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#include <uapi/linux/pidfd.h>
51
52#define CREATE_TRACE_POINTS
53#include <trace/events/signal.h>
54
55#include <asm/param.h>
56#include <linux/uaccess.h>
57#include <asm/unistd.h>
58#include <asm/siginfo.h>
59#include <asm/cacheflush.h>
60#include <asm/syscall.h> /* for syscall_get_* */
61
62#include "time/posix-timers.h"
63
64/*
65 * SLAB caches for signal bits.
66 */
67
68static struct kmem_cache *sigqueue_cachep;
69
70int print_fatal_signals __read_mostly;
71
72static void __user *sig_handler(struct task_struct *t, int sig)
73{
74 return t->sighand->action[sig - 1].sa.sa_handler;
75}
76
77static inline bool sig_handler_ignored(void __user *handler, int sig)
78{
79 /* Is it explicitly or implicitly ignored? */
80 return handler == SIG_IGN ||
81 (handler == SIG_DFL && sig_kernel_ignore(sig));
82}
83
84static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
85{
86 void __user *handler;
87
88 handler = sig_handler(t, sig);
89
90 /* SIGKILL and SIGSTOP may not be sent to the global init */
91 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
92 return true;
93
94 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
95 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
96 return true;
97
98 /* Only allow kernel generated signals to this kthread */
99 if (unlikely((t->flags & PF_KTHREAD) &&
100 (handler == SIG_KTHREAD_KERNEL) && !force))
101 return true;
102
103 return sig_handler_ignored(handler, sig);
104}
105
106static bool sig_ignored(struct task_struct *t, int sig, bool force)
107{
108 /*
109 * Blocked signals are never ignored, since the
110 * signal handler may change by the time it is
111 * unblocked.
112 */
113 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
114 return false;
115
116 /*
117 * Tracers may want to know about even ignored signal unless it
118 * is SIGKILL which can't be reported anyway but can be ignored
119 * by SIGNAL_UNKILLABLE task.
120 */
121 if (t->ptrace && sig != SIGKILL)
122 return false;
123
124 return sig_task_ignored(t, sig, force);
125}
126
127/*
128 * Re-calculate pending state from the set of locally pending
129 * signals, globally pending signals, and blocked signals.
130 */
131static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
132{
133 unsigned long ready;
134 long i;
135
136 switch (_NSIG_WORDS) {
137 default:
138 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
139 ready |= signal->sig[i] &~ blocked->sig[i];
140 break;
141
142 case 4: ready = signal->sig[3] &~ blocked->sig[3];
143 ready |= signal->sig[2] &~ blocked->sig[2];
144 ready |= signal->sig[1] &~ blocked->sig[1];
145 ready |= signal->sig[0] &~ blocked->sig[0];
146 break;
147
148 case 2: ready = signal->sig[1] &~ blocked->sig[1];
149 ready |= signal->sig[0] &~ blocked->sig[0];
150 break;
151
152 case 1: ready = signal->sig[0] &~ blocked->sig[0];
153 }
154 return ready != 0;
155}
156
157#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
158
159static bool recalc_sigpending_tsk(struct task_struct *t)
160{
161 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
162 PENDING(&t->pending, &t->blocked) ||
163 PENDING(&t->signal->shared_pending, &t->blocked) ||
164 cgroup_task_frozen(t)) {
165 set_tsk_thread_flag(t, TIF_SIGPENDING);
166 return true;
167 }
168
169 /*
170 * We must never clear the flag in another thread, or in current
171 * when it's possible the current syscall is returning -ERESTART*.
172 * So we don't clear it here, and only callers who know they should do.
173 */
174 return false;
175}
176
177void recalc_sigpending(void)
178{
179 if (!recalc_sigpending_tsk(current) && !freezing(current))
180 clear_thread_flag(TIF_SIGPENDING);
181
182}
183EXPORT_SYMBOL(recalc_sigpending);
184
185void calculate_sigpending(void)
186{
187 /* Have any signals or users of TIF_SIGPENDING been delayed
188 * until after fork?
189 */
190 spin_lock_irq(¤t->sighand->siglock);
191 set_tsk_thread_flag(current, TIF_SIGPENDING);
192 recalc_sigpending();
193 spin_unlock_irq(¤t->sighand->siglock);
194}
195
196/* Given the mask, find the first available signal that should be serviced. */
197
198#define SYNCHRONOUS_MASK \
199 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
200 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
201
202int next_signal(struct sigpending *pending, sigset_t *mask)
203{
204 unsigned long i, *s, *m, x;
205 int sig = 0;
206
207 s = pending->signal.sig;
208 m = mask->sig;
209
210 /*
211 * Handle the first word specially: it contains the
212 * synchronous signals that need to be dequeued first.
213 */
214 x = *s &~ *m;
215 if (x) {
216 if (x & SYNCHRONOUS_MASK)
217 x &= SYNCHRONOUS_MASK;
218 sig = ffz(~x) + 1;
219 return sig;
220 }
221
222 switch (_NSIG_WORDS) {
223 default:
224 for (i = 1; i < _NSIG_WORDS; ++i) {
225 x = *++s &~ *++m;
226 if (!x)
227 continue;
228 sig = ffz(~x) + i*_NSIG_BPW + 1;
229 break;
230 }
231 break;
232
233 case 2:
234 x = s[1] &~ m[1];
235 if (!x)
236 break;
237 sig = ffz(~x) + _NSIG_BPW + 1;
238 break;
239
240 case 1:
241 /* Nothing to do */
242 break;
243 }
244
245 return sig;
246}
247
248static inline void print_dropped_signal(int sig)
249{
250 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
251
252 if (!print_fatal_signals)
253 return;
254
255 if (!__ratelimit(&ratelimit_state))
256 return;
257
258 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
259 current->comm, current->pid, sig);
260}
261
262/**
263 * task_set_jobctl_pending - set jobctl pending bits
264 * @task: target task
265 * @mask: pending bits to set
266 *
267 * Clear @mask from @task->jobctl. @mask must be subset of
268 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
269 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
270 * cleared. If @task is already being killed or exiting, this function
271 * becomes noop.
272 *
273 * CONTEXT:
274 * Must be called with @task->sighand->siglock held.
275 *
276 * RETURNS:
277 * %true if @mask is set, %false if made noop because @task was dying.
278 */
279bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
280{
281 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
282 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
283 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
284
285 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
286 return false;
287
288 if (mask & JOBCTL_STOP_SIGMASK)
289 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
290
291 task->jobctl |= mask;
292 return true;
293}
294
295/**
296 * task_clear_jobctl_trapping - clear jobctl trapping bit
297 * @task: target task
298 *
299 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
300 * Clear it and wake up the ptracer. Note that we don't need any further
301 * locking. @task->siglock guarantees that @task->parent points to the
302 * ptracer.
303 *
304 * CONTEXT:
305 * Must be called with @task->sighand->siglock held.
306 */
307void task_clear_jobctl_trapping(struct task_struct *task)
308{
309 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
310 task->jobctl &= ~JOBCTL_TRAPPING;
311 smp_mb(); /* advised by wake_up_bit() */
312 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
313 }
314}
315
316/**
317 * task_clear_jobctl_pending - clear jobctl pending bits
318 * @task: target task
319 * @mask: pending bits to clear
320 *
321 * Clear @mask from @task->jobctl. @mask must be subset of
322 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
323 * STOP bits are cleared together.
324 *
325 * If clearing of @mask leaves no stop or trap pending, this function calls
326 * task_clear_jobctl_trapping().
327 *
328 * CONTEXT:
329 * Must be called with @task->sighand->siglock held.
330 */
331void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
332{
333 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
334
335 if (mask & JOBCTL_STOP_PENDING)
336 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
337
338 task->jobctl &= ~mask;
339
340 if (!(task->jobctl & JOBCTL_PENDING_MASK))
341 task_clear_jobctl_trapping(task);
342}
343
344/**
345 * task_participate_group_stop - participate in a group stop
346 * @task: task participating in a group stop
347 *
348 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
349 * Group stop states are cleared and the group stop count is consumed if
350 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
351 * stop, the appropriate `SIGNAL_*` flags are set.
352 *
353 * CONTEXT:
354 * Must be called with @task->sighand->siglock held.
355 *
356 * RETURNS:
357 * %true if group stop completion should be notified to the parent, %false
358 * otherwise.
359 */
360static bool task_participate_group_stop(struct task_struct *task)
361{
362 struct signal_struct *sig = task->signal;
363 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
364
365 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
366
367 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
368
369 if (!consume)
370 return false;
371
372 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
373 sig->group_stop_count--;
374
375 /*
376 * Tell the caller to notify completion iff we are entering into a
377 * fresh group stop. Read comment in do_signal_stop() for details.
378 */
379 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
380 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
381 return true;
382 }
383 return false;
384}
385
386void task_join_group_stop(struct task_struct *task)
387{
388 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
389 struct signal_struct *sig = current->signal;
390
391 if (sig->group_stop_count) {
392 sig->group_stop_count++;
393 mask |= JOBCTL_STOP_CONSUME;
394 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
395 return;
396
397 /* Have the new thread join an on-going signal group stop */
398 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
399}
400
401static struct ucounts *sig_get_ucounts(struct task_struct *t, int sig,
402 int override_rlimit)
403{
404 struct ucounts *ucounts;
405 long sigpending;
406
407 /*
408 * Protect access to @t credentials. This can go away when all
409 * callers hold rcu read lock.
410 *
411 * NOTE! A pending signal will hold on to the user refcount,
412 * and we get/put the refcount only when the sigpending count
413 * changes from/to zero.
414 */
415 rcu_read_lock();
416 ucounts = task_ucounts(t);
417 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING,
418 override_rlimit);
419 rcu_read_unlock();
420 if (!sigpending)
421 return NULL;
422
423 if (unlikely(!override_rlimit && sigpending > task_rlimit(t, RLIMIT_SIGPENDING))) {
424 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
425 print_dropped_signal(sig);
426 return NULL;
427 }
428
429 return ucounts;
430}
431
432static void __sigqueue_init(struct sigqueue *q, struct ucounts *ucounts,
433 const unsigned int sigqueue_flags)
434{
435 INIT_LIST_HEAD(&q->list);
436 q->flags = sigqueue_flags;
437 q->ucounts = ucounts;
438}
439
440/*
441 * allocate a new signal queue record
442 * - this may be called without locks if and only if t == current, otherwise an
443 * appropriate lock must be held to stop the target task from exiting
444 */
445static struct sigqueue *sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
446 int override_rlimit)
447{
448 struct ucounts *ucounts = sig_get_ucounts(t, sig, override_rlimit);
449 struct sigqueue *q;
450
451 if (!ucounts)
452 return NULL;
453
454 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
455 if (!q) {
456 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
457 return NULL;
458 }
459
460 __sigqueue_init(q, ucounts, 0);
461 return q;
462}
463
464static void __sigqueue_free(struct sigqueue *q)
465{
466 if (q->flags & SIGQUEUE_PREALLOC) {
467 posixtimer_sigqueue_putref(q);
468 return;
469 }
470 if (q->ucounts) {
471 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
472 q->ucounts = NULL;
473 }
474 kmem_cache_free(sigqueue_cachep, q);
475}
476
477void flush_sigqueue(struct sigpending *queue)
478{
479 struct sigqueue *q;
480
481 sigemptyset(&queue->signal);
482 while (!list_empty(&queue->list)) {
483 q = list_entry(queue->list.next, struct sigqueue , list);
484 list_del_init(&q->list);
485 __sigqueue_free(q);
486 }
487}
488
489/*
490 * Flush all pending signals for this kthread.
491 */
492void flush_signals(struct task_struct *t)
493{
494 unsigned long flags;
495
496 spin_lock_irqsave(&t->sighand->siglock, flags);
497 clear_tsk_thread_flag(t, TIF_SIGPENDING);
498 flush_sigqueue(&t->pending);
499 flush_sigqueue(&t->signal->shared_pending);
500 spin_unlock_irqrestore(&t->sighand->siglock, flags);
501}
502EXPORT_SYMBOL(flush_signals);
503
504void ignore_signals(struct task_struct *t)
505{
506 int i;
507
508 for (i = 0; i < _NSIG; ++i)
509 t->sighand->action[i].sa.sa_handler = SIG_IGN;
510
511 flush_signals(t);
512}
513
514/*
515 * Flush all handlers for a task.
516 */
517
518void
519flush_signal_handlers(struct task_struct *t, int force_default)
520{
521 int i;
522 struct k_sigaction *ka = &t->sighand->action[0];
523 for (i = _NSIG ; i != 0 ; i--) {
524 if (force_default || ka->sa.sa_handler != SIG_IGN)
525 ka->sa.sa_handler = SIG_DFL;
526 ka->sa.sa_flags = 0;
527#ifdef __ARCH_HAS_SA_RESTORER
528 ka->sa.sa_restorer = NULL;
529#endif
530 sigemptyset(&ka->sa.sa_mask);
531 ka++;
532 }
533}
534
535bool unhandled_signal(struct task_struct *tsk, int sig)
536{
537 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
538 if (is_global_init(tsk))
539 return true;
540
541 if (handler != SIG_IGN && handler != SIG_DFL)
542 return false;
543
544 /* If dying, we handle all new signals by ignoring them */
545 if (fatal_signal_pending(tsk))
546 return false;
547
548 /* if ptraced, let the tracer determine */
549 return !tsk->ptrace;
550}
551
552static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
553 struct sigqueue **timer_sigq)
554{
555 struct sigqueue *q, *first = NULL;
556
557 /*
558 * Collect the siginfo appropriate to this signal. Check if
559 * there is another siginfo for the same signal.
560 */
561 list_for_each_entry(q, &list->list, list) {
562 if (q->info.si_signo == sig) {
563 if (first)
564 goto still_pending;
565 first = q;
566 }
567 }
568
569 sigdelset(&list->signal, sig);
570
571 if (first) {
572still_pending:
573 list_del_init(&first->list);
574 copy_siginfo(info, &first->info);
575
576 /*
577 * posix-timer signals are preallocated and freed when the last
578 * reference count is dropped in posixtimer_deliver_signal() or
579 * immediately on timer deletion when the signal is not pending.
580 * Spare the extra round through __sigqueue_free() which is
581 * ignoring preallocated signals.
582 */
583 if (unlikely((first->flags & SIGQUEUE_PREALLOC) && (info->si_code == SI_TIMER)))
584 *timer_sigq = first;
585 else
586 __sigqueue_free(first);
587 } else {
588 /*
589 * Ok, it wasn't in the queue. This must be
590 * a fast-pathed signal or we must have been
591 * out of queue space. So zero out the info.
592 */
593 clear_siginfo(info);
594 info->si_signo = sig;
595 info->si_errno = 0;
596 info->si_code = SI_USER;
597 info->si_pid = 0;
598 info->si_uid = 0;
599 }
600}
601
602static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
603 kernel_siginfo_t *info, struct sigqueue **timer_sigq)
604{
605 int sig = next_signal(pending, mask);
606
607 if (sig)
608 collect_signal(sig, pending, info, timer_sigq);
609 return sig;
610}
611
612/*
613 * Try to dequeue a signal. If a deliverable signal is found fill in the
614 * caller provided siginfo and return the signal number. Otherwise return
615 * 0.
616 */
617int dequeue_signal(sigset_t *mask, kernel_siginfo_t *info, enum pid_type *type)
618{
619 struct task_struct *tsk = current;
620 struct sigqueue *timer_sigq;
621 int signr;
622
623 lockdep_assert_held(&tsk->sighand->siglock);
624
625again:
626 *type = PIDTYPE_PID;
627 timer_sigq = NULL;
628 signr = __dequeue_signal(&tsk->pending, mask, info, &timer_sigq);
629 if (!signr) {
630 *type = PIDTYPE_TGID;
631 signr = __dequeue_signal(&tsk->signal->shared_pending,
632 mask, info, &timer_sigq);
633
634 if (unlikely(signr == SIGALRM))
635 posixtimer_rearm_itimer(tsk);
636 }
637
638 recalc_sigpending();
639 if (!signr)
640 return 0;
641
642 if (unlikely(sig_kernel_stop(signr))) {
643 /*
644 * Set a marker that we have dequeued a stop signal. Our
645 * caller might release the siglock and then the pending
646 * stop signal it is about to process is no longer in the
647 * pending bitmasks, but must still be cleared by a SIGCONT
648 * (and overruled by a SIGKILL). So those cases clear this
649 * shared flag after we've set it. Note that this flag may
650 * remain set after the signal we return is ignored or
651 * handled. That doesn't matter because its only purpose
652 * is to alert stop-signal processing code when another
653 * processor has come along and cleared the flag.
654 */
655 current->jobctl |= JOBCTL_STOP_DEQUEUED;
656 }
657
658 if (IS_ENABLED(CONFIG_POSIX_TIMERS) && unlikely(timer_sigq)) {
659 if (!posixtimer_deliver_signal(info, timer_sigq))
660 goto again;
661 }
662
663 return signr;
664}
665EXPORT_SYMBOL_GPL(dequeue_signal);
666
667static int dequeue_synchronous_signal(kernel_siginfo_t *info)
668{
669 struct task_struct *tsk = current;
670 struct sigpending *pending = &tsk->pending;
671 struct sigqueue *q, *sync = NULL;
672
673 /*
674 * Might a synchronous signal be in the queue?
675 */
676 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
677 return 0;
678
679 /*
680 * Return the first synchronous signal in the queue.
681 */
682 list_for_each_entry(q, &pending->list, list) {
683 /* Synchronous signals have a positive si_code */
684 if ((q->info.si_code > SI_USER) &&
685 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
686 sync = q;
687 goto next;
688 }
689 }
690 return 0;
691next:
692 /*
693 * Check if there is another siginfo for the same signal.
694 */
695 list_for_each_entry_continue(q, &pending->list, list) {
696 if (q->info.si_signo == sync->info.si_signo)
697 goto still_pending;
698 }
699
700 sigdelset(&pending->signal, sync->info.si_signo);
701 recalc_sigpending();
702still_pending:
703 list_del_init(&sync->list);
704 copy_siginfo(info, &sync->info);
705 __sigqueue_free(sync);
706 return info->si_signo;
707}
708
709/*
710 * Tell a process that it has a new active signal..
711 *
712 * NOTE! we rely on the previous spin_lock to
713 * lock interrupts for us! We can only be called with
714 * "siglock" held, and the local interrupt must
715 * have been disabled when that got acquired!
716 *
717 * No need to set need_resched since signal event passing
718 * goes through ->blocked
719 */
720void signal_wake_up_state(struct task_struct *t, unsigned int state)
721{
722 lockdep_assert_held(&t->sighand->siglock);
723
724 set_tsk_thread_flag(t, TIF_SIGPENDING);
725
726 /*
727 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
728 * case. We don't check t->state here because there is a race with it
729 * executing another processor and just now entering stopped state.
730 * By using wake_up_state, we ensure the process will wake up and
731 * handle its death signal.
732 */
733 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
734 kick_process(t);
735}
736
737static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q);
738
739static void sigqueue_free_ignored(struct task_struct *tsk, struct sigqueue *q)
740{
741 if (likely(!(q->flags & SIGQUEUE_PREALLOC) || q->info.si_code != SI_TIMER))
742 __sigqueue_free(q);
743 else
744 posixtimer_sig_ignore(tsk, q);
745}
746
747/* Remove signals in mask from the pending set and queue. */
748static void flush_sigqueue_mask(struct task_struct *p, sigset_t *mask, struct sigpending *s)
749{
750 struct sigqueue *q, *n;
751 sigset_t m;
752
753 lockdep_assert_held(&p->sighand->siglock);
754
755 sigandsets(&m, mask, &s->signal);
756 if (sigisemptyset(&m))
757 return;
758
759 sigandnsets(&s->signal, &s->signal, mask);
760 list_for_each_entry_safe(q, n, &s->list, list) {
761 if (sigismember(mask, q->info.si_signo)) {
762 list_del_init(&q->list);
763 sigqueue_free_ignored(p, q);
764 }
765 }
766}
767
768static inline int is_si_special(const struct kernel_siginfo *info)
769{
770 return info <= SEND_SIG_PRIV;
771}
772
773static inline bool si_fromuser(const struct kernel_siginfo *info)
774{
775 return info == SEND_SIG_NOINFO ||
776 (!is_si_special(info) && SI_FROMUSER(info));
777}
778
779/*
780 * called with RCU read lock from check_kill_permission()
781 */
782static bool kill_ok_by_cred(struct task_struct *t)
783{
784 const struct cred *cred = current_cred();
785 const struct cred *tcred = __task_cred(t);
786
787 return uid_eq(cred->euid, tcred->suid) ||
788 uid_eq(cred->euid, tcred->uid) ||
789 uid_eq(cred->uid, tcred->suid) ||
790 uid_eq(cred->uid, tcred->uid) ||
791 ns_capable(tcred->user_ns, CAP_KILL);
792}
793
794/*
795 * Bad permissions for sending the signal
796 * - the caller must hold the RCU read lock
797 */
798static int check_kill_permission(int sig, struct kernel_siginfo *info,
799 struct task_struct *t)
800{
801 struct pid *sid;
802 int error;
803
804 if (!valid_signal(sig))
805 return -EINVAL;
806
807 if (!si_fromuser(info))
808 return 0;
809
810 error = audit_signal_info(sig, t); /* Let audit system see the signal */
811 if (error)
812 return error;
813
814 if (!same_thread_group(current, t) &&
815 !kill_ok_by_cred(t)) {
816 switch (sig) {
817 case SIGCONT:
818 sid = task_session(t);
819 /*
820 * We don't return the error if sid == NULL. The
821 * task was unhashed, the caller must notice this.
822 */
823 if (!sid || sid == task_session(current))
824 break;
825 fallthrough;
826 default:
827 return -EPERM;
828 }
829 }
830
831 return security_task_kill(t, info, sig, NULL);
832}
833
834/**
835 * ptrace_trap_notify - schedule trap to notify ptracer
836 * @t: tracee wanting to notify tracer
837 *
838 * This function schedules sticky ptrace trap which is cleared on the next
839 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
840 * ptracer.
841 *
842 * If @t is running, STOP trap will be taken. If trapped for STOP and
843 * ptracer is listening for events, tracee is woken up so that it can
844 * re-trap for the new event. If trapped otherwise, STOP trap will be
845 * eventually taken without returning to userland after the existing traps
846 * are finished by PTRACE_CONT.
847 *
848 * CONTEXT:
849 * Must be called with @task->sighand->siglock held.
850 */
851static void ptrace_trap_notify(struct task_struct *t)
852{
853 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
854 lockdep_assert_held(&t->sighand->siglock);
855
856 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
857 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
858}
859
860/*
861 * Handle magic process-wide effects of stop/continue signals. Unlike
862 * the signal actions, these happen immediately at signal-generation
863 * time regardless of blocking, ignoring, or handling. This does the
864 * actual continuing for SIGCONT, but not the actual stopping for stop
865 * signals. The process stop is done as a signal action for SIG_DFL.
866 *
867 * Returns true if the signal should be actually delivered, otherwise
868 * it should be dropped.
869 */
870static bool prepare_signal(int sig, struct task_struct *p, bool force)
871{
872 struct signal_struct *signal = p->signal;
873 struct task_struct *t;
874 sigset_t flush;
875
876 if (signal->flags & SIGNAL_GROUP_EXIT) {
877 if (signal->core_state)
878 return sig == SIGKILL;
879 /*
880 * The process is in the middle of dying, drop the signal.
881 */
882 return false;
883 } else if (sig_kernel_stop(sig)) {
884 /*
885 * This is a stop signal. Remove SIGCONT from all queues.
886 */
887 siginitset(&flush, sigmask(SIGCONT));
888 flush_sigqueue_mask(p, &flush, &signal->shared_pending);
889 for_each_thread(p, t)
890 flush_sigqueue_mask(p, &flush, &t->pending);
891 } else if (sig == SIGCONT) {
892 unsigned int why;
893 /*
894 * Remove all stop signals from all queues, wake all threads.
895 */
896 siginitset(&flush, SIG_KERNEL_STOP_MASK);
897 flush_sigqueue_mask(p, &flush, &signal->shared_pending);
898 for_each_thread(p, t) {
899 flush_sigqueue_mask(p, &flush, &t->pending);
900 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
901 if (likely(!(t->ptrace & PT_SEIZED))) {
902 t->jobctl &= ~JOBCTL_STOPPED;
903 wake_up_state(t, __TASK_STOPPED);
904 } else
905 ptrace_trap_notify(t);
906 }
907
908 /*
909 * Notify the parent with CLD_CONTINUED if we were stopped.
910 *
911 * If we were in the middle of a group stop, we pretend it
912 * was already finished, and then continued. Since SIGCHLD
913 * doesn't queue we report only CLD_STOPPED, as if the next
914 * CLD_CONTINUED was dropped.
915 */
916 why = 0;
917 if (signal->flags & SIGNAL_STOP_STOPPED)
918 why |= SIGNAL_CLD_CONTINUED;
919 else if (signal->group_stop_count)
920 why |= SIGNAL_CLD_STOPPED;
921
922 if (why) {
923 /*
924 * The first thread which returns from do_signal_stop()
925 * will take ->siglock, notice SIGNAL_CLD_MASK, and
926 * notify its parent. See get_signal().
927 */
928 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
929 signal->group_stop_count = 0;
930 signal->group_exit_code = 0;
931 }
932 }
933
934 return !sig_ignored(p, sig, force);
935}
936
937/*
938 * Test if P wants to take SIG. After we've checked all threads with this,
939 * it's equivalent to finding no threads not blocking SIG. Any threads not
940 * blocking SIG were ruled out because they are not running and already
941 * have pending signals. Such threads will dequeue from the shared queue
942 * as soon as they're available, so putting the signal on the shared queue
943 * will be equivalent to sending it to one such thread.
944 */
945static inline bool wants_signal(int sig, struct task_struct *p)
946{
947 if (sigismember(&p->blocked, sig))
948 return false;
949
950 if (p->flags & PF_EXITING)
951 return false;
952
953 if (sig == SIGKILL)
954 return true;
955
956 if (task_is_stopped_or_traced(p))
957 return false;
958
959 return task_curr(p) || !task_sigpending(p);
960}
961
962static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
963{
964 struct signal_struct *signal = p->signal;
965 struct task_struct *t;
966
967 /*
968 * Now find a thread we can wake up to take the signal off the queue.
969 *
970 * Try the suggested task first (may or may not be the main thread).
971 */
972 if (wants_signal(sig, p))
973 t = p;
974 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
975 /*
976 * There is just one thread and it does not need to be woken.
977 * It will dequeue unblocked signals before it runs again.
978 */
979 return;
980 else {
981 /*
982 * Otherwise try to find a suitable thread.
983 */
984 t = signal->curr_target;
985 while (!wants_signal(sig, t)) {
986 t = next_thread(t);
987 if (t == signal->curr_target)
988 /*
989 * No thread needs to be woken.
990 * Any eligible threads will see
991 * the signal in the queue soon.
992 */
993 return;
994 }
995 signal->curr_target = t;
996 }
997
998 /*
999 * Found a killable thread. If the signal will be fatal,
1000 * then start taking the whole group down immediately.
1001 */
1002 if (sig_fatal(p, sig) &&
1003 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1004 !sigismember(&t->real_blocked, sig) &&
1005 (sig == SIGKILL || !p->ptrace)) {
1006 /*
1007 * This signal will be fatal to the whole group.
1008 */
1009 if (!sig_kernel_coredump(sig)) {
1010 /*
1011 * Start a group exit and wake everybody up.
1012 * This way we don't have other threads
1013 * running and doing things after a slower
1014 * thread has the fatal signal pending.
1015 */
1016 signal->flags = SIGNAL_GROUP_EXIT;
1017 signal->group_exit_code = sig;
1018 signal->group_stop_count = 0;
1019 __for_each_thread(signal, t) {
1020 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1021 sigaddset(&t->pending.signal, SIGKILL);
1022 signal_wake_up(t, 1);
1023 }
1024 return;
1025 }
1026 }
1027
1028 /*
1029 * The signal is already in the shared-pending queue.
1030 * Tell the chosen thread to wake up and dequeue it.
1031 */
1032 signal_wake_up(t, sig == SIGKILL);
1033 return;
1034}
1035
1036static inline bool legacy_queue(struct sigpending *signals, int sig)
1037{
1038 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1039}
1040
1041static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1042 struct task_struct *t, enum pid_type type, bool force)
1043{
1044 struct sigpending *pending;
1045 struct sigqueue *q;
1046 int override_rlimit;
1047 int ret = 0, result;
1048
1049 lockdep_assert_held(&t->sighand->siglock);
1050
1051 result = TRACE_SIGNAL_IGNORED;
1052 if (!prepare_signal(sig, t, force))
1053 goto ret;
1054
1055 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1056 /*
1057 * Short-circuit ignored signals and support queuing
1058 * exactly one non-rt signal, so that we can get more
1059 * detailed information about the cause of the signal.
1060 */
1061 result = TRACE_SIGNAL_ALREADY_PENDING;
1062 if (legacy_queue(pending, sig))
1063 goto ret;
1064
1065 result = TRACE_SIGNAL_DELIVERED;
1066 /*
1067 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1068 */
1069 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1070 goto out_set;
1071
1072 /*
1073 * Real-time signals must be queued if sent by sigqueue, or
1074 * some other real-time mechanism. It is implementation
1075 * defined whether kill() does so. We attempt to do so, on
1076 * the principle of least surprise, but since kill is not
1077 * allowed to fail with EAGAIN when low on memory we just
1078 * make sure at least one signal gets delivered and don't
1079 * pass on the info struct.
1080 */
1081 if (sig < SIGRTMIN)
1082 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1083 else
1084 override_rlimit = 0;
1085
1086 q = sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit);
1087
1088 if (q) {
1089 list_add_tail(&q->list, &pending->list);
1090 switch ((unsigned long) info) {
1091 case (unsigned long) SEND_SIG_NOINFO:
1092 clear_siginfo(&q->info);
1093 q->info.si_signo = sig;
1094 q->info.si_errno = 0;
1095 q->info.si_code = SI_USER;
1096 q->info.si_pid = task_tgid_nr_ns(current,
1097 task_active_pid_ns(t));
1098 rcu_read_lock();
1099 q->info.si_uid =
1100 from_kuid_munged(task_cred_xxx(t, user_ns),
1101 current_uid());
1102 rcu_read_unlock();
1103 break;
1104 case (unsigned long) SEND_SIG_PRIV:
1105 clear_siginfo(&q->info);
1106 q->info.si_signo = sig;
1107 q->info.si_errno = 0;
1108 q->info.si_code = SI_KERNEL;
1109 q->info.si_pid = 0;
1110 q->info.si_uid = 0;
1111 break;
1112 default:
1113 copy_siginfo(&q->info, info);
1114 break;
1115 }
1116 } else if (!is_si_special(info) &&
1117 sig >= SIGRTMIN && info->si_code != SI_USER) {
1118 /*
1119 * Queue overflow, abort. We may abort if the
1120 * signal was rt and sent by user using something
1121 * other than kill().
1122 */
1123 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1124 ret = -EAGAIN;
1125 goto ret;
1126 } else {
1127 /*
1128 * This is a silent loss of information. We still
1129 * send the signal, but the *info bits are lost.
