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 int sigpending;
417
418 /*
419 * Protect access to @t credentials. This can go away when all
420 * callers hold rcu read lock.
421 *
422 * NOTE! A pending signal will hold on to the user refcount,
423 * and we get/put the refcount only when the sigpending count
424 * changes from/to zero.
425 */
426 rcu_read_lock();
427 user = __task_cred(t)->user;
428 sigpending = atomic_inc_return(&user->sigpending);
429 if (sigpending == 1)
430 get_uid(user);
431 rcu_read_unlock();
432
433 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
434 q = kmem_cache_alloc(sigqueue_cachep, flags);
435 } else {
436 print_dropped_signal(sig);
437 }
438
439 if (unlikely(q == NULL)) {
440 if (atomic_dec_and_test(&user->sigpending))
441 free_uid(user);
442 } else {
443 INIT_LIST_HEAD(&q->list);
444 q->flags = 0;
445 q->user = user;
446 }
447
448 return q;
449}
450
451static void __sigqueue_free(struct sigqueue *q)
452{
453 if (q->flags & SIGQUEUE_PREALLOC)
454 return;
455 if (atomic_dec_and_test(&q->user->sigpending))
456 free_uid(q->user);
457 kmem_cache_free(sigqueue_cachep, q);
458}
459
460void flush_sigqueue(struct sigpending *queue)
461{
462 struct sigqueue *q;
463
464 sigemptyset(&queue->signal);
465 while (!list_empty(&queue->list)) {
466 q = list_entry(queue->list.next, struct sigqueue , list);
467 list_del_init(&q->list);
468 __sigqueue_free(q);
469 }
470}
471
472/*
473 * Flush all pending signals for this kthread.
474 */
475void flush_signals(struct task_struct *t)
476{
477 unsigned long flags;
478
479 spin_lock_irqsave(&t->sighand->siglock, flags);
480 clear_tsk_thread_flag(t, TIF_SIGPENDING);
481 flush_sigqueue(&t->pending);
482 flush_sigqueue(&t->signal->shared_pending);
483 spin_unlock_irqrestore(&t->sighand->siglock, flags);
484}
485EXPORT_SYMBOL(flush_signals);
486
487#ifdef CONFIG_POSIX_TIMERS
488static void __flush_itimer_signals(struct sigpending *pending)
489{
490 sigset_t signal, retain;
491 struct sigqueue *q, *n;
492
493 signal = pending->signal;
494 sigemptyset(&retain);
495
496 list_for_each_entry_safe(q, n, &pending->list, list) {
497 int sig = q->info.si_signo;
498
499 if (likely(q->info.si_code != SI_TIMER)) {
500 sigaddset(&retain, sig);
501 } else {
502 sigdelset(&signal, sig);
503 list_del_init(&q->list);
504 __sigqueue_free(q);
505 }
506 }
507
508 sigorsets(&pending->signal, &signal, &retain);
509}
510
511void flush_itimer_signals(void)
512{
513 struct task_struct *tsk = current;
514 unsigned long flags;
515
516 spin_lock_irqsave(&tsk->sighand->siglock, flags);
517 __flush_itimer_signals(&tsk->pending);
518 __flush_itimer_signals(&tsk->signal->shared_pending);
519 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
520}
521#endif
522
523void ignore_signals(struct task_struct *t)
524{
525 int i;
526
527 for (i = 0; i < _NSIG; ++i)
528 t->sighand->action[i].sa.sa_handler = SIG_IGN;
529
530 flush_signals(t);
531}
532
533/*
534 * Flush all handlers for a task.
535 */
536
537void
538flush_signal_handlers(struct task_struct *t, int force_default)
539{
540 int i;
541 struct k_sigaction *ka = &t->sighand->action[0];
542 for (i = _NSIG ; i != 0 ; i--) {
543 if (force_default || ka->sa.sa_handler != SIG_IGN)
544 ka->sa.sa_handler = SIG_DFL;
545 ka->sa.sa_flags = 0;
546#ifdef __ARCH_HAS_SA_RESTORER
547 ka->sa.sa_restorer = NULL;
548#endif
549 sigemptyset(&ka->sa.sa_mask);
550 ka++;
551 }
552}
553
554bool unhandled_signal(struct task_struct *tsk, int sig)
555{
556 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
557 if (is_global_init(tsk))
558 return true;
559
560 if (handler != SIG_IGN && handler != SIG_DFL)
561 return false;
562
563 /* if ptraced, let the tracer determine */
564 return !tsk->ptrace;
565}
566
567static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
568 bool *resched_timer)
569{
570 struct sigqueue *q, *first = NULL;
571
572 /*
573 * Collect the siginfo appropriate to this signal. Check if
574 * there is another siginfo for the same signal.
575 */
576 list_for_each_entry(q, &list->list, list) {
577 if (q->info.si_signo == sig) {
578 if (first)
579 goto still_pending;
580 first = q;
581 }
582 }
583
584 sigdelset(&list->signal, sig);
585
586 if (first) {
587still_pending:
588 list_del_init(&first->list);
589 copy_siginfo(info, &first->info);
590
591 *resched_timer =
592 (first->flags & SIGQUEUE_PREALLOC) &&
593 (info->si_code == SI_TIMER) &&
594 (info->si_sys_private);
595
596 __sigqueue_free(first);
597 } else {
598 /*
599 * Ok, it wasn't in the queue. This must be
600 * a fast-pathed signal or we must have been
601 * out of queue space. So zero out the info.
602 */
603 clear_siginfo(info);
604 info->si_signo = sig;
605 info->si_errno = 0;
606 info->si_code = SI_USER;
607 info->si_pid = 0;
608 info->si_uid = 0;
609 }
610}
611
612static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
613 kernel_siginfo_t *info, bool *resched_timer)
614{
615 int sig = next_signal(pending, mask);
616
617 if (sig)
618 collect_signal(sig, pending, info, resched_timer);
619 return sig;
620}
621
622/*
623 * Dequeue a signal and return the element to the caller, which is
624 * expected to free it.
625 *
626 * All callers have to hold the siglock.
627 */
628int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
629{
630 bool resched_timer = false;
631 int signr;
632
633 /* We only dequeue private signals from ourselves, we don't let
634 * signalfd steal them
635 */
636 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
637 if (!signr) {
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 set_tsk_thread_flag(t, TIF_SIGPENDING);
762 /*
763 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
764 * case. We don't check t->state here because there is a race with it
765 * executing another processor and just now entering stopped state.
766 * By using wake_up_state, we ensure the process will wake up and
767 * handle its death signal.
768 */
769 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
770 kick_process(t);
771}
772
773/*
774 * Remove signals in mask from the pending set and queue.
775 * Returns 1 if any signals were found.
776 *
777 * All callers must be holding the siglock.
778 */
779static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
780{
781 struct sigqueue *q, *n;
782 sigset_t m;
783
784 sigandsets(&m, mask, &s->signal);
785 if (sigisemptyset(&m))
786 return;
787
788 sigandnsets(&s->signal, &s->signal, mask);
789 list_for_each_entry_safe(q, n, &s->list, list) {
790 if (sigismember(mask, q->info.si_signo)) {
791 list_del_init(&q->list);
792 __sigqueue_free(q);
793 }
794 }
795}
796
797static inline int is_si_special(const struct kernel_siginfo *info)
798{
799 return info <= SEND_SIG_PRIV;
800}
801
802static inline bool si_fromuser(const struct kernel_siginfo *info)
803{
804 return info == SEND_SIG_NOINFO ||
805 (!is_si_special(info) && SI_FROMUSER(info));
806}
807
808/*
809 * called with RCU read lock from check_kill_permission()
810 */
811static bool kill_ok_by_cred(struct task_struct *t)
812{
813 const struct cred *cred = current_cred();
814 const struct cred *tcred = __task_cred(t);
815
816 return uid_eq(cred->euid, tcred->suid) ||
817 uid_eq(cred->euid, tcred->uid) ||
818 uid_eq(cred->uid, tcred->suid) ||
819 uid_eq(cred->uid, tcred->uid) ||
820 ns_capable(tcred->user_ns, CAP_KILL);
821}
822
823/*
824 * Bad permissions for sending the signal
825 * - the caller must hold the RCU read lock
826 */
827static int check_kill_permission(int sig, struct kernel_siginfo *info,
828 struct task_struct *t)
829{
830 struct pid *sid;
831 int error;
832
833 if (!valid_signal(sig))
834 return -EINVAL;
835
836 if (!si_fromuser(info))
837 return 0;
838
839 error = audit_signal_info(sig, t); /* Let audit system see the signal */
840 if (error)
841 return error;
842
843 if (!same_thread_group(current, t) &&
844 !kill_ok_by_cred(t)) {
845 switch (sig) {
846 case SIGCONT:
847 sid = task_session(t);
848 /*
849 * We don't return the error if sid == NULL. The
850 * task was unhashed, the caller must notice this.
851 */
852 if (!sid || sid == task_session(current))
853 break;
854 fallthrough;
855 default:
856 return -EPERM;
857 }
858 }
859
860 return security_task_kill(t, info, sig, NULL);
861}
862
863/**
864 * ptrace_trap_notify - schedule trap to notify ptracer
865 * @t: tracee wanting to notify tracer
866 *
867 * This function schedules sticky ptrace trap which is cleared on the next
868 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
869 * ptracer.
870 *
871 * If @t is running, STOP trap will be taken. If trapped for STOP and
872 * ptracer is listening for events, tracee is woken up so that it can
873 * re-trap for the new event. If trapped otherwise, STOP trap will be
874 * eventually taken without returning to userland after the existing traps
875 * are finished by PTRACE_CONT.
876 *
877 * CONTEXT:
878 * Must be called with @task->sighand->siglock held.
879 */
880static void ptrace_trap_notify(struct task_struct *t)
881{
882 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
883 assert_spin_locked(&t->sighand->siglock);
884
885 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
886 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
887}
888
889/*
890 * Handle magic process-wide effects of stop/continue signals. Unlike
891 * the signal actions, these happen immediately at signal-generation
892 * time regardless of blocking, ignoring, or handling. This does the
893 * actual continuing for SIGCONT, but not the actual stopping for stop
894 * signals. The process stop is done as a signal action for SIG_DFL.
895 *
896 * Returns true if the signal should be actually delivered, otherwise
897 * it should be dropped.
898 */
899static bool prepare_signal(int sig, struct task_struct *p, bool force)
900{
901 struct signal_struct *signal = p->signal;
902 struct task_struct *t;
903 sigset_t flush;
904
905 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
906 if (!(signal->flags & SIGNAL_GROUP_EXIT))
907 return sig == SIGKILL;
908 /*
909 * The process is in the middle of dying, nothing to do.
910 */
911 } else if (sig_kernel_stop(sig)) {
912 /*
913 * This is a stop signal. Remove SIGCONT from all queues.
914 */
915 siginitset(&flush, sigmask(SIGCONT));
916 flush_sigqueue_mask(&flush, &signal->shared_pending);
917 for_each_thread(p, t)
918 flush_sigqueue_mask(&flush, &t->pending);
919 } else if (sig == SIGCONT) {
920 unsigned int why;
921 /*
922 * Remove all stop signals from all queues, wake all threads.
923 */
924 siginitset(&flush, SIG_KERNEL_STOP_MASK);
925 flush_sigqueue_mask(&flush, &signal->shared_pending);
926 for_each_thread(p, t) {
927 flush_sigqueue_mask(&flush, &t->pending);
928 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
929 if (likely(!(t->ptrace & PT_SEIZED)))
930 wake_up_state(t, __TASK_STOPPED);
931 else
932 ptrace_trap_notify(t);
933 }
934
935 /*
936 * Notify the parent with CLD_CONTINUED if we were stopped.
937 *
938 * If we were in the middle of a group stop, we pretend it
939 * was already finished, and then continued. Since SIGCHLD
940 * doesn't queue we report only CLD_STOPPED, as if the next
941 * CLD_CONTINUED was dropped.
942 */
943 why = 0;
944 if (signal->flags & SIGNAL_STOP_STOPPED)
945 why |= SIGNAL_CLD_CONTINUED;
946 else if (signal->group_stop_count)
947 why |= SIGNAL_CLD_STOPPED;
948
949 if (why) {
950 /*
951 * The first thread which returns from do_signal_stop()
952 * will take ->siglock, notice SIGNAL_CLD_MASK, and
953 * notify its parent. See get_signal().
954 */
955 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
956 signal->group_stop_count = 0;
957 signal->group_exit_code = 0;
958 }
959 }
960
961 return !sig_ignored(p, sig, force);
962}
963
964/*
965 * Test if P wants to take SIG. After we've checked all threads with this,
966 * it's equivalent to finding no threads not blocking SIG. Any threads not
967 * blocking SIG were ruled out because they are not running and already
968 * have pending signals. Such threads will dequeue from the shared queue
969 * as soon as they're available, so putting the signal on the shared queue
970 * will be equivalent to sending it to one such thread.
971 */
972static inline bool wants_signal(int sig, struct task_struct *p)
973{
974 if (sigismember(&p->blocked, sig))
975 return false;
976
977 if (p->flags & PF_EXITING)
978 return false;
979
980 if (sig == SIGKILL)
981 return true;
982
983 if (task_is_stopped_or_traced(p))
984 return false;
985
986 return task_curr(p) || !signal_pending(p);
987}
988
989static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
990{
991 struct signal_struct *signal = p->signal;
992 struct task_struct *t;
993
994 /*
995 * Now find a thread we can wake up to take the signal off the queue.
996 *
997 * If the main thread wants the signal, it gets first crack.
998 * Probably the least surprising to the average bear.
999 */
1000 if (wants_signal(sig, p))
1001 t = p;
1002 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1003 /*
1004 * There is just one thread and it does not need to be woken.
1005 * It will dequeue unblocked signals before it runs again.
1006 */
1007 return;
1008 else {
1009 /*
1010 * Otherwise try to find a suitable thread.
1011 */
1012 t = signal->curr_target;
1013 while (!wants_signal(sig, t)) {
1014 t = next_thread(t);
1015 if (t == signal->curr_target)
1016 /*
1017 * No thread needs to be woken.
1018 * Any eligible threads will see
1019 * the signal in the queue soon.
1020 */
1021 return;
1022 }
1023 signal->curr_target = t;
1024 }
1025
1026 /*
1027 * Found a killable thread. If the signal will be fatal,
1028 * then start taking the whole group down immediately.
1029 */
1030 if (sig_fatal(p, sig) &&
1031 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1032 !sigismember(&t->real_blocked, sig) &&
1033 (sig == SIGKILL || !p->ptrace)) {
1034 /*
1035 * This signal will be fatal to the whole group.
1036 */
1037 if (!sig_kernel_coredump(sig)) {
1038 /*
1039 * Start a group exit and wake everybody up.
1040 * This way we don't have other threads
1041 * running and doing things after a slower
1042 * thread has the fatal signal pending.
1043 */
1044 signal->flags = SIGNAL_GROUP_EXIT;
1045 signal->group_exit_code = sig;
1046 signal->group_stop_count = 0;
1047 t = p;
1048 do {
1049 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1050 sigaddset(&t->pending.signal, SIGKILL);
1051 signal_wake_up(t, 1);
1052 } while_each_thread(p, t);
1053 return;
1054 }
1055 }
1056
1057 /*
1058 * The signal is already in the shared-pending queue.
1059 * Tell the chosen thread to wake up and dequeue it.
1060 */
1061 signal_wake_up(t, sig == SIGKILL);
1062 return;
1063}
1064
1065static inline bool legacy_queue(struct sigpending *signals, int sig)
1066{
1067 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1068}
1069
1070static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1071 enum pid_type type, bool force)
1072{
1073 struct sigpending *pending;
1074 struct sigqueue *q;
1075 int override_rlimit;
1076 int ret = 0, result;
1077
1078 assert_spin_locked(&t->sighand->siglock);
1079
1080 result = TRACE_SIGNAL_IGNORED;
1081 if (!prepare_signal(sig, t, force))
1082 goto ret;
1083
1084 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1085 /*
1086 * Short-circuit ignored signals and support queuing
1087 * exactly one non-rt signal, so that we can get more
1088 * detailed information about the cause of the signal.
1089 */
1090 result = TRACE_SIGNAL_ALREADY_PENDING;
1091 if (legacy_queue(pending, sig))
1092 goto ret;
1093
1094 result = TRACE_SIGNAL_DELIVERED;
1095 /*
1096 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1097 */
1098 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1099 goto out_set;
1100
1101 /*
1102 * Real-time signals must be queued if sent by sigqueue, or
1103 * some other real-time mechanism. It is implementation
1104 * defined whether kill() does so. We attempt to do so, on
1105 * the principle of least surprise, but since kill is not
1106 * allowed to fail with EAGAIN when low on memory we just
1107 * make sure at least one signal gets delivered and don't
1108 * pass on the info struct.
1109 */
1110 if (sig < SIGRTMIN)
1111 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1112 else
1113 override_rlimit = 0;
1114
1115 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit);
1116 if (q) {
1117 list_add_tail(&q->list, &pending->list);
1118 switch ((unsigned long) info) {
1119 case (unsigned long) SEND_SIG_NOINFO:
1120 clear_siginfo(&q->info);
1121 q->info.si_signo = sig;
1122 q->info.si_errno = 0;
1123 q->info.si_code = SI_USER;
1124 q->info.si_pid = task_tgid_nr_ns(current,
1125 task_active_pid_ns(t));
1126 rcu_read_lock();
1127 q->info.si_uid =
1128 from_kuid_munged(task_cred_xxx(t, user_ns),
1129 current_uid());
1130 rcu_read_unlock();
1131 break;
1132 case (unsigned long) SEND_SIG_PRIV:
1133 clear_siginfo(&q->info);
1134 q->info.si_signo = sig;
1135 q->info.si_errno = 0;
1136 q->info.si_code = SI_KERNEL;
1137 q->info.si_pid = 0;
1138 q->info.si_uid = 0;
1139 break;
1140 default:
1141 copy_siginfo(&q->info, info);
1142 break;
1143 }
1144 } else if (!is_si_special(info) &&
1145 sig >= SIGRTMIN && info->si_code != SI_USER) {
1146 /*
1147 * Queue overflow, abort. We may abort if the
1148 * signal was rt and sent by user using something
1149 * other than kill().
1150 */
1151 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1152 ret = -EAGAIN;
1153 goto ret;
1154 } else {
1155 /*
1156 * This is a silent loss of information. We still
1157 * send the signal, but the *info bits are lost.
1158 */
1159 result = TRACE_SIGNAL_LOSE_INFO;
1160 }
1161
1162out_set:
1163 signalfd_notify(t, sig);
1164 sigaddset(&pending->signal, sig);
1165
1166 /* Let multiprocess signals appear after on-going forks */
1167 if (type > PIDTYPE_TGID) {
1168 struct multiprocess_signals *delayed;
1169 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1170 sigset_t *signal = &delayed->signal;
1171 /* Can't queue both a stop and a continue signal */
1172 if (sig == SIGCONT)
1173 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1174 else if (sig_kernel_stop(sig))
1175 sigdelset(signal, SIGCONT);
1176 sigaddset(signal, sig);
1177 }
1178 }
1179
1180 complete_signal(sig, t, type);
1181ret:
1182 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1183 return ret;
1184}
1185
1186static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1187{
1188 bool ret = false;
1189 switch (siginfo_layout(info->si_signo, info->si_code)) {
1190 case SIL_KILL:
1191 case SIL_CHLD:
1192 case SIL_RT:
1193 ret = true;
1194 break;
1195 case SIL_TIMER:
1196 case SIL_POLL:
1197 case SIL_FAULT:
1198 case SIL_FAULT_MCEERR:
1199 case SIL_FAULT_BNDERR:
1200 case SIL_FAULT_PKUERR:
1201 case SIL_SYS:
1202 ret = false;
1203 break;
1204 }
1205 return ret;
1206}
1207
1208static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1209 enum pid_type type)
1210{
1211 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1212 bool force = false;
1213
1214 if (info == SEND_SIG_NOINFO) {
1215 /* Force if sent from an ancestor pid namespace */
1216 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1217 } else if (info == SEND_SIG_PRIV) {
1218 /* Don't ignore kernel generated signals */
1219 force = true;
1220 } else if (has_si_pid_and_uid(info)) {
1221 /* SIGKILL and SIGSTOP is special or has ids */
1222 struct user_namespace *t_user_ns;
1223
1224 rcu_read_lock();
1225 t_user_ns = task_cred_xxx(t, user_ns);
1226 if (current_user_ns() != t_user_ns) {
1227 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1228 info->si_uid = from_kuid_munged(t_user_ns, uid);
1229 }
1230 rcu_read_unlock();
1231
1232 /* A kernel generated signal? */
1233 force = (info->si_code == SI_KERNEL);
1234
1235 /* From an ancestor pid namespace? */
1236 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1237 info->si_pid = 0;
1238 force = true;
1239 }
1240 }
1241 return __send_signal(sig, info, t, type, force);
1242}
1243
1244static void print_fatal_signal(int signr)
1245{
1246 struct pt_regs *regs = signal_pt_regs();
1247 pr_info("potentially unexpected fatal signal %d.\n", signr);
1248
1249#if defined(__i386__) && !defined(__arch_um__)
1250 pr_info("code at %08lx: ", regs->ip);
1251 {
1252 int i;
1253 for (i = 0; i < 16; i++) {
1254 unsigned char insn;
1255
1256 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1257 break;
1258 pr_cont("%02x ", insn);
1259 }
1260 }
1261 pr_cont("\n");
1262#endif
1263 preempt_disable();
1264 show_regs(regs);
1265 preempt_enable();
1266}
1267
1268static int __init setup_print_fatal_signals(char *str)
1269{
1270 get_option (&str, &print_fatal_signals);
1271
1272 return 1;
1273}
1274
1275__setup("print-fatal-signals=", setup_print_fatal_signals);
1276
1277int
1278__group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1279{
1280 return send_signal(sig, info, p, PIDTYPE_TGID);
1281}
1282
1283int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1284 enum pid_type type)
1285{
1286 unsigned long flags;
1287 int ret = -ESRCH;
1288
1289 if (lock_task_sighand(p, &flags)) {
1290 ret = send_signal(sig, info, p, type);
1291 unlock_task_sighand(p, &flags);
1292 }
1293
1294 return ret;
1295}
1296
1297/*
1298 * Force a signal that the process can't ignore: if necessary
1299 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1300 *
1301 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1302 * since we do not want to have a signal handler that was blocked
1303 * be invoked when user space had explicitly blocked it.
1304 *
1305 * We don't want to have recursive SIGSEGV's etc, for example,
1306 * that is why we also clear SIGNAL_UNKILLABLE.
1307 */
1308static int
1309force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1310{
1311 unsigned long int flags;
1312 int ret, blocked, ignored;
1313 struct k_sigaction *action;
1314 int sig = info->si_signo;
1315
1316 spin_lock_irqsave(&t->sighand->siglock, flags);
1317 action = &t->sighand->action[sig-1];
1318 ignored = action->sa.sa_handler == SIG_IGN;
1319 blocked = sigismember(&t->blocked, sig);
1320 if (blocked || ignored) {
1321 action->sa.sa_handler = SIG_DFL;
1322 if (blocked) {
1323 sigdelset(&t->blocked, sig);
1324 recalc_sigpending_and_wake(t);
1325 }
1326 }
1327 /*
1328 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1329 * debugging to leave init killable.
1330 */
1331 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1332 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1333 ret = send_signal(sig, info, t, PIDTYPE_PID);
1334 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1335
1336 return ret;
1337}
1338
1339int force_sig_info(struct kernel_siginfo *info)
1340{
1341 return force_sig_info_to_task(info, current);
1342}
1343
1344/*
1345 * Nuke all other threads in the group.
1346 */
1347int zap_other_threads(struct task_struct *p)
1348{
1349 struct task_struct *t = p;
1350 int count = 0;
1351
1352 p->signal->group_stop_count = 0;
1353
1354 while_each_thread(p, t) {
1355 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1356 count++;
1357
1358 /* Don't bother with already dead threads */
1359 if (t->exit_state)
1360 continue;
1361 sigaddset(&t->pending.signal, SIGKILL);
1362 signal_wake_up(t, 1);
1363 }
1364
1365 return count;
1366}
1367
1368struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1369 unsigned long *flags)
1370{
1371 struct sighand_struct *sighand;
1372
1373 rcu_read_lock();
1374 for (;;) {
1375 sighand = rcu_dereference(tsk->sighand);
1376 if (unlikely(sighand == NULL))
1377 break;
1378
1379 /*
1380 * This sighand can be already freed and even reused, but
1381 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1382 * initializes ->siglock: this slab can't go away, it has
1383 * the same object type, ->siglock can't be reinitialized.
1384 *
1385 * We need to ensure that tsk->sighand is still the same
1386 * after we take the lock, we can race with de_thread() or
1387 * __exit_signal(). In the latter case the next iteration
1388 * must see ->sighand == NULL.
1389 */
1390 spin_lock_irqsave(&sighand->siglock, *flags);
1391 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1392 break;
1393 spin_unlock_irqrestore(&sighand->siglock, *flags);
1394 }
1395 rcu_read_unlock();
1396
1397 return sighand;
1398}
1399
1400/*
1401 * send signal info to all the members of a group
1402 */
1403int group_send_sig_info(int sig, struct kernel_siginfo *info,
1404 struct task_struct *p, enum pid_type type)
1405{
1406 int ret;
1407
1408 rcu_read_lock();
1409 ret = check_kill_permission(sig, info, p);
1410 rcu_read_unlock();
1411
1412 if (!ret && sig)
1413 ret = do_send_sig_info(sig, info, p, type);
1414
1415 return ret;
1416}
1417
1418/*
1419 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1420 * control characters do (^C, ^Z etc)
1421 * - the caller must hold at least a readlock on tasklist_lock
1422 */
1423int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1424{
1425 struct task_struct *p = NULL;
1426 int retval, success;
1427
1428 success = 0;
1429 retval = -ESRCH;
1430 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1431 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1432 success |= !err;
1433 retval = err;
1434 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1435 return success ? 0 : retval;
1436}
1437
1438int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1439{
1440 int error = -ESRCH;
1441 struct task_struct *p;
1442
1443 for (;;) {
1444 rcu_read_lock();
1445 p = pid_task(pid, PIDTYPE_PID);
1446 if (p)
1447 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1448 rcu_read_unlock();
1449 if (likely(!p || error != -ESRCH))
1450 return error;
1451
1452 /*
1453 * The task was unhashed in between, try again. If it
1454 * is dead, pid_task() will return NULL, if we race with
1455 * de_thread() it will find the new leader.
1456 */
1457 }
1458}
1459
1460static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1461{
1462 int error;
1463 rcu_read_lock();
1464 error = kill_pid_info(sig, info, find_vpid(pid));
1465 rcu_read_unlock();
1466 return error;
1467}
1468
1469static inline bool kill_as_cred_perm(const struct cred *cred,
1470 struct task_struct *target)
1471{
1472 const struct cred *pcred = __task_cred(target);
1473
1474 return uid_eq(cred->euid, pcred->suid) ||
1475 uid_eq(cred->euid, pcred->uid) ||
1476 uid_eq(cred->uid, pcred->suid) ||
1477 uid_eq(cred->uid, pcred->uid);
1478}
1479
1480/*
1481 * The usb asyncio usage of siginfo is wrong. The glibc support
1482 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1483 * AKA after the generic fields:
1484 * kernel_pid_t si_pid;
1485 * kernel_uid32_t si_uid;
1486 * sigval_t si_value;
1487 *
1488 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1489 * after the generic fields is:
1490 * void __user *si_addr;
1491 *
1492 * This is a practical problem when there is a 64bit big endian kernel
1493 * and a 32bit userspace. As the 32bit address will encoded in the low
1494 * 32bits of the pointer. Those low 32bits will be stored at higher
1495 * address than appear in a 32 bit pointer. So userspace will not
1496 * see the address it was expecting for it's completions.
1497 *
1498 * There is nothing in the encoding that can allow
1499 * copy_siginfo_to_user32 to detect this confusion of formats, so
1500 * handle this by requiring the caller of kill_pid_usb_asyncio to
1501 * notice when this situration takes place and to store the 32bit
1502 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1503 * parameter.
1504 */
1505int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1506 struct pid *pid, const struct cred *cred)
1507{
1508 struct kernel_siginfo info;
1509 struct task_struct *p;
1510 unsigned long flags;
1511 int ret = -EINVAL;
1512
1513 if (!valid_signal(sig))
1514 return ret;
1515
1516 clear_siginfo(&info);
1517 info.si_signo = sig;
1518 info.si_errno = errno;
1519 info.si_code = SI_ASYNCIO;
1520 *((sigval_t *)&info.si_pid) = addr;
1521
1522 rcu_read_lock();
1523 p = pid_task(pid, PIDTYPE_PID);
1524 if (!p) {
1525 ret = -ESRCH;
1526 goto out_unlock;
1527 }
1528 if (!kill_as_cred_perm(cred, p)) {
1529 ret = -EPERM;
1530 goto out_unlock;
1531 }
1532 ret = security_task_kill(p, &info, sig, cred);
1533 if (ret)
1534 goto out_unlock;
1535
1536 if (sig) {
1537 if (lock_task_sighand(p, &flags)) {
1538 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1539 unlock_task_sighand(p, &flags);
1540 } else
1541 ret = -ESRCH;
1542 }
1543out_unlock:
1544 rcu_read_unlock();
1545 return ret;
1546}
1547EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1548
1549/*
1550 * kill_something_info() interprets pid in interesting ways just like kill(2).
1551 *
1552 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1553 * is probably wrong. Should make it like BSD or SYSV.
1554 */
1555
1556static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1557{
1558 int ret;
1559
1560 if (pid > 0)
1561 return kill_proc_info(sig, info, pid);
1562
1563 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1564 if (pid == INT_MIN)
1565 return -ESRCH;
1566
1567 read_lock(&tasklist_lock);
1568 if (pid != -1) {
1569 ret = __kill_pgrp_info(sig, info,
1570 pid ? find_vpid(-pid) : task_pgrp(current));
1571 } else {
1572 int retval = 0, count = 0;
1573 struct task_struct * p;
1574
1575 for_each_process(p) {
1576 if (task_pid_vnr(p) > 1 &&
1577 !same_thread_group(p, current)) {
1578 int err = group_send_sig_info(sig, info, p,
1579 PIDTYPE_MAX);
1580 ++count;
1581 if (err != -EPERM)
1582 retval = err;
1583 }
1584 }
1585 ret = count ? retval : -ESRCH;
1586 }
1587 read_unlock(&tasklist_lock);
1588
1589 return ret;
1590}
1591
1592/*
1593 * These are for backward compatibility with the rest of the kernel source.
1594 */
1595
1596int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1597{
1598 /*
1599 * Make sure legacy kernel users don't send in bad values
1600 * (normal paths check this in check_kill_permission).
1601 */
1602 if (!valid_signal(sig))
1603 return -EINVAL;
1604
1605 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1606}
1607EXPORT_SYMBOL(send_sig_info);
1608
1609#define __si_special(priv) \
1610 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1611
1612int
1613send_sig(int sig, struct task_struct *p, int priv)
1614{
1615 return send_sig_info(sig, __si_special(priv), p);
1616}
1617EXPORT_SYMBOL(send_sig);
1618
1619void force_sig(int sig)
1620{
1621 struct kernel_siginfo info;
1622
1623 clear_siginfo(&info);
1624 info.si_signo = sig;
1625 info.si_errno = 0;
1626 info.si_code = SI_KERNEL;
1627 info.si_pid = 0;
1628 info.si_uid = 0;
1629 force_sig_info(&info);
1630}
1631EXPORT_SYMBOL(force_sig);
1632
1633/*
1634 * When things go south during signal handling, we
1635 * will force a SIGSEGV. And if the signal that caused
1636 * the problem was already a SIGSEGV, we'll want to
1637 * make sure we don't even try to deliver the signal..
