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