1130 */
1131 result = TRACE_SIGNAL_LOSE_INFO;
1132 }
1133
1134out_set:
1135 signalfd_notify(t, sig);
1136 sigaddset(&pending->signal, sig);
1137
1138 /* Let multiprocess signals appear after on-going forks */
1139 if (type > PIDTYPE_TGID) {
1140 struct multiprocess_signals *delayed;
1141 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1142 sigset_t *signal = &delayed->signal;
1143 /* Can't queue both a stop and a continue signal */
1144 if (sig == SIGCONT)
1145 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1146 else if (sig_kernel_stop(sig))
1147 sigdelset(signal, SIGCONT);
1148 sigaddset(signal, sig);
1149 }
1150 }
1151
1152 complete_signal(sig, t, type);
1153ret:
1154 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1155 return ret;
1156}
1157
1158static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1159{
1160 bool ret = false;
1161 switch (siginfo_layout(info->si_signo, info->si_code)) {
1162 case SIL_KILL:
1163 case SIL_CHLD:
1164 case SIL_RT:
1165 ret = true;
1166 break;
1167 case SIL_TIMER:
1168 case SIL_POLL:
1169 case SIL_FAULT:
1170 case SIL_FAULT_TRAPNO:
1171 case SIL_FAULT_MCEERR:
1172 case SIL_FAULT_BNDERR:
1173 case SIL_FAULT_PKUERR:
1174 case SIL_FAULT_PERF_EVENT:
1175 case SIL_SYS:
1176 ret = false;
1177 break;
1178 }
1179 return ret;
1180}
1181
1182int send_signal_locked(int sig, struct kernel_siginfo *info,
1183 struct task_struct *t, enum pid_type type)
1184{
1185 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1186 bool force = false;
1187
1188 if (info == SEND_SIG_NOINFO) {
1189 /* Force if sent from an ancestor pid namespace */
1190 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1191 } else if (info == SEND_SIG_PRIV) {
1192 /* Don't ignore kernel generated signals */
1193 force = true;
1194 } else if (has_si_pid_and_uid(info)) {
1195 /* SIGKILL and SIGSTOP is special or has ids */
1196 struct user_namespace *t_user_ns;
1197
1198 rcu_read_lock();
1199 t_user_ns = task_cred_xxx(t, user_ns);
1200 if (current_user_ns() != t_user_ns) {
1201 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1202 info->si_uid = from_kuid_munged(t_user_ns, uid);
1203 }
1204 rcu_read_unlock();
1205
1206 /* A kernel generated signal? */
1207 force = (info->si_code == SI_KERNEL);
1208
1209 /* From an ancestor pid namespace? */
1210 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1211 info->si_pid = 0;
1212 force = true;
1213 }
1214 }
1215 return __send_signal_locked(sig, info, t, type, force);
1216}
1217
1218static void print_fatal_signal(int signr)
1219{
1220 struct pt_regs *regs = task_pt_regs(current);
1221 struct file *exe_file;
1222
1223 exe_file = get_task_exe_file(current);
1224 if (exe_file) {
1225 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1226 exe_file, current->comm, signr);
1227 fput(exe_file);
1228 } else {
1229 pr_info("%s: potentially unexpected fatal signal %d.\n",
1230 current->comm, signr);
1231 }
1232
1233#if defined(__i386__) && !defined(__arch_um__)
1234 pr_info("code at %08lx: ", regs->ip);
1235 {
1236 int i;
1237 for (i = 0; i < 16; i++) {
1238 unsigned char insn;
1239
1240 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1241 break;
1242 pr_cont("%02x ", insn);
1243 }
1244 }
1245 pr_cont("\n");
1246#endif
1247 preempt_disable();
1248 show_regs(regs);
1249 preempt_enable();
1250}
1251
1252static int __init setup_print_fatal_signals(char *str)
1253{
1254 get_option (&str, &print_fatal_signals);
1255
1256 return 1;
1257}
1258
1259__setup("print-fatal-signals=", setup_print_fatal_signals);
1260
1261int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1262 enum pid_type type)
1263{
1264 unsigned long flags;
1265 int ret = -ESRCH;
1266
1267 if (lock_task_sighand(p, &flags)) {
1268 ret = send_signal_locked(sig, info, p, type);
1269 unlock_task_sighand(p, &flags);
1270 }
1271
1272 return ret;
1273}
1274
1275enum sig_handler {
1276 HANDLER_CURRENT, /* If reachable use the current handler */
1277 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1278 HANDLER_EXIT, /* Only visible as the process exit code */
1279};
1280
1281/*
1282 * Force a signal that the process can't ignore: if necessary
1283 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1284 *
1285 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1286 * since we do not want to have a signal handler that was blocked
1287 * be invoked when user space had explicitly blocked it.
1288 *
1289 * We don't want to have recursive SIGSEGV's etc, for example,
1290 * that is why we also clear SIGNAL_UNKILLABLE.
1291 */
1292static int
1293force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1294 enum sig_handler handler)
1295{
1296 unsigned long int flags;
1297 int ret, blocked, ignored;
1298 struct k_sigaction *action;
1299 int sig = info->si_signo;
1300
1301 spin_lock_irqsave(&t->sighand->siglock, flags);
1302 action = &t->sighand->action[sig-1];
1303 ignored = action->sa.sa_handler == SIG_IGN;
1304 blocked = sigismember(&t->blocked, sig);
1305 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1306 action->sa.sa_handler = SIG_DFL;
1307 if (handler == HANDLER_EXIT)
1308 action->sa.sa_flags |= SA_IMMUTABLE;
1309 if (blocked)
1310 sigdelset(&t->blocked, sig);
1311 }
1312 /*
1313 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1314 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1315 */
1316 if (action->sa.sa_handler == SIG_DFL &&
1317 (!t->ptrace || (handler == HANDLER_EXIT)))
1318 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1319 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1320 /* This can happen if the signal was already pending and blocked */
1321 if (!task_sigpending(t))
1322 signal_wake_up(t, 0);
1323 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1324
1325 return ret;
1326}
1327
1328int force_sig_info(struct kernel_siginfo *info)
1329{
1330 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1331}
1332
1333/*
1334 * Nuke all other threads in the group.
1335 */
1336int zap_other_threads(struct task_struct *p)
1337{
1338 struct task_struct *t;
1339 int count = 0;
1340
1341 p->signal->group_stop_count = 0;
1342
1343 for_other_threads(p, t) {
1344 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1345 count++;
1346
1347 /* Don't bother with already dead threads */
1348 if (t->exit_state)
1349 continue;
1350 sigaddset(&t->pending.signal, SIGKILL);
1351 signal_wake_up(t, 1);
1352 }
1353
1354 return count;
1355}
1356
1357struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1358 unsigned long *flags)
1359{
1360 struct sighand_struct *sighand;
1361
1362 rcu_read_lock();
1363 for (;;) {
1364 sighand = rcu_dereference(tsk->sighand);
1365 if (unlikely(sighand == NULL))
1366 break;
1367
1368 /*
1369 * This sighand can be already freed and even reused, but
1370 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1371 * initializes ->siglock: this slab can't go away, it has
1372 * the same object type, ->siglock can't be reinitialized.
1373 *
1374 * We need to ensure that tsk->sighand is still the same
1375 * after we take the lock, we can race with de_thread() or
1376 * __exit_signal(). In the latter case the next iteration
1377 * must see ->sighand == NULL.
1378 */
1379 spin_lock_irqsave(&sighand->siglock, *flags);
1380 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1381 break;
1382 spin_unlock_irqrestore(&sighand->siglock, *flags);
1383 }
1384 rcu_read_unlock();
1385
1386 return sighand;
1387}
1388
1389#ifdef CONFIG_LOCKDEP
1390void lockdep_assert_task_sighand_held(struct task_struct *task)
1391{
1392 struct sighand_struct *sighand;
1393
1394 rcu_read_lock();
1395 sighand = rcu_dereference(task->sighand);
1396 if (sighand)
1397 lockdep_assert_held(&sighand->siglock);
1398 else
1399 WARN_ON_ONCE(1);
1400 rcu_read_unlock();
1401}
1402#endif
1403
1404/*
1405 * send signal info to all the members of a thread group or to the
1406 * individual thread if type == PIDTYPE_PID.
1407 */
1408int group_send_sig_info(int sig, struct kernel_siginfo *info,
1409 struct task_struct *p, enum pid_type type)
1410{
1411 int ret;
1412
1413 rcu_read_lock();
1414 ret = check_kill_permission(sig, info, p);
1415 rcu_read_unlock();
1416
1417 if (!ret && sig)
1418 ret = do_send_sig_info(sig, info, p, type);
1419
1420 return ret;
1421}
1422
1423/*
1424 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1425 * control characters do (^C, ^Z etc)
1426 * - the caller must hold at least a readlock on tasklist_lock
1427 */
1428int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1429{
1430 struct task_struct *p = NULL;
1431 int ret = -ESRCH;
1432
1433 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1434 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1435 /*
1436 * If group_send_sig_info() succeeds at least once ret
1437 * becomes 0 and after that the code below has no effect.
1438 * Otherwise we return the last err or -ESRCH if this
1439 * process group is empty.
1440 */
1441 if (ret)
1442 ret = err;
1443 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1444
1445 return ret;
1446}
1447
1448static int kill_pid_info_type(int sig, struct kernel_siginfo *info,
1449 struct pid *pid, enum pid_type type)
1450{
1451 int error = -ESRCH;
1452 struct task_struct *p;
1453
1454 for (;;) {
1455 rcu_read_lock();
1456 p = pid_task(pid, PIDTYPE_PID);
1457 if (p)
1458 error = group_send_sig_info(sig, info, p, type);
1459 rcu_read_unlock();
1460 if (likely(!p || error != -ESRCH))
1461 return error;
1462 /*
1463 * The task was unhashed in between, try again. If it
1464 * is dead, pid_task() will return NULL, if we race with
1465 * de_thread() it will find the new leader.
1466 */
1467 }
1468}
1469
1470int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1471{
1472 return kill_pid_info_type(sig, info, pid, PIDTYPE_TGID);
1473}
1474
1475static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1476{
1477 int error;
1478 rcu_read_lock();
1479 error = kill_pid_info(sig, info, find_vpid(pid));
1480 rcu_read_unlock();
1481 return error;
1482}
1483
1484static inline bool kill_as_cred_perm(const struct cred *cred,
1485 struct task_struct *target)
1486{
1487 const struct cred *pcred = __task_cred(target);
1488
1489 return uid_eq(cred->euid, pcred->suid) ||
1490 uid_eq(cred->euid, pcred->uid) ||
1491 uid_eq(cred->uid, pcred->suid) ||
1492 uid_eq(cred->uid, pcred->uid);
1493}
1494
1495/*
1496 * The usb asyncio usage of siginfo is wrong. The glibc support
1497 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1498 * AKA after the generic fields:
1499 * kernel_pid_t si_pid;
1500 * kernel_uid32_t si_uid;
1501 * sigval_t si_value;
1502 *
1503 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1504 * after the generic fields is:
1505 * void __user *si_addr;
1506 *
1507 * This is a practical problem when there is a 64bit big endian kernel
1508 * and a 32bit userspace. As the 32bit address will encoded in the low
1509 * 32bits of the pointer. Those low 32bits will be stored at higher
1510 * address than appear in a 32 bit pointer. So userspace will not
1511 * see the address it was expecting for it's completions.
1512 *
1513 * There is nothing in the encoding that can allow
1514 * copy_siginfo_to_user32 to detect this confusion of formats, so
1515 * handle this by requiring the caller of kill_pid_usb_asyncio to
1516 * notice when this situration takes place and to store the 32bit
1517 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1518 * parameter.
1519 */
1520int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1521 struct pid *pid, const struct cred *cred)
1522{
1523 struct kernel_siginfo info;
1524 struct task_struct *p;
1525 unsigned long flags;
1526 int ret = -EINVAL;
1527
1528 if (!valid_signal(sig))
1529 return ret;
1530
1531 clear_siginfo(&info);
1532 info.si_signo = sig;
1533 info.si_errno = errno;
1534 info.si_code = SI_ASYNCIO;
1535 *((sigval_t *)&info.si_pid) = addr;
1536
1537 rcu_read_lock();
1538 p = pid_task(pid, PIDTYPE_PID);
1539 if (!p) {
1540 ret = -ESRCH;
1541 goto out_unlock;
1542 }
1543 if (!kill_as_cred_perm(cred, p)) {
1544 ret = -EPERM;
1545 goto out_unlock;
1546 }
1547 ret = security_task_kill(p, &info, sig, cred);
1548 if (ret)
1549 goto out_unlock;
1550
1551 if (sig) {
1552 if (lock_task_sighand(p, &flags)) {
1553 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1554 unlock_task_sighand(p, &flags);
1555 } else
1556 ret = -ESRCH;
1557 }
1558out_unlock:
1559 rcu_read_unlock();
1560 return ret;
1561}
1562EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1563
1564/*
1565 * kill_something_info() interprets pid in interesting ways just like kill(2).
1566 *
1567 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1568 * is probably wrong. Should make it like BSD or SYSV.
1569 */
1570
1571static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1572{
1573 int ret;
1574
1575 if (pid > 0)
1576 return kill_proc_info(sig, info, pid);
1577
1578 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1579 if (pid == INT_MIN)
1580 return -ESRCH;
1581
1582 read_lock(&tasklist_lock);
1583 if (pid != -1) {
1584 ret = __kill_pgrp_info(sig, info,
1585 pid ? find_vpid(-pid) : task_pgrp(current));
1586 } else {
1587 int retval = 0, count = 0;
1588 struct task_struct * p;
1589
1590 for_each_process(p) {
1591 if (task_pid_vnr(p) > 1 &&
1592 !same_thread_group(p, current)) {
1593 int err = group_send_sig_info(sig, info, p,
1594 PIDTYPE_MAX);
1595 ++count;
1596 if (err != -EPERM)
1597 retval = err;
1598 }
1599 }
1600 ret = count ? retval : -ESRCH;
1601 }
1602 read_unlock(&tasklist_lock);
1603
1604 return ret;
1605}
1606
1607/*
1608 * These are for backward compatibility with the rest of the kernel source.
1609 */
1610
1611int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1612{
1613 /*
1614 * Make sure legacy kernel users don't send in bad values
1615 * (normal paths check this in check_kill_permission).
1616 */
1617 if (!valid_signal(sig))
1618 return -EINVAL;
1619
1620 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1621}
1622EXPORT_SYMBOL(send_sig_info);
1623
1624#define __si_special(priv) \
1625 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1626
1627int
1628send_sig(int sig, struct task_struct *p, int priv)
1629{
1630 return send_sig_info(sig, __si_special(priv), p);
1631}
1632EXPORT_SYMBOL(send_sig);
1633
1634void force_sig(int sig)
1635{
1636 struct kernel_siginfo info;
1637
1638 clear_siginfo(&info);
1639 info.si_signo = sig;
1640 info.si_errno = 0;
1641 info.si_code = SI_KERNEL;
1642 info.si_pid = 0;
1643 info.si_uid = 0;
1644 force_sig_info(&info);
1645}
1646EXPORT_SYMBOL(force_sig);
1647
1648void force_fatal_sig(int sig)
1649{
1650 struct kernel_siginfo info;
1651
1652 clear_siginfo(&info);
1653 info.si_signo = sig;
1654 info.si_errno = 0;
1655 info.si_code = SI_KERNEL;
1656 info.si_pid = 0;
1657 info.si_uid = 0;
1658 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1659}
1660
1661void force_exit_sig(int sig)
1662{
1663 struct kernel_siginfo info;
1664
1665 clear_siginfo(&info);
1666 info.si_signo = sig;
1667 info.si_errno = 0;
1668 info.si_code = SI_KERNEL;
1669 info.si_pid = 0;
1670 info.si_uid = 0;
1671 force_sig_info_to_task(&info, current, HANDLER_EXIT);
1672}
1673
1674/*
1675 * When things go south during signal handling, we
1676 * will force a SIGSEGV. And if the signal that caused
1677 * the problem was already a SIGSEGV, we'll want to
1678 * make sure we don't even try to deliver the signal..
1679 */
1680void force_sigsegv(int sig)
1681{
1682 if (sig == SIGSEGV)
1683 force_fatal_sig(SIGSEGV);
1684 else
1685 force_sig(SIGSEGV);
1686}
1687
1688int force_sig_fault_to_task(int sig, int code, void __user *addr,
1689 struct task_struct *t)
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 = code;
1697 info.si_addr = addr;
1698 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1699}
1700
1701int force_sig_fault(int sig, int code, void __user *addr)
1702{
1703 return force_sig_fault_to_task(sig, code, addr, current);
1704}
1705
1706int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t)
1707{
1708 struct kernel_siginfo info;
1709
1710 clear_siginfo(&info);
1711 info.si_signo = sig;
1712 info.si_errno = 0;
1713 info.si_code = code;
1714 info.si_addr = addr;
1715 return send_sig_info(info.si_signo, &info, t);
1716}
1717
1718int force_sig_mceerr(int code, void __user *addr, short lsb)
1719{
1720 struct kernel_siginfo info;
1721
1722 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1723 clear_siginfo(&info);
1724 info.si_signo = SIGBUS;
1725 info.si_errno = 0;
1726 info.si_code = code;
1727 info.si_addr = addr;
1728 info.si_addr_lsb = lsb;
1729 return force_sig_info(&info);
1730}
1731
1732int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1733{
1734 struct kernel_siginfo info;
1735
1736 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1737 clear_siginfo(&info);
1738 info.si_signo = SIGBUS;
1739 info.si_errno = 0;
1740 info.si_code = code;
1741 info.si_addr = addr;
1742 info.si_addr_lsb = lsb;
1743 return send_sig_info(info.si_signo, &info, t);
1744}
1745EXPORT_SYMBOL(send_sig_mceerr);
1746
1747int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1748{
1749 struct kernel_siginfo info;
1750
1751 clear_siginfo(&info);
1752 info.si_signo = SIGSEGV;
1753 info.si_errno = 0;
1754 info.si_code = SEGV_BNDERR;
1755 info.si_addr = addr;
1756 info.si_lower = lower;
1757 info.si_upper = upper;
1758 return force_sig_info(&info);
1759}
1760
1761#ifdef SEGV_PKUERR
1762int force_sig_pkuerr(void __user *addr, u32 pkey)
1763{
1764 struct kernel_siginfo info;
1765
1766 clear_siginfo(&info);
1767 info.si_signo = SIGSEGV;
1768 info.si_errno = 0;
1769 info.si_code = SEGV_PKUERR;
1770 info.si_addr = addr;
1771 info.si_pkey = pkey;
1772 return force_sig_info(&info);
1773}
1774#endif
1775
1776int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1777{
1778 struct kernel_siginfo info;
1779
1780 clear_siginfo(&info);
1781 info.si_signo = SIGTRAP;
1782 info.si_errno = 0;
1783 info.si_code = TRAP_PERF;
1784 info.si_addr = addr;
1785 info.si_perf_data = sig_data;
1786 info.si_perf_type = type;
1787
1788 /*
1789 * Signals generated by perf events should not terminate the whole
1790 * process if SIGTRAP is blocked, however, delivering the signal
1791 * asynchronously is better than not delivering at all. But tell user
1792 * space if the signal was asynchronous, so it can clearly be
1793 * distinguished from normal synchronous ones.
1794 */
1795 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1796 TRAP_PERF_FLAG_ASYNC :
1797 0;
1798
1799 return send_sig_info(info.si_signo, &info, current);
1800}
1801
1802/**
1803 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1804 * @syscall: syscall number to send to userland
1805 * @reason: filter-supplied reason code to send to userland (via si_errno)
1806 * @force_coredump: true to trigger a coredump
1807 *
1808 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1809 */
1810int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1811{
1812 struct kernel_siginfo info;
1813
1814 clear_siginfo(&info);
1815 info.si_signo = SIGSYS;
1816 info.si_code = SYS_SECCOMP;
1817 info.si_call_addr = (void __user *)KSTK_EIP(current);
1818 info.si_errno = reason;
1819 info.si_arch = syscall_get_arch(current);
1820 info.si_syscall = syscall;
1821 return force_sig_info_to_task(&info, current,
1822 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1823}
1824
1825/* For the crazy architectures that include trap information in
1826 * the errno field, instead of an actual errno value.
1827 */
1828int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1829{
1830 struct kernel_siginfo info;
1831
1832 clear_siginfo(&info);
1833 info.si_signo = SIGTRAP;
1834 info.si_errno = errno;
1835 info.si_code = TRAP_HWBKPT;
1836 info.si_addr = addr;
1837 return force_sig_info(&info);
1838}
1839
1840/* For the rare architectures that include trap information using
1841 * si_trapno.
1842 */
1843int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1844{
1845 struct kernel_siginfo info;
1846
1847 clear_siginfo(&info);
1848 info.si_signo = sig;
1849 info.si_errno = 0;
1850 info.si_code = code;
1851 info.si_addr = addr;
1852 info.si_trapno = trapno;
1853 return force_sig_info(&info);
1854}
1855
1856/* For the rare architectures that include trap information using
1857 * si_trapno.
1858 */
1859int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1860 struct task_struct *t)
1861{
1862 struct kernel_siginfo info;
1863
1864 clear_siginfo(&info);
1865 info.si_signo = sig;
1866 info.si_errno = 0;
1867 info.si_code = code;
1868 info.si_addr = addr;
1869 info.si_trapno = trapno;
1870 return send_sig_info(info.si_signo, &info, t);
1871}
1872
1873static int kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1874{
1875 int ret;
1876 read_lock(&tasklist_lock);
1877 ret = __kill_pgrp_info(sig, info, pgrp);
1878 read_unlock(&tasklist_lock);
1879 return ret;
1880}
1881
1882int kill_pgrp(struct pid *pid, int sig, int priv)
1883{
1884 return kill_pgrp_info(sig, __si_special(priv), pid);
1885}
1886EXPORT_SYMBOL(kill_pgrp);
1887
1888int kill_pid(struct pid *pid, int sig, int priv)
1889{
1890 return kill_pid_info(sig, __si_special(priv), pid);
1891}
1892EXPORT_SYMBOL(kill_pid);
1893
1894#ifdef CONFIG_POSIX_TIMERS
1895/*
1896 * These functions handle POSIX timer signals. POSIX timers use
1897 * preallocated sigqueue structs for sending signals.
1898 */
1899static void __flush_itimer_signals(struct sigpending *pending)
1900{
1901 sigset_t signal, retain;
1902 struct sigqueue *q, *n;
1903
1904 signal = pending->signal;
1905 sigemptyset(&retain);
1906
1907 list_for_each_entry_safe(q, n, &pending->list, list) {
1908 int sig = q->info.si_signo;
1909
1910 if (likely(q->info.si_code != SI_TIMER)) {
1911 sigaddset(&retain, sig);
1912 } else {
1913 sigdelset(&signal, sig);
1914 list_del_init(&q->list);
1915 __sigqueue_free(q);
1916 }
1917 }
1918
1919 sigorsets(&pending->signal, &signal, &retain);
1920}
1921
1922void flush_itimer_signals(void)
1923{
1924 struct task_struct *tsk = current;
1925
1926 guard(spinlock_irqsave)(&tsk->sighand->siglock);
1927 __flush_itimer_signals(&tsk->pending);
1928 __flush_itimer_signals(&tsk->signal->shared_pending);
1929}
1930
1931bool posixtimer_init_sigqueue(struct sigqueue *q)
1932{
1933 struct ucounts *ucounts = sig_get_ucounts(current, -1, 0);
1934
1935 if (!ucounts)
1936 return false;
1937 clear_siginfo(&q->info);
1938 __sigqueue_init(q, ucounts, SIGQUEUE_PREALLOC);
1939 return true;
1940}
1941
1942static void posixtimer_queue_sigqueue(struct sigqueue *q, struct task_struct *t, enum pid_type type)
1943{
1944 struct sigpending *pending;
1945 int sig = q->info.si_signo;
1946
1947 signalfd_notify(t, sig);
1948 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1949 list_add_tail(&q->list, &pending->list);
1950 sigaddset(&pending->signal, sig);
1951 complete_signal(sig, t, type);
1952}
1953
1954/*
1955 * This function is used by POSIX timers to deliver a timer signal.
1956 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1957 * set), the signal must be delivered to the specific thread (queues
1958 * into t->pending).
1959 *
1960 * Where type is not PIDTYPE_PID, signals must be delivered to the
1961 * process. In this case, prefer to deliver to current if it is in
1962 * the same thread group as the target process and its sighand is
1963 * stable, which avoids unnecessarily waking up a potentially idle task.
1964 */
1965static inline struct task_struct *posixtimer_get_target(struct k_itimer *tmr)
1966{
1967 struct task_struct *t = pid_task(tmr->it_pid, tmr->it_pid_type);
1968
1969 if (t && tmr->it_pid_type != PIDTYPE_PID &&
1970 same_thread_group(t, current) && !current->exit_state)
1971 t = current;
1972 return t;
1973}
1974
1975void posixtimer_send_sigqueue(struct k_itimer *tmr)
1976{
1977 struct sigqueue *q = &tmr->sigq;
1978 int sig = q->info.si_signo;
1979 struct task_struct *t;
1980 unsigned long flags;
1981 int result;
1982
1983 guard(rcu)();
1984
1985 t = posixtimer_get_target(tmr);
1986 if (!t)
1987 return;
1988
1989 if (!likely(lock_task_sighand(t, &flags)))
1990 return;
1991
1992 /*
1993 * Update @tmr::sigqueue_seq for posix timer signals with sighand
1994 * locked to prevent a race against dequeue_signal().
1995 */
1996 tmr->it_sigqueue_seq = tmr->it_signal_seq;
1997
1998 /*
1999 * Set the signal delivery status under sighand lock, so that the
2000 * ignored signal handling can distinguish between a periodic and a
2001 * non-periodic timer.
2002 */
2003 tmr->it_sig_periodic = tmr->it_status == POSIX_TIMER_REQUEUE_PENDING;
2004
2005 if (!prepare_signal(sig, t, false)) {
2006 result = TRACE_SIGNAL_IGNORED;
2007
2008 if (!list_empty(&q->list)) {
2009 /*
2010 * The signal was ignored and blocked. The timer
2011 * expiry queued it because blocked signals are
2012 * queued independent of the ignored state.
2013 *
2014 * The unblocking set SIGPENDING, but the signal
2015 * was not yet dequeued from the pending list.
2016 * So prepare_signal() sees unblocked and ignored,
2017 * which ends up here. Leave it queued like a
2018 * regular signal.
2019 *
2020 * The same happens when the task group is exiting
2021 * and the signal is already queued.
2022 * prepare_signal() treats SIGNAL_GROUP_EXIT as
2023 * ignored independent of its queued state. This
2024 * gets cleaned up in __exit_signal().
2025 */
2026 goto out;
2027 }
2028
2029 /* Periodic timers with SIG_IGN are queued on the ignored list */
2030 if (tmr->it_sig_periodic) {
2031 /*
2032 * Already queued means the timer was rearmed after
2033 * the previous expiry got it on the ignore list.
2034 * Nothing to do for that case.
2035 */
2036 if (hlist_unhashed(&tmr->ignored_list)) {
2037 /*
2038 * Take a signal reference and queue it on
2039 * the ignored list.
2040 */
2041 posixtimer_sigqueue_getref(q);
2042 posixtimer_sig_ignore(t, q);
2043 }
2044 } else if (!hlist_unhashed(&tmr->ignored_list)) {
2045 /*
2046 * Covers the case where a timer was periodic and
2047 * then the signal was ignored. Later it was rearmed
2048 * as oneshot timer. The previous signal is invalid
2049 * now, and this oneshot signal has to be dropped.
2050 * Remove it from the ignored list and drop the
2051 * reference count as the signal is not longer
2052 * queued.
2053 */
2054 hlist_del_init(&tmr->ignored_list);
2055 posixtimer_putref(tmr);
2056 }
2057 goto out;
2058 }
2059
2060 if (unlikely(!list_empty(&q->list))) {
2061 /* This holds a reference count already */
2062 result = TRACE_SIGNAL_ALREADY_PENDING;
2063 goto out;
2064 }
2065
2066 /*
2067 * If the signal is on the ignore list, it got blocked after it was
2068 * ignored earlier. But nothing lifted the ignore. Move it back to
2069 * the pending list to be consistent with the regular signal
2070 * handling. This already holds a reference count.
2071 *
2072 * If it's not on the ignore list acquire a reference count.
2073 */
2074 if (likely(hlist_unhashed(&tmr->ignored_list)))
2075 posixtimer_sigqueue_getref(q);
2076 else
2077 hlist_del_init(&tmr->ignored_list);
2078
2079 posixtimer_queue_sigqueue(q, t, tmr->it_pid_type);
2080 result = TRACE_SIGNAL_DELIVERED;
2081out:
2082 trace_signal_generate(sig, &q->info, t, tmr->it_pid_type != PIDTYPE_PID, result);
2083 unlock_task_sighand(t, &flags);
2084}
2085
2086static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q)
2087{
2088 struct k_itimer *tmr = container_of(q, struct k_itimer, sigq);
2089
2090 /*
2091 * If the timer is marked deleted already or the signal originates
2092 * from a non-periodic timer, then just drop the reference
2093 * count. Otherwise queue it on the ignored list.
2094 */
2095 if (tmr->it_signal && tmr->it_sig_periodic)
2096 hlist_add_head(&tmr->ignored_list, &tsk->signal->ignored_posix_timers);
2097 else
2098 posixtimer_putref(tmr);
2099}
2100
2101static void posixtimer_sig_unignore(struct task_struct *tsk, int sig)
2102{
2103 struct hlist_head *head = &tsk->signal->ignored_posix_timers;
2104 struct hlist_node *tmp;
2105 struct k_itimer *tmr;
2106
2107 if (likely(hlist_empty(head)))
2108 return;
2109
2110 /*
2111 * Rearming a timer with sighand lock held is not possible due to
2112 * lock ordering vs. tmr::it_lock. Just stick the sigqueue back and
2113 * let the signal delivery path deal with it whether it needs to be
2114 * rearmed or not. This cannot be decided here w/o dropping sighand
2115 * lock and creating a loop retry horror show.
2116 */
2117 hlist_for_each_entry_safe(tmr, tmp , head, ignored_list) {
2118 struct task_struct *target;
2119
2120 /*
2121 * tmr::sigq.info.si_signo is immutable, so accessing it
2122 * without holding tmr::it_lock is safe.
2123 */
2124 if (tmr->sigq.info.si_signo != sig)
2125 continue;
2126
2127 hlist_del_init(&tmr->ignored_list);
2128
2129 /* This should never happen and leaks a reference count */
2130 if (WARN_ON_ONCE(!list_empty(&tmr->sigq.list)))
2131 continue;
2132
2133 /*
2134 * Get the target for the signal. If target is a thread and
2135 * has exited by now, drop the reference count.
2136 */
2137 guard(rcu)();
2138 target = posixtimer_get_target(tmr);
2139 if (target)
2140 posixtimer_queue_sigqueue(&tmr->sigq, target, tmr->it_pid_type);
2141 else
2142 posixtimer_putref(tmr);
2143 }
2144}
2145#else /* CONFIG_POSIX_TIMERS */
2146static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q) { }
2147static inline void posixtimer_sig_unignore(struct task_struct *tsk, int sig) { }
2148#endif /* !CONFIG_POSIX_TIMERS */
2149
2150void do_notify_pidfd(struct task_struct *task)
2151{
2152 struct pid *pid = task_pid(task);
2153
2154 WARN_ON(task->exit_state == 0);
2155
2156 __wake_up(&pid->wait_pidfd, TASK_NORMAL, 0,
2157 poll_to_key(EPOLLIN | EPOLLRDNORM));
2158}
2159
2160/*
2161 * Let a parent know about the death of a child.
2162 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2163 *
2164 * Returns true if our parent ignored us and so we've switched to
2165 * self-reaping.
2166 */
2167bool do_notify_parent(struct task_struct *tsk, int sig)
2168{
2169 struct kernel_siginfo info;
2170 unsigned long flags;
2171 struct sighand_struct *psig;
2172 bool autoreap = false;
2173 u64 utime, stime;
2174
2175 WARN_ON_ONCE(sig == -1);
2176
2177 /* do_notify_parent_cldstop should have been called instead. */
2178 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2179
2180 WARN_ON_ONCE(!tsk->ptrace &&
2181 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2182 /*
2183 * tsk is a group leader and has no threads, wake up the
2184 * non-PIDFD_THREAD waiters.
2185 */
2186 if (thread_group_empty(tsk))
2187 do_notify_pidfd(tsk);
2188
2189 if (sig != SIGCHLD) {
2190 /*
2191 * This is only possible if parent == real_parent.
2192 * Check if it has changed security domain.
2193 */
2194 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2195 sig = SIGCHLD;
2196 }
2197
2198 clear_siginfo(&info);
2199 info.si_signo = sig;
2200 info.si_errno = 0;
2201 /*
2202 * We are under tasklist_lock here so our parent is tied to
2203 * us and cannot change.
2204 *
2205 * task_active_pid_ns will always return the same pid namespace
2206 * until a task passes through release_task.
2207 *
2208 * write_lock() currently calls preempt_disable() which is the
2209 * same as rcu_read_lock(), but according to Oleg, this is not
2210 * correct to rely on this
2211 */
2212 rcu_read_lock();
2213 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2214 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2215 task_uid(tsk));
2216 rcu_read_unlock();
2217
2218 task_cputime(tsk, &utime, &stime);
2219 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2220 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2221
2222 info.si_status = tsk->exit_code & 0x7f;
2223 if (tsk->exit_code & 0x80)
2224 info.si_code = CLD_DUMPED;
2225 else if (tsk->exit_code & 0x7f)
2226 info.si_code = CLD_KILLED;
2227 else {
2228 info.si_code = CLD_EXITED;
2229 info.si_status = tsk->exit_code >> 8;
2230 }
2231
2232 psig = tsk->parent->sighand;
2233 spin_lock_irqsave(&psig->siglock, flags);
2234 if (!tsk->ptrace && sig == SIGCHLD &&
2235 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2236 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2237 /*
2238 * We are exiting and our parent doesn't care. POSIX.1
2239 * defines special semantics for setting SIGCHLD to SIG_IGN
2240 * or setting the SA_NOCLDWAIT flag: we should be reaped
2241 * automatically and not left for our parent's wait4 call.
2242 * Rather than having the parent do it as a magic kind of
2243 * signal handler, we just set this to tell do_exit that we
2244 * can be cleaned up without becoming a zombie. Note that
2245 * we still call __wake_up_parent in this case, because a
2246 * blocked sys_wait4 might now return -ECHILD.
2247 *
2248 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2249 * is implementation-defined: we do (if you don't want
2250 * it, just use SIG_IGN instead).
2251 */
2252 autoreap = true;
2253 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2254 sig = 0;
2255 }
2256 /*
2257 * Send with __send_signal as si_pid and si_uid are in the
2258 * parent's namespaces.
2259 */
2260 if (valid_signal(sig) && sig)
2261 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2262 __wake_up_parent(tsk, tsk->parent);
2263 spin_unlock_irqrestore(&psig->siglock, flags);
2264
2265 return autoreap;
2266}
2267
2268/**
2269 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2270 * @tsk: task reporting the state change
2271 * @for_ptracer: the notification is for ptracer
2272 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2273 *
2274 * Notify @tsk's parent that the stopped/continued state has changed. If
2275 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2276 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2277 *
2278 * CONTEXT:
2279 * Must be called with tasklist_lock at least read locked.