1638 */
1639void force_sigsegv(int sig)
1640{
1641 struct task_struct *p = current;
1642
1643 if (sig == SIGSEGV) {
1644 unsigned long flags;
1645 spin_lock_irqsave(&p->sighand->siglock, flags);
1646 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1647 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1648 }
1649 force_sig(SIGSEGV);
1650}
1651
1652int force_sig_fault_to_task(int sig, int code, void __user *addr
1653 ___ARCH_SI_TRAPNO(int trapno)
1654 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1655 , struct task_struct *t)
1656{
1657 struct kernel_siginfo info;
1658
1659 clear_siginfo(&info);
1660 info.si_signo = sig;
1661 info.si_errno = 0;
1662 info.si_code = code;
1663 info.si_addr = addr;
1664#ifdef __ARCH_SI_TRAPNO
1665 info.si_trapno = trapno;
1666#endif
1667#ifdef __ia64__
1668 info.si_imm = imm;
1669 info.si_flags = flags;
1670 info.si_isr = isr;
1671#endif
1672 return force_sig_info_to_task(&info, t);
1673}
1674
1675int force_sig_fault(int sig, int code, void __user *addr
1676 ___ARCH_SI_TRAPNO(int trapno)
1677 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1678{
1679 return force_sig_fault_to_task(sig, code, addr
1680 ___ARCH_SI_TRAPNO(trapno)
1681 ___ARCH_SI_IA64(imm, flags, isr), current);
1682}
1683
1684int send_sig_fault(int sig, int code, void __user *addr
1685 ___ARCH_SI_TRAPNO(int trapno)
1686 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1687 , struct task_struct *t)
1688{
1689 struct kernel_siginfo info;
1690
1691 clear_siginfo(&info);
1692 info.si_signo = sig;
1693 info.si_errno = 0;
1694 info.si_code = code;
1695 info.si_addr = addr;
1696#ifdef __ARCH_SI_TRAPNO
1697 info.si_trapno = trapno;
1698#endif
1699#ifdef __ia64__
1700 info.si_imm = imm;
1701 info.si_flags = flags;
1702 info.si_isr = isr;
1703#endif
1704 return send_sig_info(info.si_signo, &info, t);
1705}
1706
1707int force_sig_mceerr(int code, void __user *addr, short lsb)
1708{
1709 struct kernel_siginfo info;
1710
1711 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1712 clear_siginfo(&info);
1713 info.si_signo = SIGBUS;
1714 info.si_errno = 0;
1715 info.si_code = code;
1716 info.si_addr = addr;
1717 info.si_addr_lsb = lsb;
1718 return force_sig_info(&info);
1719}
1720
1721int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1722{
1723 struct kernel_siginfo info;
1724
1725 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1726 clear_siginfo(&info);
1727 info.si_signo = SIGBUS;
1728 info.si_errno = 0;
1729 info.si_code = code;
1730 info.si_addr = addr;
1731 info.si_addr_lsb = lsb;
1732 return send_sig_info(info.si_signo, &info, t);
1733}
1734EXPORT_SYMBOL(send_sig_mceerr);
1735
1736int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1737{
1738 struct kernel_siginfo info;
1739
1740 clear_siginfo(&info);
1741 info.si_signo = SIGSEGV;
1742 info.si_errno = 0;
1743 info.si_code = SEGV_BNDERR;
1744 info.si_addr = addr;
1745 info.si_lower = lower;
1746 info.si_upper = upper;
1747 return force_sig_info(&info);
1748}
1749
1750#ifdef SEGV_PKUERR
1751int force_sig_pkuerr(void __user *addr, u32 pkey)
1752{
1753 struct kernel_siginfo info;
1754
1755 clear_siginfo(&info);
1756 info.si_signo = SIGSEGV;
1757 info.si_errno = 0;
1758 info.si_code = SEGV_PKUERR;
1759 info.si_addr = addr;
1760 info.si_pkey = pkey;
1761 return force_sig_info(&info);
1762}
1763#endif
1764
1765/* For the crazy architectures that include trap information in
1766 * the errno field, instead of an actual errno value.
1767 */
1768int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1769{
1770 struct kernel_siginfo info;
1771
1772 clear_siginfo(&info);
1773 info.si_signo = SIGTRAP;
1774 info.si_errno = errno;
1775 info.si_code = TRAP_HWBKPT;
1776 info.si_addr = addr;
1777 return force_sig_info(&info);
1778}
1779
1780int kill_pgrp(struct pid *pid, int sig, int priv)
1781{
1782 int ret;
1783
1784 read_lock(&tasklist_lock);
1785 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1786 read_unlock(&tasklist_lock);
1787
1788 return ret;
1789}
1790EXPORT_SYMBOL(kill_pgrp);
1791
1792int kill_pid(struct pid *pid, int sig, int priv)
1793{
1794 return kill_pid_info(sig, __si_special(priv), pid);
1795}
1796EXPORT_SYMBOL(kill_pid);
1797
1798/*
1799 * These functions support sending signals using preallocated sigqueue
1800 * structures. This is needed "because realtime applications cannot
1801 * afford to lose notifications of asynchronous events, like timer
1802 * expirations or I/O completions". In the case of POSIX Timers
1803 * we allocate the sigqueue structure from the timer_create. If this
1804 * allocation fails we are able to report the failure to the application
1805 * with an EAGAIN error.
1806 */
1807struct sigqueue *sigqueue_alloc(void)
1808{
1809 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1810
1811 if (q)
1812 q->flags |= SIGQUEUE_PREALLOC;
1813
1814 return q;
1815}
1816
1817void sigqueue_free(struct sigqueue *q)
1818{
1819 unsigned long flags;
1820 spinlock_t *lock = ¤t->sighand->siglock;
1821
1822 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1823 /*
1824 * We must hold ->siglock while testing q->list
1825 * to serialize with collect_signal() or with
1826 * __exit_signal()->flush_sigqueue().
1827 */
1828 spin_lock_irqsave(lock, flags);
1829 q->flags &= ~SIGQUEUE_PREALLOC;
1830 /*
1831 * If it is queued it will be freed when dequeued,
1832 * like the "regular" sigqueue.
1833 */
1834 if (!list_empty(&q->list))
1835 q = NULL;
1836 spin_unlock_irqrestore(lock, flags);
1837
1838 if (q)
1839 __sigqueue_free(q);
1840}
1841
1842int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1843{
1844 int sig = q->info.si_signo;
1845 struct sigpending *pending;
1846 struct task_struct *t;
1847 unsigned long flags;
1848 int ret, result;
1849
1850 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1851
1852 ret = -1;
1853 rcu_read_lock();
1854 t = pid_task(pid, type);
1855 if (!t || !likely(lock_task_sighand(t, &flags)))
1856 goto ret;
1857
1858 ret = 1; /* the signal is ignored */
1859 result = TRACE_SIGNAL_IGNORED;
1860 if (!prepare_signal(sig, t, false))
1861 goto out;
1862
1863 ret = 0;
1864 if (unlikely(!list_empty(&q->list))) {
1865 /*
1866 * If an SI_TIMER entry is already queue just increment
1867 * the overrun count.
1868 */
1869 BUG_ON(q->info.si_code != SI_TIMER);
1870 q->info.si_overrun++;
1871 result = TRACE_SIGNAL_ALREADY_PENDING;
1872 goto out;
1873 }
1874 q->info.si_overrun = 0;
1875
1876 signalfd_notify(t, sig);
1877 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1878 list_add_tail(&q->list, &pending->list);
1879 sigaddset(&pending->signal, sig);
1880 complete_signal(sig, t, type);
1881 result = TRACE_SIGNAL_DELIVERED;
1882out:
1883 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1884 unlock_task_sighand(t, &flags);
1885ret:
1886 rcu_read_unlock();
1887 return ret;
1888}
1889
1890static void do_notify_pidfd(struct task_struct *task)
1891{
1892 struct pid *pid;
1893
1894 WARN_ON(task->exit_state == 0);
1895 pid = task_pid(task);
1896 wake_up_all(&pid->wait_pidfd);
1897}
1898
1899/*
1900 * Let a parent know about the death of a child.
1901 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1902 *
1903 * Returns true if our parent ignored us and so we've switched to
1904 * self-reaping.
1905 */
1906bool do_notify_parent(struct task_struct *tsk, int sig)
1907{
1908 struct kernel_siginfo info;
1909 unsigned long flags;
1910 struct sighand_struct *psig;
1911 bool autoreap = false;
1912 u64 utime, stime;
1913
1914 BUG_ON(sig == -1);
1915
1916 /* do_notify_parent_cldstop should have been called instead. */
1917 BUG_ON(task_is_stopped_or_traced(tsk));
1918
1919 BUG_ON(!tsk->ptrace &&
1920 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1921
1922 /* Wake up all pidfd waiters */
1923 do_notify_pidfd(tsk);
1924
1925 if (sig != SIGCHLD) {
1926 /*
1927 * This is only possible if parent == real_parent.
1928 * Check if it has changed security domain.
1929 */
1930 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
1931 sig = SIGCHLD;
1932 }
1933
1934 clear_siginfo(&info);
1935 info.si_signo = sig;
1936 info.si_errno = 0;
1937 /*
1938 * We are under tasklist_lock here so our parent is tied to
1939 * us and cannot change.
1940 *
1941 * task_active_pid_ns will always return the same pid namespace
1942 * until a task passes through release_task.
1943 *
1944 * write_lock() currently calls preempt_disable() which is the
1945 * same as rcu_read_lock(), but according to Oleg, this is not
1946 * correct to rely on this
1947 */
1948 rcu_read_lock();
1949 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1950 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1951 task_uid(tsk));
1952 rcu_read_unlock();
1953
1954 task_cputime(tsk, &utime, &stime);
1955 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1956 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1957
1958 info.si_status = tsk->exit_code & 0x7f;
1959 if (tsk->exit_code & 0x80)
1960 info.si_code = CLD_DUMPED;
1961 else if (tsk->exit_code & 0x7f)
1962 info.si_code = CLD_KILLED;
1963 else {
1964 info.si_code = CLD_EXITED;
1965 info.si_status = tsk->exit_code >> 8;
1966 }
1967
1968 psig = tsk->parent->sighand;
1969 spin_lock_irqsave(&psig->siglock, flags);
1970 if (!tsk->ptrace && sig == SIGCHLD &&
1971 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1972 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1973 /*
1974 * We are exiting and our parent doesn't care. POSIX.1
1975 * defines special semantics for setting SIGCHLD to SIG_IGN
1976 * or setting the SA_NOCLDWAIT flag: we should be reaped
1977 * automatically and not left for our parent's wait4 call.
1978 * Rather than having the parent do it as a magic kind of
1979 * signal handler, we just set this to tell do_exit that we
1980 * can be cleaned up without becoming a zombie. Note that
1981 * we still call __wake_up_parent in this case, because a
1982 * blocked sys_wait4 might now return -ECHILD.
1983 *
1984 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1985 * is implementation-defined: we do (if you don't want
1986 * it, just use SIG_IGN instead).
1987 */
1988 autoreap = true;
1989 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1990 sig = 0;
1991 }
1992 /*
1993 * Send with __send_signal as si_pid and si_uid are in the
1994 * parent's namespaces.
1995 */
1996 if (valid_signal(sig) && sig)
1997 __send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
1998 __wake_up_parent(tsk, tsk->parent);
1999 spin_unlock_irqrestore(&psig->siglock, flags);
2000
2001 return autoreap;
2002}
2003
2004/**
2005 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2006 * @tsk: task reporting the state change
2007 * @for_ptracer: the notification is for ptracer
2008 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2009 *
2010 * Notify @tsk's parent that the stopped/continued state has changed. If
2011 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2012 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2013 *
2014 * CONTEXT:
2015 * Must be called with tasklist_lock at least read locked.
2016 */
2017static void do_notify_parent_cldstop(struct task_struct *tsk,
2018 bool for_ptracer, int why)
2019{
2020 struct kernel_siginfo info;
2021 unsigned long flags;
2022 struct task_struct *parent;
2023 struct sighand_struct *sighand;
2024 u64 utime, stime;
2025
2026 if (for_ptracer) {
2027 parent = tsk->parent;
2028 } else {
2029 tsk = tsk->group_leader;
2030 parent = tsk->real_parent;
2031 }
2032
2033 clear_siginfo(&info);
2034 info.si_signo = SIGCHLD;
2035 info.si_errno = 0;
2036 /*
2037 * see comment in do_notify_parent() about the following 4 lines
2038 */
2039 rcu_read_lock();
2040 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2041 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2042 rcu_read_unlock();
2043
2044 task_cputime(tsk, &utime, &stime);
2045 info.si_utime = nsec_to_clock_t(utime);
2046 info.si_stime = nsec_to_clock_t(stime);
2047
2048 info.si_code = why;
2049 switch (why) {
2050 case CLD_CONTINUED:
2051 info.si_status = SIGCONT;
2052 break;
2053 case CLD_STOPPED:
2054 info.si_status = tsk->signal->group_exit_code & 0x7f;
2055 break;
2056 case CLD_TRAPPED:
2057 info.si_status = tsk->exit_code & 0x7f;
2058 break;
2059 default:
2060 BUG();
2061 }
2062
2063 sighand = parent->sighand;
2064 spin_lock_irqsave(&sighand->siglock, flags);
2065 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2066 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2067 __group_send_sig_info(SIGCHLD, &info, parent);
2068 /*
2069 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2070 */
2071 __wake_up_parent(tsk, parent);
2072 spin_unlock_irqrestore(&sighand->siglock, flags);
2073}
2074
2075static inline bool may_ptrace_stop(void)
2076{
2077 if (!likely(current->ptrace))
2078 return false;
2079 /*
2080 * Are we in the middle of do_coredump?
2081 * If so and our tracer is also part of the coredump stopping
2082 * is a deadlock situation, and pointless because our tracer
2083 * is dead so don't allow us to stop.
2084 * If SIGKILL was already sent before the caller unlocked
2085 * ->siglock we must see ->core_state != NULL. Otherwise it
2086 * is safe to enter schedule().
2087 *
2088 * This is almost outdated, a task with the pending SIGKILL can't
2089 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2090 * after SIGKILL was already dequeued.
2091 */
2092 if (unlikely(current->mm->core_state) &&
2093 unlikely(current->mm == current->parent->mm))
2094 return false;
2095
2096 return true;
2097}
2098
2099/*
2100 * Return non-zero if there is a SIGKILL that should be waking us up.
2101 * Called with the siglock held.
2102 */
2103static bool sigkill_pending(struct task_struct *tsk)
2104{
2105 return sigismember(&tsk->pending.signal, SIGKILL) ||
2106 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2107}
2108
2109/*
2110 * This must be called with current->sighand->siglock held.
2111 *
2112 * This should be the path for all ptrace stops.
2113 * We always set current->last_siginfo while stopped here.
2114 * That makes it a way to test a stopped process for
2115 * being ptrace-stopped vs being job-control-stopped.
2116 *
2117 * If we actually decide not to stop at all because the tracer
2118 * is gone, we keep current->exit_code unless clear_code.
2119 */
2120static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2121 __releases(¤t->sighand->siglock)
2122 __acquires(¤t->sighand->siglock)
2123{
2124 bool gstop_done = false;
2125
2126 if (arch_ptrace_stop_needed(exit_code, info)) {
2127 /*
2128 * The arch code has something special to do before a
2129 * ptrace stop. This is allowed to block, e.g. for faults
2130 * on user stack pages. We can't keep the siglock while
2131 * calling arch_ptrace_stop, so we must release it now.
2132 * To preserve proper semantics, we must do this before
2133 * any signal bookkeeping like checking group_stop_count.
2134 * Meanwhile, a SIGKILL could come in before we retake the
2135 * siglock. That must prevent us from sleeping in TASK_TRACED.
2136 * So after regaining the lock, we must check for SIGKILL.
2137 */
2138 spin_unlock_irq(¤t->sighand->siglock);
2139 arch_ptrace_stop(exit_code, info);
2140 spin_lock_irq(¤t->sighand->siglock);
2141 if (sigkill_pending(current))
2142 return;
2143 }
2144
2145 set_special_state(TASK_TRACED);
2146
2147 /*
2148 * We're committing to trapping. TRACED should be visible before
2149 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2150 * Also, transition to TRACED and updates to ->jobctl should be
2151 * atomic with respect to siglock and should be done after the arch
2152 * hook as siglock is released and regrabbed across it.
2153 *
2154 * TRACER TRACEE
2155 *
2156 * ptrace_attach()
2157 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2158 * do_wait()
2159 * set_current_state() smp_wmb();
2160 * ptrace_do_wait()
2161 * wait_task_stopped()
2162 * task_stopped_code()
2163 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2164 */
2165 smp_wmb();
2166
2167 current->last_siginfo = info;
2168 current->exit_code = exit_code;
2169
2170 /*
2171 * If @why is CLD_STOPPED, we're trapping to participate in a group
2172 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2173 * across siglock relocks since INTERRUPT was scheduled, PENDING
2174 * could be clear now. We act as if SIGCONT is received after
2175 * TASK_TRACED is entered - ignore it.
2176 */
2177 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2178 gstop_done = task_participate_group_stop(current);
2179
2180 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2181 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2182 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2183 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2184
2185 /* entering a trap, clear TRAPPING */
2186 task_clear_jobctl_trapping(current);
2187
2188 spin_unlock_irq(¤t->sighand->siglock);
2189 read_lock(&tasklist_lock);
2190 if (may_ptrace_stop()) {
2191 /*
2192 * Notify parents of the stop.
2193 *
2194 * While ptraced, there are two parents - the ptracer and
2195 * the real_parent of the group_leader. The ptracer should
2196 * know about every stop while the real parent is only
2197 * interested in the completion of group stop. The states
2198 * for the two don't interact with each other. Notify
2199 * separately unless they're gonna be duplicates.
2200 */
2201 do_notify_parent_cldstop(current, true, why);
2202 if (gstop_done && ptrace_reparented(current))
2203 do_notify_parent_cldstop(current, false, why);
2204
2205 /*
2206 * Don't want to allow preemption here, because
2207 * sys_ptrace() needs this task to be inactive.
2208 *
2209 * XXX: implement read_unlock_no_resched().
2210 */
2211 preempt_disable();
2212 read_unlock(&tasklist_lock);
2213 cgroup_enter_frozen();
2214 preempt_enable_no_resched();
2215 freezable_schedule();
2216 cgroup_leave_frozen(true);
2217 } else {
2218 /*
2219 * By the time we got the lock, our tracer went away.
2220 * Don't drop the lock yet, another tracer may come.
2221 *
2222 * If @gstop_done, the ptracer went away between group stop
2223 * completion and here. During detach, it would have set
2224 * JOBCTL_STOP_PENDING on us and we'll re-enter
2225 * TASK_STOPPED in do_signal_stop() on return, so notifying
2226 * the real parent of the group stop completion is enough.
2227 */
2228 if (gstop_done)
2229 do_notify_parent_cldstop(current, false, why);
2230
2231 /* tasklist protects us from ptrace_freeze_traced() */
2232 __set_current_state(TASK_RUNNING);
2233 if (clear_code)
2234 current->exit_code = 0;
2235 read_unlock(&tasklist_lock);
2236 }
2237
2238 /*
2239 * We are back. Now reacquire the siglock before touching
2240 * last_siginfo, so that we are sure to have synchronized with
2241 * any signal-sending on another CPU that wants to examine it.
2242 */
2243 spin_lock_irq(¤t->sighand->siglock);
2244 current->last_siginfo = NULL;
2245
2246 /* LISTENING can be set only during STOP traps, clear it */
2247 current->jobctl &= ~JOBCTL_LISTENING;
2248
2249 /*
2250 * Queued signals ignored us while we were stopped for tracing.
2251 * So check for any that we should take before resuming user mode.
2252 * This sets TIF_SIGPENDING, but never clears it.
2253 */
2254 recalc_sigpending_tsk(current);
2255}
2256
2257static void ptrace_do_notify(int signr, int exit_code, int why)
2258{
2259 kernel_siginfo_t info;
2260
2261 clear_siginfo(&info);
2262 info.si_signo = signr;
2263 info.si_code = exit_code;
2264 info.si_pid = task_pid_vnr(current);
2265 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2266
2267 /* Let the debugger run. */
2268 ptrace_stop(exit_code, why, 1, &info);
2269}
2270
2271void ptrace_notify(int exit_code)
2272{
2273 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2274 if (unlikely(current->task_works))
2275 task_work_run();
2276
2277 spin_lock_irq(¤t->sighand->siglock);
2278 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2279 spin_unlock_irq(¤t->sighand->siglock);
2280}
2281
2282/**
2283 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2284 * @signr: signr causing group stop if initiating
2285 *
2286 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2287 * and participate in it. If already set, participate in the existing
2288 * group stop. If participated in a group stop (and thus slept), %true is
2289 * returned with siglock released.
2290 *
2291 * If ptraced, this function doesn't handle stop itself. Instead,
2292 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2293 * untouched. The caller must ensure that INTERRUPT trap handling takes
2294 * places afterwards.
2295 *
2296 * CONTEXT:
2297 * Must be called with @current->sighand->siglock held, which is released
2298 * on %true return.
2299 *
2300 * RETURNS:
2301 * %false if group stop is already cancelled or ptrace trap is scheduled.
2302 * %true if participated in group stop.
2303 */
2304static bool do_signal_stop(int signr)
2305 __releases(¤t->sighand->siglock)
2306{
2307 struct signal_struct *sig = current->signal;
2308
2309 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2310 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2311 struct task_struct *t;
2312
2313 /* signr will be recorded in task->jobctl for retries */
2314 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2315
2316 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2317 unlikely(signal_group_exit(sig)))
2318 return false;
2319 /*
2320 * There is no group stop already in progress. We must
2321 * initiate one now.
2322 *
2323 * While ptraced, a task may be resumed while group stop is
2324 * still in effect and then receive a stop signal and
2325 * initiate another group stop. This deviates from the
2326 * usual behavior as two consecutive stop signals can't
2327 * cause two group stops when !ptraced. That is why we
2328 * also check !task_is_stopped(t) below.
2329 *
2330 * The condition can be distinguished by testing whether
2331 * SIGNAL_STOP_STOPPED is already set. Don't generate
2332 * group_exit_code in such case.
2333 *
2334 * This is not necessary for SIGNAL_STOP_CONTINUED because
2335 * an intervening stop signal is required to cause two
2336 * continued events regardless of ptrace.
2337 */
2338 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2339 sig->group_exit_code = signr;
2340
2341 sig->group_stop_count = 0;
2342
2343 if (task_set_jobctl_pending(current, signr | gstop))
2344 sig->group_stop_count++;
2345
2346 t = current;
2347 while_each_thread(current, t) {
2348 /*
2349 * Setting state to TASK_STOPPED for a group
2350 * stop is always done with the siglock held,
2351 * so this check has no races.
2352 */
2353 if (!task_is_stopped(t) &&
2354 task_set_jobctl_pending(t, signr | gstop)) {
2355 sig->group_stop_count++;
2356 if (likely(!(t->ptrace & PT_SEIZED)))
2357 signal_wake_up(t, 0);
2358 else
2359 ptrace_trap_notify(t);
2360 }
2361 }
2362 }
2363
2364 if (likely(!current->ptrace)) {
2365 int notify = 0;
2366
2367 /*
2368 * If there are no other threads in the group, or if there
2369 * is a group stop in progress and we are the last to stop,
2370 * report to the parent.
2371 */
2372 if (task_participate_group_stop(current))
2373 notify = CLD_STOPPED;
2374
2375 set_special_state(TASK_STOPPED);
2376 spin_unlock_irq(¤t->sighand->siglock);
2377
2378 /*
2379 * Notify the parent of the group stop completion. Because
2380 * we're not holding either the siglock or tasklist_lock
2381 * here, ptracer may attach inbetween; however, this is for
2382 * group stop and should always be delivered to the real
2383 * parent of the group leader. The new ptracer will get
2384 * its notification when this task transitions into
2385 * TASK_TRACED.
2386 */
2387 if (notify) {
2388 read_lock(&tasklist_lock);
2389 do_notify_parent_cldstop(current, false, notify);
2390 read_unlock(&tasklist_lock);
2391 }
2392
2393 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2394 cgroup_enter_frozen();
2395 freezable_schedule();
2396 return true;
2397 } else {
2398 /*
2399 * While ptraced, group stop is handled by STOP trap.
2400 * Schedule it and let the caller deal with it.
2401 */
2402 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2403 return false;
2404 }
2405}
2406
2407/**
2408 * do_jobctl_trap - take care of ptrace jobctl traps
2409 *
2410 * When PT_SEIZED, it's used for both group stop and explicit
2411 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2412 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2413 * the stop signal; otherwise, %SIGTRAP.
2414 *
2415 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2416 * number as exit_code and no siginfo.
2417 *
2418 * CONTEXT:
2419 * Must be called with @current->sighand->siglock held, which may be
2420 * released and re-acquired before returning with intervening sleep.
2421 */
2422static void do_jobctl_trap(void)
2423{
2424 struct signal_struct *signal = current->signal;
2425 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2426
2427 if (current->ptrace & PT_SEIZED) {
2428 if (!signal->group_stop_count &&
2429 !(signal->flags & SIGNAL_STOP_STOPPED))
2430 signr = SIGTRAP;
2431 WARN_ON_ONCE(!signr);
2432 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2433 CLD_STOPPED);
2434 } else {
2435 WARN_ON_ONCE(!signr);
2436 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2437 current->exit_code = 0;
2438 }
2439}
2440
2441/**
2442 * do_freezer_trap - handle the freezer jobctl trap
2443 *
2444 * Puts the task into frozen state, if only the task is not about to quit.
2445 * In this case it drops JOBCTL_TRAP_FREEZE.
2446 *
2447 * CONTEXT:
2448 * Must be called with @current->sighand->siglock held,
2449 * which is always released before returning.
2450 */
2451static void do_freezer_trap(void)
2452 __releases(¤t->sighand->siglock)
2453{
2454 /*
2455 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2456 * let's make another loop to give it a chance to be handled.
2457 * In any case, we'll return back.
2458 */
2459 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2460 JOBCTL_TRAP_FREEZE) {
2461 spin_unlock_irq(¤t->sighand->siglock);
2462 return;
2463 }
2464
2465 /*
2466 * Now we're sure that there is no pending fatal signal and no
2467 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2468 * immediately (if there is a non-fatal signal pending), and
2469 * put the task into sleep.
2470 */
2471 __set_current_state(TASK_INTERRUPTIBLE);
2472 clear_thread_flag(TIF_SIGPENDING);
2473 spin_unlock_irq(¤t->sighand->siglock);
2474 cgroup_enter_frozen();
2475 freezable_schedule();
2476}
2477
2478static int ptrace_signal(int signr, kernel_siginfo_t *info)
2479{
2480 /*
2481 * We do not check sig_kernel_stop(signr) but set this marker
2482 * unconditionally because we do not know whether debugger will
2483 * change signr. This flag has no meaning unless we are going
2484 * to stop after return from ptrace_stop(). In this case it will
2485 * be checked in do_signal_stop(), we should only stop if it was
2486 * not cleared by SIGCONT while we were sleeping. See also the
2487 * comment in dequeue_signal().
2488 */
2489 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2490 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2491
2492 /* We're back. Did the debugger cancel the sig? */
2493 signr = current->exit_code;
2494 if (signr == 0)
2495 return signr;
2496
2497 current->exit_code = 0;
2498
2499 /*
2500 * Update the siginfo structure if the signal has
2501 * changed. If the debugger wanted something
2502 * specific in the siginfo structure then it should
2503 * have updated *info via PTRACE_SETSIGINFO.
2504 */
2505 if (signr != info->si_signo) {
2506 clear_siginfo(info);
2507 info->si_signo = signr;
2508 info->si_errno = 0;
2509 info->si_code = SI_USER;
2510 rcu_read_lock();
2511 info->si_pid = task_pid_vnr(current->parent);
2512 info->si_uid = from_kuid_munged(current_user_ns(),
2513 task_uid(current->parent));
2514 rcu_read_unlock();
2515 }
2516
2517 /* If the (new) signal is now blocked, requeue it. */
2518 if (sigismember(¤t->blocked, signr)) {
2519 send_signal(signr, info, current, PIDTYPE_PID);
2520 signr = 0;
2521 }
2522
2523 return signr;
2524}
2525
2526bool get_signal(struct ksignal *ksig)
2527{
2528 struct sighand_struct *sighand = current->sighand;
2529 struct signal_struct *signal = current->signal;
2530 int signr;
2531
2532 if (unlikely(uprobe_deny_signal()))
2533 return false;
2534
2535 /*
2536 * Do this once, we can't return to user-mode if freezing() == T.
2537 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2538 * thus do not need another check after return.
2539 */
2540 try_to_freeze();
2541
2542relock:
2543 spin_lock_irq(&sighand->siglock);
2544 /*
2545 * Make sure we can safely read ->jobctl() in task_work add. As Oleg
2546 * states:
2547 *
2548 * It pairs with mb (implied by cmpxchg) before READ_ONCE. So we
2549 * roughly have
2550 *
2551 * task_work_add: get_signal:
2552 * STORE(task->task_works, new_work); STORE(task->jobctl);
2553 * mb(); mb();
2554 * LOAD(task->jobctl); LOAD(task->task_works);
2555 *
2556 * and we can rely on STORE-MB-LOAD [ in task_work_add].
2557 */
2558 smp_store_mb(current->jobctl, current->jobctl & ~JOBCTL_TASK_WORK);
2559 if (unlikely(current->task_works)) {
2560 spin_unlock_irq(&sighand->siglock);
2561 task_work_run();
2562 goto relock;
2563 }
2564
2565 /*
2566 * Every stopped thread goes here after wakeup. Check to see if
2567 * we should notify the parent, prepare_signal(SIGCONT) encodes
2568 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2569 */
2570 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2571 int why;
2572
2573 if (signal->flags & SIGNAL_CLD_CONTINUED)
2574 why = CLD_CONTINUED;
2575 else
2576 why = CLD_STOPPED;
2577
2578 signal->flags &= ~SIGNAL_CLD_MASK;
2579
2580 spin_unlock_irq(&sighand->siglock);
2581
2582 /*
2583 * Notify the parent that we're continuing. This event is
2584 * always per-process and doesn't make whole lot of sense
2585 * for ptracers, who shouldn't consume the state via
2586 * wait(2) either, but, for backward compatibility, notify
2587 * the ptracer of the group leader too unless it's gonna be
2588 * a duplicate.
2589 */
2590 read_lock(&tasklist_lock);
2591 do_notify_parent_cldstop(current, false, why);
2592
2593 if (ptrace_reparented(current->group_leader))
2594 do_notify_parent_cldstop(current->group_leader,
2595 true, why);
2596 read_unlock(&tasklist_lock);
2597
2598 goto relock;
2599 }
2600
2601 /* Has this task already been marked for death? */
2602 if (signal_group_exit(signal)) {
2603 ksig->info.si_signo = signr = SIGKILL;
2604 sigdelset(¤t->pending.signal, SIGKILL);
2605 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2606 &sighand->action[SIGKILL - 1]);
2607 recalc_sigpending();
2608 goto fatal;
2609 }
2610
2611 for (;;) {
2612 struct k_sigaction *ka;
2613
2614 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2615 do_signal_stop(0))
2616 goto relock;
2617
2618 if (unlikely(current->jobctl &
2619 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2620 if (current->jobctl & JOBCTL_TRAP_MASK) {
2621 do_jobctl_trap();
2622 spin_unlock_irq(&sighand->siglock);
2623 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2624 do_freezer_trap();
2625
2626 goto relock;
2627 }
2628
2629 /*
2630 * If the task is leaving the frozen state, let's update
2631 * cgroup counters and reset the frozen bit.
2632 */
2633 if (unlikely(cgroup_task_frozen(current))) {
2634 spin_unlock_irq(&sighand->siglock);
2635 cgroup_leave_frozen(false);
2636 goto relock;
2637 }
2638
2639 /*
2640 * Signals generated by the execution of an instruction
2641 * need to be delivered before any other pending signals
2642 * so that the instruction pointer in the signal stack
2643 * frame points to the faulting instruction.
2644 */
2645 signr = dequeue_synchronous_signal(&ksig->info);
2646 if (!signr)
2647 signr = dequeue_signal(current, ¤t->blocked, &ksig->info);
2648
2649 if (!signr)
2650 break; /* will return 0 */
2651
2652 if (unlikely(current->ptrace) && signr != SIGKILL) {
2653 signr = ptrace_signal(signr, &ksig->info);
2654 if (!signr)
2655 continue;
2656 }
2657
2658 ka = &sighand->action[signr-1];
2659
2660 /* Trace actually delivered signals. */
2661 trace_signal_deliver(signr, &ksig->info, ka);
2662
2663 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2664 continue;
2665 if (ka->sa.sa_handler != SIG_DFL) {
2666 /* Run the handler. */
2667 ksig->ka = *ka;
2668
2669 if (ka->sa.sa_flags & SA_ONESHOT)
2670 ka->sa.sa_handler = SIG_DFL;
2671
2672 break; /* will return non-zero "signr" value */
2673 }
2674
2675 /*
2676 * Now we are doing the default action for this signal.
2677 */
2678 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2679 continue;
2680
2681 /*
2682 * Global init gets no signals it doesn't want.
2683 * Container-init gets no signals it doesn't want from same
2684 * container.
2685 *
2686 * Note that if global/container-init sees a sig_kernel_only()
2687 * signal here, the signal must have been generated internally
2688 * or must have come from an ancestor namespace. In either
2689 * case, the signal cannot be dropped.
2690 */
2691 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2692 !sig_kernel_only(signr))
2693 continue;
2694
2695 if (sig_kernel_stop(signr)) {
2696 /*
2697 * The default action is to stop all threads in
2698 * the thread group. The job control signals
2699 * do nothing in an orphaned pgrp, but SIGSTOP
2700 * always works. Note that siglock needs to be
2701 * dropped during the call to is_orphaned_pgrp()
2702 * because of lock ordering with tasklist_lock.
2703 * This allows an intervening SIGCONT to be posted.
2704 * We need to check for that and bail out if necessary.
2705 */
2706 if (signr != SIGSTOP) {
2707 spin_unlock_irq(&sighand->siglock);
2708
2709 /* signals can be posted during this window */
2710
2711 if (is_current_pgrp_orphaned())
2712 goto relock;
2713
2714 spin_lock_irq(&sighand->siglock);
2715 }
2716
2717 if (likely(do_signal_stop(ksig->info.si_signo))) {
2718 /* It released the siglock. */
2719 goto relock;
2720 }
2721
2722 /*
2723 * We didn't actually stop, due to a race
2724 * with SIGCONT or something like that.
