct task_struct *t)
1655{
1656 struct kernel_siginfo info;
1657
1658 clear_siginfo(&info);
1659 info.si_signo = sig;
1660 info.si_errno = 0;
1661 info.si_code = code;
1662 info.si_addr = addr;
1663#ifdef __ARCH_SI_TRAPNO
1664 info.si_trapno = trapno;
1665#endif
1666#ifdef __ia64__
1667 info.si_imm = imm;
1668 info.si_flags = flags;
1669 info.si_isr = isr;
1670#endif
1671 return force_sig_info_to_task(&info, t);
1672}
1673
1674int force_sig_fault(int sig, int code, void __user *addr
1675 ___ARCH_SI_TRAPNO(int trapno)
1676 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1677{
1678 return force_sig_fault_to_task(sig, code, addr
1679 ___ARCH_SI_TRAPNO(trapno)
1680 ___ARCH_SI_IA64(imm, flags, isr), current);
1681}
1682
1683int send_sig_fault(int sig, int code, void __user *addr
1684 ___ARCH_SI_TRAPNO(int trapno)
1685 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1686 , struct task_struct *t)
1687{
1688 struct kernel_siginfo info;
1689
1690 clear_siginfo(&info);
1691 info.si_signo = sig;
1692 info.si_errno = 0;
1693 info.si_code = code;
1694 info.si_addr = addr;
1695#ifdef __ARCH_SI_TRAPNO
1696 info.si_trapno = trapno;
1697#endif
1698#ifdef __ia64__
1699 info.si_imm = imm;
1700 info.si_flags = flags;
1701 info.si_isr = isr;
1702#endif
1703 return send_sig_info(info.si_signo, &info, t);
1704}
1705
1706int force_sig_mceerr(int code, void __user *addr, short lsb)
1707{
1708 struct kernel_siginfo info;
1709
1710 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1711 clear_siginfo(&info);
1712 info.si_signo = SIGBUS;
1713 info.si_errno = 0;
1714 info.si_code = code;
1715 info.si_addr = addr;
1716 info.si_addr_lsb = lsb;
1717 return force_sig_info(&info);
1718}
1719
1720int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1721{
1722 struct kernel_siginfo info;
1723
1724 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1725 clear_siginfo(&info);
1726 info.si_signo = SIGBUS;
1727 info.si_errno = 0;
1728 info.si_code = code;
1729 info.si_addr = addr;
1730 info.si_addr_lsb = lsb;
1731 return send_sig_info(info.si_signo, &info, t);
1732}
1733EXPORT_SYMBOL(send_sig_mceerr);
1734
1735int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1736{
1737 struct kernel_siginfo info;
1738
1739 clear_siginfo(&info);
1740 info.si_signo = SIGSEGV;
1741 info.si_errno = 0;
1742 info.si_code = SEGV_BNDERR;
1743 info.si_addr = addr;
1744 info.si_lower = lower;
1745 info.si_upper = upper;
1746 return force_sig_info(&info);
1747}
1748
1749#ifdef SEGV_PKUERR
1750int force_sig_pkuerr(void __user *addr, u32 pkey)
1751{
1752 struct kernel_siginfo info;
1753
1754 clear_siginfo(&info);
1755 info.si_signo = SIGSEGV;
1756 info.si_errno = 0;
1757 info.si_code = SEGV_PKUERR;
1758 info.si_addr = addr;
1759 info.si_pkey = pkey;
1760 return force_sig_info(&info);
1761}
1762#endif
1763
1764/* For the crazy architectures that include trap information in
1765 * the errno field, instead of an actual errno value.
1766 */
1767int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1768{
1769 struct kernel_siginfo info;
1770
1771 clear_siginfo(&info);
1772 info.si_signo = SIGTRAP;
1773 info.si_errno = errno;
1774 info.si_code = TRAP_HWBKPT;
1775 info.si_addr = addr;
1776 return force_sig_info(&info);
1777}
1778
1779int kill_pgrp(struct pid *pid, int sig, int priv)
1780{
1781 int ret;
1782
1783 read_lock(&tasklist_lock);
1784 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1785 read_unlock(&tasklist_lock);
1786
1787 return ret;
1788}
1789EXPORT_SYMBOL(kill_pgrp);
1790
1791int kill_pid(struct pid *pid, int sig, int priv)
1792{
1793 return kill_pid_info(sig, __si_special(priv), pid);
1794}
1795EXPORT_SYMBOL(kill_pid);
1796
1797/*
1798 * These functions support sending signals using preallocated sigqueue
1799 * structures. This is needed "because realtime applications cannot
1800 * afford to lose notifications of asynchronous events, like timer
1801 * expirations or I/O completions". In the case of POSIX Timers
1802 * we allocate the sigqueue structure from the timer_create. If this
1803 * allocation fails we are able to report the failure to the application
1804 * with an EAGAIN error.
1805 */
1806struct sigqueue *sigqueue_alloc(void)
1807{
1808 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1809
1810 if (q)
1811 q->flags |= SIGQUEUE_PREALLOC;
1812
1813 return q;
1814}
1815
1816void sigqueue_free(struct sigqueue *q)
1817{
1818 unsigned long flags;
1819 spinlock_t *lock = ¤t->sighand->siglock;
1820
1821 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1822 /*
1823 * We must hold ->siglock while testing q->list
1824 * to serialize with collect_signal() or with
1825 * __exit_signal()->flush_sigqueue().
1826 */
1827 spin_lock_irqsave(lock, flags);
1828 q->flags &= ~SIGQUEUE_PREALLOC;
1829 /*
1830 * If it is queued it will be freed when dequeued,
1831 * like the "regular" sigqueue.
1832 */
1833 if (!list_empty(&q->list))
1834 q = NULL;
1835 spin_unlock_irqrestore(lock, flags);
1836
1837 if (q)
1838 __sigqueue_free(q);
1839}
1840
1841int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1842{
1843 int sig = q->info.si_signo;
1844 struct sigpending *pending;
1845 struct task_struct *t;
1846 unsigned long flags;
1847 int ret, result;
1848
1849 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1850
1851 ret = -1;
1852 rcu_read_lock();
1853 t = pid_task(pid, type);
1854 if (!t || !likely(lock_task_sighand(t, &flags)))
1855 goto ret;
1856
1857 ret = 1; /* the signal is ignored */
1858 result = TRACE_SIGNAL_IGNORED;
1859 if (!prepare_signal(sig, t, false))
1860 goto out;
1861
1862 ret = 0;
1863 if (unlikely(!list_empty(&q->list))) {
1864 /*
1865 * If an SI_TIMER entry is already queue just increment
1866 * the overrun count.
1867 */
1868 BUG_ON(q->info.si_code != SI_TIMER);
1869 q->info.si_overrun++;
1870 result = TRACE_SIGNAL_ALREADY_PENDING;
1871 goto out;
1872 }
1873 q->info.si_overrun = 0;
1874
1875 signalfd_notify(t, sig);
1876 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1877 list_add_tail(&q->list, &pending->list);
1878 sigaddset(&pending->signal, sig);
1879 complete_signal(sig, t, type);
1880 result = TRACE_SIGNAL_DELIVERED;
1881out:
1882 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1883 unlock_task_sighand(t, &flags);
1884ret:
1885 rcu_read_unlock();
1886 return ret;
1887}
1888
1889static void do_notify_pidfd(struct task_struct *task)
1890{
1891 struct pid *pid;
1892
1893 WARN_ON(task->exit_state == 0);
1894 pid = task_pid(task);
1895 wake_up_all(&pid->wait_pidfd);
1896}
1897
1898/*
1899 * Let a parent know about the death of a child.
1900 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1901 *
1902 * Returns true if our parent ignored us and so we've switched to
1903 * self-reaping.
1904 */
1905bool do_notify_parent(struct task_struct *tsk, int sig)
1906{
1907 struct kernel_siginfo info;
1908 unsigned long flags;
1909 struct sighand_struct *psig;
1910 bool autoreap = false;
1911 u64 utime, stime;
1912
1913 BUG_ON(sig == -1);
1914
1915 /* do_notify_parent_cldstop should have been called instead. */
1916 BUG_ON(task_is_stopped_or_traced(tsk));
1917
1918 BUG_ON(!tsk->ptrace &&
1919 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1920
1921 /* Wake up all pidfd waiters */
1922 do_notify_pidfd(tsk);
1923
1924 if (sig != SIGCHLD) {
1925 /*
1926 * This is only possible if parent == real_parent.
1927 * Check if it has changed security domain.
1928 */
1929 if (tsk->parent_exec_id != tsk->parent->self_exec_id)
1930 sig = SIGCHLD;
1931 }
1932
1933 clear_siginfo(&info);
1934 info.si_signo = sig;
1935 info.si_errno = 0;
1936 /*
1937 * We are under tasklist_lock here so our parent is tied to
1938 * us and cannot change.
1939 *
1940 * task_active_pid_ns will always return the same pid namespace
1941 * until a task passes through release_task.
1942 *
1943 * write_lock() currently calls preempt_disable() which is the
1944 * same as rcu_read_lock(), but according to Oleg, this is not
1945 * correct to rely on this
1946 */
1947 rcu_read_lock();
1948 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1949 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1950 task_uid(tsk));
1951 rcu_read_unlock();
1952
1953 task_cputime(tsk, &utime, &stime);
1954 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1955 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1956
1957 info.si_status = tsk->exit_code & 0x7f;
1958 if (tsk->exit_code & 0x80)
1959 info.si_code = CLD_DUMPED;
1960 else if (tsk->exit_code & 0x7f)
1961 info.si_code = CLD_KILLED;
1962 else {
1963 info.si_code = CLD_EXITED;
1964 info.si_status = tsk->exit_code >> 8;
1965 }
1966
1967 psig = tsk->parent->sighand;
1968 spin_lock_irqsave(&psig->siglock, flags);
1969 if (!tsk->ptrace && sig == SIGCHLD &&
1970 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1971 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1972 /*
1973 * We are exiting and our parent doesn't care. POSIX.1
1974 * defines special semantics for setting SIGCHLD to SIG_IGN
1975 * or setting the SA_NOCLDWAIT flag: we should be reaped
1976 * automatically and not left for our parent's wait4 call.
1977 * Rather than having the parent do it as a magic kind of
1978 * signal handler, we just set this to tell do_exit that we
1979 * can be cleaned up without becoming a zombie. Note that
1980 * we still call __wake_up_parent in this case, because a
1981 * blocked sys_wait4 might now return -ECHILD.
1982 *
1983 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1984 * is implementation-defined: we do (if you don't want
1985 * it, just use SIG_IGN instead).
1986 */
1987 autoreap = true;
1988 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1989 sig = 0;
1990 }
1991 if (valid_signal(sig) && sig)
1992 __group_send_sig_info(sig, &info, tsk->parent);
1993 __wake_up_parent(tsk, tsk->parent);
1994 spin_unlock_irqrestore(&psig->siglock, flags);
1995
1996 return autoreap;
1997}
1998
1999/**
2000 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2001 * @tsk: task reporting the state change
2002 * @for_ptracer: the notification is for ptracer
2003 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2004 *
2005 * Notify @tsk's parent that the stopped/continued state has changed. If
2006 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2007 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2008 *
2009 * CONTEXT:
2010 * Must be called with tasklist_lock at least read locked.
2011 */
2012static void do_notify_parent_cldstop(struct task_struct *tsk,
2013 bool for_ptracer, int why)
2014{
2015 struct kernel_siginfo info;
2016 unsigned long flags;
2017 struct task_struct *parent;
2018 struct sighand_struct *sighand;
2019 u64 utime, stime;
2020
2021 if (for_ptracer) {
2022 parent = tsk->parent;
2023 } else {
2024 tsk = tsk->group_leader;
2025 parent = tsk->real_parent;
2026 }
2027
2028 clear_siginfo(&info);
2029 info.si_signo = SIGCHLD;
2030 info.si_errno = 0;
2031 /*
2032 * see comment in do_notify_parent() about the following 4 lines
2033 */
2034 rcu_read_lock();
2035 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2036 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2037 rcu_read_unlock();
2038
2039 task_cputime(tsk, &utime, &stime);
2040 info.si_utime = nsec_to_clock_t(utime);
2041 info.si_stime = nsec_to_clock_t(stime);
2042
2043 info.si_code = why;
2044 switch (why) {
2045 case CLD_CONTINUED:
2046 info.si_status = SIGCONT;
2047 break;
2048 case CLD_STOPPED:
2049 info.si_status = tsk->signal->group_exit_code & 0x7f;
2050 break;
2051 case CLD_TRAPPED:
2052 info.si_status = tsk->exit_code & 0x7f;
2053 break;
2054 default:
2055 BUG();
2056 }
2057
2058 sighand = parent->sighand;
2059 spin_lock_irqsave(&sighand->siglock, flags);
2060 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2061 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2062 __group_send_sig_info(SIGCHLD, &info, parent);
2063 /*
2064 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2065 */
2066 __wake_up_parent(tsk, parent);
2067 spin_unlock_irqrestore(&sighand->siglock, flags);
2068}
2069
2070static inline bool may_ptrace_stop(void)
2071{
2072 if (!likely(current->ptrace))
2073 return false;
2074 /*
2075 * Are we in the middle of do_coredump?
2076 * If so and our tracer is also part of the coredump stopping
2077 * is a deadlock situation, and pointless because our tracer
2078 * is dead so don't allow us to stop.
2079 * If SIGKILL was already sent before the caller unlocked
2080 * ->siglock we must see ->core_state != NULL. Otherwise it
2081 * is safe to enter schedule().
2082 *
2083 * This is almost outdated, a task with the pending SIGKILL can't
2084 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2085 * after SIGKILL was already dequeued.
2086 */
2087 if (unlikely(current->mm->core_state) &&
2088 unlikely(current->mm == current->parent->mm))
2089 return false;
2090
2091 return true;
2092}
2093
2094/*
2095 * Return non-zero if there is a SIGKILL that should be waking us up.
2096 * Called with the siglock held.
2097 */
2098static bool sigkill_pending(struct task_struct *tsk)
2099{
2100 return sigismember(&tsk->pending.signal, SIGKILL) ||
2101 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2102}
2103
2104/*
2105 * This must be called with current->sighand->siglock held.
2106 *
2107 * This should be the path for all ptrace stops.
2108 * We always set current->last_siginfo while stopped here.
2109 * That makes it a way to test a stopped process for
2110 * being ptrace-stopped vs being job-control-stopped.
2111 *
2112 * If we actually decide not to stop at all because the tracer
2113 * is gone, we keep current->exit_code unless clear_code.
2114 */
2115static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2116 __releases(¤t->sighand->siglock)
2117 __acquires(¤t->sighand->siglock)
2118{
2119 bool gstop_done = false;
2120
2121 if (arch_ptrace_stop_needed(exit_code, info)) {
2122 /*
2123 * The arch code has something special to do before a
2124 * ptrace stop. This is allowed to block, e.g. for faults
2125 * on user stack pages. We can't keep the siglock while
2126 * calling arch_ptrace_stop, so we must release it now.
2127 * To preserve proper semantics, we must do this before
2128 * any signal bookkeeping like checking group_stop_count.
2129 * Meanwhile, a SIGKILL could come in before we retake the
2130 * siglock. That must prevent us from sleeping in TASK_TRACED.
2131 * So after regaining the lock, we must check for SIGKILL.
2132 */
2133 spin_unlock_irq(¤t->sighand->siglock);
2134 arch_ptrace_stop(exit_code, info);
2135 spin_lock_irq(¤t->sighand->siglock);
2136 if (sigkill_pending(current))
2137 return;
2138 }
2139
2140 set_special_state(TASK_TRACED);
2141
2142 /*
2143 * We're committing to trapping. TRACED should be visible before
2144 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2145 * Also, transition to TRACED and updates to ->jobctl should be
2146 * atomic with respect to siglock and should be done after the arch
2147 * hook as siglock is released and regrabbed across it.
2148 *
2149 * TRACER TRACEE
2150 *
2151 * ptrace_attach()
2152 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2153 * do_wait()
2154 * set_current_state() smp_wmb();
2155 * ptrace_do_wait()
2156 * wait_task_stopped()
2157 * task_stopped_code()
2158 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2159 */
2160 smp_wmb();
2161
2162 current->last_siginfo = info;
2163 current->exit_code = exit_code;
2164
2165 /*
2166 * If @why is CLD_STOPPED, we're trapping to participate in a group
2167 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2168 * across siglock relocks since INTERRUPT was scheduled, PENDING
2169 * could be clear now. We act as if SIGCONT is received after
2170 * TASK_TRACED is entered - ignore it.
2171 */
2172 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2173 gstop_done = task_participate_group_stop(current);
2174
2175 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2176 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2177 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2178 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2179
2180 /* entering a trap, clear TRAPPING */
2181 task_clear_jobctl_trapping(current);
2182
2183 spin_unlock_irq(¤t->sighand->siglock);
2184 read_lock(&tasklist_lock);
2185 if (may_ptrace_stop()) {
2186 /*
2187 * Notify parents of the stop.
2188 *
2189 * While ptraced, there are two parents - the ptracer and
2190 * the real_parent of the group_leader. The ptracer should
2191 * know about every stop while the real parent is only
2192 * interested in the completion of group stop. The states
2193 * for the two don't interact with each other. Notify
2194 * separately unless they're gonna be duplicates.
2195 */
2196 do_notify_parent_cldstop(current, true, why);
2197 if (gstop_done && ptrace_reparented(current))
2198 do_notify_parent_cldstop(current, false, why);
2199
2200 /*
2201 * Don't want to allow preemption here, because
2202 * sys_ptrace() needs this task to be inactive.
2203 *
2204 * XXX: implement read_unlock_no_resched().
2205 */
2206 preempt_disable();
2207 read_unlock(&tasklist_lock);
2208 cgroup_enter_frozen();
2209 preempt_enable_no_resched();
2210 freezable_schedule();
2211 cgroup_leave_frozen(true);
2212 } else {
2213 /*
2214 * By the time we got the lock, our tracer went away.
2215 * Don't drop the lock yet, another tracer may come.
2216 *
2217 * If @gstop_done, the ptracer went away between group stop
2218 * completion and here. During detach, it would have set
2219 * JOBCTL_STOP_PENDING on us and we'll re-enter
2220 * TASK_STOPPED in do_signal_stop() on return, so notifying
2221 * the real parent of the group stop completion is enough.
2222 */
2223 if (gstop_done)
2224 do_notify_parent_cldstop(current, false, why);
2225
2226 /* tasklist protects us from ptrace_freeze_traced() */
2227 __set_current_state(TASK_RUNNING);
2228 if (clear_code)
2229 current->exit_code = 0;
2230 read_unlock(&tasklist_lock);
2231 }
2232
2233 /*
2234 * We are back. Now reacquire the siglock before touching
2235 * last_siginfo, so that we are sure to have synchronized with
2236 * any signal-sending on another CPU that wants to examine it.
2237 */
2238 spin_lock_irq(¤t->sighand->siglock);
2239 current->last_siginfo = NULL;
2240
2241 /* LISTENING can be set only during STOP traps, clear it */
2242 current->jobctl &= ~JOBCTL_LISTENING;
2243
2244 /*
2245 * Queued signals ignored us while we were stopped for tracing.
2246 * So check for any that we should take before resuming user mode.
2247 * This sets TIF_SIGPENDING, but never clears it.
2248 */
2249 recalc_sigpending_tsk(current);
2250}
2251
2252static void ptrace_do_notify(int signr, int exit_code, int why)
2253{
2254 kernel_siginfo_t info;
2255
2256 clear_siginfo(&info);
2257 info.si_signo = signr;
2258 info.si_code = exit_code;
2259 info.si_pid = task_pid_vnr(current);
2260 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2261
2262 /* Let the debugger run. */
2263 ptrace_stop(exit_code, why, 1, &info);
2264}
2265
2266void ptrace_notify(int exit_code)
2267{
2268 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2269 if (unlikely(current->task_works))
2270 task_work_run();
2271
2272 spin_lock_irq(¤t->sighand->siglock);
2273 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2274 spin_unlock_irq(¤t->sighand->siglock);
2275}
2276
2277/**
2278 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2279 * @signr: signr causing group stop if initiating
2280 *
2281 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2282 * and participate in it. If already set, participate in the existing
2283 * group stop. If participated in a group stop (and thus slept), %true is
2284 * returned with siglock released.
2285 *
2286 * If ptraced, this function doesn't handle stop itself. Instead,
2287 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2288 * untouched. The caller must ensure that INTERRUPT trap handling takes
2289 * places afterwards.
2290 *
2291 * CONTEXT:
2292 * Must be called with @current->sighand->siglock held, which is released
2293 * on %true return.
2294 *
2295 * RETURNS:
2296 * %false if group stop is already cancelled or ptrace trap is scheduled.
2297 * %true if participated in group stop.
2298 */
2299static bool do_signal_stop(int signr)
2300 __releases(¤t->sighand->siglock)
2301{
2302 struct signal_struct *sig = current->signal;
2303
2304 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2305 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2306 struct task_struct *t;
2307
2308 /* signr will be recorded in task->jobctl for retries */
2309 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2310
2311 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2312 unlikely(signal_group_exit(sig)))
2313 return false;
2314 /*
2315 * There is no group stop already in progress. We must
2316 * initiate one now.
2317 *
2318 * While ptraced, a task may be resumed while group stop is
2319 * still in effect and then receive a stop signal and
2320 * initiate another group stop. This deviates from the
2321 * usual behavior as two consecutive stop signals can't
2322 * cause two group stops when !ptraced. That is why we
2323 * also check !task_is_stopped(t) below.
2324 *
2325 * The condition can be distinguished by testing whether
2326 * SIGNAL_STOP_STOPPED is already set. Don't generate
2327 * group_exit_code in such case.
2328 *
2329 * This is not necessary for SIGNAL_STOP_CONTINUED because
2330 * an intervening stop signal is required to cause two
2331 * continued events regardless of ptrace.
2332 */
2333 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2334 sig->group_exit_code = signr;
2335
2336 sig->group_stop_count = 0;
2337
2338 if (task_set_jobctl_pending(current, signr | gstop))
2339 sig->group_stop_count++;
2340
2341 t = current;
2342 while_each_thread(current, t) {
2343 /*
2344 * Setting state to TASK_STOPPED for a group
2345 * stop is always done with the siglock held,
2346 * so this check has no races.
2347 */
2348 if (!task_is_stopped(t) &&
2349 task_set_jobctl_pending(t, signr | gstop)) {
2350 sig->group_stop_count++;
2351 if (likely(!(t->ptrace & PT_SEIZED)))
2352 signal_wake_up(t, 0);
2353 else
2354 ptrace_trap_notify(t);
2355 }
2356 }
2357 }
2358
2359 if (likely(!current->ptrace)) {
2360 int notify = 0;
2361
2362 /*
2363 * If there are no other threads in the group, or if there
2364 * is a group stop in progress and we are the last to stop,
2365 * report to the parent.
2366 */
2367 if (task_participate_group_stop(current))
2368 notify = CLD_STOPPED;
2369
2370 set_special_state(TASK_STOPPED);
2371 spin_unlock_irq(¤t->sighand->siglock);
2372
2373 /*
2374 * Notify the parent of the group stop completion. Because
2375 * we're not holding either the siglock or tasklist_lock
2376 * here, ptracer may attach inbetween; however, this is for
2377 * group stop and should always be delivered to the real
2378 * parent of the group leader. The new ptracer will get
2379 * its notification when this task transitions into
2380 * TASK_TRACED.
2381 */
2382 if (notify) {
2383 read_lock(&tasklist_lock);
2384 do_notify_parent_cldstop(current, false, notify);
2385 read_unlock(&tasklist_lock);
2386 }
2387
2388 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2389 cgroup_enter_frozen();
2390 freezable_schedule();
2391 return true;
2392 } else {
2393 /*
2394 * While ptraced, group stop is handled by STOP trap.
2395 * Schedule it and let the caller deal with it.
2396 */
2397 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2398 return false;
2399 }
2400}
2401
2402/**
2403 * do_jobctl_trap - take care of ptrace jobctl traps
2404 *
2405 * When PT_SEIZED, it's used for both group stop and explicit
2406 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2407 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2408 * the stop signal; otherwise, %SIGTRAP.
2409 *
2410 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2411 * number as exit_code and no siginfo.
2412 *
2413 * CONTEXT:
2414 * Must be called with @current->sighand->siglock held, which may be
2415 * released and re-acquired before returning with intervening sleep.
2416 */
2417static void do_jobctl_trap(void)
2418{
2419 struct signal_struct *signal = current->signal;
2420 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2421
2422 if (current->ptrace & PT_SEIZED) {
2423 if (!signal->group_stop_count &&
2424 !(signal->flags & SIGNAL_STOP_STOPPED))
2425 signr = SIGTRAP;
2426 WARN_ON_ONCE(!signr);
2427 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2428 CLD_STOPPED);
2429 } else {
2430 WARN_ON_ONCE(!signr);
2431 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2432 current->exit_code = 0;
2433 }
2434}
2435
2436/**
2437 * do_freezer_trap - handle the freezer jobctl trap
2438 *
2439 * Puts the task into frozen state, if only the task is not about to quit.
2440 * In this case it drops JOBCTL_TRAP_FREEZE.
2441 *
2442 * CONTEXT:
2443 * Must be called with @current->sighand->siglock held,
2444 * which is always released before returning.
2445 */
2446static void do_freezer_trap(void)
2447 __releases(¤t->sighand->siglock)
2448{
2449 /*
2450 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2451 * let's make another loop to give it a chance to be handled.
2452 * In any case, we'll return back.
2453 */
2454 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2455 JOBCTL_TRAP_FREEZE) {
2456 spin_unlock_irq(¤t->sighand->siglock);
2457 return;
2458 }
2459
2460 /*
2461 * Now we're sure that there is no pending fatal signal and no
2462 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2463 * immediately (if there is a non-fatal signal pending), and
2464 * put the task into sleep.
2465 */
2466 __set_current_state(TASK_INTERRUPTIBLE);
2467 clear_thread_flag(TIF_SIGPENDING);
2468 spin_unlock_irq(¤t->sighand->siglock);
2469 cgroup_enter_frozen();
2470 freezable_schedule();
2471}
2472
2473static int ptrace_signal(int signr, kernel_siginfo_t *info)
2474{
2475 /*
2476 * We do not check sig_kernel_stop(signr) but set this marker
2477 * unconditionally because we do not know whether debugger will
2478 * change signr. This flag has no meaning unless we are going
2479 * to stop after return from ptrace_stop(). In this case it will
2480 * be checked in do_signal_stop(), we should only stop if it was
2481 * not cleared by SIGCONT while we were sleeping. See also the
2482 * comment in dequeue_signal().
2483 */
2484 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2485 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2486
2487 /* We're back. Did the debugger cancel the sig? */
2488 signr = current->exit_code;
2489 if (signr == 0)
2490 return signr;
2491
2492 current->exit_code = 0;
2493
2494 /*
2495 * Update the siginfo structure if the signal has
2496 * changed. If the debugger wanted something
2497 * specific in the siginfo structure then it should
2498 * have updated *info via PTRACE_SETSIGINFO.
2499 */
2500 if (signr != info->si_signo) {
2501 clear_siginfo(info);
2502 info->si_signo = signr;
2503 info->si_errno = 0;
2504 info->si_code = SI_USER;
2505 rcu_read_lock();
2506 info->si_pid = task_pid_vnr(current->parent);
2507 info->si_uid = from_kuid_munged(current_user_ns(),
2508 task_uid(current->parent));
2509 rcu_read_unlock();
2510 }
2511
2512 /* If the (new) signal is now blocked, requeue it. */
2513 if (sigismember(¤t->blocked, signr)) {
2514 send_signal(signr, info, current, PIDTYPE_PID);
2515 signr = 0;
2516 }
2517
2518 return signr;
2519}
2520
2521bool get_signal(struct ksignal *ksig)
2522{
2523 struct sighand_struct *sighand = current->sighand;
2524 struct signal_struct *signal = current->signal;
2525 int signr;
2526
2527 if (unlikely(current->task_works))
2528 task_work_run();
2529
2530 if (unlikely(uprobe_deny_signal()))
2531 return false;
2532
2533 /*
2534 * Do this once, we can't return to user-mode if freezing() == T.
2535 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2536 * thus do not need another check after return.
2537 */
2538 try_to_freeze();
2539
2540relock:
2541 spin_lock_irq(&sighand->siglock);
2542 /*
2543 * Every stopped thread goes here after wakeup. Check to see if
2544 * we should notify the parent, prepare_signal(SIGCONT) encodes
2545 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2546 */
2547 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2548 int why;
2549
2550 if (signal->flags & SIGNAL_CLD_CONTINUED)
2551 why = CLD_CONTINUED;
2552 else
2553 why = CLD_STOPPED;
2554
2555 signal->flags &= ~SIGNAL_CLD_MASK;
2556
2557 spin_unlock_irq(&sighand->siglock);
2558
2559 /*
2560 * Notify the parent that we're continuing. This event is
2561 * always per-process and doesn't make whole lot of sense
2562 * for ptracers, who shouldn't consume the state via
2563 * wait(2) either, but, for backward compatibility, notify
2564 * the ptracer of the group leader too unless it's gonna be
2565 * a duplicate.