2280 */
2281static void do_notify_parent_cldstop(struct task_struct *tsk,
2282 bool for_ptracer, int why)
2283{
2284 struct kernel_siginfo info;
2285 unsigned long flags;
2286 struct task_struct *parent;
2287 struct sighand_struct *sighand;
2288 u64 utime, stime;
2289
2290 if (for_ptracer) {
2291 parent = tsk->parent;
2292 } else {
2293 tsk = tsk->group_leader;
2294 parent = tsk->real_parent;
2295 }
2296
2297 clear_siginfo(&info);
2298 info.si_signo = SIGCHLD;
2299 info.si_errno = 0;
2300 /*
2301 * see comment in do_notify_parent() about the following 4 lines
2302 */
2303 rcu_read_lock();
2304 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2305 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2306 rcu_read_unlock();
2307
2308 task_cputime(tsk, &utime, &stime);
2309 info.si_utime = nsec_to_clock_t(utime);
2310 info.si_stime = nsec_to_clock_t(stime);
2311
2312 info.si_code = why;
2313 switch (why) {
2314 case CLD_CONTINUED:
2315 info.si_status = SIGCONT;
2316 break;
2317 case CLD_STOPPED:
2318 info.si_status = tsk->signal->group_exit_code & 0x7f;
2319 break;
2320 case CLD_TRAPPED:
2321 info.si_status = tsk->exit_code & 0x7f;
2322 break;
2323 default:
2324 BUG();
2325 }
2326
2327 sighand = parent->sighand;
2328 spin_lock_irqsave(&sighand->siglock, flags);
2329 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2330 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2331 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2332 /*
2333 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2334 */
2335 __wake_up_parent(tsk, parent);
2336 spin_unlock_irqrestore(&sighand->siglock, flags);
2337}
2338
2339/*
2340 * This must be called with current->sighand->siglock held.
2341 *
2342 * This should be the path for all ptrace stops.
2343 * We always set current->last_siginfo while stopped here.
2344 * That makes it a way to test a stopped process for
2345 * being ptrace-stopped vs being job-control-stopped.
2346 *
2347 * Returns the signal the ptracer requested the code resume
2348 * with. If the code did not stop because the tracer is gone,
2349 * the stop signal remains unchanged unless clear_code.
2350 */
2351static int ptrace_stop(int exit_code, int why, unsigned long message,
2352 kernel_siginfo_t *info)
2353 __releases(¤t->sighand->siglock)
2354 __acquires(¤t->sighand->siglock)
2355{
2356 bool gstop_done = false;
2357
2358 if (arch_ptrace_stop_needed()) {
2359 /*
2360 * The arch code has something special to do before a
2361 * ptrace stop. This is allowed to block, e.g. for faults
2362 * on user stack pages. We can't keep the siglock while
2363 * calling arch_ptrace_stop, so we must release it now.
2364 * To preserve proper semantics, we must do this before
2365 * any signal bookkeeping like checking group_stop_count.
2366 */
2367 spin_unlock_irq(¤t->sighand->siglock);
2368 arch_ptrace_stop();
2369 spin_lock_irq(¤t->sighand->siglock);
2370 }
2371
2372 /*
2373 * After this point ptrace_signal_wake_up or signal_wake_up
2374 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2375 * signal comes in. Handle previous ptrace_unlinks and fatal
2376 * signals here to prevent ptrace_stop sleeping in schedule.
2377 */
2378 if (!current->ptrace || __fatal_signal_pending(current))
2379 return exit_code;
2380
2381 set_special_state(TASK_TRACED);
2382 current->jobctl |= JOBCTL_TRACED;
2383
2384 /*
2385 * We're committing to trapping. TRACED should be visible before
2386 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2387 * Also, transition to TRACED and updates to ->jobctl should be
2388 * atomic with respect to siglock and should be done after the arch
2389 * hook as siglock is released and regrabbed across it.
2390 *
2391 * TRACER TRACEE
2392 *
2393 * ptrace_attach()
2394 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2395 * do_wait()
2396 * set_current_state() smp_wmb();
2397 * ptrace_do_wait()
2398 * wait_task_stopped()
2399 * task_stopped_code()
2400 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2401 */
2402 smp_wmb();
2403
2404 current->ptrace_message = message;
2405 current->last_siginfo = info;
2406 current->exit_code = exit_code;
2407
2408 /*
2409 * If @why is CLD_STOPPED, we're trapping to participate in a group
2410 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2411 * across siglock relocks since INTERRUPT was scheduled, PENDING
2412 * could be clear now. We act as if SIGCONT is received after
2413 * TASK_TRACED is entered - ignore it.
2414 */
2415 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2416 gstop_done = task_participate_group_stop(current);
2417
2418 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2419 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2420 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2421 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2422
2423 /* entering a trap, clear TRAPPING */
2424 task_clear_jobctl_trapping(current);
2425
2426 spin_unlock_irq(¤t->sighand->siglock);
2427 read_lock(&tasklist_lock);
2428 /*
2429 * Notify parents of the stop.
2430 *
2431 * While ptraced, there are two parents - the ptracer and
2432 * the real_parent of the group_leader. The ptracer should
2433 * know about every stop while the real parent is only
2434 * interested in the completion of group stop. The states
2435 * for the two don't interact with each other. Notify
2436 * separately unless they're gonna be duplicates.
2437 */
2438 if (current->ptrace)
2439 do_notify_parent_cldstop(current, true, why);
2440 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2441 do_notify_parent_cldstop(current, false, why);
2442
2443 /*
2444 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2445 * One a PREEMPTION kernel this can result in preemption requirement
2446 * which will be fulfilled after read_unlock() and the ptracer will be
2447 * put on the CPU.
2448 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2449 * this task wait in schedule(). If this task gets preempted then it
2450 * remains enqueued on the runqueue. The ptracer will observe this and
2451 * then sleep for a delay of one HZ tick. In the meantime this task
2452 * gets scheduled, enters schedule() and will wait for the ptracer.
2453 *
2454 * This preemption point is not bad from a correctness point of
2455 * view but extends the runtime by one HZ tick time due to the
2456 * ptracer's sleep. The preempt-disable section ensures that there
2457 * will be no preemption between unlock and schedule() and so
2458 * improving the performance since the ptracer will observe that
2459 * the tracee is scheduled out once it gets on the CPU.
2460 *
2461 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2462 * Therefore the task can be preempted after do_notify_parent_cldstop()
2463 * before unlocking tasklist_lock so there is no benefit in doing this.
2464 *
2465 * In fact disabling preemption is harmful on PREEMPT_RT because
2466 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2467 * with preemption disabled due to the 'sleeping' spinlock
2468 * substitution of RT.
2469 */
2470 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2471 preempt_disable();
2472 read_unlock(&tasklist_lock);
2473 cgroup_enter_frozen();
2474 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2475 preempt_enable_no_resched();
2476 schedule();
2477 cgroup_leave_frozen(true);
2478
2479 /*
2480 * We are back. Now reacquire the siglock before touching
2481 * last_siginfo, so that we are sure to have synchronized with
2482 * any signal-sending on another CPU that wants to examine it.
2483 */
2484 spin_lock_irq(¤t->sighand->siglock);
2485 exit_code = current->exit_code;
2486 current->last_siginfo = NULL;
2487 current->ptrace_message = 0;
2488 current->exit_code = 0;
2489
2490 /* LISTENING can be set only during STOP traps, clear it */
2491 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2492
2493 /*
2494 * Queued signals ignored us while we were stopped for tracing.
2495 * So check for any that we should take before resuming user mode.
2496 * This sets TIF_SIGPENDING, but never clears it.
2497 */
2498 recalc_sigpending_tsk(current);
2499 return exit_code;
2500}
2501
2502static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2503{
2504 kernel_siginfo_t info;
2505
2506 clear_siginfo(&info);
2507 info.si_signo = signr;
2508 info.si_code = exit_code;
2509 info.si_pid = task_pid_vnr(current);
2510 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2511
2512 /* Let the debugger run. */
2513 return ptrace_stop(exit_code, why, message, &info);
2514}
2515
2516int ptrace_notify(int exit_code, unsigned long message)
2517{
2518 int signr;
2519
2520 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2521 if (unlikely(task_work_pending(current)))
2522 task_work_run();
2523
2524 spin_lock_irq(¤t->sighand->siglock);
2525 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2526 spin_unlock_irq(¤t->sighand->siglock);
2527 return signr;
2528}
2529
2530/**
2531 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2532 * @signr: signr causing group stop if initiating
2533 *
2534 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2535 * and participate in it. If already set, participate in the existing
2536 * group stop. If participated in a group stop (and thus slept), %true is
2537 * returned with siglock released.
2538 *
2539 * If ptraced, this function doesn't handle stop itself. Instead,
2540 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2541 * untouched. The caller must ensure that INTERRUPT trap handling takes
2542 * places afterwards.
2543 *
2544 * CONTEXT:
2545 * Must be called with @current->sighand->siglock held, which is released
2546 * on %true return.
2547 *
2548 * RETURNS:
2549 * %false if group stop is already cancelled or ptrace trap is scheduled.
2550 * %true if participated in group stop.
2551 */
2552static bool do_signal_stop(int signr)
2553 __releases(¤t->sighand->siglock)
2554{
2555 struct signal_struct *sig = current->signal;
2556
2557 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2558 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2559 struct task_struct *t;
2560
2561 /* signr will be recorded in task->jobctl for retries */
2562 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2563
2564 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2565 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2566 unlikely(sig->group_exec_task))
2567 return false;
2568 /*
2569 * There is no group stop already in progress. We must
2570 * initiate one now.
2571 *
2572 * While ptraced, a task may be resumed while group stop is
2573 * still in effect and then receive a stop signal and
2574 * initiate another group stop. This deviates from the
2575 * usual behavior as two consecutive stop signals can't
2576 * cause two group stops when !ptraced. That is why we
2577 * also check !task_is_stopped(t) below.
2578 *
2579 * The condition can be distinguished by testing whether
2580 * SIGNAL_STOP_STOPPED is already set. Don't generate
2581 * group_exit_code in such case.
2582 *
2583 * This is not necessary for SIGNAL_STOP_CONTINUED because
2584 * an intervening stop signal is required to cause two
2585 * continued events regardless of ptrace.
2586 */
2587 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2588 sig->group_exit_code = signr;
2589
2590 sig->group_stop_count = 0;
2591 if (task_set_jobctl_pending(current, signr | gstop))
2592 sig->group_stop_count++;
2593
2594 for_other_threads(current, t) {
2595 /*
2596 * Setting state to TASK_STOPPED for a group
2597 * stop is always done with the siglock held,
2598 * so this check has no races.
2599 */
2600 if (!task_is_stopped(t) &&
2601 task_set_jobctl_pending(t, signr | gstop)) {
2602 sig->group_stop_count++;
2603 if (likely(!(t->ptrace & PT_SEIZED)))
2604 signal_wake_up(t, 0);
2605 else
2606 ptrace_trap_notify(t);
2607 }
2608 }
2609 }
2610
2611 if (likely(!current->ptrace)) {
2612 int notify = 0;
2613
2614 /*
2615 * If there are no other threads in the group, or if there
2616 * is a group stop in progress and we are the last to stop,
2617 * report to the parent.
2618 */
2619 if (task_participate_group_stop(current))
2620 notify = CLD_STOPPED;
2621
2622 current->jobctl |= JOBCTL_STOPPED;
2623 set_special_state(TASK_STOPPED);
2624 spin_unlock_irq(¤t->sighand->siglock);
2625
2626 /*
2627 * Notify the parent of the group stop completion. Because
2628 * we're not holding either the siglock or tasklist_lock
2629 * here, ptracer may attach inbetween; however, this is for
2630 * group stop and should always be delivered to the real
2631 * parent of the group leader. The new ptracer will get
2632 * its notification when this task transitions into
2633 * TASK_TRACED.
2634 */
2635 if (notify) {
2636 read_lock(&tasklist_lock);
2637 do_notify_parent_cldstop(current, false, notify);
2638 read_unlock(&tasklist_lock);
2639 }
2640
2641 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2642 cgroup_enter_frozen();
2643 schedule();
2644 return true;
2645 } else {
2646 /*
2647 * While ptraced, group stop is handled by STOP trap.
2648 * Schedule it and let the caller deal with it.
2649 */
2650 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2651 return false;
2652 }
2653}
2654
2655/**
2656 * do_jobctl_trap - take care of ptrace jobctl traps
2657 *
2658 * When PT_SEIZED, it's used for both group stop and explicit
2659 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2660 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2661 * the stop signal; otherwise, %SIGTRAP.
2662 *
2663 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2664 * number as exit_code and no siginfo.
2665 *
2666 * CONTEXT:
2667 * Must be called with @current->sighand->siglock held, which may be
2668 * released and re-acquired before returning with intervening sleep.
2669 */
2670static void do_jobctl_trap(void)
2671{
2672 struct signal_struct *signal = current->signal;
2673 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2674
2675 if (current->ptrace & PT_SEIZED) {
2676 if (!signal->group_stop_count &&
2677 !(signal->flags & SIGNAL_STOP_STOPPED))
2678 signr = SIGTRAP;
2679 WARN_ON_ONCE(!signr);
2680 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2681 CLD_STOPPED, 0);
2682 } else {
2683 WARN_ON_ONCE(!signr);
2684 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2685 }
2686}
2687
2688/**
2689 * do_freezer_trap - handle the freezer jobctl trap
2690 *
2691 * Puts the task into frozen state, if only the task is not about to quit.
2692 * In this case it drops JOBCTL_TRAP_FREEZE.
2693 *
2694 * CONTEXT:
2695 * Must be called with @current->sighand->siglock held,
2696 * which is always released before returning.
2697 */
2698static void do_freezer_trap(void)
2699 __releases(¤t->sighand->siglock)
2700{
2701 /*
2702 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2703 * let's make another loop to give it a chance to be handled.
2704 * In any case, we'll return back.
2705 */
2706 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2707 JOBCTL_TRAP_FREEZE) {
2708 spin_unlock_irq(¤t->sighand->siglock);
2709 return;
2710 }
2711
2712 /*
2713 * Now we're sure that there is no pending fatal signal and no
2714 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2715 * immediately (if there is a non-fatal signal pending), and
2716 * put the task into sleep.
2717 */
2718 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2719 clear_thread_flag(TIF_SIGPENDING);
2720 spin_unlock_irq(¤t->sighand->siglock);
2721 cgroup_enter_frozen();
2722 schedule();
2723
2724 /*
2725 * We could've been woken by task_work, run it to clear
2726 * TIF_NOTIFY_SIGNAL. The caller will retry if necessary.
2727 */
2728 clear_notify_signal();
2729 if (unlikely(task_work_pending(current)))
2730 task_work_run();
2731}
2732
2733static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2734{
2735 /*
2736 * We do not check sig_kernel_stop(signr) but set this marker
2737 * unconditionally because we do not know whether debugger will
2738 * change signr. This flag has no meaning unless we are going
2739 * to stop after return from ptrace_stop(). In this case it will
2740 * be checked in do_signal_stop(), we should only stop if it was
2741 * not cleared by SIGCONT while we were sleeping. See also the
2742 * comment in dequeue_signal().
2743 */
2744 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2745 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2746
2747 /* We're back. Did the debugger cancel the sig? */
2748 if (signr == 0)
2749 return signr;
2750
2751 /*
2752 * Update the siginfo structure if the signal has
2753 * changed. If the debugger wanted something
2754 * specific in the siginfo structure then it should
2755 * have updated *info via PTRACE_SETSIGINFO.
2756 */
2757 if (signr != info->si_signo) {
2758 clear_siginfo(info);
2759 info->si_signo = signr;
2760 info->si_errno = 0;
2761 info->si_code = SI_USER;
2762 rcu_read_lock();
2763 info->si_pid = task_pid_vnr(current->parent);
2764 info->si_uid = from_kuid_munged(current_user_ns(),
2765 task_uid(current->parent));
2766 rcu_read_unlock();
2767 }
2768
2769 /* If the (new) signal is now blocked, requeue it. */
2770 if (sigismember(¤t->blocked, signr) ||
2771 fatal_signal_pending(current)) {
2772 send_signal_locked(signr, info, current, type);
2773 signr = 0;
2774 }
2775
2776 return signr;
2777}
2778
2779static void hide_si_addr_tag_bits(struct ksignal *ksig)
2780{
2781 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2782 case SIL_FAULT:
2783 case SIL_FAULT_TRAPNO:
2784 case SIL_FAULT_MCEERR:
2785 case SIL_FAULT_BNDERR:
2786 case SIL_FAULT_PKUERR:
2787 case SIL_FAULT_PERF_EVENT:
2788 ksig->info.si_addr = arch_untagged_si_addr(
2789 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2790 break;
2791 case SIL_KILL:
2792 case SIL_TIMER:
2793 case SIL_POLL:
2794 case SIL_CHLD:
2795 case SIL_RT:
2796 case SIL_SYS:
2797 break;
2798 }
2799}
2800
2801bool get_signal(struct ksignal *ksig)
2802{
2803 struct sighand_struct *sighand = current->sighand;
2804 struct signal_struct *signal = current->signal;
2805 int signr;
2806
2807 clear_notify_signal();
2808 if (unlikely(task_work_pending(current)))
2809 task_work_run();
2810
2811 if (!task_sigpending(current))
2812 return false;
2813
2814 if (unlikely(uprobe_deny_signal()))
2815 return false;
2816
2817 /*
2818 * Do this once, we can't return to user-mode if freezing() == T.
2819 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2820 * thus do not need another check after return.
2821 */
2822 try_to_freeze();
2823
2824relock:
2825 spin_lock_irq(&sighand->siglock);
2826
2827 /*
2828 * Every stopped thread goes here after wakeup. Check to see if
2829 * we should notify the parent, prepare_signal(SIGCONT) encodes
2830 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2831 */
2832 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2833 int why;
2834
2835 if (signal->flags & SIGNAL_CLD_CONTINUED)
2836 why = CLD_CONTINUED;
2837 else
2838 why = CLD_STOPPED;
2839
2840 signal->flags &= ~SIGNAL_CLD_MASK;
2841
2842 spin_unlock_irq(&sighand->siglock);
2843
2844 /*
2845 * Notify the parent that we're continuing. This event is
2846 * always per-process and doesn't make whole lot of sense
2847 * for ptracers, who shouldn't consume the state via
2848 * wait(2) either, but, for backward compatibility, notify
2849 * the ptracer of the group leader too unless it's gonna be
2850 * a duplicate.
2851 */
2852 read_lock(&tasklist_lock);
2853 do_notify_parent_cldstop(current, false, why);
2854
2855 if (ptrace_reparented(current->group_leader))
2856 do_notify_parent_cldstop(current->group_leader,
2857 true, why);
2858 read_unlock(&tasklist_lock);
2859
2860 goto relock;
2861 }
2862
2863 for (;;) {
2864 struct k_sigaction *ka;
2865 enum pid_type type;
2866
2867 /* Has this task already been marked for death? */
2868 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2869 signal->group_exec_task) {
2870 signr = SIGKILL;
2871 sigdelset(¤t->pending.signal, SIGKILL);
2872 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2873 &sighand->action[SIGKILL-1]);
2874 recalc_sigpending();
2875 /*
2876 * implies do_group_exit() or return to PF_USER_WORKER,
2877 * no need to initialize ksig->info/etc.
2878 */
2879 goto fatal;
2880 }
2881
2882 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2883 do_signal_stop(0))
2884 goto relock;
2885
2886 if (unlikely(current->jobctl &
2887 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2888 if (current->jobctl & JOBCTL_TRAP_MASK) {
2889 do_jobctl_trap();
2890 spin_unlock_irq(&sighand->siglock);
2891 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2892 do_freezer_trap();
2893
2894 goto relock;
2895 }
2896
2897 /*
2898 * If the task is leaving the frozen state, let's update
2899 * cgroup counters and reset the frozen bit.
2900 */
2901 if (unlikely(cgroup_task_frozen(current))) {
2902 spin_unlock_irq(&sighand->siglock);
2903 cgroup_leave_frozen(false);
2904 goto relock;
2905 }
2906
2907 /*
2908 * Signals generated by the execution of an instruction
2909 * need to be delivered before any other pending signals
2910 * so that the instruction pointer in the signal stack
2911 * frame points to the faulting instruction.
2912 */
2913 type = PIDTYPE_PID;
2914 signr = dequeue_synchronous_signal(&ksig->info);
2915 if (!signr)
2916 signr = dequeue_signal(¤t->blocked, &ksig->info, &type);
2917
2918 if (!signr)
2919 break; /* will return 0 */
2920
2921 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2922 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2923 signr = ptrace_signal(signr, &ksig->info, type);
2924 if (!signr)
2925 continue;
2926 }
2927
2928 ka = &sighand->action[signr-1];
2929
2930 /* Trace actually delivered signals. */
2931 trace_signal_deliver(signr, &ksig->info, ka);
2932
2933 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2934 continue;
2935 if (ka->sa.sa_handler != SIG_DFL) {
2936 /* Run the handler. */
2937 ksig->ka = *ka;
2938
2939 if (ka->sa.sa_flags & SA_ONESHOT)
2940 ka->sa.sa_handler = SIG_DFL;
2941
2942 break; /* will return non-zero "signr" value */
2943 }
2944
2945 /*
2946 * Now we are doing the default action for this signal.
2947 */
2948 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2949 continue;
2950
2951 /*
2952 * Global init gets no signals it doesn't want.
2953 * Container-init gets no signals it doesn't want from same
2954 * container.
2955 *
2956 * Note that if global/container-init sees a sig_kernel_only()
2957 * signal here, the signal must have been generated internally
2958 * or must have come from an ancestor namespace. In either
2959 * case, the signal cannot be dropped.
2960 */
2961 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2962 !sig_kernel_only(signr))
2963 continue;
2964
2965 if (sig_kernel_stop(signr)) {
2966 /*
2967 * The default action is to stop all threads in
2968 * the thread group. The job control signals
2969 * do nothing in an orphaned pgrp, but SIGSTOP
2970 * always works. Note that siglock needs to be
2971 * dropped during the call to is_orphaned_pgrp()
2972 * because of lock ordering with tasklist_lock.
2973 * This allows an intervening SIGCONT to be posted.
2974 * We need to check for that and bail out if necessary.
2975 */
2976 if (signr != SIGSTOP) {
2977 spin_unlock_irq(&sighand->siglock);
2978
2979 /* signals can be posted during this window */
2980
2981 if (is_current_pgrp_orphaned())
2982 goto relock;
2983
2984 spin_lock_irq(&sighand->siglock);
2985 }
2986
2987 if (likely(do_signal_stop(signr))) {
2988 /* It released the siglock. */
2989 goto relock;
2990 }
2991
2992 /*
2993 * We didn't actually stop, due to a race
2994 * with SIGCONT or something like that.
2995 */
2996 continue;
2997 }
2998
2999 fatal:
3000 spin_unlock_irq(&sighand->siglock);
3001 if (unlikely(cgroup_task_frozen(current)))
3002 cgroup_leave_frozen(true);
3003
3004 /*
3005 * Anything else is fatal, maybe with a core dump.
3006 */
3007 current->flags |= PF_SIGNALED;
3008
3009 if (sig_kernel_coredump(signr)) {
3010 if (print_fatal_signals)
3011 print_fatal_signal(signr);
3012 proc_coredump_connector(current);
3013 /*
3014 * If it was able to dump core, this kills all
3015 * other threads in the group and synchronizes with
3016 * their demise. If we lost the race with another
3017 * thread getting here, it set group_exit_code
3018 * first and our do_group_exit call below will use
3019 * that value and ignore the one we pass it.
3020 */
3021 do_coredump(&ksig->info);
3022 }
3023
3024 /*
3025 * PF_USER_WORKER threads will catch and exit on fatal signals
3026 * themselves. They have cleanup that must be performed, so we
3027 * cannot call do_exit() on their behalf. Note that ksig won't
3028 * be properly initialized, PF_USER_WORKER's shouldn't use it.
3029 */
3030 if (current->flags & PF_USER_WORKER)
3031 goto out;
3032
3033 /*
3034 * Death signals, no core dump.
3035 */
3036 do_group_exit(signr);
3037 /* NOTREACHED */
3038 }
3039 spin_unlock_irq(&sighand->siglock);
3040
3041 ksig->sig = signr;
3042
3043 if (signr && !(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
3044 hide_si_addr_tag_bits(ksig);
3045out:
3046 return signr > 0;
3047}
3048
3049/**
3050 * signal_delivered - called after signal delivery to update blocked signals
3051 * @ksig: kernel signal struct
3052 * @stepping: nonzero if debugger single-step or block-step in use
3053 *
3054 * This function should be called when a signal has successfully been
3055 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
3056 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
3057 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
3058 */
3059static void signal_delivered(struct ksignal *ksig, int stepping)
3060{
3061 sigset_t blocked;
3062
3063 /* A signal was successfully delivered, and the
3064 saved sigmask was stored on the signal frame,
3065 and will be restored by sigreturn. So we can
3066 simply clear the restore sigmask flag. */
3067 clear_restore_sigmask();
3068
3069 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
3070 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
3071 sigaddset(&blocked, ksig->sig);
3072 set_current_blocked(&blocked);
3073 if (current->sas_ss_flags & SS_AUTODISARM)
3074 sas_ss_reset(current);
3075 if (stepping)
3076 ptrace_notify(SIGTRAP, 0);
3077}
3078
3079void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
3080{
3081 if (failed)
3082 force_sigsegv(ksig->sig);
3083 else
3084 signal_delivered(ksig, stepping);
3085}
3086
3087/*
3088 * It could be that complete_signal() picked us to notify about the
3089 * group-wide signal. Other threads should be notified now to take
3090 * the shared signals in @which since we will not.
3091 */
3092static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
3093{
3094 sigset_t retarget;
3095 struct task_struct *t;
3096
3097 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
3098 if (sigisemptyset(&retarget))
3099 return;
3100
3101 for_other_threads(tsk, t) {
3102 if (t->flags & PF_EXITING)
3103 continue;
3104
3105 if (!has_pending_signals(&retarget, &t->blocked))
3106 continue;
3107 /* Remove the signals this thread can handle. */
3108 sigandsets(&retarget, &retarget, &t->blocked);
3109
3110 if (!task_sigpending(t))
3111 signal_wake_up(t, 0);
3112
3113 if (sigisemptyset(&retarget))
3114 break;
3115 }
3116}
3117
3118void exit_signals(struct task_struct *tsk)
3119{
3120 int group_stop = 0;
3121 sigset_t unblocked;
3122
3123 /*
3124 * @tsk is about to have PF_EXITING set - lock out users which
3125 * expect stable threadgroup.
3126 */
3127 cgroup_threadgroup_change_begin(tsk);
3128
3129 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
3130 sched_mm_cid_exit_signals(tsk);
3131 tsk->flags |= PF_EXITING;
3132 cgroup_threadgroup_change_end(tsk);
3133 return;
3134 }
3135
3136 spin_lock_irq(&tsk->sighand->siglock);
3137 /*
3138 * From now this task is not visible for group-wide signals,
3139 * see wants_signal(), do_signal_stop().
3140 */
3141 sched_mm_cid_exit_signals(tsk);
3142 tsk->flags |= PF_EXITING;
3143
3144 cgroup_threadgroup_change_end(tsk);
3145
3146 if (!task_sigpending(tsk))
3147 goto out;
3148
3149 unblocked = tsk->blocked;
3150 signotset(&unblocked);
3151 retarget_shared_pending(tsk, &unblocked);
3152
3153 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3154 task_participate_group_stop(tsk))
3155 group_stop = CLD_STOPPED;
3156out:
3157 spin_unlock_irq(&tsk->sighand->siglock);
3158
3159 /*
3160 * If group stop has completed, deliver the notification. This
3161 * should always go to the real parent of the group leader.
3162 */
3163 if (unlikely(group_stop)) {
3164 read_lock(&tasklist_lock);
3165 do_notify_parent_cldstop(tsk, false, group_stop);
3166 read_unlock(&tasklist_lock);
3167 }
3168}
3169
3170/*
3171 * System call entry points.
3172 */
3173
3174/**
3175 * sys_restart_syscall - restart a system call
3176 */
3177SYSCALL_DEFINE0(restart_syscall)
3178{
3179 struct restart_block *restart = ¤t->restart_block;
3180 return restart->fn(restart);
3181}
3182
3183long do_no_restart_syscall(struct restart_block *param)
3184{
3185 return -EINTR;
3186}
3187
3188static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3189{
3190 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3191 sigset_t newblocked;
3192 /* A set of now blocked but previously unblocked signals. */
3193 sigandnsets(&newblocked, newset, ¤t->blocked);
3194 retarget_shared_pending(tsk, &newblocked);
3195 }
3196 tsk->blocked = *newset;
3197 recalc_sigpending();
3198}
3199
3200/**
3201 * set_current_blocked - change current->blocked mask
3202 * @newset: new mask
3203 *
3204 * It is wrong to change ->blocked directly, this helper should be used
3205 * to ensure the process can't miss a shared signal we are going to block.
3206 */
3207void set_current_blocked(sigset_t *newset)
3208{
3209 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3210 __set_current_blocked(newset);
3211}
3212
3213void __set_current_blocked(const sigset_t *newset)
3214{
3215 struct task_struct *tsk = current;
3216
3217 /*
3218 * In case the signal mask hasn't changed, there is nothing we need
3219 * to do. The current->blocked shouldn't be modified by other task.
3220 */
3221 if (sigequalsets(&tsk->blocked, newset))
3222 return;
3223
3224 spin_lock_irq(&tsk->sighand->siglock);
3225 __set_task_blocked(tsk, newset);
3226 spin_unlock_irq(&tsk->sighand->siglock);
3227}
3228
3229/*
3230 * This is also useful for kernel threads that want to temporarily
3231 * (or permanently) block certain signals.
3232 *
3233 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3234 * interface happily blocks "unblockable" signals like SIGKILL
3235 * and friends.
3236 */
3237int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3238{
3239 struct task_struct *tsk = current;
3240 sigset_t newset;
3241
3242 /* Lockless, only current can change ->blocked, never from irq */
3243 if (oldset)
3244 *oldset = tsk->blocked;
3245
3246 switch (how) {
3247 case SIG_BLOCK:
3248 sigorsets(&newset, &tsk->blocked, set);
3249 break;
3250 case SIG_UNBLOCK:
3251 sigandnsets(&newset, &tsk->blocked, set);
3252 break;
3253 case SIG_SETMASK:
3254 newset = *set;
3255 break;
3256 default:
3257 return -EINVAL;
3258 }
3259
3260 __set_current_blocked(&newset);
3261 return 0;
3262}
3263EXPORT_SYMBOL(sigprocmask);
3264
3265/*
3266 * The api helps set app-provided sigmasks.
3267 *
3268 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3269 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3270 *
3271 * Note that it does set_restore_sigmask() in advance, so it must be always
3272 * paired with restore_saved_sigmask_unless() before return from syscall.