2725 */
2726 continue;
2727 }
2728
2729 fatal:
2730 spin_unlock_irq(&sighand->siglock);
2731 if (unlikely(cgroup_task_frozen(current)))
2732 cgroup_leave_frozen(true);
2733
2734 /*
2735 * Anything else is fatal, maybe with a core dump.
2736 */
2737 current->flags |= PF_SIGNALED;
2738
2739 if (sig_kernel_coredump(signr)) {
2740 if (print_fatal_signals)
2741 print_fatal_signal(ksig->info.si_signo);
2742 proc_coredump_connector(current);
2743 /*
2744 * If it was able to dump core, this kills all
2745 * other threads in the group and synchronizes with
2746 * their demise. If we lost the race with another
2747 * thread getting here, it set group_exit_code
2748 * first and our do_group_exit call below will use
2749 * that value and ignore the one we pass it.
2750 */
2751 do_coredump(&ksig->info);
2752 }
2753
2754 /*
2755 * Death signals, no core dump.
2756 */
2757 do_group_exit(ksig->info.si_signo);
2758 /* NOTREACHED */
2759 }
2760 spin_unlock_irq(&sighand->siglock);
2761
2762 ksig->sig = signr;
2763 return ksig->sig > 0;
2764}
2765
2766/**
2767 * signal_delivered -
2768 * @ksig: kernel signal struct
2769 * @stepping: nonzero if debugger single-step or block-step in use
2770 *
2771 * This function should be called when a signal has successfully been
2772 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2773 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2774 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2775 */
2776static void signal_delivered(struct ksignal *ksig, int stepping)
2777{
2778 sigset_t blocked;
2779
2780 /* A signal was successfully delivered, and the
2781 saved sigmask was stored on the signal frame,
2782 and will be restored by sigreturn. So we can
2783 simply clear the restore sigmask flag. */
2784 clear_restore_sigmask();
2785
2786 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2787 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2788 sigaddset(&blocked, ksig->sig);
2789 set_current_blocked(&blocked);
2790 tracehook_signal_handler(stepping);
2791}
2792
2793void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2794{
2795 if (failed)
2796 force_sigsegv(ksig->sig);
2797 else
2798 signal_delivered(ksig, stepping);
2799}
2800
2801/*
2802 * It could be that complete_signal() picked us to notify about the
2803 * group-wide signal. Other threads should be notified now to take
2804 * the shared signals in @which since we will not.
2805 */
2806static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2807{
2808 sigset_t retarget;
2809 struct task_struct *t;
2810
2811 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2812 if (sigisemptyset(&retarget))
2813 return;
2814
2815 t = tsk;
2816 while_each_thread(tsk, t) {
2817 if (t->flags & PF_EXITING)
2818 continue;
2819
2820 if (!has_pending_signals(&retarget, &t->blocked))
2821 continue;
2822 /* Remove the signals this thread can handle. */
2823 sigandsets(&retarget, &retarget, &t->blocked);
2824
2825 if (!signal_pending(t))
2826 signal_wake_up(t, 0);
2827
2828 if (sigisemptyset(&retarget))
2829 break;
2830 }
2831}
2832
2833void exit_signals(struct task_struct *tsk)
2834{
2835 int group_stop = 0;
2836 sigset_t unblocked;
2837
2838 /*
2839 * @tsk is about to have PF_EXITING set - lock out users which
2840 * expect stable threadgroup.
2841 */
2842 cgroup_threadgroup_change_begin(tsk);
2843
2844 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2845 tsk->flags |= PF_EXITING;
2846 cgroup_threadgroup_change_end(tsk);
2847 return;
2848 }
2849
2850 spin_lock_irq(&tsk->sighand->siglock);
2851 /*
2852 * From now this task is not visible for group-wide signals,
2853 * see wants_signal(), do_signal_stop().
2854 */
2855 tsk->flags |= PF_EXITING;
2856
2857 cgroup_threadgroup_change_end(tsk);
2858
2859 if (!signal_pending(tsk))
2860 goto out;
2861
2862 unblocked = tsk->blocked;
2863 signotset(&unblocked);
2864 retarget_shared_pending(tsk, &unblocked);
2865
2866 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2867 task_participate_group_stop(tsk))
2868 group_stop = CLD_STOPPED;
2869out:
2870 spin_unlock_irq(&tsk->sighand->siglock);
2871
2872 /*
2873 * If group stop has completed, deliver the notification. This
2874 * should always go to the real parent of the group leader.
2875 */
2876 if (unlikely(group_stop)) {
2877 read_lock(&tasklist_lock);
2878 do_notify_parent_cldstop(tsk, false, group_stop);
2879 read_unlock(&tasklist_lock);
2880 }
2881}
2882
2883/*
2884 * System call entry points.
2885 */
2886
2887/**
2888 * sys_restart_syscall - restart a system call
2889 */
2890SYSCALL_DEFINE0(restart_syscall)
2891{
2892 struct restart_block *restart = ¤t->restart_block;
2893 return restart->fn(restart);
2894}
2895
2896long do_no_restart_syscall(struct restart_block *param)
2897{
2898 return -EINTR;
2899}
2900
2901static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2902{
2903 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2904 sigset_t newblocked;
2905 /* A set of now blocked but previously unblocked signals. */
2906 sigandnsets(&newblocked, newset, ¤t->blocked);
2907 retarget_shared_pending(tsk, &newblocked);
2908 }
2909 tsk->blocked = *newset;
2910 recalc_sigpending();
2911}
2912
2913/**
2914 * set_current_blocked - change current->blocked mask
2915 * @newset: new mask
2916 *
2917 * It is wrong to change ->blocked directly, this helper should be used
2918 * to ensure the process can't miss a shared signal we are going to block.
2919 */
2920void set_current_blocked(sigset_t *newset)
2921{
2922 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2923 __set_current_blocked(newset);
2924}
2925
2926void __set_current_blocked(const sigset_t *newset)
2927{
2928 struct task_struct *tsk = current;
2929
2930 /*
2931 * In case the signal mask hasn't changed, there is nothing we need
2932 * to do. The current->blocked shouldn't be modified by other task.
2933 */
2934 if (sigequalsets(&tsk->blocked, newset))
2935 return;
2936
2937 spin_lock_irq(&tsk->sighand->siglock);
2938 __set_task_blocked(tsk, newset);
2939 spin_unlock_irq(&tsk->sighand->siglock);
2940}
2941
2942/*
2943 * This is also useful for kernel threads that want to temporarily
2944 * (or permanently) block certain signals.
2945 *
2946 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2947 * interface happily blocks "unblockable" signals like SIGKILL
2948 * and friends.
2949 */
2950int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2951{
2952 struct task_struct *tsk = current;
2953 sigset_t newset;
2954
2955 /* Lockless, only current can change ->blocked, never from irq */
2956 if (oldset)
2957 *oldset = tsk->blocked;
2958
2959 switch (how) {
2960 case SIG_BLOCK:
2961 sigorsets(&newset, &tsk->blocked, set);
2962 break;
2963 case SIG_UNBLOCK:
2964 sigandnsets(&newset, &tsk->blocked, set);
2965 break;
2966 case SIG_SETMASK:
2967 newset = *set;
2968 break;
2969 default:
2970 return -EINVAL;
2971 }
2972
2973 __set_current_blocked(&newset);
2974 return 0;
2975}
2976EXPORT_SYMBOL(sigprocmask);
2977
2978/*
2979 * The api helps set app-provided sigmasks.
2980 *
2981 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
2982 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
2983 *
2984 * Note that it does set_restore_sigmask() in advance, so it must be always
2985 * paired with restore_saved_sigmask_unless() before return from syscall.
2986 */
2987int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
2988{
2989 sigset_t kmask;
2990
2991 if (!umask)
2992 return 0;
2993 if (sigsetsize != sizeof(sigset_t))
2994 return -EINVAL;
2995 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
2996 return -EFAULT;
2997
2998 set_restore_sigmask();
2999 current->saved_sigmask = current->blocked;
3000 set_current_blocked(&kmask);
3001
3002 return 0;
3003}
3004
3005#ifdef CONFIG_COMPAT
3006int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3007 size_t sigsetsize)
3008{
3009 sigset_t kmask;
3010
3011 if (!umask)
3012 return 0;
3013 if (sigsetsize != sizeof(compat_sigset_t))
3014 return -EINVAL;
3015 if (get_compat_sigset(&kmask, umask))
3016 return -EFAULT;
3017
3018 set_restore_sigmask();
3019 current->saved_sigmask = current->blocked;
3020 set_current_blocked(&kmask);
3021
3022 return 0;
3023}
3024#endif
3025
3026/**
3027 * sys_rt_sigprocmask - change the list of currently blocked signals
3028 * @how: whether to add, remove, or set signals
3029 * @nset: stores pending signals
3030 * @oset: previous value of signal mask if non-null
3031 * @sigsetsize: size of sigset_t type
3032 */
3033SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3034 sigset_t __user *, oset, size_t, sigsetsize)
3035{
3036 sigset_t old_set, new_set;
3037 int error;
3038
3039 /* XXX: Don't preclude handling different sized sigset_t's. */
3040 if (sigsetsize != sizeof(sigset_t))
3041 return -EINVAL;
3042
3043 old_set = current->blocked;
3044
3045 if (nset) {
3046 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3047 return -EFAULT;
3048 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3049
3050 error = sigprocmask(how, &new_set, NULL);
3051 if (error)
3052 return error;
3053 }
3054
3055 if (oset) {
3056 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3057 return -EFAULT;
3058 }
3059
3060 return 0;
3061}
3062
3063#ifdef CONFIG_COMPAT
3064COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3065 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3066{
3067 sigset_t old_set = current->blocked;
3068
3069 /* XXX: Don't preclude handling different sized sigset_t's. */
3070 if (sigsetsize != sizeof(sigset_t))
3071 return -EINVAL;
3072
3073 if (nset) {
3074 sigset_t new_set;
3075 int error;
3076 if (get_compat_sigset(&new_set, nset))
3077 return -EFAULT;
3078 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3079
3080 error = sigprocmask(how, &new_set, NULL);
3081 if (error)
3082 return error;
3083 }
3084 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3085}
3086#endif
3087
3088static void do_sigpending(sigset_t *set)
3089{
3090 spin_lock_irq(¤t->sighand->siglock);
3091 sigorsets(set, ¤t->pending.signal,
3092 ¤t->signal->shared_pending.signal);
3093 spin_unlock_irq(¤t->sighand->siglock);
3094
3095 /* Outside the lock because only this thread touches it. */
3096 sigandsets(set, ¤t->blocked, set);
3097}
3098
3099/**
3100 * sys_rt_sigpending - examine a pending signal that has been raised
3101 * while blocked
3102 * @uset: stores pending signals
3103 * @sigsetsize: size of sigset_t type or larger
3104 */
3105SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3106{
3107 sigset_t set;
3108
3109 if (sigsetsize > sizeof(*uset))
3110 return -EINVAL;
3111
3112 do_sigpending(&set);
3113
3114 if (copy_to_user(uset, &set, sigsetsize))
3115 return -EFAULT;
3116
3117 return 0;
3118}
3119
3120#ifdef CONFIG_COMPAT
3121COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3122 compat_size_t, sigsetsize)
3123{
3124 sigset_t set;
3125
3126 if (sigsetsize > sizeof(*uset))
3127 return -EINVAL;
3128
3129 do_sigpending(&set);
3130
3131 return put_compat_sigset(uset, &set, sigsetsize);
3132}
3133#endif
3134
3135static const struct {
3136 unsigned char limit, layout;
3137} sig_sicodes[] = {
3138 [SIGILL] = { NSIGILL, SIL_FAULT },
3139 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3140 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3141 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3142 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3143#if defined(SIGEMT)
3144 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3145#endif
3146 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3147 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3148 [SIGSYS] = { NSIGSYS, SIL_SYS },
3149};
3150
3151static bool known_siginfo_layout(unsigned sig, int si_code)
3152{
3153 if (si_code == SI_KERNEL)
3154 return true;
3155 else if ((si_code > SI_USER)) {
3156 if (sig_specific_sicodes(sig)) {
3157 if (si_code <= sig_sicodes[sig].limit)
3158 return true;
3159 }
3160 else if (si_code <= NSIGPOLL)
3161 return true;
3162 }
3163 else if (si_code >= SI_DETHREAD)
3164 return true;
3165 else if (si_code == SI_ASYNCNL)
3166 return true;
3167 return false;
3168}
3169
3170enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3171{
3172 enum siginfo_layout layout = SIL_KILL;
3173 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3174 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3175 (si_code <= sig_sicodes[sig].limit)) {
3176 layout = sig_sicodes[sig].layout;
3177 /* Handle the exceptions */
3178 if ((sig == SIGBUS) &&
3179 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3180 layout = SIL_FAULT_MCEERR;
3181 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3182 layout = SIL_FAULT_BNDERR;
3183#ifdef SEGV_PKUERR
3184 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3185 layout = SIL_FAULT_PKUERR;
3186#endif
3187 }
3188 else if (si_code <= NSIGPOLL)
3189 layout = SIL_POLL;
3190 } else {
3191 if (si_code == SI_TIMER)
3192 layout = SIL_TIMER;
3193 else if (si_code == SI_SIGIO)
3194 layout = SIL_POLL;
3195 else if (si_code < 0)
3196 layout = SIL_RT;
3197 }
3198 return layout;
3199}
3200
3201static inline char __user *si_expansion(const siginfo_t __user *info)
3202{
3203 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3204}
3205
3206int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3207{
3208 char __user *expansion = si_expansion(to);
3209 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3210 return -EFAULT;
3211 if (clear_user(expansion, SI_EXPANSION_SIZE))
3212 return -EFAULT;
3213 return 0;
3214}
3215
3216static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3217 const siginfo_t __user *from)
3218{
3219 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3220 char __user *expansion = si_expansion(from);
3221 char buf[SI_EXPANSION_SIZE];
3222 int i;
3223 /*
3224 * An unknown si_code might need more than
3225 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3226 * extra bytes are 0. This guarantees copy_siginfo_to_user
3227 * will return this data to userspace exactly.
3228 */
3229 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3230 return -EFAULT;
3231 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3232 if (buf[i] != 0)
3233 return -E2BIG;
3234 }
3235 }
3236 return 0;
3237}
3238
3239static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3240 const siginfo_t __user *from)
3241{
3242 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3243 return -EFAULT;
3244 to->si_signo = signo;
3245 return post_copy_siginfo_from_user(to, from);
3246}
3247
3248int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3249{
3250 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3251 return -EFAULT;
3252 return post_copy_siginfo_from_user(to, from);
3253}
3254
3255#ifdef CONFIG_COMPAT
3256/**
3257 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3258 * @to: compat siginfo destination
3259 * @from: kernel siginfo source
3260 *
3261 * Note: This function does not work properly for the SIGCHLD on x32, but
3262 * fortunately it doesn't have to. The only valid callers for this function are
3263 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3264 * The latter does not care because SIGCHLD will never cause a coredump.
3265 */
3266void copy_siginfo_to_external32(struct compat_siginfo *to,
3267 const struct kernel_siginfo *from)
3268{
3269 memset(to, 0, sizeof(*to));
3270
3271 to->si_signo = from->si_signo;
3272 to->si_errno = from->si_errno;
3273 to->si_code = from->si_code;
3274 switch(siginfo_layout(from->si_signo, from->si_code)) {
3275 case SIL_KILL:
3276 to->si_pid = from->si_pid;
3277 to->si_uid = from->si_uid;
3278 break;
3279 case SIL_TIMER:
3280 to->si_tid = from->si_tid;
3281 to->si_overrun = from->si_overrun;
3282 to->si_int = from->si_int;
3283 break;
3284 case SIL_POLL:
3285 to->si_band = from->si_band;
3286 to->si_fd = from->si_fd;
3287 break;
3288 case SIL_FAULT:
3289 to->si_addr = ptr_to_compat(from->si_addr);
3290#ifdef __ARCH_SI_TRAPNO
3291 to->si_trapno = from->si_trapno;
3292#endif
3293 break;
3294 case SIL_FAULT_MCEERR:
3295 to->si_addr = ptr_to_compat(from->si_addr);
3296#ifdef __ARCH_SI_TRAPNO
3297 to->si_trapno = from->si_trapno;
3298#endif
3299 to->si_addr_lsb = from->si_addr_lsb;
3300 break;
3301 case SIL_FAULT_BNDERR:
3302 to->si_addr = ptr_to_compat(from->si_addr);
3303#ifdef __ARCH_SI_TRAPNO
3304 to->si_trapno = from->si_trapno;
3305#endif
3306 to->si_lower = ptr_to_compat(from->si_lower);
3307 to->si_upper = ptr_to_compat(from->si_upper);
3308 break;
3309 case SIL_FAULT_PKUERR:
3310 to->si_addr = ptr_to_compat(from->si_addr);
3311#ifdef __ARCH_SI_TRAPNO
3312 to->si_trapno = from->si_trapno;
3313#endif
3314 to->si_pkey = from->si_pkey;
3315 break;
3316 case SIL_CHLD:
3317 to->si_pid = from->si_pid;
3318 to->si_uid = from->si_uid;
3319 to->si_status = from->si_status;
3320 to->si_utime = from->si_utime;
3321 to->si_stime = from->si_stime;
3322 break;
3323 case SIL_RT:
3324 to->si_pid = from->si_pid;
3325 to->si_uid = from->si_uid;
3326 to->si_int = from->si_int;
3327 break;
3328 case SIL_SYS:
3329 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3330 to->si_syscall = from->si_syscall;
3331 to->si_arch = from->si_arch;
3332 break;
3333 }
3334}
3335
3336int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3337 const struct kernel_siginfo *from)
3338{
3339 struct compat_siginfo new;
3340
3341 copy_siginfo_to_external32(&new, from);
3342 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3343 return -EFAULT;
3344 return 0;
3345}
3346
3347static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3348 const struct compat_siginfo *from)
3349{
3350 clear_siginfo(to);
3351 to->si_signo = from->si_signo;
3352 to->si_errno = from->si_errno;
3353 to->si_code = from->si_code;
3354 switch(siginfo_layout(from->si_signo, from->si_code)) {
3355 case SIL_KILL:
3356 to->si_pid = from->si_pid;
3357 to->si_uid = from->si_uid;
3358 break;
3359 case SIL_TIMER:
3360 to->si_tid = from->si_tid;
3361 to->si_overrun = from->si_overrun;
3362 to->si_int = from->si_int;
3363 break;
3364 case SIL_POLL:
3365 to->si_band = from->si_band;
3366 to->si_fd = from->si_fd;
3367 break;
3368 case SIL_FAULT:
3369 to->si_addr = compat_ptr(from->si_addr);
3370#ifdef __ARCH_SI_TRAPNO
3371 to->si_trapno = from->si_trapno;
3372#endif
3373 break;
3374 case SIL_FAULT_MCEERR:
3375 to->si_addr = compat_ptr(from->si_addr);
3376#ifdef __ARCH_SI_TRAPNO
3377 to->si_trapno = from->si_trapno;
3378#endif
3379 to->si_addr_lsb = from->si_addr_lsb;
3380 break;
3381 case SIL_FAULT_BNDERR:
3382 to->si_addr = compat_ptr(from->si_addr);
3383#ifdef __ARCH_SI_TRAPNO
3384 to->si_trapno = from->si_trapno;
3385#endif
3386 to->si_lower = compat_ptr(from->si_lower);
3387 to->si_upper = compat_ptr(from->si_upper);
3388 break;
3389 case SIL_FAULT_PKUERR:
3390 to->si_addr = compat_ptr(from->si_addr);
3391#ifdef __ARCH_SI_TRAPNO
3392 to->si_trapno = from->si_trapno;
3393#endif
3394 to->si_pkey = from->si_pkey;
3395 break;
3396 case SIL_CHLD:
3397 to->si_pid = from->si_pid;
3398 to->si_uid = from->si_uid;
3399 to->si_status = from->si_status;
3400#ifdef CONFIG_X86_X32_ABI
3401 if (in_x32_syscall()) {
3402 to->si_utime = from->_sifields._sigchld_x32._utime;
3403 to->si_stime = from->_sifields._sigchld_x32._stime;
3404 } else
3405#endif
3406 {
3407 to->si_utime = from->si_utime;
3408 to->si_stime = from->si_stime;
3409 }
3410 break;
3411 case SIL_RT:
3412 to->si_pid = from->si_pid;
3413 to->si_uid = from->si_uid;
3414 to->si_int = from->si_int;
3415 break;
3416 case SIL_SYS:
3417 to->si_call_addr = compat_ptr(from->si_call_addr);
3418 to->si_syscall = from->si_syscall;
3419 to->si_arch = from->si_arch;
3420 break;
3421 }
3422 return 0;
3423}
3424
3425static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3426 const struct compat_siginfo __user *ufrom)
3427{
3428 struct compat_siginfo from;
3429
3430 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3431 return -EFAULT;
3432
3433 from.si_signo = signo;
3434 return post_copy_siginfo_from_user32(to, &from);
3435}
3436
3437int copy_siginfo_from_user32(struct kernel_siginfo *to,
3438 const struct compat_siginfo __user *ufrom)
3439{
3440 struct compat_siginfo from;
3441
3442 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3443 return -EFAULT;
3444
3445 return post_copy_siginfo_from_user32(to, &from);
3446}
3447#endif /* CONFIG_COMPAT */
3448
3449/**
3450 * do_sigtimedwait - wait for queued signals specified in @which
3451 * @which: queued signals to wait for
3452 * @info: if non-null, the signal's siginfo is returned here
3453 * @ts: upper bound on process time suspension
3454 */
3455static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3456 const struct timespec64 *ts)
3457{
3458 ktime_t *to = NULL, timeout = KTIME_MAX;
3459 struct task_struct *tsk = current;
3460 sigset_t mask = *which;
3461 int sig, ret = 0;
3462
3463 if (ts) {
3464 if (!timespec64_valid(ts))
3465 return -EINVAL;
3466 timeout = timespec64_to_ktime(*ts);
3467 to = &timeout;
3468 }
3469
3470 /*
3471 * Invert the set of allowed signals to get those we want to block.
3472 */
3473 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3474 signotset(&mask);
3475
3476 spin_lock_irq(&tsk->sighand->siglock);
3477 sig = dequeue_signal(tsk, &mask, info);
3478 if (!sig && timeout) {
3479 /*
3480 * None ready, temporarily unblock those we're interested
3481 * while we are sleeping in so that we'll be awakened when
3482 * they arrive. Unblocking is always fine, we can avoid
3483 * set_current_blocked().
3484 */
3485 tsk->real_blocked = tsk->blocked;
3486 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3487 recalc_sigpending();
3488 spin_unlock_irq(&tsk->sighand->siglock);
3489
3490 __set_current_state(TASK_INTERRUPTIBLE);
3491 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3492 HRTIMER_MODE_REL);
3493 spin_lock_irq(&tsk->sighand->siglock);
3494 __set_task_blocked(tsk, &tsk->real_blocked);
3495 sigemptyset(&tsk->real_blocked);
3496 sig = dequeue_signal(tsk, &mask, info);
3497 }
3498 spin_unlock_irq(&tsk->sighand->siglock);
3499
3500 if (sig)
3501 return sig;
3502 return ret ? -EINTR : -EAGAIN;
3503}
3504
3505/**
3506 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3507 * in @uthese
3508 * @uthese: queued signals to wait for
3509 * @uinfo: if non-null, the signal's siginfo is returned here
3510 * @uts: upper bound on process time suspension
3511 * @sigsetsize: size of sigset_t type
3512 */
3513SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3514 siginfo_t __user *, uinfo,
3515 const struct __kernel_timespec __user *, uts,
3516 size_t, sigsetsize)
3517{
3518 sigset_t these;
3519 struct timespec64 ts;
3520 kernel_siginfo_t info;
3521 int ret;
3522
3523 /* XXX: Don't preclude handling different sized sigset_t's. */
3524 if (sigsetsize != sizeof(sigset_t))
3525 return -EINVAL;
3526
3527 if (copy_from_user(&these, uthese, sizeof(these)))
3528 return -EFAULT;
3529
3530 if (uts) {
3531 if (get_timespec64(&ts, uts))
3532 return -EFAULT;
3533 }
3534
3535 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3536
3537 if (ret > 0 && uinfo) {
3538 if (copy_siginfo_to_user(uinfo, &info))
3539 ret = -EFAULT;
3540 }
3541
3542 return ret;
3543}
3544
3545#ifdef CONFIG_COMPAT_32BIT_TIME
3546SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3547 siginfo_t __user *, uinfo,
3548 const struct old_timespec32 __user *, uts,
3549 size_t, sigsetsize)
3550{
3551 sigset_t these;
3552 struct timespec64 ts;
3553 kernel_siginfo_t info;
3554 int ret;
3555
3556 if (sigsetsize != sizeof(sigset_t))
3557 return -EINVAL;
3558
3559 if (copy_from_user(&these, uthese, sizeof(these)))
3560 return -EFAULT;
3561
3562 if (uts) {
3563 if (get_old_timespec32(&ts, uts))
3564 return -EFAULT;
3565 }
3566
3567 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3568
3569 if (ret > 0 && uinfo) {
3570 if (copy_siginfo_to_user(uinfo, &info))
3571 ret = -EFAULT;
3572 }
3573
3574 return ret;
3575}
3576#endif
3577
3578#ifdef CONFIG_COMPAT
3579COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3580 struct compat_siginfo __user *, uinfo,
3581 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3582{
3583 sigset_t s;
3584 struct timespec64 t;
3585 kernel_siginfo_t info;
3586 long ret;
3587
3588 if (sigsetsize != sizeof(sigset_t))
3589 return -EINVAL;
3590
3591 if (get_compat_sigset(&s, uthese))
3592 return -EFAULT;
3593
3594 if (uts) {
3595 if (get_timespec64(&t, uts))
3596 return -EFAULT;
3597 }
3598
3599 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3600
3601 if (ret > 0 && uinfo) {
3602 if (copy_siginfo_to_user32(uinfo, &info))
3603 ret = -EFAULT;
3604 }
3605
3606 return ret;
3607}
3608
3609#ifdef CONFIG_COMPAT_32BIT_TIME
3610COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3611 struct compat_siginfo __user *, uinfo,
3612 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3613{
3614 sigset_t s;
3615 struct timespec64 t;
3616 kernel_siginfo_t info;
3617 long ret;
3618
3619 if (sigsetsize != sizeof(sigset_t))
3620 return -EINVAL;
3621
3622 if (get_compat_sigset(&s, uthese))
3623 return -EFAULT;
3624
3625 if (uts) {
3626 if (get_old_timespec32(&t, uts))
3627 return -EFAULT;
3628 }
3629
3630 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3631
3632 if (ret > 0 && uinfo) {
3633 if (copy_siginfo_to_user32(uinfo, &info))
3634 ret = -EFAULT;
3635 }
3636
3637 return ret;
3638}
3639#endif
3640#endif
3641
3642static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3643{
3644 clear_siginfo(info);
3645 info->si_signo = sig;
3646 info->si_errno = 0;
3647 info->si_code = SI_USER;
3648 info->si_pid = task_tgid_vnr(current);
3649 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3650}
3651
3652/**
3653 * sys_kill - send a signal to a process
3654 * @pid: the PID of the process
3655 * @sig: signal to be sent
3656 */
3657SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3658{
3659 struct kernel_siginfo info;
3660
3661 prepare_kill_siginfo(sig, &info);
3662
3663 return kill_something_info(sig, &info, pid);
3664}
3665
3666/*
3667 * Verify that the signaler and signalee either are in the same pid namespace
3668 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3669 * namespace.
3670 */
3671static bool access_pidfd_pidns(struct pid *pid)
3672{
3673 struct pid_namespace *active = task_active_pid_ns(current);
3674 struct pid_namespace *p = ns_of_pid(pid);
3675
3676 for (;;) {
3677 if (!p)
3678 return false;
3679 if (p == active)
3680 break;
3681 p = p->parent;
3682 }
3683
3684 return true;
3685}
3686
3687static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo, siginfo_t *info)
3688{
3689#ifdef CONFIG_COMPAT
3690 /*
3691 * Avoid hooking up compat syscalls and instead handle necessary
3692 * conversions here. Note, this is a stop-gap measure and should not be
3693 * considered a generic solution.
3694 */
3695 if (in_compat_syscall())
3696 return copy_siginfo_from_user32(
3697 kinfo, (struct compat_siginfo __user *)info);
3698#endif
3699 return copy_siginfo_from_user(kinfo, info);
3700}
3701
3702static struct pid *pidfd_to_pid(const struct file *file)
3703{
3704 struct pid *pid;
3705
3706 pid = pidfd_pid(file);
3707 if (!IS_ERR(pid))
3708 return pid;
3709
3710 return tgid_pidfd_to_pid(file);
3711}
3712
3713/**
3714 * sys_pidfd_send_signal - Signal a process through a pidfd
3715 * @pidfd: file descriptor of the process
3716 * @sig: signal to send
3717 * @info: signal info
3718 * @flags: future flags
3719 *
3720 * The syscall currently only signals via PIDTYPE_PID which covers
3721 * kill(<positive-pid>, <signal>. It does not signal threads or process
3722 * groups.
3723 * In order to extend the syscall to threads and process groups the @flags
3724 * argument should be used. In essence, the @flags argument will determine
3725 * what is signaled and not the file descriptor itself. Put in other words,
3726 * grouping is a property of the flags argument not a property of the file
3727 * descriptor.
3728 *
3729 * Return: 0 on success, negative errno on failure
3730 */
3731SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3732 siginfo_t __user *, info, unsigned int, flags)
3733{
3734 int ret;
3735 struct fd f;
3736 struct pid *pid;
3737 kernel_siginfo_t kinfo;
3738
3739 /* Enforce flags be set to 0 until we add an extension. */
3740 if (flags)
3741 return -EINVAL;
3742
3743 f = fdget(pidfd);
3744 if (!f.file)
3745 return -EBADF;
3746
3747 /* Is this a pidfd? */
3748 pid = pidfd_to_pid(f.file);
3749 if (IS_ERR(pid)) {
3750 ret = PTR_ERR(pid);
3751 goto err;
3752 }
3753
3754 ret = -EINVAL;
3755 if (!access_pidfd_pidns(pid))
3756 goto err;
3757
3758 if (info) {
3759 ret = copy_siginfo_from_user_any(&kinfo, info);
3760 if (unlikely(ret))
3761 goto err;
3762
3763 ret = -EINVAL;
3764 if (unlikely(sig != kinfo.si_signo))
3765 goto err;
3766
3767 /* Only allow sending arbitrary signals to yourself. */
3768 ret = -EPERM;
3769 if ((task_pid(current) != pid) &&
3770 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3771 goto err;
3772 } else {
3773 prepare_kill_siginfo(sig, &kinfo);
3774 }
3775
3776 ret = kill_pid_info(sig, &kinfo, pid);
3777
3778err:
3779 fdput(f);
3780 return ret;
3781}
3782
3783static int
3784do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3785{
3786 struct task_struct *p;
3787 int error = -ESRCH;
3788
3789 rcu_read_lock();
3790 p = find_task_by_vpid(pid);
3791 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3792 error = check_kill_permission(sig, info, p);
3793 /*
3794 * The null signal is a permissions and process existence
3795 * probe. No signal is actually delivered.
3796 */
3797 if (!error && sig) {
3798 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3799 /*
3800 * If lock_task_sighand() failed we pretend the task
3801 * dies after receiving the signal. The window is tiny,
3802 * and the signal is private anyway.
3803 */
3804 if (unlikely(error == -ESRCH))
3805 error = 0;
3806 }
3807 }
3808 rcu_read_unlock();
3809
3810 return error;
3811}
3812
3813static int do_tkill(pid_t tgid, pid_t pid, int sig)
3814{
3815 struct kernel_siginfo info;
3816
3817 clear_siginfo(&info);
3818 info.si_signo = sig;
3819 info.si_errno = 0;
3820 info.si_code = SI_TKILL;
3821 info.si_pid = task_tgid_vnr(current);
3822 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3823
3824 return do_send_specific(tgid, pid, sig, &info);
3825}
3826
3827/**
3828 * sys_tgkill - send signal to one specific thread
3829 * @tgid: the thread group ID of the thread
3830 * @pid: the PID of the thread
3831 * @sig: signal to be sent
3832 *
3833 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3834 * exists but it's not belonging to the target process anymore. This
3835 * method solves the problem of threads exiting and PIDs getting reused.
3836 */
3837SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3838{
3839 /* This is only valid for single tasks */
3840 if (pid <= 0 || tgid <= 0)
3841 return -EINVAL;
3842
3843 return do_tkill(tgid, pid, sig);
3844}
3845
3846/**
3847 * sys_tkill - send signal to one specific task
3848 * @pid: the PID of the task
3849 * @sig: signal to be sent
3850 *
3851 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3852 */
3853SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3854{
3855 /* This is only valid for single tasks */
3856 if (pid <= 0)
3857 return -EINVAL;
3858
3859 return do_tkill(0, pid, sig);
3860}
3861
3862static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3863{
3864 /* Not even root can pretend to send signals from the kernel.