2566 */
2567 read_lock(&tasklist_lock);
2568 do_notify_parent_cldstop(current, false, why);
2569
2570 if (ptrace_reparented(current->group_leader))
2571 do_notify_parent_cldstop(current->group_leader,
2572 true, why);
2573 read_unlock(&tasklist_lock);
2574
2575 goto relock;
2576 }
2577
2578 /* Has this task already been marked for death? */
2579 if (signal_group_exit(signal)) {
2580 ksig->info.si_signo = signr = SIGKILL;
2581 sigdelset(¤t->pending.signal, SIGKILL);
2582 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2583 &sighand->action[SIGKILL - 1]);
2584 recalc_sigpending();
2585 goto fatal;
2586 }
2587
2588 for (;;) {
2589 struct k_sigaction *ka;
2590
2591 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2592 do_signal_stop(0))
2593 goto relock;
2594
2595 if (unlikely(current->jobctl &
2596 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2597 if (current->jobctl & JOBCTL_TRAP_MASK) {
2598 do_jobctl_trap();
2599 spin_unlock_irq(&sighand->siglock);
2600 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2601 do_freezer_trap();
2602
2603 goto relock;
2604 }
2605
2606 /*
2607 * If the task is leaving the frozen state, let's update
2608 * cgroup counters and reset the frozen bit.
2609 */
2610 if (unlikely(cgroup_task_frozen(current))) {
2611 spin_unlock_irq(&sighand->siglock);
2612 cgroup_leave_frozen(false);
2613 goto relock;
2614 }
2615
2616 /*
2617 * Signals generated by the execution of an instruction
2618 * need to be delivered before any other pending signals
2619 * so that the instruction pointer in the signal stack
2620 * frame points to the faulting instruction.
2621 */
2622 signr = dequeue_synchronous_signal(&ksig->info);
2623 if (!signr)
2624 signr = dequeue_signal(current, ¤t->blocked, &ksig->info);
2625
2626 if (!signr)
2627 break; /* will return 0 */
2628
2629 if (unlikely(current->ptrace) && signr != SIGKILL) {
2630 signr = ptrace_signal(signr, &ksig->info);
2631 if (!signr)
2632 continue;
2633 }
2634
2635 ka = &sighand->action[signr-1];
2636
2637 /* Trace actually delivered signals. */
2638 trace_signal_deliver(signr, &ksig->info, ka);
2639
2640 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2641 continue;
2642 if (ka->sa.sa_handler != SIG_DFL) {
2643 /* Run the handler. */
2644 ksig->ka = *ka;
2645
2646 if (ka->sa.sa_flags & SA_ONESHOT)
2647 ka->sa.sa_handler = SIG_DFL;
2648
2649 break; /* will return non-zero "signr" value */
2650 }
2651
2652 /*
2653 * Now we are doing the default action for this signal.
2654 */
2655 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2656 continue;
2657
2658 /*
2659 * Global init gets no signals it doesn't want.
2660 * Container-init gets no signals it doesn't want from same
2661 * container.
2662 *
2663 * Note that if global/container-init sees a sig_kernel_only()
2664 * signal here, the signal must have been generated internally
2665 * or must have come from an ancestor namespace. In either
2666 * case, the signal cannot be dropped.
2667 */
2668 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2669 !sig_kernel_only(signr))
2670 continue;
2671
2672 if (sig_kernel_stop(signr)) {
2673 /*
2674 * The default action is to stop all threads in
2675 * the thread group. The job control signals
2676 * do nothing in an orphaned pgrp, but SIGSTOP
2677 * always works. Note that siglock needs to be
2678 * dropped during the call to is_orphaned_pgrp()
2679 * because of lock ordering with tasklist_lock.
2680 * This allows an intervening SIGCONT to be posted.
2681 * We need to check for that and bail out if necessary.
2682 */
2683 if (signr != SIGSTOP) {
2684 spin_unlock_irq(&sighand->siglock);
2685
2686 /* signals can be posted during this window */
2687
2688 if (is_current_pgrp_orphaned())
2689 goto relock;
2690
2691 spin_lock_irq(&sighand->siglock);
2692 }
2693
2694 if (likely(do_signal_stop(ksig->info.si_signo))) {
2695 /* It released the siglock. */
2696 goto relock;
2697 }
2698
2699 /*
2700 * We didn't actually stop, due to a race
2701 * with SIGCONT or something like that.
2702 */
2703 continue;
2704 }
2705
2706 fatal:
2707 spin_unlock_irq(&sighand->siglock);
2708 if (unlikely(cgroup_task_frozen(current)))
2709 cgroup_leave_frozen(true);
2710
2711 /*
2712 * Anything else is fatal, maybe with a core dump.
2713 */
2714 current->flags |= PF_SIGNALED;
2715
2716 if (sig_kernel_coredump(signr)) {
2717 if (print_fatal_signals)
2718 print_fatal_signal(ksig->info.si_signo);
2719 proc_coredump_connector(current);
2720 /*
2721 * If it was able to dump core, this kills all
2722 * other threads in the group and synchronizes with
2723 * their demise. If we lost the race with another
2724 * thread getting here, it set group_exit_code
2725 * first and our do_group_exit call below will use
2726 * that value and ignore the one we pass it.
2727 */
2728 do_coredump(&ksig->info);
2729 }
2730
2731 /*
2732 * Death signals, no core dump.
2733 */
2734 do_group_exit(ksig->info.si_signo);
2735 /* NOTREACHED */
2736 }
2737 spin_unlock_irq(&sighand->siglock);
2738
2739 ksig->sig = signr;
2740 return ksig->sig > 0;
2741}
2742
2743/**
2744 * signal_delivered -
2745 * @ksig: kernel signal struct
2746 * @stepping: nonzero if debugger single-step or block-step in use
2747 *
2748 * This function should be called when a signal has successfully been
2749 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2750 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2751 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2752 */
2753static void signal_delivered(struct ksignal *ksig, int stepping)
2754{
2755 sigset_t blocked;
2756
2757 /* A signal was successfully delivered, and the
2758 saved sigmask was stored on the signal frame,
2759 and will be restored by sigreturn. So we can
2760 simply clear the restore sigmask flag. */
2761 clear_restore_sigmask();
2762
2763 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2764 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2765 sigaddset(&blocked, ksig->sig);
2766 set_current_blocked(&blocked);
2767 tracehook_signal_handler(stepping);
2768}
2769
2770void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2771{
2772 if (failed)
2773 force_sigsegv(ksig->sig);
2774 else
2775 signal_delivered(ksig, stepping);
2776}
2777
2778/*
2779 * It could be that complete_signal() picked us to notify about the
2780 * group-wide signal. Other threads should be notified now to take
2781 * the shared signals in @which since we will not.
2782 */
2783static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2784{
2785 sigset_t retarget;
2786 struct task_struct *t;
2787
2788 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2789 if (sigisemptyset(&retarget))
2790 return;
2791
2792 t = tsk;
2793 while_each_thread(tsk, t) {
2794 if (t->flags & PF_EXITING)
2795 continue;
2796
2797 if (!has_pending_signals(&retarget, &t->blocked))
2798 continue;
2799 /* Remove the signals this thread can handle. */
2800 sigandsets(&retarget, &retarget, &t->blocked);
2801
2802 if (!signal_pending(t))
2803 signal_wake_up(t, 0);
2804
2805 if (sigisemptyset(&retarget))
2806 break;
2807 }
2808}
2809
2810void exit_signals(struct task_struct *tsk)
2811{
2812 int group_stop = 0;
2813 sigset_t unblocked;
2814
2815 /*
2816 * @tsk is about to have PF_EXITING set - lock out users which
2817 * expect stable threadgroup.
2818 */
2819 cgroup_threadgroup_change_begin(tsk);
2820
2821 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2822 tsk->flags |= PF_EXITING;
2823 cgroup_threadgroup_change_end(tsk);
2824 return;
2825 }
2826
2827 spin_lock_irq(&tsk->sighand->siglock);
2828 /*
2829 * From now this task is not visible for group-wide signals,
2830 * see wants_signal(), do_signal_stop().
2831 */
2832 tsk->flags |= PF_EXITING;
2833
2834 cgroup_threadgroup_change_end(tsk);
2835
2836 if (!signal_pending(tsk))
2837 goto out;
2838
2839 unblocked = tsk->blocked;
2840 signotset(&unblocked);
2841 retarget_shared_pending(tsk, &unblocked);
2842
2843 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2844 task_participate_group_stop(tsk))
2845 group_stop = CLD_STOPPED;
2846out:
2847 spin_unlock_irq(&tsk->sighand->siglock);
2848
2849 /*
2850 * If group stop has completed, deliver the notification. This
2851 * should always go to the real parent of the group leader.
2852 */
2853 if (unlikely(group_stop)) {
2854 read_lock(&tasklist_lock);
2855 do_notify_parent_cldstop(tsk, false, group_stop);
2856 read_unlock(&tasklist_lock);
2857 }
2858}
2859
2860/*
2861 * System call entry points.
2862 */
2863
2864/**
2865 * sys_restart_syscall - restart a system call
2866 */
2867SYSCALL_DEFINE0(restart_syscall)
2868{
2869 struct restart_block *restart = ¤t->restart_block;
2870 return restart->fn(restart);
2871}
2872
2873long do_no_restart_syscall(struct restart_block *param)
2874{
2875 return -EINTR;
2876}
2877
2878static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2879{
2880 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2881 sigset_t newblocked;
2882 /* A set of now blocked but previously unblocked signals. */
2883 sigandnsets(&newblocked, newset, ¤t->blocked);
2884 retarget_shared_pending(tsk, &newblocked);
2885 }
2886 tsk->blocked = *newset;
2887 recalc_sigpending();
2888}
2889
2890/**
2891 * set_current_blocked - change current->blocked mask
2892 * @newset: new mask
2893 *
2894 * It is wrong to change ->blocked directly, this helper should be used
2895 * to ensure the process can't miss a shared signal we are going to block.
2896 */
2897void set_current_blocked(sigset_t *newset)
2898{
2899 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2900 __set_current_blocked(newset);
2901}
2902
2903void __set_current_blocked(const sigset_t *newset)
2904{
2905 struct task_struct *tsk = current;
2906
2907 /*
2908 * In case the signal mask hasn't changed, there is nothing we need
2909 * to do. The current->blocked shouldn't be modified by other task.
2910 */
2911 if (sigequalsets(&tsk->blocked, newset))
2912 return;
2913
2914 spin_lock_irq(&tsk->sighand->siglock);
2915 __set_task_blocked(tsk, newset);
2916 spin_unlock_irq(&tsk->sighand->siglock);
2917}
2918
2919/*
2920 * This is also useful for kernel threads that want to temporarily
2921 * (or permanently) block certain signals.
2922 *
2923 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2924 * interface happily blocks "unblockable" signals like SIGKILL
2925 * and friends.
2926 */
2927int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2928{
2929 struct task_struct *tsk = current;
2930 sigset_t newset;
2931
2932 /* Lockless, only current can change ->blocked, never from irq */
2933 if (oldset)
2934 *oldset = tsk->blocked;
2935
2936 switch (how) {
2937 case SIG_BLOCK:
2938 sigorsets(&newset, &tsk->blocked, set);
2939 break;
2940 case SIG_UNBLOCK:
2941 sigandnsets(&newset, &tsk->blocked, set);
2942 break;
2943 case SIG_SETMASK:
2944 newset = *set;
2945 break;
2946 default:
2947 return -EINVAL;
2948 }
2949
2950 __set_current_blocked(&newset);
2951 return 0;
2952}
2953EXPORT_SYMBOL(sigprocmask);
2954
2955/*
2956 * The api helps set app-provided sigmasks.
2957 *
2958 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
2959 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
2960 *
2961 * Note that it does set_restore_sigmask() in advance, so it must be always
2962 * paired with restore_saved_sigmask_unless() before return from syscall.
2963 */
2964int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
2965{
2966 sigset_t kmask;
2967
2968 if (!umask)
2969 return 0;
2970 if (sigsetsize != sizeof(sigset_t))
2971 return -EINVAL;
2972 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
2973 return -EFAULT;
2974
2975 set_restore_sigmask();
2976 current->saved_sigmask = current->blocked;
2977 set_current_blocked(&kmask);
2978
2979 return 0;
2980}
2981
2982#ifdef CONFIG_COMPAT
2983int set_compat_user_sigmask(const compat_sigset_t __user *umask,
2984 size_t sigsetsize)
2985{
2986 sigset_t kmask;
2987
2988 if (!umask)
2989 return 0;
2990 if (sigsetsize != sizeof(compat_sigset_t))
2991 return -EINVAL;
2992 if (get_compat_sigset(&kmask, umask))
2993 return -EFAULT;
2994
2995 set_restore_sigmask();
2996 current->saved_sigmask = current->blocked;
2997 set_current_blocked(&kmask);
2998
2999 return 0;
3000}
3001#endif
3002
3003/**
3004 * sys_rt_sigprocmask - change the list of currently blocked signals
3005 * @how: whether to add, remove, or set signals
3006 * @nset: stores pending signals
3007 * @oset: previous value of signal mask if non-null
3008 * @sigsetsize: size of sigset_t type
3009 */
3010SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3011 sigset_t __user *, oset, size_t, sigsetsize)
3012{
3013 sigset_t old_set, new_set;
3014 int error;
3015
3016 /* XXX: Don't preclude handling different sized sigset_t's. */
3017 if (sigsetsize != sizeof(sigset_t))
3018 return -EINVAL;
3019
3020 old_set = current->blocked;
3021
3022 if (nset) {
3023 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3024 return -EFAULT;
3025 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3026
3027 error = sigprocmask(how, &new_set, NULL);
3028 if (error)
3029 return error;
3030 }
3031
3032 if (oset) {
3033 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3034 return -EFAULT;
3035 }
3036
3037 return 0;
3038}
3039
3040#ifdef CONFIG_COMPAT
3041COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3042 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3043{
3044 sigset_t old_set = current->blocked;
3045
3046 /* XXX: Don't preclude handling different sized sigset_t's. */
3047 if (sigsetsize != sizeof(sigset_t))
3048 return -EINVAL;
3049
3050 if (nset) {
3051 sigset_t new_set;
3052 int error;
3053 if (get_compat_sigset(&new_set, nset))
3054 return -EFAULT;
3055 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3056
3057 error = sigprocmask(how, &new_set, NULL);
3058 if (error)
3059 return error;
3060 }
3061 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3062}
3063#endif
3064
3065static void do_sigpending(sigset_t *set)
3066{
3067 spin_lock_irq(¤t->sighand->siglock);
3068 sigorsets(set, ¤t->pending.signal,
3069 ¤t->signal->shared_pending.signal);
3070 spin_unlock_irq(¤t->sighand->siglock);
3071
3072 /* Outside the lock because only this thread touches it. */
3073 sigandsets(set, ¤t->blocked, set);
3074}
3075
3076/**
3077 * sys_rt_sigpending - examine a pending signal that has been raised
3078 * while blocked
3079 * @uset: stores pending signals
3080 * @sigsetsize: size of sigset_t type or larger
3081 */
3082SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3083{
3084 sigset_t set;
3085
3086 if (sigsetsize > sizeof(*uset))
3087 return -EINVAL;
3088
3089 do_sigpending(&set);
3090
3091 if (copy_to_user(uset, &set, sigsetsize))
3092 return -EFAULT;
3093
3094 return 0;
3095}
3096
3097#ifdef CONFIG_COMPAT
3098COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3099 compat_size_t, sigsetsize)
3100{
3101 sigset_t set;
3102
3103 if (sigsetsize > sizeof(*uset))
3104 return -EINVAL;
3105
3106 do_sigpending(&set);
3107
3108 return put_compat_sigset(uset, &set, sigsetsize);
3109}
3110#endif
3111
3112static const struct {
3113 unsigned char limit, layout;
3114} sig_sicodes[] = {
3115 [SIGILL] = { NSIGILL, SIL_FAULT },
3116 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3117 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3118 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3119 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3120#if defined(SIGEMT)
3121 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3122#endif
3123 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3124 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3125 [SIGSYS] = { NSIGSYS, SIL_SYS },
3126};
3127
3128static bool known_siginfo_layout(unsigned sig, int si_code)
3129{
3130 if (si_code == SI_KERNEL)
3131 return true;
3132 else if ((si_code > SI_USER)) {
3133 if (sig_specific_sicodes(sig)) {
3134 if (si_code <= sig_sicodes[sig].limit)
3135 return true;
3136 }
3137 else if (si_code <= NSIGPOLL)
3138 return true;
3139 }
3140 else if (si_code >= SI_DETHREAD)
3141 return true;
3142 else if (si_code == SI_ASYNCNL)
3143 return true;
3144 return false;
3145}
3146
3147enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3148{
3149 enum siginfo_layout layout = SIL_KILL;
3150 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3151 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3152 (si_code <= sig_sicodes[sig].limit)) {
3153 layout = sig_sicodes[sig].layout;
3154 /* Handle the exceptions */
3155 if ((sig == SIGBUS) &&
3156 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3157 layout = SIL_FAULT_MCEERR;
3158 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3159 layout = SIL_FAULT_BNDERR;
3160#ifdef SEGV_PKUERR
3161 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3162 layout = SIL_FAULT_PKUERR;
3163#endif
3164 }
3165 else if (si_code <= NSIGPOLL)
3166 layout = SIL_POLL;
3167 } else {
3168 if (si_code == SI_TIMER)
3169 layout = SIL_TIMER;
3170 else if (si_code == SI_SIGIO)
3171 layout = SIL_POLL;
3172 else if (si_code < 0)
3173 layout = SIL_RT;
3174 }
3175 return layout;
3176}
3177
3178static inline char __user *si_expansion(const siginfo_t __user *info)
3179{
3180 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3181}
3182
3183int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3184{
3185 char __user *expansion = si_expansion(to);
3186 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3187 return -EFAULT;
3188 if (clear_user(expansion, SI_EXPANSION_SIZE))
3189 return -EFAULT;
3190 return 0;
3191}
3192
3193static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3194 const siginfo_t __user *from)
3195{
3196 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3197 char __user *expansion = si_expansion(from);
3198 char buf[SI_EXPANSION_SIZE];
3199 int i;
3200 /*
3201 * An unknown si_code might need more than
3202 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3203 * extra bytes are 0. This guarantees copy_siginfo_to_user
3204 * will return this data to userspace exactly.
3205 */
3206 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3207 return -EFAULT;
3208 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3209 if (buf[i] != 0)
3210 return -E2BIG;
3211 }
3212 }
3213 return 0;
3214}
3215
3216static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3217 const siginfo_t __user *from)
3218{
3219 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3220 return -EFAULT;
3221 to->si_signo = signo;
3222 return post_copy_siginfo_from_user(to, from);
3223}
3224
3225int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3226{
3227 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3228 return -EFAULT;
3229 return post_copy_siginfo_from_user(to, from);
3230}
3231
3232#ifdef CONFIG_COMPAT
3233int copy_siginfo_to_user32(struct compat_siginfo __user *to,
3234 const struct kernel_siginfo *from)
3235#if defined(CONFIG_X86_X32_ABI) || defined(CONFIG_IA32_EMULATION)
3236{
3237 return __copy_siginfo_to_user32(to, from, in_x32_syscall());
3238}
3239int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3240 const struct kernel_siginfo *from, bool x32_ABI)
3241#endif
3242{
3243 struct compat_siginfo new;
3244 memset(&new, 0, sizeof(new));
3245
3246 new.si_signo = from->si_signo;
3247 new.si_errno = from->si_errno;
3248 new.si_code = from->si_code;
3249 switch(siginfo_layout(from->si_signo, from->si_code)) {
3250 case SIL_KILL:
3251 new.si_pid = from->si_pid;
3252 new.si_uid = from->si_uid;
3253 break;
3254 case SIL_TIMER:
3255 new.si_tid = from->si_tid;
3256 new.si_overrun = from->si_overrun;
3257 new.si_int = from->si_int;
3258 break;
3259 case SIL_POLL:
3260 new.si_band = from->si_band;
3261 new.si_fd = from->si_fd;
3262 break;
3263 case SIL_FAULT:
3264 new.si_addr = ptr_to_compat(from->si_addr);
3265#ifdef __ARCH_SI_TRAPNO
3266 new.si_trapno = from->si_trapno;
3267#endif
3268 break;
3269 case SIL_FAULT_MCEERR:
3270 new.si_addr = ptr_to_compat(from->si_addr);
3271#ifdef __ARCH_SI_TRAPNO
3272 new.si_trapno = from->si_trapno;
3273#endif
3274 new.si_addr_lsb = from->si_addr_lsb;
3275 break;
3276 case SIL_FAULT_BNDERR:
3277 new.si_addr = ptr_to_compat(from->si_addr);
3278#ifdef __ARCH_SI_TRAPNO
3279 new.si_trapno = from->si_trapno;
3280#endif
3281 new.si_lower = ptr_to_compat(from->si_lower);
3282 new.si_upper = ptr_to_compat(from->si_upper);
3283 break;
3284 case SIL_FAULT_PKUERR:
3285 new.si_addr = ptr_to_compat(from->si_addr);
3286#ifdef __ARCH_SI_TRAPNO
3287 new.si_trapno = from->si_trapno;
3288#endif
3289 new.si_pkey = from->si_pkey;
3290 break;
3291 case SIL_CHLD:
3292 new.si_pid = from->si_pid;
3293 new.si_uid = from->si_uid;
3294 new.si_status = from->si_status;
3295#ifdef CONFIG_X86_X32_ABI
3296 if (x32_ABI) {
3297 new._sifields._sigchld_x32._utime = from->si_utime;
3298 new._sifields._sigchld_x32._stime = from->si_stime;
3299 } else
3300#endif
3301 {
3302 new.si_utime = from->si_utime;
3303 new.si_stime = from->si_stime;
3304 }
3305 break;
3306 case SIL_RT:
3307 new.si_pid = from->si_pid;
3308 new.si_uid = from->si_uid;
3309 new.si_int = from->si_int;
3310 break;
3311 case SIL_SYS:
3312 new.si_call_addr = ptr_to_compat(from->si_call_addr);
3313 new.si_syscall = from->si_syscall;
3314 new.si_arch = from->si_arch;
3315 break;
3316 }
3317
3318 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3319 return -EFAULT;
3320
3321 return 0;
3322}
3323
3324static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3325 const struct compat_siginfo *from)
3326{
3327 clear_siginfo(to);
3328 to->si_signo = from->si_signo;
3329 to->si_errno = from->si_errno;
3330 to->si_code = from->si_code;
3331 switch(siginfo_layout(from->si_signo, from->si_code)) {
3332 case SIL_KILL:
3333 to->si_pid = from->si_pid;
3334 to->si_uid = from->si_uid;
3335 break;
3336 case SIL_TIMER:
3337 to->si_tid = from->si_tid;
3338 to->si_overrun = from->si_overrun;
3339 to->si_int = from->si_int;
3340 break;
3341 case SIL_POLL:
3342 to->si_band = from->si_band;
3343 to->si_fd = from->si_fd;
3344 break;
3345 case SIL_FAULT:
3346 to->si_addr = compat_ptr(from->si_addr);
3347#ifdef __ARCH_SI_TRAPNO
3348 to->si_trapno = from->si_trapno;
3349#endif
3350 break;
3351 case SIL_FAULT_MCEERR:
3352 to->si_addr = compat_ptr(from->si_addr);
3353#ifdef __ARCH_SI_TRAPNO
3354 to->si_trapno = from->si_trapno;
3355#endif
3356 to->si_addr_lsb = from->si_addr_lsb;
3357 break;
3358 case SIL_FAULT_BNDERR:
3359 to->si_addr = compat_ptr(from->si_addr);
3360#ifdef __ARCH_SI_TRAPNO
3361 to->si_trapno = from->si_trapno;
3362#endif
3363 to->si_lower = compat_ptr(from->si_lower);
3364 to->si_upper = compat_ptr(from->si_upper);
3365 break;
3366 case SIL_FAULT_PKUERR:
3367 to->si_addr = compat_ptr(from->si_addr);
3368#ifdef __ARCH_SI_TRAPNO
3369 to->si_trapno = from->si_trapno;
3370#endif
3371 to->si_pkey = from->si_pkey;
3372 break;
3373 case SIL_CHLD:
3374 to->si_pid = from->si_pid;
3375 to->si_uid = from->si_uid;
3376 to->si_status = from->si_status;
3377#ifdef CONFIG_X86_X32_ABI
3378 if (in_x32_syscall()) {
3379 to->si_utime = from->_sifields._sigchld_x32._utime;
3380 to->si_stime = from->_sifields._sigchld_x32._stime;
3381 } else
3382#endif
3383 {
3384 to->si_utime = from->si_utime;
3385 to->si_stime = from->si_stime;
3386 }
3387 break;
3388 case SIL_RT:
3389 to->si_pid = from->si_pid;
3390 to->si_uid = from->si_uid;
3391 to->si_int = from->si_int;
3392 break;
3393 case SIL_SYS:
3394 to->si_call_addr = compat_ptr(from->si_call_addr);
3395 to->si_syscall = from->si_syscall;
3396 to->si_arch = from->si_arch;
3397 break;
3398 }
3399 return 0;
3400}
3401
3402static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3403 const struct compat_siginfo __user *ufrom)
3404{
3405 struct compat_siginfo from;
3406
3407 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3408 return -EFAULT;
3409
3410 from.si_signo = signo;
3411 return post_copy_siginfo_from_user32(to, &from);
3412}
3413
3414int copy_siginfo_from_user32(struct kernel_siginfo *to,
3415 const struct compat_siginfo __user *ufrom)
3416{
3417 struct compat_siginfo from;
3418
3419 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3420 return -EFAULT;
3421
3422 return post_copy_siginfo_from_user32(to, &from);
3423}
3424#endif /* CONFIG_COMPAT */
3425
3426/**
3427 * do_sigtimedwait - wait for queued signals specified in @which
3428 * @which: queued signals to wait for
3429 * @info: if non-null, the signal's siginfo is returned here
3430 * @ts: upper bound on process time suspension
3431 */
3432static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3433 const struct timespec64 *ts)
3434{
3435 ktime_t *to = NULL, timeout = KTIME_MAX;
3436 struct task_struct *tsk = current;
3437 sigset_t mask = *which;
3438 int sig, ret = 0;
3439
3440 if (ts) {
3441 if (!timespec64_valid(ts))
3442 return -EINVAL;
3443 timeout = timespec64_to_ktime(*ts);
3444 to = &timeout;
3445 }
3446
3447 /*
3448 * Invert the set of allowed signals to get those we want to block.
3449 */
3450 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3451 signotset(&mask);
3452
3453 spin_lock_irq(&tsk->sighand->siglock);
3454 sig = dequeue_signal(tsk, &mask, info);
3455 if (!sig && timeout) {
3456 /*
3457 * None ready, temporarily unblock those we're interested
3458 * while we are sleeping in so that we'll be awakened when
3459 * they arrive. Unblocking is always fine, we can avoid
3460 * set_current_blocked().