3273 */
3274int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3275{
3276 sigset_t kmask;
3277
3278 if (!umask)
3279 return 0;
3280 if (sigsetsize != sizeof(sigset_t))
3281 return -EINVAL;
3282 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3283 return -EFAULT;
3284
3285 set_restore_sigmask();
3286 current->saved_sigmask = current->blocked;
3287 set_current_blocked(&kmask);
3288
3289 return 0;
3290}
3291
3292#ifdef CONFIG_COMPAT
3293int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3294 size_t sigsetsize)
3295{
3296 sigset_t kmask;
3297
3298 if (!umask)
3299 return 0;
3300 if (sigsetsize != sizeof(compat_sigset_t))
3301 return -EINVAL;
3302 if (get_compat_sigset(&kmask, umask))
3303 return -EFAULT;
3304
3305 set_restore_sigmask();
3306 current->saved_sigmask = current->blocked;
3307 set_current_blocked(&kmask);
3308
3309 return 0;
3310}
3311#endif
3312
3313/**
3314 * sys_rt_sigprocmask - change the list of currently blocked signals
3315 * @how: whether to add, remove, or set signals
3316 * @nset: stores pending signals
3317 * @oset: previous value of signal mask if non-null
3318 * @sigsetsize: size of sigset_t type
3319 */
3320SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3321 sigset_t __user *, oset, size_t, sigsetsize)
3322{
3323 sigset_t old_set, new_set;
3324 int error;
3325
3326 /* XXX: Don't preclude handling different sized sigset_t's. */
3327 if (sigsetsize != sizeof(sigset_t))
3328 return -EINVAL;
3329
3330 old_set = current->blocked;
3331
3332 if (nset) {
3333 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3334 return -EFAULT;
3335 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3336
3337 error = sigprocmask(how, &new_set, NULL);
3338 if (error)
3339 return error;
3340 }
3341
3342 if (oset) {
3343 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3344 return -EFAULT;
3345 }
3346
3347 return 0;
3348}
3349
3350#ifdef CONFIG_COMPAT
3351COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3352 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3353{
3354 sigset_t old_set = current->blocked;
3355
3356 /* XXX: Don't preclude handling different sized sigset_t's. */
3357 if (sigsetsize != sizeof(sigset_t))
3358 return -EINVAL;
3359
3360 if (nset) {
3361 sigset_t new_set;
3362 int error;
3363 if (get_compat_sigset(&new_set, nset))
3364 return -EFAULT;
3365 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3366
3367 error = sigprocmask(how, &new_set, NULL);
3368 if (error)
3369 return error;
3370 }
3371 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3372}
3373#endif
3374
3375static void do_sigpending(sigset_t *set)
3376{
3377 spin_lock_irq(¤t->sighand->siglock);
3378 sigorsets(set, ¤t->pending.signal,
3379 ¤t->signal->shared_pending.signal);
3380 spin_unlock_irq(¤t->sighand->siglock);
3381
3382 /* Outside the lock because only this thread touches it. */
3383 sigandsets(set, ¤t->blocked, set);
3384}
3385
3386/**
3387 * sys_rt_sigpending - examine a pending signal that has been raised
3388 * while blocked
3389 * @uset: stores pending signals
3390 * @sigsetsize: size of sigset_t type or larger
3391 */
3392SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3393{
3394 sigset_t set;
3395
3396 if (sigsetsize > sizeof(*uset))
3397 return -EINVAL;
3398
3399 do_sigpending(&set);
3400
3401 if (copy_to_user(uset, &set, sigsetsize))
3402 return -EFAULT;
3403
3404 return 0;
3405}
3406
3407#ifdef CONFIG_COMPAT
3408COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3409 compat_size_t, sigsetsize)
3410{
3411 sigset_t set;
3412
3413 if (sigsetsize > sizeof(*uset))
3414 return -EINVAL;
3415
3416 do_sigpending(&set);
3417
3418 return put_compat_sigset(uset, &set, sigsetsize);
3419}
3420#endif
3421
3422static const struct {
3423 unsigned char limit, layout;
3424} sig_sicodes[] = {
3425 [SIGILL] = { NSIGILL, SIL_FAULT },
3426 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3427 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3428 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3429 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3430#if defined(SIGEMT)
3431 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3432#endif
3433 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3434 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3435 [SIGSYS] = { NSIGSYS, SIL_SYS },
3436};
3437
3438static bool known_siginfo_layout(unsigned sig, int si_code)
3439{
3440 if (si_code == SI_KERNEL)
3441 return true;
3442 else if ((si_code > SI_USER)) {
3443 if (sig_specific_sicodes(sig)) {
3444 if (si_code <= sig_sicodes[sig].limit)
3445 return true;
3446 }
3447 else if (si_code <= NSIGPOLL)
3448 return true;
3449 }
3450 else if (si_code >= SI_DETHREAD)
3451 return true;
3452 else if (si_code == SI_ASYNCNL)
3453 return true;
3454 return false;
3455}
3456
3457enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3458{
3459 enum siginfo_layout layout = SIL_KILL;
3460 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3461 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3462 (si_code <= sig_sicodes[sig].limit)) {
3463 layout = sig_sicodes[sig].layout;
3464 /* Handle the exceptions */
3465 if ((sig == SIGBUS) &&
3466 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3467 layout = SIL_FAULT_MCEERR;
3468 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3469 layout = SIL_FAULT_BNDERR;
3470#ifdef SEGV_PKUERR
3471 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3472 layout = SIL_FAULT_PKUERR;
3473#endif
3474 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3475 layout = SIL_FAULT_PERF_EVENT;
3476 else if (IS_ENABLED(CONFIG_SPARC) &&
3477 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3478 layout = SIL_FAULT_TRAPNO;
3479 else if (IS_ENABLED(CONFIG_ALPHA) &&
3480 ((sig == SIGFPE) ||
3481 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3482 layout = SIL_FAULT_TRAPNO;
3483 }
3484 else if (si_code <= NSIGPOLL)
3485 layout = SIL_POLL;
3486 } else {
3487 if (si_code == SI_TIMER)
3488 layout = SIL_TIMER;
3489 else if (si_code == SI_SIGIO)
3490 layout = SIL_POLL;
3491 else if (si_code < 0)
3492 layout = SIL_RT;
3493 }
3494 return layout;
3495}
3496
3497static inline char __user *si_expansion(const siginfo_t __user *info)
3498{
3499 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3500}
3501
3502int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3503{
3504 char __user *expansion = si_expansion(to);
3505 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3506 return -EFAULT;
3507 if (clear_user(expansion, SI_EXPANSION_SIZE))
3508 return -EFAULT;
3509 return 0;
3510}
3511
3512static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3513 const siginfo_t __user *from)
3514{
3515 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3516 char __user *expansion = si_expansion(from);
3517 char buf[SI_EXPANSION_SIZE];
3518 int i;
3519 /*
3520 * An unknown si_code might need more than
3521 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3522 * extra bytes are 0. This guarantees copy_siginfo_to_user
3523 * will return this data to userspace exactly.
3524 */
3525 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3526 return -EFAULT;
3527 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3528 if (buf[i] != 0)
3529 return -E2BIG;
3530 }
3531 }
3532 return 0;
3533}
3534
3535static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3536 const siginfo_t __user *from)
3537{
3538 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3539 return -EFAULT;
3540 to->si_signo = signo;
3541 return post_copy_siginfo_from_user(to, from);
3542}
3543
3544int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3545{
3546 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3547 return -EFAULT;
3548 return post_copy_siginfo_from_user(to, from);
3549}
3550
3551#ifdef CONFIG_COMPAT
3552/**
3553 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3554 * @to: compat siginfo destination
3555 * @from: kernel siginfo source
3556 *
3557 * Note: This function does not work properly for the SIGCHLD on x32, but
3558 * fortunately it doesn't have to. The only valid callers for this function are
3559 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3560 * The latter does not care because SIGCHLD will never cause a coredump.
3561 */
3562void copy_siginfo_to_external32(struct compat_siginfo *to,
3563 const struct kernel_siginfo *from)
3564{
3565 memset(to, 0, sizeof(*to));
3566
3567 to->si_signo = from->si_signo;
3568 to->si_errno = from->si_errno;
3569 to->si_code = from->si_code;
3570 switch(siginfo_layout(from->si_signo, from->si_code)) {
3571 case SIL_KILL:
3572 to->si_pid = from->si_pid;
3573 to->si_uid = from->si_uid;
3574 break;
3575 case SIL_TIMER:
3576 to->si_tid = from->si_tid;
3577 to->si_overrun = from->si_overrun;
3578 to->si_int = from->si_int;
3579 break;
3580 case SIL_POLL:
3581 to->si_band = from->si_band;
3582 to->si_fd = from->si_fd;
3583 break;
3584 case SIL_FAULT:
3585 to->si_addr = ptr_to_compat(from->si_addr);
3586 break;
3587 case SIL_FAULT_TRAPNO:
3588 to->si_addr = ptr_to_compat(from->si_addr);
3589 to->si_trapno = from->si_trapno;
3590 break;
3591 case SIL_FAULT_MCEERR:
3592 to->si_addr = ptr_to_compat(from->si_addr);
3593 to->si_addr_lsb = from->si_addr_lsb;
3594 break;
3595 case SIL_FAULT_BNDERR:
3596 to->si_addr = ptr_to_compat(from->si_addr);
3597 to->si_lower = ptr_to_compat(from->si_lower);
3598 to->si_upper = ptr_to_compat(from->si_upper);
3599 break;
3600 case SIL_FAULT_PKUERR:
3601 to->si_addr = ptr_to_compat(from->si_addr);
3602 to->si_pkey = from->si_pkey;
3603 break;
3604 case SIL_FAULT_PERF_EVENT:
3605 to->si_addr = ptr_to_compat(from->si_addr);
3606 to->si_perf_data = from->si_perf_data;
3607 to->si_perf_type = from->si_perf_type;
3608 to->si_perf_flags = from->si_perf_flags;
3609 break;
3610 case SIL_CHLD:
3611 to->si_pid = from->si_pid;
3612 to->si_uid = from->si_uid;
3613 to->si_status = from->si_status;
3614 to->si_utime = from->si_utime;
3615 to->si_stime = from->si_stime;
3616 break;
3617 case SIL_RT:
3618 to->si_pid = from->si_pid;
3619 to->si_uid = from->si_uid;
3620 to->si_int = from->si_int;
3621 break;
3622 case SIL_SYS:
3623 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3624 to->si_syscall = from->si_syscall;
3625 to->si_arch = from->si_arch;
3626 break;
3627 }
3628}
3629
3630int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3631 const struct kernel_siginfo *from)
3632{
3633 struct compat_siginfo new;
3634
3635 copy_siginfo_to_external32(&new, from);
3636 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3637 return -EFAULT;
3638 return 0;
3639}
3640
3641static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3642 const struct compat_siginfo *from)
3643{
3644 clear_siginfo(to);
3645 to->si_signo = from->si_signo;
3646 to->si_errno = from->si_errno;
3647 to->si_code = from->si_code;
3648 switch(siginfo_layout(from->si_signo, from->si_code)) {
3649 case SIL_KILL:
3650 to->si_pid = from->si_pid;
3651 to->si_uid = from->si_uid;
3652 break;
3653 case SIL_TIMER:
3654 to->si_tid = from->si_tid;
3655 to->si_overrun = from->si_overrun;
3656 to->si_int = from->si_int;
3657 break;
3658 case SIL_POLL:
3659 to->si_band = from->si_band;
3660 to->si_fd = from->si_fd;
3661 break;
3662 case SIL_FAULT:
3663 to->si_addr = compat_ptr(from->si_addr);
3664 break;
3665 case SIL_FAULT_TRAPNO:
3666 to->si_addr = compat_ptr(from->si_addr);
3667 to->si_trapno = from->si_trapno;
3668 break;
3669 case SIL_FAULT_MCEERR:
3670 to->si_addr = compat_ptr(from->si_addr);
3671 to->si_addr_lsb = from->si_addr_lsb;
3672 break;
3673 case SIL_FAULT_BNDERR:
3674 to->si_addr = compat_ptr(from->si_addr);
3675 to->si_lower = compat_ptr(from->si_lower);
3676 to->si_upper = compat_ptr(from->si_upper);
3677 break;
3678 case SIL_FAULT_PKUERR:
3679 to->si_addr = compat_ptr(from->si_addr);
3680 to->si_pkey = from->si_pkey;
3681 break;
3682 case SIL_FAULT_PERF_EVENT:
3683 to->si_addr = compat_ptr(from->si_addr);
3684 to->si_perf_data = from->si_perf_data;
3685 to->si_perf_type = from->si_perf_type;
3686 to->si_perf_flags = from->si_perf_flags;
3687 break;
3688 case SIL_CHLD:
3689 to->si_pid = from->si_pid;
3690 to->si_uid = from->si_uid;
3691 to->si_status = from->si_status;
3692#ifdef CONFIG_X86_X32_ABI
3693 if (in_x32_syscall()) {
3694 to->si_utime = from->_sifields._sigchld_x32._utime;
3695 to->si_stime = from->_sifields._sigchld_x32._stime;
3696 } else
3697#endif
3698 {
3699 to->si_utime = from->si_utime;
3700 to->si_stime = from->si_stime;
3701 }
3702 break;
3703 case SIL_RT:
3704 to->si_pid = from->si_pid;
3705 to->si_uid = from->si_uid;
3706 to->si_int = from->si_int;
3707 break;
3708 case SIL_SYS:
3709 to->si_call_addr = compat_ptr(from->si_call_addr);
3710 to->si_syscall = from->si_syscall;
3711 to->si_arch = from->si_arch;
3712 break;
3713 }
3714 return 0;
3715}
3716
3717static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3718 const struct compat_siginfo __user *ufrom)
3719{
3720 struct compat_siginfo from;
3721
3722 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3723 return -EFAULT;
3724
3725 from.si_signo = signo;
3726 return post_copy_siginfo_from_user32(to, &from);
3727}
3728
3729int copy_siginfo_from_user32(struct kernel_siginfo *to,
3730 const struct compat_siginfo __user *ufrom)
3731{
3732 struct compat_siginfo from;
3733
3734 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3735 return -EFAULT;
3736
3737 return post_copy_siginfo_from_user32(to, &from);
3738}
3739#endif /* CONFIG_COMPAT */
3740
3741/**
3742 * do_sigtimedwait - wait for queued signals specified in @which
3743 * @which: queued signals to wait for
3744 * @info: if non-null, the signal's siginfo is returned here
3745 * @ts: upper bound on process time suspension
3746 */
3747static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3748 const struct timespec64 *ts)
3749{
3750 ktime_t *to = NULL, timeout = KTIME_MAX;
3751 struct task_struct *tsk = current;
3752 sigset_t mask = *which;
3753 enum pid_type type;
3754 int sig, ret = 0;
3755
3756 if (ts) {
3757 if (!timespec64_valid(ts))
3758 return -EINVAL;
3759 timeout = timespec64_to_ktime(*ts);
3760 to = &timeout;
3761 }
3762
3763 /*
3764 * Invert the set of allowed signals to get those we want to block.
3765 */
3766 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3767 signotset(&mask);
3768
3769 spin_lock_irq(&tsk->sighand->siglock);
3770 sig = dequeue_signal(&mask, info, &type);
3771 if (!sig && timeout) {
3772 /*
3773 * None ready, temporarily unblock those we're interested
3774 * while we are sleeping in so that we'll be awakened when
3775 * they arrive. Unblocking is always fine, we can avoid
3776 * set_current_blocked().
3777 */
3778 tsk->real_blocked = tsk->blocked;
3779 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3780 recalc_sigpending();
3781 spin_unlock_irq(&tsk->sighand->siglock);
3782
3783 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3784 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3785 HRTIMER_MODE_REL);
3786 spin_lock_irq(&tsk->sighand->siglock);
3787 __set_task_blocked(tsk, &tsk->real_blocked);
3788 sigemptyset(&tsk->real_blocked);
3789 sig = dequeue_signal(&mask, info, &type);
3790 }
3791 spin_unlock_irq(&tsk->sighand->siglock);
3792
3793 if (sig)
3794 return sig;
3795 return ret ? -EINTR : -EAGAIN;
3796}
3797
3798/**
3799 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3800 * in @uthese
3801 * @uthese: queued signals to wait for
3802 * @uinfo: if non-null, the signal's siginfo is returned here
3803 * @uts: upper bound on process time suspension
3804 * @sigsetsize: size of sigset_t type
3805 */
3806SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3807 siginfo_t __user *, uinfo,
3808 const struct __kernel_timespec __user *, uts,
3809 size_t, sigsetsize)
3810{
3811 sigset_t these;
3812 struct timespec64 ts;
3813 kernel_siginfo_t info;
3814 int ret;
3815
3816 /* XXX: Don't preclude handling different sized sigset_t's. */
3817 if (sigsetsize != sizeof(sigset_t))
3818 return -EINVAL;
3819
3820 if (copy_from_user(&these, uthese, sizeof(these)))
3821 return -EFAULT;
3822
3823 if (uts) {
3824 if (get_timespec64(&ts, uts))
3825 return -EFAULT;
3826 }
3827
3828 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3829
3830 if (ret > 0 && uinfo) {
3831 if (copy_siginfo_to_user(uinfo, &info))
3832 ret = -EFAULT;
3833 }
3834
3835 return ret;
3836}
3837
3838#ifdef CONFIG_COMPAT_32BIT_TIME
3839SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3840 siginfo_t __user *, uinfo,
3841 const struct old_timespec32 __user *, uts,
3842 size_t, sigsetsize)
3843{
3844 sigset_t these;
3845 struct timespec64 ts;
3846 kernel_siginfo_t info;
3847 int ret;
3848
3849 if (sigsetsize != sizeof(sigset_t))
3850 return -EINVAL;
3851
3852 if (copy_from_user(&these, uthese, sizeof(these)))
3853 return -EFAULT;
3854
3855 if (uts) {
3856 if (get_old_timespec32(&ts, uts))
3857 return -EFAULT;
3858 }
3859
3860 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3861
3862 if (ret > 0 && uinfo) {
3863 if (copy_siginfo_to_user(uinfo, &info))
3864 ret = -EFAULT;
3865 }
3866
3867 return ret;
3868}
3869#endif
3870
3871#ifdef CONFIG_COMPAT
3872COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3873 struct compat_siginfo __user *, uinfo,
3874 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3875{
3876 sigset_t s;
3877 struct timespec64 t;
3878 kernel_siginfo_t info;
3879 long ret;
3880
3881 if (sigsetsize != sizeof(sigset_t))
3882 return -EINVAL;
3883
3884 if (get_compat_sigset(&s, uthese))
3885 return -EFAULT;
3886
3887 if (uts) {
3888 if (get_timespec64(&t, uts))
3889 return -EFAULT;
3890 }
3891
3892 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3893
3894 if (ret > 0 && uinfo) {
3895 if (copy_siginfo_to_user32(uinfo, &info))
3896 ret = -EFAULT;
3897 }
3898
3899 return ret;
3900}
3901
3902#ifdef CONFIG_COMPAT_32BIT_TIME
3903COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3904 struct compat_siginfo __user *, uinfo,
3905 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3906{
3907 sigset_t s;
3908 struct timespec64 t;
3909 kernel_siginfo_t info;
3910 long ret;
3911
3912 if (sigsetsize != sizeof(sigset_t))
3913 return -EINVAL;
3914
3915 if (get_compat_sigset(&s, uthese))
3916 return -EFAULT;
3917
3918 if (uts) {
3919 if (get_old_timespec32(&t, uts))
3920 return -EFAULT;
3921 }
3922
3923 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3924
3925 if (ret > 0 && uinfo) {
3926 if (copy_siginfo_to_user32(uinfo, &info))
3927 ret = -EFAULT;
3928 }
3929
3930 return ret;
3931}
3932#endif
3933#endif
3934
3935static void prepare_kill_siginfo(int sig, struct kernel_siginfo *info,
3936 enum pid_type type)
3937{
3938 clear_siginfo(info);
3939 info->si_signo = sig;
3940 info->si_errno = 0;
3941 info->si_code = (type == PIDTYPE_PID) ? SI_TKILL : SI_USER;
3942 info->si_pid = task_tgid_vnr(current);
3943 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3944}
3945
3946/**
3947 * sys_kill - send a signal to a process
3948 * @pid: the PID of the process
3949 * @sig: signal to be sent
3950 */
3951SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3952{
3953 struct kernel_siginfo info;
3954
3955 prepare_kill_siginfo(sig, &info, PIDTYPE_TGID);
3956
3957 return kill_something_info(sig, &info, pid);
3958}
3959
3960/*
3961 * Verify that the signaler and signalee either are in the same pid namespace
3962 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3963 * namespace.
3964 */
3965static bool access_pidfd_pidns(struct pid *pid)
3966{
3967 struct pid_namespace *active = task_active_pid_ns(current);
3968 struct pid_namespace *p = ns_of_pid(pid);
3969
3970 for (;;) {
3971 if (!p)
3972 return false;
3973 if (p == active)
3974 break;
3975 p = p->parent;
3976 }
3977
3978 return true;
3979}
3980
3981static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3982 siginfo_t __user *info)
3983{
3984#ifdef CONFIG_COMPAT
3985 /*
3986 * Avoid hooking up compat syscalls and instead handle necessary
3987 * conversions here. Note, this is a stop-gap measure and should not be
3988 * considered a generic solution.
3989 */
3990 if (in_compat_syscall())
3991 return copy_siginfo_from_user32(
3992 kinfo, (struct compat_siginfo __user *)info);
3993#endif
3994 return copy_siginfo_from_user(kinfo, info);
3995}
3996
3997static struct pid *pidfd_to_pid(const struct file *file)
3998{
3999 struct pid *pid;
4000
4001 pid = pidfd_pid(file);
4002 if (!IS_ERR(pid))
4003 return pid;
4004
4005 return tgid_pidfd_to_pid(file);
4006}
4007
4008#define PIDFD_SEND_SIGNAL_FLAGS \
4009 (PIDFD_SIGNAL_THREAD | PIDFD_SIGNAL_THREAD_GROUP | \
4010 PIDFD_SIGNAL_PROCESS_GROUP)
4011
4012/**
4013 * sys_pidfd_send_signal - Signal a process through a pidfd
4014 * @pidfd: file descriptor of the process
4015 * @sig: signal to send
4016 * @info: signal info
4017 * @flags: future flags
4018 *
4019 * Send the signal to the thread group or to the individual thread depending
4020 * on PIDFD_THREAD.
4021 * In the future extension to @flags may be used to override the default scope
4022 * of @pidfd.
4023 *
4024 * Return: 0 on success, negative errno on failure
4025 */
4026SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
4027 siginfo_t __user *, info, unsigned int, flags)
4028{
4029 int ret;
4030 struct pid *pid;
4031 kernel_siginfo_t kinfo;
4032 enum pid_type type;
4033
4034 /* Enforce flags be set to 0 until we add an extension. */
4035 if (flags & ~PIDFD_SEND_SIGNAL_FLAGS)
4036 return -EINVAL;
4037
4038 /* Ensure that only a single signal scope determining flag is set. */
4039 if (hweight32(flags & PIDFD_SEND_SIGNAL_FLAGS) > 1)
4040 return -EINVAL;
4041
4042 CLASS(fd, f)(pidfd);
4043 if (fd_empty(f))
4044 return -EBADF;
4045
4046 /* Is this a pidfd? */
4047 pid = pidfd_to_pid(fd_file(f));
4048 if (IS_ERR(pid))
4049 return PTR_ERR(pid);
4050
4051 if (!access_pidfd_pidns(pid))
4052 return -EINVAL;
4053
4054 switch (flags) {
4055 case 0:
4056 /* Infer scope from the type of pidfd. */
4057 if (fd_file(f)->f_flags & PIDFD_THREAD)
4058 type = PIDTYPE_PID;
4059 else
4060 type = PIDTYPE_TGID;
4061 break;
4062 case PIDFD_SIGNAL_THREAD:
4063 type = PIDTYPE_PID;
4064 break;
4065 case PIDFD_SIGNAL_THREAD_GROUP:
4066 type = PIDTYPE_TGID;
4067 break;
4068 case PIDFD_SIGNAL_PROCESS_GROUP:
4069 type = PIDTYPE_PGID;
4070 break;
4071 }
4072
4073 if (info) {
4074 ret = copy_siginfo_from_user_any(&kinfo, info);
4075 if (unlikely(ret))
4076 return ret;
4077
4078 if (unlikely(sig != kinfo.si_signo))
4079 return -EINVAL;
4080
4081 /* Only allow sending arbitrary signals to yourself. */
4082 if ((task_pid(current) != pid || type > PIDTYPE_TGID) &&
4083 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
4084 return -EPERM;
4085 } else {
4086 prepare_kill_siginfo(sig, &kinfo, type);
4087 }
4088
4089 if (type == PIDTYPE_PGID)
4090 return kill_pgrp_info(sig, &kinfo, pid);
4091 else
4092 return kill_pid_info_type(sig, &kinfo, pid, type);
4093}
4094
4095static int
4096do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
4097{
4098 struct task_struct *p;
4099 int error = -ESRCH;
4100
4101 rcu_read_lock();
4102 p = find_task_by_vpid(pid);
4103 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
4104 error = check_kill_permission(sig, info, p);
4105 /*
4106 * The null signal is a permissions and process existence
4107 * probe. No signal is actually delivered.
4108 */
4109 if (!error && sig) {
4110 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
4111 /*
4112 * If lock_task_sighand() failed we pretend the task
4113 * dies after receiving the signal. The window is tiny,
4114 * and the signal is private anyway.
4115 */
4116 if (unlikely(error == -ESRCH))
4117 error = 0;
4118 }
4119 }
4120 rcu_read_unlock();
4121
4122 return error;
4123}
4124
4125static int do_tkill(pid_t tgid, pid_t pid, int sig)
4126{
4127 struct kernel_siginfo info;
4128
4129 prepare_kill_siginfo(sig, &info, PIDTYPE_PID);
4130
4131 return do_send_specific(tgid, pid, sig, &info);
4132}
4133
4134/**
4135 * sys_tgkill - send signal to one specific thread
4136 * @tgid: the thread group ID of the thread
4137 * @pid: the PID of the thread
4138 * @sig: signal to be sent
4139 *
4140 * This syscall also checks the @tgid and returns -ESRCH even if the PID
4141 * exists but it's not belonging to the target process anymore. This
4142 * method solves the problem of threads exiting and PIDs getting reused.
4143 */
4144SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
4145{
4146 /* This is only valid for single tasks */
4147 if (pid <= 0 || tgid <= 0)
4148 return -EINVAL;
4149
4150 return do_tkill(tgid, pid, sig);
4151}
4152
4153/**
4154 * sys_tkill - send signal to one specific task
4155 * @pid: the PID of the task
4156 * @sig: signal to be sent
4157 *
4158 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4159 */
4160SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4161{
4162 /* This is only valid for single tasks */
4163 if (pid <= 0)
4164 return -EINVAL;
4165
4166 return do_tkill(0, pid, sig);
4167}
4168
4169static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4170{
4171 /* Not even root can pretend to send signals from the kernel.
4172 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4173 */
4174 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4175 (task_pid_vnr(current) != pid))
4176 return -EPERM;
4177
4178 /* POSIX.1b doesn't mention process groups. */
4179 return kill_proc_info(sig, info, pid);
4180}
4181
4182/**
4183 * sys_rt_sigqueueinfo - send signal information to a signal
4184 * @pid: the PID of the thread
4185 * @sig: signal to be sent
4186 * @uinfo: signal info to be sent
4187 */
4188SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4189 siginfo_t __user *, uinfo)
4190{
4191 kernel_siginfo_t info;
4192 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4193 if (unlikely(ret))
4194 return ret;
4195 return do_rt_sigqueueinfo(pid, sig, &info);
4196}
4197
4198#ifdef CONFIG_COMPAT
4199COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4200 compat_pid_t, pid,
4201 int, sig,
4202 struct compat_siginfo __user *, uinfo)
4203{
4204 kernel_siginfo_t info;
4205 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4206 if (unlikely(ret))
4207 return ret;
4208 return do_rt_sigqueueinfo(pid, sig, &info);
4209}
4210#endif
4211
4212static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4213{
4214 /* This is only valid for single tasks */
4215 if (pid <= 0 || tgid <= 0)
4216 return -EINVAL;
4217
4218 /* Not even root can pretend to send signals from the kernel.
4219 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4220 */
4221 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4222 (task_pid_vnr(current) != pid))
4223 return -EPERM;
4224
4225 return do_send_specific(tgid, pid, sig, info);
4226}
4227
4228SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4229 siginfo_t __user *, uinfo)
4230{
4231 kernel_siginfo_t info;
4232 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4233 if (unlikely(ret))
4234 return ret;
4235 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4236}
4237
4238#ifdef CONFIG_COMPAT
4239COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4240 compat_pid_t, tgid,
4241 compat_pid_t, pid,
4242 int, sig,
4243 struct compat_siginfo __user *, uinfo)
4244{
4245 kernel_siginfo_t info;
4246 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4247 if (unlikely(ret))
4248 return ret;
4249 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4250}
4251#endif
4252
4253/*
4254 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4255 */
4256void kernel_sigaction(int sig, __sighandler_t action)
4257{
4258 spin_lock_irq(¤t->sighand->siglock);
4259 current->sighand->action[sig - 1].sa.sa_handler = action;
4260 if (action == SIG_IGN) {
4261 sigset_t mask;
4262
4263 sigemptyset(&mask);
4264 sigaddset(&mask, sig);
4265
4266 flush_sigqueue_mask(current, &mask, ¤t->signal->shared_pending);
4267 flush_sigqueue_mask(current, &mask, ¤t->pending);
4268 recalc_sigpending();
4269 }
4270 spin_unlock_irq(¤t->sighand->siglock);
4271}
4272EXPORT_SYMBOL(kernel_sigaction);
4273
4274void __weak sigaction_compat_abi(struct k_sigaction *act,
4275 struct k_sigaction *oact)
4276{
4277}
4278
4279int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4280{
4281 struct task_struct *p = current, *t;
4282 struct k_sigaction *k;
4283 sigset_t mask;
4284
4285 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4286 return -EINVAL;
4287
4288 k = &p->sighand->action[sig-1];
4289
4290 spin_lock_irq(&p->sighand->siglock);
4291 if (k->sa.sa_flags & SA_IMMUTABLE) {
4292 spin_unlock_irq(&p->sighand->siglock);
4293 return -EINVAL;
4294 }
4295 if (oact)
4296 *oact = *k;
4297
4298 /*
4299 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4300 * e.g. by having an architecture use the bit in their uapi.
4301 */
4302 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4303
4304 /*
4305 * Clear unknown flag bits in order to allow userspace to detect missing
4306 * support for flag bits and to allow the kernel to use non-uapi bits
4307 * internally.
4308 */
4309 if (act)
4310 act->sa.sa_flags &= UAPI_SA_FLAGS;
4311 if (oact)
4312 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4313
4314 sigaction_compat_abi(act, oact);
4315
4316 if (act) {
4317 bool was_ignored = k->sa.sa_handler == SIG_IGN;
4318
4319 sigdelsetmask(&act->sa.sa_mask,
4320 sigmask(SIGKILL) | sigmask(SIGSTOP));
4321 *k = *act;
4322 /*
4323 * POSIX 3.3.1.3:
4324 * "Setting a signal action to SIG_IGN for a signal that is
4325 * pending shall cause the pending signal to be discarded,
4326 * whether or not it is blocked."
4327 *
4328 * "Setting a signal action to SIG_DFL for a signal that is
4329 * pending and whose default action is to ignore the signal
4330 * (for example, SIGCHLD), shall cause the pending signal to
4331 * be discarded, whether or not it is blocked"
4332 */
4333 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4334 sigemptyset(&mask);
4335 sigaddset(&mask, sig);
4336 flush_sigqueue_mask(p, &mask, &p->signal->shared_pending);
4337 for_each_thread(p, t)
4338 flush_sigqueue_mask(p, &mask, &t->pending);
4339 } else if (was_ignored) {
4340 posixtimer_sig_unignore(p, sig);
4341 }
4342 }
4343
4344 spin_unlock_irq(&p->sighand->siglock);
4345 return 0;
4346}
4347
4348#ifdef CONFIG_DYNAMIC_SIGFRAME
4349static inline void sigaltstack_lock(void)
4350 __acquires(¤t->sighand->siglock)
4351{
4352 spin_lock_irq(¤t->sighand->siglock);
4353}
4354
4355static inline void sigaltstack_unlock(void)
4356 __releases(¤t->sighand->siglock)
4357{
4358 spin_unlock_irq(¤t->sighand->siglock);
4359}
4360#else
4361static inline void sigaltstack_lock(void) { }
4362static inline void sigaltstack_unlock(void) { }
4363#endif
4364
4365static int
4366do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4367 size_t min_ss_size)
4368{
4369 struct task_struct *t = current;
4370 int ret = 0;
4371
4372 if (oss) {
4373 memset(oss, 0, sizeof(stack_t));
4374 oss->ss_sp = (void __user *) t->sas_ss_sp;
4375 oss->ss_size = t->sas_ss_size;
4376 oss->ss_flags = sas_ss_flags(sp) |
4377 (current->sas_ss_flags & SS_FLAG_BITS);
4378 }
4379
4380 if (ss) {
4381 void __user *ss_sp = ss->ss_sp;
4382 size_t ss_size = ss->ss_size;
4383 unsigned ss_flags = ss->ss_flags;
4384 int ss_mode;
4385
4386 if (unlikely(on_sig_stack(sp)))
4387 return -EPERM;
4388
4389 ss_mode = ss_flags & ~SS_FLAG_BITS;
4390 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4391 ss_mode != 0))
4392 return -EINVAL;
4393
4394 /*
4395 * Return before taking any locks if no actual
4396 * sigaltstack changes were requested.