3865 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3866 */
3867 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3868 (task_pid_vnr(current) != pid))
3869 return -EPERM;
3870
3871 /* POSIX.1b doesn't mention process groups. */
3872 return kill_proc_info(sig, info, pid);
3873}
3874
3875/**
3876 * sys_rt_sigqueueinfo - send signal information to a signal
3877 * @pid: the PID of the thread
3878 * @sig: signal to be sent
3879 * @uinfo: signal info to be sent
3880 */
3881SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3882 siginfo_t __user *, uinfo)
3883{
3884 kernel_siginfo_t info;
3885 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3886 if (unlikely(ret))
3887 return ret;
3888 return do_rt_sigqueueinfo(pid, sig, &info);
3889}
3890
3891#ifdef CONFIG_COMPAT
3892COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3893 compat_pid_t, pid,
3894 int, sig,
3895 struct compat_siginfo __user *, uinfo)
3896{
3897 kernel_siginfo_t info;
3898 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3899 if (unlikely(ret))
3900 return ret;
3901 return do_rt_sigqueueinfo(pid, sig, &info);
3902}
3903#endif
3904
3905static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3906{
3907 /* This is only valid for single tasks */
3908 if (pid <= 0 || tgid <= 0)
3909 return -EINVAL;
3910
3911 /* Not even root can pretend to send signals from the kernel.
3912 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3913 */
3914 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3915 (task_pid_vnr(current) != pid))
3916 return -EPERM;
3917
3918 return do_send_specific(tgid, pid, sig, info);
3919}
3920
3921SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3922 siginfo_t __user *, uinfo)
3923{
3924 kernel_siginfo_t info;
3925 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3926 if (unlikely(ret))
3927 return ret;
3928 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3929}
3930
3931#ifdef CONFIG_COMPAT
3932COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3933 compat_pid_t, tgid,
3934 compat_pid_t, pid,
3935 int, sig,
3936 struct compat_siginfo __user *, uinfo)
3937{
3938 kernel_siginfo_t info;
3939 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3940 if (unlikely(ret))
3941 return ret;
3942 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3943}
3944#endif
3945
3946/*
3947 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
3948 */
3949void kernel_sigaction(int sig, __sighandler_t action)
3950{
3951 spin_lock_irq(¤t->sighand->siglock);
3952 current->sighand->action[sig - 1].sa.sa_handler = action;
3953 if (action == SIG_IGN) {
3954 sigset_t mask;
3955
3956 sigemptyset(&mask);
3957 sigaddset(&mask, sig);
3958
3959 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
3960 flush_sigqueue_mask(&mask, ¤t->pending);
3961 recalc_sigpending();
3962 }
3963 spin_unlock_irq(¤t->sighand->siglock);
3964}
3965EXPORT_SYMBOL(kernel_sigaction);
3966
3967void __weak sigaction_compat_abi(struct k_sigaction *act,
3968 struct k_sigaction *oact)
3969{
3970}
3971
3972int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
3973{
3974 struct task_struct *p = current, *t;
3975 struct k_sigaction *k;
3976 sigset_t mask;
3977
3978 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
3979 return -EINVAL;
3980
3981 k = &p->sighand->action[sig-1];
3982
3983 spin_lock_irq(&p->sighand->siglock);
3984 if (oact)
3985 *oact = *k;
3986
3987 sigaction_compat_abi(act, oact);
3988
3989 if (act) {
3990 sigdelsetmask(&act->sa.sa_mask,
3991 sigmask(SIGKILL) | sigmask(SIGSTOP));
3992 *k = *act;
3993 /*
3994 * POSIX 3.3.1.3:
3995 * "Setting a signal action to SIG_IGN for a signal that is
3996 * pending shall cause the pending signal to be discarded,
3997 * whether or not it is blocked."
3998 *
3999 * "Setting a signal action to SIG_DFL for a signal that is
4000 * pending and whose default action is to ignore the signal
4001 * (for example, SIGCHLD), shall cause the pending signal to
4002 * be discarded, whether or not it is blocked"
4003 */
4004 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4005 sigemptyset(&mask);
4006 sigaddset(&mask, sig);
4007 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4008 for_each_thread(p, t)
4009 flush_sigqueue_mask(&mask, &t->pending);
4010 }
4011 }
4012
4013 spin_unlock_irq(&p->sighand->siglock);
4014 return 0;
4015}
4016
4017static int
4018do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4019 size_t min_ss_size)
4020{
4021 struct task_struct *t = current;
4022
4023 if (oss) {
4024 memset(oss, 0, sizeof(stack_t));
4025 oss->ss_sp = (void __user *) t->sas_ss_sp;
4026 oss->ss_size = t->sas_ss_size;
4027 oss->ss_flags = sas_ss_flags(sp) |
4028 (current->sas_ss_flags & SS_FLAG_BITS);
4029 }
4030
4031 if (ss) {
4032 void __user *ss_sp = ss->ss_sp;
4033 size_t ss_size = ss->ss_size;
4034 unsigned ss_flags = ss->ss_flags;
4035 int ss_mode;
4036
4037 if (unlikely(on_sig_stack(sp)))
4038 return -EPERM;
4039
4040 ss_mode = ss_flags & ~SS_FLAG_BITS;
4041 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4042 ss_mode != 0))
4043 return -EINVAL;
4044
4045 if (ss_mode == SS_DISABLE) {
4046 ss_size = 0;
4047 ss_sp = NULL;
4048 } else {
4049 if (unlikely(ss_size < min_ss_size))
4050 return -ENOMEM;
4051 }
4052
4053 t->sas_ss_sp = (unsigned long) ss_sp;
4054 t->sas_ss_size = ss_size;
4055 t->sas_ss_flags = ss_flags;
4056 }
4057 return 0;
4058}
4059
4060SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4061{
4062 stack_t new, old;
4063 int err;
4064 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4065 return -EFAULT;
4066 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4067 current_user_stack_pointer(),
4068 MINSIGSTKSZ);
4069 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4070 err = -EFAULT;
4071 return err;
4072}
4073
4074int restore_altstack(const stack_t __user *uss)
4075{
4076 stack_t new;
4077 if (copy_from_user(&new, uss, sizeof(stack_t)))
4078 return -EFAULT;
4079 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4080 MINSIGSTKSZ);
4081 /* squash all but EFAULT for now */
4082 return 0;
4083}
4084
4085int __save_altstack(stack_t __user *uss, unsigned long sp)
4086{
4087 struct task_struct *t = current;
4088 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4089 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4090 __put_user(t->sas_ss_size, &uss->ss_size);
4091 if (err)
4092 return err;
4093 if (t->sas_ss_flags & SS_AUTODISARM)
4094 sas_ss_reset(t);
4095 return 0;
4096}
4097
4098#ifdef CONFIG_COMPAT
4099static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4100 compat_stack_t __user *uoss_ptr)
4101{
4102 stack_t uss, uoss;
4103 int ret;
4104
4105 if (uss_ptr) {
4106 compat_stack_t uss32;
4107 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4108 return -EFAULT;
4109 uss.ss_sp = compat_ptr(uss32.ss_sp);
4110 uss.ss_flags = uss32.ss_flags;
4111 uss.ss_size = uss32.ss_size;
4112 }
4113 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4114 compat_user_stack_pointer(),
4115 COMPAT_MINSIGSTKSZ);
4116 if (ret >= 0 && uoss_ptr) {
4117 compat_stack_t old;
4118 memset(&old, 0, sizeof(old));
4119 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4120 old.ss_flags = uoss.ss_flags;
4121 old.ss_size = uoss.ss_size;
4122 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4123 ret = -EFAULT;
4124 }
4125 return ret;
4126}
4127
4128COMPAT_SYSCALL_DEFINE2(sigaltstack,
4129 const compat_stack_t __user *, uss_ptr,
4130 compat_stack_t __user *, uoss_ptr)
4131{
4132 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4133}
4134
4135int compat_restore_altstack(const compat_stack_t __user *uss)
4136{
4137 int err = do_compat_sigaltstack(uss, NULL);
4138 /* squash all but -EFAULT for now */
4139 return err == -EFAULT ? err : 0;
4140}
4141
4142int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4143{
4144 int err;
4145 struct task_struct *t = current;
4146 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4147 &uss->ss_sp) |
4148 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4149 __put_user(t->sas_ss_size, &uss->ss_size);
4150 if (err)
4151 return err;
4152 if (t->sas_ss_flags & SS_AUTODISARM)
4153 sas_ss_reset(t);
4154 return 0;
4155}
4156#endif
4157
4158#ifdef __ARCH_WANT_SYS_SIGPENDING
4159
4160/**
4161 * sys_sigpending - examine pending signals
4162 * @uset: where mask of pending signal is returned
4163 */
4164SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4165{
4166 sigset_t set;
4167
4168 if (sizeof(old_sigset_t) > sizeof(*uset))
4169 return -EINVAL;
4170
4171 do_sigpending(&set);
4172
4173 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4174 return -EFAULT;
4175
4176 return 0;
4177}
4178
4179#ifdef CONFIG_COMPAT
4180COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4181{
4182 sigset_t set;
4183
4184 do_sigpending(&set);
4185
4186 return put_user(set.sig[0], set32);
4187}
4188#endif
4189
4190#endif
4191
4192#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4193/**
4194 * sys_sigprocmask - examine and change blocked signals
4195 * @how: whether to add, remove, or set signals
4196 * @nset: signals to add or remove (if non-null)
4197 * @oset: previous value of signal mask if non-null
4198 *
4199 * Some platforms have their own version with special arguments;
4200 * others support only sys_rt_sigprocmask.
4201 */
4202
4203SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4204 old_sigset_t __user *, oset)
4205{
4206 old_sigset_t old_set, new_set;
4207 sigset_t new_blocked;
4208
4209 old_set = current->blocked.sig[0];
4210
4211 if (nset) {
4212 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4213 return -EFAULT;
4214
4215 new_blocked = current->blocked;
4216
4217 switch (how) {
4218 case SIG_BLOCK:
4219 sigaddsetmask(&new_blocked, new_set);
4220 break;
4221 case SIG_UNBLOCK:
4222 sigdelsetmask(&new_blocked, new_set);
4223 break;
4224 case SIG_SETMASK:
4225 new_blocked.sig[0] = new_set;
4226 break;
4227 default:
4228 return -EINVAL;
4229 }
4230
4231 set_current_blocked(&new_blocked);
4232 }
4233
4234 if (oset) {
4235 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4236 return -EFAULT;
4237 }
4238
4239 return 0;
4240}
4241#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4242
4243#ifndef CONFIG_ODD_RT_SIGACTION
4244/**
4245 * sys_rt_sigaction - alter an action taken by a process
4246 * @sig: signal to be sent
4247 * @act: new sigaction
4248 * @oact: used to save the previous sigaction
4249 * @sigsetsize: size of sigset_t type
4250 */
4251SYSCALL_DEFINE4(rt_sigaction, int, sig,
4252 const struct sigaction __user *, act,
4253 struct sigaction __user *, oact,
4254 size_t, sigsetsize)
4255{
4256 struct k_sigaction new_sa, old_sa;
4257 int ret;
4258
4259 /* XXX: Don't preclude handling different sized sigset_t's. */
4260 if (sigsetsize != sizeof(sigset_t))
4261 return -EINVAL;
4262
4263 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4264 return -EFAULT;
4265
4266 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4267 if (ret)
4268 return ret;
4269
4270 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4271 return -EFAULT;
4272
4273 return 0;
4274}
4275#ifdef CONFIG_COMPAT
4276COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4277 const struct compat_sigaction __user *, act,
4278 struct compat_sigaction __user *, oact,
4279 compat_size_t, sigsetsize)
4280{
4281 struct k_sigaction new_ka, old_ka;
4282#ifdef __ARCH_HAS_SA_RESTORER
4283 compat_uptr_t restorer;
4284#endif
4285 int ret;
4286
4287 /* XXX: Don't preclude handling different sized sigset_t's. */
4288 if (sigsetsize != sizeof(compat_sigset_t))
4289 return -EINVAL;
4290
4291 if (act) {
4292 compat_uptr_t handler;
4293 ret = get_user(handler, &act->sa_handler);
4294 new_ka.sa.sa_handler = compat_ptr(handler);
4295#ifdef __ARCH_HAS_SA_RESTORER
4296 ret |= get_user(restorer, &act->sa_restorer);
4297 new_ka.sa.sa_restorer = compat_ptr(restorer);
4298#endif
4299 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4300 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4301 if (ret)
4302 return -EFAULT;
4303 }
4304
4305 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4306 if (!ret && oact) {
4307 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4308 &oact->sa_handler);
4309 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4310 sizeof(oact->sa_mask));
4311 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4312#ifdef __ARCH_HAS_SA_RESTORER
4313 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4314 &oact->sa_restorer);
4315#endif
4316 }
4317 return ret;
4318}
4319#endif
4320#endif /* !CONFIG_ODD_RT_SIGACTION */
4321
4322#ifdef CONFIG_OLD_SIGACTION
4323SYSCALL_DEFINE3(sigaction, int, sig,
4324 const struct old_sigaction __user *, act,
4325 struct old_sigaction __user *, oact)
4326{
4327 struct k_sigaction new_ka, old_ka;
4328 int ret;
4329
4330 if (act) {
4331 old_sigset_t mask;
4332 if (!access_ok(act, sizeof(*act)) ||
4333 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4334 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4335 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4336 __get_user(mask, &act->sa_mask))
4337 return -EFAULT;
4338#ifdef __ARCH_HAS_KA_RESTORER
4339 new_ka.ka_restorer = NULL;
4340#endif
4341 siginitset(&new_ka.sa.sa_mask, mask);
4342 }
4343
4344 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4345
4346 if (!ret && oact) {
4347 if (!access_ok(oact, sizeof(*oact)) ||
4348 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4349 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4350 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4351 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4352 return -EFAULT;
4353 }
4354
4355 return ret;
4356}
4357#endif
4358#ifdef CONFIG_COMPAT_OLD_SIGACTION
4359COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4360 const struct compat_old_sigaction __user *, act,
4361 struct compat_old_sigaction __user *, oact)
4362{
4363 struct k_sigaction new_ka, old_ka;
4364 int ret;
4365 compat_old_sigset_t mask;
4366 compat_uptr_t handler, restorer;
4367
4368 if (act) {
4369 if (!access_ok(act, sizeof(*act)) ||
4370 __get_user(handler, &act->sa_handler) ||
4371 __get_user(restorer, &act->sa_restorer) ||
4372 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4373 __get_user(mask, &act->sa_mask))
4374 return -EFAULT;
4375
4376#ifdef __ARCH_HAS_KA_RESTORER
4377 new_ka.ka_restorer = NULL;
4378#endif
4379 new_ka.sa.sa_handler = compat_ptr(handler);
4380 new_ka.sa.sa_restorer = compat_ptr(restorer);
4381 siginitset(&new_ka.sa.sa_mask, mask);
4382 }
4383
4384 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4385
4386 if (!ret && oact) {
4387 if (!access_ok(oact, sizeof(*oact)) ||
4388 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4389 &oact->sa_handler) ||
4390 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4391 &oact->sa_restorer) ||
4392 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4393 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4394 return -EFAULT;
4395 }
4396 return ret;
4397}
4398#endif
4399
4400#ifdef CONFIG_SGETMASK_SYSCALL
4401
4402/*
4403 * For backwards compatibility. Functionality superseded by sigprocmask.
4404 */
4405SYSCALL_DEFINE0(sgetmask)
4406{
4407 /* SMP safe */
4408 return current->blocked.sig[0];
4409}
4410
4411SYSCALL_DEFINE1(ssetmask, int, newmask)
4412{
4413 int old = current->blocked.sig[0];
4414 sigset_t newset;
4415
4416 siginitset(&newset, newmask);
4417 set_current_blocked(&newset);
4418
4419 return old;
4420}
4421#endif /* CONFIG_SGETMASK_SYSCALL */
4422
4423#ifdef __ARCH_WANT_SYS_SIGNAL
4424/*
4425 * For backwards compatibility. Functionality superseded by sigaction.
4426 */
4427SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4428{
4429 struct k_sigaction new_sa, old_sa;
4430 int ret;
4431
4432 new_sa.sa.sa_handler = handler;
4433 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4434 sigemptyset(&new_sa.sa.sa_mask);
4435
4436 ret = do_sigaction(sig, &new_sa, &old_sa);
4437
4438 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4439}
4440#endif /* __ARCH_WANT_SYS_SIGNAL */
4441
4442#ifdef __ARCH_WANT_SYS_PAUSE
4443
4444SYSCALL_DEFINE0(pause)
4445{
4446 while (!signal_pending(current)) {
4447 __set_current_state(TASK_INTERRUPTIBLE);
4448 schedule();
4449 }
4450 return -ERESTARTNOHAND;
4451}
4452
4453#endif
4454
4455static int sigsuspend(sigset_t *set)
4456{
4457 current->saved_sigmask = current->blocked;
4458 set_current_blocked(set);
4459
4460 while (!signal_pending(current)) {
4461 __set_current_state(TASK_INTERRUPTIBLE);
4462 schedule();
4463 }
4464 set_restore_sigmask();
4465 return -ERESTARTNOHAND;
4466}
4467
4468/**
4469 * sys_rt_sigsuspend - replace the signal mask for a value with the
4470 * @unewset value until a signal is received
4471 * @unewset: new signal mask value
4472 * @sigsetsize: size of sigset_t type
4473 */
4474SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4475{
4476 sigset_t newset;
4477
4478 /* XXX: Don't preclude handling different sized sigset_t's. */
4479 if (sigsetsize != sizeof(sigset_t))
4480 return -EINVAL;
4481
4482 if (copy_from_user(&newset, unewset, sizeof(newset)))
4483 return -EFAULT;
4484 return sigsuspend(&newset);
4485}
4486
4487#ifdef CONFIG_COMPAT
4488COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4489{
4490 sigset_t newset;
4491
4492 /* XXX: Don't preclude handling different sized sigset_t's. */
4493 if (sigsetsize != sizeof(sigset_t))
4494 return -EINVAL;
4495
4496 if (get_compat_sigset(&newset, unewset))
4497 return -EFAULT;
4498 return sigsuspend(&newset);
4499}
4500#endif
4501
4502#ifdef CONFIG_OLD_SIGSUSPEND
4503SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4504{
4505 sigset_t blocked;
4506 siginitset(&blocked, mask);
4507 return sigsuspend(&blocked);
4508}
4509#endif
4510#ifdef CONFIG_OLD_SIGSUSPEND3
4511SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4512{
4513 sigset_t blocked;
4514 siginitset(&blocked, mask);
4515 return sigsuspend(&blocked);
4516}
4517#endif
4518
4519__weak const char *arch_vma_name(struct vm_area_struct *vma)
4520{
4521 return NULL;
4522}
4523
4524static inline void siginfo_buildtime_checks(void)
4525{
4526 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4527
4528 /* Verify the offsets in the two siginfos match */
4529#define CHECK_OFFSET(field) \
4530 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4531
4532 /* kill */
4533 CHECK_OFFSET(si_pid);
4534 CHECK_OFFSET(si_uid);
4535
4536 /* timer */
4537 CHECK_OFFSET(si_tid);
4538 CHECK_OFFSET(si_overrun);
4539 CHECK_OFFSET(si_value);
4540
4541 /* rt */
4542 CHECK_OFFSET(si_pid);
4543 CHECK_OFFSET(si_uid);
4544 CHECK_OFFSET(si_value);
4545
4546 /* sigchld */
4547 CHECK_OFFSET(si_pid);
4548 CHECK_OFFSET(si_uid);
4549 CHECK_OFFSET(si_status);
4550 CHECK_OFFSET(si_utime);
4551 CHECK_OFFSET(si_stime);
4552
4553 /* sigfault */
4554 CHECK_OFFSET(si_addr);
4555 CHECK_OFFSET(si_addr_lsb);
4556 CHECK_OFFSET(si_lower);
4557 CHECK_OFFSET(si_upper);
4558 CHECK_OFFSET(si_pkey);
4559
4560 /* sigpoll */
4561 CHECK_OFFSET(si_band);
4562 CHECK_OFFSET(si_fd);
4563
4564 /* sigsys */
4565 CHECK_OFFSET(si_call_addr);
4566 CHECK_OFFSET(si_syscall);
4567 CHECK_OFFSET(si_arch);
4568#undef CHECK_OFFSET
4569
4570 /* usb asyncio */
4571 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4572 offsetof(struct siginfo, si_addr));
4573 if (sizeof(int) == sizeof(void __user *)) {
4574 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4575 sizeof(void __user *));
4576 } else {
4577 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4578 sizeof_field(struct siginfo, si_uid)) !=
4579 sizeof(void __user *));
4580 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4581 offsetof(struct siginfo, si_uid));
4582 }
4583#ifdef CONFIG_COMPAT
4584 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4585 offsetof(struct compat_siginfo, si_addr));
4586 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4587 sizeof(compat_uptr_t));
4588 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4589 sizeof_field(struct siginfo, si_pid));
4590#endif
4591}
4592
4593void __init signals_init(void)
4594{
4595 siginfo_buildtime_checks();
4596
4597 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4598}
4599
4600#ifdef CONFIG_KGDB_KDB
4601#include <linux/kdb.h>
4602/*
4603 * kdb_send_sig - Allows kdb to send signals without exposing
4604 * signal internals. This function checks if the required locks are
4605 * available before calling the main signal code, to avoid kdb
4606 * deadlocks.
4607 */
4608void kdb_send_sig(struct task_struct *t, int sig)
4609{
4610 static struct task_struct *kdb_prev_t;
4611 int new_t, ret;
4612 if (!spin_trylock(&t->sighand->siglock)) {
4613 kdb_printf("Can't do kill command now.\n"
4614 "The sigmask lock is held somewhere else in "
4615 "kernel, try again later\n");
4616 return;
4617 }
4618 new_t = kdb_prev_t != t;
4619 kdb_prev_t = t;
4620 if (t->state != TASK_RUNNING && new_t) {
4621 spin_unlock(&t->sighand->siglock);
4622 kdb_printf("Process is not RUNNING, sending a signal from "
4623 "kdb risks deadlock\n"
4624 "on the run queue locks. "
4625 "The signal has _not_ been sent.\n"
4626 "Reissue the kill command if you want to risk "
4627 "the deadlock.\n");
4628 return;
4629 }
4630 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4631 spin_unlock(&t->sighand->siglock);
4632 if (ret)
4633 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4634 sig, t->pid);
4635 else
4636 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4637}
4638#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#include <uapi/linux/pidfd.h>
51
52#define CREATE_TRACE_POINTS
53#include <trace/events/signal.h>
54
55#include <asm/param.h>
56#include <linux/uaccess.h>
57#include <asm/unistd.h>
58#include <asm/siginfo.h>
59#include <asm/cacheflush.h>
60#include <asm/syscall.h> /* for syscall_get_* */
61
62/*
63 * SLAB caches for signal bits.
64 */
65
66static struct kmem_cache *sigqueue_cachep;
67
68int print_fatal_signals __read_mostly;
69
70static void __user *sig_handler(struct task_struct *t, int sig)
71{
72 return t->sighand->action[sig - 1].sa.sa_handler;
73}
74
75static inline bool sig_handler_ignored(void __user *handler, int sig)
76{
77 /* Is it explicitly or implicitly ignored? */
78 return handler == SIG_IGN ||
79 (handler == SIG_DFL && sig_kernel_ignore(sig));
80}
81
82static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
83{
84 void __user *handler;
85
86 handler = sig_handler(t, sig);
87
88 /* SIGKILL and SIGSTOP may not be sent to the global init */
89 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
90 return true;
91
92 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
93 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
94 return true;
95
96 /* Only allow kernel generated signals to this kthread */
97 if (unlikely((t->flags & PF_KTHREAD) &&
98 (handler == SIG_KTHREAD_KERNEL) && !force))
99 return true;
100
101 return sig_handler_ignored(handler, sig);
102}
103
104static bool sig_ignored(struct task_struct *t, int sig, bool force)
105{
106 /*
107 * Blocked signals are never ignored, since the
108 * signal handler may change by the time it is
109 * unblocked.
110 */
111 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
112 return false;
113
114 /*
115 * Tracers may want to know about even ignored signal unless it
116 * is SIGKILL which can't be reported anyway but can be ignored
117 * by SIGNAL_UNKILLABLE task.
118 */
119 if (t->ptrace && sig != SIGKILL)
120 return false;
121
122 return sig_task_ignored(t, sig, force);
123}
124
125/*
126 * Re-calculate pending state from the set of locally pending
127 * signals, globally pending signals, and blocked signals.
128 */
129static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
130{
131 unsigned long ready;
132 long i;
133
134 switch (_NSIG_WORDS) {
135 default:
136 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
137 ready |= signal->sig[i] &~ blocked->sig[i];
138 break;
139
140 case 4: ready = signal->sig[3] &~ blocked->sig[3];
141 ready |= signal->sig[2] &~ blocked->sig[2];
142 ready |= signal->sig[1] &~ blocked->sig[1];
143 ready |= signal->sig[0] &~ blocked->sig[0];
144 break;
145
146 case 2: ready = signal->sig[1] &~ blocked->sig[1];
147 ready |= signal->sig[0] &~ blocked->sig[0];
148 break;
149
150 case 1: ready = signal->sig[0] &~ blocked->sig[0];
151 }
152 return ready != 0;
153}
154
155#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
156
157static bool recalc_sigpending_tsk(struct task_struct *t)
158{
159 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
160 PENDING(&t->pending, &t->blocked) ||
161 PENDING(&t->signal->shared_pending, &t->blocked) ||
162 cgroup_task_frozen(t)) {
163 set_tsk_thread_flag(t, TIF_SIGPENDING);
164 return true;
165 }
166
167 /*
168 * We must never clear the flag in another thread, or in current
169 * when it's possible the current syscall is returning -ERESTART*.
170 * So we don't clear it here, and only callers who know they should do.
171 */
172 return false;
173}
174
175void recalc_sigpending(void)
176{
177 if (!recalc_sigpending_tsk(current) && !freezing(current))
178 clear_thread_flag(TIF_SIGPENDING);
179
180}
181EXPORT_SYMBOL(recalc_sigpending);
182
183void calculate_sigpending(void)
184{
185 /* Have any signals or users of TIF_SIGPENDING been delayed
186 * until after fork?
187 */
188 spin_lock_irq(¤t->sighand->siglock);
189 set_tsk_thread_flag(current, TIF_SIGPENDING);
190 recalc_sigpending();
191 spin_unlock_irq(¤t->sighand->siglock);
192}
193
194/* Given the mask, find the first available signal that should be serviced. */
195
196#define SYNCHRONOUS_MASK \
197 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
198 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
199
200int next_signal(struct sigpending *pending, sigset_t *mask)
201{
202 unsigned long i, *s, *m, x;
203 int sig = 0;
204
205 s = pending->signal.sig;
206 m = mask->sig;
207
208 /*
209 * Handle the first word specially: it contains the
210 * synchronous signals that need to be dequeued first.
211 */
212 x = *s &~ *m;
213 if (x) {
214 if (x & SYNCHRONOUS_MASK)
215 x &= SYNCHRONOUS_MASK;
216 sig = ffz(~x) + 1;
217 return sig;
218 }
219
220 switch (_NSIG_WORDS) {
221 default:
222 for (i = 1; i < _NSIG_WORDS; ++i) {
223 x = *++s &~ *++m;
224 if (!x)
225 continue;
226 sig = ffz(~x) + i*_NSIG_BPW + 1;
227 break;
228 }
229 break;
230
231 case 2:
232 x = s[1] &~ m[1];
233 if (!x)
234 break;
235 sig = ffz(~x) + _NSIG_BPW + 1;
236 break;
237
238 case 1:
239 /* Nothing to do */
240 break;
241 }
242
243 return sig;
244}
245
246static inline void print_dropped_signal(int sig)
247{
248 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
249
250 if (!print_fatal_signals)
251 return;
252
253 if (!__ratelimit(&ratelimit_state))
254 return;
255
256 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
257 current->comm, current->pid, sig);
258}
259
260/**
261 * task_set_jobctl_pending - set jobctl pending bits
262 * @task: target task
263 * @mask: pending bits to set
264 *
265 * Clear @mask from @task->jobctl. @mask must be subset of
266 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
267 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
268 * cleared. If @task is already being killed or exiting, this function
269 * becomes noop.
270 *
271 * CONTEXT:
272 * Must be called with @task->sighand->siglock held.
273 *
274 * RETURNS:
275 * %true if @mask is set, %false if made noop because @task was dying.
276 */
277bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
278{
279 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
280 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
281 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
282
283 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
284 return false;
285
286 if (mask & JOBCTL_STOP_SIGMASK)
287 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
288
289 task->jobctl |= mask;
290 return true;
291}
292
293/**
294 * task_clear_jobctl_trapping - clear jobctl trapping bit
295 * @task: target task
296 *
297 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
298 * Clear it and wake up the ptracer. Note that we don't need any further
299 * locking. @task->siglock guarantees that @task->parent points to the
300 * ptracer.
301 *
302 * CONTEXT:
303 * Must be called with @task->sighand->siglock held.
304 */
305void task_clear_jobctl_trapping(struct task_struct *task)
306{
307 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
308 task->jobctl &= ~JOBCTL_TRAPPING;
309 smp_mb(); /* advised by wake_up_bit() */
310 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
311 }
312}
313
314/**
315 * task_clear_jobctl_pending - clear jobctl pending bits
316 * @task: target task
317 * @mask: pending bits to clear
318 *
319 * Clear @mask from @task->jobctl. @mask must be subset of
320 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
321 * STOP bits are cleared together.
322 *
323 * If clearing of @mask leaves no stop or trap pending, this function calls
324 * task_clear_jobctl_trapping().
325 *
326 * CONTEXT:
327 * Must be called with @task->sighand->siglock held.
328 */
329void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
330{
331 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
332
333 if (mask & JOBCTL_STOP_PENDING)
334 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
335
336 task->jobctl &= ~mask;
337
338 if (!(task->jobctl & JOBCTL_PENDING_MASK))
339 task_clear_jobctl_trapping(task);
340}
341
342/**
343 * task_participate_group_stop - participate in a group stop
344 * @task: task participating in a group stop
345 *
346 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
347 * Group stop states are cleared and the group stop count is consumed if
348 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
349 * stop, the appropriate `SIGNAL_*` flags are set.
350 *
351 * CONTEXT:
352 * Must be called with @task->sighand->siglock held.
353 *
354 * RETURNS:
355 * %true if group stop completion should be notified to the parent, %false
356 * otherwise.
357 */
358static bool task_participate_group_stop(struct task_struct *task)
359{
360 struct signal_struct *sig = task->signal;
361 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
362
363 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
364
365 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
366
367 if (!consume)
368 return false;
369
370 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
371 sig->group_stop_count--;
372
373 /*
374 * Tell the caller to notify completion iff we are entering into a
375 * fresh group stop. Read comment in do_signal_stop() for details.
376 */
377 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
378 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
379 return true;
380 }
381 return false;
382}
383
384void task_join_group_stop(struct task_struct *task)
385{
386 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
387 struct signal_struct *sig = current->signal;
388
389 if (sig->group_stop_count) {
390 sig->group_stop_count++;
391 mask |= JOBCTL_STOP_CONSUME;
392 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
393 return;
394
395 /* Have the new thread join an on-going signal group stop */
396 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
397}
398
399/*
400 * allocate a new signal queue record
401 * - this may be called without locks if and only if t == current, otherwise an
402 * appropriate lock must be held to stop the target task from exiting
403 */
404static struct sigqueue *
405__sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
406 int override_rlimit, const unsigned int sigqueue_flags)
407{
408 struct sigqueue *q = NULL;
409 struct ucounts *ucounts;
410 long sigpending;
411
412 /*
413 * Protect access to @t credentials. This can go away when all
414 * callers hold rcu read lock.
415 *
416 * NOTE! A pending signal will hold on to the user refcount,
417 * and we get/put the refcount only when the sigpending count
418 * changes from/to zero.
419 */
420 rcu_read_lock();
421 ucounts = task_ucounts(t);
422 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
423 rcu_read_unlock();
424 if (!sigpending)
425 return NULL;
426
427 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
428 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
429 } else {
430 print_dropped_signal(sig);
431 }
432
433 if (unlikely(q == NULL)) {
434 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
435 } else {
436 INIT_LIST_HEAD(&q->list);
437 q->flags = sigqueue_flags;
438 q->ucounts = ucounts;
439 }
440 return q;
441}
442
443static void __sigqueue_free(struct sigqueue *q)
444{
445 if (q->flags & SIGQUEUE_PREALLOC)
446 return;
447 if (q->ucounts) {
448 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
449 q->ucounts = NULL;
450 }
451 kmem_cache_free(sigqueue_cachep, q);
452}
453
454void flush_sigqueue(struct sigpending *queue)
455{
456 struct sigqueue *q;
457
458 sigemptyset(&queue->signal);
459 while (!list_empty(&queue->list)) {
460 q = list_entry(queue->list.next, struct sigqueue , list);
461 list_del_init(&q->list);
462 __sigqueue_free(q);
463 }
464}
465
466/*
467 * Flush all pending signals for this kthread.