3461 */
3462 tsk->real_blocked = tsk->blocked;
3463 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3464 recalc_sigpending();
3465 spin_unlock_irq(&tsk->sighand->siglock);
3466
3467 __set_current_state(TASK_INTERRUPTIBLE);
3468 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3469 HRTIMER_MODE_REL);
3470 spin_lock_irq(&tsk->sighand->siglock);
3471 __set_task_blocked(tsk, &tsk->real_blocked);
3472 sigemptyset(&tsk->real_blocked);
3473 sig = dequeue_signal(tsk, &mask, info);
3474 }
3475 spin_unlock_irq(&tsk->sighand->siglock);
3476
3477 if (sig)
3478 return sig;
3479 return ret ? -EINTR : -EAGAIN;
3480}
3481
3482/**
3483 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3484 * in @uthese
3485 * @uthese: queued signals to wait for
3486 * @uinfo: if non-null, the signal's siginfo is returned here
3487 * @uts: upper bound on process time suspension
3488 * @sigsetsize: size of sigset_t type
3489 */
3490SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3491 siginfo_t __user *, uinfo,
3492 const struct __kernel_timespec __user *, uts,
3493 size_t, sigsetsize)
3494{
3495 sigset_t these;
3496 struct timespec64 ts;
3497 kernel_siginfo_t info;
3498 int ret;
3499
3500 /* XXX: Don't preclude handling different sized sigset_t's. */
3501 if (sigsetsize != sizeof(sigset_t))
3502 return -EINVAL;
3503
3504 if (copy_from_user(&these, uthese, sizeof(these)))
3505 return -EFAULT;
3506
3507 if (uts) {
3508 if (get_timespec64(&ts, uts))
3509 return -EFAULT;
3510 }
3511
3512 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3513
3514 if (ret > 0 && uinfo) {
3515 if (copy_siginfo_to_user(uinfo, &info))
3516 ret = -EFAULT;
3517 }
3518
3519 return ret;
3520}
3521
3522#ifdef CONFIG_COMPAT_32BIT_TIME
3523SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3524 siginfo_t __user *, uinfo,
3525 const struct old_timespec32 __user *, uts,
3526 size_t, sigsetsize)
3527{
3528 sigset_t these;
3529 struct timespec64 ts;
3530 kernel_siginfo_t info;
3531 int ret;
3532
3533 if (sigsetsize != sizeof(sigset_t))
3534 return -EINVAL;
3535
3536 if (copy_from_user(&these, uthese, sizeof(these)))
3537 return -EFAULT;
3538
3539 if (uts) {
3540 if (get_old_timespec32(&ts, uts))
3541 return -EFAULT;
3542 }
3543
3544 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3545
3546 if (ret > 0 && uinfo) {
3547 if (copy_siginfo_to_user(uinfo, &info))
3548 ret = -EFAULT;
3549 }
3550
3551 return ret;
3552}
3553#endif
3554
3555#ifdef CONFIG_COMPAT
3556COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3557 struct compat_siginfo __user *, uinfo,
3558 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3559{
3560 sigset_t s;
3561 struct timespec64 t;
3562 kernel_siginfo_t info;
3563 long ret;
3564
3565 if (sigsetsize != sizeof(sigset_t))
3566 return -EINVAL;
3567
3568 if (get_compat_sigset(&s, uthese))
3569 return -EFAULT;
3570
3571 if (uts) {
3572 if (get_timespec64(&t, uts))
3573 return -EFAULT;
3574 }
3575
3576 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3577
3578 if (ret > 0 && uinfo) {
3579 if (copy_siginfo_to_user32(uinfo, &info))
3580 ret = -EFAULT;
3581 }
3582
3583 return ret;
3584}
3585
3586#ifdef CONFIG_COMPAT_32BIT_TIME
3587COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3588 struct compat_siginfo __user *, uinfo,
3589 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3590{
3591 sigset_t s;
3592 struct timespec64 t;
3593 kernel_siginfo_t info;
3594 long ret;
3595
3596 if (sigsetsize != sizeof(sigset_t))
3597 return -EINVAL;
3598
3599 if (get_compat_sigset(&s, uthese))
3600 return -EFAULT;
3601
3602 if (uts) {
3603 if (get_old_timespec32(&t, uts))
3604 return -EFAULT;
3605 }
3606
3607 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3608
3609 if (ret > 0 && uinfo) {
3610 if (copy_siginfo_to_user32(uinfo, &info))
3611 ret = -EFAULT;
3612 }
3613
3614 return ret;
3615}
3616#endif
3617#endif
3618
3619static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3620{
3621 clear_siginfo(info);
3622 info->si_signo = sig;
3623 info->si_errno = 0;
3624 info->si_code = SI_USER;
3625 info->si_pid = task_tgid_vnr(current);
3626 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3627}
3628
3629/**
3630 * sys_kill - send a signal to a process
3631 * @pid: the PID of the process
3632 * @sig: signal to be sent
3633 */
3634SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3635{
3636 struct kernel_siginfo info;
3637
3638 prepare_kill_siginfo(sig, &info);
3639
3640 return kill_something_info(sig, &info, pid);
3641}
3642
3643/*
3644 * Verify that the signaler and signalee either are in the same pid namespace
3645 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3646 * namespace.
3647 */
3648static bool access_pidfd_pidns(struct pid *pid)
3649{
3650 struct pid_namespace *active = task_active_pid_ns(current);
3651 struct pid_namespace *p = ns_of_pid(pid);
3652
3653 for (;;) {
3654 if (!p)
3655 return false;
3656 if (p == active)
3657 break;
3658 p = p->parent;
3659 }
3660
3661 return true;
3662}
3663
3664static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo, siginfo_t *info)
3665{
3666#ifdef CONFIG_COMPAT
3667 /*
3668 * Avoid hooking up compat syscalls and instead handle necessary
3669 * conversions here. Note, this is a stop-gap measure and should not be
3670 * considered a generic solution.
3671 */
3672 if (in_compat_syscall())
3673 return copy_siginfo_from_user32(
3674 kinfo, (struct compat_siginfo __user *)info);
3675#endif
3676 return copy_siginfo_from_user(kinfo, info);
3677}
3678
3679static struct pid *pidfd_to_pid(const struct file *file)
3680{
3681 struct pid *pid;
3682
3683 pid = pidfd_pid(file);
3684 if (!IS_ERR(pid))
3685 return pid;
3686
3687 return tgid_pidfd_to_pid(file);
3688}
3689
3690/**
3691 * sys_pidfd_send_signal - Signal a process through a pidfd
3692 * @pidfd: file descriptor of the process
3693 * @sig: signal to send
3694 * @info: signal info
3695 * @flags: future flags
3696 *
3697 * The syscall currently only signals via PIDTYPE_PID which covers
3698 * kill(<positive-pid>, <signal>. It does not signal threads or process
3699 * groups.
3700 * In order to extend the syscall to threads and process groups the @flags
3701 * argument should be used. In essence, the @flags argument will determine
3702 * what is signaled and not the file descriptor itself. Put in other words,
3703 * grouping is a property of the flags argument not a property of the file
3704 * descriptor.
3705 *
3706 * Return: 0 on success, negative errno on failure
3707 */
3708SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3709 siginfo_t __user *, info, unsigned int, flags)
3710{
3711 int ret;
3712 struct fd f;
3713 struct pid *pid;
3714 kernel_siginfo_t kinfo;
3715
3716 /* Enforce flags be set to 0 until we add an extension. */
3717 if (flags)
3718 return -EINVAL;
3719
3720 f = fdget(pidfd);
3721 if (!f.file)
3722 return -EBADF;
3723
3724 /* Is this a pidfd? */
3725 pid = pidfd_to_pid(f.file);
3726 if (IS_ERR(pid)) {
3727 ret = PTR_ERR(pid);
3728 goto err;
3729 }
3730
3731 ret = -EINVAL;
3732 if (!access_pidfd_pidns(pid))
3733 goto err;
3734
3735 if (info) {
3736 ret = copy_siginfo_from_user_any(&kinfo, info);
3737 if (unlikely(ret))
3738 goto err;
3739
3740 ret = -EINVAL;
3741 if (unlikely(sig != kinfo.si_signo))
3742 goto err;
3743
3744 /* Only allow sending arbitrary signals to yourself. */
3745 ret = -EPERM;
3746 if ((task_pid(current) != pid) &&
3747 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3748 goto err;
3749 } else {
3750 prepare_kill_siginfo(sig, &kinfo);
3751 }
3752
3753 ret = kill_pid_info(sig, &kinfo, pid);
3754
3755err:
3756 fdput(f);
3757 return ret;
3758}
3759
3760static int
3761do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3762{
3763 struct task_struct *p;
3764 int error = -ESRCH;
3765
3766 rcu_read_lock();
3767 p = find_task_by_vpid(pid);
3768 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3769 error = check_kill_permission(sig, info, p);
3770 /*
3771 * The null signal is a permissions and process existence
3772 * probe. No signal is actually delivered.
3773 */
3774 if (!error && sig) {
3775 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3776 /*
3777 * If lock_task_sighand() failed we pretend the task
3778 * dies after receiving the signal. The window is tiny,
3779 * and the signal is private anyway.
3780 */
3781 if (unlikely(error == -ESRCH))
3782 error = 0;
3783 }
3784 }
3785 rcu_read_unlock();
3786
3787 return error;
3788}
3789
3790static int do_tkill(pid_t tgid, pid_t pid, int sig)
3791{
3792 struct kernel_siginfo info;
3793
3794 clear_siginfo(&info);
3795 info.si_signo = sig;
3796 info.si_errno = 0;
3797 info.si_code = SI_TKILL;
3798 info.si_pid = task_tgid_vnr(current);
3799 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3800
3801 return do_send_specific(tgid, pid, sig, &info);
3802}
3803
3804/**
3805 * sys_tgkill - send signal to one specific thread
3806 * @tgid: the thread group ID of the thread
3807 * @pid: the PID of the thread
3808 * @sig: signal to be sent
3809 *
3810 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3811 * exists but it's not belonging to the target process anymore. This
3812 * method solves the problem of threads exiting and PIDs getting reused.
3813 */
3814SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3815{
3816 /* This is only valid for single tasks */
3817 if (pid <= 0 || tgid <= 0)
3818 return -EINVAL;
3819
3820 return do_tkill(tgid, pid, sig);
3821}
3822
3823/**
3824 * sys_tkill - send signal to one specific task
3825 * @pid: the PID of the task
3826 * @sig: signal to be sent
3827 *
3828 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3829 */
3830SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3831{
3832 /* This is only valid for single tasks */
3833 if (pid <= 0)
3834 return -EINVAL;
3835
3836 return do_tkill(0, pid, sig);
3837}
3838
3839static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3840{
3841 /* Not even root can pretend to send signals from the kernel.
3842 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3843 */
3844 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3845 (task_pid_vnr(current) != pid))
3846 return -EPERM;
3847
3848 /* POSIX.1b doesn't mention process groups. */
3849 return kill_proc_info(sig, info, pid);
3850}
3851
3852/**
3853 * sys_rt_sigqueueinfo - send signal information to a signal
3854 * @pid: the PID of the thread
3855 * @sig: signal to be sent
3856 * @uinfo: signal info to be sent
3857 */
3858SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3859 siginfo_t __user *, uinfo)
3860{
3861 kernel_siginfo_t info;
3862 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3863 if (unlikely(ret))
3864 return ret;
3865 return do_rt_sigqueueinfo(pid, sig, &info);
3866}
3867
3868#ifdef CONFIG_COMPAT
3869COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3870 compat_pid_t, pid,
3871 int, sig,
3872 struct compat_siginfo __user *, uinfo)
3873{
3874 kernel_siginfo_t info;
3875 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3876 if (unlikely(ret))
3877 return ret;
3878 return do_rt_sigqueueinfo(pid, sig, &info);
3879}
3880#endif
3881
3882static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3883{
3884 /* This is only valid for single tasks */
3885 if (pid <= 0 || tgid <= 0)
3886 return -EINVAL;
3887
3888 /* Not even root can pretend to send signals from the kernel.
3889 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3890 */
3891 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3892 (task_pid_vnr(current) != pid))
3893 return -EPERM;
3894
3895 return do_send_specific(tgid, pid, sig, info);
3896}
3897
3898SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3899 siginfo_t __user *, uinfo)
3900{
3901 kernel_siginfo_t info;
3902 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3903 if (unlikely(ret))
3904 return ret;
3905 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3906}
3907
3908#ifdef CONFIG_COMPAT
3909COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3910 compat_pid_t, tgid,
3911 compat_pid_t, pid,
3912 int, sig,
3913 struct compat_siginfo __user *, uinfo)
3914{
3915 kernel_siginfo_t info;
3916 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3917 if (unlikely(ret))
3918 return ret;
3919 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3920}
3921#endif
3922
3923/*
3924 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
3925 */
3926void kernel_sigaction(int sig, __sighandler_t action)
3927{
3928 spin_lock_irq(¤t->sighand->siglock);
3929 current->sighand->action[sig - 1].sa.sa_handler = action;
3930 if (action == SIG_IGN) {
3931 sigset_t mask;
3932
3933 sigemptyset(&mask);
3934 sigaddset(&mask, sig);
3935
3936 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
3937 flush_sigqueue_mask(&mask, ¤t->pending);
3938 recalc_sigpending();
3939 }
3940 spin_unlock_irq(¤t->sighand->siglock);
3941}
3942EXPORT_SYMBOL(kernel_sigaction);
3943
3944void __weak sigaction_compat_abi(struct k_sigaction *act,
3945 struct k_sigaction *oact)
3946{
3947}
3948
3949int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
3950{
3951 struct task_struct *p = current, *t;
3952 struct k_sigaction *k;
3953 sigset_t mask;
3954
3955 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
3956 return -EINVAL;
3957
3958 k = &p->sighand->action[sig-1];
3959
3960 spin_lock_irq(&p->sighand->siglock);
3961 if (oact)
3962 *oact = *k;
3963
3964 sigaction_compat_abi(act, oact);
3965
3966 if (act) {
3967 sigdelsetmask(&act->sa.sa_mask,
3968 sigmask(SIGKILL) | sigmask(SIGSTOP));
3969 *k = *act;
3970 /*
3971 * POSIX 3.3.1.3:
3972 * "Setting a signal action to SIG_IGN for a signal that is
3973 * pending shall cause the pending signal to be discarded,
3974 * whether or not it is blocked."
3975 *
3976 * "Setting a signal action to SIG_DFL for a signal that is
3977 * pending and whose default action is to ignore the signal
3978 * (for example, SIGCHLD), shall cause the pending signal to
3979 * be discarded, whether or not it is blocked"
3980 */
3981 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
3982 sigemptyset(&mask);
3983 sigaddset(&mask, sig);
3984 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
3985 for_each_thread(p, t)
3986 flush_sigqueue_mask(&mask, &t->pending);
3987 }
3988 }
3989
3990 spin_unlock_irq(&p->sighand->siglock);
3991 return 0;
3992}
3993
3994static int
3995do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
3996 size_t min_ss_size)
3997{
3998 struct task_struct *t = current;
3999
4000 if (oss) {
4001 memset(oss, 0, sizeof(stack_t));
4002 oss->ss_sp = (void __user *) t->sas_ss_sp;
4003 oss->ss_size = t->sas_ss_size;
4004 oss->ss_flags = sas_ss_flags(sp) |
4005 (current->sas_ss_flags & SS_FLAG_BITS);
4006 }
4007
4008 if (ss) {
4009 void __user *ss_sp = ss->ss_sp;
4010 size_t ss_size = ss->ss_size;
4011 unsigned ss_flags = ss->ss_flags;
4012 int ss_mode;
4013
4014 if (unlikely(on_sig_stack(sp)))
4015 return -EPERM;
4016
4017 ss_mode = ss_flags & ~SS_FLAG_BITS;
4018 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4019 ss_mode != 0))
4020 return -EINVAL;
4021
4022 if (ss_mode == SS_DISABLE) {
4023 ss_size = 0;
4024 ss_sp = NULL;
4025 } else {
4026 if (unlikely(ss_size < min_ss_size))
4027 return -ENOMEM;
4028 }
4029
4030 t->sas_ss_sp = (unsigned long) ss_sp;
4031 t->sas_ss_size = ss_size;
4032 t->sas_ss_flags = ss_flags;
4033 }
4034 return 0;
4035}
4036
4037SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4038{
4039 stack_t new, old;
4040 int err;
4041 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4042 return -EFAULT;
4043 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4044 current_user_stack_pointer(),
4045 MINSIGSTKSZ);
4046 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4047 err = -EFAULT;
4048 return err;
4049}
4050
4051int restore_altstack(const stack_t __user *uss)
4052{
4053 stack_t new;
4054 if (copy_from_user(&new, uss, sizeof(stack_t)))
4055 return -EFAULT;
4056 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4057 MINSIGSTKSZ);
4058 /* squash all but EFAULT for now */
4059 return 0;
4060}
4061
4062int __save_altstack(stack_t __user *uss, unsigned long sp)
4063{
4064 struct task_struct *t = current;
4065 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4066 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4067 __put_user(t->sas_ss_size, &uss->ss_size);
4068 if (err)
4069 return err;
4070 if (t->sas_ss_flags & SS_AUTODISARM)
4071 sas_ss_reset(t);
4072 return 0;
4073}
4074
4075#ifdef CONFIG_COMPAT
4076static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4077 compat_stack_t __user *uoss_ptr)
4078{
4079 stack_t uss, uoss;
4080 int ret;
4081
4082 if (uss_ptr) {
4083 compat_stack_t uss32;
4084 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4085 return -EFAULT;
4086 uss.ss_sp = compat_ptr(uss32.ss_sp);
4087 uss.ss_flags = uss32.ss_flags;
4088 uss.ss_size = uss32.ss_size;
4089 }
4090 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4091 compat_user_stack_pointer(),
4092 COMPAT_MINSIGSTKSZ);
4093 if (ret >= 0 && uoss_ptr) {
4094 compat_stack_t old;
4095 memset(&old, 0, sizeof(old));
4096 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4097 old.ss_flags = uoss.ss_flags;
4098 old.ss_size = uoss.ss_size;
4099 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4100 ret = -EFAULT;
4101 }
4102 return ret;
4103}
4104
4105COMPAT_SYSCALL_DEFINE2(sigaltstack,
4106 const compat_stack_t __user *, uss_ptr,
4107 compat_stack_t __user *, uoss_ptr)
4108{
4109 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4110}
4111
4112int compat_restore_altstack(const compat_stack_t __user *uss)
4113{
4114 int err = do_compat_sigaltstack(uss, NULL);
4115 /* squash all but -EFAULT for now */
4116 return err == -EFAULT ? err : 0;
4117}
4118
4119int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4120{
4121 int err;
4122 struct task_struct *t = current;
4123 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4124 &uss->ss_sp) |
4125 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4126 __put_user(t->sas_ss_size, &uss->ss_size);
4127 if (err)
4128 return err;
4129 if (t->sas_ss_flags & SS_AUTODISARM)
4130 sas_ss_reset(t);
4131 return 0;
4132}
4133#endif
4134
4135#ifdef __ARCH_WANT_SYS_SIGPENDING
4136
4137/**
4138 * sys_sigpending - examine pending signals
4139 * @uset: where mask of pending signal is returned
4140 */
4141SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4142{
4143 sigset_t set;
4144
4145 if (sizeof(old_sigset_t) > sizeof(*uset))
4146 return -EINVAL;
4147
4148 do_sigpending(&set);
4149
4150 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4151 return -EFAULT;
4152
4153 return 0;
4154}
4155
4156#ifdef CONFIG_COMPAT
4157COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4158{
4159 sigset_t set;
4160
4161 do_sigpending(&set);
4162
4163 return put_user(set.sig[0], set32);
4164}
4165#endif
4166
4167#endif
4168
4169#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4170/**
4171 * sys_sigprocmask - examine and change blocked signals
4172 * @how: whether to add, remove, or set signals
4173 * @nset: signals to add or remove (if non-null)
4174 * @oset: previous value of signal mask if non-null
4175 *
4176 * Some platforms have their own version with special arguments;
4177 * others support only sys_rt_sigprocmask.
4178 */
4179
4180SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4181 old_sigset_t __user *, oset)
4182{
4183 old_sigset_t old_set, new_set;
4184 sigset_t new_blocked;
4185
4186 old_set = current->blocked.sig[0];
4187
4188 if (nset) {
4189 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4190 return -EFAULT;
4191
4192 new_blocked = current->blocked;
4193
4194 switch (how) {
4195 case SIG_BLOCK:
4196 sigaddsetmask(&new_blocked, new_set);
4197 break;
4198 case SIG_UNBLOCK:
4199 sigdelsetmask(&new_blocked, new_set);
4200 break;
4201 case SIG_SETMASK:
4202 new_blocked.sig[0] = new_set;
4203 break;
4204 default:
4205 return -EINVAL;
4206 }
4207
4208 set_current_blocked(&new_blocked);
4209 }
4210
4211 if (oset) {
4212 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4213 return -EFAULT;
4214 }
4215
4216 return 0;
4217}
4218#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4219
4220#ifndef CONFIG_ODD_RT_SIGACTION
4221/**
4222 * sys_rt_sigaction - alter an action taken by a process
4223 * @sig: signal to be sent
4224 * @act: new sigaction
4225 * @oact: used to save the previous sigaction
4226 * @sigsetsize: size of sigset_t type
4227 */
4228SYSCALL_DEFINE4(rt_sigaction, int, sig,
4229 const struct sigaction __user *, act,
4230 struct sigaction __user *, oact,
4231 size_t, sigsetsize)
4232{
4233 struct k_sigaction new_sa, old_sa;
4234 int ret;
4235
4236 /* XXX: Don't preclude handling different sized sigset_t's. */
4237 if (sigsetsize != sizeof(sigset_t))
4238 return -EINVAL;
4239
4240 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4241 return -EFAULT;
4242
4243 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4244 if (ret)
4245 return ret;
4246
4247 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4248 return -EFAULT;
4249
4250 return 0;
4251}
4252#ifdef CONFIG_COMPAT
4253COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4254 const struct compat_sigaction __user *, act,
4255 struct compat_sigaction __user *, oact,
4256 compat_size_t, sigsetsize)
4257{
4258 struct k_sigaction new_ka, old_ka;
4259#ifdef __ARCH_HAS_SA_RESTORER
4260 compat_uptr_t restorer;
4261#endif
4262 int ret;
4263
4264 /* XXX: Don't preclude handling different sized sigset_t's. */
4265 if (sigsetsize != sizeof(compat_sigset_t))
4266 return -EINVAL;
4267
4268 if (act) {
4269 compat_uptr_t handler;
4270 ret = get_user(handler, &act->sa_handler);
4271 new_ka.sa.sa_handler = compat_ptr(handler);
4272#ifdef __ARCH_HAS_SA_RESTORER
4273 ret |= get_user(restorer, &act->sa_restorer);
4274 new_ka.sa.sa_restorer = compat_ptr(restorer);
4275#endif
4276 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4277 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4278 if (ret)
4279 return -EFAULT;
4280 }
4281
4282 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4283 if (!ret && oact) {
4284 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4285 &oact->sa_handler);
4286 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4287 sizeof(oact->sa_mask));
4288 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4289#ifdef __ARCH_HAS_SA_RESTORER
4290 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4291 &oact->sa_restorer);
4292#endif
4293 }
4294 return ret;
4295}
4296#endif
4297#endif /* !CONFIG_ODD_RT_SIGACTION */
4298
4299#ifdef CONFIG_OLD_SIGACTION
4300SYSCALL_DEFINE3(sigaction, int, sig,
4301 const struct old_sigaction __user *, act,
4302 struct old_sigaction __user *, oact)
4303{
4304 struct k_sigaction new_ka, old_ka;
4305 int ret;
4306
4307 if (act) {
4308 old_sigset_t mask;
4309 if (!access_ok(act, sizeof(*act)) ||
4310 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4311 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4312 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4313 __get_user(mask, &act->sa_mask))
4314 return -EFAULT;
4315#ifdef __ARCH_HAS_KA_RESTORER
4316 new_ka.ka_restorer = NULL;
4317#endif
4318 siginitset(&new_ka.sa.sa_mask, mask);
4319 }
4320
4321 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4322
4323 if (!ret && oact) {
4324 if (!access_ok(oact, sizeof(*oact)) ||
4325 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4326 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4327 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4328 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4329 return -EFAULT;
4330 }
4331
4332 return ret;
4333}
4334#endif
4335#ifdef CONFIG_COMPAT_OLD_SIGACTION
4336COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4337 const struct compat_old_sigaction __user *, act,
4338 struct compat_old_sigaction __user *, oact)
4339{
4340 struct k_sigaction new_ka, old_ka;
4341 int ret;
4342 compat_old_sigset_t mask;
4343 compat_uptr_t handler, restorer;
4344
4345 if (act) {
4346 if (!access_ok(act, sizeof(*act)) ||
4347 __get_user(handler, &act->sa_handler) ||
4348 __get_user(restorer, &act->sa_restorer) ||
4349 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4350 __get_user(mask, &act->sa_mask))
4351 return -EFAULT;
4352
4353#ifdef __ARCH_HAS_KA_RESTORER
4354 new_ka.ka_restorer = NULL;
4355#endif
4356 new_ka.sa.sa_handler = compat_ptr(handler);
4357 new_ka.sa.sa_restorer = compat_ptr(restorer);
4358 siginitset(&new_ka.sa.sa_mask, mask);
4359 }
4360
4361 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4362
4363 if (!ret && oact) {
4364 if (!access_ok(oact, sizeof(*oact)) ||
4365 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4366 &oact->sa_handler) ||
4367 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4368 &oact->sa_restorer) ||
4369 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4370 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4371 return -EFAULT;
4372 }
4373 return ret;
4374}
4375#endif
4376
4377#ifdef CONFIG_SGETMASK_SYSCALL
4378
4379/*
4380 * For backwards compatibility. Functionality superseded by sigprocmask.
4381 */
4382SYSCALL_DEFINE0(sgetmask)
4383{
4384 /* SMP safe */
4385 return current->blocked.sig[0];
4386}
4387
4388SYSCALL_DEFINE1(ssetmask, int, newmask)
4389{
4390 int old = current->blocked.sig[0];
4391 sigset_t newset;
4392
4393 siginitset(&newset, newmask);
4394 set_current_blocked(&newset);
4395
4396 return old;
4397}
4398#endif /* CONFIG_SGETMASK_SYSCALL */
4399
4400#ifdef __ARCH_WANT_SYS_SIGNAL
4401/*
4402 * For backwards compatibility. Functionality superseded by sigaction.
4403 */
4404SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4405{
4406 struct k_sigaction new_sa, old_sa;
4407 int ret;
4408
4409 new_sa.sa.sa_handler = handler;
4410 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4411 sigemptyset(&new_sa.sa.sa_mask);
4412
4413 ret = do_sigaction(sig, &new_sa, &old_sa);
4414
4415 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4416}
4417#endif /* __ARCH_WANT_SYS_SIGNAL */
4418
4419#ifdef __ARCH_WANT_SYS_PAUSE
4420
4421SYSCALL_DEFINE0(pause)
4422{
4423 while (!signal_pending(current)) {
4424 __set_current_state(TASK_INTERRUPTIBLE);
4425 schedule();
4426 }
4427 return -ERESTARTNOHAND;
4428}
4429
4430#endif
4431
4432static int sigsuspend(sigset_t *set)
4433{
4434 current->saved_sigmask = current->blocked;
4435 set_current_blocked(set);
4436
4437 while (!signal_pending(current)) {
4438 __set_current_state(TASK_INTERRUPTIBLE);
4439 schedule();
4440 }
4441 set_restore_sigmask();
4442 return -ERESTARTNOHAND;
4443}
4444
4445/**
4446 * sys_rt_sigsuspend - replace the signal mask for a value with the
4447 * @unewset value until a signal is received
4448 * @unewset: new signal mask value
4449 * @sigsetsize: size of sigset_t type
4450 */
4451SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4452{
4453 sigset_t newset;
4454
4455 /* XXX: Don't preclude handling different sized sigset_t's. */
4456 if (sigsetsize != sizeof(sigset_t))
4457 return -EINVAL;
4458
4459 if (copy_from_user(&newset, unewset, sizeof(newset)))
4460 return -EFAULT;
4461 return sigsuspend(&newset);
4462}
4463
4464#ifdef CONFIG_COMPAT
4465COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4466{
4467 sigset_t newset;
4468
4469 /* XXX: Don't preclude handling different sized sigset_t's. */
4470 if (sigsetsize != sizeof(sigset_t))
4471 return -EINVAL;
4472
4473 if (get_compat_sigset(&newset, unewset))
4474 return -EFAULT;
4475 return sigsuspend(&newset);
4476}
4477#endif
4478
4479#ifdef CONFIG_OLD_SIGSUSPEND
4480SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4481{
4482 sigset_t blocked;
4483 siginitset(&blocked, mask);
4484 return sigsuspend(&blocked);
4485}
4486#endif
4487#ifdef CONFIG_OLD_SIGSUSPEND3
4488SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4489{
4490 sigset_t blocked;
4491 siginitset(&blocked, mask);
4492 return sigsuspend(&blocked);
4493}
4494#endif
4495
4496__weak const char *arch_vma_name(struct vm_area_struct *vma)
4497{
4498 return NULL;
4499}
4500
4501static inline void siginfo_buildtime_checks(void)
4502{
4503 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4504
4505 /* Verify the offsets in the two siginfos match */
4506#define CHECK_OFFSET(field) \
4507 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4508
4509 /* kill */
4510 CHECK_OFFSET(si_pid);
4511 CHECK_OFFSET(si_uid);
4512
4513 /* timer */
4514 CHECK_OFFSET(si_tid);
4515 CHECK_OFFSET(si_overrun);
4516 CHECK_OFFSET(si_value);
4517
4518 /* rt */
4519 CHECK_OFFSET(si_pid);
4520 CHECK_OFFSET(si_uid);
4521 CHECK_OFFSET(si_value);
4522
4523 /* sigchld */
4524 CHECK_OFFSET(si_pid);
4525 CHECK_OFFSET(si_uid);
4526 CHECK_OFFSET(si_status);
4527 CHECK_OFFSET(si_utime);
4528 CHECK_OFFSET(si_stime);
4529
4530 /* sigfault */
4531 CHECK_OFFSET(si_addr);
4532 CHECK_OFFSET(si_addr_lsb);
4533 CHECK_OFFSET(si_lower);
4534 CHECK_OFFSET(si_upper);
4535 CHECK_OFFSET(si_pkey);
4536
4537 /* sigpoll */
4538 CHECK_OFFSET(si_band);
4539 CHECK_OFFSET(si_fd);
4540
4541 /* sigsys */
4542 CHECK_OFFSET(si_call_addr);
4543 CHECK_OFFSET(si_syscall);
4544 CHECK_OFFSET(si_arch);
4545#undef CHECK_OFFSET
4546
4547 /* usb asyncio */
4548 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4549 offsetof(struct siginfo, si_addr));
4550 if (sizeof(int) == sizeof(void __user *)) {
4551 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4552 sizeof(void __user *));
4553 } else {
4554 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4555 sizeof_field(struct siginfo, si_uid)) !=
4556 sizeof(void __user *));
4557 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4558 offsetof(struct siginfo, si_uid));
4559 }
4560#ifdef CONFIG_COMPAT
4561 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4562 offsetof(struct compat_siginfo, si_addr));
4563 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4564 sizeof(compat_uptr_t));
4565 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4566 sizeof_field(struct siginfo, si_pid));
4567#endif
4568}
4569
4570void __init signals_init(void)
4571{
4572 siginfo_buildtime_checks();
4573
4574 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4575}
4576
4577#ifdef CONFIG_KGDB_KDB
4578#include <linux/kdb.h>
4579/*
4580 * kdb_send_sig - Allows kdb to send signals without exposing
4581 * signal internals. This function checks if the required locks are
4582 * available before calling the main signal code, to avoid kdb
4583 * deadlocks.