4397 */
4398 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4399 t->sas_ss_size == ss_size &&
4400 t->sas_ss_flags == ss_flags)
4401 return 0;
4402
4403 sigaltstack_lock();
4404 if (ss_mode == SS_DISABLE) {
4405 ss_size = 0;
4406 ss_sp = NULL;
4407 } else {
4408 if (unlikely(ss_size < min_ss_size))
4409 ret = -ENOMEM;
4410 if (!sigaltstack_size_valid(ss_size))
4411 ret = -ENOMEM;
4412 }
4413 if (!ret) {
4414 t->sas_ss_sp = (unsigned long) ss_sp;
4415 t->sas_ss_size = ss_size;
4416 t->sas_ss_flags = ss_flags;
4417 }
4418 sigaltstack_unlock();
4419 }
4420 return ret;
4421}
4422
4423SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4424{
4425 stack_t new, old;
4426 int err;
4427 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4428 return -EFAULT;
4429 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4430 current_user_stack_pointer(),
4431 MINSIGSTKSZ);
4432 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4433 err = -EFAULT;
4434 return err;
4435}
4436
4437int restore_altstack(const stack_t __user *uss)
4438{
4439 stack_t new;
4440 if (copy_from_user(&new, uss, sizeof(stack_t)))
4441 return -EFAULT;
4442 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4443 MINSIGSTKSZ);
4444 /* squash all but EFAULT for now */
4445 return 0;
4446}
4447
4448int __save_altstack(stack_t __user *uss, unsigned long sp)
4449{
4450 struct task_struct *t = current;
4451 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4452 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4453 __put_user(t->sas_ss_size, &uss->ss_size);
4454 return err;
4455}
4456
4457#ifdef CONFIG_COMPAT
4458static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4459 compat_stack_t __user *uoss_ptr)
4460{
4461 stack_t uss, uoss;
4462 int ret;
4463
4464 if (uss_ptr) {
4465 compat_stack_t uss32;
4466 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4467 return -EFAULT;
4468 uss.ss_sp = compat_ptr(uss32.ss_sp);
4469 uss.ss_flags = uss32.ss_flags;
4470 uss.ss_size = uss32.ss_size;
4471 }
4472 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4473 compat_user_stack_pointer(),
4474 COMPAT_MINSIGSTKSZ);
4475 if (ret >= 0 && uoss_ptr) {
4476 compat_stack_t old;
4477 memset(&old, 0, sizeof(old));
4478 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4479 old.ss_flags = uoss.ss_flags;
4480 old.ss_size = uoss.ss_size;
4481 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4482 ret = -EFAULT;
4483 }
4484 return ret;
4485}
4486
4487COMPAT_SYSCALL_DEFINE2(sigaltstack,
4488 const compat_stack_t __user *, uss_ptr,
4489 compat_stack_t __user *, uoss_ptr)
4490{
4491 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4492}
4493
4494int compat_restore_altstack(const compat_stack_t __user *uss)
4495{
4496 int err = do_compat_sigaltstack(uss, NULL);
4497 /* squash all but -EFAULT for now */
4498 return err == -EFAULT ? err : 0;
4499}
4500
4501int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4502{
4503 int err;
4504 struct task_struct *t = current;
4505 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4506 &uss->ss_sp) |
4507 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4508 __put_user(t->sas_ss_size, &uss->ss_size);
4509 return err;
4510}
4511#endif
4512
4513#ifdef __ARCH_WANT_SYS_SIGPENDING
4514
4515/**
4516 * sys_sigpending - examine pending signals
4517 * @uset: where mask of pending signal is returned
4518 */
4519SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4520{
4521 sigset_t set;
4522
4523 if (sizeof(old_sigset_t) > sizeof(*uset))
4524 return -EINVAL;
4525
4526 do_sigpending(&set);
4527
4528 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4529 return -EFAULT;
4530
4531 return 0;
4532}
4533
4534#ifdef CONFIG_COMPAT
4535COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4536{
4537 sigset_t set;
4538
4539 do_sigpending(&set);
4540
4541 return put_user(set.sig[0], set32);
4542}
4543#endif
4544
4545#endif
4546
4547#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4548/**
4549 * sys_sigprocmask - examine and change blocked signals
4550 * @how: whether to add, remove, or set signals
4551 * @nset: signals to add or remove (if non-null)
4552 * @oset: previous value of signal mask if non-null
4553 *
4554 * Some platforms have their own version with special arguments;
4555 * others support only sys_rt_sigprocmask.
4556 */
4557
4558SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4559 old_sigset_t __user *, oset)
4560{
4561 old_sigset_t old_set, new_set;
4562 sigset_t new_blocked;
4563
4564 old_set = current->blocked.sig[0];
4565
4566 if (nset) {
4567 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4568 return -EFAULT;
4569
4570 new_blocked = current->blocked;
4571
4572 switch (how) {
4573 case SIG_BLOCK:
4574 sigaddsetmask(&new_blocked, new_set);
4575 break;
4576 case SIG_UNBLOCK:
4577 sigdelsetmask(&new_blocked, new_set);
4578 break;
4579 case SIG_SETMASK:
4580 new_blocked.sig[0] = new_set;
4581 break;
4582 default:
4583 return -EINVAL;
4584 }
4585
4586 set_current_blocked(&new_blocked);
4587 }
4588
4589 if (oset) {
4590 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4591 return -EFAULT;
4592 }
4593
4594 return 0;
4595}
4596#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4597
4598#ifndef CONFIG_ODD_RT_SIGACTION
4599/**
4600 * sys_rt_sigaction - alter an action taken by a process
4601 * @sig: signal to be sent
4602 * @act: new sigaction
4603 * @oact: used to save the previous sigaction
4604 * @sigsetsize: size of sigset_t type
4605 */
4606SYSCALL_DEFINE4(rt_sigaction, int, sig,
4607 const struct sigaction __user *, act,
4608 struct sigaction __user *, oact,
4609 size_t, sigsetsize)
4610{
4611 struct k_sigaction new_sa, old_sa;
4612 int ret;
4613
4614 /* XXX: Don't preclude handling different sized sigset_t's. */
4615 if (sigsetsize != sizeof(sigset_t))
4616 return -EINVAL;
4617
4618 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4619 return -EFAULT;
4620
4621 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4622 if (ret)
4623 return ret;
4624
4625 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4626 return -EFAULT;
4627
4628 return 0;
4629}
4630#ifdef CONFIG_COMPAT
4631COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4632 const struct compat_sigaction __user *, act,
4633 struct compat_sigaction __user *, oact,
4634 compat_size_t, sigsetsize)
4635{
4636 struct k_sigaction new_ka, old_ka;
4637#ifdef __ARCH_HAS_SA_RESTORER
4638 compat_uptr_t restorer;
4639#endif
4640 int ret;
4641
4642 /* XXX: Don't preclude handling different sized sigset_t's. */
4643 if (sigsetsize != sizeof(compat_sigset_t))
4644 return -EINVAL;
4645
4646 if (act) {
4647 compat_uptr_t handler;
4648 ret = get_user(handler, &act->sa_handler);
4649 new_ka.sa.sa_handler = compat_ptr(handler);
4650#ifdef __ARCH_HAS_SA_RESTORER
4651 ret |= get_user(restorer, &act->sa_restorer);
4652 new_ka.sa.sa_restorer = compat_ptr(restorer);
4653#endif
4654 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4655 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4656 if (ret)
4657 return -EFAULT;
4658 }
4659
4660 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4661 if (!ret && oact) {
4662 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4663 &oact->sa_handler);
4664 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4665 sizeof(oact->sa_mask));
4666 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4667#ifdef __ARCH_HAS_SA_RESTORER
4668 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4669 &oact->sa_restorer);
4670#endif
4671 }
4672 return ret;
4673}
4674#endif
4675#endif /* !CONFIG_ODD_RT_SIGACTION */
4676
4677#ifdef CONFIG_OLD_SIGACTION
4678SYSCALL_DEFINE3(sigaction, int, sig,
4679 const struct old_sigaction __user *, act,
4680 struct old_sigaction __user *, oact)
4681{
4682 struct k_sigaction new_ka, old_ka;
4683 int ret;
4684
4685 if (act) {
4686 old_sigset_t mask;
4687 if (!access_ok(act, sizeof(*act)) ||
4688 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4689 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4690 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4691 __get_user(mask, &act->sa_mask))
4692 return -EFAULT;
4693#ifdef __ARCH_HAS_KA_RESTORER
4694 new_ka.ka_restorer = NULL;
4695#endif
4696 siginitset(&new_ka.sa.sa_mask, mask);
4697 }
4698
4699 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4700
4701 if (!ret && oact) {
4702 if (!access_ok(oact, sizeof(*oact)) ||
4703 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4704 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4705 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4706 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4707 return -EFAULT;
4708 }
4709
4710 return ret;
4711}
4712#endif
4713#ifdef CONFIG_COMPAT_OLD_SIGACTION
4714COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4715 const struct compat_old_sigaction __user *, act,
4716 struct compat_old_sigaction __user *, oact)
4717{
4718 struct k_sigaction new_ka, old_ka;
4719 int ret;
4720 compat_old_sigset_t mask;
4721 compat_uptr_t handler, restorer;
4722
4723 if (act) {
4724 if (!access_ok(act, sizeof(*act)) ||
4725 __get_user(handler, &act->sa_handler) ||
4726 __get_user(restorer, &act->sa_restorer) ||
4727 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4728 __get_user(mask, &act->sa_mask))
4729 return -EFAULT;
4730
4731#ifdef __ARCH_HAS_KA_RESTORER
4732 new_ka.ka_restorer = NULL;
4733#endif
4734 new_ka.sa.sa_handler = compat_ptr(handler);
4735 new_ka.sa.sa_restorer = compat_ptr(restorer);
4736 siginitset(&new_ka.sa.sa_mask, mask);
4737 }
4738
4739 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4740
4741 if (!ret && oact) {
4742 if (!access_ok(oact, sizeof(*oact)) ||
4743 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4744 &oact->sa_handler) ||
4745 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4746 &oact->sa_restorer) ||
4747 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4748 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4749 return -EFAULT;
4750 }
4751 return ret;
4752}
4753#endif
4754
4755#ifdef CONFIG_SGETMASK_SYSCALL
4756
4757/*
4758 * For backwards compatibility. Functionality superseded by sigprocmask.
4759 */
4760SYSCALL_DEFINE0(sgetmask)
4761{
4762 /* SMP safe */
4763 return current->blocked.sig[0];
4764}
4765
4766SYSCALL_DEFINE1(ssetmask, int, newmask)
4767{
4768 int old = current->blocked.sig[0];
4769 sigset_t newset;
4770
4771 siginitset(&newset, newmask);
4772 set_current_blocked(&newset);
4773
4774 return old;
4775}
4776#endif /* CONFIG_SGETMASK_SYSCALL */
4777
4778#ifdef __ARCH_WANT_SYS_SIGNAL
4779/*
4780 * For backwards compatibility. Functionality superseded by sigaction.
4781 */
4782SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4783{
4784 struct k_sigaction new_sa, old_sa;
4785 int ret;
4786
4787 new_sa.sa.sa_handler = handler;
4788 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4789 sigemptyset(&new_sa.sa.sa_mask);
4790
4791 ret = do_sigaction(sig, &new_sa, &old_sa);
4792
4793 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4794}
4795#endif /* __ARCH_WANT_SYS_SIGNAL */
4796
4797#ifdef __ARCH_WANT_SYS_PAUSE
4798
4799SYSCALL_DEFINE0(pause)
4800{
4801 while (!signal_pending(current)) {
4802 __set_current_state(TASK_INTERRUPTIBLE);
4803 schedule();
4804 }
4805 return -ERESTARTNOHAND;
4806}
4807
4808#endif
4809
4810static int sigsuspend(sigset_t *set)
4811{
4812 current->saved_sigmask = current->blocked;
4813 set_current_blocked(set);
4814
4815 while (!signal_pending(current)) {
4816 __set_current_state(TASK_INTERRUPTIBLE);
4817 schedule();
4818 }
4819 set_restore_sigmask();
4820 return -ERESTARTNOHAND;
4821}
4822
4823/**
4824 * sys_rt_sigsuspend - replace the signal mask for a value with the
4825 * @unewset value until a signal is received
4826 * @unewset: new signal mask value
4827 * @sigsetsize: size of sigset_t type
4828 */
4829SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4830{
4831 sigset_t newset;
4832
4833 /* XXX: Don't preclude handling different sized sigset_t's. */
4834 if (sigsetsize != sizeof(sigset_t))
4835 return -EINVAL;
4836
4837 if (copy_from_user(&newset, unewset, sizeof(newset)))
4838 return -EFAULT;
4839 return sigsuspend(&newset);
4840}
4841
4842#ifdef CONFIG_COMPAT
4843COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4844{
4845 sigset_t newset;
4846
4847 /* XXX: Don't preclude handling different sized sigset_t's. */
4848 if (sigsetsize != sizeof(sigset_t))
4849 return -EINVAL;
4850
4851 if (get_compat_sigset(&newset, unewset))
4852 return -EFAULT;
4853 return sigsuspend(&newset);
4854}
4855#endif
4856
4857#ifdef CONFIG_OLD_SIGSUSPEND
4858SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4859{
4860 sigset_t blocked;
4861 siginitset(&blocked, mask);
4862 return sigsuspend(&blocked);
4863}
4864#endif
4865#ifdef CONFIG_OLD_SIGSUSPEND3
4866SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4867{
4868 sigset_t blocked;
4869 siginitset(&blocked, mask);
4870 return sigsuspend(&blocked);
4871}
4872#endif
4873
4874__weak const char *arch_vma_name(struct vm_area_struct *vma)
4875{
4876 return NULL;
4877}
4878
4879static inline void siginfo_buildtime_checks(void)
4880{
4881 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4882
4883 /* Verify the offsets in the two siginfos match */
4884#define CHECK_OFFSET(field) \
4885 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4886
4887 /* kill */
4888 CHECK_OFFSET(si_pid);
4889 CHECK_OFFSET(si_uid);
4890
4891 /* timer */
4892 CHECK_OFFSET(si_tid);
4893 CHECK_OFFSET(si_overrun);
4894 CHECK_OFFSET(si_value);
4895
4896 /* rt */
4897 CHECK_OFFSET(si_pid);
4898 CHECK_OFFSET(si_uid);
4899 CHECK_OFFSET(si_value);
4900
4901 /* sigchld */
4902 CHECK_OFFSET(si_pid);
4903 CHECK_OFFSET(si_uid);
4904 CHECK_OFFSET(si_status);
4905 CHECK_OFFSET(si_utime);
4906 CHECK_OFFSET(si_stime);
4907
4908 /* sigfault */
4909 CHECK_OFFSET(si_addr);
4910 CHECK_OFFSET(si_trapno);
4911 CHECK_OFFSET(si_addr_lsb);
4912 CHECK_OFFSET(si_lower);
4913 CHECK_OFFSET(si_upper);
4914 CHECK_OFFSET(si_pkey);
4915 CHECK_OFFSET(si_perf_data);
4916 CHECK_OFFSET(si_perf_type);
4917 CHECK_OFFSET(si_perf_flags);
4918
4919 /* sigpoll */
4920 CHECK_OFFSET(si_band);
4921 CHECK_OFFSET(si_fd);
4922
4923 /* sigsys */
4924 CHECK_OFFSET(si_call_addr);
4925 CHECK_OFFSET(si_syscall);
4926 CHECK_OFFSET(si_arch);
4927#undef CHECK_OFFSET
4928
4929 /* usb asyncio */
4930 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4931 offsetof(struct siginfo, si_addr));
4932 if (sizeof(int) == sizeof(void __user *)) {
4933 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4934 sizeof(void __user *));
4935 } else {
4936 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4937 sizeof_field(struct siginfo, si_uid)) !=
4938 sizeof(void __user *));
4939 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4940 offsetof(struct siginfo, si_uid));
4941 }
4942#ifdef CONFIG_COMPAT
4943 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4944 offsetof(struct compat_siginfo, si_addr));
4945 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4946 sizeof(compat_uptr_t));
4947 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4948 sizeof_field(struct siginfo, si_pid));
4949#endif
4950}
4951
4952#if defined(CONFIG_SYSCTL)
4953static struct ctl_table signal_debug_table[] = {
4954#ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4955 {
4956 .procname = "exception-trace",
4957 .data = &show_unhandled_signals,
4958 .maxlen = sizeof(int),
4959 .mode = 0644,
4960 .proc_handler = proc_dointvec
4961 },
4962#endif
4963};
4964
4965static int __init init_signal_sysctls(void)
4966{
4967 register_sysctl_init("debug", signal_debug_table);
4968 return 0;
4969}
4970early_initcall(init_signal_sysctls);
4971#endif /* CONFIG_SYSCTL */
4972
4973void __init signals_init(void)
4974{
4975 siginfo_buildtime_checks();
4976
4977 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4978}
4979
4980#ifdef CONFIG_KGDB_KDB
4981#include <linux/kdb.h>
4982/*
4983 * kdb_send_sig - Allows kdb to send signals without exposing
4984 * signal internals. This function checks if the required locks are
4985 * available before calling the main signal code, to avoid kdb
4986 * deadlocks.
4987 */
4988void kdb_send_sig(struct task_struct *t, int sig)
4989{
4990 static struct task_struct *kdb_prev_t;
4991 int new_t, ret;
4992 if (!spin_trylock(&t->sighand->siglock)) {
4993 kdb_printf("Can't do kill command now.\n"
4994 "The sigmask lock is held somewhere else in "
4995 "kernel, try again later\n");
4996 return;
4997 }
4998 new_t = kdb_prev_t != t;
4999 kdb_prev_t = t;
5000 if (!task_is_running(t) && new_t) {
5001 spin_unlock(&t->sighand->siglock);
5002 kdb_printf("Process is not RUNNING, sending a signal from "
5003 "kdb risks deadlock\n"
5004 "on the run queue locks. "
5005 "The signal has _not_ been sent.\n"
5006 "Reissue the kill command if you want to risk "
5007 "the deadlock.\n");
5008 return;
5009 }
5010 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
5011 spin_unlock(&t->sighand->siglock);
5012 if (ret)
5013 kdb_printf("Fail to deliver Signal %d to process %d.\n",
5014 sig, t->pid);
5015 else
5016 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
5017}
5018#endif /* CONFIG_KGDB_KDB */
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/signal.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 *
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
12 */
13
14#include <linux/slab.h>
15#include <linux/export.h>
16#include <linux/init.h>
17#include <linux/sched/mm.h>
18#include <linux/sched/user.h>
19#include <linux/sched/debug.h>
20#include <linux/sched/task.h>
21#include <linux/sched/task_stack.h>
22#include <linux/sched/cputime.h>
23#include <linux/file.h>
24#include <linux/fs.h>
25#include <linux/mm.h>
26#include <linux/proc_fs.h>
27#include <linux/tty.h>
28#include <linux/binfmts.h>
29#include <linux/coredump.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
32#include <linux/ptrace.h>
33#include <linux/signal.h>
34#include <linux/signalfd.h>
35#include <linux/ratelimit.h>
36#include <linux/task_work.h>
37#include <linux/capability.h>
38#include <linux/freezer.h>
39#include <linux/pid_namespace.h>
40#include <linux/nsproxy.h>
41#include <linux/user_namespace.h>
42#include <linux/uprobes.h>
43#include <linux/compat.h>
44#include <linux/cn_proc.h>
45#include <linux/compiler.h>
46#include <linux/posix-timers.h>
47#include <linux/cgroup.h>
48#include <linux/audit.h>
49#include <linux/sysctl.h>
50
51#define CREATE_TRACE_POINTS
52#include <trace/events/signal.h>
53
54#include <asm/param.h>
55#include <linux/uaccess.h>
56#include <asm/unistd.h>
57#include <asm/siginfo.h>
58#include <asm/cacheflush.h>
59#include <asm/syscall.h> /* for syscall_get_* */
60
61/*
62 * SLAB caches for signal bits.
63 */
64
65static struct kmem_cache *sigqueue_cachep;
66
67int print_fatal_signals __read_mostly;
68
69static void __user *sig_handler(struct task_struct *t, int sig)
70{
71 return t->sighand->action[sig - 1].sa.sa_handler;
72}
73
74static inline bool sig_handler_ignored(void __user *handler, int sig)
75{
76 /* Is it explicitly or implicitly ignored? */
77 return handler == SIG_IGN ||
78 (handler == SIG_DFL && sig_kernel_ignore(sig));
79}
80
81static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
82{
83 void __user *handler;
84
85 handler = sig_handler(t, sig);
86
87 /* SIGKILL and SIGSTOP may not be sent to the global init */
88 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
89 return true;
90
91 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
92 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
93 return true;
94
95 /* Only allow kernel generated signals to this kthread */
96 if (unlikely((t->flags & PF_KTHREAD) &&
97 (handler == SIG_KTHREAD_KERNEL) && !force))
98 return true;
99
100 return sig_handler_ignored(handler, sig);
101}
102
103static bool sig_ignored(struct task_struct *t, int sig, bool force)
104{
105 /*
106 * Blocked signals are never ignored, since the
107 * signal handler may change by the time it is
108 * unblocked.
109 */
110 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
111 return false;
112
113 /*
114 * Tracers may want to know about even ignored signal unless it
115 * is SIGKILL which can't be reported anyway but can be ignored
116 * by SIGNAL_UNKILLABLE task.
117 */
118 if (t->ptrace && sig != SIGKILL)
119 return false;
120
121 return sig_task_ignored(t, sig, force);
122}
123
124/*
125 * Re-calculate pending state from the set of locally pending
126 * signals, globally pending signals, and blocked signals.
127 */
128static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
129{
130 unsigned long ready;
131 long i;
132
133 switch (_NSIG_WORDS) {
134 default:
135 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
136 ready |= signal->sig[i] &~ blocked->sig[i];
137 break;
138
139 case 4: ready = signal->sig[3] &~ blocked->sig[3];
140 ready |= signal->sig[2] &~ blocked->sig[2];
141 ready |= signal->sig[1] &~ blocked->sig[1];
142 ready |= signal->sig[0] &~ blocked->sig[0];
143 break;
144
145 case 2: ready = signal->sig[1] &~ blocked->sig[1];
146 ready |= signal->sig[0] &~ blocked->sig[0];
147 break;
148
149 case 1: ready = signal->sig[0] &~ blocked->sig[0];
150 }
151 return ready != 0;
152}
153
154#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
155
156static bool recalc_sigpending_tsk(struct task_struct *t)
157{
158 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
159 PENDING(&t->pending, &t->blocked) ||
160 PENDING(&t->signal->shared_pending, &t->blocked) ||
161 cgroup_task_frozen(t)) {
162 set_tsk_thread_flag(t, TIF_SIGPENDING);
163 return true;
164 }
165
166 /*
167 * We must never clear the flag in another thread, or in current
168 * when it's possible the current syscall is returning -ERESTART*.
169 * So we don't clear it here, and only callers who know they should do.
170 */
171 return false;
172}
173
174void recalc_sigpending(void)
175{
176 if (!recalc_sigpending_tsk(current) && !freezing(current))
177 clear_thread_flag(TIF_SIGPENDING);
178
179}
180EXPORT_SYMBOL(recalc_sigpending);
181
182void calculate_sigpending(void)
183{
184 /* Have any signals or users of TIF_SIGPENDING been delayed
185 * until after fork?
186 */
187 spin_lock_irq(¤t->sighand->siglock);
188 set_tsk_thread_flag(current, TIF_SIGPENDING);
189 recalc_sigpending();
190 spin_unlock_irq(¤t->sighand->siglock);
191}
192
193/* Given the mask, find the first available signal that should be serviced. */
194
195#define SYNCHRONOUS_MASK \
196 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
197 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
198
199int next_signal(struct sigpending *pending, sigset_t *mask)
200{
201 unsigned long i, *s, *m, x;
202 int sig = 0;
203
204 s = pending->signal.sig;
205 m = mask->sig;
206
207 /*
208 * Handle the first word specially: it contains the
209 * synchronous signals that need to be dequeued first.
210 */
211 x = *s &~ *m;
212 if (x) {
213 if (x & SYNCHRONOUS_MASK)
214 x &= SYNCHRONOUS_MASK;
215 sig = ffz(~x) + 1;
216 return sig;
217 }
218
219 switch (_NSIG_WORDS) {
220 default:
221 for (i = 1; i < _NSIG_WORDS; ++i) {
222 x = *++s &~ *++m;
223 if (!x)
224 continue;
225 sig = ffz(~x) + i*_NSIG_BPW + 1;
226 break;
227 }
228 break;
229
230 case 2:
231 x = s[1] &~ m[1];
232 if (!x)
233 break;
234 sig = ffz(~x) + _NSIG_BPW + 1;
235 break;
236
237 case 1:
238 /* Nothing to do */
239 break;
240 }
241
242 return sig;
243}
244
245static inline void print_dropped_signal(int sig)
246{
247 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
248
249 if (!print_fatal_signals)
250 return;
251
252 if (!__ratelimit(&ratelimit_state))
253 return;
254
255 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
256 current->comm, current->pid, sig);
257}
258
259/**
260 * task_set_jobctl_pending - set jobctl pending bits
261 * @task: target task
262 * @mask: pending bits to set
263 *
264 * Clear @mask from @task->jobctl. @mask must be subset of
265 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
266 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
267 * cleared. If @task is already being killed or exiting, this function
268 * becomes noop.
269 *
270 * CONTEXT:
271 * Must be called with @task->sighand->siglock held.
272 *
273 * RETURNS:
274 * %true if @mask is set, %false if made noop because @task was dying.
275 */
276bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
277{
278 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
279 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
280 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
281
282 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
283 return false;
284
285 if (mask & JOBCTL_STOP_SIGMASK)
286 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
287
288 task->jobctl |= mask;
289 return true;
290}
291
292/**
293 * task_clear_jobctl_trapping - clear jobctl trapping bit
294 * @task: target task
295 *
296 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
297 * Clear it and wake up the ptracer. Note that we don't need any further
298 * locking. @task->siglock guarantees that @task->parent points to the
299 * ptracer.
300 *
301 * CONTEXT:
302 * Must be called with @task->sighand->siglock held.
303 */
304void task_clear_jobctl_trapping(struct task_struct *task)
305{
306 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
307 task->jobctl &= ~JOBCTL_TRAPPING;
308 smp_mb(); /* advised by wake_up_bit() */
309 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
310 }
311}
312
313/**
314 * task_clear_jobctl_pending - clear jobctl pending bits
315 * @task: target task
316 * @mask: pending bits to clear
317 *
318 * Clear @mask from @task->jobctl. @mask must be subset of
319 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
320 * STOP bits are cleared together.
321 *
322 * If clearing of @mask leaves no stop or trap pending, this function calls
323 * task_clear_jobctl_trapping().
324 *
325 * CONTEXT:
326 * Must be called with @task->sighand->siglock held.
327 */
328void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
329{
330 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
331
332 if (mask & JOBCTL_STOP_PENDING)
333 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
334
335 task->jobctl &= ~mask;
336
337 if (!(task->jobctl & JOBCTL_PENDING_MASK))
338 task_clear_jobctl_trapping(task);
339}
340
341/**
342 * task_participate_group_stop - participate in a group stop
343 * @task: task participating in a group stop
344 *
345 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
346 * Group stop states are cleared and the group stop count is consumed if
347 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
348 * stop, the appropriate `SIGNAL_*` flags are set.
349 *
350 * CONTEXT:
351 * Must be called with @task->sighand->siglock held.
352 *
353 * RETURNS:
354 * %true if group stop completion should be notified to the parent, %false
355 * otherwise.
356 */
357static bool task_participate_group_stop(struct task_struct *task)
358{
359 struct signal_struct *sig = task->signal;
360 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
361
362 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
363
364 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
365
366 if (!consume)
367 return false;
368
369 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
370 sig->group_stop_count--;
371
372 /*
373 * Tell the caller to notify completion iff we are entering into a
374 * fresh group stop. Read comment in do_signal_stop() for details.
375 */
376 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
377 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
378 return true;
379 }
380 return false;
381}
382
383void task_join_group_stop(struct task_struct *task)
384{
385 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
386 struct signal_struct *sig = current->signal;
387
388 if (sig->group_stop_count) {
389 sig->group_stop_count++;
390 mask |= JOBCTL_STOP_CONSUME;
391 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
392 return;
393
394 /* Have the new thread join an on-going signal group stop */
395 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
396}
397
398/*
399 * allocate a new signal queue record
400 * - this may be called without locks if and only if t == current, otherwise an
401 * appropriate lock must be held to stop the target task from exiting
402 */
403static struct sigqueue *
404__sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
405 int override_rlimit, const unsigned int sigqueue_flags)
406{
407 struct sigqueue *q = NULL;
408 struct ucounts *ucounts;
409 long sigpending;
410
411 /*
412 * Protect access to @t credentials. This can go away when all
413 * callers hold rcu read lock.
414 *
415 * NOTE! A pending signal will hold on to the user refcount,
416 * and we get/put the refcount only when the sigpending count
417 * changes from/to zero.
418 */
419 rcu_read_lock();
420 ucounts = task_ucounts(t);
421 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
422 rcu_read_unlock();
423 if (!sigpending)
424 return NULL;
425
426 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
427 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
428 } else {
429 print_dropped_signal(sig);
430 }
431
432 if (unlikely(q == NULL)) {
433 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
434 } else {
435 INIT_LIST_HEAD(&q->list);
436 q->flags = sigqueue_flags;
437 q->ucounts = ucounts;
438 }
439 return q;
440}
441
442static void __sigqueue_free(struct sigqueue *q)
443{
444 if (q->flags & SIGQUEUE_PREALLOC)
445 return;
446 if (q->ucounts) {
447 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
448 q->ucounts = NULL;
449 }
450 kmem_cache_free(sigqueue_cachep, q);
451}
452
453void flush_sigqueue(struct sigpending *queue)
454{
455 struct sigqueue *q;
456
457 sigemptyset(&queue->signal);
458 while (!list_empty(&queue->list)) {
459 q = list_entry(queue->list.next, struct sigqueue , list);
460 list_del_init(&q->list);
461 __sigqueue_free(q);
462 }
463}
464
465/*
466 * Flush all pending signals for this kthread.
467 */
468void flush_signals(struct task_struct *t)
469{
470 unsigned long flags;
471
472 spin_lock_irqsave(&t->sighand->siglock, flags);
473 clear_tsk_thread_flag(t, TIF_SIGPENDING);
474 flush_sigqueue(&t->pending);
475 flush_sigqueue(&t->signal->shared_pending);
476 spin_unlock_irqrestore(&t->sighand->siglock, flags);
477}
478EXPORT_SYMBOL(flush_signals);
479
480#ifdef CONFIG_POSIX_TIMERS
481static void __flush_itimer_signals(struct sigpending *pending)
482{
483 sigset_t signal, retain;
484 struct sigqueue *q, *n;
485
486 signal = pending->signal;
487 sigemptyset(&retain);
488
489 list_for_each_entry_safe(q, n, &pending->list, list) {
490 int sig = q->info.si_signo;
491
492 if (likely(q->info.si_code != SI_TIMER)) {
493 sigaddset(&retain, sig);
494 } else {
495 sigdelset(&signal, sig);
496 list_del_init(&q->list);
497 __sigqueue_free(q);
498 }
499 }
500
501 sigorsets(&pending->signal, &signal, &retain);
502}
503
504void flush_itimer_signals(void)
505{
506 struct task_struct *tsk = current;
507 unsigned long flags;
508
509 spin_lock_irqsave(&tsk->sighand->siglock, flags);
510 __flush_itimer_signals(&tsk->pending);
511 __flush_itimer_signals(&tsk->signal->shared_pending);
512 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
513}
514#endif
515
516void ignore_signals(struct task_struct *t)
517{
518 int i;
519
520 for (i = 0; i < _NSIG; ++i)
521 t->sighand->action[i].sa.sa_handler = SIG_IGN;
522
523 flush_signals(t);
524}
525
526/*
527 * Flush all handlers for a task.
528 */
529
530void
531flush_signal_handlers(struct task_struct *t, int force_default)
532{
533 int i;
534 struct k_sigaction *ka = &t->sighand->action[0];
535 for (i = _NSIG ; i != 0 ; i--) {
536 if (force_default || ka->sa.sa_handler != SIG_IGN)
537 ka->sa.sa_handler = SIG_DFL;
538 ka->sa.sa_flags = 0;
539#ifdef __ARCH_HAS_SA_RESTORER
540 ka->sa.sa_restorer = NULL;
541#endif
542 sigemptyset(&ka->sa.sa_mask);
543 ka++;
544 }
545}
546
547bool unhandled_signal(struct task_struct *tsk, int sig)
548{
549 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
550 if (is_global_init(tsk))
551 return true;
552
553 if (handler != SIG_IGN && handler != SIG_DFL)
554 return false;
555
556 /* If dying, we handle all new signals by ignoring them */
557 if (fatal_signal_pending(tsk))
558 return false;
559
560 /* if ptraced, let the tracer determine */
561 return !tsk->ptrace;
562}
563
564static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
565 bool *resched_timer)
566{
567 struct sigqueue *q, *first = NULL;
568
569 /*
570 * Collect the siginfo appropriate to this signal. Check if
571 * there is another siginfo for the same signal.
572 */
573 list_for_each_entry(q, &list->list, list) {
574 if (q->info.si_signo == sig) {
575 if (first)
576 goto still_pending;
577 first = q;
578 }
579 }
580
581 sigdelset(&list->signal, sig);
582
583 if (first) {
584still_pending:
585 list_del_init(&first->list);
586 copy_siginfo(info, &first->info);
587
588 *resched_timer =
589 (first->flags & SIGQUEUE_PREALLOC) &&
590 (info->si_code == SI_TIMER) &&
591 (info->si_sys_private);
592
593 __sigqueue_free(first);
594 } else {
595 /*
596 * Ok, it wasn't in the queue. This must be
597 * a fast-pathed signal or we must have been
598 * out of queue space. So zero out the info.
599 */
600 clear_siginfo(info);
601 info->si_signo = sig;
602 info->si_errno = 0;
603 info->si_code = SI_USER;
604 info->si_pid = 0;
605 info->si_uid = 0;
606 }
607}
608
609static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
610 kernel_siginfo_t *info, bool *resched_timer)
611{
612 int sig = next_signal(pending, mask);
613
614 if (sig)
615 collect_signal(sig, pending, info, resched_timer);
616 return sig;
617}
618
619/*
620 * Dequeue a signal and return the element to the caller, which is
621 * expected to free it.
622 *
623 * All callers have to hold the siglock.
624 */
625int dequeue_signal(struct task_struct *tsk, sigset_t *mask,
626 kernel_siginfo_t *info, enum pid_type *type)
627{
628 bool resched_timer = false;
629 int signr;
630
631 /* We only dequeue private signals from ourselves, we don't let
632 * signalfd steal them
633 */
634 *type = PIDTYPE_PID;
635 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
636 if (!signr) {
637 *type = PIDTYPE_TGID;
638 signr = __dequeue_signal(&tsk->signal->shared_pending,
639 mask, info, &resched_timer);
640#ifdef CONFIG_POSIX_TIMERS
641 /*
642 * itimer signal ?