468 */
469void flush_signals(struct task_struct *t)
470{
471 unsigned long flags;
472
473 spin_lock_irqsave(&t->sighand->siglock, flags);
474 clear_tsk_thread_flag(t, TIF_SIGPENDING);
475 flush_sigqueue(&t->pending);
476 flush_sigqueue(&t->signal->shared_pending);
477 spin_unlock_irqrestore(&t->sighand->siglock, flags);
478}
479EXPORT_SYMBOL(flush_signals);
480
481#ifdef CONFIG_POSIX_TIMERS
482static void __flush_itimer_signals(struct sigpending *pending)
483{
484 sigset_t signal, retain;
485 struct sigqueue *q, *n;
486
487 signal = pending->signal;
488 sigemptyset(&retain);
489
490 list_for_each_entry_safe(q, n, &pending->list, list) {
491 int sig = q->info.si_signo;
492
493 if (likely(q->info.si_code != SI_TIMER)) {
494 sigaddset(&retain, sig);
495 } else {
496 sigdelset(&signal, sig);
497 list_del_init(&q->list);
498 __sigqueue_free(q);
499 }
500 }
501
502 sigorsets(&pending->signal, &signal, &retain);
503}
504
505void flush_itimer_signals(void)
506{
507 struct task_struct *tsk = current;
508 unsigned long flags;
509
510 spin_lock_irqsave(&tsk->sighand->siglock, flags);
511 __flush_itimer_signals(&tsk->pending);
512 __flush_itimer_signals(&tsk->signal->shared_pending);
513 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
514}
515#endif
516
517void ignore_signals(struct task_struct *t)
518{
519 int i;
520
521 for (i = 0; i < _NSIG; ++i)
522 t->sighand->action[i].sa.sa_handler = SIG_IGN;
523
524 flush_signals(t);
525}
526
527/*
528 * Flush all handlers for a task.
529 */
530
531void
532flush_signal_handlers(struct task_struct *t, int force_default)
533{
534 int i;
535 struct k_sigaction *ka = &t->sighand->action[0];
536 for (i = _NSIG ; i != 0 ; i--) {
537 if (force_default || ka->sa.sa_handler != SIG_IGN)
538 ka->sa.sa_handler = SIG_DFL;
539 ka->sa.sa_flags = 0;
540#ifdef __ARCH_HAS_SA_RESTORER
541 ka->sa.sa_restorer = NULL;
542#endif
543 sigemptyset(&ka->sa.sa_mask);
544 ka++;
545 }
546}
547
548bool unhandled_signal(struct task_struct *tsk, int sig)
549{
550 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
551 if (is_global_init(tsk))
552 return true;
553
554 if (handler != SIG_IGN && handler != SIG_DFL)
555 return false;
556
557 /* If dying, we handle all new signals by ignoring them */
558 if (fatal_signal_pending(tsk))
559 return false;
560
561 /* if ptraced, let the tracer determine */
562 return !tsk->ptrace;
563}
564
565static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
566 bool *resched_timer)
567{
568 struct sigqueue *q, *first = NULL;
569
570 /*
571 * Collect the siginfo appropriate to this signal. Check if
572 * there is another siginfo for the same signal.
573 */
574 list_for_each_entry(q, &list->list, list) {
575 if (q->info.si_signo == sig) {
576 if (first)
577 goto still_pending;
578 first = q;
579 }
580 }
581
582 sigdelset(&list->signal, sig);
583
584 if (first) {
585still_pending:
586 list_del_init(&first->list);
587 copy_siginfo(info, &first->info);
588
589 *resched_timer =
590 (first->flags & SIGQUEUE_PREALLOC) &&
591 (info->si_code == SI_TIMER) &&
592 (info->si_sys_private);
593
594 __sigqueue_free(first);
595 } else {
596 /*
597 * Ok, it wasn't in the queue. This must be
598 * a fast-pathed signal or we must have been
599 * out of queue space. So zero out the info.
600 */
601 clear_siginfo(info);
602 info->si_signo = sig;
603 info->si_errno = 0;
604 info->si_code = SI_USER;
605 info->si_pid = 0;
606 info->si_uid = 0;
607 }
608}
609
610static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
611 kernel_siginfo_t *info, bool *resched_timer)
612{
613 int sig = next_signal(pending, mask);
614
615 if (sig)
616 collect_signal(sig, pending, info, resched_timer);
617 return sig;
618}
619
620/*
621 * Dequeue a signal and return the element to the caller, which is
622 * expected to free it.
623 *
624 * All callers have to hold the siglock.
625 */
626int dequeue_signal(struct task_struct *tsk, sigset_t *mask,
627 kernel_siginfo_t *info, enum pid_type *type)
628{
629 bool resched_timer = false;
630 int signr;
631
632 /* We only dequeue private signals from ourselves, we don't let
633 * signalfd steal them
634 */
635 *type = PIDTYPE_PID;
636 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
637 if (!signr) {
638 *type = PIDTYPE_TGID;
639 signr = __dequeue_signal(&tsk->signal->shared_pending,
640 mask, info, &resched_timer);
641#ifdef CONFIG_POSIX_TIMERS
642 /*
643 * itimer signal ?
644 *
645 * itimers are process shared and we restart periodic
646 * itimers in the signal delivery path to prevent DoS
647 * attacks in the high resolution timer case. This is
648 * compliant with the old way of self-restarting
649 * itimers, as the SIGALRM is a legacy signal and only
650 * queued once. Changing the restart behaviour to
651 * restart the timer in the signal dequeue path is
652 * reducing the timer noise on heavy loaded !highres
653 * systems too.
654 */
655 if (unlikely(signr == SIGALRM)) {
656 struct hrtimer *tmr = &tsk->signal->real_timer;
657
658 if (!hrtimer_is_queued(tmr) &&
659 tsk->signal->it_real_incr != 0) {
660 hrtimer_forward(tmr, tmr->base->get_time(),
661 tsk->signal->it_real_incr);
662 hrtimer_restart(tmr);
663 }
664 }
665#endif
666 }
667
668 recalc_sigpending();
669 if (!signr)
670 return 0;
671
672 if (unlikely(sig_kernel_stop(signr))) {
673 /*
674 * Set a marker that we have dequeued a stop signal. Our
675 * caller might release the siglock and then the pending
676 * stop signal it is about to process is no longer in the
677 * pending bitmasks, but must still be cleared by a SIGCONT
678 * (and overruled by a SIGKILL). So those cases clear this
679 * shared flag after we've set it. Note that this flag may
680 * remain set after the signal we return is ignored or
681 * handled. That doesn't matter because its only purpose
682 * is to alert stop-signal processing code when another
683 * processor has come along and cleared the flag.
684 */
685 current->jobctl |= JOBCTL_STOP_DEQUEUED;
686 }
687#ifdef CONFIG_POSIX_TIMERS
688 if (resched_timer) {
689 /*
690 * Release the siglock to ensure proper locking order
691 * of timer locks outside of siglocks. Note, we leave
692 * irqs disabled here, since the posix-timers code is
693 * about to disable them again anyway.
694 */
695 spin_unlock(&tsk->sighand->siglock);
696 posixtimer_rearm(info);
697 spin_lock(&tsk->sighand->siglock);
698
699 /* Don't expose the si_sys_private value to userspace */
700 info->si_sys_private = 0;
701 }
702#endif
703 return signr;
704}
705EXPORT_SYMBOL_GPL(dequeue_signal);
706
707static int dequeue_synchronous_signal(kernel_siginfo_t *info)
708{
709 struct task_struct *tsk = current;
710 struct sigpending *pending = &tsk->pending;
711 struct sigqueue *q, *sync = NULL;
712
713 /*
714 * Might a synchronous signal be in the queue?
715 */
716 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
717 return 0;
718
719 /*
720 * Return the first synchronous signal in the queue.
721 */
722 list_for_each_entry(q, &pending->list, list) {
723 /* Synchronous signals have a positive si_code */
724 if ((q->info.si_code > SI_USER) &&
725 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
726 sync = q;
727 goto next;
728 }
729 }
730 return 0;
731next:
732 /*
733 * Check if there is another siginfo for the same signal.
734 */
735 list_for_each_entry_continue(q, &pending->list, list) {
736 if (q->info.si_signo == sync->info.si_signo)
737 goto still_pending;
738 }
739
740 sigdelset(&pending->signal, sync->info.si_signo);
741 recalc_sigpending();
742still_pending:
743 list_del_init(&sync->list);
744 copy_siginfo(info, &sync->info);
745 __sigqueue_free(sync);
746 return info->si_signo;
747}
748
749/*
750 * Tell a process that it has a new active signal..
751 *
752 * NOTE! we rely on the previous spin_lock to
753 * lock interrupts for us! We can only be called with
754 * "siglock" held, and the local interrupt must
755 * have been disabled when that got acquired!
756 *
757 * No need to set need_resched since signal event passing
758 * goes through ->blocked
759 */
760void signal_wake_up_state(struct task_struct *t, unsigned int state)
761{
762 lockdep_assert_held(&t->sighand->siglock);
763
764 set_tsk_thread_flag(t, TIF_SIGPENDING);
765
766 /*
767 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
768 * case. We don't check t->state here because there is a race with it
769 * executing another processor and just now entering stopped state.
770 * By using wake_up_state, we ensure the process will wake up and
771 * handle its death signal.
772 */
773 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
774 kick_process(t);
775}
776
777/*
778 * Remove signals in mask from the pending set and queue.
779 * Returns 1 if any signals were found.
780 *
781 * All callers must be holding the siglock.
782 */
783static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
784{
785 struct sigqueue *q, *n;
786 sigset_t m;
787
788 sigandsets(&m, mask, &s->signal);
789 if (sigisemptyset(&m))
790 return;
791
792 sigandnsets(&s->signal, &s->signal, mask);
793 list_for_each_entry_safe(q, n, &s->list, list) {
794 if (sigismember(mask, q->info.si_signo)) {
795 list_del_init(&q->list);
796 __sigqueue_free(q);
797 }
798 }
799}
800
801static inline int is_si_special(const struct kernel_siginfo *info)
802{
803 return info <= SEND_SIG_PRIV;
804}
805
806static inline bool si_fromuser(const struct kernel_siginfo *info)
807{
808 return info == SEND_SIG_NOINFO ||
809 (!is_si_special(info) && SI_FROMUSER(info));
810}
811
812/*
813 * called with RCU read lock from check_kill_permission()
814 */
815static bool kill_ok_by_cred(struct task_struct *t)
816{
817 const struct cred *cred = current_cred();
818 const struct cred *tcred = __task_cred(t);
819
820 return uid_eq(cred->euid, tcred->suid) ||
821 uid_eq(cred->euid, tcred->uid) ||
822 uid_eq(cred->uid, tcred->suid) ||
823 uid_eq(cred->uid, tcred->uid) ||
824 ns_capable(tcred->user_ns, CAP_KILL);
825}
826
827/*
828 * Bad permissions for sending the signal
829 * - the caller must hold the RCU read lock
830 */
831static int check_kill_permission(int sig, struct kernel_siginfo *info,
832 struct task_struct *t)
833{
834 struct pid *sid;
835 int error;
836
837 if (!valid_signal(sig))
838 return -EINVAL;
839
840 if (!si_fromuser(info))
841 return 0;
842
843 error = audit_signal_info(sig, t); /* Let audit system see the signal */
844 if (error)
845 return error;
846
847 if (!same_thread_group(current, t) &&
848 !kill_ok_by_cred(t)) {
849 switch (sig) {
850 case SIGCONT:
851 sid = task_session(t);
852 /*
853 * We don't return the error if sid == NULL. The
854 * task was unhashed, the caller must notice this.
855 */
856 if (!sid || sid == task_session(current))
857 break;
858 fallthrough;
859 default:
860 return -EPERM;
861 }
862 }
863
864 return security_task_kill(t, info, sig, NULL);
865}
866
867/**
868 * ptrace_trap_notify - schedule trap to notify ptracer
869 * @t: tracee wanting to notify tracer
870 *
871 * This function schedules sticky ptrace trap which is cleared on the next
872 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
873 * ptracer.
874 *
875 * If @t is running, STOP trap will be taken. If trapped for STOP and
876 * ptracer is listening for events, tracee is woken up so that it can
877 * re-trap for the new event. If trapped otherwise, STOP trap will be
878 * eventually taken without returning to userland after the existing traps
879 * are finished by PTRACE_CONT.
880 *
881 * CONTEXT:
882 * Must be called with @task->sighand->siglock held.
883 */
884static void ptrace_trap_notify(struct task_struct *t)
885{
886 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
887 lockdep_assert_held(&t->sighand->siglock);
888
889 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
890 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
891}
892
893/*
894 * Handle magic process-wide effects of stop/continue signals. Unlike
895 * the signal actions, these happen immediately at signal-generation
896 * time regardless of blocking, ignoring, or handling. This does the
897 * actual continuing for SIGCONT, but not the actual stopping for stop
898 * signals. The process stop is done as a signal action for SIG_DFL.
899 *
900 * Returns true if the signal should be actually delivered, otherwise
901 * it should be dropped.
902 */
903static bool prepare_signal(int sig, struct task_struct *p, bool force)
904{
905 struct signal_struct *signal = p->signal;
906 struct task_struct *t;
907 sigset_t flush;
908
909 if (signal->flags & SIGNAL_GROUP_EXIT) {
910 if (signal->core_state)
911 return sig == SIGKILL;
912 /*
913 * The process is in the middle of dying, drop the signal.
914 */
915 return false;
916 } else if (sig_kernel_stop(sig)) {
917 /*
918 * This is a stop signal. Remove SIGCONT from all queues.
919 */
920 siginitset(&flush, sigmask(SIGCONT));
921 flush_sigqueue_mask(&flush, &signal->shared_pending);
922 for_each_thread(p, t)
923 flush_sigqueue_mask(&flush, &t->pending);
924 } else if (sig == SIGCONT) {
925 unsigned int why;
926 /*
927 * Remove all stop signals from all queues, wake all threads.
928 */
929 siginitset(&flush, SIG_KERNEL_STOP_MASK);
930 flush_sigqueue_mask(&flush, &signal->shared_pending);
931 for_each_thread(p, t) {
932 flush_sigqueue_mask(&flush, &t->pending);
933 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
934 if (likely(!(t->ptrace & PT_SEIZED))) {
935 t->jobctl &= ~JOBCTL_STOPPED;
936 wake_up_state(t, __TASK_STOPPED);
937 } else
938 ptrace_trap_notify(t);
939 }
940
941 /*
942 * Notify the parent with CLD_CONTINUED if we were stopped.
943 *
944 * If we were in the middle of a group stop, we pretend it
945 * was already finished, and then continued. Since SIGCHLD
946 * doesn't queue we report only CLD_STOPPED, as if the next
947 * CLD_CONTINUED was dropped.
948 */
949 why = 0;
950 if (signal->flags & SIGNAL_STOP_STOPPED)
951 why |= SIGNAL_CLD_CONTINUED;
952 else if (signal->group_stop_count)
953 why |= SIGNAL_CLD_STOPPED;
954
955 if (why) {
956 /*
957 * The first thread which returns from do_signal_stop()
958 * will take ->siglock, notice SIGNAL_CLD_MASK, and
959 * notify its parent. See get_signal().
960 */
961 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
962 signal->group_stop_count = 0;
963 signal->group_exit_code = 0;
964 }
965 }
966
967 return !sig_ignored(p, sig, force);
968}
969
970/*
971 * Test if P wants to take SIG. After we've checked all threads with this,
972 * it's equivalent to finding no threads not blocking SIG. Any threads not
973 * blocking SIG were ruled out because they are not running and already
974 * have pending signals. Such threads will dequeue from the shared queue
975 * as soon as they're available, so putting the signal on the shared queue
976 * will be equivalent to sending it to one such thread.
977 */
978static inline bool wants_signal(int sig, struct task_struct *p)
979{
980 if (sigismember(&p->blocked, sig))
981 return false;
982
983 if (p->flags & PF_EXITING)
984 return false;
985
986 if (sig == SIGKILL)
987 return true;
988
989 if (task_is_stopped_or_traced(p))
990 return false;
991
992 return task_curr(p) || !task_sigpending(p);
993}
994
995static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
996{
997 struct signal_struct *signal = p->signal;
998 struct task_struct *t;
999
1000 /*
1001 * Now find a thread we can wake up to take the signal off the queue.
1002 *
1003 * Try the suggested task first (may or may not be the main thread).
1004 */
1005 if (wants_signal(sig, p))
1006 t = p;
1007 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1008 /*
1009 * There is just one thread and it does not need to be woken.
1010 * It will dequeue unblocked signals before it runs again.
1011 */
1012 return;
1013 else {
1014 /*
1015 * Otherwise try to find a suitable thread.
1016 */
1017 t = signal->curr_target;
1018 while (!wants_signal(sig, t)) {
1019 t = next_thread(t);
1020 if (t == signal->curr_target)
1021 /*
1022 * No thread needs to be woken.
1023 * Any eligible threads will see
1024 * the signal in the queue soon.
1025 */
1026 return;
1027 }
1028 signal->curr_target = t;
1029 }
1030
1031 /*
1032 * Found a killable thread. If the signal will be fatal,
1033 * then start taking the whole group down immediately.
1034 */
1035 if (sig_fatal(p, sig) &&
1036 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1037 !sigismember(&t->real_blocked, sig) &&
1038 (sig == SIGKILL || !p->ptrace)) {
1039 /*
1040 * This signal will be fatal to the whole group.
1041 */
1042 if (!sig_kernel_coredump(sig)) {
1043 /*
1044 * Start a group exit and wake everybody up.
1045 * This way we don't have other threads
1046 * running and doing things after a slower
1047 * thread has the fatal signal pending.
1048 */
1049 signal->flags = SIGNAL_GROUP_EXIT;
1050 signal->group_exit_code = sig;
1051 signal->group_stop_count = 0;
1052 __for_each_thread(signal, t) {
1053 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1054 sigaddset(&t->pending.signal, SIGKILL);
1055 signal_wake_up(t, 1);
1056 }
1057 return;
1058 }
1059 }
1060
1061 /*
1062 * The signal is already in the shared-pending queue.
1063 * Tell the chosen thread to wake up and dequeue it.
1064 */
1065 signal_wake_up(t, sig == SIGKILL);
1066 return;
1067}
1068
1069static inline bool legacy_queue(struct sigpending *signals, int sig)
1070{
1071 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1072}
1073
1074static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1075 struct task_struct *t, enum pid_type type, bool force)
1076{
1077 struct sigpending *pending;
1078 struct sigqueue *q;
1079 int override_rlimit;
1080 int ret = 0, result;
1081
1082 lockdep_assert_held(&t->sighand->siglock);
1083
1084 result = TRACE_SIGNAL_IGNORED;
1085 if (!prepare_signal(sig, t, force))
1086 goto ret;
1087
1088 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1089 /*
1090 * Short-circuit ignored signals and support queuing
1091 * exactly one non-rt signal, so that we can get more
1092 * detailed information about the cause of the signal.
1093 */
1094 result = TRACE_SIGNAL_ALREADY_PENDING;
1095 if (legacy_queue(pending, sig))
1096 goto ret;
1097
1098 result = TRACE_SIGNAL_DELIVERED;
1099 /*
1100 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1101 */
1102 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1103 goto out_set;
1104
1105 /*
1106 * Real-time signals must be queued if sent by sigqueue, or
1107 * some other real-time mechanism. It is implementation
1108 * defined whether kill() does so. We attempt to do so, on
1109 * the principle of least surprise, but since kill is not
1110 * allowed to fail with EAGAIN when low on memory we just
1111 * make sure at least one signal gets delivered and don't
1112 * pass on the info struct.
1113 */
1114 if (sig < SIGRTMIN)
1115 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1116 else
1117 override_rlimit = 0;
1118
1119 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1120
1121 if (q) {
1122 list_add_tail(&q->list, &pending->list);
1123 switch ((unsigned long) info) {
1124 case (unsigned long) SEND_SIG_NOINFO:
1125 clear_siginfo(&q->info);
1126 q->info.si_signo = sig;
1127 q->info.si_errno = 0;
1128 q->info.si_code = SI_USER;
1129 q->info.si_pid = task_tgid_nr_ns(current,
1130 task_active_pid_ns(t));
1131 rcu_read_lock();
1132 q->info.si_uid =
1133 from_kuid_munged(task_cred_xxx(t, user_ns),
1134 current_uid());
1135 rcu_read_unlock();
1136 break;
1137 case (unsigned long) SEND_SIG_PRIV:
1138 clear_siginfo(&q->info);
1139 q->info.si_signo = sig;
1140 q->info.si_errno = 0;
1141 q->info.si_code = SI_KERNEL;
1142 q->info.si_pid = 0;
1143 q->info.si_uid = 0;
1144 break;
1145 default:
1146 copy_siginfo(&q->info, info);
1147 break;
1148 }
1149 } else if (!is_si_special(info) &&
1150 sig >= SIGRTMIN && info->si_code != SI_USER) {
1151 /*
1152 * Queue overflow, abort. We may abort if the
1153 * signal was rt and sent by user using something
1154 * other than kill().
1155 */
1156 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1157 ret = -EAGAIN;
1158 goto ret;
1159 } else {
1160 /*
1161 * This is a silent loss of information. We still
1162 * send the signal, but the *info bits are lost.
1163 */
1164 result = TRACE_SIGNAL_LOSE_INFO;
1165 }
1166
1167out_set:
1168 signalfd_notify(t, sig);
1169 sigaddset(&pending->signal, sig);
1170
1171 /* Let multiprocess signals appear after on-going forks */
1172 if (type > PIDTYPE_TGID) {
1173 struct multiprocess_signals *delayed;
1174 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1175 sigset_t *signal = &delayed->signal;
1176 /* Can't queue both a stop and a continue signal */
1177 if (sig == SIGCONT)
1178 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1179 else if (sig_kernel_stop(sig))
1180 sigdelset(signal, SIGCONT);
1181 sigaddset(signal, sig);
1182 }
1183 }
1184
1185 complete_signal(sig, t, type);
1186ret:
1187 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1188 return ret;
1189}
1190
1191static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1192{
1193 bool ret = false;
1194 switch (siginfo_layout(info->si_signo, info->si_code)) {
1195 case SIL_KILL:
1196 case SIL_CHLD:
1197 case SIL_RT:
1198 ret = true;
1199 break;
1200 case SIL_TIMER:
1201 case SIL_POLL:
1202 case SIL_FAULT:
1203 case SIL_FAULT_TRAPNO:
1204 case SIL_FAULT_MCEERR:
1205 case SIL_FAULT_BNDERR:
1206 case SIL_FAULT_PKUERR:
1207 case SIL_FAULT_PERF_EVENT:
1208 case SIL_SYS:
1209 ret = false;
1210 break;
1211 }
1212 return ret;
1213}
1214
1215int send_signal_locked(int sig, struct kernel_siginfo *info,
1216 struct task_struct *t, enum pid_type type)
1217{
1218 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1219 bool force = false;
1220
1221 if (info == SEND_SIG_NOINFO) {
1222 /* Force if sent from an ancestor pid namespace */
1223 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1224 } else if (info == SEND_SIG_PRIV) {
1225 /* Don't ignore kernel generated signals */
1226 force = true;
1227 } else if (has_si_pid_and_uid(info)) {
1228 /* SIGKILL and SIGSTOP is special or has ids */
1229 struct user_namespace *t_user_ns;
1230
1231 rcu_read_lock();
1232 t_user_ns = task_cred_xxx(t, user_ns);
1233 if (current_user_ns() != t_user_ns) {
1234 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1235 info->si_uid = from_kuid_munged(t_user_ns, uid);
1236 }
1237 rcu_read_unlock();
1238
1239 /* A kernel generated signal? */
1240 force = (info->si_code == SI_KERNEL);
1241
1242 /* From an ancestor pid namespace? */
1243 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1244 info->si_pid = 0;
1245 force = true;
1246 }
1247 }
1248 return __send_signal_locked(sig, info, t, type, force);
1249}
1250
1251static void print_fatal_signal(int signr)
1252{
1253 struct pt_regs *regs = task_pt_regs(current);
1254 struct file *exe_file;
1255
1256 exe_file = get_task_exe_file(current);
1257 if (exe_file) {
1258 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1259 exe_file, current->comm, signr);
1260 fput(exe_file);
1261 } else {
1262 pr_info("%s: potentially unexpected fatal signal %d.\n",
1263 current->comm, signr);
1264 }
1265
1266#if defined(__i386__) && !defined(__arch_um__)
1267 pr_info("code at %08lx: ", regs->ip);
1268 {
1269 int i;
1270 for (i = 0; i < 16; i++) {
1271 unsigned char insn;
1272
1273 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1274 break;
1275 pr_cont("%02x ", insn);
1276 }
1277 }
1278 pr_cont("\n");
1279#endif
1280 preempt_disable();
1281 show_regs(regs);
1282 preempt_enable();
1283}
1284
1285static int __init setup_print_fatal_signals(char *str)
1286{
1287 get_option (&str, &print_fatal_signals);
1288
1289 return 1;
1290}
1291
1292__setup("print-fatal-signals=", setup_print_fatal_signals);
1293
1294int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1295 enum pid_type type)
1296{
1297 unsigned long flags;
1298 int ret = -ESRCH;
1299
1300 if (lock_task_sighand(p, &flags)) {
1301 ret = send_signal_locked(sig, info, p, type);
1302 unlock_task_sighand(p, &flags);
1303 }
1304
1305 return ret;
1306}
1307
1308enum sig_handler {
1309 HANDLER_CURRENT, /* If reachable use the current handler */
1310 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1311 HANDLER_EXIT, /* Only visible as the process exit code */
1312};
1313
1314/*
1315 * Force a signal that the process can't ignore: if necessary
1316 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1317 *
1318 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1319 * since we do not want to have a signal handler that was blocked
1320 * be invoked when user space had explicitly blocked it.
1321 *
1322 * We don't want to have recursive SIGSEGV's etc, for example,
1323 * that is why we also clear SIGNAL_UNKILLABLE.
1324 */
1325static int
1326force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1327 enum sig_handler handler)
1328{
1329 unsigned long int flags;
1330 int ret, blocked, ignored;
1331 struct k_sigaction *action;
1332 int sig = info->si_signo;
1333
1334 spin_lock_irqsave(&t->sighand->siglock, flags);
1335 action = &t->sighand->action[sig-1];
1336 ignored = action->sa.sa_handler == SIG_IGN;
1337 blocked = sigismember(&t->blocked, sig);
1338 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1339 action->sa.sa_handler = SIG_DFL;
1340 if (handler == HANDLER_EXIT)
1341 action->sa.sa_flags |= SA_IMMUTABLE;
1342 if (blocked)
1343 sigdelset(&t->blocked, sig);
1344 }
1345 /*
1346 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1347 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1348 */
1349 if (action->sa.sa_handler == SIG_DFL &&
1350 (!t->ptrace || (handler == HANDLER_EXIT)))
1351 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1352 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1353 /* This can happen if the signal was already pending and blocked */
1354 if (!task_sigpending(t))
1355 signal_wake_up(t, 0);
1356 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1357
1358 return ret;
1359}
1360
1361int force_sig_info(struct kernel_siginfo *info)
1362{
1363 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1364}
1365
1366/*
1367 * Nuke all other threads in the group.
1368 */
1369int zap_other_threads(struct task_struct *p)
1370{
1371 struct task_struct *t;
1372 int count = 0;
1373
1374 p->signal->group_stop_count = 0;
1375
1376 for_other_threads(p, t) {
1377 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1378 /* Don't require de_thread to wait for the vhost_worker */
1379 if ((t->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)
1380 count++;
1381
1382 /* Don't bother with already dead threads */
1383 if (t->exit_state)
1384 continue;
1385 sigaddset(&t->pending.signal, SIGKILL);
1386 signal_wake_up(t, 1);
1387 }
1388
1389 return count;
1390}
1391
1392struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1393 unsigned long *flags)
1394{
1395 struct sighand_struct *sighand;
1396
1397 rcu_read_lock();
1398 for (;;) {
1399 sighand = rcu_dereference(tsk->sighand);
1400 if (unlikely(sighand == NULL))
1401 break;
1402
1403 /*
1404 * This sighand can be already freed and even reused, but
1405 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1406 * initializes ->siglock: this slab can't go away, it has
1407 * the same object type, ->siglock can't be reinitialized.
1408 *
1409 * We need to ensure that tsk->sighand is still the same
1410 * after we take the lock, we can race with de_thread() or
1411 * __exit_signal(). In the latter case the next iteration
1412 * must see ->sighand == NULL.
1413 */
1414 spin_lock_irqsave(&sighand->siglock, *flags);
1415 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1416 break;
1417 spin_unlock_irqrestore(&sighand->siglock, *flags);
1418 }
1419 rcu_read_unlock();
1420
1421 return sighand;
1422}
1423
1424#ifdef CONFIG_LOCKDEP
1425void lockdep_assert_task_sighand_held(struct task_struct *task)
1426{
1427 struct sighand_struct *sighand;
1428
1429 rcu_read_lock();
1430 sighand = rcu_dereference(task->sighand);
1431 if (sighand)
1432 lockdep_assert_held(&sighand->siglock);
1433 else
1434 WARN_ON_ONCE(1);
1435 rcu_read_unlock();
1436}
1437#endif
1438
1439/*
1440 * send signal info to all the members of a thread group or to the
1441 * individual thread if type == PIDTYPE_PID.
1442 */
1443int group_send_sig_info(int sig, struct kernel_siginfo *info,
1444 struct task_struct *p, enum pid_type type)
1445{
1446 int ret;
1447
1448 rcu_read_lock();
1449 ret = check_kill_permission(sig, info, p);
1450 rcu_read_unlock();
1451
1452 if (!ret && sig)
1453 ret = do_send_sig_info(sig, info, p, type);
1454
1455 return ret;
1456}
1457
1458/*
1459 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1460 * control characters do (^C, ^Z etc)
1461 * - the caller must hold at least a readlock on tasklist_lock
1462 */
1463int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1464{
1465 struct task_struct *p = NULL;
1466 int ret = -ESRCH;
1467
1468 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1469 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1470 /*
1471 * If group_send_sig_info() succeeds at least once ret
1472 * becomes 0 and after that the code below has no effect.
1473 * Otherwise we return the last err or -ESRCH if this
1474 * process group is empty.
1475 */
1476 if (ret)
1477 ret = err;
1478 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1479
1480 return ret;
1481}
1482
1483static int kill_pid_info_type(int sig, struct kernel_siginfo *info,
1484 struct pid *pid, enum pid_type type)
1485{
1486 int error = -ESRCH;
1487 struct task_struct *p;
1488
1489 for (;;) {
1490 rcu_read_lock();
1491 p = pid_task(pid, PIDTYPE_PID);
1492 if (p)
1493 error = group_send_sig_info(sig, info, p, type);
1494 rcu_read_unlock();
1495 if (likely(!p || error != -ESRCH))
1496 return error;
1497 /*
1498 * The task was unhashed in between, try again. If it
1499 * is dead, pid_task() will return NULL, if we race with
1500 * de_thread() it will find the new leader.
1501 */
1502 }
1503}
1504
1505int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1506{
1507 return kill_pid_info_type(sig, info, pid, PIDTYPE_TGID);
1508}
1509
1510static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1511{
1512 int error;
1513 rcu_read_lock();
1514 error = kill_pid_info(sig, info, find_vpid(pid));
1515 rcu_read_unlock();
1516 return error;
1517}
1518
1519static inline bool kill_as_cred_perm(const struct cred *cred,
1520 struct task_struct *target)
1521{
1522 const struct cred *pcred = __task_cred(target);
1523
1524 return uid_eq(cred->euid, pcred->suid) ||
1525 uid_eq(cred->euid, pcred->uid) ||
1526 uid_eq(cred->uid, pcred->suid) ||
1527 uid_eq(cred->uid, pcred->uid);
1528}
1529
1530/*
1531 * The usb asyncio usage of siginfo is wrong. The glibc support
1532 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1533 * AKA after the generic fields:
1534 * kernel_pid_t si_pid;
1535 * kernel_uid32_t si_uid;
1536 * sigval_t si_value;
1537 *
1538 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1539 * after the generic fields is:
1540 * void __user *si_addr;
1541 *
1542 * This is a practical problem when there is a 64bit big endian kernel
1543 * and a 32bit userspace. As the 32bit address will encoded in the low
1544 * 32bits of the pointer. Those low 32bits will be stored at higher
1545 * address than appear in a 32 bit pointer. So userspace will not
1546 * see the address it was expecting for it's completions.
1547 *
1548 * There is nothing in the encoding that can allow
1549 * copy_siginfo_to_user32 to detect this confusion of formats, so
1550 * handle this by requiring the caller of kill_pid_usb_asyncio to
1551 * notice when this situration takes place and to store the 32bit
1552 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1553 * parameter.