4584 */
4585void kdb_send_sig(struct task_struct *t, int sig)
4586{
4587 static struct task_struct *kdb_prev_t;
4588 int new_t, ret;
4589 if (!spin_trylock(&t->sighand->siglock)) {
4590 kdb_printf("Can't do kill command now.\n"
4591 "The sigmask lock is held somewhere else in "
4592 "kernel, try again later\n");
4593 return;
4594 }
4595 new_t = kdb_prev_t != t;
4596 kdb_prev_t = t;
4597 if (t->state != TASK_RUNNING && new_t) {
4598 spin_unlock(&t->sighand->siglock);
4599 kdb_printf("Process is not RUNNING, sending a signal from "
4600 "kdb risks deadlock\n"
4601 "on the run queue locks. "
4602 "The signal has _not_ been sent.\n"
4603 "Reissue the kill command if you want to risk "
4604 "the deadlock.\n");
4605 return;
4606 }
4607 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4608 spin_unlock(&t->sighand->siglock);
4609 if (ret)
4610 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4611 sig, t->pid);
4612 else
4613 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4614}
4615#endif /* CONFIG_KGDB_KDB */
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/signal.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 *
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
12 */
13
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>
24#include <linux/fs.h>
25#include <linux/mm.h>
26#include <linux/proc_fs.h>
27#include <linux/tty.h>
28#include <linux/binfmts.h>
29#include <linux/coredump.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
32#include <linux/ptrace.h>
33#include <linux/signal.h>
34#include <linux/signalfd.h>
35#include <linux/ratelimit.h>
36#include <linux/task_work.h>
37#include <linux/capability.h>
38#include <linux/freezer.h>
39#include <linux/pid_namespace.h>
40#include <linux/nsproxy.h>
41#include <linux/user_namespace.h>
42#include <linux/uprobes.h>
43#include <linux/compat.h>
44#include <linux/cn_proc.h>
45#include <linux/compiler.h>
46#include <linux/posix-timers.h>
47#include <linux/cgroup.h>
48#include <linux/audit.h>
49#include <linux/sysctl.h>
50
51#define CREATE_TRACE_POINTS
52#include <trace/events/signal.h>
53
54#include <asm/param.h>
55#include <linux/uaccess.h>
56#include <asm/unistd.h>
57#include <asm/siginfo.h>
58#include <asm/cacheflush.h>
59#include <asm/syscall.h> /* for syscall_get_* */
60
61/*
62 * SLAB caches for signal bits.
63 */
64
65static struct kmem_cache *sigqueue_cachep;
66
67int print_fatal_signals __read_mostly;
68
69static void __user *sig_handler(struct task_struct *t, int sig)
70{
71 return t->sighand->action[sig - 1].sa.sa_handler;
72}
73
74static inline bool sig_handler_ignored(void __user *handler, int sig)
75{
76 /* Is it explicitly or implicitly ignored? */
77 return handler == SIG_IGN ||
78 (handler == SIG_DFL && sig_kernel_ignore(sig));
79}
80
81static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
82{
83 void __user *handler;
84
85 handler = sig_handler(t, sig);
86
87 /* SIGKILL and SIGSTOP may not be sent to the global init */
88 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
89 return true;
90
91 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
92 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
93 return true;
94
95 /* Only allow kernel generated signals to this kthread */
96 if (unlikely((t->flags & PF_KTHREAD) &&
97 (handler == SIG_KTHREAD_KERNEL) && !force))
98 return true;
99
100 return sig_handler_ignored(handler, sig);
101}
102
103static bool sig_ignored(struct task_struct *t, int sig, bool force)
104{
105 /*
106 * Blocked signals are never ignored, since the
107 * signal handler may change by the time it is
108 * unblocked.
109 */
110 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
111 return false;
112
113 /*
114 * Tracers may want to know about even ignored signal unless it
115 * is SIGKILL which can't be reported anyway but can be ignored
116 * by SIGNAL_UNKILLABLE task.
117 */
118 if (t->ptrace && sig != SIGKILL)
119 return false;
120
121 return sig_task_ignored(t, sig, force);
122}
123
124/*
125 * Re-calculate pending state from the set of locally pending
126 * signals, globally pending signals, and blocked signals.
127 */
128static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
129{
130 unsigned long ready;
131 long i;
132
133 switch (_NSIG_WORDS) {
134 default:
135 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
136 ready |= signal->sig[i] &~ blocked->sig[i];
137 break;
138
139 case 4: ready = signal->sig[3] &~ blocked->sig[3];
140 ready |= signal->sig[2] &~ blocked->sig[2];
141 ready |= signal->sig[1] &~ blocked->sig[1];
142 ready |= signal->sig[0] &~ blocked->sig[0];
143 break;
144
145 case 2: ready = signal->sig[1] &~ blocked->sig[1];
146 ready |= signal->sig[0] &~ blocked->sig[0];
147 break;
148
149 case 1: ready = signal->sig[0] &~ blocked->sig[0];
150 }
151 return ready != 0;
152}
153
154#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
155
156static bool recalc_sigpending_tsk(struct task_struct *t)
157{
158 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
159 PENDING(&t->pending, &t->blocked) ||
160 PENDING(&t->signal->shared_pending, &t->blocked) ||
161 cgroup_task_frozen(t)) {
162 set_tsk_thread_flag(t, TIF_SIGPENDING);
163 return true;
164 }
165
166 /*
167 * We must never clear the flag in another thread, or in current
168 * when it's possible the current syscall is returning -ERESTART*.
169 * So we don't clear it here, and only callers who know they should do.
170 */
171 return false;
172}
173
174void recalc_sigpending(void)
175{
176 if (!recalc_sigpending_tsk(current) && !freezing(current))
177 clear_thread_flag(TIF_SIGPENDING);
178
179}
180EXPORT_SYMBOL(recalc_sigpending);
181
182void calculate_sigpending(void)
183{
184 /* Have any signals or users of TIF_SIGPENDING been delayed
185 * until after fork?
186 */
187 spin_lock_irq(¤t->sighand->siglock);
188 set_tsk_thread_flag(current, TIF_SIGPENDING);
189 recalc_sigpending();
190 spin_unlock_irq(¤t->sighand->siglock);
191}
192
193/* Given the mask, find the first available signal that should be serviced. */
194
195#define SYNCHRONOUS_MASK \
196 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
197 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
198
199int next_signal(struct sigpending *pending, sigset_t *mask)
200{
201 unsigned long i, *s, *m, x;
202 int sig = 0;
203
204 s = pending->signal.sig;
205 m = mask->sig;
206
207 /*
208 * Handle the first word specially: it contains the
209 * synchronous signals that need to be dequeued first.
210 */
211 x = *s &~ *m;
212 if (x) {
213 if (x & SYNCHRONOUS_MASK)
214 x &= SYNCHRONOUS_MASK;
215 sig = ffz(~x) + 1;
216 return sig;
217 }
218
219 switch (_NSIG_WORDS) {
220 default:
221 for (i = 1; i < _NSIG_WORDS; ++i) {
222 x = *++s &~ *++m;
223 if (!x)
224 continue;
225 sig = ffz(~x) + i*_NSIG_BPW + 1;
226 break;
227 }
228 break;
229
230 case 2:
231 x = s[1] &~ m[1];
232 if (!x)
233 break;
234 sig = ffz(~x) + _NSIG_BPW + 1;
235 break;
236
237 case 1:
238 /* Nothing to do */
239 break;
240 }
241
242 return sig;
243}
244
245static inline void print_dropped_signal(int sig)
246{
247 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
248
249 if (!print_fatal_signals)
250 return;
251
252 if (!__ratelimit(&ratelimit_state))
253 return;
254
255 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
256 current->comm, current->pid, sig);
257}
258
259/**
260 * task_set_jobctl_pending - set jobctl pending bits
261 * @task: target task
262 * @mask: pending bits to set
263 *
264 * Clear @mask from @task->jobctl. @mask must be subset of
265 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
266 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
267 * cleared. If @task is already being killed or exiting, this function
268 * becomes noop.
269 *
270 * CONTEXT:
271 * Must be called with @task->sighand->siglock held.
272 *
273 * RETURNS:
274 * %true if @mask is set, %false if made noop because @task was dying.
275 */
276bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
277{
278 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
279 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
280 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
281
282 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
283 return false;
284
285 if (mask & JOBCTL_STOP_SIGMASK)
286 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
287
288 task->jobctl |= mask;
289 return true;
290}
291
292/**
293 * task_clear_jobctl_trapping - clear jobctl trapping bit
294 * @task: target task
295 *
296 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
297 * Clear it and wake up the ptracer. Note that we don't need any further
298 * locking. @task->siglock guarantees that @task->parent points to the
299 * ptracer.
300 *
301 * CONTEXT:
302 * Must be called with @task->sighand->siglock held.
303 */
304void task_clear_jobctl_trapping(struct task_struct *task)
305{
306 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
307 task->jobctl &= ~JOBCTL_TRAPPING;
308 smp_mb(); /* advised by wake_up_bit() */
309 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
310 }
311}
312
313/**
314 * task_clear_jobctl_pending - clear jobctl pending bits
315 * @task: target task
316 * @mask: pending bits to clear
317 *
318 * Clear @mask from @task->jobctl. @mask must be subset of
319 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
320 * STOP bits are cleared together.
321 *
322 * If clearing of @mask leaves no stop or trap pending, this function calls
323 * task_clear_jobctl_trapping().
324 *
325 * CONTEXT:
326 * Must be called with @task->sighand->siglock held.
327 */
328void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
329{
330 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
331
332 if (mask & JOBCTL_STOP_PENDING)
333 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
334
335 task->jobctl &= ~mask;
336
337 if (!(task->jobctl & JOBCTL_PENDING_MASK))
338 task_clear_jobctl_trapping(task);
339}
340
341/**
342 * task_participate_group_stop - participate in a group stop
343 * @task: task participating in a group stop
344 *
345 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
346 * Group stop states are cleared and the group stop count is consumed if
347 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
348 * stop, the appropriate `SIGNAL_*` flags are set.
349 *
350 * CONTEXT:
351 * Must be called with @task->sighand->siglock held.
352 *
353 * RETURNS:
354 * %true if group stop completion should be notified to the parent, %false
355 * otherwise.
356 */
357static bool task_participate_group_stop(struct task_struct *task)
358{
359 struct signal_struct *sig = task->signal;
360 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
361
362 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
363
364 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
365
366 if (!consume)
367 return false;
368
369 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
370 sig->group_stop_count--;
371
372 /*
373 * Tell the caller to notify completion iff we are entering into a
374 * fresh group stop. Read comment in do_signal_stop() for details.
375 */
376 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
377 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
378 return true;
379 }
380 return false;
381}
382
383void task_join_group_stop(struct task_struct *task)
384{
385 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
386 struct signal_struct *sig = current->signal;
387
388 if (sig->group_stop_count) {
389 sig->group_stop_count++;
390 mask |= JOBCTL_STOP_CONSUME;
391 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
392 return;
393
394 /* Have the new thread join an on-going signal group stop */
395 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
396}
397
398/*
399 * allocate a new signal queue record
400 * - this may be called without locks if and only if t == current, otherwise an
401 * appropriate lock must be held to stop the target task from exiting
402 */
403static struct sigqueue *
404__sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
405 int override_rlimit, const unsigned int sigqueue_flags)
406{
407 struct sigqueue *q = NULL;
408 struct ucounts *ucounts;
409 long sigpending;
410
411 /*
412 * Protect access to @t credentials. This can go away when all
413 * callers hold rcu read lock.
414 *
415 * NOTE! A pending signal will hold on to the user refcount,
416 * and we get/put the refcount only when the sigpending count
417 * changes from/to zero.
418 */
419 rcu_read_lock();
420 ucounts = task_ucounts(t);
421 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
422 rcu_read_unlock();
423 if (!sigpending)
424 return NULL;
425
426 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
427 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
428 } else {
429 print_dropped_signal(sig);
430 }
431
432 if (unlikely(q == NULL)) {
433 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
434 } else {
435 INIT_LIST_HEAD(&q->list);
436 q->flags = sigqueue_flags;
437 q->ucounts = ucounts;
438 }
439 return q;
440}
441
442static void __sigqueue_free(struct sigqueue *q)
443{
444 if (q->flags & SIGQUEUE_PREALLOC)
445 return;
446 if (q->ucounts) {
447 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
448 q->ucounts = NULL;
449 }
450 kmem_cache_free(sigqueue_cachep, q);
451}
452
453void flush_sigqueue(struct sigpending *queue)
454{
455 struct sigqueue *q;
456
457 sigemptyset(&queue->signal);
458 while (!list_empty(&queue->list)) {
459 q = list_entry(queue->list.next, struct sigqueue , list);
460 list_del_init(&q->list);
461 __sigqueue_free(q);
462 }
463}
464
465/*
466 * Flush all pending signals for this kthread.
467 */
468void flush_signals(struct task_struct *t)
469{
470 unsigned long flags;
471
472 spin_lock_irqsave(&t->sighand->siglock, flags);
473 clear_tsk_thread_flag(t, TIF_SIGPENDING);
474 flush_sigqueue(&t->pending);
475 flush_sigqueue(&t->signal->shared_pending);
476 spin_unlock_irqrestore(&t->sighand->siglock, flags);
477}
478EXPORT_SYMBOL(flush_signals);
479
480#ifdef CONFIG_POSIX_TIMERS
481static void __flush_itimer_signals(struct sigpending *pending)
482{
483 sigset_t signal, retain;
484 struct sigqueue *q, *n;
485
486 signal = pending->signal;
487 sigemptyset(&retain);
488
489 list_for_each_entry_safe(q, n, &pending->list, list) {
490 int sig = q->info.si_signo;
491
492 if (likely(q->info.si_code != SI_TIMER)) {
493 sigaddset(&retain, sig);
494 } else {
495 sigdelset(&signal, sig);
496 list_del_init(&q->list);
497 __sigqueue_free(q);
498 }
499 }
500
501 sigorsets(&pending->signal, &signal, &retain);
502}
503
504void flush_itimer_signals(void)
505{
506 struct task_struct *tsk = current;
507 unsigned long flags;
508
509 spin_lock_irqsave(&tsk->sighand->siglock, flags);
510 __flush_itimer_signals(&tsk->pending);
511 __flush_itimer_signals(&tsk->signal->shared_pending);
512 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
513}
514#endif
515
516void ignore_signals(struct task_struct *t)
517{
518 int i;
519
520 for (i = 0; i < _NSIG; ++i)
521 t->sighand->action[i].sa.sa_handler = SIG_IGN;
522
523 flush_signals(t);
524}
525
526/*
527 * Flush all handlers for a task.
528 */
529
530void
531flush_signal_handlers(struct task_struct *t, int force_default)
532{
533 int i;
534 struct k_sigaction *ka = &t->sighand->action[0];
535 for (i = _NSIG ; i != 0 ; i--) {
536 if (force_default || ka->sa.sa_handler != SIG_IGN)
537 ka->sa.sa_handler = SIG_DFL;
538 ka->sa.sa_flags = 0;
539#ifdef __ARCH_HAS_SA_RESTORER
540 ka->sa.sa_restorer = NULL;
541#endif
542 sigemptyset(&ka->sa.sa_mask);
543 ka++;
544 }
545}
546
547bool unhandled_signal(struct task_struct *tsk, int sig)
548{
549 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
550 if (is_global_init(tsk))
551 return true;
552
553 if (handler != SIG_IGN && handler != SIG_DFL)
554 return false;
555
556 /* If dying, we handle all new signals by ignoring them */
557 if (fatal_signal_pending(tsk))
558 return false;
559
560 /* if ptraced, let the tracer determine */
561 return !tsk->ptrace;
562}
563
564static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
565 bool *resched_timer)
566{
567 struct sigqueue *q, *first = NULL;
568
569 /*
570 * Collect the siginfo appropriate to this signal. Check if
571 * there is another siginfo for the same signal.
572 */
573 list_for_each_entry(q, &list->list, list) {
574 if (q->info.si_signo == sig) {
575 if (first)
576 goto still_pending;
577 first = q;
578 }
579 }
580
581 sigdelset(&list->signal, sig);
582
583 if (first) {
584still_pending:
585 list_del_init(&first->list);
586 copy_siginfo(info, &first->info);
587
588 *resched_timer =
589 (first->flags & SIGQUEUE_PREALLOC) &&
590 (info->si_code == SI_TIMER) &&
591 (info->si_sys_private);
592
593 __sigqueue_free(first);
594 } else {
595 /*
596 * Ok, it wasn't in the queue. This must be
597 * a fast-pathed signal or we must have been
598 * out of queue space. So zero out the info.
599 */
600 clear_siginfo(info);
601 info->si_signo = sig;
602 info->si_errno = 0;
603 info->si_code = SI_USER;
604 info->si_pid = 0;
605 info->si_uid = 0;
606 }
607}
608
609static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
610 kernel_siginfo_t *info, bool *resched_timer)
611{
612 int sig = next_signal(pending, mask);
613
614 if (sig)
615 collect_signal(sig, pending, info, resched_timer);
616 return sig;
617}
618
619/*
620 * Dequeue a signal and return the element to the caller, which is
621 * expected to free it.
622 *
623 * All callers have to hold the siglock.
624 */
625int dequeue_signal(struct task_struct *tsk, sigset_t *mask,
626 kernel_siginfo_t *info, enum pid_type *type)
627{
628 bool resched_timer = false;
629 int signr;
630
631 /* We only dequeue private signals from ourselves, we don't let
632 * signalfd steal them
633 */
634 *type = PIDTYPE_PID;
635 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
636 if (!signr) {
637 *type = PIDTYPE_TGID;
638 signr = __dequeue_signal(&tsk->signal->shared_pending,
639 mask, info, &resched_timer);
640#ifdef CONFIG_POSIX_TIMERS
641 /*
642 * itimer signal ?
643 *
644 * itimers are process shared and we restart periodic
645 * itimers in the signal delivery path to prevent DoS
646 * attacks in the high resolution timer case. This is
647 * compliant with the old way of self-restarting
648 * itimers, as the SIGALRM is a legacy signal and only
649 * queued once. Changing the restart behaviour to
650 * restart the timer in the signal dequeue path is
651 * reducing the timer noise on heavy loaded !highres
652 * systems too.
653 */
654 if (unlikely(signr == SIGALRM)) {
655 struct hrtimer *tmr = &tsk->signal->real_timer;
656
657 if (!hrtimer_is_queued(tmr) &&
658 tsk->signal->it_real_incr != 0) {
659 hrtimer_forward(tmr, tmr->base->get_time(),
660 tsk->signal->it_real_incr);
661 hrtimer_restart(tmr);
662 }
663 }
664#endif
665 }
666
667 recalc_sigpending();
668 if (!signr)
669 return 0;
670
671 if (unlikely(sig_kernel_stop(signr))) {
672 /*
673 * Set a marker that we have dequeued a stop signal. Our
674 * caller might release the siglock and then the pending
675 * stop signal it is about to process is no longer in the
676 * pending bitmasks, but must still be cleared by a SIGCONT
677 * (and overruled by a SIGKILL). So those cases clear this
678 * shared flag after we've set it. Note that this flag may
679 * remain set after the signal we return is ignored or
680 * handled. That doesn't matter because its only purpose
681 * is to alert stop-signal processing code when another
682 * processor has come along and cleared the flag.
683 */
684 current->jobctl |= JOBCTL_STOP_DEQUEUED;
685 }
686#ifdef CONFIG_POSIX_TIMERS
687 if (resched_timer) {
688 /*
689 * Release the siglock to ensure proper locking order
690 * of timer locks outside of siglocks. Note, we leave
691 * irqs disabled here, since the posix-timers code is
692 * about to disable them again anyway.
693 */
694 spin_unlock(&tsk->sighand->siglock);
695 posixtimer_rearm(info);
696 spin_lock(&tsk->sighand->siglock);
697
698 /* Don't expose the si_sys_private value to userspace */
699 info->si_sys_private = 0;
700 }
701#endif
702 return signr;
703}
704EXPORT_SYMBOL_GPL(dequeue_signal);
705
706static int dequeue_synchronous_signal(kernel_siginfo_t *info)
707{
708 struct task_struct *tsk = current;
709 struct sigpending *pending = &tsk->pending;
710 struct sigqueue *q, *sync = NULL;
711
712 /*
713 * Might a synchronous signal be in the queue?
714 */
715 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
716 return 0;
717
718 /*
719 * Return the first synchronous signal in the queue.
720 */
721 list_for_each_entry(q, &pending->list, list) {
722 /* Synchronous signals have a positive si_code */
723 if ((q->info.si_code > SI_USER) &&
724 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
725 sync = q;
726 goto next;
727 }
728 }
729 return 0;
730next:
731 /*
732 * Check if there is another siginfo for the same signal.
733 */
734 list_for_each_entry_continue(q, &pending->list, list) {
735 if (q->info.si_signo == sync->info.si_signo)
736 goto still_pending;
737 }
738
739 sigdelset(&pending->signal, sync->info.si_signo);
740 recalc_sigpending();
741still_pending:
742 list_del_init(&sync->list);
743 copy_siginfo(info, &sync->info);
744 __sigqueue_free(sync);
745 return info->si_signo;
746}
747
748/*
749 * Tell a process that it has a new active signal..
750 *
751 * NOTE! we rely on the previous spin_lock to
752 * lock interrupts for us! We can only be called with
753 * "siglock" held, and the local interrupt must
754 * have been disabled when that got acquired!
755 *
756 * No need to set need_resched since signal event passing
757 * goes through ->blocked
758 */
759void signal_wake_up_state(struct task_struct *t, unsigned int state)
760{
761 lockdep_assert_held(&t->sighand->siglock);
762
763 set_tsk_thread_flag(t, TIF_SIGPENDING);
764
765 /*
766 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
767 * case. We don't check t->state here because there is a race with it
768 * executing another processor and just now entering stopped state.
769 * By using wake_up_state, we ensure the process will wake up and
770 * handle its death signal.
771 */
772 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
773 kick_process(t);
774}
775
776/*
777 * Remove signals in mask from the pending set and queue.
778 * Returns 1 if any signals were found.
779 *
780 * All callers must be holding the siglock.
781 */
782static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
783{
784 struct sigqueue *q, *n;
785 sigset_t m;
786
787 sigandsets(&m, mask, &s->signal);
788 if (sigisemptyset(&m))
789 return;
790
791 sigandnsets(&s->signal, &s->signal, mask);
792 list_for_each_entry_safe(q, n, &s->list, list) {
793 if (sigismember(mask, q->info.si_signo)) {
794 list_del_init(&q->list);
795 __sigqueue_free(q);
796 }
797 }
798}
799
800static inline int is_si_special(const struct kernel_siginfo *info)
801{
802 return info <= SEND_SIG_PRIV;
803}
804
805static inline bool si_fromuser(const struct kernel_siginfo *info)
806{
807 return info == SEND_SIG_NOINFO ||
808 (!is_si_special(info) && SI_FROMUSER(info));
809}
810
811/*
812 * called with RCU read lock from check_kill_permission()
813 */
814static bool kill_ok_by_cred(struct task_struct *t)
815{
816 const struct cred *cred = current_cred();
817 const struct cred *tcred = __task_cred(t);
818
819 return uid_eq(cred->euid, tcred->suid) ||
820 uid_eq(cred->euid, tcred->uid) ||
821 uid_eq(cred->uid, tcred->suid) ||
822 uid_eq(cred->uid, tcred->uid) ||
823 ns_capable(tcred->user_ns, CAP_KILL);
824}
825
826/*
827 * Bad permissions for sending the signal
828 * - the caller must hold the RCU read lock
829 */
830static int check_kill_permission(int sig, struct kernel_siginfo *info,
831 struct task_struct *t)
832{
833 struct pid *sid;
834 int error;
835
836 if (!valid_signal(sig))
837 return -EINVAL;
838
839 if (!si_fromuser(info))
840 return 0;
841
842 error = audit_signal_info(sig, t); /* Let audit system see the signal */
843 if (error)
844 return error;
845
846 if (!same_thread_group(current, t) &&
847 !kill_ok_by_cred(t)) {
848 switch (sig) {
849 case SIGCONT:
850 sid = task_session(t);
851 /*
852 * We don't return the error if sid == NULL. The
853 * task was unhashed, the caller must notice this.
854 */
855 if (!sid || sid == task_session(current))
856 break;
857 fallthrough;
858 default:
859 return -EPERM;
860 }
861 }
862
863 return security_task_kill(t, info, sig, NULL);
864}
865
866/**
867 * ptrace_trap_notify - schedule trap to notify ptracer
868 * @t: tracee wanting to notify tracer
869 *
870 * This function schedules sticky ptrace trap which is cleared on the next
871 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
872 * ptracer.
873 *
874 * If @t is running, STOP trap will be taken. If trapped for STOP and
875 * ptracer is listening for events, tracee is woken up so that it can
876 * re-trap for the new event. If trapped otherwise, STOP trap will be
877 * eventually taken without returning to userland after the existing traps
878 * are finished by PTRACE_CONT.
879 *
880 * CONTEXT:
881 * Must be called with @task->sighand->siglock held.
882 */
883static void ptrace_trap_notify(struct task_struct *t)
884{
885 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
886 lockdep_assert_held(&t->sighand->siglock);
887
888 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
889 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
890}
891
892/*
893 * Handle magic process-wide effects of stop/continue signals. Unlike
894 * the signal actions, these happen immediately at signal-generation
895 * time regardless of blocking, ignoring, or handling. This does the
896 * actual continuing for SIGCONT, but not the actual stopping for stop
897 * signals. The process stop is done as a signal action for SIG_DFL.
898 *
899 * Returns true if the signal should be actually delivered, otherwise
900 * it should be dropped.
901 */
902static bool prepare_signal(int sig, struct task_struct *p, bool force)
903{
904 struct signal_struct *signal = p->signal;
905 struct task_struct *t;
906 sigset_t flush;
907
908 if (signal->flags & SIGNAL_GROUP_EXIT) {
909 if (signal->core_state)
910 return sig == SIGKILL;
911 /*
912 * The process is in the middle of dying, drop the signal.
913 */
914 return false;
915 } else if (sig_kernel_stop(sig)) {
916 /*
917 * This is a stop signal. Remove SIGCONT from all queues.
918 */
919 siginitset(&flush, sigmask(SIGCONT));
920 flush_sigqueue_mask(&flush, &signal->shared_pending);
921 for_each_thread(p, t)
922 flush_sigqueue_mask(&flush, &t->pending);
923 } else if (sig == SIGCONT) {
924 unsigned int why;
925 /*
926 * Remove all stop signals from all queues, wake all threads.
927 */
928 siginitset(&flush, SIG_KERNEL_STOP_MASK);
929 flush_sigqueue_mask(&flush, &signal->shared_pending);
930 for_each_thread(p, t) {
931 flush_sigqueue_mask(&flush, &t->pending);
932 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
933 if (likely(!(t->ptrace & PT_SEIZED))) {
934 t->jobctl &= ~JOBCTL_STOPPED;
935 wake_up_state(t, __TASK_STOPPED);
936 } else
937 ptrace_trap_notify(t);
938 }
939
940 /*
941 * Notify the parent with CLD_CONTINUED if we were stopped.
942 *
943 * If we were in the middle of a group stop, we pretend it
944 * was already finished, and then continued. Since SIGCHLD
945 * doesn't queue we report only CLD_STOPPED, as if the next
946 * CLD_CONTINUED was dropped.
947 */
948 why = 0;
949 if (signal->flags & SIGNAL_STOP_STOPPED)
950 why |= SIGNAL_CLD_CONTINUED;
951 else if (signal->group_stop_count)
952 why |= SIGNAL_CLD_STOPPED;
953
954 if (why) {
955 /*
956 * The first thread which returns from do_signal_stop()
957 * will take ->siglock, notice SIGNAL_CLD_MASK, and
958 * notify its parent. See get_signal().