643 *
644 * itimers are process shared and we restart periodic
645 * itimers in the signal delivery path to prevent DoS
646 * attacks in the high resolution timer case. This is
647 * compliant with the old way of self-restarting
648 * itimers, as the SIGALRM is a legacy signal and only
649 * queued once. Changing the restart behaviour to
650 * restart the timer in the signal dequeue path is
651 * reducing the timer noise on heavy loaded !highres
652 * systems too.
653 */
654 if (unlikely(signr == SIGALRM)) {
655 struct hrtimer *tmr = &tsk->signal->real_timer;
656
657 if (!hrtimer_is_queued(tmr) &&
658 tsk->signal->it_real_incr != 0) {
659 hrtimer_forward(tmr, tmr->base->get_time(),
660 tsk->signal->it_real_incr);
661 hrtimer_restart(tmr);
662 }
663 }
664#endif
665 }
666
667 recalc_sigpending();
668 if (!signr)
669 return 0;
670
671 if (unlikely(sig_kernel_stop(signr))) {
672 /*
673 * Set a marker that we have dequeued a stop signal. Our
674 * caller might release the siglock and then the pending
675 * stop signal it is about to process is no longer in the
676 * pending bitmasks, but must still be cleared by a SIGCONT
677 * (and overruled by a SIGKILL). So those cases clear this
678 * shared flag after we've set it. Note that this flag may
679 * remain set after the signal we return is ignored or
680 * handled. That doesn't matter because its only purpose
681 * is to alert stop-signal processing code when another
682 * processor has come along and cleared the flag.
683 */
684 current->jobctl |= JOBCTL_STOP_DEQUEUED;
685 }
686#ifdef CONFIG_POSIX_TIMERS
687 if (resched_timer) {
688 /*
689 * Release the siglock to ensure proper locking order
690 * of timer locks outside of siglocks. Note, we leave
691 * irqs disabled here, since the posix-timers code is
692 * about to disable them again anyway.
693 */
694 spin_unlock(&tsk->sighand->siglock);
695 posixtimer_rearm(info);
696 spin_lock(&tsk->sighand->siglock);
697
698 /* Don't expose the si_sys_private value to userspace */
699 info->si_sys_private = 0;
700 }
701#endif
702 return signr;
703}
704EXPORT_SYMBOL_GPL(dequeue_signal);
705
706static int dequeue_synchronous_signal(kernel_siginfo_t *info)
707{
708 struct task_struct *tsk = current;
709 struct sigpending *pending = &tsk->pending;
710 struct sigqueue *q, *sync = NULL;
711
712 /*
713 * Might a synchronous signal be in the queue?
714 */
715 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
716 return 0;
717
718 /*
719 * Return the first synchronous signal in the queue.
720 */
721 list_for_each_entry(q, &pending->list, list) {
722 /* Synchronous signals have a positive si_code */
723 if ((q->info.si_code > SI_USER) &&
724 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
725 sync = q;
726 goto next;
727 }
728 }
729 return 0;
730next:
731 /*
732 * Check if there is another siginfo for the same signal.
733 */
734 list_for_each_entry_continue(q, &pending->list, list) {
735 if (q->info.si_signo == sync->info.si_signo)
736 goto still_pending;
737 }
738
739 sigdelset(&pending->signal, sync->info.si_signo);
740 recalc_sigpending();
741still_pending:
742 list_del_init(&sync->list);
743 copy_siginfo(info, &sync->info);
744 __sigqueue_free(sync);
745 return info->si_signo;
746}
747
748/*
749 * Tell a process that it has a new active signal..
750 *
751 * NOTE! we rely on the previous spin_lock to
752 * lock interrupts for us! We can only be called with
753 * "siglock" held, and the local interrupt must
754 * have been disabled when that got acquired!
755 *
756 * No need to set need_resched since signal event passing
757 * goes through ->blocked
758 */
759void signal_wake_up_state(struct task_struct *t, unsigned int state)
760{
761 lockdep_assert_held(&t->sighand->siglock);
762
763 set_tsk_thread_flag(t, TIF_SIGPENDING);
764
765 /*
766 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
767 * case. We don't check t->state here because there is a race with it
768 * executing another processor and just now entering stopped state.
769 * By using wake_up_state, we ensure the process will wake up and
770 * handle its death signal.
771 */
772 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
773 kick_process(t);
774}
775
776/*
777 * Remove signals in mask from the pending set and queue.
778 * Returns 1 if any signals were found.
779 *
780 * All callers must be holding the siglock.
781 */
782static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
783{
784 struct sigqueue *q, *n;
785 sigset_t m;
786
787 sigandsets(&m, mask, &s->signal);
788 if (sigisemptyset(&m))
789 return;
790
791 sigandnsets(&s->signal, &s->signal, mask);
792 list_for_each_entry_safe(q, n, &s->list, list) {
793 if (sigismember(mask, q->info.si_signo)) {
794 list_del_init(&q->list);
795 __sigqueue_free(q);
796 }
797 }
798}
799
800static inline int is_si_special(const struct kernel_siginfo *info)
801{
802 return info <= SEND_SIG_PRIV;
803}
804
805static inline bool si_fromuser(const struct kernel_siginfo *info)
806{
807 return info == SEND_SIG_NOINFO ||
808 (!is_si_special(info) && SI_FROMUSER(info));
809}
810
811/*
812 * called with RCU read lock from check_kill_permission()
813 */
814static bool kill_ok_by_cred(struct task_struct *t)
815{
816 const struct cred *cred = current_cred();
817 const struct cred *tcred = __task_cred(t);
818
819 return uid_eq(cred->euid, tcred->suid) ||
820 uid_eq(cred->euid, tcred->uid) ||
821 uid_eq(cred->uid, tcred->suid) ||
822 uid_eq(cred->uid, tcred->uid) ||
823 ns_capable(tcred->user_ns, CAP_KILL);
824}
825
826/*
827 * Bad permissions for sending the signal
828 * - the caller must hold the RCU read lock
829 */
830static int check_kill_permission(int sig, struct kernel_siginfo *info,
831 struct task_struct *t)
832{
833 struct pid *sid;
834 int error;
835
836 if (!valid_signal(sig))
837 return -EINVAL;
838
839 if (!si_fromuser(info))
840 return 0;
841
842 error = audit_signal_info(sig, t); /* Let audit system see the signal */
843 if (error)
844 return error;
845
846 if (!same_thread_group(current, t) &&
847 !kill_ok_by_cred(t)) {
848 switch (sig) {
849 case SIGCONT:
850 sid = task_session(t);
851 /*
852 * We don't return the error if sid == NULL. The
853 * task was unhashed, the caller must notice this.
854 */
855 if (!sid || sid == task_session(current))
856 break;
857 fallthrough;
858 default:
859 return -EPERM;
860 }
861 }
862
863 return security_task_kill(t, info, sig, NULL);
864}
865
866/**
867 * ptrace_trap_notify - schedule trap to notify ptracer
868 * @t: tracee wanting to notify tracer
869 *
870 * This function schedules sticky ptrace trap which is cleared on the next
871 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
872 * ptracer.
873 *
874 * If @t is running, STOP trap will be taken. If trapped for STOP and
875 * ptracer is listening for events, tracee is woken up so that it can
876 * re-trap for the new event. If trapped otherwise, STOP trap will be
877 * eventually taken without returning to userland after the existing traps
878 * are finished by PTRACE_CONT.
879 *
880 * CONTEXT:
881 * Must be called with @task->sighand->siglock held.
882 */
883static void ptrace_trap_notify(struct task_struct *t)
884{
885 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
886 lockdep_assert_held(&t->sighand->siglock);
887
888 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
889 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
890}
891
892/*
893 * Handle magic process-wide effects of stop/continue signals. Unlike
894 * the signal actions, these happen immediately at signal-generation
895 * time regardless of blocking, ignoring, or handling. This does the
896 * actual continuing for SIGCONT, but not the actual stopping for stop
897 * signals. The process stop is done as a signal action for SIG_DFL.
898 *
899 * Returns true if the signal should be actually delivered, otherwise
900 * it should be dropped.
901 */
902static bool prepare_signal(int sig, struct task_struct *p, bool force)
903{
904 struct signal_struct *signal = p->signal;
905 struct task_struct *t;
906 sigset_t flush;
907
908 if (signal->flags & SIGNAL_GROUP_EXIT) {
909 if (signal->core_state)
910 return sig == SIGKILL;
911 /*
912 * The process is in the middle of dying, drop the signal.
913 */
914 return false;
915 } else if (sig_kernel_stop(sig)) {
916 /*
917 * This is a stop signal. Remove SIGCONT from all queues.
918 */
919 siginitset(&flush, sigmask(SIGCONT));
920 flush_sigqueue_mask(&flush, &signal->shared_pending);
921 for_each_thread(p, t)
922 flush_sigqueue_mask(&flush, &t->pending);
923 } else if (sig == SIGCONT) {
924 unsigned int why;
925 /*
926 * Remove all stop signals from all queues, wake all threads.
927 */
928 siginitset(&flush, SIG_KERNEL_STOP_MASK);
929 flush_sigqueue_mask(&flush, &signal->shared_pending);
930 for_each_thread(p, t) {
931 flush_sigqueue_mask(&flush, &t->pending);
932 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
933 if (likely(!(t->ptrace & PT_SEIZED))) {
934 t->jobctl &= ~JOBCTL_STOPPED;
935 wake_up_state(t, __TASK_STOPPED);
936 } else
937 ptrace_trap_notify(t);
938 }
939
940 /*
941 * Notify the parent with CLD_CONTINUED if we were stopped.
942 *
943 * If we were in the middle of a group stop, we pretend it
944 * was already finished, and then continued. Since SIGCHLD
945 * doesn't queue we report only CLD_STOPPED, as if the next
946 * CLD_CONTINUED was dropped.
947 */
948 why = 0;
949 if (signal->flags & SIGNAL_STOP_STOPPED)
950 why |= SIGNAL_CLD_CONTINUED;
951 else if (signal->group_stop_count)
952 why |= SIGNAL_CLD_STOPPED;
953
954 if (why) {
955 /*
956 * The first thread which returns from do_signal_stop()
957 * will take ->siglock, notice SIGNAL_CLD_MASK, and
958 * notify its parent. See get_signal().
959 */
960 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
961 signal->group_stop_count = 0;
962 signal->group_exit_code = 0;
963 }
964 }
965
966 return !sig_ignored(p, sig, force);
967}
968
969/*
970 * Test if P wants to take SIG. After we've checked all threads with this,
971 * it's equivalent to finding no threads not blocking SIG. Any threads not
972 * blocking SIG were ruled out because they are not running and already
973 * have pending signals. Such threads will dequeue from the shared queue
974 * as soon as they're available, so putting the signal on the shared queue
975 * will be equivalent to sending it to one such thread.
976 */
977static inline bool wants_signal(int sig, struct task_struct *p)
978{
979 if (sigismember(&p->blocked, sig))
980 return false;
981
982 if (p->flags & PF_EXITING)
983 return false;
984
985 if (sig == SIGKILL)
986 return true;
987
988 if (task_is_stopped_or_traced(p))
989 return false;
990
991 return task_curr(p) || !task_sigpending(p);
992}
993
994static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
995{
996 struct signal_struct *signal = p->signal;
997 struct task_struct *t;
998
999 /*
1000 * Now find a thread we can wake up to take the signal off the queue.
1001 *
1002 * Try the suggested task first (may or may not be the main thread).
1003 */
1004 if (wants_signal(sig, p))
1005 t = p;
1006 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1007 /*
1008 * There is just one thread and it does not need to be woken.
1009 * It will dequeue unblocked signals before it runs again.
1010 */
1011 return;
1012 else {
1013 /*
1014 * Otherwise try to find a suitable thread.
1015 */
1016 t = signal->curr_target;
1017 while (!wants_signal(sig, t)) {
1018 t = next_thread(t);
1019 if (t == signal->curr_target)
1020 /*
1021 * No thread needs to be woken.
1022 * Any eligible threads will see
1023 * the signal in the queue soon.
1024 */
1025 return;
1026 }
1027 signal->curr_target = t;
1028 }
1029
1030 /*
1031 * Found a killable thread. If the signal will be fatal,
1032 * then start taking the whole group down immediately.
1033 */
1034 if (sig_fatal(p, sig) &&
1035 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1036 !sigismember(&t->real_blocked, sig) &&
1037 (sig == SIGKILL || !p->ptrace)) {
1038 /*
1039 * This signal will be fatal to the whole group.
1040 */
1041 if (!sig_kernel_coredump(sig)) {
1042 /*
1043 * Start a group exit and wake everybody up.
1044 * This way we don't have other threads
1045 * running and doing things after a slower
1046 * thread has the fatal signal pending.
1047 */
1048 signal->flags = SIGNAL_GROUP_EXIT;
1049 signal->group_exit_code = sig;
1050 signal->group_stop_count = 0;
1051 __for_each_thread(signal, t) {
1052 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1053 sigaddset(&t->pending.signal, SIGKILL);
1054 signal_wake_up(t, 1);
1055 }
1056 return;
1057 }
1058 }
1059
1060 /*
1061 * The signal is already in the shared-pending queue.
1062 * Tell the chosen thread to wake up and dequeue it.
1063 */
1064 signal_wake_up(t, sig == SIGKILL);
1065 return;
1066}
1067
1068static inline bool legacy_queue(struct sigpending *signals, int sig)
1069{
1070 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1071}
1072
1073static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1074 struct task_struct *t, enum pid_type type, bool force)
1075{
1076 struct sigpending *pending;
1077 struct sigqueue *q;
1078 int override_rlimit;
1079 int ret = 0, result;
1080
1081 lockdep_assert_held(&t->sighand->siglock);
1082
1083 result = TRACE_SIGNAL_IGNORED;
1084 if (!prepare_signal(sig, t, force))
1085 goto ret;
1086
1087 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1088 /*
1089 * Short-circuit ignored signals and support queuing
1090 * exactly one non-rt signal, so that we can get more
1091 * detailed information about the cause of the signal.
1092 */
1093 result = TRACE_SIGNAL_ALREADY_PENDING;
1094 if (legacy_queue(pending, sig))
1095 goto ret;
1096
1097 result = TRACE_SIGNAL_DELIVERED;
1098 /*
1099 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1100 */
1101 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1102 goto out_set;
1103
1104 /*
1105 * Real-time signals must be queued if sent by sigqueue, or
1106 * some other real-time mechanism. It is implementation
1107 * defined whether kill() does so. We attempt to do so, on
1108 * the principle of least surprise, but since kill is not
1109 * allowed to fail with EAGAIN when low on memory we just
1110 * make sure at least one signal gets delivered and don't
1111 * pass on the info struct.
1112 */
1113 if (sig < SIGRTMIN)
1114 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1115 else
1116 override_rlimit = 0;
1117
1118 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1119
1120 if (q) {
1121 list_add_tail(&q->list, &pending->list);
1122 switch ((unsigned long) info) {
1123 case (unsigned long) SEND_SIG_NOINFO:
1124 clear_siginfo(&q->info);
1125 q->info.si_signo = sig;
1126 q->info.si_errno = 0;
1127 q->info.si_code = SI_USER;
1128 q->info.si_pid = task_tgid_nr_ns(current,
1129 task_active_pid_ns(t));
1130 rcu_read_lock();
1131 q->info.si_uid =
1132 from_kuid_munged(task_cred_xxx(t, user_ns),
1133 current_uid());
1134 rcu_read_unlock();
1135 break;
1136 case (unsigned long) SEND_SIG_PRIV:
1137 clear_siginfo(&q->info);
1138 q->info.si_signo = sig;
1139 q->info.si_errno = 0;
1140 q->info.si_code = SI_KERNEL;
1141 q->info.si_pid = 0;
1142 q->info.si_uid = 0;
1143 break;
1144 default:
1145 copy_siginfo(&q->info, info);
1146 break;
1147 }
1148 } else if (!is_si_special(info) &&
1149 sig >= SIGRTMIN && info->si_code != SI_USER) {
1150 /*
1151 * Queue overflow, abort. We may abort if the
1152 * signal was rt and sent by user using something
1153 * other than kill().
1154 */
1155 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1156 ret = -EAGAIN;
1157 goto ret;
1158 } else {
1159 /*
1160 * This is a silent loss of information. We still
1161 * send the signal, but the *info bits are lost.
1162 */
1163 result = TRACE_SIGNAL_LOSE_INFO;
1164 }
1165
1166out_set:
1167 signalfd_notify(t, sig);
1168 sigaddset(&pending->signal, sig);
1169
1170 /* Let multiprocess signals appear after on-going forks */
1171 if (type > PIDTYPE_TGID) {
1172 struct multiprocess_signals *delayed;
1173 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1174 sigset_t *signal = &delayed->signal;
1175 /* Can't queue both a stop and a continue signal */
1176 if (sig == SIGCONT)
1177 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1178 else if (sig_kernel_stop(sig))
1179 sigdelset(signal, SIGCONT);
1180 sigaddset(signal, sig);
1181 }
1182 }
1183
1184 complete_signal(sig, t, type);
1185ret:
1186 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1187 return ret;
1188}
1189
1190static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1191{
1192 bool ret = false;
1193 switch (siginfo_layout(info->si_signo, info->si_code)) {
1194 case SIL_KILL:
1195 case SIL_CHLD:
1196 case SIL_RT:
1197 ret = true;
1198 break;
1199 case SIL_TIMER:
1200 case SIL_POLL:
1201 case SIL_FAULT:
1202 case SIL_FAULT_TRAPNO:
1203 case SIL_FAULT_MCEERR:
1204 case SIL_FAULT_BNDERR:
1205 case SIL_FAULT_PKUERR:
1206 case SIL_FAULT_PERF_EVENT:
1207 case SIL_SYS:
1208 ret = false;
1209 break;
1210 }
1211 return ret;
1212}
1213
1214int send_signal_locked(int sig, struct kernel_siginfo *info,
1215 struct task_struct *t, enum pid_type type)
1216{
1217 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1218 bool force = false;
1219
1220 if (info == SEND_SIG_NOINFO) {
1221 /* Force if sent from an ancestor pid namespace */
1222 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1223 } else if (info == SEND_SIG_PRIV) {
1224 /* Don't ignore kernel generated signals */
1225 force = true;
1226 } else if (has_si_pid_and_uid(info)) {
1227 /* SIGKILL and SIGSTOP is special or has ids */
1228 struct user_namespace *t_user_ns;
1229
1230 rcu_read_lock();
1231 t_user_ns = task_cred_xxx(t, user_ns);
1232 if (current_user_ns() != t_user_ns) {
1233 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1234 info->si_uid = from_kuid_munged(t_user_ns, uid);
1235 }
1236 rcu_read_unlock();
1237
1238 /* A kernel generated signal? */
1239 force = (info->si_code == SI_KERNEL);
1240
1241 /* From an ancestor pid namespace? */
1242 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1243 info->si_pid = 0;
1244 force = true;
1245 }
1246 }
1247 return __send_signal_locked(sig, info, t, type, force);
1248}
1249
1250static void print_fatal_signal(int signr)
1251{
1252 struct pt_regs *regs = task_pt_regs(current);
1253 struct file *exe_file;
1254
1255 exe_file = get_task_exe_file(current);
1256 if (exe_file) {
1257 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1258 exe_file, current->comm, signr);
1259 fput(exe_file);
1260 } else {
1261 pr_info("%s: potentially unexpected fatal signal %d.\n",
1262 current->comm, signr);
1263 }
1264
1265#if defined(__i386__) && !defined(__arch_um__)
1266 pr_info("code at %08lx: ", regs->ip);
1267 {
1268 int i;
1269 for (i = 0; i < 16; i++) {
1270 unsigned char insn;
1271
1272 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1273 break;
1274 pr_cont("%02x ", insn);
1275 }
1276 }
1277 pr_cont("\n");
1278#endif
1279 preempt_disable();
1280 show_regs(regs);
1281 preempt_enable();
1282}
1283
1284static int __init setup_print_fatal_signals(char *str)
1285{
1286 get_option (&str, &print_fatal_signals);
1287
1288 return 1;
1289}
1290
1291__setup("print-fatal-signals=", setup_print_fatal_signals);
1292
1293int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1294 enum pid_type type)
1295{
1296 unsigned long flags;
1297 int ret = -ESRCH;
1298
1299 if (lock_task_sighand(p, &flags)) {
1300 ret = send_signal_locked(sig, info, p, type);
1301 unlock_task_sighand(p, &flags);
1302 }
1303
1304 return ret;
1305}
1306
1307enum sig_handler {
1308 HANDLER_CURRENT, /* If reachable use the current handler */
1309 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1310 HANDLER_EXIT, /* Only visible as the process exit code */
1311};
1312
1313/*
1314 * Force a signal that the process can't ignore: if necessary
1315 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1316 *
1317 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1318 * since we do not want to have a signal handler that was blocked
1319 * be invoked when user space had explicitly blocked it.
1320 *
1321 * We don't want to have recursive SIGSEGV's etc, for example,
1322 * that is why we also clear SIGNAL_UNKILLABLE.
1323 */
1324static int
1325force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1326 enum sig_handler handler)
1327{
1328 unsigned long int flags;
1329 int ret, blocked, ignored;
1330 struct k_sigaction *action;
1331 int sig = info->si_signo;
1332
1333 spin_lock_irqsave(&t->sighand->siglock, flags);
1334 action = &t->sighand->action[sig-1];
1335 ignored = action->sa.sa_handler == SIG_IGN;
1336 blocked = sigismember(&t->blocked, sig);
1337 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1338 action->sa.sa_handler = SIG_DFL;
1339 if (handler == HANDLER_EXIT)
1340 action->sa.sa_flags |= SA_IMMUTABLE;
1341 if (blocked)
1342 sigdelset(&t->blocked, sig);
1343 }
1344 /*
1345 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1346 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1347 */
1348 if (action->sa.sa_handler == SIG_DFL &&
1349 (!t->ptrace || (handler == HANDLER_EXIT)))
1350 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1351 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1352 /* This can happen if the signal was already pending and blocked */
1353 if (!task_sigpending(t))
1354 signal_wake_up(t, 0);
1355 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1356
1357 return ret;
1358}
1359
1360int force_sig_info(struct kernel_siginfo *info)
1361{
1362 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1363}
1364
1365/*
1366 * Nuke all other threads in the group.
1367 */
1368int zap_other_threads(struct task_struct *p)
1369{
1370 struct task_struct *t;
1371 int count = 0;
1372
1373 p->signal->group_stop_count = 0;
1374
1375 for_other_threads(p, t) {
1376 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1377 /* Don't require de_thread to wait for the vhost_worker */
1378 if ((t->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)
1379 count++;
1380
1381 /* Don't bother with already dead threads */
1382 if (t->exit_state)
1383 continue;
1384 sigaddset(&t->pending.signal, SIGKILL);
1385 signal_wake_up(t, 1);
1386 }
1387
1388 return count;
1389}
1390
1391struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1392 unsigned long *flags)
1393{
1394 struct sighand_struct *sighand;
1395
1396 rcu_read_lock();
1397 for (;;) {
1398 sighand = rcu_dereference(tsk->sighand);
1399 if (unlikely(sighand == NULL))
1400 break;
1401
1402 /*
1403 * This sighand can be already freed and even reused, but
1404 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1405 * initializes ->siglock: this slab can't go away, it has
1406 * the same object type, ->siglock can't be reinitialized.
1407 *
1408 * We need to ensure that tsk->sighand is still the same
1409 * after we take the lock, we can race with de_thread() or
1410 * __exit_signal(). In the latter case the next iteration
1411 * must see ->sighand == NULL.
1412 */
1413 spin_lock_irqsave(&sighand->siglock, *flags);
1414 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1415 break;
1416 spin_unlock_irqrestore(&sighand->siglock, *flags);
1417 }
1418 rcu_read_unlock();
1419
1420 return sighand;
1421}
1422
1423#ifdef CONFIG_LOCKDEP
1424void lockdep_assert_task_sighand_held(struct task_struct *task)
1425{
1426 struct sighand_struct *sighand;
1427
1428 rcu_read_lock();
1429 sighand = rcu_dereference(task->sighand);
1430 if (sighand)
1431 lockdep_assert_held(&sighand->siglock);
1432 else
1433 WARN_ON_ONCE(1);
1434 rcu_read_unlock();
1435}
1436#endif
1437
1438/*
1439 * send signal info to all the members of a group
1440 */
1441int group_send_sig_info(int sig, struct kernel_siginfo *info,
1442 struct task_struct *p, enum pid_type type)
1443{
1444 int ret;
1445
1446 rcu_read_lock();
1447 ret = check_kill_permission(sig, info, p);
1448 rcu_read_unlock();
1449
1450 if (!ret && sig)
1451 ret = do_send_sig_info(sig, info, p, type);
1452
1453 return ret;
1454}
1455
1456/*
1457 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1458 * control characters do (^C, ^Z etc)
1459 * - the caller must hold at least a readlock on tasklist_lock
1460 */
1461int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1462{
1463 struct task_struct *p = NULL;
1464 int ret = -ESRCH;
1465
1466 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1467 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1468 /*
1469 * If group_send_sig_info() succeeds at least once ret
1470 * becomes 0 and after that the code below has no effect.
1471 * Otherwise we return the last err or -ESRCH if this
1472 * process group is empty.
1473 */
1474 if (ret)
1475 ret = err;
1476 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1477
1478 return ret;
1479}
1480
1481int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1482{
1483 int error = -ESRCH;
1484 struct task_struct *p;
1485
1486 for (;;) {
1487 rcu_read_lock();
1488 p = pid_task(pid, PIDTYPE_PID);
1489 if (p)
1490 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1491 rcu_read_unlock();
1492 if (likely(!p || error != -ESRCH))
1493 return error;
1494
1495 /*
1496 * The task was unhashed in between, try again. If it
1497 * is dead, pid_task() will return NULL, if we race with
1498 * de_thread() it will find the new leader.
1499 */
1500 }
1501}
1502
1503static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1504{
1505 int error;
1506 rcu_read_lock();
1507 error = kill_pid_info(sig, info, find_vpid(pid));
1508 rcu_read_unlock();
1509 return error;
1510}
1511
1512static inline bool kill_as_cred_perm(const struct cred *cred,
1513 struct task_struct *target)
1514{
1515 const struct cred *pcred = __task_cred(target);
1516
1517 return uid_eq(cred->euid, pcred->suid) ||
1518 uid_eq(cred->euid, pcred->uid) ||
1519 uid_eq(cred->uid, pcred->suid) ||
1520 uid_eq(cred->uid, pcred->uid);
1521}
1522
1523/*
1524 * The usb asyncio usage of siginfo is wrong. The glibc support
1525 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1526 * AKA after the generic fields:
1527 * kernel_pid_t si_pid;
1528 * kernel_uid32_t si_uid;
1529 * sigval_t si_value;
1530 *
1531 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1532 * after the generic fields is:
1533 * void __user *si_addr;
1534 *
1535 * This is a practical problem when there is a 64bit big endian kernel
1536 * and a 32bit userspace. As the 32bit address will encoded in the low
1537 * 32bits of the pointer. Those low 32bits will be stored at higher
1538 * address than appear in a 32 bit pointer. So userspace will not
1539 * see the address it was expecting for it's completions.
1540 *
1541 * There is nothing in the encoding that can allow
1542 * copy_siginfo_to_user32 to detect this confusion of formats, so
1543 * handle this by requiring the caller of kill_pid_usb_asyncio to
1544 * notice when this situration takes place and to store the 32bit
1545 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1546 * parameter.
1547 */
1548int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1549 struct pid *pid, const struct cred *cred)
1550{
1551 struct kernel_siginfo info;
1552 struct task_struct *p;
1553 unsigned long flags;
1554 int ret = -EINVAL;
1555
1556 if (!valid_signal(sig))
1557 return ret;
1558
1559 clear_siginfo(&info);
1560 info.si_signo = sig;
1561 info.si_errno = errno;
1562 info.si_code = SI_ASYNCIO;
1563 *((sigval_t *)&info.si_pid) = addr;
1564
1565 rcu_read_lock();
1566 p = pid_task(pid, PIDTYPE_PID);
1567 if (!p) {
1568 ret = -ESRCH;
1569 goto out_unlock;
1570 }
1571 if (!kill_as_cred_perm(cred, p)) {
1572 ret = -EPERM;
1573 goto out_unlock;
1574 }
1575 ret = security_task_kill(p, &info, sig, cred);
1576 if (ret)
1577 goto out_unlock;
1578
1579 if (sig) {
1580 if (lock_task_sighand(p, &flags)) {
1581 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1582 unlock_task_sighand(p, &flags);
1583 } else
1584 ret = -ESRCH;
1585 }
1586out_unlock:
1587 rcu_read_unlock();
1588 return ret;
1589}
1590EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1591
1592/*
1593 * kill_something_info() interprets pid in interesting ways just like kill(2).
1594 *
1595 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1596 * is probably wrong. Should make it like BSD or SYSV.
1597 */
1598
1599static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1600{
1601 int ret;
1602
1603 if (pid > 0)
1604 return kill_proc_info(sig, info, pid);
1605
1606 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1607 if (pid == INT_MIN)
1608 return -ESRCH;
1609
1610 read_lock(&tasklist_lock);
1611 if (pid != -1) {
1612 ret = __kill_pgrp_info(sig, info,
1613 pid ? find_vpid(-pid) : task_pgrp(current));
1614 } else {
1615 int retval = 0, count = 0;
1616 struct task_struct * p;
1617
1618 for_each_process(p) {
1619 if (task_pid_vnr(p) > 1 &&
1620 !same_thread_group(p, current)) {
1621 int err = group_send_sig_info(sig, info, p,
1622 PIDTYPE_MAX);
1623 ++count;
1624 if (err != -EPERM)
1625 retval = err;
1626 }
1627 }
1628 ret = count ? retval : -ESRCH;
1629 }
1630 read_unlock(&tasklist_lock);
1631
1632 return ret;
1633}
1634
1635/*
1636 * These are for backward compatibility with the rest of the kernel source.
1637 */
1638
1639int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1640{
1641 /*
1642 * Make sure legacy kernel users don't send in bad values
1643 * (normal paths check this in check_kill_permission).
1644 */
1645 if (!valid_signal(sig))
1646 return -EINVAL;
1647
1648 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1649}
1650EXPORT_SYMBOL(send_sig_info);
1651
1652#define __si_special(priv) \
1653 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1654
1655int
1656send_sig(int sig, struct task_struct *p, int priv)
1657{
1658 return send_sig_info(sig, __si_special(priv), p);
1659}
1660EXPORT_SYMBOL(send_sig);
1661
1662void force_sig(int sig)
1663{
1664 struct kernel_siginfo info;
1665
1666 clear_siginfo(&info);
1667 info.si_signo = sig;
1668 info.si_errno = 0;
1669 info.si_code = SI_KERNEL;
1670 info.si_pid = 0;
1671 info.si_uid = 0;
1672 force_sig_info(&info);
1673}
1674EXPORT_SYMBOL(force_sig);
1675
1676void force_fatal_sig(int sig)
1677{
1678 struct kernel_siginfo info;
1679
1680 clear_siginfo(&info);
1681 info.si_signo = sig;
1682 info.si_errno = 0;
1683 info.si_code = SI_KERNEL;
1684 info.si_pid = 0;
1685 info.si_uid = 0;
1686 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1687}
1688
1689void force_exit_sig(int sig)
1690{
1691 struct kernel_siginfo info;
1692
1693 clear_siginfo(&info);
1694 info.si_signo = sig;
1695 info.si_errno = 0;
1696 info.si_code = SI_KERNEL;
1697 info.si_pid = 0;
1698 info.si_uid = 0;
1699 force_sig_info_to_task(&info, current, HANDLER_EXIT);
1700}
1701
1702/*
1703 * When things go south during signal handling, we
1704 * will force a SIGSEGV. And if the signal that caused
1705 * the problem was already a SIGSEGV, we'll want to
1706 * make sure we don't even try to deliver the signal..