1554 */
1555int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1556 struct pid *pid, const struct cred *cred)
1557{
1558 struct kernel_siginfo info;
1559 struct task_struct *p;
1560 unsigned long flags;
1561 int ret = -EINVAL;
1562
1563 if (!valid_signal(sig))
1564 return ret;
1565
1566 clear_siginfo(&info);
1567 info.si_signo = sig;
1568 info.si_errno = errno;
1569 info.si_code = SI_ASYNCIO;
1570 *((sigval_t *)&info.si_pid) = addr;
1571
1572 rcu_read_lock();
1573 p = pid_task(pid, PIDTYPE_PID);
1574 if (!p) {
1575 ret = -ESRCH;
1576 goto out_unlock;
1577 }
1578 if (!kill_as_cred_perm(cred, p)) {
1579 ret = -EPERM;
1580 goto out_unlock;
1581 }
1582 ret = security_task_kill(p, &info, sig, cred);
1583 if (ret)
1584 goto out_unlock;
1585
1586 if (sig) {
1587 if (lock_task_sighand(p, &flags)) {
1588 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1589 unlock_task_sighand(p, &flags);
1590 } else
1591 ret = -ESRCH;
1592 }
1593out_unlock:
1594 rcu_read_unlock();
1595 return ret;
1596}
1597EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1598
1599/*
1600 * kill_something_info() interprets pid in interesting ways just like kill(2).
1601 *
1602 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1603 * is probably wrong. Should make it like BSD or SYSV.
1604 */
1605
1606static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1607{
1608 int ret;
1609
1610 if (pid > 0)
1611 return kill_proc_info(sig, info, pid);
1612
1613 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1614 if (pid == INT_MIN)
1615 return -ESRCH;
1616
1617 read_lock(&tasklist_lock);
1618 if (pid != -1) {
1619 ret = __kill_pgrp_info(sig, info,
1620 pid ? find_vpid(-pid) : task_pgrp(current));
1621 } else {
1622 int retval = 0, count = 0;
1623 struct task_struct * p;
1624
1625 for_each_process(p) {
1626 if (task_pid_vnr(p) > 1 &&
1627 !same_thread_group(p, current)) {
1628 int err = group_send_sig_info(sig, info, p,
1629 PIDTYPE_MAX);
1630 ++count;
1631 if (err != -EPERM)
1632 retval = err;
1633 }
1634 }
1635 ret = count ? retval : -ESRCH;
1636 }
1637 read_unlock(&tasklist_lock);
1638
1639 return ret;
1640}
1641
1642/*
1643 * These are for backward compatibility with the rest of the kernel source.
1644 */
1645
1646int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1647{
1648 /*
1649 * Make sure legacy kernel users don't send in bad values
1650 * (normal paths check this in check_kill_permission).
1651 */
1652 if (!valid_signal(sig))
1653 return -EINVAL;
1654
1655 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1656}
1657EXPORT_SYMBOL(send_sig_info);
1658
1659#define __si_special(priv) \
1660 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1661
1662int
1663send_sig(int sig, struct task_struct *p, int priv)
1664{
1665 return send_sig_info(sig, __si_special(priv), p);
1666}
1667EXPORT_SYMBOL(send_sig);
1668
1669void force_sig(int sig)
1670{
1671 struct kernel_siginfo info;
1672
1673 clear_siginfo(&info);
1674 info.si_signo = sig;
1675 info.si_errno = 0;
1676 info.si_code = SI_KERNEL;
1677 info.si_pid = 0;
1678 info.si_uid = 0;
1679 force_sig_info(&info);
1680}
1681EXPORT_SYMBOL(force_sig);
1682
1683void force_fatal_sig(int sig)
1684{
1685 struct kernel_siginfo info;
1686
1687 clear_siginfo(&info);
1688 info.si_signo = sig;
1689 info.si_errno = 0;
1690 info.si_code = SI_KERNEL;
1691 info.si_pid = 0;
1692 info.si_uid = 0;
1693 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1694}
1695
1696void force_exit_sig(int sig)
1697{
1698 struct kernel_siginfo info;
1699
1700 clear_siginfo(&info);
1701 info.si_signo = sig;
1702 info.si_errno = 0;
1703 info.si_code = SI_KERNEL;
1704 info.si_pid = 0;
1705 info.si_uid = 0;
1706 force_sig_info_to_task(&info, current, HANDLER_EXIT);
1707}
1708
1709/*
1710 * When things go south during signal handling, we
1711 * will force a SIGSEGV. And if the signal that caused
1712 * the problem was already a SIGSEGV, we'll want to
1713 * make sure we don't even try to deliver the signal..
1714 */
1715void force_sigsegv(int sig)
1716{
1717 if (sig == SIGSEGV)
1718 force_fatal_sig(SIGSEGV);
1719 else
1720 force_sig(SIGSEGV);
1721}
1722
1723int force_sig_fault_to_task(int sig, int code, void __user *addr,
1724 struct task_struct *t)
1725{
1726 struct kernel_siginfo info;
1727
1728 clear_siginfo(&info);
1729 info.si_signo = sig;
1730 info.si_errno = 0;
1731 info.si_code = code;
1732 info.si_addr = addr;
1733 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1734}
1735
1736int force_sig_fault(int sig, int code, void __user *addr)
1737{
1738 return force_sig_fault_to_task(sig, code, addr, current);
1739}
1740
1741int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t)
1742{
1743 struct kernel_siginfo info;
1744
1745 clear_siginfo(&info);
1746 info.si_signo = sig;
1747 info.si_errno = 0;
1748 info.si_code = code;
1749 info.si_addr = addr;
1750 return send_sig_info(info.si_signo, &info, t);
1751}
1752
1753int force_sig_mceerr(int code, void __user *addr, short lsb)
1754{
1755 struct kernel_siginfo info;
1756
1757 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1758 clear_siginfo(&info);
1759 info.si_signo = SIGBUS;
1760 info.si_errno = 0;
1761 info.si_code = code;
1762 info.si_addr = addr;
1763 info.si_addr_lsb = lsb;
1764 return force_sig_info(&info);
1765}
1766
1767int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1768{
1769 struct kernel_siginfo info;
1770
1771 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1772 clear_siginfo(&info);
1773 info.si_signo = SIGBUS;
1774 info.si_errno = 0;
1775 info.si_code = code;
1776 info.si_addr = addr;
1777 info.si_addr_lsb = lsb;
1778 return send_sig_info(info.si_signo, &info, t);
1779}
1780EXPORT_SYMBOL(send_sig_mceerr);
1781
1782int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1783{
1784 struct kernel_siginfo info;
1785
1786 clear_siginfo(&info);
1787 info.si_signo = SIGSEGV;
1788 info.si_errno = 0;
1789 info.si_code = SEGV_BNDERR;
1790 info.si_addr = addr;
1791 info.si_lower = lower;
1792 info.si_upper = upper;
1793 return force_sig_info(&info);
1794}
1795
1796#ifdef SEGV_PKUERR
1797int force_sig_pkuerr(void __user *addr, u32 pkey)
1798{
1799 struct kernel_siginfo info;
1800
1801 clear_siginfo(&info);
1802 info.si_signo = SIGSEGV;
1803 info.si_errno = 0;
1804 info.si_code = SEGV_PKUERR;
1805 info.si_addr = addr;
1806 info.si_pkey = pkey;
1807 return force_sig_info(&info);
1808}
1809#endif
1810
1811int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1812{
1813 struct kernel_siginfo info;
1814
1815 clear_siginfo(&info);
1816 info.si_signo = SIGTRAP;
1817 info.si_errno = 0;
1818 info.si_code = TRAP_PERF;
1819 info.si_addr = addr;
1820 info.si_perf_data = sig_data;
1821 info.si_perf_type = type;
1822
1823 /*
1824 * Signals generated by perf events should not terminate the whole
1825 * process if SIGTRAP is blocked, however, delivering the signal
1826 * asynchronously is better than not delivering at all. But tell user
1827 * space if the signal was asynchronous, so it can clearly be
1828 * distinguished from normal synchronous ones.
1829 */
1830 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1831 TRAP_PERF_FLAG_ASYNC :
1832 0;
1833
1834 return send_sig_info(info.si_signo, &info, current);
1835}
1836
1837/**
1838 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1839 * @syscall: syscall number to send to userland
1840 * @reason: filter-supplied reason code to send to userland (via si_errno)
1841 * @force_coredump: true to trigger a coredump
1842 *
1843 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1844 */
1845int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1846{
1847 struct kernel_siginfo info;
1848
1849 clear_siginfo(&info);
1850 info.si_signo = SIGSYS;
1851 info.si_code = SYS_SECCOMP;
1852 info.si_call_addr = (void __user *)KSTK_EIP(current);
1853 info.si_errno = reason;
1854 info.si_arch = syscall_get_arch(current);
1855 info.si_syscall = syscall;
1856 return force_sig_info_to_task(&info, current,
1857 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1858}
1859
1860/* For the crazy architectures that include trap information in
1861 * the errno field, instead of an actual errno value.
1862 */
1863int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1864{
1865 struct kernel_siginfo info;
1866
1867 clear_siginfo(&info);
1868 info.si_signo = SIGTRAP;
1869 info.si_errno = errno;
1870 info.si_code = TRAP_HWBKPT;
1871 info.si_addr = addr;
1872 return force_sig_info(&info);
1873}
1874
1875/* For the rare architectures that include trap information using
1876 * si_trapno.
1877 */
1878int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1879{
1880 struct kernel_siginfo info;
1881
1882 clear_siginfo(&info);
1883 info.si_signo = sig;
1884 info.si_errno = 0;
1885 info.si_code = code;
1886 info.si_addr = addr;
1887 info.si_trapno = trapno;
1888 return force_sig_info(&info);
1889}
1890
1891/* For the rare architectures that include trap information using
1892 * si_trapno.
1893 */
1894int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1895 struct task_struct *t)
1896{
1897 struct kernel_siginfo info;
1898
1899 clear_siginfo(&info);
1900 info.si_signo = sig;
1901 info.si_errno = 0;
1902 info.si_code = code;
1903 info.si_addr = addr;
1904 info.si_trapno = trapno;
1905 return send_sig_info(info.si_signo, &info, t);
1906}
1907
1908static int kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1909{
1910 int ret;
1911 read_lock(&tasklist_lock);
1912 ret = __kill_pgrp_info(sig, info, pgrp);
1913 read_unlock(&tasklist_lock);
1914 return ret;
1915}
1916
1917int kill_pgrp(struct pid *pid, int sig, int priv)
1918{
1919 return kill_pgrp_info(sig, __si_special(priv), pid);
1920}
1921EXPORT_SYMBOL(kill_pgrp);
1922
1923int kill_pid(struct pid *pid, int sig, int priv)
1924{
1925 return kill_pid_info(sig, __si_special(priv), pid);
1926}
1927EXPORT_SYMBOL(kill_pid);
1928
1929/*
1930 * These functions support sending signals using preallocated sigqueue
1931 * structures. This is needed "because realtime applications cannot
1932 * afford to lose notifications of asynchronous events, like timer
1933 * expirations or I/O completions". In the case of POSIX Timers
1934 * we allocate the sigqueue structure from the timer_create. If this
1935 * allocation fails we are able to report the failure to the application
1936 * with an EAGAIN error.
1937 */
1938struct sigqueue *sigqueue_alloc(void)
1939{
1940 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1941}
1942
1943void sigqueue_free(struct sigqueue *q)
1944{
1945 unsigned long flags;
1946 spinlock_t *lock = ¤t->sighand->siglock;
1947
1948 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1949 /*
1950 * We must hold ->siglock while testing q->list
1951 * to serialize with collect_signal() or with
1952 * __exit_signal()->flush_sigqueue().
1953 */
1954 spin_lock_irqsave(lock, flags);
1955 q->flags &= ~SIGQUEUE_PREALLOC;
1956 /*
1957 * If it is queued it will be freed when dequeued,
1958 * like the "regular" sigqueue.
1959 */
1960 if (!list_empty(&q->list))
1961 q = NULL;
1962 spin_unlock_irqrestore(lock, flags);
1963
1964 if (q)
1965 __sigqueue_free(q);
1966}
1967
1968int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1969{
1970 int sig = q->info.si_signo;
1971 struct sigpending *pending;
1972 struct task_struct *t;
1973 unsigned long flags;
1974 int ret, result;
1975
1976 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1977
1978 ret = -1;
1979 rcu_read_lock();
1980
1981 /*
1982 * This function is used by POSIX timers to deliver a timer signal.
1983 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1984 * set), the signal must be delivered to the specific thread (queues
1985 * into t->pending).
1986 *
1987 * Where type is not PIDTYPE_PID, signals must be delivered to the
1988 * process. In this case, prefer to deliver to current if it is in
1989 * the same thread group as the target process, which avoids
1990 * unnecessarily waking up a potentially idle task.
1991 */
1992 t = pid_task(pid, type);
1993 if (!t)
1994 goto ret;
1995 if (type != PIDTYPE_PID && same_thread_group(t, current))
1996 t = current;
1997 if (!likely(lock_task_sighand(t, &flags)))
1998 goto ret;
1999
2000 ret = 1; /* the signal is ignored */
2001 result = TRACE_SIGNAL_IGNORED;
2002 if (!prepare_signal(sig, t, false))
2003 goto out;
2004
2005 ret = 0;
2006 if (unlikely(!list_empty(&q->list))) {
2007 /*
2008 * If an SI_TIMER entry is already queue just increment
2009 * the overrun count.
2010 */
2011 BUG_ON(q->info.si_code != SI_TIMER);
2012 q->info.si_overrun++;
2013 result = TRACE_SIGNAL_ALREADY_PENDING;
2014 goto out;
2015 }
2016 q->info.si_overrun = 0;
2017
2018 signalfd_notify(t, sig);
2019 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
2020 list_add_tail(&q->list, &pending->list);
2021 sigaddset(&pending->signal, sig);
2022 complete_signal(sig, t, type);
2023 result = TRACE_SIGNAL_DELIVERED;
2024out:
2025 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
2026 unlock_task_sighand(t, &flags);
2027ret:
2028 rcu_read_unlock();
2029 return ret;
2030}
2031
2032void do_notify_pidfd(struct task_struct *task)
2033{
2034 struct pid *pid = task_pid(task);
2035
2036 WARN_ON(task->exit_state == 0);
2037
2038 __wake_up(&pid->wait_pidfd, TASK_NORMAL, 0,
2039 poll_to_key(EPOLLIN | EPOLLRDNORM));
2040}
2041
2042/*
2043 * Let a parent know about the death of a child.
2044 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2045 *
2046 * Returns true if our parent ignored us and so we've switched to
2047 * self-reaping.
2048 */
2049bool do_notify_parent(struct task_struct *tsk, int sig)
2050{
2051 struct kernel_siginfo info;
2052 unsigned long flags;
2053 struct sighand_struct *psig;
2054 bool autoreap = false;
2055 u64 utime, stime;
2056
2057 WARN_ON_ONCE(sig == -1);
2058
2059 /* do_notify_parent_cldstop should have been called instead. */
2060 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2061
2062 WARN_ON_ONCE(!tsk->ptrace &&
2063 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2064 /*
2065 * tsk is a group leader and has no threads, wake up the
2066 * non-PIDFD_THREAD waiters.
2067 */
2068 if (thread_group_empty(tsk))
2069 do_notify_pidfd(tsk);
2070
2071 if (sig != SIGCHLD) {
2072 /*
2073 * This is only possible if parent == real_parent.
2074 * Check if it has changed security domain.
2075 */
2076 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2077 sig = SIGCHLD;
2078 }
2079
2080 clear_siginfo(&info);
2081 info.si_signo = sig;
2082 info.si_errno = 0;
2083 /*
2084 * We are under tasklist_lock here so our parent is tied to
2085 * us and cannot change.
2086 *
2087 * task_active_pid_ns will always return the same pid namespace
2088 * until a task passes through release_task.
2089 *
2090 * write_lock() currently calls preempt_disable() which is the
2091 * same as rcu_read_lock(), but according to Oleg, this is not
2092 * correct to rely on this
2093 */
2094 rcu_read_lock();
2095 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2096 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2097 task_uid(tsk));
2098 rcu_read_unlock();
2099
2100 task_cputime(tsk, &utime, &stime);
2101 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2102 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2103
2104 info.si_status = tsk->exit_code & 0x7f;
2105 if (tsk->exit_code & 0x80)
2106 info.si_code = CLD_DUMPED;
2107 else if (tsk->exit_code & 0x7f)
2108 info.si_code = CLD_KILLED;
2109 else {
2110 info.si_code = CLD_EXITED;
2111 info.si_status = tsk->exit_code >> 8;
2112 }
2113
2114 psig = tsk->parent->sighand;
2115 spin_lock_irqsave(&psig->siglock, flags);
2116 if (!tsk->ptrace && sig == SIGCHLD &&
2117 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2118 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2119 /*
2120 * We are exiting and our parent doesn't care. POSIX.1
2121 * defines special semantics for setting SIGCHLD to SIG_IGN
2122 * or setting the SA_NOCLDWAIT flag: we should be reaped
2123 * automatically and not left for our parent's wait4 call.
2124 * Rather than having the parent do it as a magic kind of
2125 * signal handler, we just set this to tell do_exit that we
2126 * can be cleaned up without becoming a zombie. Note that
2127 * we still call __wake_up_parent in this case, because a
2128 * blocked sys_wait4 might now return -ECHILD.
2129 *
2130 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2131 * is implementation-defined: we do (if you don't want
2132 * it, just use SIG_IGN instead).
2133 */
2134 autoreap = true;
2135 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2136 sig = 0;
2137 }
2138 /*
2139 * Send with __send_signal as si_pid and si_uid are in the
2140 * parent's namespaces.
2141 */
2142 if (valid_signal(sig) && sig)
2143 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2144 __wake_up_parent(tsk, tsk->parent);
2145 spin_unlock_irqrestore(&psig->siglock, flags);
2146
2147 return autoreap;
2148}
2149
2150/**
2151 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2152 * @tsk: task reporting the state change
2153 * @for_ptracer: the notification is for ptracer
2154 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2155 *
2156 * Notify @tsk's parent that the stopped/continued state has changed. If
2157 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2158 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2159 *
2160 * CONTEXT:
2161 * Must be called with tasklist_lock at least read locked.
2162 */
2163static void do_notify_parent_cldstop(struct task_struct *tsk,
2164 bool for_ptracer, int why)
2165{
2166 struct kernel_siginfo info;
2167 unsigned long flags;
2168 struct task_struct *parent;
2169 struct sighand_struct *sighand;
2170 u64 utime, stime;
2171
2172 if (for_ptracer) {
2173 parent = tsk->parent;
2174 } else {
2175 tsk = tsk->group_leader;
2176 parent = tsk->real_parent;
2177 }
2178
2179 clear_siginfo(&info);
2180 info.si_signo = SIGCHLD;
2181 info.si_errno = 0;
2182 /*
2183 * see comment in do_notify_parent() about the following 4 lines
2184 */
2185 rcu_read_lock();
2186 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2187 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2188 rcu_read_unlock();
2189
2190 task_cputime(tsk, &utime, &stime);
2191 info.si_utime = nsec_to_clock_t(utime);
2192 info.si_stime = nsec_to_clock_t(stime);
2193
2194 info.si_code = why;
2195 switch (why) {
2196 case CLD_CONTINUED:
2197 info.si_status = SIGCONT;
2198 break;
2199 case CLD_STOPPED:
2200 info.si_status = tsk->signal->group_exit_code & 0x7f;
2201 break;
2202 case CLD_TRAPPED:
2203 info.si_status = tsk->exit_code & 0x7f;
2204 break;
2205 default:
2206 BUG();
2207 }
2208
2209 sighand = parent->sighand;
2210 spin_lock_irqsave(&sighand->siglock, flags);
2211 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2212 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2213 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2214 /*
2215 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2216 */
2217 __wake_up_parent(tsk, parent);
2218 spin_unlock_irqrestore(&sighand->siglock, flags);
2219}
2220
2221/*
2222 * This must be called with current->sighand->siglock held.
2223 *
2224 * This should be the path for all ptrace stops.
2225 * We always set current->last_siginfo while stopped here.
2226 * That makes it a way to test a stopped process for
2227 * being ptrace-stopped vs being job-control-stopped.
2228 *
2229 * Returns the signal the ptracer requested the code resume
2230 * with. If the code did not stop because the tracer is gone,
2231 * the stop signal remains unchanged unless clear_code.
2232 */
2233static int ptrace_stop(int exit_code, int why, unsigned long message,
2234 kernel_siginfo_t *info)
2235 __releases(¤t->sighand->siglock)
2236 __acquires(¤t->sighand->siglock)
2237{
2238 bool gstop_done = false;
2239
2240 if (arch_ptrace_stop_needed()) {
2241 /*
2242 * The arch code has something special to do before a
2243 * ptrace stop. This is allowed to block, e.g. for faults
2244 * on user stack pages. We can't keep the siglock while
2245 * calling arch_ptrace_stop, so we must release it now.
2246 * To preserve proper semantics, we must do this before
2247 * any signal bookkeeping like checking group_stop_count.
2248 */
2249 spin_unlock_irq(¤t->sighand->siglock);
2250 arch_ptrace_stop();
2251 spin_lock_irq(¤t->sighand->siglock);
2252 }
2253
2254 /*
2255 * After this point ptrace_signal_wake_up or signal_wake_up
2256 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2257 * signal comes in. Handle previous ptrace_unlinks and fatal
2258 * signals here to prevent ptrace_stop sleeping in schedule.
2259 */
2260 if (!current->ptrace || __fatal_signal_pending(current))
2261 return exit_code;
2262
2263 set_special_state(TASK_TRACED);
2264 current->jobctl |= JOBCTL_TRACED;
2265
2266 /*
2267 * We're committing to trapping. TRACED should be visible before
2268 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2269 * Also, transition to TRACED and updates to ->jobctl should be
2270 * atomic with respect to siglock and should be done after the arch
2271 * hook as siglock is released and regrabbed across it.
2272 *
2273 * TRACER TRACEE
2274 *
2275 * ptrace_attach()
2276 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2277 * do_wait()
2278 * set_current_state() smp_wmb();
2279 * ptrace_do_wait()
2280 * wait_task_stopped()
2281 * task_stopped_code()
2282 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2283 */
2284 smp_wmb();
2285
2286 current->ptrace_message = message;
2287 current->last_siginfo = info;
2288 current->exit_code = exit_code;
2289
2290 /*
2291 * If @why is CLD_STOPPED, we're trapping to participate in a group
2292 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2293 * across siglock relocks since INTERRUPT was scheduled, PENDING
2294 * could be clear now. We act as if SIGCONT is received after
2295 * TASK_TRACED is entered - ignore it.
2296 */
2297 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2298 gstop_done = task_participate_group_stop(current);
2299
2300 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2301 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2302 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2303 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2304
2305 /* entering a trap, clear TRAPPING */
2306 task_clear_jobctl_trapping(current);
2307
2308 spin_unlock_irq(¤t->sighand->siglock);
2309 read_lock(&tasklist_lock);
2310 /*
2311 * Notify parents of the stop.
2312 *
2313 * While ptraced, there are two parents - the ptracer and
2314 * the real_parent of the group_leader. The ptracer should
2315 * know about every stop while the real parent is only
2316 * interested in the completion of group stop. The states
2317 * for the two don't interact with each other. Notify
2318 * separately unless they're gonna be duplicates.
2319 */
2320 if (current->ptrace)
2321 do_notify_parent_cldstop(current, true, why);
2322 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2323 do_notify_parent_cldstop(current, false, why);
2324
2325 /*
2326 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2327 * One a PREEMPTION kernel this can result in preemption requirement
2328 * which will be fulfilled after read_unlock() and the ptracer will be
2329 * put on the CPU.
2330 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2331 * this task wait in schedule(). If this task gets preempted then it
2332 * remains enqueued on the runqueue. The ptracer will observe this and
2333 * then sleep for a delay of one HZ tick. In the meantime this task
2334 * gets scheduled, enters schedule() and will wait for the ptracer.
2335 *
2336 * This preemption point is not bad from a correctness point of
2337 * view but extends the runtime by one HZ tick time due to the
2338 * ptracer's sleep. The preempt-disable section ensures that there
2339 * will be no preemption between unlock and schedule() and so
2340 * improving the performance since the ptracer will observe that
2341 * the tracee is scheduled out once it gets on the CPU.
2342 *
2343 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2344 * Therefore the task can be preempted after do_notify_parent_cldstop()
2345 * before unlocking tasklist_lock so there is no benefit in doing this.
2346 *
2347 * In fact disabling preemption is harmful on PREEMPT_RT because
2348 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2349 * with preemption disabled due to the 'sleeping' spinlock
2350 * substitution of RT.
2351 */
2352 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2353 preempt_disable();
2354 read_unlock(&tasklist_lock);
2355 cgroup_enter_frozen();
2356 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2357 preempt_enable_no_resched();
2358 schedule();
2359 cgroup_leave_frozen(true);
2360
2361 /*
2362 * We are back. Now reacquire the siglock before touching
2363 * last_siginfo, so that we are sure to have synchronized with
2364 * any signal-sending on another CPU that wants to examine it.
2365 */
2366 spin_lock_irq(¤t->sighand->siglock);
2367 exit_code = current->exit_code;
2368 current->last_siginfo = NULL;
2369 current->ptrace_message = 0;
2370 current->exit_code = 0;
2371
2372 /* LISTENING can be set only during STOP traps, clear it */
2373 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2374
2375 /*
2376 * Queued signals ignored us while we were stopped for tracing.
2377 * So check for any that we should take before resuming user mode.
2378 * This sets TIF_SIGPENDING, but never clears it.
2379 */
2380 recalc_sigpending_tsk(current);
2381 return exit_code;
2382}
2383
2384static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2385{
2386 kernel_siginfo_t info;
2387
2388 clear_siginfo(&info);
2389 info.si_signo = signr;
2390 info.si_code = exit_code;
2391 info.si_pid = task_pid_vnr(current);
2392 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2393
2394 /* Let the debugger run. */
2395 return ptrace_stop(exit_code, why, message, &info);
2396}
2397
2398int ptrace_notify(int exit_code, unsigned long message)
2399{
2400 int signr;
2401
2402 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2403 if (unlikely(task_work_pending(current)))
2404 task_work_run();
2405
2406 spin_lock_irq(¤t->sighand->siglock);
2407 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2408 spin_unlock_irq(¤t->sighand->siglock);
2409 return signr;
2410}
2411
2412/**
2413 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2414 * @signr: signr causing group stop if initiating
2415 *
2416 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2417 * and participate in it. If already set, participate in the existing
2418 * group stop. If participated in a group stop (and thus slept), %true is
2419 * returned with siglock released.
2420 *
2421 * If ptraced, this function doesn't handle stop itself. Instead,
2422 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2423 * untouched. The caller must ensure that INTERRUPT trap handling takes
2424 * places afterwards.
2425 *
2426 * CONTEXT:
2427 * Must be called with @current->sighand->siglock held, which is released
2428 * on %true return.
2429 *
2430 * RETURNS:
2431 * %false if group stop is already cancelled or ptrace trap is scheduled.
2432 * %true if participated in group stop.
2433 */
2434static bool do_signal_stop(int signr)
2435 __releases(¤t->sighand->siglock)
2436{
2437 struct signal_struct *sig = current->signal;
2438
2439 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2440 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2441 struct task_struct *t;
2442
2443 /* signr will be recorded in task->jobctl for retries */
2444 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2445
2446 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2447 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2448 unlikely(sig->group_exec_task))
2449 return false;
2450 /*
2451 * There is no group stop already in progress. We must
2452 * initiate one now.
2453 *
2454 * While ptraced, a task may be resumed while group stop is
2455 * still in effect and then receive a stop signal and
2456 * initiate another group stop. This deviates from the
2457 * usual behavior as two consecutive stop signals can't
2458 * cause two group stops when !ptraced. That is why we
2459 * also check !task_is_stopped(t) below.
2460 *
2461 * The condition can be distinguished by testing whether
2462 * SIGNAL_STOP_STOPPED is already set. Don't generate
2463 * group_exit_code in such case.
2464 *
2465 * This is not necessary for SIGNAL_STOP_CONTINUED because
2466 * an intervening stop signal is required to cause two
2467 * continued events regardless of ptrace.
2468 */
2469 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2470 sig->group_exit_code = signr;
2471
2472 sig->group_stop_count = 0;
2473 if (task_set_jobctl_pending(current, signr | gstop))
2474 sig->group_stop_count++;
2475
2476 for_other_threads(current, t) {
2477 /*
2478 * Setting state to TASK_STOPPED for a group
2479 * stop is always done with the siglock held,
2480 * so this check has no races.
2481 */
2482 if (!task_is_stopped(t) &&
2483 task_set_jobctl_pending(t, signr | gstop)) {
2484 sig->group_stop_count++;
2485 if (likely(!(t->ptrace & PT_SEIZED)))
2486 signal_wake_up(t, 0);
2487 else
2488 ptrace_trap_notify(t);
2489 }
2490 }
2491 }
2492
2493 if (likely(!current->ptrace)) {
2494 int notify = 0;
2495
2496 /*
2497 * If there are no other threads in the group, or if there
2498 * is a group stop in progress and we are the last to stop,
2499 * report to the parent.
2500 */
2501 if (task_participate_group_stop(current))
2502 notify = CLD_STOPPED;
2503
2504 current->jobctl |= JOBCTL_STOPPED;
2505 set_special_state(TASK_STOPPED);
2506 spin_unlock_irq(¤t->sighand->siglock);
2507
2508 /*
2509 * Notify the parent of the group stop completion. Because
2510 * we're not holding either the siglock or tasklist_lock
2511 * here, ptracer may attach inbetween; however, this is for
2512 * group stop and should always be delivered to the real
2513 * parent of the group leader. The new ptracer will get
2514 * its notification when this task transitions into
2515 * TASK_TRACED.
2516 */
2517 if (notify) {
2518 read_lock(&tasklist_lock);
2519 do_notify_parent_cldstop(current, false, notify);
2520 read_unlock(&tasklist_lock);
2521 }
2522
2523 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2524 cgroup_enter_frozen();
2525 schedule();
2526 return true;
2527 } else {
2528 /*
2529 * While ptraced, group stop is handled by STOP trap.
2530 * Schedule it and let the caller deal with it.
2531 */
2532 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2533 return false;
2534 }
2535}
2536
2537/**
2538 * do_jobctl_trap - take care of ptrace jobctl traps
2539 *
2540 * When PT_SEIZED, it's used for both group stop and explicit
2541 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2542 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2543 * the stop signal; otherwise, %SIGTRAP.
2544 *
2545 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2546 * number as exit_code and no siginfo.
2547 *
2548 * CONTEXT:
2549 * Must be called with @current->sighand->siglock held, which may be
2550 * released and re-acquired before returning with intervening sleep.
2551 */
2552static void do_jobctl_trap(void)
2553{
2554 struct signal_struct *signal = current->signal;
2555 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2556
2557 if (current->ptrace & PT_SEIZED) {
2558 if (!signal->group_stop_count &&
2559 !(signal->flags & SIGNAL_STOP_STOPPED))
2560 signr = SIGTRAP;
2561 WARN_ON_ONCE(!signr);
2562 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2563 CLD_STOPPED, 0);
2564 } else {
2565 WARN_ON_ONCE(!signr);
2566 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2567 }
2568}
2569
2570/**
2571 * do_freezer_trap - handle the freezer jobctl trap
2572 *
2573 * Puts the task into frozen state, if only the task is not about to quit.
2574 * In this case it drops JOBCTL_TRAP_FREEZE.
2575 *
2576 * CONTEXT:
2577 * Must be called with @current->sighand->siglock held,
2578 * which is always released before returning.
2579 */
2580static void do_freezer_trap(void)
2581 __releases(¤t->sighand->siglock)
2582{
2583 /*
2584 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2585 * let's make another loop to give it a chance to be handled.
2586 * In any case, we'll return back.
2587 */
2588 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2589 JOBCTL_TRAP_FREEZE) {
2590 spin_unlock_irq(¤t->sighand->siglock);
2591 return;
2592 }
2593
2594 /*
2595 * Now we're sure that there is no pending fatal signal and no
2596 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2597 * immediately (if there is a non-fatal signal pending), and
2598 * put the task into sleep.
2599 */
2600 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2601 clear_thread_flag(TIF_SIGPENDING);
2602 spin_unlock_irq(¤t->sighand->siglock);
2603 cgroup_enter_frozen();
2604 schedule();
2605}
2606
2607static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2608{
2609 /*
2610 * We do not check sig_kernel_stop(signr) but set this marker
2611 * unconditionally because we do not know whether debugger will
2612 * change signr. This flag has no meaning unless we are going
2613 * to stop after return from ptrace_stop(). In this case it will
2614 * be checked in do_signal_stop(), we should only stop if it was
2615 * not cleared by SIGCONT while we were sleeping. See also the
2616 * comment in dequeue_signal().
2617 */
2618 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2619 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2620
2621 /* We're back. Did the debugger cancel the sig? */
2622 if (signr == 0)
2623 return signr;
2624
2625 /*
2626 * Update the siginfo structure if the signal has
2627 * changed. If the debugger wanted something
2628 * specific in the siginfo structure then it should
2629 * have updated *info via PTRACE_SETSIGINFO.