959 */
960 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
961 signal->group_stop_count = 0;
962 signal->group_exit_code = 0;
963 }
964 }
965
966 return !sig_ignored(p, sig, force);
967}
968
969/*
970 * Test if P wants to take SIG. After we've checked all threads with this,
971 * it's equivalent to finding no threads not blocking SIG. Any threads not
972 * blocking SIG were ruled out because they are not running and already
973 * have pending signals. Such threads will dequeue from the shared queue
974 * as soon as they're available, so putting the signal on the shared queue
975 * will be equivalent to sending it to one such thread.
976 */
977static inline bool wants_signal(int sig, struct task_struct *p)
978{
979 if (sigismember(&p->blocked, sig))
980 return false;
981
982 if (p->flags & PF_EXITING)
983 return false;
984
985 if (sig == SIGKILL)
986 return true;
987
988 if (task_is_stopped_or_traced(p))
989 return false;
990
991 return task_curr(p) || !task_sigpending(p);
992}
993
994static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
995{
996 struct signal_struct *signal = p->signal;
997 struct task_struct *t;
998
999 /*
1000 * Now find a thread we can wake up to take the signal off the queue.
1001 *
1002 * Try the suggested task first (may or may not be the main thread).
1003 */
1004 if (wants_signal(sig, p))
1005 t = p;
1006 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1007 /*
1008 * There is just one thread and it does not need to be woken.
1009 * It will dequeue unblocked signals before it runs again.
1010 */
1011 return;
1012 else {
1013 /*
1014 * Otherwise try to find a suitable thread.
1015 */
1016 t = signal->curr_target;
1017 while (!wants_signal(sig, t)) {
1018 t = next_thread(t);
1019 if (t == signal->curr_target)
1020 /*
1021 * No thread needs to be woken.
1022 * Any eligible threads will see
1023 * the signal in the queue soon.
1024 */
1025 return;
1026 }
1027 signal->curr_target = t;
1028 }
1029
1030 /*
1031 * Found a killable thread. If the signal will be fatal,
1032 * then start taking the whole group down immediately.
1033 */
1034 if (sig_fatal(p, sig) &&
1035 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1036 !sigismember(&t->real_blocked, sig) &&
1037 (sig == SIGKILL || !p->ptrace)) {
1038 /*
1039 * This signal will be fatal to the whole group.
1040 */
1041 if (!sig_kernel_coredump(sig)) {
1042 /*
1043 * Start a group exit and wake everybody up.
1044 * This way we don't have other threads
1045 * running and doing things after a slower
1046 * thread has the fatal signal pending.
1047 */
1048 signal->flags = SIGNAL_GROUP_EXIT;
1049 signal->group_exit_code = sig;
1050 signal->group_stop_count = 0;
1051 __for_each_thread(signal, t) {
1052 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1053 sigaddset(&t->pending.signal, SIGKILL);
1054 signal_wake_up(t, 1);
1055 }
1056 return;
1057 }
1058 }
1059
1060 /*
1061 * The signal is already in the shared-pending queue.
1062 * Tell the chosen thread to wake up and dequeue it.
1063 */
1064 signal_wake_up(t, sig == SIGKILL);
1065 return;
1066}
1067
1068static inline bool legacy_queue(struct sigpending *signals, int sig)
1069{
1070 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1071}
1072
1073static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1074 struct task_struct *t, enum pid_type type, bool force)
1075{
1076 struct sigpending *pending;
1077 struct sigqueue *q;
1078 int override_rlimit;
1079 int ret = 0, result;
1080
1081 lockdep_assert_held(&t->sighand->siglock);
1082
1083 result = TRACE_SIGNAL_IGNORED;
1084 if (!prepare_signal(sig, t, force))
1085 goto ret;
1086
1087 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1088 /*
1089 * Short-circuit ignored signals and support queuing
1090 * exactly one non-rt signal, so that we can get more
1091 * detailed information about the cause of the signal.
1092 */
1093 result = TRACE_SIGNAL_ALREADY_PENDING;
1094 if (legacy_queue(pending, sig))
1095 goto ret;
1096
1097 result = TRACE_SIGNAL_DELIVERED;
1098 /*
1099 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1100 */
1101 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1102 goto out_set;
1103
1104 /*
1105 * Real-time signals must be queued if sent by sigqueue, or
1106 * some other real-time mechanism. It is implementation
1107 * defined whether kill() does so. We attempt to do so, on
1108 * the principle of least surprise, but since kill is not
1109 * allowed to fail with EAGAIN when low on memory we just
1110 * make sure at least one signal gets delivered and don't
1111 * pass on the info struct.
1112 */
1113 if (sig < SIGRTMIN)
1114 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1115 else
1116 override_rlimit = 0;
1117
1118 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1119
1120 if (q) {
1121 list_add_tail(&q->list, &pending->list);
1122 switch ((unsigned long) info) {
1123 case (unsigned long) SEND_SIG_NOINFO:
1124 clear_siginfo(&q->info);
1125 q->info.si_signo = sig;
1126 q->info.si_errno = 0;
1127 q->info.si_code = SI_USER;
1128 q->info.si_pid = task_tgid_nr_ns(current,
1129 task_active_pid_ns(t));
1130 rcu_read_lock();
1131 q->info.si_uid =
1132 from_kuid_munged(task_cred_xxx(t, user_ns),
1133 current_uid());
1134 rcu_read_unlock();
1135 break;
1136 case (unsigned long) SEND_SIG_PRIV:
1137 clear_siginfo(&q->info);
1138 q->info.si_signo = sig;
1139 q->info.si_errno = 0;
1140 q->info.si_code = SI_KERNEL;
1141 q->info.si_pid = 0;
1142 q->info.si_uid = 0;
1143 break;
1144 default:
1145 copy_siginfo(&q->info, info);
1146 break;
1147 }
1148 } else if (!is_si_special(info) &&
1149 sig >= SIGRTMIN && info->si_code != SI_USER) {
1150 /*
1151 * Queue overflow, abort. We may abort if the
1152 * signal was rt and sent by user using something
1153 * other than kill().
1154 */
1155 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1156 ret = -EAGAIN;
1157 goto ret;
1158 } else {
1159 /*
1160 * This is a silent loss of information. We still
1161 * send the signal, but the *info bits are lost.
1162 */
1163 result = TRACE_SIGNAL_LOSE_INFO;
1164 }
1165
1166out_set:
1167 signalfd_notify(t, sig);
1168 sigaddset(&pending->signal, sig);
1169
1170 /* Let multiprocess signals appear after on-going forks */
1171 if (type > PIDTYPE_TGID) {
1172 struct multiprocess_signals *delayed;
1173 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1174 sigset_t *signal = &delayed->signal;
1175 /* Can't queue both a stop and a continue signal */
1176 if (sig == SIGCONT)
1177 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1178 else if (sig_kernel_stop(sig))
1179 sigdelset(signal, SIGCONT);
1180 sigaddset(signal, sig);
1181 }
1182 }
1183
1184 complete_signal(sig, t, type);
1185ret:
1186 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1187 return ret;
1188}
1189
1190static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1191{
1192 bool ret = false;
1193 switch (siginfo_layout(info->si_signo, info->si_code)) {
1194 case SIL_KILL:
1195 case SIL_CHLD:
1196 case SIL_RT:
1197 ret = true;
1198 break;
1199 case SIL_TIMER:
1200 case SIL_POLL:
1201 case SIL_FAULT:
1202 case SIL_FAULT_TRAPNO:
1203 case SIL_FAULT_MCEERR:
1204 case SIL_FAULT_BNDERR:
1205 case SIL_FAULT_PKUERR:
1206 case SIL_FAULT_PERF_EVENT:
1207 case SIL_SYS:
1208 ret = false;
1209 break;
1210 }
1211 return ret;
1212}
1213
1214int send_signal_locked(int sig, struct kernel_siginfo *info,
1215 struct task_struct *t, enum pid_type type)
1216{
1217 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1218 bool force = false;
1219
1220 if (info == SEND_SIG_NOINFO) {
1221 /* Force if sent from an ancestor pid namespace */
1222 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1223 } else if (info == SEND_SIG_PRIV) {
1224 /* Don't ignore kernel generated signals */
1225 force = true;
1226 } else if (has_si_pid_and_uid(info)) {
1227 /* SIGKILL and SIGSTOP is special or has ids */
1228 struct user_namespace *t_user_ns;
1229
1230 rcu_read_lock();
1231 t_user_ns = task_cred_xxx(t, user_ns);
1232 if (current_user_ns() != t_user_ns) {
1233 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1234 info->si_uid = from_kuid_munged(t_user_ns, uid);
1235 }
1236 rcu_read_unlock();
1237
1238 /* A kernel generated signal? */
1239 force = (info->si_code == SI_KERNEL);
1240
1241 /* From an ancestor pid namespace? */
1242 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1243 info->si_pid = 0;
1244 force = true;
1245 }
1246 }
1247 return __send_signal_locked(sig, info, t, type, force);
1248}
1249
1250static void print_fatal_signal(int signr)
1251{
1252 struct pt_regs *regs = task_pt_regs(current);
1253 struct file *exe_file;
1254
1255 exe_file = get_task_exe_file(current);
1256 if (exe_file) {
1257 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1258 exe_file, current->comm, signr);
1259 fput(exe_file);
1260 } else {
1261 pr_info("%s: potentially unexpected fatal signal %d.\n",
1262 current->comm, signr);
1263 }
1264
1265#if defined(__i386__) && !defined(__arch_um__)
1266 pr_info("code at %08lx: ", regs->ip);
1267 {
1268 int i;
1269 for (i = 0; i < 16; i++) {
1270 unsigned char insn;
1271
1272 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1273 break;
1274 pr_cont("%02x ", insn);
1275 }
1276 }
1277 pr_cont("\n");
1278#endif
1279 preempt_disable();
1280 show_regs(regs);
1281 preempt_enable();
1282}
1283
1284static int __init setup_print_fatal_signals(char *str)
1285{
1286 get_option (&str, &print_fatal_signals);
1287
1288 return 1;
1289}
1290
1291__setup("print-fatal-signals=", setup_print_fatal_signals);
1292
1293int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1294 enum pid_type type)
1295{
1296 unsigned long flags;
1297 int ret = -ESRCH;
1298
1299 if (lock_task_sighand(p, &flags)) {
1300 ret = send_signal_locked(sig, info, p, type);
1301 unlock_task_sighand(p, &flags);
1302 }
1303
1304 return ret;
1305}
1306
1307enum sig_handler {
1308 HANDLER_CURRENT, /* If reachable use the current handler */
1309 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1310 HANDLER_EXIT, /* Only visible as the process exit code */
1311};
1312
1313/*
1314 * Force a signal that the process can't ignore: if necessary
1315 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1316 *
1317 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1318 * since we do not want to have a signal handler that was blocked
1319 * be invoked when user space had explicitly blocked it.
1320 *
1321 * We don't want to have recursive SIGSEGV's etc, for example,
1322 * that is why we also clear SIGNAL_UNKILLABLE.
1323 */
1324static int
1325force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1326 enum sig_handler handler)
1327{
1328 unsigned long int flags;
1329 int ret, blocked, ignored;
1330 struct k_sigaction *action;
1331 int sig = info->si_signo;
1332
1333 spin_lock_irqsave(&t->sighand->siglock, flags);
1334 action = &t->sighand->action[sig-1];
1335 ignored = action->sa.sa_handler == SIG_IGN;
1336 blocked = sigismember(&t->blocked, sig);
1337 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1338 action->sa.sa_handler = SIG_DFL;
1339 if (handler == HANDLER_EXIT)
1340 action->sa.sa_flags |= SA_IMMUTABLE;
1341 if (blocked)
1342 sigdelset(&t->blocked, sig);
1343 }
1344 /*
1345 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1346 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1347 */
1348 if (action->sa.sa_handler == SIG_DFL &&
1349 (!t->ptrace || (handler == HANDLER_EXIT)))
1350 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1351 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1352 /* This can happen if the signal was already pending and blocked */
1353 if (!task_sigpending(t))
1354 signal_wake_up(t, 0);
1355 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1356
1357 return ret;
1358}
1359
1360int force_sig_info(struct kernel_siginfo *info)
1361{
1362 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1363}
1364
1365/*
1366 * Nuke all other threads in the group.
1367 */
1368int zap_other_threads(struct task_struct *p)
1369{
1370 struct task_struct *t;
1371 int count = 0;
1372
1373 p->signal->group_stop_count = 0;
1374
1375 for_other_threads(p, t) {
1376 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1377 /* Don't require de_thread to wait for the vhost_worker */
1378 if ((t->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)
1379 count++;
1380
1381 /* Don't bother with already dead threads */
1382 if (t->exit_state)
1383 continue;
1384 sigaddset(&t->pending.signal, SIGKILL);
1385 signal_wake_up(t, 1);
1386 }
1387
1388 return count;
1389}
1390
1391struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1392 unsigned long *flags)
1393{
1394 struct sighand_struct *sighand;
1395
1396 rcu_read_lock();
1397 for (;;) {
1398 sighand = rcu_dereference(tsk->sighand);
1399 if (unlikely(sighand == NULL))
1400 break;
1401
1402 /*
1403 * This sighand can be already freed and even reused, but
1404 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1405 * initializes ->siglock: this slab can't go away, it has
1406 * the same object type, ->siglock can't be reinitialized.
1407 *
1408 * We need to ensure that tsk->sighand is still the same
1409 * after we take the lock, we can race with de_thread() or
1410 * __exit_signal(). In the latter case the next iteration
1411 * must see ->sighand == NULL.
1412 */
1413 spin_lock_irqsave(&sighand->siglock, *flags);
1414 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1415 break;
1416 spin_unlock_irqrestore(&sighand->siglock, *flags);
1417 }
1418 rcu_read_unlock();
1419
1420 return sighand;
1421}
1422
1423#ifdef CONFIG_LOCKDEP
1424void lockdep_assert_task_sighand_held(struct task_struct *task)
1425{
1426 struct sighand_struct *sighand;
1427
1428 rcu_read_lock();
1429 sighand = rcu_dereference(task->sighand);
1430 if (sighand)
1431 lockdep_assert_held(&sighand->siglock);
1432 else
1433 WARN_ON_ONCE(1);
1434 rcu_read_unlock();
1435}
1436#endif
1437
1438/*
1439 * send signal info to all the members of a group
1440 */
1441int group_send_sig_info(int sig, struct kernel_siginfo *info,
1442 struct task_struct *p, enum pid_type type)
1443{
1444 int ret;
1445
1446 rcu_read_lock();
1447 ret = check_kill_permission(sig, info, p);
1448 rcu_read_unlock();
1449
1450 if (!ret && sig)
1451 ret = do_send_sig_info(sig, info, p, type);
1452
1453 return ret;
1454}
1455
1456/*
1457 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1458 * control characters do (^C, ^Z etc)
1459 * - the caller must hold at least a readlock on tasklist_lock
1460 */
1461int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1462{
1463 struct task_struct *p = NULL;
1464 int ret = -ESRCH;
1465
1466 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1467 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1468 /*
1469 * If group_send_sig_info() succeeds at least once ret
1470 * becomes 0 and after that the code below has no effect.
1471 * Otherwise we return the last err or -ESRCH if this
1472 * process group is empty.
1473 */
1474 if (ret)
1475 ret = err;
1476 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1477
1478 return ret;
1479}
1480
1481int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1482{
1483 int error = -ESRCH;
1484 struct task_struct *p;
1485
1486 for (;;) {
1487 rcu_read_lock();
1488 p = pid_task(pid, PIDTYPE_PID);
1489 if (p)
1490 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1491 rcu_read_unlock();
1492 if (likely(!p || error != -ESRCH))
1493 return error;
1494
1495 /*
1496 * The task was unhashed in between, try again. If it
1497 * is dead, pid_task() will return NULL, if we race with
1498 * de_thread() it will find the new leader.
1499 */
1500 }
1501}
1502
1503static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1504{
1505 int error;
1506 rcu_read_lock();
1507 error = kill_pid_info(sig, info, find_vpid(pid));
1508 rcu_read_unlock();
1509 return error;
1510}
1511
1512static inline bool kill_as_cred_perm(const struct cred *cred,
1513 struct task_struct *target)
1514{
1515 const struct cred *pcred = __task_cred(target);
1516
1517 return uid_eq(cred->euid, pcred->suid) ||
1518 uid_eq(cred->euid, pcred->uid) ||
1519 uid_eq(cred->uid, pcred->suid) ||
1520 uid_eq(cred->uid, pcred->uid);
1521}
1522
1523/*
1524 * The usb asyncio usage of siginfo is wrong. The glibc support
1525 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1526 * AKA after the generic fields:
1527 * kernel_pid_t si_pid;
1528 * kernel_uid32_t si_uid;
1529 * sigval_t si_value;
1530 *
1531 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1532 * after the generic fields is:
1533 * void __user *si_addr;
1534 *
1535 * This is a practical problem when there is a 64bit big endian kernel
1536 * and a 32bit userspace. As the 32bit address will encoded in the low
1537 * 32bits of the pointer. Those low 32bits will be stored at higher
1538 * address than appear in a 32 bit pointer. So userspace will not
1539 * see the address it was expecting for it's completions.
1540 *
1541 * There is nothing in the encoding that can allow
1542 * copy_siginfo_to_user32 to detect this confusion of formats, so
1543 * handle this by requiring the caller of kill_pid_usb_asyncio to
1544 * notice when this situration takes place and to store the 32bit
1545 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1546 * parameter.
1547 */
1548int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1549 struct pid *pid, const struct cred *cred)
1550{
1551 struct kernel_siginfo info;
1552 struct task_struct *p;
1553 unsigned long flags;
1554 int ret = -EINVAL;
1555
1556 if (!valid_signal(sig))
1557 return ret;
1558
1559 clear_siginfo(&info);
1560 info.si_signo = sig;
1561 info.si_errno = errno;
1562 info.si_code = SI_ASYNCIO;
1563 *((sigval_t *)&info.si_pid) = addr;
1564
1565 rcu_read_lock();
1566 p = pid_task(pid, PIDTYPE_PID);
1567 if (!p) {
1568 ret = -ESRCH;
1569 goto out_unlock;
1570 }
1571 if (!kill_as_cred_perm(cred, p)) {
1572 ret = -EPERM;
1573 goto out_unlock;
1574 }
1575 ret = security_task_kill(p, &info, sig, cred);
1576 if (ret)
1577 goto out_unlock;
1578
1579 if (sig) {
1580 if (lock_task_sighand(p, &flags)) {
1581 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1582 unlock_task_sighand(p, &flags);
1583 } else
1584 ret = -ESRCH;
1585 }
1586out_unlock:
1587 rcu_read_unlock();
1588 return ret;
1589}
1590EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1591
1592/*
1593 * kill_something_info() interprets pid in interesting ways just like kill(2).
1594 *
1595 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1596 * is probably wrong. Should make it like BSD or SYSV.
1597 */
1598
1599static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1600{
1601 int ret;
1602
1603 if (pid > 0)
1604 return kill_proc_info(sig, info, pid);
1605
1606 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1607 if (pid == INT_MIN)
1608 return -ESRCH;
1609
1610 read_lock(&tasklist_lock);
1611 if (pid != -1) {
1612 ret = __kill_pgrp_info(sig, info,
1613 pid ? find_vpid(-pid) : task_pgrp(current));
1614 } else {
1615 int retval = 0, count = 0;
1616 struct task_struct * p;
1617
1618 for_each_process(p) {
1619 if (task_pid_vnr(p) > 1 &&
1620 !same_thread_group(p, current)) {
1621 int err = group_send_sig_info(sig, info, p,
1622 PIDTYPE_MAX);
1623 ++count;
1624 if (err != -EPERM)
1625 retval = err;
1626 }
1627 }
1628 ret = count ? retval : -ESRCH;
1629 }
1630 read_unlock(&tasklist_lock);
1631
1632 return ret;
1633}
1634
1635/*
1636 * These are for backward compatibility with the rest of the kernel source.
1637 */
1638
1639int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1640{
1641 /*
1642 * Make sure legacy kernel users don't send in bad values
1643 * (normal paths check this in check_kill_permission).
1644 */
1645 if (!valid_signal(sig))
1646 return -EINVAL;
1647
1648 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1649}
1650EXPORT_SYMBOL(send_sig_info);
1651
1652#define __si_special(priv) \
1653 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1654
1655int
1656send_sig(int sig, struct task_struct *p, int priv)
1657{
1658 return send_sig_info(sig, __si_special(priv), p);
1659}
1660EXPORT_SYMBOL(send_sig);
1661
1662void force_sig(int sig)
1663{
1664 struct kernel_siginfo info;
1665
1666 clear_siginfo(&info);
1667 info.si_signo = sig;
1668 info.si_errno = 0;
1669 info.si_code = SI_KERNEL;
1670 info.si_pid = 0;
1671 info.si_uid = 0;
1672 force_sig_info(&info);
1673}
1674EXPORT_SYMBOL(force_sig);
1675
1676void force_fatal_sig(int sig)
1677{
1678 struct kernel_siginfo info;
1679
1680 clear_siginfo(&info);
1681 info.si_signo = sig;
1682 info.si_errno = 0;
1683 info.si_code = SI_KERNEL;
1684 info.si_pid = 0;
1685 info.si_uid = 0;
1686 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1687}
1688
1689void force_exit_sig(int sig)
1690{
1691 struct kernel_siginfo info;
1692
1693 clear_siginfo(&info);
1694 info.si_signo = sig;
1695 info.si_errno = 0;
1696 info.si_code = SI_KERNEL;
1697 info.si_pid = 0;
1698 info.si_uid = 0;
1699 force_sig_info_to_task(&info, current, HANDLER_EXIT);
1700}
1701
1702/*
1703 * When things go south during signal handling, we
1704 * will force a SIGSEGV. And if the signal that caused
1705 * the problem was already a SIGSEGV, we'll want to
1706 * make sure we don't even try to deliver the signal..
1707 */
1708void force_sigsegv(int sig)
1709{
1710 if (sig == SIGSEGV)
1711 force_fatal_sig(SIGSEGV);
1712 else
1713 force_sig(SIGSEGV);
1714}
1715
1716int force_sig_fault_to_task(int sig, int code, void __user *addr,
1717 struct task_struct *t)
1718{
1719 struct kernel_siginfo info;
1720
1721 clear_siginfo(&info);
1722 info.si_signo = sig;
1723 info.si_errno = 0;
1724 info.si_code = code;
1725 info.si_addr = addr;
1726 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1727}
1728
1729int force_sig_fault(int sig, int code, void __user *addr)
1730{
1731 return force_sig_fault_to_task(sig, code, addr, current);
1732}
1733
1734int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t)
1735{
1736 struct kernel_siginfo info;
1737
1738 clear_siginfo(&info);
1739 info.si_signo = sig;
1740 info.si_errno = 0;
1741 info.si_code = code;
1742 info.si_addr = addr;
1743 return send_sig_info(info.si_signo, &info, t);
1744}
1745
1746int force_sig_mceerr(int code, void __user *addr, short lsb)
1747{
1748 struct kernel_siginfo info;
1749
1750 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1751 clear_siginfo(&info);
1752 info.si_signo = SIGBUS;
1753 info.si_errno = 0;
1754 info.si_code = code;
1755 info.si_addr = addr;
1756 info.si_addr_lsb = lsb;
1757 return force_sig_info(&info);
1758}
1759
1760int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1761{
1762 struct kernel_siginfo info;
1763
1764 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1765 clear_siginfo(&info);
1766 info.si_signo = SIGBUS;
1767 info.si_errno = 0;
1768 info.si_code = code;
1769 info.si_addr = addr;
1770 info.si_addr_lsb = lsb;
1771 return send_sig_info(info.si_signo, &info, t);
1772}
1773EXPORT_SYMBOL(send_sig_mceerr);
1774
1775int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1776{
1777 struct kernel_siginfo info;
1778
1779 clear_siginfo(&info);
1780 info.si_signo = SIGSEGV;
1781 info.si_errno = 0;
1782 info.si_code = SEGV_BNDERR;
1783 info.si_addr = addr;
1784 info.si_lower = lower;
1785 info.si_upper = upper;
1786 return force_sig_info(&info);
1787}
1788
1789#ifdef SEGV_PKUERR
1790int force_sig_pkuerr(void __user *addr, u32 pkey)
1791{
1792 struct kernel_siginfo info;
1793
1794 clear_siginfo(&info);
1795 info.si_signo = SIGSEGV;
1796 info.si_errno = 0;
1797 info.si_code = SEGV_PKUERR;
1798 info.si_addr = addr;
1799 info.si_pkey = pkey;
1800 return force_sig_info(&info);
1801}
1802#endif
1803
1804int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1805{
1806 struct kernel_siginfo info;
1807
1808 clear_siginfo(&info);
1809 info.si_signo = SIGTRAP;
1810 info.si_errno = 0;
1811 info.si_code = TRAP_PERF;
1812 info.si_addr = addr;
1813 info.si_perf_data = sig_data;
1814 info.si_perf_type = type;
1815
1816 /*
1817 * Signals generated by perf events should not terminate the whole
1818 * process if SIGTRAP is blocked, however, delivering the signal
1819 * asynchronously is better than not delivering at all. But tell user
1820 * space if the signal was asynchronous, so it can clearly be
1821 * distinguished from normal synchronous ones.
1822 */
1823 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1824 TRAP_PERF_FLAG_ASYNC :
1825 0;
1826
1827 return send_sig_info(info.si_signo, &info, current);
1828}
1829
1830/**
1831 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1832 * @syscall: syscall number to send to userland
1833 * @reason: filter-supplied reason code to send to userland (via si_errno)
1834 * @force_coredump: true to trigger a coredump
1835 *
1836 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1837 */
1838int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1839{
1840 struct kernel_siginfo info;
1841
1842 clear_siginfo(&info);
1843 info.si_signo = SIGSYS;
1844 info.si_code = SYS_SECCOMP;
1845 info.si_call_addr = (void __user *)KSTK_EIP(current);
1846 info.si_errno = reason;
1847 info.si_arch = syscall_get_arch(current);
1848 info.si_syscall = syscall;
1849 return force_sig_info_to_task(&info, current,
1850 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1851}
1852
1853/* For the crazy architectures that include trap information in
1854 * the errno field, instead of an actual errno value.
1855 */
1856int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1857{
1858 struct kernel_siginfo info;
1859
1860 clear_siginfo(&info);
1861 info.si_signo = SIGTRAP;
1862 info.si_errno = errno;
1863 info.si_code = TRAP_HWBKPT;
1864 info.si_addr = addr;
1865 return force_sig_info(&info);
1866}
1867
1868/* For the rare architectures that include trap information using
1869 * si_trapno.
1870 */
1871int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1872{
1873 struct kernel_siginfo info;
1874
1875 clear_siginfo(&info);
1876 info.si_signo = sig;
1877 info.si_errno = 0;
1878 info.si_code = code;
1879 info.si_addr = addr;
1880 info.si_trapno = trapno;
1881 return force_sig_info(&info);
1882}
1883
1884/* For the rare architectures that include trap information using
1885 * si_trapno.
1886 */
1887int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1888 struct task_struct *t)
1889{
1890 struct kernel_siginfo info;
1891
1892 clear_siginfo(&info);
1893 info.si_signo = sig;
1894 info.si_errno = 0;
1895 info.si_code = code;
1896 info.si_addr = addr;
1897 info.si_trapno = trapno;
1898 return send_sig_info(info.si_signo, &info, t);
1899}
1900
1901int kill_pgrp(struct pid *pid, int sig, int priv)
1902{
1903 int ret;
1904
1905 read_lock(&tasklist_lock);
1906 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1907 read_unlock(&tasklist_lock);
1908
1909 return ret;
1910}
1911EXPORT_SYMBOL(kill_pgrp);
1912
1913int kill_pid(struct pid *pid, int sig, int priv)
1914{
1915 return kill_pid_info(sig, __si_special(priv), pid);
1916}
1917EXPORT_SYMBOL(kill_pid);
1918
1919/*
1920 * These functions support sending signals using preallocated sigqueue
1921 * structures. This is needed "because realtime applications cannot
1922 * afford to lose notifications of asynchronous events, like timer
1923 * expirations or I/O completions". In the case of POSIX Timers
1924 * we allocate the sigqueue structure from the timer_create. If this
1925 * allocation fails we are able to report the failure to the application
1926 * with an EAGAIN error.
1927 */
1928struct sigqueue *sigqueue_alloc(void)
1929{
1930 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1931}
1932
1933void sigqueue_free(struct sigqueue *q)
1934{
1935 unsigned long flags;
1936 spinlock_t *lock = ¤t->sighand->siglock;
1937
1938 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1939 /*
1940 * We must hold ->siglock while testing q->list
1941 * to serialize with collect_signal() or with
1942 * __exit_signal()->flush_sigqueue().
1943 */
1944 spin_lock_irqsave(lock, flags);
1945 q->flags &= ~SIGQUEUE_PREALLOC;
1946 /*
1947 * If it is queued it will be freed when dequeued,
1948 * like the "regular" sigqueue.