1707 */
1708void force_sigsegv(int sig)
1709{
1710 if (sig == SIGSEGV)
1711 force_fatal_sig(SIGSEGV);
1712 else
1713 force_sig(SIGSEGV);
1714}
1715
1716int force_sig_fault_to_task(int sig, int code, void __user *addr,
1717 struct task_struct *t)
1718{
1719 struct kernel_siginfo info;
1720
1721 clear_siginfo(&info);
1722 info.si_signo = sig;
1723 info.si_errno = 0;
1724 info.si_code = code;
1725 info.si_addr = addr;
1726 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1727}
1728
1729int force_sig_fault(int sig, int code, void __user *addr)
1730{
1731 return force_sig_fault_to_task(sig, code, addr, current);
1732}
1733
1734int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t)
1735{
1736 struct kernel_siginfo info;
1737
1738 clear_siginfo(&info);
1739 info.si_signo = sig;
1740 info.si_errno = 0;
1741 info.si_code = code;
1742 info.si_addr = addr;
1743 return send_sig_info(info.si_signo, &info, t);
1744}
1745
1746int force_sig_mceerr(int code, void __user *addr, short lsb)
1747{
1748 struct kernel_siginfo info;
1749
1750 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1751 clear_siginfo(&info);
1752 info.si_signo = SIGBUS;
1753 info.si_errno = 0;
1754 info.si_code = code;
1755 info.si_addr = addr;
1756 info.si_addr_lsb = lsb;
1757 return force_sig_info(&info);
1758}
1759
1760int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1761{
1762 struct kernel_siginfo info;
1763
1764 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1765 clear_siginfo(&info);
1766 info.si_signo = SIGBUS;
1767 info.si_errno = 0;
1768 info.si_code = code;
1769 info.si_addr = addr;
1770 info.si_addr_lsb = lsb;
1771 return send_sig_info(info.si_signo, &info, t);
1772}
1773EXPORT_SYMBOL(send_sig_mceerr);
1774
1775int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1776{
1777 struct kernel_siginfo info;
1778
1779 clear_siginfo(&info);
1780 info.si_signo = SIGSEGV;
1781 info.si_errno = 0;
1782 info.si_code = SEGV_BNDERR;
1783 info.si_addr = addr;
1784 info.si_lower = lower;
1785 info.si_upper = upper;
1786 return force_sig_info(&info);
1787}
1788
1789#ifdef SEGV_PKUERR
1790int force_sig_pkuerr(void __user *addr, u32 pkey)
1791{
1792 struct kernel_siginfo info;
1793
1794 clear_siginfo(&info);
1795 info.si_signo = SIGSEGV;
1796 info.si_errno = 0;
1797 info.si_code = SEGV_PKUERR;
1798 info.si_addr = addr;
1799 info.si_pkey = pkey;
1800 return force_sig_info(&info);
1801}
1802#endif
1803
1804int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1805{
1806 struct kernel_siginfo info;
1807
1808 clear_siginfo(&info);
1809 info.si_signo = SIGTRAP;
1810 info.si_errno = 0;
1811 info.si_code = TRAP_PERF;
1812 info.si_addr = addr;
1813 info.si_perf_data = sig_data;
1814 info.si_perf_type = type;
1815
1816 /*
1817 * Signals generated by perf events should not terminate the whole
1818 * process if SIGTRAP is blocked, however, delivering the signal
1819 * asynchronously is better than not delivering at all. But tell user
1820 * space if the signal was asynchronous, so it can clearly be
1821 * distinguished from normal synchronous ones.
1822 */
1823 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1824 TRAP_PERF_FLAG_ASYNC :
1825 0;
1826
1827 return send_sig_info(info.si_signo, &info, current);
1828}
1829
1830/**
1831 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1832 * @syscall: syscall number to send to userland
1833 * @reason: filter-supplied reason code to send to userland (via si_errno)
1834 * @force_coredump: true to trigger a coredump
1835 *
1836 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1837 */
1838int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1839{
1840 struct kernel_siginfo info;
1841
1842 clear_siginfo(&info);
1843 info.si_signo = SIGSYS;
1844 info.si_code = SYS_SECCOMP;
1845 info.si_call_addr = (void __user *)KSTK_EIP(current);
1846 info.si_errno = reason;
1847 info.si_arch = syscall_get_arch(current);
1848 info.si_syscall = syscall;
1849 return force_sig_info_to_task(&info, current,
1850 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1851}
1852
1853/* For the crazy architectures that include trap information in
1854 * the errno field, instead of an actual errno value.
1855 */
1856int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1857{
1858 struct kernel_siginfo info;
1859
1860 clear_siginfo(&info);
1861 info.si_signo = SIGTRAP;
1862 info.si_errno = errno;
1863 info.si_code = TRAP_HWBKPT;
1864 info.si_addr = addr;
1865 return force_sig_info(&info);
1866}
1867
1868/* For the rare architectures that include trap information using
1869 * si_trapno.
1870 */
1871int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1872{
1873 struct kernel_siginfo info;
1874
1875 clear_siginfo(&info);
1876 info.si_signo = sig;
1877 info.si_errno = 0;
1878 info.si_code = code;
1879 info.si_addr = addr;
1880 info.si_trapno = trapno;
1881 return force_sig_info(&info);
1882}
1883
1884/* For the rare architectures that include trap information using
1885 * si_trapno.
1886 */
1887int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1888 struct task_struct *t)
1889{
1890 struct kernel_siginfo info;
1891
1892 clear_siginfo(&info);
1893 info.si_signo = sig;
1894 info.si_errno = 0;
1895 info.si_code = code;
1896 info.si_addr = addr;
1897 info.si_trapno = trapno;
1898 return send_sig_info(info.si_signo, &info, t);
1899}
1900
1901int kill_pgrp(struct pid *pid, int sig, int priv)
1902{
1903 int ret;
1904
1905 read_lock(&tasklist_lock);
1906 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1907 read_unlock(&tasklist_lock);
1908
1909 return ret;
1910}
1911EXPORT_SYMBOL(kill_pgrp);
1912
1913int kill_pid(struct pid *pid, int sig, int priv)
1914{
1915 return kill_pid_info(sig, __si_special(priv), pid);
1916}
1917EXPORT_SYMBOL(kill_pid);
1918
1919/*
1920 * These functions support sending signals using preallocated sigqueue
1921 * structures. This is needed "because realtime applications cannot
1922 * afford to lose notifications of asynchronous events, like timer
1923 * expirations or I/O completions". In the case of POSIX Timers
1924 * we allocate the sigqueue structure from the timer_create. If this
1925 * allocation fails we are able to report the failure to the application
1926 * with an EAGAIN error.
1927 */
1928struct sigqueue *sigqueue_alloc(void)
1929{
1930 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1931}
1932
1933void sigqueue_free(struct sigqueue *q)
1934{
1935 unsigned long flags;
1936 spinlock_t *lock = ¤t->sighand->siglock;
1937
1938 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1939 /*
1940 * We must hold ->siglock while testing q->list
1941 * to serialize with collect_signal() or with
1942 * __exit_signal()->flush_sigqueue().
1943 */
1944 spin_lock_irqsave(lock, flags);
1945 q->flags &= ~SIGQUEUE_PREALLOC;
1946 /*
1947 * If it is queued it will be freed when dequeued,
1948 * like the "regular" sigqueue.
1949 */
1950 if (!list_empty(&q->list))
1951 q = NULL;
1952 spin_unlock_irqrestore(lock, flags);
1953
1954 if (q)
1955 __sigqueue_free(q);
1956}
1957
1958int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1959{
1960 int sig = q->info.si_signo;
1961 struct sigpending *pending;
1962 struct task_struct *t;
1963 unsigned long flags;
1964 int ret, result;
1965
1966 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1967
1968 ret = -1;
1969 rcu_read_lock();
1970
1971 /*
1972 * This function is used by POSIX timers to deliver a timer signal.
1973 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1974 * set), the signal must be delivered to the specific thread (queues
1975 * into t->pending).
1976 *
1977 * Where type is not PIDTYPE_PID, signals must be delivered to the
1978 * process. In this case, prefer to deliver to current if it is in
1979 * the same thread group as the target process, which avoids
1980 * unnecessarily waking up a potentially idle task.
1981 */
1982 t = pid_task(pid, type);
1983 if (!t)
1984 goto ret;
1985 if (type != PIDTYPE_PID && same_thread_group(t, current))
1986 t = current;
1987 if (!likely(lock_task_sighand(t, &flags)))
1988 goto ret;
1989
1990 ret = 1; /* the signal is ignored */
1991 result = TRACE_SIGNAL_IGNORED;
1992 if (!prepare_signal(sig, t, false))
1993 goto out;
1994
1995 ret = 0;
1996 if (unlikely(!list_empty(&q->list))) {
1997 /*
1998 * If an SI_TIMER entry is already queue just increment
1999 * the overrun count.
2000 */
2001 BUG_ON(q->info.si_code != SI_TIMER);
2002 q->info.si_overrun++;
2003 result = TRACE_SIGNAL_ALREADY_PENDING;
2004 goto out;
2005 }
2006 q->info.si_overrun = 0;
2007
2008 signalfd_notify(t, sig);
2009 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
2010 list_add_tail(&q->list, &pending->list);
2011 sigaddset(&pending->signal, sig);
2012 complete_signal(sig, t, type);
2013 result = TRACE_SIGNAL_DELIVERED;
2014out:
2015 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
2016 unlock_task_sighand(t, &flags);
2017ret:
2018 rcu_read_unlock();
2019 return ret;
2020}
2021
2022static void do_notify_pidfd(struct task_struct *task)
2023{
2024 struct pid *pid;
2025
2026 WARN_ON(task->exit_state == 0);
2027 pid = task_pid(task);
2028 wake_up_all(&pid->wait_pidfd);
2029}
2030
2031/*
2032 * Let a parent know about the death of a child.
2033 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2034 *
2035 * Returns true if our parent ignored us and so we've switched to
2036 * self-reaping.
2037 */
2038bool do_notify_parent(struct task_struct *tsk, int sig)
2039{
2040 struct kernel_siginfo info;
2041 unsigned long flags;
2042 struct sighand_struct *psig;
2043 bool autoreap = false;
2044 u64 utime, stime;
2045
2046 WARN_ON_ONCE(sig == -1);
2047
2048 /* do_notify_parent_cldstop should have been called instead. */
2049 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2050
2051 WARN_ON_ONCE(!tsk->ptrace &&
2052 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2053
2054 /* Wake up all pidfd waiters */
2055 do_notify_pidfd(tsk);
2056
2057 if (sig != SIGCHLD) {
2058 /*
2059 * This is only possible if parent == real_parent.
2060 * Check if it has changed security domain.
2061 */
2062 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2063 sig = SIGCHLD;
2064 }
2065
2066 clear_siginfo(&info);
2067 info.si_signo = sig;
2068 info.si_errno = 0;
2069 /*
2070 * We are under tasklist_lock here so our parent is tied to
2071 * us and cannot change.
2072 *
2073 * task_active_pid_ns will always return the same pid namespace
2074 * until a task passes through release_task.
2075 *
2076 * write_lock() currently calls preempt_disable() which is the
2077 * same as rcu_read_lock(), but according to Oleg, this is not
2078 * correct to rely on this
2079 */
2080 rcu_read_lock();
2081 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2082 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2083 task_uid(tsk));
2084 rcu_read_unlock();
2085
2086 task_cputime(tsk, &utime, &stime);
2087 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2088 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2089
2090 info.si_status = tsk->exit_code & 0x7f;
2091 if (tsk->exit_code & 0x80)
2092 info.si_code = CLD_DUMPED;
2093 else if (tsk->exit_code & 0x7f)
2094 info.si_code = CLD_KILLED;
2095 else {
2096 info.si_code = CLD_EXITED;
2097 info.si_status = tsk->exit_code >> 8;
2098 }
2099
2100 psig = tsk->parent->sighand;
2101 spin_lock_irqsave(&psig->siglock, flags);
2102 if (!tsk->ptrace && sig == SIGCHLD &&
2103 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2104 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2105 /*
2106 * We are exiting and our parent doesn't care. POSIX.1
2107 * defines special semantics for setting SIGCHLD to SIG_IGN
2108 * or setting the SA_NOCLDWAIT flag: we should be reaped
2109 * automatically and not left for our parent's wait4 call.
2110 * Rather than having the parent do it as a magic kind of
2111 * signal handler, we just set this to tell do_exit that we
2112 * can be cleaned up without becoming a zombie. Note that
2113 * we still call __wake_up_parent in this case, because a
2114 * blocked sys_wait4 might now return -ECHILD.
2115 *
2116 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2117 * is implementation-defined: we do (if you don't want
2118 * it, just use SIG_IGN instead).
2119 */
2120 autoreap = true;
2121 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2122 sig = 0;
2123 }
2124 /*
2125 * Send with __send_signal as si_pid and si_uid are in the
2126 * parent's namespaces.
2127 */
2128 if (valid_signal(sig) && sig)
2129 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2130 __wake_up_parent(tsk, tsk->parent);
2131 spin_unlock_irqrestore(&psig->siglock, flags);
2132
2133 return autoreap;
2134}
2135
2136/**
2137 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2138 * @tsk: task reporting the state change
2139 * @for_ptracer: the notification is for ptracer
2140 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2141 *
2142 * Notify @tsk's parent that the stopped/continued state has changed. If
2143 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2144 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2145 *
2146 * CONTEXT:
2147 * Must be called with tasklist_lock at least read locked.
2148 */
2149static void do_notify_parent_cldstop(struct task_struct *tsk,
2150 bool for_ptracer, int why)
2151{
2152 struct kernel_siginfo info;
2153 unsigned long flags;
2154 struct task_struct *parent;
2155 struct sighand_struct *sighand;
2156 u64 utime, stime;
2157
2158 if (for_ptracer) {
2159 parent = tsk->parent;
2160 } else {
2161 tsk = tsk->group_leader;
2162 parent = tsk->real_parent;
2163 }
2164
2165 clear_siginfo(&info);
2166 info.si_signo = SIGCHLD;
2167 info.si_errno = 0;
2168 /*
2169 * see comment in do_notify_parent() about the following 4 lines
2170 */
2171 rcu_read_lock();
2172 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2173 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2174 rcu_read_unlock();
2175
2176 task_cputime(tsk, &utime, &stime);
2177 info.si_utime = nsec_to_clock_t(utime);
2178 info.si_stime = nsec_to_clock_t(stime);
2179
2180 info.si_code = why;
2181 switch (why) {
2182 case CLD_CONTINUED:
2183 info.si_status = SIGCONT;
2184 break;
2185 case CLD_STOPPED:
2186 info.si_status = tsk->signal->group_exit_code & 0x7f;
2187 break;
2188 case CLD_TRAPPED:
2189 info.si_status = tsk->exit_code & 0x7f;
2190 break;
2191 default:
2192 BUG();
2193 }
2194
2195 sighand = parent->sighand;
2196 spin_lock_irqsave(&sighand->siglock, flags);
2197 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2198 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2199 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2200 /*
2201 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2202 */
2203 __wake_up_parent(tsk, parent);
2204 spin_unlock_irqrestore(&sighand->siglock, flags);
2205}
2206
2207/*
2208 * This must be called with current->sighand->siglock held.
2209 *
2210 * This should be the path for all ptrace stops.
2211 * We always set current->last_siginfo while stopped here.
2212 * That makes it a way to test a stopped process for
2213 * being ptrace-stopped vs being job-control-stopped.
2214 *
2215 * Returns the signal the ptracer requested the code resume
2216 * with. If the code did not stop because the tracer is gone,
2217 * the stop signal remains unchanged unless clear_code.
2218 */
2219static int ptrace_stop(int exit_code, int why, unsigned long message,
2220 kernel_siginfo_t *info)
2221 __releases(¤t->sighand->siglock)
2222 __acquires(¤t->sighand->siglock)
2223{
2224 bool gstop_done = false;
2225
2226 if (arch_ptrace_stop_needed()) {
2227 /*
2228 * The arch code has something special to do before a
2229 * ptrace stop. This is allowed to block, e.g. for faults
2230 * on user stack pages. We can't keep the siglock while
2231 * calling arch_ptrace_stop, so we must release it now.
2232 * To preserve proper semantics, we must do this before
2233 * any signal bookkeeping like checking group_stop_count.
2234 */
2235 spin_unlock_irq(¤t->sighand->siglock);
2236 arch_ptrace_stop();
2237 spin_lock_irq(¤t->sighand->siglock);
2238 }
2239
2240 /*
2241 * After this point ptrace_signal_wake_up or signal_wake_up
2242 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2243 * signal comes in. Handle previous ptrace_unlinks and fatal
2244 * signals here to prevent ptrace_stop sleeping in schedule.
2245 */
2246 if (!current->ptrace || __fatal_signal_pending(current))
2247 return exit_code;
2248
2249 set_special_state(TASK_TRACED);
2250 current->jobctl |= JOBCTL_TRACED;
2251
2252 /*
2253 * We're committing to trapping. TRACED should be visible before
2254 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2255 * Also, transition to TRACED and updates to ->jobctl should be
2256 * atomic with respect to siglock and should be done after the arch
2257 * hook as siglock is released and regrabbed across it.
2258 *
2259 * TRACER TRACEE
2260 *
2261 * ptrace_attach()
2262 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2263 * do_wait()
2264 * set_current_state() smp_wmb();
2265 * ptrace_do_wait()
2266 * wait_task_stopped()
2267 * task_stopped_code()
2268 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2269 */
2270 smp_wmb();
2271
2272 current->ptrace_message = message;
2273 current->last_siginfo = info;
2274 current->exit_code = exit_code;
2275
2276 /*
2277 * If @why is CLD_STOPPED, we're trapping to participate in a group
2278 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2279 * across siglock relocks since INTERRUPT was scheduled, PENDING
2280 * could be clear now. We act as if SIGCONT is received after
2281 * TASK_TRACED is entered - ignore it.
2282 */
2283 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2284 gstop_done = task_participate_group_stop(current);
2285
2286 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2287 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2288 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2289 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2290
2291 /* entering a trap, clear TRAPPING */
2292 task_clear_jobctl_trapping(current);
2293
2294 spin_unlock_irq(¤t->sighand->siglock);
2295 read_lock(&tasklist_lock);
2296 /*
2297 * Notify parents of the stop.
2298 *
2299 * While ptraced, there are two parents - the ptracer and
2300 * the real_parent of the group_leader. The ptracer should
2301 * know about every stop while the real parent is only
2302 * interested in the completion of group stop. The states
2303 * for the two don't interact with each other. Notify
2304 * separately unless they're gonna be duplicates.
2305 */
2306 if (current->ptrace)
2307 do_notify_parent_cldstop(current, true, why);
2308 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2309 do_notify_parent_cldstop(current, false, why);
2310
2311 /*
2312 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2313 * One a PREEMPTION kernel this can result in preemption requirement
2314 * which will be fulfilled after read_unlock() and the ptracer will be
2315 * put on the CPU.
2316 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2317 * this task wait in schedule(). If this task gets preempted then it
2318 * remains enqueued on the runqueue. The ptracer will observe this and
2319 * then sleep for a delay of one HZ tick. In the meantime this task
2320 * gets scheduled, enters schedule() and will wait for the ptracer.
2321 *
2322 * This preemption point is not bad from a correctness point of
2323 * view but extends the runtime by one HZ tick time due to the
2324 * ptracer's sleep. The preempt-disable section ensures that there
2325 * will be no preemption between unlock and schedule() and so
2326 * improving the performance since the ptracer will observe that
2327 * the tracee is scheduled out once it gets on the CPU.
2328 *
2329 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2330 * Therefore the task can be preempted after do_notify_parent_cldstop()
2331 * before unlocking tasklist_lock so there is no benefit in doing this.
2332 *
2333 * In fact disabling preemption is harmful on PREEMPT_RT because
2334 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2335 * with preemption disabled due to the 'sleeping' spinlock
2336 * substitution of RT.
2337 */
2338 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2339 preempt_disable();
2340 read_unlock(&tasklist_lock);
2341 cgroup_enter_frozen();
2342 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2343 preempt_enable_no_resched();
2344 schedule();
2345 cgroup_leave_frozen(true);
2346
2347 /*
2348 * We are back. Now reacquire the siglock before touching
2349 * last_siginfo, so that we are sure to have synchronized with
2350 * any signal-sending on another CPU that wants to examine it.
2351 */
2352 spin_lock_irq(¤t->sighand->siglock);
2353 exit_code = current->exit_code;
2354 current->last_siginfo = NULL;
2355 current->ptrace_message = 0;
2356 current->exit_code = 0;
2357
2358 /* LISTENING can be set only during STOP traps, clear it */
2359 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2360
2361 /*
2362 * Queued signals ignored us while we were stopped for tracing.
2363 * So check for any that we should take before resuming user mode.
2364 * This sets TIF_SIGPENDING, but never clears it.
2365 */
2366 recalc_sigpending_tsk(current);
2367 return exit_code;
2368}
2369
2370static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2371{
2372 kernel_siginfo_t info;
2373
2374 clear_siginfo(&info);
2375 info.si_signo = signr;
2376 info.si_code = exit_code;
2377 info.si_pid = task_pid_vnr(current);
2378 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2379
2380 /* Let the debugger run. */
2381 return ptrace_stop(exit_code, why, message, &info);
2382}
2383
2384int ptrace_notify(int exit_code, unsigned long message)
2385{
2386 int signr;
2387
2388 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2389 if (unlikely(task_work_pending(current)))
2390 task_work_run();
2391
2392 spin_lock_irq(¤t->sighand->siglock);
2393 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2394 spin_unlock_irq(¤t->sighand->siglock);
2395 return signr;
2396}
2397
2398/**
2399 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2400 * @signr: signr causing group stop if initiating
2401 *
2402 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2403 * and participate in it. If already set, participate in the existing
2404 * group stop. If participated in a group stop (and thus slept), %true is
2405 * returned with siglock released.
2406 *
2407 * If ptraced, this function doesn't handle stop itself. Instead,
2408 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2409 * untouched. The caller must ensure that INTERRUPT trap handling takes
2410 * places afterwards.
2411 *
2412 * CONTEXT:
2413 * Must be called with @current->sighand->siglock held, which is released
2414 * on %true return.
2415 *
2416 * RETURNS:
2417 * %false if group stop is already cancelled or ptrace trap is scheduled.
2418 * %true if participated in group stop.
2419 */
2420static bool do_signal_stop(int signr)
2421 __releases(¤t->sighand->siglock)
2422{
2423 struct signal_struct *sig = current->signal;
2424
2425 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2426 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2427 struct task_struct *t;
2428
2429 /* signr will be recorded in task->jobctl for retries */
2430 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2431
2432 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2433 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2434 unlikely(sig->group_exec_task))
2435 return false;
2436 /*
2437 * There is no group stop already in progress. We must
2438 * initiate one now.
2439 *
2440 * While ptraced, a task may be resumed while group stop is
2441 * still in effect and then receive a stop signal and
2442 * initiate another group stop. This deviates from the
2443 * usual behavior as two consecutive stop signals can't
2444 * cause two group stops when !ptraced. That is why we
2445 * also check !task_is_stopped(t) below.
2446 *
2447 * The condition can be distinguished by testing whether
2448 * SIGNAL_STOP_STOPPED is already set. Don't generate
2449 * group_exit_code in such case.
2450 *
2451 * This is not necessary for SIGNAL_STOP_CONTINUED because
2452 * an intervening stop signal is required to cause two
2453 * continued events regardless of ptrace.
2454 */
2455 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2456 sig->group_exit_code = signr;
2457
2458 sig->group_stop_count = 0;
2459 if (task_set_jobctl_pending(current, signr | gstop))
2460 sig->group_stop_count++;
2461
2462 for_other_threads(current, t) {
2463 /*
2464 * Setting state to TASK_STOPPED for a group
2465 * stop is always done with the siglock held,
2466 * so this check has no races.
2467 */
2468 if (!task_is_stopped(t) &&
2469 task_set_jobctl_pending(t, signr | gstop)) {
2470 sig->group_stop_count++;
2471 if (likely(!(t->ptrace & PT_SEIZED)))
2472 signal_wake_up(t, 0);
2473 else
2474 ptrace_trap_notify(t);
2475 }
2476 }
2477 }
2478
2479 if (likely(!current->ptrace)) {
2480 int notify = 0;
2481
2482 /*
2483 * If there are no other threads in the group, or if there
2484 * is a group stop in progress and we are the last to stop,
2485 * report to the parent.
2486 */
2487 if (task_participate_group_stop(current))
2488 notify = CLD_STOPPED;
2489
2490 current->jobctl |= JOBCTL_STOPPED;
2491 set_special_state(TASK_STOPPED);
2492 spin_unlock_irq(¤t->sighand->siglock);
2493
2494 /*
2495 * Notify the parent of the group stop completion. Because
2496 * we're not holding either the siglock or tasklist_lock
2497 * here, ptracer may attach inbetween; however, this is for
2498 * group stop and should always be delivered to the real
2499 * parent of the group leader. The new ptracer will get
2500 * its notification when this task transitions into
2501 * TASK_TRACED.
2502 */
2503 if (notify) {
2504 read_lock(&tasklist_lock);
2505 do_notify_parent_cldstop(current, false, notify);
2506 read_unlock(&tasklist_lock);
2507 }
2508
2509 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2510 cgroup_enter_frozen();
2511 schedule();
2512 return true;
2513 } else {
2514 /*
2515 * While ptraced, group stop is handled by STOP trap.
2516 * Schedule it and let the caller deal with it.
2517 */
2518 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2519 return false;
2520 }
2521}
2522
2523/**
2524 * do_jobctl_trap - take care of ptrace jobctl traps
2525 *
2526 * When PT_SEIZED, it's used for both group stop and explicit
2527 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2528 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2529 * the stop signal; otherwise, %SIGTRAP.
2530 *
2531 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2532 * number as exit_code and no siginfo.
2533 *
2534 * CONTEXT:
2535 * Must be called with @current->sighand->siglock held, which may be
2536 * released and re-acquired before returning with intervening sleep.
2537 */
2538static void do_jobctl_trap(void)
2539{
2540 struct signal_struct *signal = current->signal;
2541 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2542
2543 if (current->ptrace & PT_SEIZED) {
2544 if (!signal->group_stop_count &&
2545 !(signal->flags & SIGNAL_STOP_STOPPED))
2546 signr = SIGTRAP;
2547 WARN_ON_ONCE(!signr);
2548 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2549 CLD_STOPPED, 0);
2550 } else {
2551 WARN_ON_ONCE(!signr);
2552 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2553 }
2554}
2555
2556/**
2557 * do_freezer_trap - handle the freezer jobctl trap
2558 *
2559 * Puts the task into frozen state, if only the task is not about to quit.
2560 * In this case it drops JOBCTL_TRAP_FREEZE.
2561 *
2562 * CONTEXT:
2563 * Must be called with @current->sighand->siglock held,
2564 * which is always released before returning.
2565 */
2566static void do_freezer_trap(void)
2567 __releases(¤t->sighand->siglock)
2568{
2569 /*
2570 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2571 * let's make another loop to give it a chance to be handled.
2572 * In any case, we'll return back.
2573 */
2574 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2575 JOBCTL_TRAP_FREEZE) {
2576 spin_unlock_irq(¤t->sighand->siglock);
2577 return;
2578 }
2579
2580 /*
2581 * Now we're sure that there is no pending fatal signal and no
2582 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2583 * immediately (if there is a non-fatal signal pending), and
2584 * put the task into sleep.
2585 */
2586 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2587 clear_thread_flag(TIF_SIGPENDING);
2588 spin_unlock_irq(¤t->sighand->siglock);
2589 cgroup_enter_frozen();
2590 schedule();
2591}
2592
2593static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2594{
2595 /*
2596 * We do not check sig_kernel_stop(signr) but set this marker
2597 * unconditionally because we do not know whether debugger will
2598 * change signr. This flag has no meaning unless we are going
2599 * to stop after return from ptrace_stop(). In this case it will
2600 * be checked in do_signal_stop(), we should only stop if it was
2601 * not cleared by SIGCONT while we were sleeping. See also the
2602 * comment in dequeue_signal().
2603 */
2604 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2605 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2606
2607 /* We're back. Did the debugger cancel the sig? */
2608 if (signr == 0)
2609 return signr;
2610
2611 /*
2612 * Update the siginfo structure if the signal has
2613 * changed. If the debugger wanted something
2614 * specific in the siginfo structure then it should
2615 * have updated *info via PTRACE_SETSIGINFO.
2616 */
2617 if (signr != info->si_signo) {
2618 clear_siginfo(info);
2619 info->si_signo = signr;
2620 info->si_errno = 0;
2621 info->si_code = SI_USER;
2622 rcu_read_lock();
2623 info->si_pid = task_pid_vnr(current->parent);
2624 info->si_uid = from_kuid_munged(current_user_ns(),
2625 task_uid(current->parent));
2626 rcu_read_unlock();
2627 }
2628
2629 /* If the (new) signal is now blocked, requeue it. */
2630 if (sigismember(¤t->blocked, signr) ||
2631 fatal_signal_pending(current)) {
2632 send_signal_locked(signr, info, current, type);
2633 signr = 0;
2634 }
2635
2636 return signr;
2637}
2638
2639static void hide_si_addr_tag_bits(struct ksignal *ksig)
2640{
2641 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2642 case SIL_FAULT:
2643 case SIL_FAULT_TRAPNO:
2644 case SIL_FAULT_MCEERR:
2645 case SIL_FAULT_BNDERR:
2646 case SIL_FAULT_PKUERR:
2647 case SIL_FAULT_PERF_EVENT:
2648 ksig->info.si_addr = arch_untagged_si_addr(
2649 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2650 break;
2651 case SIL_KILL:
2652 case SIL_TIMER:
2653 case SIL_POLL:
2654 case SIL_CHLD:
2655 case SIL_RT:
2656 case SIL_SYS:
2657 break;
2658 }
2659}
2660
2661bool get_signal(struct ksignal *ksig)
2662{
2663 struct sighand_struct *sighand = current->sighand;
2664 struct signal_struct *signal = current->signal;
2665 int signr;
2666
2667 clear_notify_signal();
2668 if (unlikely(task_work_pending(current)))
2669 task_work_run();
2670
2671 if (!task_sigpending(current))
2672 return false;
2673
2674 if (unlikely(uprobe_deny_signal()))
2675 return false;
2676
2677 /*
2678 * Do this once, we can't return to user-mode if freezing() == T.
2679 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2680 * thus do not need another check after return.
2681 */
2682 try_to_freeze();
2683
2684relock:
2685 spin_lock_irq(&sighand->siglock);
2686
2687 /*
2688 * Every stopped thread goes here after wakeup. Check to see if
2689 * we should notify the parent, prepare_signal(SIGCONT) encodes
2690 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2691 */
2692 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2693 int why;
2694
2695 if (signal->flags & SIGNAL_CLD_CONTINUED)
2696 why = CLD_CONTINUED;
2697 else
2698 why = CLD_STOPPED;
2699
2700 signal->flags &= ~SIGNAL_CLD_MASK;
2701
2702 spin_unlock_irq(&sighand->siglock);
2703
2704 /*
2705 * Notify the parent that we're continuing. This event is
2706 * always per-process and doesn't make whole lot of sense
2707 * for ptracers, who shouldn't consume the state via
2708 * wait(2) either, but, for backward compatibility, notify
2709 * the ptracer of the group leader too unless it's gonna be
2710 * a duplicate.
2711 */
2712 read_lock(&tasklist_lock);
2713 do_notify_parent_cldstop(current, false, why);
2714
2715 if (ptrace_reparented(current->group_leader))
2716 do_notify_parent_cldstop(current->group_leader,
2717 true, why);
2718 read_unlock(&tasklist_lock);
2719
2720 goto relock;
2721 }
2722
2723 for (;;) {
2724 struct k_sigaction *ka;
2725 enum pid_type type;
2726
2727 /* Has this task already been marked for death? */
2728 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2729 signal->group_exec_task) {
2730 clear_siginfo(&ksig->info);
2731 ksig->info.si_signo = signr = SIGKILL;
2732 sigdelset(¤t->pending.signal, SIGKILL);
2733 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2734 &sighand->action[SIGKILL - 1]);
2735 recalc_sigpending();
2736 goto fatal;
2737 }
2738
2739 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2740 do_signal_stop(0))
2741 goto relock;
2742
2743 if (unlikely(current->jobctl &
2744 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2745 if (current->jobctl & JOBCTL_TRAP_MASK) {
2746 do_jobctl_trap();
2747 spin_unlock_irq(&sighand->siglock);
2748 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2749 do_freezer_trap();
2750
2751 goto relock;
2752 }
2753
2754 /*
2755 * If the task is leaving the frozen state, let's update
2756 * cgroup counters and reset the frozen bit.
2757 */
2758 if (unlikely(cgroup_task_frozen(current))) {
2759 spin_unlock_irq(&sighand->siglock);
2760 cgroup_leave_frozen(false);
2761 goto relock;
2762 }
2763
2764 /*
2765 * Signals generated by the execution of an instruction
2766 * need to be delivered before any other pending signals
2767 * so that the instruction pointer in the signal stack
2768 * frame points to the faulting instruction.
2769 */
2770 type = PIDTYPE_PID;
2771 signr = dequeue_synchronous_signal(&ksig->info);
2772 if (!signr)
2773 signr = dequeue_signal(current, ¤t->blocked,
2774 &ksig->info, &type);
2775
2776 if (!signr)
2777 break; /* will return 0 */
2778
2779 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2780 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2781 signr = ptrace_signal(signr, &ksig->info, type);
2782 if (!signr)
2783 continue;
2784 }
2785
2786 ka = &sighand->action[signr-1];
2787
2788 /* Trace actually delivered signals. */
2789 trace_signal_deliver(signr, &ksig->info, ka);
2790
2791 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2792 continue;
2793 if (ka->sa.sa_handler != SIG_DFL) {
2794 /* Run the handler. */
2795 ksig->ka = *ka;
2796
2797 if (ka->sa.sa_flags & SA_ONESHOT)
2798 ka->sa.sa_handler = SIG_DFL;
2799
2800 break; /* will return non-zero "signr" value */
2801 }
2802
2803 /*
2804 * Now we are doing the default action for this signal.
2805 */
2806 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2807 continue;
2808
2809 /*
2810 * Global init gets no signals it doesn't want.