2630 */
2631 if (signr != info->si_signo) {
2632 clear_siginfo(info);
2633 info->si_signo = signr;
2634 info->si_errno = 0;
2635 info->si_code = SI_USER;
2636 rcu_read_lock();
2637 info->si_pid = task_pid_vnr(current->parent);
2638 info->si_uid = from_kuid_munged(current_user_ns(),
2639 task_uid(current->parent));
2640 rcu_read_unlock();
2641 }
2642
2643 /* If the (new) signal is now blocked, requeue it. */
2644 if (sigismember(¤t->blocked, signr) ||
2645 fatal_signal_pending(current)) {
2646 send_signal_locked(signr, info, current, type);
2647 signr = 0;
2648 }
2649
2650 return signr;
2651}
2652
2653static void hide_si_addr_tag_bits(struct ksignal *ksig)
2654{
2655 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2656 case SIL_FAULT:
2657 case SIL_FAULT_TRAPNO:
2658 case SIL_FAULT_MCEERR:
2659 case SIL_FAULT_BNDERR:
2660 case SIL_FAULT_PKUERR:
2661 case SIL_FAULT_PERF_EVENT:
2662 ksig->info.si_addr = arch_untagged_si_addr(
2663 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2664 break;
2665 case SIL_KILL:
2666 case SIL_TIMER:
2667 case SIL_POLL:
2668 case SIL_CHLD:
2669 case SIL_RT:
2670 case SIL_SYS:
2671 break;
2672 }
2673}
2674
2675bool get_signal(struct ksignal *ksig)
2676{
2677 struct sighand_struct *sighand = current->sighand;
2678 struct signal_struct *signal = current->signal;
2679 int signr;
2680
2681 clear_notify_signal();
2682 if (unlikely(task_work_pending(current)))
2683 task_work_run();
2684
2685 if (!task_sigpending(current))
2686 return false;
2687
2688 if (unlikely(uprobe_deny_signal()))
2689 return false;
2690
2691 /*
2692 * Do this once, we can't return to user-mode if freezing() == T.
2693 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2694 * thus do not need another check after return.
2695 */
2696 try_to_freeze();
2697
2698relock:
2699 spin_lock_irq(&sighand->siglock);
2700
2701 /*
2702 * Every stopped thread goes here after wakeup. Check to see if
2703 * we should notify the parent, prepare_signal(SIGCONT) encodes
2704 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2705 */
2706 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2707 int why;
2708
2709 if (signal->flags & SIGNAL_CLD_CONTINUED)
2710 why = CLD_CONTINUED;
2711 else
2712 why = CLD_STOPPED;
2713
2714 signal->flags &= ~SIGNAL_CLD_MASK;
2715
2716 spin_unlock_irq(&sighand->siglock);
2717
2718 /*
2719 * Notify the parent that we're continuing. This event is
2720 * always per-process and doesn't make whole lot of sense
2721 * for ptracers, who shouldn't consume the state via
2722 * wait(2) either, but, for backward compatibility, notify
2723 * the ptracer of the group leader too unless it's gonna be
2724 * a duplicate.
2725 */
2726 read_lock(&tasklist_lock);
2727 do_notify_parent_cldstop(current, false, why);
2728
2729 if (ptrace_reparented(current->group_leader))
2730 do_notify_parent_cldstop(current->group_leader,
2731 true, why);
2732 read_unlock(&tasklist_lock);
2733
2734 goto relock;
2735 }
2736
2737 for (;;) {
2738 struct k_sigaction *ka;
2739 enum pid_type type;
2740
2741 /* Has this task already been marked for death? */
2742 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2743 signal->group_exec_task) {
2744 signr = SIGKILL;
2745 sigdelset(¤t->pending.signal, SIGKILL);
2746 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2747 &sighand->action[SIGKILL-1]);
2748 recalc_sigpending();
2749 /*
2750 * implies do_group_exit() or return to PF_USER_WORKER,
2751 * no need to initialize ksig->info/etc.
2752 */
2753 goto fatal;
2754 }
2755
2756 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2757 do_signal_stop(0))
2758 goto relock;
2759
2760 if (unlikely(current->jobctl &
2761 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2762 if (current->jobctl & JOBCTL_TRAP_MASK) {
2763 do_jobctl_trap();
2764 spin_unlock_irq(&sighand->siglock);
2765 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2766 do_freezer_trap();
2767
2768 goto relock;
2769 }
2770
2771 /*
2772 * If the task is leaving the frozen state, let's update
2773 * cgroup counters and reset the frozen bit.
2774 */
2775 if (unlikely(cgroup_task_frozen(current))) {
2776 spin_unlock_irq(&sighand->siglock);
2777 cgroup_leave_frozen(false);
2778 goto relock;
2779 }
2780
2781 /*
2782 * Signals generated by the execution of an instruction
2783 * need to be delivered before any other pending signals
2784 * so that the instruction pointer in the signal stack
2785 * frame points to the faulting instruction.
2786 */
2787 type = PIDTYPE_PID;
2788 signr = dequeue_synchronous_signal(&ksig->info);
2789 if (!signr)
2790 signr = dequeue_signal(current, ¤t->blocked,
2791 &ksig->info, &type);
2792
2793 if (!signr)
2794 break; /* will return 0 */
2795
2796 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2797 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2798 signr = ptrace_signal(signr, &ksig->info, type);
2799 if (!signr)
2800 continue;
2801 }
2802
2803 ka = &sighand->action[signr-1];
2804
2805 /* Trace actually delivered signals. */
2806 trace_signal_deliver(signr, &ksig->info, ka);
2807
2808 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2809 continue;
2810 if (ka->sa.sa_handler != SIG_DFL) {
2811 /* Run the handler. */
2812 ksig->ka = *ka;
2813
2814 if (ka->sa.sa_flags & SA_ONESHOT)
2815 ka->sa.sa_handler = SIG_DFL;
2816
2817 break; /* will return non-zero "signr" value */
2818 }
2819
2820 /*
2821 * Now we are doing the default action for this signal.
2822 */
2823 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2824 continue;
2825
2826 /*
2827 * Global init gets no signals it doesn't want.
2828 * Container-init gets no signals it doesn't want from same
2829 * container.
2830 *
2831 * Note that if global/container-init sees a sig_kernel_only()
2832 * signal here, the signal must have been generated internally
2833 * or must have come from an ancestor namespace. In either
2834 * case, the signal cannot be dropped.
2835 */
2836 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2837 !sig_kernel_only(signr))
2838 continue;
2839
2840 if (sig_kernel_stop(signr)) {
2841 /*
2842 * The default action is to stop all threads in
2843 * the thread group. The job control signals
2844 * do nothing in an orphaned pgrp, but SIGSTOP
2845 * always works. Note that siglock needs to be
2846 * dropped during the call to is_orphaned_pgrp()
2847 * because of lock ordering with tasklist_lock.
2848 * This allows an intervening SIGCONT to be posted.
2849 * We need to check for that and bail out if necessary.
2850 */
2851 if (signr != SIGSTOP) {
2852 spin_unlock_irq(&sighand->siglock);
2853
2854 /* signals can be posted during this window */
2855
2856 if (is_current_pgrp_orphaned())
2857 goto relock;
2858
2859 spin_lock_irq(&sighand->siglock);
2860 }
2861
2862 if (likely(do_signal_stop(signr))) {
2863 /* It released the siglock. */
2864 goto relock;
2865 }
2866
2867 /*
2868 * We didn't actually stop, due to a race
2869 * with SIGCONT or something like that.
2870 */
2871 continue;
2872 }
2873
2874 fatal:
2875 spin_unlock_irq(&sighand->siglock);
2876 if (unlikely(cgroup_task_frozen(current)))
2877 cgroup_leave_frozen(true);
2878
2879 /*
2880 * Anything else is fatal, maybe with a core dump.
2881 */
2882 current->flags |= PF_SIGNALED;
2883
2884 if (sig_kernel_coredump(signr)) {
2885 if (print_fatal_signals)
2886 print_fatal_signal(signr);
2887 proc_coredump_connector(current);
2888 /*
2889 * If it was able to dump core, this kills all
2890 * other threads in the group and synchronizes with
2891 * their demise. If we lost the race with another
2892 * thread getting here, it set group_exit_code
2893 * first and our do_group_exit call below will use
2894 * that value and ignore the one we pass it.
2895 */
2896 do_coredump(&ksig->info);
2897 }
2898
2899 /*
2900 * PF_USER_WORKER threads will catch and exit on fatal signals
2901 * themselves. They have cleanup that must be performed, so we
2902 * cannot call do_exit() on their behalf. Note that ksig won't
2903 * be properly initialized, PF_USER_WORKER's shouldn't use it.
2904 */
2905 if (current->flags & PF_USER_WORKER)
2906 goto out;
2907
2908 /*
2909 * Death signals, no core dump.
2910 */
2911 do_group_exit(signr);
2912 /* NOTREACHED */
2913 }
2914 spin_unlock_irq(&sighand->siglock);
2915
2916 ksig->sig = signr;
2917
2918 if (signr && !(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2919 hide_si_addr_tag_bits(ksig);
2920out:
2921 return signr > 0;
2922}
2923
2924/**
2925 * signal_delivered - called after signal delivery to update blocked signals
2926 * @ksig: kernel signal struct
2927 * @stepping: nonzero if debugger single-step or block-step in use
2928 *
2929 * This function should be called when a signal has successfully been
2930 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2931 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2932 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2933 */
2934static void signal_delivered(struct ksignal *ksig, int stepping)
2935{
2936 sigset_t blocked;
2937
2938 /* A signal was successfully delivered, and the
2939 saved sigmask was stored on the signal frame,
2940 and will be restored by sigreturn. So we can
2941 simply clear the restore sigmask flag. */
2942 clear_restore_sigmask();
2943
2944 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2945 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2946 sigaddset(&blocked, ksig->sig);
2947 set_current_blocked(&blocked);
2948 if (current->sas_ss_flags & SS_AUTODISARM)
2949 sas_ss_reset(current);
2950 if (stepping)
2951 ptrace_notify(SIGTRAP, 0);
2952}
2953
2954void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2955{
2956 if (failed)
2957 force_sigsegv(ksig->sig);
2958 else
2959 signal_delivered(ksig, stepping);
2960}
2961
2962/*
2963 * It could be that complete_signal() picked us to notify about the
2964 * group-wide signal. Other threads should be notified now to take
2965 * the shared signals in @which since we will not.
2966 */
2967static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2968{
2969 sigset_t retarget;
2970 struct task_struct *t;
2971
2972 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2973 if (sigisemptyset(&retarget))
2974 return;
2975
2976 for_other_threads(tsk, t) {
2977 if (t->flags & PF_EXITING)
2978 continue;
2979
2980 if (!has_pending_signals(&retarget, &t->blocked))
2981 continue;
2982 /* Remove the signals this thread can handle. */
2983 sigandsets(&retarget, &retarget, &t->blocked);
2984
2985 if (!task_sigpending(t))
2986 signal_wake_up(t, 0);
2987
2988 if (sigisemptyset(&retarget))
2989 break;
2990 }
2991}
2992
2993void exit_signals(struct task_struct *tsk)
2994{
2995 int group_stop = 0;
2996 sigset_t unblocked;
2997
2998 /*
2999 * @tsk is about to have PF_EXITING set - lock out users which
3000 * expect stable threadgroup.
3001 */
3002 cgroup_threadgroup_change_begin(tsk);
3003
3004 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
3005 sched_mm_cid_exit_signals(tsk);
3006 tsk->flags |= PF_EXITING;
3007 cgroup_threadgroup_change_end(tsk);
3008 return;
3009 }
3010
3011 spin_lock_irq(&tsk->sighand->siglock);
3012 /*
3013 * From now this task is not visible for group-wide signals,
3014 * see wants_signal(), do_signal_stop().
3015 */
3016 sched_mm_cid_exit_signals(tsk);
3017 tsk->flags |= PF_EXITING;
3018
3019 cgroup_threadgroup_change_end(tsk);
3020
3021 if (!task_sigpending(tsk))
3022 goto out;
3023
3024 unblocked = tsk->blocked;
3025 signotset(&unblocked);
3026 retarget_shared_pending(tsk, &unblocked);
3027
3028 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3029 task_participate_group_stop(tsk))
3030 group_stop = CLD_STOPPED;
3031out:
3032 spin_unlock_irq(&tsk->sighand->siglock);
3033
3034 /*
3035 * If group stop has completed, deliver the notification. This
3036 * should always go to the real parent of the group leader.
3037 */
3038 if (unlikely(group_stop)) {
3039 read_lock(&tasklist_lock);
3040 do_notify_parent_cldstop(tsk, false, group_stop);
3041 read_unlock(&tasklist_lock);
3042 }
3043}
3044
3045/*
3046 * System call entry points.
3047 */
3048
3049/**
3050 * sys_restart_syscall - restart a system call
3051 */
3052SYSCALL_DEFINE0(restart_syscall)
3053{
3054 struct restart_block *restart = ¤t->restart_block;
3055 return restart->fn(restart);
3056}
3057
3058long do_no_restart_syscall(struct restart_block *param)
3059{
3060 return -EINTR;
3061}
3062
3063static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3064{
3065 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3066 sigset_t newblocked;
3067 /* A set of now blocked but previously unblocked signals. */
3068 sigandnsets(&newblocked, newset, ¤t->blocked);
3069 retarget_shared_pending(tsk, &newblocked);
3070 }
3071 tsk->blocked = *newset;
3072 recalc_sigpending();
3073}
3074
3075/**
3076 * set_current_blocked - change current->blocked mask
3077 * @newset: new mask
3078 *
3079 * It is wrong to change ->blocked directly, this helper should be used
3080 * to ensure the process can't miss a shared signal we are going to block.
3081 */
3082void set_current_blocked(sigset_t *newset)
3083{
3084 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3085 __set_current_blocked(newset);
3086}
3087
3088void __set_current_blocked(const sigset_t *newset)
3089{
3090 struct task_struct *tsk = current;
3091
3092 /*
3093 * In case the signal mask hasn't changed, there is nothing we need
3094 * to do. The current->blocked shouldn't be modified by other task.
3095 */
3096 if (sigequalsets(&tsk->blocked, newset))
3097 return;
3098
3099 spin_lock_irq(&tsk->sighand->siglock);
3100 __set_task_blocked(tsk, newset);
3101 spin_unlock_irq(&tsk->sighand->siglock);
3102}
3103
3104/*
3105 * This is also useful for kernel threads that want to temporarily
3106 * (or permanently) block certain signals.
3107 *
3108 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3109 * interface happily blocks "unblockable" signals like SIGKILL
3110 * and friends.
3111 */
3112int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3113{
3114 struct task_struct *tsk = current;
3115 sigset_t newset;
3116
3117 /* Lockless, only current can change ->blocked, never from irq */
3118 if (oldset)
3119 *oldset = tsk->blocked;
3120
3121 switch (how) {
3122 case SIG_BLOCK:
3123 sigorsets(&newset, &tsk->blocked, set);
3124 break;
3125 case SIG_UNBLOCK:
3126 sigandnsets(&newset, &tsk->blocked, set);
3127 break;
3128 case SIG_SETMASK:
3129 newset = *set;
3130 break;
3131 default:
3132 return -EINVAL;
3133 }
3134
3135 __set_current_blocked(&newset);
3136 return 0;
3137}
3138EXPORT_SYMBOL(sigprocmask);
3139
3140/*
3141 * The api helps set app-provided sigmasks.
3142 *
3143 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3144 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3145 *
3146 * Note that it does set_restore_sigmask() in advance, so it must be always
3147 * paired with restore_saved_sigmask_unless() before return from syscall.
3148 */
3149int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3150{
3151 sigset_t kmask;
3152
3153 if (!umask)
3154 return 0;
3155 if (sigsetsize != sizeof(sigset_t))
3156 return -EINVAL;
3157 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3158 return -EFAULT;
3159
3160 set_restore_sigmask();
3161 current->saved_sigmask = current->blocked;
3162 set_current_blocked(&kmask);
3163
3164 return 0;
3165}
3166
3167#ifdef CONFIG_COMPAT
3168int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3169 size_t sigsetsize)
3170{
3171 sigset_t kmask;
3172
3173 if (!umask)
3174 return 0;
3175 if (sigsetsize != sizeof(compat_sigset_t))
3176 return -EINVAL;
3177 if (get_compat_sigset(&kmask, umask))
3178 return -EFAULT;
3179
3180 set_restore_sigmask();
3181 current->saved_sigmask = current->blocked;
3182 set_current_blocked(&kmask);
3183
3184 return 0;
3185}
3186#endif
3187
3188/**
3189 * sys_rt_sigprocmask - change the list of currently blocked signals
3190 * @how: whether to add, remove, or set signals
3191 * @nset: stores pending signals
3192 * @oset: previous value of signal mask if non-null
3193 * @sigsetsize: size of sigset_t type
3194 */
3195SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3196 sigset_t __user *, oset, size_t, sigsetsize)
3197{
3198 sigset_t old_set, new_set;
3199 int error;
3200
3201 /* XXX: Don't preclude handling different sized sigset_t's. */
3202 if (sigsetsize != sizeof(sigset_t))
3203 return -EINVAL;
3204
3205 old_set = current->blocked;
3206
3207 if (nset) {
3208 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3209 return -EFAULT;
3210 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3211
3212 error = sigprocmask(how, &new_set, NULL);
3213 if (error)
3214 return error;
3215 }
3216
3217 if (oset) {
3218 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3219 return -EFAULT;
3220 }
3221
3222 return 0;
3223}
3224
3225#ifdef CONFIG_COMPAT
3226COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3227 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3228{
3229 sigset_t old_set = current->blocked;
3230
3231 /* XXX: Don't preclude handling different sized sigset_t's. */
3232 if (sigsetsize != sizeof(sigset_t))
3233 return -EINVAL;
3234
3235 if (nset) {
3236 sigset_t new_set;
3237 int error;
3238 if (get_compat_sigset(&new_set, nset))
3239 return -EFAULT;
3240 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3241
3242 error = sigprocmask(how, &new_set, NULL);
3243 if (error)
3244 return error;
3245 }
3246 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3247}
3248#endif
3249
3250static void do_sigpending(sigset_t *set)
3251{
3252 spin_lock_irq(¤t->sighand->siglock);
3253 sigorsets(set, ¤t->pending.signal,
3254 ¤t->signal->shared_pending.signal);
3255 spin_unlock_irq(¤t->sighand->siglock);
3256
3257 /* Outside the lock because only this thread touches it. */
3258 sigandsets(set, ¤t->blocked, set);
3259}
3260
3261/**
3262 * sys_rt_sigpending - examine a pending signal that has been raised
3263 * while blocked
3264 * @uset: stores pending signals
3265 * @sigsetsize: size of sigset_t type or larger
3266 */
3267SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3268{
3269 sigset_t set;
3270
3271 if (sigsetsize > sizeof(*uset))
3272 return -EINVAL;
3273
3274 do_sigpending(&set);
3275
3276 if (copy_to_user(uset, &set, sigsetsize))
3277 return -EFAULT;
3278
3279 return 0;
3280}
3281
3282#ifdef CONFIG_COMPAT
3283COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3284 compat_size_t, sigsetsize)
3285{
3286 sigset_t set;
3287
3288 if (sigsetsize > sizeof(*uset))
3289 return -EINVAL;
3290
3291 do_sigpending(&set);
3292
3293 return put_compat_sigset(uset, &set, sigsetsize);
3294}
3295#endif
3296
3297static const struct {
3298 unsigned char limit, layout;
3299} sig_sicodes[] = {
3300 [SIGILL] = { NSIGILL, SIL_FAULT },
3301 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3302 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3303 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3304 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3305#if defined(SIGEMT)
3306 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3307#endif
3308 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3309 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3310 [SIGSYS] = { NSIGSYS, SIL_SYS },
3311};
3312
3313static bool known_siginfo_layout(unsigned sig, int si_code)
3314{
3315 if (si_code == SI_KERNEL)
3316 return true;
3317 else if ((si_code > SI_USER)) {
3318 if (sig_specific_sicodes(sig)) {
3319 if (si_code <= sig_sicodes[sig].limit)
3320 return true;
3321 }
3322 else if (si_code <= NSIGPOLL)
3323 return true;
3324 }
3325 else if (si_code >= SI_DETHREAD)
3326 return true;
3327 else if (si_code == SI_ASYNCNL)
3328 return true;
3329 return false;
3330}
3331
3332enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3333{
3334 enum siginfo_layout layout = SIL_KILL;
3335 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3336 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3337 (si_code <= sig_sicodes[sig].limit)) {
3338 layout = sig_sicodes[sig].layout;
3339 /* Handle the exceptions */
3340 if ((sig == SIGBUS) &&
3341 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3342 layout = SIL_FAULT_MCEERR;
3343 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3344 layout = SIL_FAULT_BNDERR;
3345#ifdef SEGV_PKUERR
3346 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3347 layout = SIL_FAULT_PKUERR;
3348#endif
3349 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3350 layout = SIL_FAULT_PERF_EVENT;
3351 else if (IS_ENABLED(CONFIG_SPARC) &&
3352 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3353 layout = SIL_FAULT_TRAPNO;
3354 else if (IS_ENABLED(CONFIG_ALPHA) &&
3355 ((sig == SIGFPE) ||
3356 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3357 layout = SIL_FAULT_TRAPNO;
3358 }
3359 else if (si_code <= NSIGPOLL)
3360 layout = SIL_POLL;
3361 } else {
3362 if (si_code == SI_TIMER)
3363 layout = SIL_TIMER;
3364 else if (si_code == SI_SIGIO)
3365 layout = SIL_POLL;
3366 else if (si_code < 0)
3367 layout = SIL_RT;
3368 }
3369 return layout;
3370}
3371
3372static inline char __user *si_expansion(const siginfo_t __user *info)
3373{
3374 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3375}
3376
3377int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3378{
3379 char __user *expansion = si_expansion(to);
3380 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3381 return -EFAULT;
3382 if (clear_user(expansion, SI_EXPANSION_SIZE))
3383 return -EFAULT;
3384 return 0;
3385}
3386
3387static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3388 const siginfo_t __user *from)
3389{
3390 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3391 char __user *expansion = si_expansion(from);
3392 char buf[SI_EXPANSION_SIZE];
3393 int i;
3394 /*
3395 * An unknown si_code might need more than
3396 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3397 * extra bytes are 0. This guarantees copy_siginfo_to_user
3398 * will return this data to userspace exactly.
3399 */
3400 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3401 return -EFAULT;
3402 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3403 if (buf[i] != 0)
3404 return -E2BIG;
3405 }
3406 }
3407 return 0;
3408}
3409
3410static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3411 const siginfo_t __user *from)
3412{
3413 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3414 return -EFAULT;
3415 to->si_signo = signo;
3416 return post_copy_siginfo_from_user(to, from);
3417}
3418
3419int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3420{
3421 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3422 return -EFAULT;
3423 return post_copy_siginfo_from_user(to, from);
3424}
3425
3426#ifdef CONFIG_COMPAT
3427/**
3428 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3429 * @to: compat siginfo destination
3430 * @from: kernel siginfo source
3431 *
3432 * Note: This function does not work properly for the SIGCHLD on x32, but
3433 * fortunately it doesn't have to. The only valid callers for this function are
3434 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3435 * The latter does not care because SIGCHLD will never cause a coredump.
3436 */
3437void copy_siginfo_to_external32(struct compat_siginfo *to,
3438 const struct kernel_siginfo *from)
3439{
3440 memset(to, 0, sizeof(*to));
3441
3442 to->si_signo = from->si_signo;
3443 to->si_errno = from->si_errno;
3444 to->si_code = from->si_code;
3445 switch(siginfo_layout(from->si_signo, from->si_code)) {
3446 case SIL_KILL:
3447 to->si_pid = from->si_pid;
3448 to->si_uid = from->si_uid;
3449 break;
3450 case SIL_TIMER:
3451 to->si_tid = from->si_tid;
3452 to->si_overrun = from->si_overrun;
3453 to->si_int = from->si_int;
3454 break;
3455 case SIL_POLL:
3456 to->si_band = from->si_band;
3457 to->si_fd = from->si_fd;
3458 break;
3459 case SIL_FAULT:
3460 to->si_addr = ptr_to_compat(from->si_addr);
3461 break;
3462 case SIL_FAULT_TRAPNO:
3463 to->si_addr = ptr_to_compat(from->si_addr);
3464 to->si_trapno = from->si_trapno;
3465 break;
3466 case SIL_FAULT_MCEERR:
3467 to->si_addr = ptr_to_compat(from->si_addr);
3468 to->si_addr_lsb = from->si_addr_lsb;
3469 break;
3470 case SIL_FAULT_BNDERR:
3471 to->si_addr = ptr_to_compat(from->si_addr);
3472 to->si_lower = ptr_to_compat(from->si_lower);
3473 to->si_upper = ptr_to_compat(from->si_upper);
3474 break;
3475 case SIL_FAULT_PKUERR:
3476 to->si_addr = ptr_to_compat(from->si_addr);
3477 to->si_pkey = from->si_pkey;
3478 break;
3479 case SIL_FAULT_PERF_EVENT:
3480 to->si_addr = ptr_to_compat(from->si_addr);
3481 to->si_perf_data = from->si_perf_data;
3482 to->si_perf_type = from->si_perf_type;
3483 to->si_perf_flags = from->si_perf_flags;
3484 break;
3485 case SIL_CHLD:
3486 to->si_pid = from->si_pid;
3487 to->si_uid = from->si_uid;
3488 to->si_status = from->si_status;
3489 to->si_utime = from->si_utime;
3490 to->si_stime = from->si_stime;
3491 break;
3492 case SIL_RT:
3493 to->si_pid = from->si_pid;
3494 to->si_uid = from->si_uid;
3495 to->si_int = from->si_int;
3496 break;
3497 case SIL_SYS:
3498 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3499 to->si_syscall = from->si_syscall;
3500 to->si_arch = from->si_arch;
3501 break;
3502 }
3503}
3504
3505int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3506 const struct kernel_siginfo *from)
3507{
3508 struct compat_siginfo new;
3509
3510 copy_siginfo_to_external32(&new, from);
3511 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3512 return -EFAULT;
3513 return 0;
3514}
3515
3516static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3517 const struct compat_siginfo *from)
3518{
3519 clear_siginfo(to);
3520 to->si_signo = from->si_signo;
3521 to->si_errno = from->si_errno;
3522 to->si_code = from->si_code;
3523 switch(siginfo_layout(from->si_signo, from->si_code)) {
3524 case SIL_KILL:
3525 to->si_pid = from->si_pid;
3526 to->si_uid = from->si_uid;
3527 break;
3528 case SIL_TIMER:
3529 to->si_tid = from->si_tid;
3530 to->si_overrun = from->si_overrun;
3531 to->si_int = from->si_int;
3532 break;
3533 case SIL_POLL:
3534 to->si_band = from->si_band;
3535 to->si_fd = from->si_fd;
3536 break;
3537 case SIL_FAULT:
3538 to->si_addr = compat_ptr(from->si_addr);
3539 break;
3540 case SIL_FAULT_TRAPNO:
3541 to->si_addr = compat_ptr(from->si_addr);
3542 to->si_trapno = from->si_trapno;
3543 break;
3544 case SIL_FAULT_MCEERR:
3545 to->si_addr = compat_ptr(from->si_addr);
3546 to->si_addr_lsb = from->si_addr_lsb;
3547 break;
3548 case SIL_FAULT_BNDERR:
3549 to->si_addr = compat_ptr(from->si_addr);
3550 to->si_lower = compat_ptr(from->si_lower);
3551 to->si_upper = compat_ptr(from->si_upper);
3552 break;
3553 case SIL_FAULT_PKUERR:
3554 to->si_addr = compat_ptr(from->si_addr);
3555 to->si_pkey = from->si_pkey;
3556 break;
3557 case SIL_FAULT_PERF_EVENT:
3558 to->si_addr = compat_ptr(from->si_addr);
3559 to->si_perf_data = from->si_perf_data;
3560 to->si_perf_type = from->si_perf_type;
3561 to->si_perf_flags = from->si_perf_flags;
3562 break;
3563 case SIL_CHLD:
3564 to->si_pid = from->si_pid;
3565 to->si_uid = from->si_uid;
3566 to->si_status = from->si_status;
3567#ifdef CONFIG_X86_X32_ABI
3568 if (in_x32_syscall()) {
3569 to->si_utime = from->_sifields._sigchld_x32._utime;
3570 to->si_stime = from->_sifields._sigchld_x32._stime;
3571 } else
3572#endif
3573 {
3574 to->si_utime = from->si_utime;
3575 to->si_stime = from->si_stime;
3576 }
3577 break;
3578 case SIL_RT:
3579 to->si_pid = from->si_pid;
3580 to->si_uid = from->si_uid;
3581 to->si_int = from->si_int;
3582 break;
3583 case SIL_SYS:
3584 to->si_call_addr = compat_ptr(from->si_call_addr);
3585 to->si_syscall = from->si_syscall;
3586 to->si_arch = from->si_arch;
3587 break;
3588 }
3589 return 0;
3590}
3591
3592static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3593 const struct compat_siginfo __user *ufrom)
3594{
3595 struct compat_siginfo from;
3596
3597 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3598 return -EFAULT;
3599
3600 from.si_signo = signo;
3601 return post_copy_siginfo_from_user32(to, &from);
3602}
3603
3604int copy_siginfo_from_user32(struct kernel_siginfo *to,
3605 const struct compat_siginfo __user *ufrom)
3606{
3607 struct compat_siginfo from;
3608
3609 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3610 return -EFAULT;
3611
3612 return post_copy_siginfo_from_user32(to, &from);
3613}
3614#endif /* CONFIG_COMPAT */
3615
3616/**
3617 * do_sigtimedwait - wait for queued signals specified in @which
3618 * @which: queued signals to wait for
3619 * @info: if non-null, the signal's siginfo is returned here
3620 * @ts: upper bound on process time suspension
3621 */
3622static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3623 const struct timespec64 *ts)
3624{
3625 ktime_t *to = NULL, timeout = KTIME_MAX;
3626 struct task_struct *tsk = current;
3627 sigset_t mask = *which;
3628 enum pid_type type;
3629 int sig, ret = 0;
3630
3631 if (ts) {
3632 if (!timespec64_valid(ts))
3633 return -EINVAL;
3634 timeout = timespec64_to_ktime(*ts);
3635 to = &timeout;
3636 }
3637
3638 /*
3639 * Invert the set of allowed signals to get those we want to block.
3640 */
3641 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3642 signotset(&mask);
3643
3644 spin_lock_irq(&tsk->sighand->siglock);
3645 sig = dequeue_signal(tsk, &mask, info, &type);
3646 if (!sig && timeout) {
3647 /*
3648 * None ready, temporarily unblock those we're interested
3649 * while we are sleeping in so that we'll be awakened when
3650 * they arrive. Unblocking is always fine, we can avoid
3651 * set_current_blocked().
3652 */
3653 tsk->real_blocked = tsk->blocked;
3654 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3655 recalc_sigpending();
3656 spin_unlock_irq(&tsk->sighand->siglock);
3657
3658 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3659 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3660 HRTIMER_MODE_REL);
3661 spin_lock_irq(&tsk->sighand->siglock);
3662 __set_task_blocked(tsk, &tsk->real_blocked);
3663 sigemptyset(&tsk->real_blocked);
3664 sig = dequeue_signal(tsk, &mask, info, &type);
3665 }
3666 spin_unlock_irq(&tsk->sighand->siglock);
3667
3668 if (sig)
3669 return sig;
3670 return ret ? -EINTR : -EAGAIN;
3671}
3672
3673/**
3674 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3675 * in @uthese
3676 * @uthese: queued signals to wait for
3677 * @uinfo: if non-null, the signal's siginfo is returned here
3678 * @uts: upper bound on process time suspension
3679 * @sigsetsize: size of sigset_t type
3680 */
3681SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3682 siginfo_t __user *, uinfo,
3683 const struct __kernel_timespec __user *, uts,
3684 size_t, sigsetsize)
3685{
3686 sigset_t these;
3687 struct timespec64 ts;
3688 kernel_siginfo_t info;
3689 int ret;
3690
3691 /* XXX: Don't preclude handling different sized sigset_t's. */
3692 if (sigsetsize != sizeof(sigset_t))
3693 return -EINVAL;
3694
3695 if (copy_from_user(&these, uthese, sizeof(these)))
3696 return -EFAULT;
3697
3698 if (uts) {
3699 if (get_timespec64(&ts, uts))
3700 return -EFAULT;
3701 }
3702
3703 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3704
3705 if (ret > 0 && uinfo) {
3706 if (copy_siginfo_to_user(uinfo, &info))
3707 ret = -EFAULT;
3708 }
3709
3710 return ret;
3711}
3712
3713#ifdef CONFIG_COMPAT_32BIT_TIME
3714SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3715 siginfo_t __user *, uinfo,
3716 const struct old_timespec32 __user *, uts,
3717 size_t, sigsetsize)
3718{
3719 sigset_t these;
3720 struct timespec64 ts;
3721 kernel_siginfo_t info;
3722 int ret;
3723
3724 if (sigsetsize != sizeof(sigset_t))
3725 return -EINVAL;
3726
3727 if (copy_from_user(&these, uthese, sizeof(these)))
3728 return -EFAULT;
3729
3730 if (uts) {
3731 if (get_old_timespec32(&ts, uts))
3732 return -EFAULT;
3733 }
3734
3735 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3736
3737 if (ret > 0 && uinfo) {
3738 if (copy_siginfo_to_user(uinfo, &info))
3739 ret = -EFAULT;
3740 }
3741
3742 return ret;
3743}
3744#endif
3745
3746#ifdef CONFIG_COMPAT
3747COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3748 struct compat_siginfo __user *, uinfo,
3749 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3750{
3751 sigset_t s;
3752 struct timespec64 t;
3753 kernel_siginfo_t info;
3754 long ret;
3755
3756 if (sigsetsize != sizeof(sigset_t))
3757 return -EINVAL;
3758
3759 if (get_compat_sigset(&s, uthese))
3760 return -EFAULT;
3761
3762 if (uts) {
3763 if (get_timespec64(&t, uts))
3764 return -EFAULT;
3765 }
3766
3767 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3768
3769 if (ret > 0 && uinfo) {
3770 if (copy_siginfo_to_user32(uinfo, &info))
3771 ret = -EFAULT;
3772 }
3773
3774 return ret;
3775}
3776
3777#ifdef CONFIG_COMPAT_32BIT_TIME
3778COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3779 struct compat_siginfo __user *, uinfo,
3780 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3781{
3782 sigset_t s;
3783 struct timespec64 t;
3784 kernel_siginfo_t info;
3785 long ret;
3786
3787 if (sigsetsize != sizeof(sigset_t))
3788 return -EINVAL;
3789
3790 if (get_compat_sigset(&s, uthese))
3791 return -EFAULT;
3792
3793 if (uts) {
3794 if (get_old_timespec32(&t, uts))
3795 return -EFAULT;
3796 }
3797
3798 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3799
3800 if (ret > 0 && uinfo) {
3801 if (copy_siginfo_to_user32(uinfo, &info))
3802 ret = -EFAULT;
3803 }
3804
3805 return ret;
3806}
3807#endif
3808#endif
3809
3810static void prepare_kill_siginfo(int sig, struct kernel_siginfo *info,
3811 enum pid_type type)
3812{
3813 clear_siginfo(info);
3814 info->si_signo = sig;
3815 info->si_errno = 0;
3816 info->si_code = (type == PIDTYPE_PID) ? SI_TKILL : SI_USER;
3817 info->si_pid = task_tgid_vnr(current);
3818 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3819}
3820
3821/**
3822 * sys_kill - send a signal to a process
3823 * @pid: the PID of the process
3824 * @sig: signal to be sent
3825 */
3826SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3827{
3828 struct kernel_siginfo info;
3829
3830 prepare_kill_siginfo(sig, &info, PIDTYPE_TGID);
3831
3832 return kill_something_info(sig, &info, pid);
3833}
3834
3835/*
3836 * Verify that the signaler and signalee either are in the same pid namespace
3837 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3838 * namespace.