1949 */
1950 if (!list_empty(&q->list))
1951 q = NULL;
1952 spin_unlock_irqrestore(lock, flags);
1953
1954 if (q)
1955 __sigqueue_free(q);
1956}
1957
1958int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1959{
1960 int sig = q->info.si_signo;
1961 struct sigpending *pending;
1962 struct task_struct *t;
1963 unsigned long flags;
1964 int ret, result;
1965
1966 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1967
1968 ret = -1;
1969 rcu_read_lock();
1970
1971 /*
1972 * This function is used by POSIX timers to deliver a timer signal.
1973 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1974 * set), the signal must be delivered to the specific thread (queues
1975 * into t->pending).
1976 *
1977 * Where type is not PIDTYPE_PID, signals must be delivered to the
1978 * process. In this case, prefer to deliver to current if it is in
1979 * the same thread group as the target process, which avoids
1980 * unnecessarily waking up a potentially idle task.
1981 */
1982 t = pid_task(pid, type);
1983 if (!t)
1984 goto ret;
1985 if (type != PIDTYPE_PID && same_thread_group(t, current))
1986 t = current;
1987 if (!likely(lock_task_sighand(t, &flags)))
1988 goto ret;
1989
1990 ret = 1; /* the signal is ignored */
1991 result = TRACE_SIGNAL_IGNORED;
1992 if (!prepare_signal(sig, t, false))
1993 goto out;
1994
1995 ret = 0;
1996 if (unlikely(!list_empty(&q->list))) {
1997 /*
1998 * If an SI_TIMER entry is already queue just increment
1999 * the overrun count.
2000 */
2001 BUG_ON(q->info.si_code != SI_TIMER);
2002 q->info.si_overrun++;
2003 result = TRACE_SIGNAL_ALREADY_PENDING;
2004 goto out;
2005 }
2006 q->info.si_overrun = 0;
2007
2008 signalfd_notify(t, sig);
2009 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
2010 list_add_tail(&q->list, &pending->list);
2011 sigaddset(&pending->signal, sig);
2012 complete_signal(sig, t, type);
2013 result = TRACE_SIGNAL_DELIVERED;
2014out:
2015 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
2016 unlock_task_sighand(t, &flags);
2017ret:
2018 rcu_read_unlock();
2019 return ret;
2020}
2021
2022static void do_notify_pidfd(struct task_struct *task)
2023{
2024 struct pid *pid;
2025
2026 WARN_ON(task->exit_state == 0);
2027 pid = task_pid(task);
2028 wake_up_all(&pid->wait_pidfd);
2029}
2030
2031/*
2032 * Let a parent know about the death of a child.
2033 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2034 *
2035 * Returns true if our parent ignored us and so we've switched to
2036 * self-reaping.
2037 */
2038bool do_notify_parent(struct task_struct *tsk, int sig)
2039{
2040 struct kernel_siginfo info;
2041 unsigned long flags;
2042 struct sighand_struct *psig;
2043 bool autoreap = false;
2044 u64 utime, stime;
2045
2046 WARN_ON_ONCE(sig == -1);
2047
2048 /* do_notify_parent_cldstop should have been called instead. */
2049 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2050
2051 WARN_ON_ONCE(!tsk->ptrace &&
2052 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2053
2054 /* Wake up all pidfd waiters */
2055 do_notify_pidfd(tsk);
2056
2057 if (sig != SIGCHLD) {
2058 /*
2059 * This is only possible if parent == real_parent.
2060 * Check if it has changed security domain.
2061 */
2062 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2063 sig = SIGCHLD;
2064 }
2065
2066 clear_siginfo(&info);
2067 info.si_signo = sig;
2068 info.si_errno = 0;
2069 /*
2070 * We are under tasklist_lock here so our parent is tied to
2071 * us and cannot change.
2072 *
2073 * task_active_pid_ns will always return the same pid namespace
2074 * until a task passes through release_task.
2075 *
2076 * write_lock() currently calls preempt_disable() which is the
2077 * same as rcu_read_lock(), but according to Oleg, this is not
2078 * correct to rely on this
2079 */
2080 rcu_read_lock();
2081 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2082 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2083 task_uid(tsk));
2084 rcu_read_unlock();
2085
2086 task_cputime(tsk, &utime, &stime);
2087 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2088 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2089
2090 info.si_status = tsk->exit_code & 0x7f;
2091 if (tsk->exit_code & 0x80)
2092 info.si_code = CLD_DUMPED;
2093 else if (tsk->exit_code & 0x7f)
2094 info.si_code = CLD_KILLED;
2095 else {
2096 info.si_code = CLD_EXITED;
2097 info.si_status = tsk->exit_code >> 8;
2098 }
2099
2100 psig = tsk->parent->sighand;
2101 spin_lock_irqsave(&psig->siglock, flags);
2102 if (!tsk->ptrace && sig == SIGCHLD &&
2103 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2104 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2105 /*
2106 * We are exiting and our parent doesn't care. POSIX.1
2107 * defines special semantics for setting SIGCHLD to SIG_IGN
2108 * or setting the SA_NOCLDWAIT flag: we should be reaped
2109 * automatically and not left for our parent's wait4 call.
2110 * Rather than having the parent do it as a magic kind of
2111 * signal handler, we just set this to tell do_exit that we
2112 * can be cleaned up without becoming a zombie. Note that
2113 * we still call __wake_up_parent in this case, because a
2114 * blocked sys_wait4 might now return -ECHILD.
2115 *
2116 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2117 * is implementation-defined: we do (if you don't want
2118 * it, just use SIG_IGN instead).
2119 */
2120 autoreap = true;
2121 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2122 sig = 0;
2123 }
2124 /*
2125 * Send with __send_signal as si_pid and si_uid are in the
2126 * parent's namespaces.
2127 */
2128 if (valid_signal(sig) && sig)
2129 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2130 __wake_up_parent(tsk, tsk->parent);
2131 spin_unlock_irqrestore(&psig->siglock, flags);
2132
2133 return autoreap;
2134}
2135
2136/**
2137 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2138 * @tsk: task reporting the state change
2139 * @for_ptracer: the notification is for ptracer
2140 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2141 *
2142 * Notify @tsk's parent that the stopped/continued state has changed. If
2143 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2144 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2145 *
2146 * CONTEXT:
2147 * Must be called with tasklist_lock at least read locked.
2148 */
2149static void do_notify_parent_cldstop(struct task_struct *tsk,
2150 bool for_ptracer, int why)
2151{
2152 struct kernel_siginfo info;
2153 unsigned long flags;
2154 struct task_struct *parent;
2155 struct sighand_struct *sighand;
2156 u64 utime, stime;
2157
2158 if (for_ptracer) {
2159 parent = tsk->parent;
2160 } else {
2161 tsk = tsk->group_leader;
2162 parent = tsk->real_parent;
2163 }
2164
2165 clear_siginfo(&info);
2166 info.si_signo = SIGCHLD;
2167 info.si_errno = 0;
2168 /*
2169 * see comment in do_notify_parent() about the following 4 lines
2170 */
2171 rcu_read_lock();
2172 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2173 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2174 rcu_read_unlock();
2175
2176 task_cputime(tsk, &utime, &stime);
2177 info.si_utime = nsec_to_clock_t(utime);
2178 info.si_stime = nsec_to_clock_t(stime);
2179
2180 info.si_code = why;
2181 switch (why) {
2182 case CLD_CONTINUED:
2183 info.si_status = SIGCONT;
2184 break;
2185 case CLD_STOPPED:
2186 info.si_status = tsk->signal->group_exit_code & 0x7f;
2187 break;
2188 case CLD_TRAPPED:
2189 info.si_status = tsk->exit_code & 0x7f;
2190 break;
2191 default:
2192 BUG();
2193 }
2194
2195 sighand = parent->sighand;
2196 spin_lock_irqsave(&sighand->siglock, flags);
2197 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2198 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2199 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2200 /*
2201 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2202 */
2203 __wake_up_parent(tsk, parent);
2204 spin_unlock_irqrestore(&sighand->siglock, flags);
2205}
2206
2207/*
2208 * This must be called with current->sighand->siglock held.
2209 *
2210 * This should be the path for all ptrace stops.
2211 * We always set current->last_siginfo while stopped here.
2212 * That makes it a way to test a stopped process for
2213 * being ptrace-stopped vs being job-control-stopped.
2214 *
2215 * Returns the signal the ptracer requested the code resume
2216 * with. If the code did not stop because the tracer is gone,
2217 * the stop signal remains unchanged unless clear_code.
2218 */
2219static int ptrace_stop(int exit_code, int why, unsigned long message,
2220 kernel_siginfo_t *info)
2221 __releases(¤t->sighand->siglock)
2222 __acquires(¤t->sighand->siglock)
2223{
2224 bool gstop_done = false;
2225
2226 if (arch_ptrace_stop_needed()) {
2227 /*
2228 * The arch code has something special to do before a
2229 * ptrace stop. This is allowed to block, e.g. for faults
2230 * on user stack pages. We can't keep the siglock while
2231 * calling arch_ptrace_stop, so we must release it now.
2232 * To preserve proper semantics, we must do this before
2233 * any signal bookkeeping like checking group_stop_count.
2234 */
2235 spin_unlock_irq(¤t->sighand->siglock);
2236 arch_ptrace_stop();
2237 spin_lock_irq(¤t->sighand->siglock);
2238 }
2239
2240 /*
2241 * After this point ptrace_signal_wake_up or signal_wake_up
2242 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2243 * signal comes in. Handle previous ptrace_unlinks and fatal
2244 * signals here to prevent ptrace_stop sleeping in schedule.
2245 */
2246 if (!current->ptrace || __fatal_signal_pending(current))
2247 return exit_code;
2248
2249 set_special_state(TASK_TRACED);
2250 current->jobctl |= JOBCTL_TRACED;
2251
2252 /*
2253 * We're committing to trapping. TRACED should be visible before
2254 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2255 * Also, transition to TRACED and updates to ->jobctl should be
2256 * atomic with respect to siglock and should be done after the arch
2257 * hook as siglock is released and regrabbed across it.
2258 *
2259 * TRACER TRACEE
2260 *
2261 * ptrace_attach()
2262 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2263 * do_wait()
2264 * set_current_state() smp_wmb();
2265 * ptrace_do_wait()
2266 * wait_task_stopped()
2267 * task_stopped_code()
2268 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2269 */
2270 smp_wmb();
2271
2272 current->ptrace_message = message;
2273 current->last_siginfo = info;
2274 current->exit_code = exit_code;
2275
2276 /*
2277 * If @why is CLD_STOPPED, we're trapping to participate in a group
2278 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2279 * across siglock relocks since INTERRUPT was scheduled, PENDING
2280 * could be clear now. We act as if SIGCONT is received after
2281 * TASK_TRACED is entered - ignore it.
2282 */
2283 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2284 gstop_done = task_participate_group_stop(current);
2285
2286 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2287 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2288 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2289 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2290
2291 /* entering a trap, clear TRAPPING */
2292 task_clear_jobctl_trapping(current);
2293
2294 spin_unlock_irq(¤t->sighand->siglock);
2295 read_lock(&tasklist_lock);
2296 /*
2297 * Notify parents of the stop.
2298 *
2299 * While ptraced, there are two parents - the ptracer and
2300 * the real_parent of the group_leader. The ptracer should
2301 * know about every stop while the real parent is only
2302 * interested in the completion of group stop. The states
2303 * for the two don't interact with each other. Notify
2304 * separately unless they're gonna be duplicates.
2305 */
2306 if (current->ptrace)
2307 do_notify_parent_cldstop(current, true, why);
2308 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2309 do_notify_parent_cldstop(current, false, why);
2310
2311 /*
2312 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2313 * One a PREEMPTION kernel this can result in preemption requirement
2314 * which will be fulfilled after read_unlock() and the ptracer will be
2315 * put on the CPU.
2316 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2317 * this task wait in schedule(). If this task gets preempted then it
2318 * remains enqueued on the runqueue. The ptracer will observe this and
2319 * then sleep for a delay of one HZ tick. In the meantime this task
2320 * gets scheduled, enters schedule() and will wait for the ptracer.
2321 *
2322 * This preemption point is not bad from a correctness point of
2323 * view but extends the runtime by one HZ tick time due to the
2324 * ptracer's sleep. The preempt-disable section ensures that there
2325 * will be no preemption between unlock and schedule() and so
2326 * improving the performance since the ptracer will observe that
2327 * the tracee is scheduled out once it gets on the CPU.
2328 *
2329 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2330 * Therefore the task can be preempted after do_notify_parent_cldstop()
2331 * before unlocking tasklist_lock so there is no benefit in doing this.
2332 *
2333 * In fact disabling preemption is harmful on PREEMPT_RT because
2334 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2335 * with preemption disabled due to the 'sleeping' spinlock
2336 * substitution of RT.
2337 */
2338 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2339 preempt_disable();
2340 read_unlock(&tasklist_lock);
2341 cgroup_enter_frozen();
2342 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2343 preempt_enable_no_resched();
2344 schedule();
2345 cgroup_leave_frozen(true);
2346
2347 /*
2348 * We are back. Now reacquire the siglock before touching
2349 * last_siginfo, so that we are sure to have synchronized with
2350 * any signal-sending on another CPU that wants to examine it.
2351 */
2352 spin_lock_irq(¤t->sighand->siglock);
2353 exit_code = current->exit_code;
2354 current->last_siginfo = NULL;
2355 current->ptrace_message = 0;
2356 current->exit_code = 0;
2357
2358 /* LISTENING can be set only during STOP traps, clear it */
2359 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2360
2361 /*
2362 * Queued signals ignored us while we were stopped for tracing.
2363 * So check for any that we should take before resuming user mode.
2364 * This sets TIF_SIGPENDING, but never clears it.
2365 */
2366 recalc_sigpending_tsk(current);
2367 return exit_code;
2368}
2369
2370static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2371{
2372 kernel_siginfo_t info;
2373
2374 clear_siginfo(&info);
2375 info.si_signo = signr;
2376 info.si_code = exit_code;
2377 info.si_pid = task_pid_vnr(current);
2378 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2379
2380 /* Let the debugger run. */
2381 return ptrace_stop(exit_code, why, message, &info);
2382}
2383
2384int ptrace_notify(int exit_code, unsigned long message)
2385{
2386 int signr;
2387
2388 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2389 if (unlikely(task_work_pending(current)))
2390 task_work_run();
2391
2392 spin_lock_irq(¤t->sighand->siglock);
2393 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2394 spin_unlock_irq(¤t->sighand->siglock);
2395 return signr;
2396}
2397
2398/**
2399 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2400 * @signr: signr causing group stop if initiating
2401 *
2402 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2403 * and participate in it. If already set, participate in the existing
2404 * group stop. If participated in a group stop (and thus slept), %true is
2405 * returned with siglock released.
2406 *
2407 * If ptraced, this function doesn't handle stop itself. Instead,
2408 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2409 * untouched. The caller must ensure that INTERRUPT trap handling takes
2410 * places afterwards.
2411 *
2412 * CONTEXT:
2413 * Must be called with @current->sighand->siglock held, which is released
2414 * on %true return.
2415 *
2416 * RETURNS:
2417 * %false if group stop is already cancelled or ptrace trap is scheduled.
2418 * %true if participated in group stop.
2419 */
2420static bool do_signal_stop(int signr)
2421 __releases(¤t->sighand->siglock)
2422{
2423 struct signal_struct *sig = current->signal;
2424
2425 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2426 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2427 struct task_struct *t;
2428
2429 /* signr will be recorded in task->jobctl for retries */
2430 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2431
2432 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2433 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2434 unlikely(sig->group_exec_task))
2435 return false;
2436 /*
2437 * There is no group stop already in progress. We must
2438 * initiate one now.
2439 *
2440 * While ptraced, a task may be resumed while group stop is
2441 * still in effect and then receive a stop signal and
2442 * initiate another group stop. This deviates from the
2443 * usual behavior as two consecutive stop signals can't
2444 * cause two group stops when !ptraced. That is why we
2445 * also check !task_is_stopped(t) below.
2446 *
2447 * The condition can be distinguished by testing whether
2448 * SIGNAL_STOP_STOPPED is already set. Don't generate
2449 * group_exit_code in such case.
2450 *
2451 * This is not necessary for SIGNAL_STOP_CONTINUED because
2452 * an intervening stop signal is required to cause two
2453 * continued events regardless of ptrace.
2454 */
2455 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2456 sig->group_exit_code = signr;
2457
2458 sig->group_stop_count = 0;
2459 if (task_set_jobctl_pending(current, signr | gstop))
2460 sig->group_stop_count++;
2461
2462 for_other_threads(current, t) {
2463 /*
2464 * Setting state to TASK_STOPPED for a group
2465 * stop is always done with the siglock held,
2466 * so this check has no races.
2467 */
2468 if (!task_is_stopped(t) &&
2469 task_set_jobctl_pending(t, signr | gstop)) {
2470 sig->group_stop_count++;
2471 if (likely(!(t->ptrace & PT_SEIZED)))
2472 signal_wake_up(t, 0);
2473 else
2474 ptrace_trap_notify(t);
2475 }
2476 }
2477 }
2478
2479 if (likely(!current->ptrace)) {
2480 int notify = 0;
2481
2482 /*
2483 * If there are no other threads in the group, or if there
2484 * is a group stop in progress and we are the last to stop,
2485 * report to the parent.
2486 */
2487 if (task_participate_group_stop(current))
2488 notify = CLD_STOPPED;
2489
2490 current->jobctl |= JOBCTL_STOPPED;
2491 set_special_state(TASK_STOPPED);
2492 spin_unlock_irq(¤t->sighand->siglock);
2493
2494 /*
2495 * Notify the parent of the group stop completion. Because
2496 * we're not holding either the siglock or tasklist_lock
2497 * here, ptracer may attach inbetween; however, this is for
2498 * group stop and should always be delivered to the real
2499 * parent of the group leader. The new ptracer will get
2500 * its notification when this task transitions into
2501 * TASK_TRACED.
2502 */
2503 if (notify) {
2504 read_lock(&tasklist_lock);
2505 do_notify_parent_cldstop(current, false, notify);
2506 read_unlock(&tasklist_lock);
2507 }
2508
2509 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2510 cgroup_enter_frozen();
2511 schedule();
2512 return true;
2513 } else {
2514 /*
2515 * While ptraced, group stop is handled by STOP trap.
2516 * Schedule it and let the caller deal with it.
2517 */
2518 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2519 return false;
2520 }
2521}
2522
2523/**
2524 * do_jobctl_trap - take care of ptrace jobctl traps
2525 *
2526 * When PT_SEIZED, it's used for both group stop and explicit
2527 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2528 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2529 * the stop signal; otherwise, %SIGTRAP.
2530 *
2531 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2532 * number as exit_code and no siginfo.
2533 *
2534 * CONTEXT:
2535 * Must be called with @current->sighand->siglock held, which may be
2536 * released and re-acquired before returning with intervening sleep.
2537 */
2538static void do_jobctl_trap(void)
2539{
2540 struct signal_struct *signal = current->signal;
2541 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2542
2543 if (current->ptrace & PT_SEIZED) {
2544 if (!signal->group_stop_count &&
2545 !(signal->flags & SIGNAL_STOP_STOPPED))
2546 signr = SIGTRAP;
2547 WARN_ON_ONCE(!signr);
2548 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2549 CLD_STOPPED, 0);
2550 } else {
2551 WARN_ON_ONCE(!signr);
2552 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2553 }
2554}
2555
2556/**
2557 * do_freezer_trap - handle the freezer jobctl trap
2558 *
2559 * Puts the task into frozen state, if only the task is not about to quit.
2560 * In this case it drops JOBCTL_TRAP_FREEZE.
2561 *
2562 * CONTEXT:
2563 * Must be called with @current->sighand->siglock held,
2564 * which is always released before returning.
2565 */
2566static void do_freezer_trap(void)
2567 __releases(¤t->sighand->siglock)
2568{
2569 /*
2570 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2571 * let's make another loop to give it a chance to be handled.
2572 * In any case, we'll return back.
2573 */
2574 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2575 JOBCTL_TRAP_FREEZE) {
2576 spin_unlock_irq(¤t->sighand->siglock);
2577 return;
2578 }
2579
2580 /*
2581 * Now we're sure that there is no pending fatal signal and no
2582 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2583 * immediately (if there is a non-fatal signal pending), and
2584 * put the task into sleep.
2585 */
2586 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2587 clear_thread_flag(TIF_SIGPENDING);
2588 spin_unlock_irq(¤t->sighand->siglock);
2589 cgroup_enter_frozen();
2590 schedule();
2591}
2592
2593static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2594{
2595 /*
2596 * We do not check sig_kernel_stop(signr) but set this marker
2597 * unconditionally because we do not know whether debugger will
2598 * change signr. This flag has no meaning unless we are going
2599 * to stop after return from ptrace_stop(). In this case it will
2600 * be checked in do_signal_stop(), we should only stop if it was
2601 * not cleared by SIGCONT while we were sleeping. See also the
2602 * comment in dequeue_signal().
2603 */
2604 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2605 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2606
2607 /* We're back. Did the debugger cancel the sig? */
2608 if (signr == 0)
2609 return signr;
2610
2611 /*
2612 * Update the siginfo structure if the signal has
2613 * changed. If the debugger wanted something
2614 * specific in the siginfo structure then it should
2615 * have updated *info via PTRACE_SETSIGINFO.
2616 */
2617 if (signr != info->si_signo) {
2618 clear_siginfo(info);
2619 info->si_signo = signr;
2620 info->si_errno = 0;
2621 info->si_code = SI_USER;
2622 rcu_read_lock();
2623 info->si_pid = task_pid_vnr(current->parent);
2624 info->si_uid = from_kuid_munged(current_user_ns(),
2625 task_uid(current->parent));
2626 rcu_read_unlock();
2627 }
2628
2629 /* If the (new) signal is now blocked, requeue it. */
2630 if (sigismember(¤t->blocked, signr) ||
2631 fatal_signal_pending(current)) {
2632 send_signal_locked(signr, info, current, type);
2633 signr = 0;
2634 }
2635
2636 return signr;
2637}
2638
2639static void hide_si_addr_tag_bits(struct ksignal *ksig)
2640{
2641 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2642 case SIL_FAULT:
2643 case SIL_FAULT_TRAPNO:
2644 case SIL_FAULT_MCEERR:
2645 case SIL_FAULT_BNDERR:
2646 case SIL_FAULT_PKUERR:
2647 case SIL_FAULT_PERF_EVENT:
2648 ksig->info.si_addr = arch_untagged_si_addr(
2649 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2650 break;
2651 case SIL_KILL:
2652 case SIL_TIMER:
2653 case SIL_POLL:
2654 case SIL_CHLD:
2655 case SIL_RT:
2656 case SIL_SYS:
2657 break;
2658 }
2659}
2660
2661bool get_signal(struct ksignal *ksig)
2662{
2663 struct sighand_struct *sighand = current->sighand;
2664 struct signal_struct *signal = current->signal;
2665 int signr;
2666
2667 clear_notify_signal();
2668 if (unlikely(task_work_pending(current)))
2669 task_work_run();
2670
2671 if (!task_sigpending(current))
2672 return false;
2673
2674 if (unlikely(uprobe_deny_signal()))
2675 return false;
2676
2677 /*
2678 * Do this once, we can't return to user-mode if freezing() == T.
2679 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2680 * thus do not need another check after return.
2681 */
2682 try_to_freeze();
2683
2684relock:
2685 spin_lock_irq(&sighand->siglock);
2686
2687 /*
2688 * Every stopped thread goes here after wakeup. Check to see if
2689 * we should notify the parent, prepare_signal(SIGCONT) encodes
2690 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2691 */
2692 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2693 int why;
2694
2695 if (signal->flags & SIGNAL_CLD_CONTINUED)
2696 why = CLD_CONTINUED;
2697 else
2698 why = CLD_STOPPED;
2699
2700 signal->flags &= ~SIGNAL_CLD_MASK;
2701
2702 spin_unlock_irq(&sighand->siglock);
2703
2704 /*
2705 * Notify the parent that we're continuing. This event is
2706 * always per-process and doesn't make whole lot of sense
2707 * for ptracers, who shouldn't consume the state via
2708 * wait(2) either, but, for backward compatibility, notify
2709 * the ptracer of the group leader too unless it's gonna be
2710 * a duplicate.
2711 */
2712 read_lock(&tasklist_lock);
2713 do_notify_parent_cldstop(current, false, why);
2714
2715 if (ptrace_reparented(current->group_leader))
2716 do_notify_parent_cldstop(current->group_leader,
2717 true, why);
2718 read_unlock(&tasklist_lock);
2719
2720 goto relock;
2721 }
2722
2723 for (;;) {
2724 struct k_sigaction *ka;
2725 enum pid_type type;
2726
2727 /* Has this task already been marked for death? */
2728 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2729 signal->group_exec_task) {
2730 clear_siginfo(&ksig->info);
2731 ksig->info.si_signo = signr = SIGKILL;
2732 sigdelset(¤t->pending.signal, SIGKILL);
2733 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2734 &sighand->action[SIGKILL - 1]);
2735 recalc_sigpending();
2736 goto fatal;
2737 }
2738
2739 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2740 do_signal_stop(0))
2741 goto relock;
2742
2743 if (unlikely(current->jobctl &
2744 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2745 if (current->jobctl & JOBCTL_TRAP_MASK) {
2746 do_jobctl_trap();
2747 spin_unlock_irq(&sighand->siglock);
2748 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2749 do_freezer_trap();
2750
2751 goto relock;
2752 }
2753
2754 /*
2755 * If the task is leaving the frozen state, let's update
2756 * cgroup counters and reset the frozen bit.
2757 */
2758 if (unlikely(cgroup_task_frozen(current))) {
2759 spin_unlock_irq(&sighand->siglock);
2760 cgroup_leave_frozen(false);
2761 goto relock;
2762 }
2763
2764 /*
2765 * Signals generated by the execution of an instruction
2766 * need to be delivered before any other pending signals
2767 * so that the instruction pointer in the signal stack
2768 * frame points to the faulting instruction.
2769 */
2770 type = PIDTYPE_PID;
2771 signr = dequeue_synchronous_signal(&ksig->info);
2772 if (!signr)
2773 signr = dequeue_signal(current, ¤t->blocked,
2774 &ksig->info, &type);
2775
2776 if (!signr)
2777 break; /* will return 0 */
2778
2779 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2780 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2781 signr = ptrace_signal(signr, &ksig->info, type);
2782 if (!signr)
2783 continue;
2784 }
2785
2786 ka = &sighand->action[signr-1];
2787
2788 /* Trace actually delivered signals. */
2789 trace_signal_deliver(signr, &ksig->info, ka);
2790
2791 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2792 continue;
2793 if (ka->sa.sa_handler != SIG_DFL) {
2794 /* Run the handler. */
2795 ksig->ka = *ka;
2796
2797 if (ka->sa.sa_flags & SA_ONESHOT)
2798 ka->sa.sa_handler = SIG_DFL;
2799
2800 break; /* will return non-zero "signr" value */
2801 }
2802
2803 /*
2804 * Now we are doing the default action for this signal.
2805 */
2806 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2807 continue;
2808
2809 /*
2810 * Global init gets no signals it doesn't want.
2811 * Container-init gets no signals it doesn't want from same
2812 * container.
2813 *
2814 * Note that if global/container-init sees a sig_kernel_only()
2815 * signal here, the signal must have been generated internally
2816 * or must have come from an ancestor namespace. In either
2817 * case, the signal cannot be dropped.
2818 */
2819 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2820 !sig_kernel_only(signr))
2821 continue;
2822
2823 if (sig_kernel_stop(signr)) {
2824 /*
2825 * The default action is to stop all threads in
2826 * the thread group. The job control signals
2827 * do nothing in an orphaned pgrp, but SIGSTOP
2828 * always works. Note that siglock needs to be
2829 * dropped during the call to is_orphaned_pgrp()
2830 * because of lock ordering with tasklist_lock.
2831 * This allows an intervening SIGCONT to be posted.
2832 * We need to check for that and bail out if necessary.
2833 */
2834 if (signr != SIGSTOP) {
2835 spin_unlock_irq(&sighand->siglock);
2836
2837 /* signals can be posted during this window */
2838
2839 if (is_current_pgrp_orphaned())
2840 goto relock;
2841
2842 spin_lock_irq(&sighand->siglock);
2843 }
2844
2845 if (likely(do_signal_stop(ksig->info.si_signo))) {
2846 /* It released the siglock. */
2847 goto relock;
2848 }
2849
2850 /*
2851 * We didn't actually stop, due to a race
2852 * with SIGCONT or something like that.
2853 */
2854 continue;
2855 }
2856
2857 fatal:
2858 spin_unlock_irq(&sighand->siglock);
2859 if (unlikely(cgroup_task_frozen(current)))
2860 cgroup_leave_frozen(true);
2861
2862 /*
2863 * Anything else is fatal, maybe with a core dump.
2864 */
2865 current->flags |= PF_SIGNALED;
2866
2867 if (sig_kernel_coredump(signr)) {
2868 if (print_fatal_signals)
2869 print_fatal_signal(ksig->info.si_signo);
2870 proc_coredump_connector(current);
2871 /*
2872 * If it was able to dump core, this kills all
2873 * other threads in the group and synchronizes with
2874 * their demise. If we lost the race with another
2875 * thread getting here, it set group_exit_code
2876 * first and our do_group_exit call below will use
2877 * that value and ignore the one we pass it.
