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