3839 */
3840static bool access_pidfd_pidns(struct pid *pid)
3841{
3842 struct pid_namespace *active = task_active_pid_ns(current);
3843 struct pid_namespace *p = ns_of_pid(pid);
3844
3845 for (;;) {
3846 if (!p)
3847 return false;
3848 if (p == active)
3849 break;
3850 p = p->parent;
3851 }
3852
3853 return true;
3854}
3855
3856static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3857 siginfo_t __user *info)
3858{
3859#ifdef CONFIG_COMPAT
3860 /*
3861 * Avoid hooking up compat syscalls and instead handle necessary
3862 * conversions here. Note, this is a stop-gap measure and should not be
3863 * considered a generic solution.
3864 */
3865 if (in_compat_syscall())
3866 return copy_siginfo_from_user32(
3867 kinfo, (struct compat_siginfo __user *)info);
3868#endif
3869 return copy_siginfo_from_user(kinfo, info);
3870}
3871
3872static struct pid *pidfd_to_pid(const struct file *file)
3873{
3874 struct pid *pid;
3875
3876 pid = pidfd_pid(file);
3877 if (!IS_ERR(pid))
3878 return pid;
3879
3880 return tgid_pidfd_to_pid(file);
3881}
3882
3883#define PIDFD_SEND_SIGNAL_FLAGS \
3884 (PIDFD_SIGNAL_THREAD | PIDFD_SIGNAL_THREAD_GROUP | \
3885 PIDFD_SIGNAL_PROCESS_GROUP)
3886
3887/**
3888 * sys_pidfd_send_signal - Signal a process through a pidfd
3889 * @pidfd: file descriptor of the process
3890 * @sig: signal to send
3891 * @info: signal info
3892 * @flags: future flags
3893 *
3894 * Send the signal to the thread group or to the individual thread depending
3895 * on PIDFD_THREAD.
3896 * In the future extension to @flags may be used to override the default scope
3897 * of @pidfd.
3898 *
3899 * Return: 0 on success, negative errno on failure
3900 */
3901SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3902 siginfo_t __user *, info, unsigned int, flags)
3903{
3904 int ret;
3905 struct fd f;
3906 struct pid *pid;
3907 kernel_siginfo_t kinfo;
3908 enum pid_type type;
3909
3910 /* Enforce flags be set to 0 until we add an extension. */
3911 if (flags & ~PIDFD_SEND_SIGNAL_FLAGS)
3912 return -EINVAL;
3913
3914 /* Ensure that only a single signal scope determining flag is set. */
3915 if (hweight32(flags & PIDFD_SEND_SIGNAL_FLAGS) > 1)
3916 return -EINVAL;
3917
3918 f = fdget(pidfd);
3919 if (!f.file)
3920 return -EBADF;
3921
3922 /* Is this a pidfd? */
3923 pid = pidfd_to_pid(f.file);
3924 if (IS_ERR(pid)) {
3925 ret = PTR_ERR(pid);
3926 goto err;
3927 }
3928
3929 ret = -EINVAL;
3930 if (!access_pidfd_pidns(pid))
3931 goto err;
3932
3933 switch (flags) {
3934 case 0:
3935 /* Infer scope from the type of pidfd. */
3936 if (f.file->f_flags & PIDFD_THREAD)
3937 type = PIDTYPE_PID;
3938 else
3939 type = PIDTYPE_TGID;
3940 break;
3941 case PIDFD_SIGNAL_THREAD:
3942 type = PIDTYPE_PID;
3943 break;
3944 case PIDFD_SIGNAL_THREAD_GROUP:
3945 type = PIDTYPE_TGID;
3946 break;
3947 case PIDFD_SIGNAL_PROCESS_GROUP:
3948 type = PIDTYPE_PGID;
3949 break;
3950 }
3951
3952 if (info) {
3953 ret = copy_siginfo_from_user_any(&kinfo, info);
3954 if (unlikely(ret))
3955 goto err;
3956
3957 ret = -EINVAL;
3958 if (unlikely(sig != kinfo.si_signo))
3959 goto err;
3960
3961 /* Only allow sending arbitrary signals to yourself. */
3962 ret = -EPERM;
3963 if ((task_pid(current) != pid || type > PIDTYPE_TGID) &&
3964 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3965 goto err;
3966 } else {
3967 prepare_kill_siginfo(sig, &kinfo, type);
3968 }
3969
3970 if (type == PIDTYPE_PGID)
3971 ret = kill_pgrp_info(sig, &kinfo, pid);
3972 else
3973 ret = kill_pid_info_type(sig, &kinfo, pid, type);
3974err:
3975 fdput(f);
3976 return ret;
3977}
3978
3979static int
3980do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3981{
3982 struct task_struct *p;
3983 int error = -ESRCH;
3984
3985 rcu_read_lock();
3986 p = find_task_by_vpid(pid);
3987 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3988 error = check_kill_permission(sig, info, p);
3989 /*
3990 * The null signal is a permissions and process existence
3991 * probe. No signal is actually delivered.
3992 */
3993 if (!error && sig) {
3994 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3995 /*
3996 * If lock_task_sighand() failed we pretend the task
3997 * dies after receiving the signal. The window is tiny,
3998 * and the signal is private anyway.
3999 */
4000 if (unlikely(error == -ESRCH))
4001 error = 0;
4002 }
4003 }
4004 rcu_read_unlock();
4005
4006 return error;
4007}
4008
4009static int do_tkill(pid_t tgid, pid_t pid, int sig)
4010{
4011 struct kernel_siginfo info;
4012
4013 prepare_kill_siginfo(sig, &info, PIDTYPE_PID);
4014
4015 return do_send_specific(tgid, pid, sig, &info);
4016}
4017
4018/**
4019 * sys_tgkill - send signal to one specific thread
4020 * @tgid: the thread group ID of the thread
4021 * @pid: the PID of the thread
4022 * @sig: signal to be sent
4023 *
4024 * This syscall also checks the @tgid and returns -ESRCH even if the PID
4025 * exists but it's not belonging to the target process anymore. This
4026 * method solves the problem of threads exiting and PIDs getting reused.
4027 */
4028SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
4029{
4030 /* This is only valid for single tasks */
4031 if (pid <= 0 || tgid <= 0)
4032 return -EINVAL;
4033
4034 return do_tkill(tgid, pid, sig);
4035}
4036
4037/**
4038 * sys_tkill - send signal to one specific task
4039 * @pid: the PID of the task
4040 * @sig: signal to be sent
4041 *
4042 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4043 */
4044SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4045{
4046 /* This is only valid for single tasks */
4047 if (pid <= 0)
4048 return -EINVAL;
4049
4050 return do_tkill(0, pid, sig);
4051}
4052
4053static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4054{
4055 /* Not even root can pretend to send signals from the kernel.
4056 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4057 */
4058 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4059 (task_pid_vnr(current) != pid))
4060 return -EPERM;
4061
4062 /* POSIX.1b doesn't mention process groups. */
4063 return kill_proc_info(sig, info, pid);
4064}
4065
4066/**
4067 * sys_rt_sigqueueinfo - send signal information to a signal
4068 * @pid: the PID of the thread
4069 * @sig: signal to be sent
4070 * @uinfo: signal info to be sent
4071 */
4072SYSCALL_DEFINE3(rt_sigqueueinfo, 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_sigqueueinfo(pid, sig, &info);
4080}
4081
4082#ifdef CONFIG_COMPAT
4083COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4084 compat_pid_t, pid,
4085 int, sig,
4086 struct compat_siginfo __user *, uinfo)
4087{
4088 kernel_siginfo_t info;
4089 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4090 if (unlikely(ret))
4091 return ret;
4092 return do_rt_sigqueueinfo(pid, sig, &info);
4093}
4094#endif
4095
4096static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4097{
4098 /* This is only valid for single tasks */
4099 if (pid <= 0 || tgid <= 0)
4100 return -EINVAL;
4101
4102 /* Not even root can pretend to send signals from the kernel.
4103 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4104 */
4105 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4106 (task_pid_vnr(current) != pid))
4107 return -EPERM;
4108
4109 return do_send_specific(tgid, pid, sig, info);
4110}
4111
4112SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4113 siginfo_t __user *, uinfo)
4114{
4115 kernel_siginfo_t info;
4116 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4117 if (unlikely(ret))
4118 return ret;
4119 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4120}
4121
4122#ifdef CONFIG_COMPAT
4123COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4124 compat_pid_t, tgid,
4125 compat_pid_t, pid,
4126 int, sig,
4127 struct compat_siginfo __user *, uinfo)
4128{
4129 kernel_siginfo_t info;
4130 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4131 if (unlikely(ret))
4132 return ret;
4133 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4134}
4135#endif
4136
4137/*
4138 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4139 */
4140void kernel_sigaction(int sig, __sighandler_t action)
4141{
4142 spin_lock_irq(¤t->sighand->siglock);
4143 current->sighand->action[sig - 1].sa.sa_handler = action;
4144 if (action == SIG_IGN) {
4145 sigset_t mask;
4146
4147 sigemptyset(&mask);
4148 sigaddset(&mask, sig);
4149
4150 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
4151 flush_sigqueue_mask(&mask, ¤t->pending);
4152 recalc_sigpending();
4153 }
4154 spin_unlock_irq(¤t->sighand->siglock);
4155}
4156EXPORT_SYMBOL(kernel_sigaction);
4157
4158void __weak sigaction_compat_abi(struct k_sigaction *act,
4159 struct k_sigaction *oact)
4160{
4161}
4162
4163int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4164{
4165 struct task_struct *p = current, *t;
4166 struct k_sigaction *k;
4167 sigset_t mask;
4168
4169 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4170 return -EINVAL;
4171
4172 k = &p->sighand->action[sig-1];
4173
4174 spin_lock_irq(&p->sighand->siglock);
4175 if (k->sa.sa_flags & SA_IMMUTABLE) {
4176 spin_unlock_irq(&p->sighand->siglock);
4177 return -EINVAL;
4178 }
4179 if (oact)
4180 *oact = *k;
4181
4182 /*
4183 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4184 * e.g. by having an architecture use the bit in their uapi.
4185 */
4186 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4187
4188 /*
4189 * Clear unknown flag bits in order to allow userspace to detect missing
4190 * support for flag bits and to allow the kernel to use non-uapi bits
4191 * internally.
4192 */
4193 if (act)
4194 act->sa.sa_flags &= UAPI_SA_FLAGS;
4195 if (oact)
4196 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4197
4198 sigaction_compat_abi(act, oact);
4199
4200 if (act) {
4201 sigdelsetmask(&act->sa.sa_mask,
4202 sigmask(SIGKILL) | sigmask(SIGSTOP));
4203 *k = *act;
4204 /*
4205 * POSIX 3.3.1.3:
4206 * "Setting a signal action to SIG_IGN for a signal that is
4207 * pending shall cause the pending signal to be discarded,
4208 * whether or not it is blocked."
4209 *
4210 * "Setting a signal action to SIG_DFL for a signal that is
4211 * pending and whose default action is to ignore the signal
4212 * (for example, SIGCHLD), shall cause the pending signal to
4213 * be discarded, whether or not it is blocked"
4214 */
4215 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4216 sigemptyset(&mask);
4217 sigaddset(&mask, sig);
4218 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4219 for_each_thread(p, t)
4220 flush_sigqueue_mask(&mask, &t->pending);
4221 }
4222 }
4223
4224 spin_unlock_irq(&p->sighand->siglock);
4225 return 0;
4226}
4227
4228#ifdef CONFIG_DYNAMIC_SIGFRAME
4229static inline void sigaltstack_lock(void)
4230 __acquires(¤t->sighand->siglock)
4231{
4232 spin_lock_irq(¤t->sighand->siglock);
4233}
4234
4235static inline void sigaltstack_unlock(void)
4236 __releases(¤t->sighand->siglock)
4237{
4238 spin_unlock_irq(¤t->sighand->siglock);
4239}
4240#else
4241static inline void sigaltstack_lock(void) { }
4242static inline void sigaltstack_unlock(void) { }
4243#endif
4244
4245static int
4246do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4247 size_t min_ss_size)
4248{
4249 struct task_struct *t = current;
4250 int ret = 0;
4251
4252 if (oss) {
4253 memset(oss, 0, sizeof(stack_t));
4254 oss->ss_sp = (void __user *) t->sas_ss_sp;
4255 oss->ss_size = t->sas_ss_size;
4256 oss->ss_flags = sas_ss_flags(sp) |
4257 (current->sas_ss_flags & SS_FLAG_BITS);
4258 }
4259
4260 if (ss) {
4261 void __user *ss_sp = ss->ss_sp;
4262 size_t ss_size = ss->ss_size;
4263 unsigned ss_flags = ss->ss_flags;
4264 int ss_mode;
4265
4266 if (unlikely(on_sig_stack(sp)))
4267 return -EPERM;
4268
4269 ss_mode = ss_flags & ~SS_FLAG_BITS;
4270 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4271 ss_mode != 0))
4272 return -EINVAL;
4273
4274 /*
4275 * Return before taking any locks if no actual
4276 * sigaltstack changes were requested.
4277 */
4278 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4279 t->sas_ss_size == ss_size &&
4280 t->sas_ss_flags == ss_flags)
4281 return 0;
4282
4283 sigaltstack_lock();
4284 if (ss_mode == SS_DISABLE) {
4285 ss_size = 0;
4286 ss_sp = NULL;
4287 } else {
4288 if (unlikely(ss_size < min_ss_size))
4289 ret = -ENOMEM;
4290 if (!sigaltstack_size_valid(ss_size))
4291 ret = -ENOMEM;
4292 }
4293 if (!ret) {
4294 t->sas_ss_sp = (unsigned long) ss_sp;
4295 t->sas_ss_size = ss_size;
4296 t->sas_ss_flags = ss_flags;
4297 }
4298 sigaltstack_unlock();
4299 }
4300 return ret;
4301}
4302
4303SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4304{
4305 stack_t new, old;
4306 int err;
4307 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4308 return -EFAULT;
4309 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4310 current_user_stack_pointer(),
4311 MINSIGSTKSZ);
4312 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4313 err = -EFAULT;
4314 return err;
4315}
4316
4317int restore_altstack(const stack_t __user *uss)
4318{
4319 stack_t new;
4320 if (copy_from_user(&new, uss, sizeof(stack_t)))
4321 return -EFAULT;
4322 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4323 MINSIGSTKSZ);
4324 /* squash all but EFAULT for now */
4325 return 0;
4326}
4327
4328int __save_altstack(stack_t __user *uss, unsigned long sp)
4329{
4330 struct task_struct *t = current;
4331 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4332 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4333 __put_user(t->sas_ss_size, &uss->ss_size);
4334 return err;
4335}
4336
4337#ifdef CONFIG_COMPAT
4338static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4339 compat_stack_t __user *uoss_ptr)
4340{
4341 stack_t uss, uoss;
4342 int ret;
4343
4344 if (uss_ptr) {
4345 compat_stack_t uss32;
4346 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4347 return -EFAULT;
4348 uss.ss_sp = compat_ptr(uss32.ss_sp);
4349 uss.ss_flags = uss32.ss_flags;
4350 uss.ss_size = uss32.ss_size;
4351 }
4352 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4353 compat_user_stack_pointer(),
4354 COMPAT_MINSIGSTKSZ);
4355 if (ret >= 0 && uoss_ptr) {
4356 compat_stack_t old;
4357 memset(&old, 0, sizeof(old));
4358 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4359 old.ss_flags = uoss.ss_flags;
4360 old.ss_size = uoss.ss_size;
4361 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4362 ret = -EFAULT;
4363 }
4364 return ret;
4365}
4366
4367COMPAT_SYSCALL_DEFINE2(sigaltstack,
4368 const compat_stack_t __user *, uss_ptr,
4369 compat_stack_t __user *, uoss_ptr)
4370{
4371 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4372}
4373
4374int compat_restore_altstack(const compat_stack_t __user *uss)
4375{
4376 int err = do_compat_sigaltstack(uss, NULL);
4377 /* squash all but -EFAULT for now */
4378 return err == -EFAULT ? err : 0;
4379}
4380
4381int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4382{
4383 int err;
4384 struct task_struct *t = current;
4385 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4386 &uss->ss_sp) |
4387 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4388 __put_user(t->sas_ss_size, &uss->ss_size);
4389 return err;
4390}
4391#endif
4392
4393#ifdef __ARCH_WANT_SYS_SIGPENDING
4394
4395/**
4396 * sys_sigpending - examine pending signals
4397 * @uset: where mask of pending signal is returned
4398 */
4399SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4400{
4401 sigset_t set;
4402
4403 if (sizeof(old_sigset_t) > sizeof(*uset))
4404 return -EINVAL;
4405
4406 do_sigpending(&set);
4407
4408 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4409 return -EFAULT;
4410
4411 return 0;
4412}
4413
4414#ifdef CONFIG_COMPAT
4415COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4416{
4417 sigset_t set;
4418
4419 do_sigpending(&set);
4420
4421 return put_user(set.sig[0], set32);
4422}
4423#endif
4424
4425#endif
4426
4427#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4428/**
4429 * sys_sigprocmask - examine and change blocked signals
4430 * @how: whether to add, remove, or set signals
4431 * @nset: signals to add or remove (if non-null)
4432 * @oset: previous value of signal mask if non-null
4433 *
4434 * Some platforms have their own version with special arguments;
4435 * others support only sys_rt_sigprocmask.
4436 */
4437
4438SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4439 old_sigset_t __user *, oset)
4440{
4441 old_sigset_t old_set, new_set;
4442 sigset_t new_blocked;
4443
4444 old_set = current->blocked.sig[0];
4445
4446 if (nset) {
4447 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4448 return -EFAULT;
4449
4450 new_blocked = current->blocked;
4451
4452 switch (how) {
4453 case SIG_BLOCK:
4454 sigaddsetmask(&new_blocked, new_set);
4455 break;
4456 case SIG_UNBLOCK:
4457 sigdelsetmask(&new_blocked, new_set);
4458 break;
4459 case SIG_SETMASK:
4460 new_blocked.sig[0] = new_set;
4461 break;
4462 default:
4463 return -EINVAL;
4464 }
4465
4466 set_current_blocked(&new_blocked);
4467 }
4468
4469 if (oset) {
4470 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4471 return -EFAULT;
4472 }
4473
4474 return 0;
4475}
4476#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4477
4478#ifndef CONFIG_ODD_RT_SIGACTION
4479/**
4480 * sys_rt_sigaction - alter an action taken by a process
4481 * @sig: signal to be sent
4482 * @act: new sigaction
4483 * @oact: used to save the previous sigaction
4484 * @sigsetsize: size of sigset_t type
4485 */
4486SYSCALL_DEFINE4(rt_sigaction, int, sig,
4487 const struct sigaction __user *, act,
4488 struct sigaction __user *, oact,
4489 size_t, sigsetsize)
4490{
4491 struct k_sigaction new_sa, old_sa;
4492 int ret;
4493
4494 /* XXX: Don't preclude handling different sized sigset_t's. */
4495 if (sigsetsize != sizeof(sigset_t))
4496 return -EINVAL;
4497
4498 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4499 return -EFAULT;
4500
4501 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4502 if (ret)
4503 return ret;
4504
4505 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4506 return -EFAULT;
4507
4508 return 0;
4509}
4510#ifdef CONFIG_COMPAT
4511COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4512 const struct compat_sigaction __user *, act,
4513 struct compat_sigaction __user *, oact,
4514 compat_size_t, sigsetsize)
4515{
4516 struct k_sigaction new_ka, old_ka;
4517#ifdef __ARCH_HAS_SA_RESTORER
4518 compat_uptr_t restorer;
4519#endif
4520 int ret;
4521
4522 /* XXX: Don't preclude handling different sized sigset_t's. */
4523 if (sigsetsize != sizeof(compat_sigset_t))
4524 return -EINVAL;
4525
4526 if (act) {
4527 compat_uptr_t handler;
4528 ret = get_user(handler, &act->sa_handler);
4529 new_ka.sa.sa_handler = compat_ptr(handler);
4530#ifdef __ARCH_HAS_SA_RESTORER
4531 ret |= get_user(restorer, &act->sa_restorer);
4532 new_ka.sa.sa_restorer = compat_ptr(restorer);
4533#endif
4534 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4535 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4536 if (ret)
4537 return -EFAULT;
4538 }
4539
4540 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4541 if (!ret && oact) {
4542 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4543 &oact->sa_handler);
4544 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4545 sizeof(oact->sa_mask));
4546 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4547#ifdef __ARCH_HAS_SA_RESTORER
4548 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4549 &oact->sa_restorer);
4550#endif
4551 }
4552 return ret;
4553}
4554#endif
4555#endif /* !CONFIG_ODD_RT_SIGACTION */
4556
4557#ifdef CONFIG_OLD_SIGACTION
4558SYSCALL_DEFINE3(sigaction, int, sig,
4559 const struct old_sigaction __user *, act,
4560 struct old_sigaction __user *, oact)
4561{
4562 struct k_sigaction new_ka, old_ka;
4563 int ret;
4564
4565 if (act) {
4566 old_sigset_t mask;
4567 if (!access_ok(act, sizeof(*act)) ||
4568 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4569 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4570 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4571 __get_user(mask, &act->sa_mask))
4572 return -EFAULT;
4573#ifdef __ARCH_HAS_KA_RESTORER
4574 new_ka.ka_restorer = NULL;
4575#endif
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(old_ka.sa.sa_handler, &oact->sa_handler) ||
4584 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4585 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4586 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4587 return -EFAULT;
4588 }
4589
4590 return ret;
4591}
4592#endif
4593#ifdef CONFIG_COMPAT_OLD_SIGACTION
4594COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4595 const struct compat_old_sigaction __user *, act,
4596 struct compat_old_sigaction __user *, oact)
4597{
4598 struct k_sigaction new_ka, old_ka;
4599 int ret;
4600 compat_old_sigset_t mask;
4601 compat_uptr_t handler, restorer;
4602
4603 if (act) {
4604 if (!access_ok(act, sizeof(*act)) ||
4605 __get_user(handler, &act->sa_handler) ||
4606 __get_user(restorer, &act->sa_restorer) ||
4607 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4608 __get_user(mask, &act->sa_mask))
4609 return -EFAULT;
4610
4611#ifdef __ARCH_HAS_KA_RESTORER
4612 new_ka.ka_restorer = NULL;
4613#endif
4614 new_ka.sa.sa_handler = compat_ptr(handler);
4615 new_ka.sa.sa_restorer = compat_ptr(restorer);
4616 siginitset(&new_ka.sa.sa_mask, mask);
4617 }
4618
4619 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4620
4621 if (!ret && oact) {
4622 if (!access_ok(oact, sizeof(*oact)) ||
4623 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4624 &oact->sa_handler) ||
4625 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4626 &oact->sa_restorer) ||
4627 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4628 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4629 return -EFAULT;
4630 }
4631 return ret;
4632}
4633#endif
4634
4635#ifdef CONFIG_SGETMASK_SYSCALL
4636
4637/*
4638 * For backwards compatibility. Functionality superseded by sigprocmask.
4639 */
4640SYSCALL_DEFINE0(sgetmask)
4641{
4642 /* SMP safe */
4643 return current->blocked.sig[0];
4644}
4645
4646SYSCALL_DEFINE1(ssetmask, int, newmask)
4647{
4648 int old = current->blocked.sig[0];
4649 sigset_t newset;
4650
4651 siginitset(&newset, newmask);
4652 set_current_blocked(&newset);
4653
4654 return old;
4655}
4656#endif /* CONFIG_SGETMASK_SYSCALL */
4657
4658#ifdef __ARCH_WANT_SYS_SIGNAL
4659/*
4660 * For backwards compatibility. Functionality superseded by sigaction.
4661 */
4662SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4663{
4664 struct k_sigaction new_sa, old_sa;
4665 int ret;
4666
4667 new_sa.sa.sa_handler = handler;
4668 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4669 sigemptyset(&new_sa.sa.sa_mask);
4670
4671 ret = do_sigaction(sig, &new_sa, &old_sa);
4672
4673 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4674}
4675#endif /* __ARCH_WANT_SYS_SIGNAL */
4676
4677#ifdef __ARCH_WANT_SYS_PAUSE
4678
4679SYSCALL_DEFINE0(pause)
4680{
4681 while (!signal_pending(current)) {
4682 __set_current_state(TASK_INTERRUPTIBLE);
4683 schedule();
4684 }
4685 return -ERESTARTNOHAND;
4686}
4687
4688#endif
4689
4690static int sigsuspend(sigset_t *set)
4691{
4692 current->saved_sigmask = current->blocked;
4693 set_current_blocked(set);
4694
4695 while (!signal_pending(current)) {
4696 __set_current_state(TASK_INTERRUPTIBLE);
4697 schedule();
4698 }
4699 set_restore_sigmask();
4700 return -ERESTARTNOHAND;
4701}
4702
4703/**
4704 * sys_rt_sigsuspend - replace the signal mask for a value with the
4705 * @unewset value until a signal is received
4706 * @unewset: new signal mask value
4707 * @sigsetsize: size of sigset_t type
4708 */
4709SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4710{
4711 sigset_t newset;
4712
4713 /* XXX: Don't preclude handling different sized sigset_t's. */
4714 if (sigsetsize != sizeof(sigset_t))
4715 return -EINVAL;
4716
4717 if (copy_from_user(&newset, unewset, sizeof(newset)))
4718 return -EFAULT;
4719 return sigsuspend(&newset);
4720}
4721
4722#ifdef CONFIG_COMPAT
4723COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4724{
4725 sigset_t newset;
4726
4727 /* XXX: Don't preclude handling different sized sigset_t's. */
4728 if (sigsetsize != sizeof(sigset_t))
4729 return -EINVAL;
4730
4731 if (get_compat_sigset(&newset, unewset))
4732 return -EFAULT;
4733 return sigsuspend(&newset);
4734}
4735#endif
4736
4737#ifdef CONFIG_OLD_SIGSUSPEND
4738SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4739{
4740 sigset_t blocked;
4741 siginitset(&blocked, mask);
4742 return sigsuspend(&blocked);
4743}
4744#endif
4745#ifdef CONFIG_OLD_SIGSUSPEND3
4746SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4747{
4748 sigset_t blocked;
4749 siginitset(&blocked, mask);
4750 return sigsuspend(&blocked);
4751}
4752#endif
4753
4754__weak const char *arch_vma_name(struct vm_area_struct *vma)
4755{
4756 return NULL;
4757}
4758
4759static inline void siginfo_buildtime_checks(void)
4760{
4761 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4762
4763 /* Verify the offsets in the two siginfos match */
4764#define CHECK_OFFSET(field) \
4765 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4766
4767 /* kill */
4768 CHECK_OFFSET(si_pid);
4769 CHECK_OFFSET(si_uid);
4770
4771 /* timer */
4772 CHECK_OFFSET(si_tid);
4773 CHECK_OFFSET(si_overrun);
4774 CHECK_OFFSET(si_value);
4775
4776 /* rt */
4777 CHECK_OFFSET(si_pid);
4778 CHECK_OFFSET(si_uid);
4779 CHECK_OFFSET(si_value);
4780
4781 /* sigchld */
4782 CHECK_OFFSET(si_pid);
4783 CHECK_OFFSET(si_uid);
4784 CHECK_OFFSET(si_status);
4785 CHECK_OFFSET(si_utime);
4786 CHECK_OFFSET(si_stime);
4787
4788 /* sigfault */
4789 CHECK_OFFSET(si_addr);
4790 CHECK_OFFSET(si_trapno);
4791 CHECK_OFFSET(si_addr_lsb);
4792 CHECK_OFFSET(si_lower);
4793 CHECK_OFFSET(si_upper);
4794 CHECK_OFFSET(si_pkey);
4795 CHECK_OFFSET(si_perf_data);
4796 CHECK_OFFSET(si_perf_type);
4797 CHECK_OFFSET(si_perf_flags);
4798
4799 /* sigpoll */
4800 CHECK_OFFSET(si_band);
4801 CHECK_OFFSET(si_fd);
4802
4803 /* sigsys */
4804 CHECK_OFFSET(si_call_addr);
4805 CHECK_OFFSET(si_syscall);
4806 CHECK_OFFSET(si_arch);
4807#undef CHECK_OFFSET
4808
4809 /* usb asyncio */
4810 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4811 offsetof(struct siginfo, si_addr));
4812 if (sizeof(int) == sizeof(void __user *)) {
4813 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4814 sizeof(void __user *));
4815 } else {
4816 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4817 sizeof_field(struct siginfo, si_uid)) !=
4818 sizeof(void __user *));
4819 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4820 offsetof(struct siginfo, si_uid));
4821 }
4822#ifdef CONFIG_COMPAT
4823 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4824 offsetof(struct compat_siginfo, si_addr));
4825 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4826 sizeof(compat_uptr_t));
4827 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4828 sizeof_field(struct siginfo, si_pid));
4829#endif
4830}
4831
4832#if defined(CONFIG_SYSCTL)
4833static struct ctl_table signal_debug_table[] = {
4834#ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4835 {
4836 .procname = "exception-trace",
4837 .data = &show_unhandled_signals,
4838 .maxlen = sizeof(int),
4839 .mode = 0644,
4840 .proc_handler = proc_dointvec
4841 },
4842#endif
4843 { }
4844};
4845
4846static int __init init_signal_sysctls(void)
4847{
4848 register_sysctl_init("debug", signal_debug_table);
4849 return 0;
4850}
4851early_initcall(init_signal_sysctls);
4852#endif /* CONFIG_SYSCTL */
4853
4854void __init signals_init(void)
4855{
4856 siginfo_buildtime_checks();
4857
4858 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4859}
4860
4861#ifdef CONFIG_KGDB_KDB
4862#include <linux/kdb.h>
4863/*
4864 * kdb_send_sig - Allows kdb to send signals without exposing
4865 * signal internals. This function checks if the required locks are
4866 * available before calling the main signal code, to avoid kdb
4867 * deadlocks.
4868 */
4869void kdb_send_sig(struct task_struct *t, int sig)
4870{
4871 static struct task_struct *kdb_prev_t;
4872 int new_t, ret;
4873 if (!spin_trylock(&t->sighand->siglock)) {
4874 kdb_printf("Can't do kill command now.\n"
4875 "The sigmask lock is held somewhere else in "
4876 "kernel, try again later\n");
4877 return;
4878 }
4879 new_t = kdb_prev_t != t;
4880 kdb_prev_t = t;
4881 if (!task_is_running(t) && new_t) {
4882 spin_unlock(&t->sighand->siglock);
4883 kdb_printf("Process is not RUNNING, sending a signal from "
4884 "kdb risks deadlock\n"
4885 "on the run queue locks. "
4886 "The signal has _not_ been sent.\n"
4887 "Reissue the kill command if you want to risk "
4888 "the deadlock.\n");
4889 return;
4890 }
4891 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4892 spin_unlock(&t->sighand->siglock);
4893 if (ret)
4894 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4895 sig, t->pid);
4896 else
4897 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4898}
4899#endif /* CONFIG_KGDB_KDB */