2878 */
2879 do_coredump(&ksig->info);
2880 }
2881
2882 /*
2883 * PF_USER_WORKER threads will catch and exit on fatal signals
2884 * themselves. They have cleanup that must be performed, so
2885 * we cannot call do_exit() on their behalf.
2886 */
2887 if (current->flags & PF_USER_WORKER)
2888 goto out;
2889
2890 /*
2891 * Death signals, no core dump.
2892 */
2893 do_group_exit(ksig->info.si_signo);
2894 /* NOTREACHED */
2895 }
2896 spin_unlock_irq(&sighand->siglock);
2897out:
2898 ksig->sig = signr;
2899
2900 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2901 hide_si_addr_tag_bits(ksig);
2902
2903 return ksig->sig > 0;
2904}
2905
2906/**
2907 * signal_delivered - called after signal delivery to update blocked signals
2908 * @ksig: kernel signal struct
2909 * @stepping: nonzero if debugger single-step or block-step in use
2910 *
2911 * This function should be called when a signal has successfully been
2912 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2913 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2914 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2915 */
2916static void signal_delivered(struct ksignal *ksig, int stepping)
2917{
2918 sigset_t blocked;
2919
2920 /* A signal was successfully delivered, and the
2921 saved sigmask was stored on the signal frame,
2922 and will be restored by sigreturn. So we can
2923 simply clear the restore sigmask flag. */
2924 clear_restore_sigmask();
2925
2926 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2927 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2928 sigaddset(&blocked, ksig->sig);
2929 set_current_blocked(&blocked);
2930 if (current->sas_ss_flags & SS_AUTODISARM)
2931 sas_ss_reset(current);
2932 if (stepping)
2933 ptrace_notify(SIGTRAP, 0);
2934}
2935
2936void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2937{
2938 if (failed)
2939 force_sigsegv(ksig->sig);
2940 else
2941 signal_delivered(ksig, stepping);
2942}
2943
2944/*
2945 * It could be that complete_signal() picked us to notify about the
2946 * group-wide signal. Other threads should be notified now to take
2947 * the shared signals in @which since we will not.
2948 */
2949static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2950{
2951 sigset_t retarget;
2952 struct task_struct *t;
2953
2954 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2955 if (sigisemptyset(&retarget))
2956 return;
2957
2958 for_other_threads(tsk, t) {
2959 if (t->flags & PF_EXITING)
2960 continue;
2961
2962 if (!has_pending_signals(&retarget, &t->blocked))
2963 continue;
2964 /* Remove the signals this thread can handle. */
2965 sigandsets(&retarget, &retarget, &t->blocked);
2966
2967 if (!task_sigpending(t))
2968 signal_wake_up(t, 0);
2969
2970 if (sigisemptyset(&retarget))
2971 break;
2972 }
2973}
2974
2975void exit_signals(struct task_struct *tsk)
2976{
2977 int group_stop = 0;
2978 sigset_t unblocked;
2979
2980 /*
2981 * @tsk is about to have PF_EXITING set - lock out users which
2982 * expect stable threadgroup.
2983 */
2984 cgroup_threadgroup_change_begin(tsk);
2985
2986 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
2987 sched_mm_cid_exit_signals(tsk);
2988 tsk->flags |= PF_EXITING;
2989 cgroup_threadgroup_change_end(tsk);
2990 return;
2991 }
2992
2993 spin_lock_irq(&tsk->sighand->siglock);
2994 /*
2995 * From now this task is not visible for group-wide signals,
2996 * see wants_signal(), do_signal_stop().
2997 */
2998 sched_mm_cid_exit_signals(tsk);
2999 tsk->flags |= PF_EXITING;
3000
3001 cgroup_threadgroup_change_end(tsk);
3002
3003 if (!task_sigpending(tsk))
3004 goto out;
3005
3006 unblocked = tsk->blocked;
3007 signotset(&unblocked);
3008 retarget_shared_pending(tsk, &unblocked);
3009
3010 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3011 task_participate_group_stop(tsk))
3012 group_stop = CLD_STOPPED;
3013out:
3014 spin_unlock_irq(&tsk->sighand->siglock);
3015
3016 /*
3017 * If group stop has completed, deliver the notification. This
3018 * should always go to the real parent of the group leader.
3019 */
3020 if (unlikely(group_stop)) {
3021 read_lock(&tasklist_lock);
3022 do_notify_parent_cldstop(tsk, false, group_stop);
3023 read_unlock(&tasklist_lock);
3024 }
3025}
3026
3027/*
3028 * System call entry points.
3029 */
3030
3031/**
3032 * sys_restart_syscall - restart a system call
3033 */
3034SYSCALL_DEFINE0(restart_syscall)
3035{
3036 struct restart_block *restart = ¤t->restart_block;
3037 return restart->fn(restart);
3038}
3039
3040long do_no_restart_syscall(struct restart_block *param)
3041{
3042 return -EINTR;
3043}
3044
3045static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3046{
3047 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3048 sigset_t newblocked;
3049 /* A set of now blocked but previously unblocked signals. */
3050 sigandnsets(&newblocked, newset, ¤t->blocked);
3051 retarget_shared_pending(tsk, &newblocked);
3052 }
3053 tsk->blocked = *newset;
3054 recalc_sigpending();
3055}
3056
3057/**
3058 * set_current_blocked - change current->blocked mask
3059 * @newset: new mask
3060 *
3061 * It is wrong to change ->blocked directly, this helper should be used
3062 * to ensure the process can't miss a shared signal we are going to block.
3063 */
3064void set_current_blocked(sigset_t *newset)
3065{
3066 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3067 __set_current_blocked(newset);
3068}
3069
3070void __set_current_blocked(const sigset_t *newset)
3071{
3072 struct task_struct *tsk = current;
3073
3074 /*
3075 * In case the signal mask hasn't changed, there is nothing we need
3076 * to do. The current->blocked shouldn't be modified by other task.
3077 */
3078 if (sigequalsets(&tsk->blocked, newset))
3079 return;
3080
3081 spin_lock_irq(&tsk->sighand->siglock);
3082 __set_task_blocked(tsk, newset);
3083 spin_unlock_irq(&tsk->sighand->siglock);
3084}
3085
3086/*
3087 * This is also useful for kernel threads that want to temporarily
3088 * (or permanently) block certain signals.
3089 *
3090 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3091 * interface happily blocks "unblockable" signals like SIGKILL
3092 * and friends.
3093 */
3094int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3095{
3096 struct task_struct *tsk = current;
3097 sigset_t newset;
3098
3099 /* Lockless, only current can change ->blocked, never from irq */
3100 if (oldset)
3101 *oldset = tsk->blocked;
3102
3103 switch (how) {
3104 case SIG_BLOCK:
3105 sigorsets(&newset, &tsk->blocked, set);
3106 break;
3107 case SIG_UNBLOCK:
3108 sigandnsets(&newset, &tsk->blocked, set);
3109 break;
3110 case SIG_SETMASK:
3111 newset = *set;
3112 break;
3113 default:
3114 return -EINVAL;
3115 }
3116
3117 __set_current_blocked(&newset);
3118 return 0;
3119}
3120EXPORT_SYMBOL(sigprocmask);
3121
3122/*
3123 * The api helps set app-provided sigmasks.
3124 *
3125 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3126 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3127 *
3128 * Note that it does set_restore_sigmask() in advance, so it must be always
3129 * paired with restore_saved_sigmask_unless() before return from syscall.
3130 */
3131int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3132{
3133 sigset_t kmask;
3134
3135 if (!umask)
3136 return 0;
3137 if (sigsetsize != sizeof(sigset_t))
3138 return -EINVAL;
3139 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3140 return -EFAULT;
3141
3142 set_restore_sigmask();
3143 current->saved_sigmask = current->blocked;
3144 set_current_blocked(&kmask);
3145
3146 return 0;
3147}
3148
3149#ifdef CONFIG_COMPAT
3150int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3151 size_t sigsetsize)
3152{
3153 sigset_t kmask;
3154
3155 if (!umask)
3156 return 0;
3157 if (sigsetsize != sizeof(compat_sigset_t))
3158 return -EINVAL;
3159 if (get_compat_sigset(&kmask, umask))
3160 return -EFAULT;
3161
3162 set_restore_sigmask();
3163 current->saved_sigmask = current->blocked;
3164 set_current_blocked(&kmask);
3165
3166 return 0;
3167}
3168#endif
3169
3170/**
3171 * sys_rt_sigprocmask - change the list of currently blocked signals
3172 * @how: whether to add, remove, or set signals
3173 * @nset: stores pending signals
3174 * @oset: previous value of signal mask if non-null
3175 * @sigsetsize: size of sigset_t type
3176 */
3177SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3178 sigset_t __user *, oset, size_t, sigsetsize)
3179{
3180 sigset_t old_set, new_set;
3181 int error;
3182
3183 /* XXX: Don't preclude handling different sized sigset_t's. */
3184 if (sigsetsize != sizeof(sigset_t))
3185 return -EINVAL;
3186
3187 old_set = current->blocked;
3188
3189 if (nset) {
3190 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3191 return -EFAULT;
3192 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3193
3194 error = sigprocmask(how, &new_set, NULL);
3195 if (error)
3196 return error;
3197 }
3198
3199 if (oset) {
3200 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3201 return -EFAULT;
3202 }
3203
3204 return 0;
3205}
3206
3207#ifdef CONFIG_COMPAT
3208COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3209 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3210{
3211 sigset_t old_set = current->blocked;
3212
3213 /* XXX: Don't preclude handling different sized sigset_t's. */
3214 if (sigsetsize != sizeof(sigset_t))
3215 return -EINVAL;
3216
3217 if (nset) {
3218 sigset_t new_set;
3219 int error;
3220 if (get_compat_sigset(&new_set, nset))
3221 return -EFAULT;
3222 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3223
3224 error = sigprocmask(how, &new_set, NULL);
3225 if (error)
3226 return error;
3227 }
3228 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3229}
3230#endif
3231
3232static void do_sigpending(sigset_t *set)
3233{
3234 spin_lock_irq(¤t->sighand->siglock);
3235 sigorsets(set, ¤t->pending.signal,
3236 ¤t->signal->shared_pending.signal);
3237 spin_unlock_irq(¤t->sighand->siglock);
3238
3239 /* Outside the lock because only this thread touches it. */
3240 sigandsets(set, ¤t->blocked, set);
3241}
3242
3243/**
3244 * sys_rt_sigpending - examine a pending signal that has been raised
3245 * while blocked
3246 * @uset: stores pending signals
3247 * @sigsetsize: size of sigset_t type or larger
3248 */
3249SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3250{
3251 sigset_t set;
3252
3253 if (sigsetsize > sizeof(*uset))
3254 return -EINVAL;
3255
3256 do_sigpending(&set);
3257
3258 if (copy_to_user(uset, &set, sigsetsize))
3259 return -EFAULT;
3260
3261 return 0;
3262}
3263
3264#ifdef CONFIG_COMPAT
3265COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3266 compat_size_t, sigsetsize)
3267{
3268 sigset_t set;
3269
3270 if (sigsetsize > sizeof(*uset))
3271 return -EINVAL;
3272
3273 do_sigpending(&set);
3274
3275 return put_compat_sigset(uset, &set, sigsetsize);
3276}
3277#endif
3278
3279static const struct {
3280 unsigned char limit, layout;
3281} sig_sicodes[] = {
3282 [SIGILL] = { NSIGILL, SIL_FAULT },
3283 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3284 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3285 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3286 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3287#if defined(SIGEMT)
3288 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3289#endif
3290 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3291 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3292 [SIGSYS] = { NSIGSYS, SIL_SYS },
3293};
3294
3295static bool known_siginfo_layout(unsigned sig, int si_code)
3296{
3297 if (si_code == SI_KERNEL)
3298 return true;
3299 else if ((si_code > SI_USER)) {
3300 if (sig_specific_sicodes(sig)) {
3301 if (si_code <= sig_sicodes[sig].limit)
3302 return true;
3303 }
3304 else if (si_code <= NSIGPOLL)
3305 return true;
3306 }
3307 else if (si_code >= SI_DETHREAD)
3308 return true;
3309 else if (si_code == SI_ASYNCNL)
3310 return true;
3311 return false;
3312}
3313
3314enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3315{
3316 enum siginfo_layout layout = SIL_KILL;
3317 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3318 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3319 (si_code <= sig_sicodes[sig].limit)) {
3320 layout = sig_sicodes[sig].layout;
3321 /* Handle the exceptions */
3322 if ((sig == SIGBUS) &&
3323 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3324 layout = SIL_FAULT_MCEERR;
3325 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3326 layout = SIL_FAULT_BNDERR;
3327#ifdef SEGV_PKUERR
3328 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3329 layout = SIL_FAULT_PKUERR;
3330#endif
3331 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3332 layout = SIL_FAULT_PERF_EVENT;
3333 else if (IS_ENABLED(CONFIG_SPARC) &&
3334 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3335 layout = SIL_FAULT_TRAPNO;
3336 else if (IS_ENABLED(CONFIG_ALPHA) &&
3337 ((sig == SIGFPE) ||
3338 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3339 layout = SIL_FAULT_TRAPNO;
3340 }
3341 else if (si_code <= NSIGPOLL)
3342 layout = SIL_POLL;
3343 } else {
3344 if (si_code == SI_TIMER)
3345 layout = SIL_TIMER;
3346 else if (si_code == SI_SIGIO)
3347 layout = SIL_POLL;
3348 else if (si_code < 0)
3349 layout = SIL_RT;
3350 }
3351 return layout;
3352}
3353
3354static inline char __user *si_expansion(const siginfo_t __user *info)
3355{
3356 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3357}
3358
3359int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3360{
3361 char __user *expansion = si_expansion(to);
3362 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3363 return -EFAULT;
3364 if (clear_user(expansion, SI_EXPANSION_SIZE))
3365 return -EFAULT;
3366 return 0;
3367}
3368
3369static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3370 const siginfo_t __user *from)
3371{
3372 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3373 char __user *expansion = si_expansion(from);
3374 char buf[SI_EXPANSION_SIZE];
3375 int i;
3376 /*
3377 * An unknown si_code might need more than
3378 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3379 * extra bytes are 0. This guarantees copy_siginfo_to_user
3380 * will return this data to userspace exactly.
3381 */
3382 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3383 return -EFAULT;
3384 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3385 if (buf[i] != 0)
3386 return -E2BIG;
3387 }
3388 }
3389 return 0;
3390}
3391
3392static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3393 const siginfo_t __user *from)
3394{
3395 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3396 return -EFAULT;
3397 to->si_signo = signo;
3398 return post_copy_siginfo_from_user(to, from);
3399}
3400
3401int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3402{
3403 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3404 return -EFAULT;
3405 return post_copy_siginfo_from_user(to, from);
3406}
3407
3408#ifdef CONFIG_COMPAT
3409/**
3410 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3411 * @to: compat siginfo destination
3412 * @from: kernel siginfo source
3413 *
3414 * Note: This function does not work properly for the SIGCHLD on x32, but
3415 * fortunately it doesn't have to. The only valid callers for this function are
3416 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3417 * The latter does not care because SIGCHLD will never cause a coredump.
3418 */
3419void copy_siginfo_to_external32(struct compat_siginfo *to,
3420 const struct kernel_siginfo *from)
3421{
3422 memset(to, 0, sizeof(*to));
3423
3424 to->si_signo = from->si_signo;
3425 to->si_errno = from->si_errno;
3426 to->si_code = from->si_code;
3427 switch(siginfo_layout(from->si_signo, from->si_code)) {
3428 case SIL_KILL:
3429 to->si_pid = from->si_pid;
3430 to->si_uid = from->si_uid;
3431 break;
3432 case SIL_TIMER:
3433 to->si_tid = from->si_tid;
3434 to->si_overrun = from->si_overrun;
3435 to->si_int = from->si_int;
3436 break;
3437 case SIL_POLL:
3438 to->si_band = from->si_band;
3439 to->si_fd = from->si_fd;
3440 break;
3441 case SIL_FAULT:
3442 to->si_addr = ptr_to_compat(from->si_addr);
3443 break;
3444 case SIL_FAULT_TRAPNO:
3445 to->si_addr = ptr_to_compat(from->si_addr);
3446 to->si_trapno = from->si_trapno;
3447 break;
3448 case SIL_FAULT_MCEERR:
3449 to->si_addr = ptr_to_compat(from->si_addr);
3450 to->si_addr_lsb = from->si_addr_lsb;
3451 break;
3452 case SIL_FAULT_BNDERR:
3453 to->si_addr = ptr_to_compat(from->si_addr);
3454 to->si_lower = ptr_to_compat(from->si_lower);
3455 to->si_upper = ptr_to_compat(from->si_upper);
3456 break;
3457 case SIL_FAULT_PKUERR:
3458 to->si_addr = ptr_to_compat(from->si_addr);
3459 to->si_pkey = from->si_pkey;
3460 break;
3461 case SIL_FAULT_PERF_EVENT:
3462 to->si_addr = ptr_to_compat(from->si_addr);
3463 to->si_perf_data = from->si_perf_data;
3464 to->si_perf_type = from->si_perf_type;
3465 to->si_perf_flags = from->si_perf_flags;
3466 break;
3467 case SIL_CHLD:
3468 to->si_pid = from->si_pid;
3469 to->si_uid = from->si_uid;
3470 to->si_status = from->si_status;
3471 to->si_utime = from->si_utime;
3472 to->si_stime = from->si_stime;
3473 break;
3474 case SIL_RT:
3475 to->si_pid = from->si_pid;
3476 to->si_uid = from->si_uid;
3477 to->si_int = from->si_int;
3478 break;
3479 case SIL_SYS:
3480 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3481 to->si_syscall = from->si_syscall;
3482 to->si_arch = from->si_arch;
3483 break;
3484 }
3485}
3486
3487int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3488 const struct kernel_siginfo *from)
3489{
3490 struct compat_siginfo new;
3491
3492 copy_siginfo_to_external32(&new, from);
3493 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3494 return -EFAULT;
3495 return 0;
3496}
3497
3498static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3499 const struct compat_siginfo *from)
3500{
3501 clear_siginfo(to);
3502 to->si_signo = from->si_signo;
3503 to->si_errno = from->si_errno;
3504 to->si_code = from->si_code;
3505 switch(siginfo_layout(from->si_signo, from->si_code)) {
3506 case SIL_KILL:
3507 to->si_pid = from->si_pid;
3508 to->si_uid = from->si_uid;
3509 break;
3510 case SIL_TIMER:
3511 to->si_tid = from->si_tid;
3512 to->si_overrun = from->si_overrun;
3513 to->si_int = from->si_int;
3514 break;
3515 case SIL_POLL:
3516 to->si_band = from->si_band;
3517 to->si_fd = from->si_fd;
3518 break;
3519 case SIL_FAULT:
3520 to->si_addr = compat_ptr(from->si_addr);
3521 break;
3522 case SIL_FAULT_TRAPNO:
3523 to->si_addr = compat_ptr(from->si_addr);
3524 to->si_trapno = from->si_trapno;
3525 break;
3526 case SIL_FAULT_MCEERR:
3527 to->si_addr = compat_ptr(from->si_addr);
3528 to->si_addr_lsb = from->si_addr_lsb;
3529 break;
3530 case SIL_FAULT_BNDERR:
3531 to->si_addr = compat_ptr(from->si_addr);
3532 to->si_lower = compat_ptr(from->si_lower);
3533 to->si_upper = compat_ptr(from->si_upper);
3534 break;
3535 case SIL_FAULT_PKUERR:
3536 to->si_addr = compat_ptr(from->si_addr);
3537 to->si_pkey = from->si_pkey;
3538 break;
3539 case SIL_FAULT_PERF_EVENT:
3540 to->si_addr = compat_ptr(from->si_addr);
3541 to->si_perf_data = from->si_perf_data;
3542 to->si_perf_type = from->si_perf_type;
3543 to->si_perf_flags = from->si_perf_flags;
3544 break;
3545 case SIL_CHLD:
3546 to->si_pid = from->si_pid;
3547 to->si_uid = from->si_uid;
3548 to->si_status = from->si_status;
3549#ifdef CONFIG_X86_X32_ABI
3550 if (in_x32_syscall()) {
3551 to->si_utime = from->_sifields._sigchld_x32._utime;
3552 to->si_stime = from->_sifields._sigchld_x32._stime;
3553 } else
3554#endif
3555 {
3556 to->si_utime = from->si_utime;
3557 to->si_stime = from->si_stime;
3558 }
3559 break;
3560 case SIL_RT:
3561 to->si_pid = from->si_pid;
3562 to->si_uid = from->si_uid;
3563 to->si_int = from->si_int;
3564 break;
3565 case SIL_SYS:
3566 to->si_call_addr = compat_ptr(from->si_call_addr);
3567 to->si_syscall = from->si_syscall;
3568 to->si_arch = from->si_arch;
3569 break;
3570 }
3571 return 0;
3572}
3573
3574static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3575 const struct compat_siginfo __user *ufrom)
3576{
3577 struct compat_siginfo from;
3578
3579 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3580 return -EFAULT;
3581
3582 from.si_signo = signo;
3583 return post_copy_siginfo_from_user32(to, &from);
3584}
3585
3586int copy_siginfo_from_user32(struct kernel_siginfo *to,
3587 const struct compat_siginfo __user *ufrom)
3588{
3589 struct compat_siginfo from;
3590
3591 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3592 return -EFAULT;
3593
3594 return post_copy_siginfo_from_user32(to, &from);
3595}
3596#endif /* CONFIG_COMPAT */
3597
3598/**
3599 * do_sigtimedwait - wait for queued signals specified in @which
3600 * @which: queued signals to wait for
3601 * @info: if non-null, the signal's siginfo is returned here
3602 * @ts: upper bound on process time suspension
3603 */
3604static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3605 const struct timespec64 *ts)
3606{
3607 ktime_t *to = NULL, timeout = KTIME_MAX;
3608 struct task_struct *tsk = current;
3609 sigset_t mask = *which;
3610 enum pid_type type;
3611 int sig, ret = 0;
3612
3613 if (ts) {
3614 if (!timespec64_valid(ts))
3615 return -EINVAL;
3616 timeout = timespec64_to_ktime(*ts);
3617 to = &timeout;
3618 }
3619
3620 /*
3621 * Invert the set of allowed signals to get those we want to block.
3622 */
3623 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3624 signotset(&mask);
3625
3626 spin_lock_irq(&tsk->sighand->siglock);
3627 sig = dequeue_signal(tsk, &mask, info, &type);
3628 if (!sig && timeout) {
3629 /*
3630 * None ready, temporarily unblock those we're interested
3631 * while we are sleeping in so that we'll be awakened when
3632 * they arrive. Unblocking is always fine, we can avoid
3633 * set_current_blocked().
3634 */
3635 tsk->real_blocked = tsk->blocked;
3636 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3637 recalc_sigpending();
3638 spin_unlock_irq(&tsk->sighand->siglock);
3639
3640 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3641 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3642 HRTIMER_MODE_REL);
3643 spin_lock_irq(&tsk->sighand->siglock);
3644 __set_task_blocked(tsk, &tsk->real_blocked);
3645 sigemptyset(&tsk->real_blocked);
3646 sig = dequeue_signal(tsk, &mask, info, &type);
3647 }
3648 spin_unlock_irq(&tsk->sighand->siglock);
3649
3650 if (sig)
3651 return sig;
3652 return ret ? -EINTR : -EAGAIN;
3653}
3654
3655/**
3656 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3657 * in @uthese
3658 * @uthese: queued signals to wait for
3659 * @uinfo: if non-null, the signal's siginfo is returned here
3660 * @uts: upper bound on process time suspension
3661 * @sigsetsize: size of sigset_t type
3662 */
3663SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3664 siginfo_t __user *, uinfo,
3665 const struct __kernel_timespec __user *, uts,
3666 size_t, sigsetsize)
3667{
3668 sigset_t these;
3669 struct timespec64 ts;
3670 kernel_siginfo_t info;
3671 int ret;
3672
3673 /* XXX: Don't preclude handling different sized sigset_t's. */
3674 if (sigsetsize != sizeof(sigset_t))
3675 return -EINVAL;
3676
3677 if (copy_from_user(&these, uthese, sizeof(these)))
3678 return -EFAULT;
3679
3680 if (uts) {
3681 if (get_timespec64(&ts, uts))
3682 return -EFAULT;
3683 }
3684
3685 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3686
3687 if (ret > 0 && uinfo) {
3688 if (copy_siginfo_to_user(uinfo, &info))
3689 ret = -EFAULT;
3690 }
3691
3692 return ret;
3693}
3694
3695#ifdef CONFIG_COMPAT_32BIT_TIME
3696SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3697 siginfo_t __user *, uinfo,
3698 const struct old_timespec32 __user *, uts,
3699 size_t, sigsetsize)
3700{
3701 sigset_t these;
3702 struct timespec64 ts;
3703 kernel_siginfo_t info;
3704 int ret;
3705
3706 if (sigsetsize != sizeof(sigset_t))
3707 return -EINVAL;
3708
3709 if (copy_from_user(&these, uthese, sizeof(these)))
3710 return -EFAULT;
3711
3712 if (uts) {
3713 if (get_old_timespec32(&ts, uts))
3714 return -EFAULT;
3715 }
3716
3717 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3718
3719 if (ret > 0 && uinfo) {
3720 if (copy_siginfo_to_user(uinfo, &info))
3721 ret = -EFAULT;
3722 }
3723
3724 return ret;
3725}
3726#endif
3727
3728#ifdef CONFIG_COMPAT
3729COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3730 struct compat_siginfo __user *, uinfo,
3731 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3732{
3733 sigset_t s;
3734 struct timespec64 t;
3735 kernel_siginfo_t info;
3736 long ret;
3737
3738 if (sigsetsize != sizeof(sigset_t))
3739 return -EINVAL;
3740
3741 if (get_compat_sigset(&s, uthese))
3742 return -EFAULT;
3743
3744 if (uts) {
3745 if (get_timespec64(&t, uts))
3746 return -EFAULT;
3747 }
3748
3749 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3750
3751 if (ret > 0 && uinfo) {
3752 if (copy_siginfo_to_user32(uinfo, &info))
3753 ret = -EFAULT;
3754 }
3755
3756 return ret;
3757}
3758
3759#ifdef CONFIG_COMPAT_32BIT_TIME
3760COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3761 struct compat_siginfo __user *, uinfo,
3762 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3763{
3764 sigset_t s;
3765 struct timespec64 t;
3766 kernel_siginfo_t info;
3767 long ret;
3768
3769 if (sigsetsize != sizeof(sigset_t))
3770 return -EINVAL;
3771
3772 if (get_compat_sigset(&s, uthese))
3773 return -EFAULT;
3774
3775 if (uts) {
3776 if (get_old_timespec32(&t, uts))
3777 return -EFAULT;
3778 }
3779
3780 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3781
3782 if (ret > 0 && uinfo) {
3783 if (copy_siginfo_to_user32(uinfo, &info))
3784 ret = -EFAULT;
3785 }
3786
3787 return ret;
3788}
3789#endif
3790#endif
3791
3792static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3793{
3794 clear_siginfo(info);
3795 info->si_signo = sig;
3796 info->si_errno = 0;
3797 info->si_code = SI_USER;
3798 info->si_pid = task_tgid_vnr(current);
3799 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3800}
3801
3802/**
3803 * sys_kill - send a signal to a process
3804 * @pid: the PID of the process
3805 * @sig: signal to be sent
3806 */
3807SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3808{
3809 struct kernel_siginfo info;
3810
3811 prepare_kill_siginfo(sig, &info);
3812
3813 return kill_something_info(sig, &info, pid);
3814}
3815
3816/*
3817 * Verify that the signaler and signalee either are in the same pid namespace
3818 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3819 * namespace.
3820 */
3821static bool access_pidfd_pidns(struct pid *pid)
3822{
3823 struct pid_namespace *active = task_active_pid_ns(current);
3824 struct pid_namespace *p = ns_of_pid(pid);
3825
3826 for (;;) {
3827 if (!p)
3828 return false;
3829 if (p == active)
3830 break;
3831 p = p->parent;
3832 }
3833
3834 return true;
3835}
3836
3837static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3838 siginfo_t __user *info)
3839{
3840#ifdef CONFIG_COMPAT
3841 /*
3842 * Avoid hooking up compat syscalls and instead handle necessary
3843 * conversions here. Note, this is a stop-gap measure and should not be
3844 * considered a generic solution.
3845 */
3846 if (in_compat_syscall())
3847 return copy_siginfo_from_user32(
3848 kinfo, (struct compat_siginfo __user *)info);
3849#endif
3850 return copy_siginfo_from_user(kinfo, info);
3851}
3852
3853static struct pid *pidfd_to_pid(const struct file *file)
3854{
3855 struct pid *pid;
3856
3857 pid = pidfd_pid(file);
3858 if (!IS_ERR(pid))
3859 return pid;
3860
3861 return tgid_pidfd_to_pid(file);
3862}
3863
3864/**
3865 * sys_pidfd_send_signal - Signal a process through a pidfd
3866 * @pidfd: file descriptor of the process
3867 * @sig: signal to send
3868 * @info: signal info
3869 * @flags: future flags
3870 *
3871 * The syscall currently only signals via PIDTYPE_PID which covers
3872 * kill(<positive-pid>, <signal>. It does not signal threads or process
3873 * groups.
3874 * In order to extend the syscall to threads and process groups the @flags
3875 * argument should be used. In essence, the @flags argument will determine
3876 * what is signaled and not the file descriptor itself. Put in other words,
3877 * grouping is a property of the flags argument not a property of the file
3878 * descriptor.
3879 *
3880 * Return: 0 on success, negative errno on failure
3881 */
3882SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3883 siginfo_t __user *, info, unsigned int, flags)
3884{
3885 int ret;
3886 struct fd f;
3887 struct pid *pid;
3888 kernel_siginfo_t kinfo;
3889
3890 /* Enforce flags be set to 0 until we add an extension. */
3891 if (flags)
3892 return -EINVAL;
3893
3894 f = fdget(pidfd);
3895 if (!f.file)
3896 return -EBADF;
3897
3898 /* Is this a pidfd? */
3899 pid = pidfd_to_pid(f.file);
3900 if (IS_ERR(pid)) {
3901 ret = PTR_ERR(pid);
3902 goto err;
3903 }
3904
3905 ret = -EINVAL;
3906 if (!access_pidfd_pidns(pid))
3907 goto err;
3908
3909 if (info) {
3910 ret = copy_siginfo_from_user_any(&kinfo, info);
3911 if (unlikely(ret))
3912 goto err;
3913
3914 ret = -EINVAL;
3915 if (unlikely(sig != kinfo.si_signo))
3916 goto err;
3917
3918 /* Only allow sending arbitrary signals to yourself. */
3919 ret = -EPERM;
3920 if ((task_pid(current) != pid) &&
3921 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3922 goto err;
3923 } else {
3924 prepare_kill_siginfo(sig, &kinfo);
3925 }
3926
3927 ret = kill_pid_info(sig, &kinfo, pid);
3928
3929err:
3930 fdput(f);
3931 return ret;
3932}
3933
3934static int
3935do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3936{
3937 struct task_struct *p;
3938 int error = -ESRCH;
3939
3940 rcu_read_lock();
3941 p = find_task_by_vpid(pid);
3942 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3943 error = check_kill_permission(sig, info, p);
3944 /*
3945 * The null signal is a permissions and process existence
3946 * probe. No signal is actually delivered.
3947 */
3948 if (!error && sig) {
3949 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3950 /*
3951 * If lock_task_sighand() failed we pretend the task
3952 * dies after receiving the signal. The window is tiny,
3953 * and the signal is private anyway.
3954 */
3955 if (unlikely(error == -ESRCH))
3956 error = 0;
3957 }
3958 }
3959 rcu_read_unlock();
3960
3961 return error;
3962}
3963
3964static int do_tkill(pid_t tgid, pid_t pid, int sig)
3965{
3966 struct kernel_siginfo info;
3967
3968 clear_siginfo(&info);
3969 info.si_signo = sig;
3970 info.si_errno = 0;
3971 info.si_code = SI_TKILL;
3972 info.si_pid = task_tgid_vnr(current);
3973 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3974
3975 return do_send_specific(tgid, pid, sig, &info);
3976}
3977
3978/**
3979 * sys_tgkill - send signal to one specific thread
3980 * @tgid: the thread group ID of the thread
3981 * @pid: the PID of the thread
3982 * @sig: signal to be sent
3983 *
3984 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3985 * exists but it's not belonging to the target process anymore. This
3986 * method solves the problem of threads exiting and PIDs getting reused.
3987 */
3988SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3989{
3990 /* This is only valid for single tasks */
3991 if (pid <= 0 || tgid <= 0)
3992 return -EINVAL;
3993
3994 return do_tkill(tgid, pid, sig);
3995}
3996
3997/**
3998 * sys_tkill - send signal to one specific task
3999 * @pid: the PID of the task
4000 * @sig: signal to be sent
4001 *
4002 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4003 */
4004SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4005{
4006 /* This is only valid for single tasks */
4007 if (pid <= 0)
4008 return -EINVAL;
4009
4010 return do_tkill(0, pid, sig);
4011}
4012
4013static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4014{
4015 /* Not even root can pretend to send signals from the kernel.
4016 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4017 */
4018 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4019 (task_pid_vnr(current) != pid))
4020 return -EPERM;
4021
4022 /* POSIX.1b doesn't mention process groups. */
4023 return kill_proc_info(sig, info, pid);
4024}
4025
4026/**
4027 * sys_rt_sigqueueinfo - send signal information to a signal
4028 * @pid: the PID of the thread
4029 * @sig: signal to be sent
4030 * @uinfo: signal info to be sent
4031 */
4032SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4033 siginfo_t __user *, uinfo)
4034{
4035 kernel_siginfo_t info;
4036 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4037 if (unlikely(ret))
4038 return ret;
4039 return do_rt_sigqueueinfo(pid, sig, &info);
4040}
4041
4042#ifdef CONFIG_COMPAT
4043COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4044 compat_pid_t, pid,
4045 int, sig,
4046 struct compat_siginfo __user *, uinfo)
4047{
4048 kernel_siginfo_t info;
4049 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4050 if (unlikely(ret))
4051 return ret;
4052 return do_rt_sigqueueinfo(pid, sig, &info);
4053}
4054#endif
4055
4056static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4057{
4058 /* This is only valid for single tasks */
4059 if (pid <= 0 || tgid <= 0)
4060 return -EINVAL;
4061
4062 /* Not even root can pretend to send signals from the kernel.
4063 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4064 */
4065 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4066 (task_pid_vnr(current) != pid))
4067 return -EPERM;
4068
4069 return do_send_specific(tgid, pid, sig, info);
4070}
4071
4072SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4073 siginfo_t __user *, uinfo)
4074{
4075 kernel_siginfo_t info;
4076 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4077 if (unlikely(ret))
4078 return ret;
4079 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4080}
4081
4082#ifdef CONFIG_COMPAT
4083COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4084 compat_pid_t, tgid,
4085 compat_pid_t, pid,
4086 int, sig,
4087 struct compat_siginfo __user *, uinfo)
4088{
4089 kernel_siginfo_t info;
4090 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4091 if (unlikely(ret))
4092 return ret;
4093 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4094}
4095#endif
4096
4097/*
4098 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4099 */
4100void kernel_sigaction(int sig, __sighandler_t action)
4101{
4102 spin_lock_irq(¤t->sighand->siglock);
4103 current->sighand->action[sig - 1].sa.sa_handler = action;
4104 if (action == SIG_IGN) {
4105 sigset_t mask;
4106
4107 sigemptyset(&mask);
4108 sigaddset(&mask, sig);
4109
4110 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
4111 flush_sigqueue_mask(&mask, ¤t->pending);
4112 recalc_sigpending();
4113 }
4114 spin_unlock_irq(¤t->sighand->siglock);
4115}
4116EXPORT_SYMBOL(kernel_sigaction);
4117
4118void __weak sigaction_compat_abi(struct k_sigaction *act,
4119 struct k_sigaction *oact)
4120{
4121}
4122
4123int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4124{
4125 struct task_struct *p = current, *t;
4126 struct k_sigaction *k;
4127 sigset_t mask;
4128
4129 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4130 return -EINVAL;
4131
4132 k = &p->sighand->action[sig-1];
4133
4134 spin_lock_irq(&p->sighand->siglock);
4135 if (k->sa.sa_flags & SA_IMMUTABLE) {
4136 spin_unlock_irq(&p->sighand->siglock);
4137 return -EINVAL;
4138 }
4139 if (oact)
4140 *oact = *k;
4141
4142 /*
4143 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4144 * e.g. by having an architecture use the bit in their uapi.
4145 */
4146 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4147
4148 /*
4149 * Clear unknown flag bits in order to allow userspace to detect missing
4150 * support for flag bits and to allow the kernel to use non-uapi bits
4151 * internally.
4152 */
4153 if (act)
4154 act->sa.sa_flags &= UAPI_SA_FLAGS;
4155 if (oact)
4156 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4157
4158 sigaction_compat_abi(act, oact);
4159
4160 if (act) {
4161 sigdelsetmask(&act->sa.sa_mask,
4162 sigmask(SIGKILL) | sigmask(SIGSTOP));
4163 *k = *act;
4164 /*
4165 * POSIX 3.3.1.3:
4166 * "Setting a signal action to SIG_IGN for a signal that is
4167 * pending shall cause the pending signal to be discarded,
4168 * whether or not it is blocked."
4169 *
4170 * "Setting a signal action to SIG_DFL for a signal that is
4171 * pending and whose default action is to ignore the signal
4172 * (for example, SIGCHLD), shall cause the pending signal to
4173 * be discarded, whether or not it is blocked"
4174 */
4175 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4176 sigemptyset(&mask);
4177 sigaddset(&mask, sig);
4178 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4179 for_each_thread(p, t)
4180 flush_sigqueue_mask(&mask, &t->pending);
4181 }
4182 }
4183
4184 spin_unlock_irq(&p->sighand->siglock);
4185 return 0;
4186}
4187
4188#ifdef CONFIG_DYNAMIC_SIGFRAME
4189static inline void sigaltstack_lock(void)
4190 __acquires(¤t->sighand->siglock)
4191{
4192 spin_lock_irq(¤t->sighand->siglock);
4193}
4194
4195static inline void sigaltstack_unlock(void)
4196 __releases(¤t->sighand->siglock)
4197{
4198 spin_unlock_irq(¤t->sighand->siglock);
4199}
4200#else
4201static inline void sigaltstack_lock(void) { }
4202static inline void sigaltstack_unlock(void) { }
4203#endif
4204
4205static int
4206do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4207 size_t min_ss_size)
4208{
4209 struct task_struct *t = current;
4210 int ret = 0;
4211
4212 if (oss) {
4213 memset(oss, 0, sizeof(stack_t));
4214 oss->ss_sp = (void __user *) t->sas_ss_sp;
4215 oss->ss_size = t->sas_ss_size;
4216 oss->ss_flags = sas_ss_flags(sp) |
4217 (current->sas_ss_flags & SS_FLAG_BITS);
4218 }
4219
4220 if (ss) {
4221 void __user *ss_sp = ss->ss_sp;
4222 size_t ss_size = ss->ss_size;
4223 unsigned ss_flags = ss->ss_flags;
4224 int ss_mode;
4225
4226 if (unlikely(on_sig_stack(sp)))
4227 return -EPERM;
4228
4229 ss_mode = ss_flags & ~SS_FLAG_BITS;
4230 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4231 ss_mode != 0))
4232 return -EINVAL;
4233
4234 /*
4235 * Return before taking any locks if no actual
4236 * sigaltstack changes were requested.
4237 */
4238 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4239 t->sas_ss_size == ss_size &&
4240 t->sas_ss_flags == ss_flags)
4241 return 0;
4242
4243 sigaltstack_lock();
4244 if (ss_mode == SS_DISABLE) {
4245 ss_size = 0;
4246 ss_sp = NULL;
4247 } else {
4248 if (unlikely(ss_size < min_ss_size))
4249 ret = -ENOMEM;
4250 if (!sigaltstack_size_valid(ss_size))
4251 ret = -ENOMEM;
4252 }
4253 if (!ret) {
4254 t->sas_ss_sp = (unsigned long) ss_sp;
4255 t->sas_ss_size = ss_size;
4256 t->sas_ss_flags = ss_flags;
4257 }
4258 sigaltstack_unlock();
4259 }
4260 return ret;
4261}
4262
4263SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4264{
4265 stack_t new, old;
4266 int err;
4267 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4268 return -EFAULT;
4269 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4270 current_user_stack_pointer(),
4271 MINSIGSTKSZ);
4272 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4273 err = -EFAULT;
4274 return err;
4275}
4276
4277int restore_altstack(const stack_t __user *uss)
4278{
4279 stack_t new;
4280 if (copy_from_user(&new, uss, sizeof(stack_t)))
4281 return -EFAULT;
4282 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4283 MINSIGSTKSZ);
4284 /* squash all but EFAULT for now */
4285 return 0;
4286}
4287
4288int __save_altstack(stack_t __user *uss, unsigned long sp)
4289{
4290 struct task_struct *t = current;
4291 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4292 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4293 __put_user(t->sas_ss_size, &uss->ss_size);
4294 return err;
4295}
4296
4297#ifdef CONFIG_COMPAT
4298static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4299 compat_stack_t __user *uoss_ptr)
4300{
4301 stack_t uss, uoss;
4302 int ret;
4303
4304 if (uss_ptr) {
4305 compat_stack_t uss32;
4306 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4307 return -EFAULT;
4308 uss.ss_sp = compat_ptr(uss32.ss_sp);
4309 uss.ss_flags = uss32.ss_flags;
4310 uss.ss_size = uss32.ss_size;
4311 }
4312 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4313 compat_user_stack_pointer(),
4314 COMPAT_MINSIGSTKSZ);
4315 if (ret >= 0 && uoss_ptr) {
4316 compat_stack_t old;
4317 memset(&old, 0, sizeof(old));
4318 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4319 old.ss_flags = uoss.ss_flags;
4320 old.ss_size = uoss.ss_size;
4321 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4322 ret = -EFAULT;
4323 }
4324 return ret;
4325}
4326
4327COMPAT_SYSCALL_DEFINE2(sigaltstack,
4328 const compat_stack_t __user *, uss_ptr,
4329 compat_stack_t __user *, uoss_ptr)
4330{
4331 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4332}
4333
4334int compat_restore_altstack(const compat_stack_t __user *uss)
4335{
4336 int err = do_compat_sigaltstack(uss, NULL);
4337 /* squash all but -EFAULT for now */
4338 return err == -EFAULT ? err : 0;
4339}
4340
4341int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4342{
4343 int err;
4344 struct task_struct *t = current;
4345 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4346 &uss->ss_sp) |
4347 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4348 __put_user(t->sas_ss_size, &uss->ss_size);
4349 return err;
4350}
4351#endif
4352
4353#ifdef __ARCH_WANT_SYS_SIGPENDING
4354
4355/**
4356 * sys_sigpending - examine pending signals
4357 * @uset: where mask of pending signal is returned
4358 */
4359SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4360{
4361 sigset_t set;
4362
4363 if (sizeof(old_sigset_t) > sizeof(*uset))
4364 return -EINVAL;
4365
4366 do_sigpending(&set);
4367
4368 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4369 return -EFAULT;
4370
4371 return 0;
4372}
4373
4374#ifdef CONFIG_COMPAT
4375COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4376{
4377 sigset_t set;
4378
4379 do_sigpending(&set);
4380
4381 return put_user(set.sig[0], set32);
4382}
4383#endif
4384
4385#endif
4386
4387#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4388/**
4389 * sys_sigprocmask - examine and change blocked signals
4390 * @how: whether to add, remove, or set signals
4391 * @nset: signals to add or remove (if non-null)
4392 * @oset: previous value of signal mask if non-null
4393 *
4394 * Some platforms have their own version with special arguments;
4395 * others support only sys_rt_sigprocmask.
4396 */
4397
4398SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4399 old_sigset_t __user *, oset)
4400{
4401 old_sigset_t old_set, new_set;
4402 sigset_t new_blocked;
4403
4404 old_set = current->blocked.sig[0];
4405
4406 if (nset) {
4407 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4408 return -EFAULT;
4409
4410 new_blocked = current->blocked;
4411
4412 switch (how) {
4413 case SIG_BLOCK:
4414 sigaddsetmask(&new_blocked, new_set);
4415 break;
4416 case SIG_UNBLOCK:
4417 sigdelsetmask(&new_blocked, new_set);
4418 break;
4419 case SIG_SETMASK:
4420 new_blocked.sig[0] = new_set;
4421 break;
4422 default:
4423 return -EINVAL;
4424 }
4425
4426 set_current_blocked(&new_blocked);
4427 }
4428
4429 if (oset) {
4430 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4431 return -EFAULT;
4432 }
4433
4434 return 0;
4435}
4436#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4437
4438#ifndef CONFIG_ODD_RT_SIGACTION
4439/**
4440 * sys_rt_sigaction - alter an action taken by a process
4441 * @sig: signal to be sent
4442 * @act: new sigaction
4443 * @oact: used to save the previous sigaction
4444 * @sigsetsize: size of sigset_t type
4445 */
4446SYSCALL_DEFINE4(rt_sigaction, int, sig,
4447 const struct sigaction __user *, act,
4448 struct sigaction __user *, oact,
4449 size_t, sigsetsize)
4450{
4451 struct k_sigaction new_sa, old_sa;
4452 int ret;
4453
4454 /* XXX: Don't preclude handling different sized sigset_t's. */
4455 if (sigsetsize != sizeof(sigset_t))
4456 return -EINVAL;
4457
4458 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4459 return -EFAULT;
4460
4461 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4462 if (ret)
4463 return ret;
4464
4465 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4466 return -EFAULT;
4467
4468 return 0;
4469}
4470#ifdef CONFIG_COMPAT
4471COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4472 const struct compat_sigaction __user *, act,
4473 struct compat_sigaction __user *, oact,
4474 compat_size_t, sigsetsize)
4475{
4476 struct k_sigaction new_ka, old_ka;
4477#ifdef __ARCH_HAS_SA_RESTORER
4478 compat_uptr_t restorer;
4479#endif
4480 int ret;
4481
4482 /* XXX: Don't preclude handling different sized sigset_t's. */
4483 if (sigsetsize != sizeof(compat_sigset_t))
4484 return -EINVAL;
4485
4486 if (act) {
4487 compat_uptr_t handler;
4488 ret = get_user(handler, &act->sa_handler);
4489 new_ka.sa.sa_handler = compat_ptr(handler);
4490#ifdef __ARCH_HAS_SA_RESTORER
4491 ret |= get_user(restorer, &act->sa_restorer);
4492 new_ka.sa.sa_restorer = compat_ptr(restorer);
4493#endif
4494 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4495 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4496 if (ret)
4497 return -EFAULT;
4498 }
4499
4500 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4501 if (!ret && oact) {
4502 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4503 &oact->sa_handler);
4504 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4505 sizeof(oact->sa_mask));
4506 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4507#ifdef __ARCH_HAS_SA_RESTORER
4508 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4509 &oact->sa_restorer);
4510#endif
4511 }
4512 return ret;
4513}
4514#endif
4515#endif /* !CONFIG_ODD_RT_SIGACTION */
4516
4517#ifdef CONFIG_OLD_SIGACTION
4518SYSCALL_DEFINE3(sigaction, int, sig,
4519 const struct old_sigaction __user *, act,
4520 struct old_sigaction __user *, oact)
4521{
4522 struct k_sigaction new_ka, old_ka;
4523 int ret;
4524
4525 if (act) {
4526 old_sigset_t mask;
4527 if (!access_ok(act, sizeof(*act)) ||
4528 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4529 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4530 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4531 __get_user(mask, &act->sa_mask))
4532 return -EFAULT;
4533#ifdef __ARCH_HAS_KA_RESTORER
4534 new_ka.ka_restorer = NULL;
4535#endif
4536 siginitset(&new_ka.sa.sa_mask, mask);
4537 }
4538
4539 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4540
4541 if (!ret && oact) {
4542 if (!access_ok(oact, sizeof(*oact)) ||
4543 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4544 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4545 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4546 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4547 return -EFAULT;
4548 }
4549
4550 return ret;
4551}
4552#endif
4553#ifdef CONFIG_COMPAT_OLD_SIGACTION
4554COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4555 const struct compat_old_sigaction __user *, act,
4556 struct compat_old_sigaction __user *, oact)
4557{
4558 struct k_sigaction new_ka, old_ka;
4559 int ret;
4560 compat_old_sigset_t mask;
4561 compat_uptr_t handler, restorer;
4562
4563 if (act) {
4564 if (!access_ok(act, sizeof(*act)) ||
4565 __get_user(handler, &act->sa_handler) ||
4566 __get_user(restorer, &act->sa_restorer) ||
4567 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4568 __get_user(mask, &act->sa_mask))
4569 return -EFAULT;
4570
4571#ifdef __ARCH_HAS_KA_RESTORER
4572 new_ka.ka_restorer = NULL;
4573#endif
4574 new_ka.sa.sa_handler = compat_ptr(handler);
4575 new_ka.sa.sa_restorer = compat_ptr(restorer);
4576 siginitset(&new_ka.sa.sa_mask, mask);
4577 }
4578
4579 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4580
4581 if (!ret && oact) {
4582 if (!access_ok(oact, sizeof(*oact)) ||
4583 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4584 &oact->sa_handler) ||
4585 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4586 &oact->sa_restorer) ||
4587 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4588 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4589 return -EFAULT;
4590 }
4591 return ret;
4592}
4593#endif
4594
4595#ifdef CONFIG_SGETMASK_SYSCALL
4596
4597/*
4598 * For backwards compatibility. Functionality superseded by sigprocmask.
4599 */
4600SYSCALL_DEFINE0(sgetmask)
4601{
4602 /* SMP safe */
4603 return current->blocked.sig[0];
4604}
4605
4606SYSCALL_DEFINE1(ssetmask, int, newmask)
4607{
4608 int old = current->blocked.sig[0];
4609 sigset_t newset;
4610
4611 siginitset(&newset, newmask);
4612 set_current_blocked(&newset);
4613
4614 return old;
4615}
4616#endif /* CONFIG_SGETMASK_SYSCALL */
4617
4618#ifdef __ARCH_WANT_SYS_SIGNAL
4619/*
4620 * For backwards compatibility. Functionality superseded by sigaction.
4621 */
4622SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4623{
4624 struct k_sigaction new_sa, old_sa;
4625 int ret;
4626
4627 new_sa.sa.sa_handler = handler;
4628 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4629 sigemptyset(&new_sa.sa.sa_mask);
4630
4631 ret = do_sigaction(sig, &new_sa, &old_sa);
4632
4633 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4634}
4635#endif /* __ARCH_WANT_SYS_SIGNAL */
4636
4637#ifdef __ARCH_WANT_SYS_PAUSE
4638
4639SYSCALL_DEFINE0(pause)
4640{
4641 while (!signal_pending(current)) {
4642 __set_current_state(TASK_INTERRUPTIBLE);
4643 schedule();
4644 }
4645 return -ERESTARTNOHAND;
4646}
4647
4648#endif
4649
4650static int sigsuspend(sigset_t *set)
4651{
4652 current->saved_sigmask = current->blocked;
4653 set_current_blocked(set);
4654
4655 while (!signal_pending(current)) {
4656 __set_current_state(TASK_INTERRUPTIBLE);
4657 schedule();
4658 }
4659 set_restore_sigmask();
4660 return -ERESTARTNOHAND;
4661}
4662
4663/**
4664 * sys_rt_sigsuspend - replace the signal mask for a value with the
4665 * @unewset value until a signal is received
4666 * @unewset: new signal mask value
4667 * @sigsetsize: size of sigset_t type
4668 */
4669SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4670{
4671 sigset_t newset;
4672
4673 /* XXX: Don't preclude handling different sized sigset_t's. */
4674 if (sigsetsize != sizeof(sigset_t))
4675 return -EINVAL;
4676
4677 if (copy_from_user(&newset, unewset, sizeof(newset)))
4678 return -EFAULT;
4679 return sigsuspend(&newset);
4680}
4681
4682#ifdef CONFIG_COMPAT
4683COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4684{
4685 sigset_t newset;
4686
4687 /* XXX: Don't preclude handling different sized sigset_t's. */
4688 if (sigsetsize != sizeof(sigset_t))
4689 return -EINVAL;
4690
4691 if (get_compat_sigset(&newset, unewset))
4692 return -EFAULT;
4693 return sigsuspend(&newset);
4694}
4695#endif
4696
4697#ifdef CONFIG_OLD_SIGSUSPEND
4698SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4699{
4700 sigset_t blocked;
4701 siginitset(&blocked, mask);
4702 return sigsuspend(&blocked);
4703}
4704#endif
4705#ifdef CONFIG_OLD_SIGSUSPEND3
4706SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4707{
4708 sigset_t blocked;
4709 siginitset(&blocked, mask);
4710 return sigsuspend(&blocked);
4711}
4712#endif
4713
4714__weak const char *arch_vma_name(struct vm_area_struct *vma)
4715{
4716 return NULL;
4717}
4718
4719static inline void siginfo_buildtime_checks(void)
4720{
4721 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4722
4723 /* Verify the offsets in the two siginfos match */
4724#define CHECK_OFFSET(field) \
4725 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4726
4727 /* kill */
4728 CHECK_OFFSET(si_pid);
4729 CHECK_OFFSET(si_uid);
4730
4731 /* timer */
4732 CHECK_OFFSET(si_tid);
4733 CHECK_OFFSET(si_overrun);
4734 CHECK_OFFSET(si_value);
4735
4736 /* rt */
4737 CHECK_OFFSET(si_pid);
4738 CHECK_OFFSET(si_uid);
4739 CHECK_OFFSET(si_value);
4740
4741 /* sigchld */
4742 CHECK_OFFSET(si_pid);
4743 CHECK_OFFSET(si_uid);
4744 CHECK_OFFSET(si_status);
4745 CHECK_OFFSET(si_utime);
4746 CHECK_OFFSET(si_stime);
4747
4748 /* sigfault */
4749 CHECK_OFFSET(si_addr);
4750 CHECK_OFFSET(si_trapno);
4751 CHECK_OFFSET(si_addr_lsb);
4752 CHECK_OFFSET(si_lower);
4753 CHECK_OFFSET(si_upper);
4754 CHECK_OFFSET(si_pkey);
4755 CHECK_OFFSET(si_perf_data);
4756 CHECK_OFFSET(si_perf_type);
4757 CHECK_OFFSET(si_perf_flags);
4758
4759 /* sigpoll */
4760 CHECK_OFFSET(si_band);
4761 CHECK_OFFSET(si_fd);
4762
4763 /* sigsys */
4764 CHECK_OFFSET(si_call_addr);
4765 CHECK_OFFSET(si_syscall);
4766 CHECK_OFFSET(si_arch);
4767#undef CHECK_OFFSET
4768
4769 /* usb asyncio */
4770 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4771 offsetof(struct siginfo, si_addr));
4772 if (sizeof(int) == sizeof(void __user *)) {
4773 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4774 sizeof(void __user *));
4775 } else {
4776 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4777 sizeof_field(struct siginfo, si_uid)) !=
4778 sizeof(void __user *));
4779 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4780 offsetof(struct siginfo, si_uid));
4781 }
4782#ifdef CONFIG_COMPAT
4783 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4784 offsetof(struct compat_siginfo, si_addr));
4785 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4786 sizeof(compat_uptr_t));
4787 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4788 sizeof_field(struct siginfo, si_pid));
4789#endif
4790}
4791
4792#if defined(CONFIG_SYSCTL)
4793static struct ctl_table signal_debug_table[] = {
4794#ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4795 {
4796 .procname = "exception-trace",
4797 .data = &show_unhandled_signals,
4798 .maxlen = sizeof(int),
4799 .mode = 0644,
4800 .proc_handler = proc_dointvec
4801 },
4802#endif
4803 { }
4804};
4805
4806static int __init init_signal_sysctls(void)
4807{
4808 register_sysctl_init("debug", signal_debug_table);
4809 return 0;
4810}
4811early_initcall(init_signal_sysctls);
4812#endif /* CONFIG_SYSCTL */
4813
4814void __init signals_init(void)
4815{
4816 siginfo_buildtime_checks();
4817
4818 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4819}
4820
4821#ifdef CONFIG_KGDB_KDB
4822#include <linux/kdb.h>
4823/*
4824 * kdb_send_sig - Allows kdb to send signals without exposing
4825 * signal internals. This function checks if the required locks are
4826 * available before calling the main signal code, to avoid kdb
4827 * deadlocks.
4828 */
4829void kdb_send_sig(struct task_struct *t, int sig)
4830{
4831 static struct task_struct *kdb_prev_t;
4832 int new_t, ret;
4833 if (!spin_trylock(&t->sighand->siglock)) {
4834 kdb_printf("Can't do kill command now.\n"
4835 "The sigmask lock is held somewhere else in "
4836 "kernel, try again later\n");
4837 return;
4838 }
4839 new_t = kdb_prev_t != t;
4840 kdb_prev_t = t;
4841 if (!task_is_running(t) && new_t) {
4842 spin_unlock(&t->sighand->siglock);
4843 kdb_printf("Process is not RUNNING, sending a signal from "
4844 "kdb risks deadlock\n"
4845 "on the run queue locks. "
4846 "The signal has _not_ been sent.\n"
4847 "Reissue the kill command if you want to risk "
4848 "the deadlock.\n");
4849 return;
4850 }
4851 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4852 spin_unlock(&t->sighand->siglock);
4853 if (ret)
4854 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4855 sig, t->pid);
4856 else
4857 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4858}
4859#endif /* CONFIG_KGDB_KDB */