2811 * Container-init gets no signals it doesn't want from same
2812 * container.
2813 *
2814 * Note that if global/container-init sees a sig_kernel_only()
2815 * signal here, the signal must have been generated internally
2816 * or must have come from an ancestor namespace. In either
2817 * case, the signal cannot be dropped.
2818 */
2819 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2820 !sig_kernel_only(signr))
2821 continue;
2822
2823 if (sig_kernel_stop(signr)) {
2824 /*
2825 * The default action is to stop all threads in
2826 * the thread group. The job control signals
2827 * do nothing in an orphaned pgrp, but SIGSTOP
2828 * always works. Note that siglock needs to be
2829 * dropped during the call to is_orphaned_pgrp()
2830 * because of lock ordering with tasklist_lock.
2831 * This allows an intervening SIGCONT to be posted.
2832 * We need to check for that and bail out if necessary.
2833 */
2834 if (signr != SIGSTOP) {
2835 spin_unlock_irq(&sighand->siglock);
2836
2837 /* signals can be posted during this window */
2838
2839 if (is_current_pgrp_orphaned())
2840 goto relock;
2841
2842 spin_lock_irq(&sighand->siglock);
2843 }
2844
2845 if (likely(do_signal_stop(ksig->info.si_signo))) {
2846 /* It released the siglock. */
2847 goto relock;
2848 }
2849
2850 /*
2851 * We didn't actually stop, due to a race
2852 * with SIGCONT or something like that.
2853 */
2854 continue;
2855 }
2856
2857 fatal:
2858 spin_unlock_irq(&sighand->siglock);
2859 if (unlikely(cgroup_task_frozen(current)))
2860 cgroup_leave_frozen(true);
2861
2862 /*
2863 * Anything else is fatal, maybe with a core dump.
2864 */
2865 current->flags |= PF_SIGNALED;
2866
2867 if (sig_kernel_coredump(signr)) {
2868 if (print_fatal_signals)
2869 print_fatal_signal(ksig->info.si_signo);
2870 proc_coredump_connector(current);
2871 /*
2872 * If it was able to dump core, this kills all
2873 * other threads in the group and synchronizes with
2874 * their demise. If we lost the race with another
2875 * thread getting here, it set group_exit_code
2876 * first and our do_group_exit call below will use
2877 * that value and ignore the one we pass it.
2878 */
2879 do_coredump(&ksig->info);
2880 }
2881
2882 /*
2883 * PF_USER_WORKER threads will catch and exit on fatal signals
2884 * themselves. They have cleanup that must be performed, so
2885 * we cannot call do_exit() on their behalf.
2886 */
2887 if (current->flags & PF_USER_WORKER)
2888 goto out;
2889
2890 /*
2891 * Death signals, no core dump.
2892 */
2893 do_group_exit(ksig->info.si_signo);
2894 /* NOTREACHED */
2895 }
2896 spin_unlock_irq(&sighand->siglock);
2897out:
2898 ksig->sig = signr;
2899
2900 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2901 hide_si_addr_tag_bits(ksig);
2902
2903 return ksig->sig > 0;
2904}
2905
2906/**
2907 * signal_delivered - called after signal delivery to update blocked signals
2908 * @ksig: kernel signal struct
2909 * @stepping: nonzero if debugger single-step or block-step in use
2910 *
2911 * This function should be called when a signal has successfully been
2912 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2913 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2914 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2915 */
2916static void signal_delivered(struct ksignal *ksig, int stepping)
2917{
2918 sigset_t blocked;
2919
2920 /* A signal was successfully delivered, and the
2921 saved sigmask was stored on the signal frame,
2922 and will be restored by sigreturn. So we can
2923 simply clear the restore sigmask flag. */
2924 clear_restore_sigmask();
2925
2926 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2927 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2928 sigaddset(&blocked, ksig->sig);
2929 set_current_blocked(&blocked);
2930 if (current->sas_ss_flags & SS_AUTODISARM)
2931 sas_ss_reset(current);
2932 if (stepping)
2933 ptrace_notify(SIGTRAP, 0);
2934}
2935
2936void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2937{
2938 if (failed)
2939 force_sigsegv(ksig->sig);
2940 else
2941 signal_delivered(ksig, stepping);
2942}
2943
2944/*
2945 * It could be that complete_signal() picked us to notify about the
2946 * group-wide signal. Other threads should be notified now to take
2947 * the shared signals in @which since we will not.
2948 */
2949static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2950{
2951 sigset_t retarget;
2952 struct task_struct *t;
2953
2954 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2955 if (sigisemptyset(&retarget))
2956 return;
2957
2958 for_other_threads(tsk, t) {
2959 if (t->flags & PF_EXITING)
2960 continue;
2961
2962 if (!has_pending_signals(&retarget, &t->blocked))
2963 continue;
2964 /* Remove the signals this thread can handle. */
2965 sigandsets(&retarget, &retarget, &t->blocked);
2966
2967 if (!task_sigpending(t))
2968 signal_wake_up(t, 0);
2969
2970 if (sigisemptyset(&retarget))
2971 break;
2972 }
2973}
2974
2975void exit_signals(struct task_struct *tsk)
2976{
2977 int group_stop = 0;
2978 sigset_t unblocked;
2979
2980 /*
2981 * @tsk is about to have PF_EXITING set - lock out users which
2982 * expect stable threadgroup.
2983 */
2984 cgroup_threadgroup_change_begin(tsk);
2985
2986 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
2987 sched_mm_cid_exit_signals(tsk);
2988 tsk->flags |= PF_EXITING;
2989 cgroup_threadgroup_change_end(tsk);
2990 return;
2991 }
2992
2993 spin_lock_irq(&tsk->sighand->siglock);
2994 /*
2995 * From now this task is not visible for group-wide signals,
2996 * see wants_signal(), do_signal_stop().
2997 */
2998 sched_mm_cid_exit_signals(tsk);
2999 tsk->flags |= PF_EXITING;
3000
3001 cgroup_threadgroup_change_end(tsk);
3002
3003 if (!task_sigpending(tsk))
3004 goto out;
3005
3006 unblocked = tsk->blocked;
3007 signotset(&unblocked);
3008 retarget_shared_pending(tsk, &unblocked);
3009
3010 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3011 task_participate_group_stop(tsk))
3012 group_stop = CLD_STOPPED;
3013out:
3014 spin_unlock_irq(&tsk->sighand->siglock);
3015
3016 /*
3017 * If group stop has completed, deliver the notification. This
3018 * should always go to the real parent of the group leader.
3019 */
3020 if (unlikely(group_stop)) {
3021 read_lock(&tasklist_lock);
3022 do_notify_parent_cldstop(tsk, false, group_stop);
3023 read_unlock(&tasklist_lock);
3024 }
3025}
3026
3027/*
3028 * System call entry points.
3029 */
3030
3031/**
3032 * sys_restart_syscall - restart a system call
3033 */
3034SYSCALL_DEFINE0(restart_syscall)
3035{
3036 struct restart_block *restart = ¤t->restart_block;
3037 return restart->fn(restart);
3038}
3039
3040long do_no_restart_syscall(struct restart_block *param)
3041{
3042 return -EINTR;
3043}
3044
3045static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3046{
3047 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3048 sigset_t newblocked;
3049 /* A set of now blocked but previously unblocked signals. */
3050 sigandnsets(&newblocked, newset, ¤t->blocked);
3051 retarget_shared_pending(tsk, &newblocked);
3052 }
3053 tsk->blocked = *newset;
3054 recalc_sigpending();
3055}
3056
3057/**
3058 * set_current_blocked - change current->blocked mask
3059 * @newset: new mask
3060 *
3061 * It is wrong to change ->blocked directly, this helper should be used
3062 * to ensure the process can't miss a shared signal we are going to block.
3063 */
3064void set_current_blocked(sigset_t *newset)
3065{
3066 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3067 __set_current_blocked(newset);
3068}
3069
3070void __set_current_blocked(const sigset_t *newset)
3071{
3072 struct task_struct *tsk = current;
3073
3074 /*
3075 * In case the signal mask hasn't changed, there is nothing we need
3076 * to do. The current->blocked shouldn't be modified by other task.
3077 */
3078 if (sigequalsets(&tsk->blocked, newset))
3079 return;
3080
3081 spin_lock_irq(&tsk->sighand->siglock);
3082 __set_task_blocked(tsk, newset);
3083 spin_unlock_irq(&tsk->sighand->siglock);
3084}
3085
3086/*
3087 * This is also useful for kernel threads that want to temporarily
3088 * (or permanently) block certain signals.
3089 *
3090 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3091 * interface happily blocks "unblockable" signals like SIGKILL
3092 * and friends.
3093 */
3094int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3095{
3096 struct task_struct *tsk = current;
3097 sigset_t newset;
3098
3099 /* Lockless, only current can change ->blocked, never from irq */
3100 if (oldset)
3101 *oldset = tsk->blocked;
3102
3103 switch (how) {
3104 case SIG_BLOCK:
3105 sigorsets(&newset, &tsk->blocked, set);
3106 break;
3107 case SIG_UNBLOCK:
3108 sigandnsets(&newset, &tsk->blocked, set);
3109 break;
3110 case SIG_SETMASK:
3111 newset = *set;
3112 break;
3113 default:
3114 return -EINVAL;
3115 }
3116
3117 __set_current_blocked(&newset);
3118 return 0;
3119}
3120EXPORT_SYMBOL(sigprocmask);
3121
3122/*
3123 * The api helps set app-provided sigmasks.
3124 *
3125 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3126 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3127 *
3128 * Note that it does set_restore_sigmask() in advance, so it must be always
3129 * paired with restore_saved_sigmask_unless() before return from syscall.
3130 */
3131int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3132{
3133 sigset_t kmask;
3134
3135 if (!umask)
3136 return 0;
3137 if (sigsetsize != sizeof(sigset_t))
3138 return -EINVAL;
3139 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3140 return -EFAULT;
3141
3142 set_restore_sigmask();
3143 current->saved_sigmask = current->blocked;
3144 set_current_blocked(&kmask);
3145
3146 return 0;
3147}
3148
3149#ifdef CONFIG_COMPAT
3150int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3151 size_t sigsetsize)
3152{
3153 sigset_t kmask;
3154
3155 if (!umask)
3156 return 0;
3157 if (sigsetsize != sizeof(compat_sigset_t))
3158 return -EINVAL;
3159 if (get_compat_sigset(&kmask, umask))
3160 return -EFAULT;
3161
3162 set_restore_sigmask();
3163 current->saved_sigmask = current->blocked;
3164 set_current_blocked(&kmask);
3165
3166 return 0;
3167}
3168#endif
3169
3170/**
3171 * sys_rt_sigprocmask - change the list of currently blocked signals
3172 * @how: whether to add, remove, or set signals
3173 * @nset: stores pending signals
3174 * @oset: previous value of signal mask if non-null
3175 * @sigsetsize: size of sigset_t type
3176 */
3177SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3178 sigset_t __user *, oset, size_t, sigsetsize)
3179{
3180 sigset_t old_set, new_set;
3181 int error;
3182
3183 /* XXX: Don't preclude handling different sized sigset_t's. */
3184 if (sigsetsize != sizeof(sigset_t))
3185 return -EINVAL;
3186
3187 old_set = current->blocked;
3188
3189 if (nset) {
3190 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3191 return -EFAULT;
3192 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3193
3194 error = sigprocmask(how, &new_set, NULL);
3195 if (error)
3196 return error;
3197 }
3198
3199 if (oset) {
3200 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3201 return -EFAULT;
3202 }
3203
3204 return 0;
3205}
3206
3207#ifdef CONFIG_COMPAT
3208COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3209 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3210{
3211 sigset_t old_set = current->blocked;
3212
3213 /* XXX: Don't preclude handling different sized sigset_t's. */
3214 if (sigsetsize != sizeof(sigset_t))
3215 return -EINVAL;
3216
3217 if (nset) {
3218 sigset_t new_set;
3219 int error;
3220 if (get_compat_sigset(&new_set, nset))
3221 return -EFAULT;
3222 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3223
3224 error = sigprocmask(how, &new_set, NULL);
3225 if (error)
3226 return error;
3227 }
3228 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3229}
3230#endif
3231
3232static void do_sigpending(sigset_t *set)
3233{
3234 spin_lock_irq(¤t->sighand->siglock);
3235 sigorsets(set, ¤t->pending.signal,
3236 ¤t->signal->shared_pending.signal);
3237 spin_unlock_irq(¤t->sighand->siglock);
3238
3239 /* Outside the lock because only this thread touches it. */
3240 sigandsets(set, ¤t->blocked, set);
3241}
3242
3243/**
3244 * sys_rt_sigpending - examine a pending signal that has been raised
3245 * while blocked
3246 * @uset: stores pending signals
3247 * @sigsetsize: size of sigset_t type or larger
3248 */
3249SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3250{
3251 sigset_t set;
3252
3253 if (sigsetsize > sizeof(*uset))
3254 return -EINVAL;
3255
3256 do_sigpending(&set);
3257
3258 if (copy_to_user(uset, &set, sigsetsize))
3259 return -EFAULT;
3260
3261 return 0;
3262}
3263
3264#ifdef CONFIG_COMPAT
3265COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3266 compat_size_t, sigsetsize)
3267{
3268 sigset_t set;
3269
3270 if (sigsetsize > sizeof(*uset))
3271 return -EINVAL;
3272
3273 do_sigpending(&set);
3274
3275 return put_compat_sigset(uset, &set, sigsetsize);
3276}
3277#endif
3278
3279static const struct {
3280 unsigned char limit, layout;
3281} sig_sicodes[] = {
3282 [SIGILL] = { NSIGILL, SIL_FAULT },
3283 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3284 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3285 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3286 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3287#if defined(SIGEMT)
3288 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3289#endif
3290 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3291 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3292 [SIGSYS] = { NSIGSYS, SIL_SYS },
3293};
3294
3295static bool known_siginfo_layout(unsigned sig, int si_code)
3296{
3297 if (si_code == SI_KERNEL)
3298 return true;
3299 else if ((si_code > SI_USER)) {
3300 if (sig_specific_sicodes(sig)) {
3301 if (si_code <= sig_sicodes[sig].limit)
3302 return true;
3303 }
3304 else if (si_code <= NSIGPOLL)
3305 return true;
3306 }
3307 else if (si_code >= SI_DETHREAD)
3308 return true;
3309 else if (si_code == SI_ASYNCNL)
3310 return true;
3311 return false;
3312}
3313
3314enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3315{
3316 enum siginfo_layout layout = SIL_KILL;
3317 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3318 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3319 (si_code <= sig_sicodes[sig].limit)) {
3320 layout = sig_sicodes[sig].layout;
3321 /* Handle the exceptions */
3322 if ((sig == SIGBUS) &&
3323 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3324 layout = SIL_FAULT_MCEERR;
3325 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3326 layout = SIL_FAULT_BNDERR;
3327#ifdef SEGV_PKUERR
3328 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3329 layout = SIL_FAULT_PKUERR;
3330#endif
3331 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3332 layout = SIL_FAULT_PERF_EVENT;
3333 else if (IS_ENABLED(CONFIG_SPARC) &&
3334 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3335 layout = SIL_FAULT_TRAPNO;
3336 else if (IS_ENABLED(CONFIG_ALPHA) &&
3337 ((sig == SIGFPE) ||
3338 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3339 layout = SIL_FAULT_TRAPNO;
3340 }
3341 else if (si_code <= NSIGPOLL)
3342 layout = SIL_POLL;
3343 } else {
3344 if (si_code == SI_TIMER)
3345 layout = SIL_TIMER;
3346 else if (si_code == SI_SIGIO)
3347 layout = SIL_POLL;
3348 else if (si_code < 0)
3349 layout = SIL_RT;
3350 }
3351 return layout;
3352}
3353
3354static inline char __user *si_expansion(const siginfo_t __user *info)
3355{
3356 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3357}
3358
3359int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3360{
3361 char __user *expansion = si_expansion(to);
3362 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3363 return -EFAULT;
3364 if (clear_user(expansion, SI_EXPANSION_SIZE))
3365 return -EFAULT;
3366 return 0;
3367}
3368
3369static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3370 const siginfo_t __user *from)
3371{
3372 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3373 char __user *expansion = si_expansion(from);
3374 char buf[SI_EXPANSION_SIZE];
3375 int i;
3376 /*
3377 * An unknown si_code might need more than
3378 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3379 * extra bytes are 0. This guarantees copy_siginfo_to_user
3380 * will return this data to userspace exactly.
3381 */
3382 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3383 return -EFAULT;
3384 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3385 if (buf[i] != 0)
3386 return -E2BIG;
3387 }
3388 }
3389 return 0;
3390}
3391
3392static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3393 const siginfo_t __user *from)
3394{
3395 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3396 return -EFAULT;
3397 to->si_signo = signo;
3398 return post_copy_siginfo_from_user(to, from);
3399}
3400
3401int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3402{
3403 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3404 return -EFAULT;
3405 return post_copy_siginfo_from_user(to, from);
3406}
3407
3408#ifdef CONFIG_COMPAT
3409/**
3410 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3411 * @to: compat siginfo destination
3412 * @from: kernel siginfo source
3413 *
3414 * Note: This function does not work properly for the SIGCHLD on x32, but
3415 * fortunately it doesn't have to. The only valid callers for this function are
3416 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3417 * The latter does not care because SIGCHLD will never cause a coredump.
3418 */
3419void copy_siginfo_to_external32(struct compat_siginfo *to,
3420 const struct kernel_siginfo *from)
3421{
3422 memset(to, 0, sizeof(*to));
3423
3424 to->si_signo = from->si_signo;
3425 to->si_errno = from->si_errno;
3426 to->si_code = from->si_code;
3427 switch(siginfo_layout(from->si_signo, from->si_code)) {
3428 case SIL_KILL:
3429 to->si_pid = from->si_pid;
3430 to->si_uid = from->si_uid;
3431 break;
3432 case SIL_TIMER:
3433 to->si_tid = from->si_tid;
3434 to->si_overrun = from->si_overrun;
3435 to->si_int = from->si_int;
3436 break;
3437 case SIL_POLL:
3438 to->si_band = from->si_band;
3439 to->si_fd = from->si_fd;
3440 break;
3441 case SIL_FAULT:
3442 to->si_addr = ptr_to_compat(from->si_addr);
3443 break;
3444 case SIL_FAULT_TRAPNO:
3445 to->si_addr = ptr_to_compat(from->si_addr);
3446 to->si_trapno = from->si_trapno;
3447 break;
3448 case SIL_FAULT_MCEERR:
3449 to->si_addr = ptr_to_compat(from->si_addr);
3450 to->si_addr_lsb = from->si_addr_lsb;
3451 break;
3452 case SIL_FAULT_BNDERR:
3453 to->si_addr = ptr_to_compat(from->si_addr);
3454 to->si_lower = ptr_to_compat(from->si_lower);
3455 to->si_upper = ptr_to_compat(from->si_upper);
3456 break;
3457 case SIL_FAULT_PKUERR:
3458 to->si_addr = ptr_to_compat(from->si_addr);
3459 to->si_pkey = from->si_pkey;
3460 break;
3461 case SIL_FAULT_PERF_EVENT:
3462 to->si_addr = ptr_to_compat(from->si_addr);
3463 to->si_perf_data = from->si_perf_data;
3464 to->si_perf_type = from->si_perf_type;
3465 to->si_perf_flags = from->si_perf_flags;
3466 break;
3467 case SIL_CHLD:
3468 to->si_pid = from->si_pid;
3469 to->si_uid = from->si_uid;
3470 to->si_status = from->si_status;
3471 to->si_utime = from->si_utime;
3472 to->si_stime = from->si_stime;
3473 break;
3474 case SIL_RT:
3475 to->si_pid = from->si_pid;
3476 to->si_uid = from->si_uid;
3477 to->si_int = from->si_int;
3478 break;
3479 case SIL_SYS:
3480 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3481 to->si_syscall = from->si_syscall;
3482 to->si_arch = from->si_arch;
3483 break;
3484 }
3485}
3486
3487int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3488 const struct kernel_siginfo *from)
3489{
3490 struct compat_siginfo new;
3491
3492 copy_siginfo_to_external32(&new, from);
3493 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3494 return -EFAULT;
3495 return 0;
3496}
3497
3498static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3499 const struct compat_siginfo *from)
3500{
3501 clear_siginfo(to);
3502 to->si_signo = from->si_signo;
3503 to->si_errno = from->si_errno;
3504 to->si_code = from->si_code;
3505 switch(siginfo_layout(from->si_signo, from->si_code)) {
3506 case SIL_KILL:
3507 to->si_pid = from->si_pid;
3508 to->si_uid = from->si_uid;
3509 break;
3510 case SIL_TIMER:
3511 to->si_tid = from->si_tid;
3512 to->si_overrun = from->si_overrun;
3513 to->si_int = from->si_int;
3514 break;
3515 case SIL_POLL:
3516 to->si_band = from->si_band;
3517 to->si_fd = from->si_fd;
3518 break;
3519 case SIL_FAULT:
3520 to->si_addr = compat_ptr(from->si_addr);
3521 break;
3522 case SIL_FAULT_TRAPNO:
3523 to->si_addr = compat_ptr(from->si_addr);
3524 to->si_trapno = from->si_trapno;
3525 break;
3526 case SIL_FAULT_MCEERR:
3527 to->si_addr = compat_ptr(from->si_addr);
3528 to->si_addr_lsb = from->si_addr_lsb;
3529 break;
3530 case SIL_FAULT_BNDERR:
3531 to->si_addr = compat_ptr(from->si_addr);
3532 to->si_lower = compat_ptr(from->si_lower);
3533 to->si_upper = compat_ptr(from->si_upper);
3534 break;
3535 case SIL_FAULT_PKUERR:
3536 to->si_addr = compat_ptr(from->si_addr);
3537 to->si_pkey = from->si_pkey;
3538 break;
3539 case SIL_FAULT_PERF_EVENT:
3540 to->si_addr = compat_ptr(from->si_addr);
3541 to->si_perf_data = from->si_perf_data;
3542 to->si_perf_type = from->si_perf_type;
3543 to->si_perf_flags = from->si_perf_flags;
3544 break;
3545 case SIL_CHLD:
3546 to->si_pid = from->si_pid;
3547 to->si_uid = from->si_uid;
3548 to->si_status = from->si_status;
3549#ifdef CONFIG_X86_X32_ABI
3550 if (in_x32_syscall()) {
3551 to->si_utime = from->_sifields._sigchld_x32._utime;
3552 to->si_stime = from->_sifields._sigchld_x32._stime;
3553 } else
3554#endif
3555 {
3556 to->si_utime = from->si_utime;
3557 to->si_stime = from->si_stime;
3558 }
3559 break;
3560 case SIL_RT:
3561 to->si_pid = from->si_pid;
3562 to->si_uid = from->si_uid;
3563 to->si_int = from->si_int;
3564 break;
3565 case SIL_SYS:
3566 to->si_call_addr = compat_ptr(from->si_call_addr);
3567 to->si_syscall = from->si_syscall;
3568 to->si_arch = from->si_arch;
3569 break;
3570 }
3571 return 0;
3572}
3573
3574static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3575 const struct compat_siginfo __user *ufrom)
3576{
3577 struct compat_siginfo from;
3578
3579 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3580 return -EFAULT;
3581
3582 from.si_signo = signo;
3583 return post_copy_siginfo_from_user32(to, &from);
3584}
3585
3586int copy_siginfo_from_user32(struct kernel_siginfo *to,
3587 const struct compat_siginfo __user *ufrom)
3588{
3589 struct compat_siginfo from;
3590
3591 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3592 return -EFAULT;
3593
3594 return post_copy_siginfo_from_user32(to, &from);
3595}
3596#endif /* CONFIG_COMPAT */
3597
3598/**
3599 * do_sigtimedwait - wait for queued signals specified in @which
3600 * @which: queued signals to wait for
3601 * @info: if non-null, the signal's siginfo is returned here
3602 * @ts: upper bound on process time suspension
3603 */
3604static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3605 const struct timespec64 *ts)
3606{
3607 ktime_t *to = NULL, timeout = KTIME_MAX;
3608 struct task_struct *tsk = current;
3609 sigset_t mask = *which;
3610 enum pid_type type;
3611 int sig, ret = 0;
3612
3613 if (ts) {
3614 if (!timespec64_valid(ts))
3615 return -EINVAL;
3616 timeout = timespec64_to_ktime(*ts);
3617 to = &timeout;
3618 }
3619
3620 /*
3621 * Invert the set of allowed signals to get those we want to block.
3622 */
3623 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3624 signotset(&mask);
3625
3626 spin_lock_irq(&tsk->sighand->siglock);
3627 sig = dequeue_signal(tsk, &mask, info, &type);
3628 if (!sig && timeout) {
3629 /*
3630 * None ready, temporarily unblock those we're interested
3631 * while we are sleeping in so that we'll be awakened when
3632 * they arrive. Unblocking is always fine, we can avoid
3633 * set_current_blocked().
3634 */
3635 tsk->real_blocked = tsk->blocked;
3636 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3637 recalc_sigpending();
3638 spin_unlock_irq(&tsk->sighand->siglock);
3639
3640 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3641 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3642 HRTIMER_MODE_REL);
3643 spin_lock_irq(&tsk->sighand->siglock);
3644 __set_task_blocked(tsk, &tsk->real_blocked);
3645 sigemptyset(&tsk->real_blocked);
3646 sig = dequeue_signal(tsk, &mask, info, &type);
3647 }
3648 spin_unlock_irq(&tsk->sighand->siglock);
3649
3650 if (sig)
3651 return sig;
3652 return ret ? -EINTR : -EAGAIN;
3653}
3654
3655/**
3656 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3657 * in @uthese
3658 * @uthese: queued signals to wait for
3659 * @uinfo: if non-null, the signal's siginfo is returned here
3660 * @uts: upper bound on process time suspension
3661 * @sigsetsize: size of sigset_t type
3662 */
3663SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3664 siginfo_t __user *, uinfo,
3665 const struct __kernel_timespec __user *, uts,
3666 size_t, sigsetsize)
3667{
3668 sigset_t these;
3669 struct timespec64 ts;
3670 kernel_siginfo_t info;
3671 int ret;
3672
3673 /* XXX: Don't preclude handling different sized sigset_t's. */
3674 if (sigsetsize != sizeof(sigset_t))
3675 return -EINVAL;
3676
3677 if (copy_from_user(&these, uthese, sizeof(these)))
3678 return -EFAULT;
3679
3680 if (uts) {
3681 if (get_timespec64(&ts, uts))
3682 return -EFAULT;
3683 }
3684
3685 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3686
3687 if (ret > 0 && uinfo) {
3688 if (copy_siginfo_to_user(uinfo, &info))
3689 ret = -EFAULT;
3690 }
3691
3692 return ret;
3693}
3694
3695#ifdef CONFIG_COMPAT_32BIT_TIME
3696SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3697 siginfo_t __user *, uinfo,
3698 const struct old_timespec32 __user *, uts,
3699 size_t, sigsetsize)
3700{
3701 sigset_t these;
3702 struct timespec64 ts;
3703 kernel_siginfo_t info;
3704 int ret;
3705
3706 if (sigsetsize != sizeof(sigset_t))
3707 return -EINVAL;
3708
3709 if (copy_from_user(&these, uthese, sizeof(these)))
3710 return -EFAULT;
3711
3712 if (uts) {
3713 if (get_old_timespec32(&ts, uts))
3714 return -EFAULT;
3715 }
3716
3717 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3718
3719 if (ret > 0 && uinfo) {
3720 if (copy_siginfo_to_user(uinfo, &info))
3721 ret = -EFAULT;
3722 }
3723
3724 return ret;
3725}
3726#endif
3727
3728#ifdef CONFIG_COMPAT
3729COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3730 struct compat_siginfo __user *, uinfo,
3731 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3732{
3733 sigset_t s;
3734 struct timespec64 t;
3735 kernel_siginfo_t info;
3736 long ret;
3737
3738 if (sigsetsize != sizeof(sigset_t))
3739 return -EINVAL;
3740
3741 if (get_compat_sigset(&s, uthese))
3742 return -EFAULT;
3743
3744 if (uts) {
3745 if (get_timespec64(&t, uts))
3746 return -EFAULT;
3747 }
3748
3749 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3750
3751 if (ret > 0 && uinfo) {
3752 if (copy_siginfo_to_user32(uinfo, &info))
3753 ret = -EFAULT;
3754 }
3755
3756 return ret;
3757}
3758
3759#ifdef CONFIG_COMPAT_32BIT_TIME
3760COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3761 struct compat_siginfo __user *, uinfo,
3762 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3763{
3764 sigset_t s;
3765 struct timespec64 t;
3766 kernel_siginfo_t info;
3767 long ret;
3768
3769 if (sigsetsize != sizeof(sigset_t))
3770 return -EINVAL;
3771
3772 if (get_compat_sigset(&s, uthese))
3773 return -EFAULT;
3774
3775 if (uts) {
3776 if (get_old_timespec32(&t, uts))
3777 return -EFAULT;
3778 }
3779
3780 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3781
3782 if (ret > 0 && uinfo) {
3783 if (copy_siginfo_to_user32(uinfo, &info))
3784 ret = -EFAULT;
3785 }
3786
3787 return ret;
3788}
3789#endif
3790#endif
3791
3792static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3793{
3794 clear_siginfo(info);
3795 info->si_signo = sig;
3796 info->si_errno = 0;
3797 info->si_code = SI_USER;
3798 info->si_pid = task_tgid_vnr(current);
3799 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3800}
3801
3802/**
3803 * sys_kill - send a signal to a process
3804 * @pid: the PID of the process
3805 * @sig: signal to be sent
3806 */
3807SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3808{
3809 struct kernel_siginfo info;
3810
3811 prepare_kill_siginfo(sig, &info);
3812
3813 return kill_something_info(sig, &info, pid);
3814}
3815
3816/*
3817 * Verify that the signaler and signalee either are in the same pid namespace
3818 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3819 * namespace.
3820 */
3821static bool access_pidfd_pidns(struct pid *pid)
3822{
3823 struct pid_namespace *active = task_active_pid_ns(current);
3824 struct pid_namespace *p = ns_of_pid(pid);
3825
3826 for (;;) {
3827 if (!p)
3828 return false;
3829 if (p == active)
3830 break;
3831 p = p->parent;
3832 }
3833
3834 return true;
3835}
3836
3837static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3838 siginfo_t __user *info)
3839{
3840#ifdef CONFIG_COMPAT
3841 /*
3842 * Avoid hooking up compat syscalls and instead handle necessary
3843 * conversions here. Note, this is a stop-gap measure and should not be
3844 * considered a generic solution.
3845 */
3846 if (in_compat_syscall())
3847 return copy_siginfo_from_user32(
3848 kinfo, (struct compat_siginfo __user *)info);
3849#endif
3850 return copy_siginfo_from_user(kinfo, info);
3851}
3852
3853static struct pid *pidfd_to_pid(const struct file *file)
3854{
3855 struct pid *pid;
3856
3857 pid = pidfd_pid(file);
3858 if (!IS_ERR(pid))
3859 return pid;
3860
3861 return tgid_pidfd_to_pid(file);
3862}
3863
3864/**
3865 * sys_pidfd_send_signal - Signal a process through a pidfd
3866 * @pidfd: file descriptor of the process
3867 * @sig: signal to send
3868 * @info: signal info
3869 * @flags: future flags
3870 *
3871 * The syscall currently only signals via PIDTYPE_PID which covers
3872 * kill(<positive-pid>, <signal>. It does not signal threads or process
3873 * groups.
3874 * In order to extend the syscall to threads and process groups the @flags
3875 * argument should be used. In essence, the @flags argument will determine
3876 * what is signaled and not the file descriptor itself. Put in other words,
3877 * grouping is a property of the flags argument not a property of the file
3878 * descriptor.
3879 *
3880 * Return: 0 on success, negative errno on failure
3881 */
3882SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3883 siginfo_t __user *, info, unsigned int, flags)
3884{
3885 int ret;
3886 struct fd f;
3887 struct pid *pid;
3888 kernel_siginfo_t kinfo;
3889
3890 /* Enforce flags be set to 0 until we add an extension. */
3891 if (flags)
3892 return -EINVAL;
3893
3894 f = fdget(pidfd);
3895 if (!f.file)
3896 return -EBADF;
3897
3898 /* Is this a pidfd? */
3899 pid = pidfd_to_pid(f.file);
3900 if (IS_ERR(pid)) {
3901 ret = PTR_ERR(pid);
3902 goto err;
3903 }
3904
3905 ret = -EINVAL;
3906 if (!access_pidfd_pidns(pid))
3907 goto err;
3908
3909 if (info) {
3910 ret = copy_siginfo_from_user_any(&kinfo, info);
3911 if (unlikely(ret))
3912 goto err;
3913
3914 ret = -EINVAL;
3915 if (unlikely(sig != kinfo.si_signo))
3916 goto err;
3917
3918 /* Only allow sending arbitrary signals to yourself. */
3919 ret = -EPERM;
3920 if ((task_pid(current) != pid) &&
3921 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3922 goto err;
3923 } else {
3924 prepare_kill_siginfo(sig, &kinfo);
3925 }
3926
3927 ret = kill_pid_info(sig, &kinfo, pid);
3928
3929err:
3930 fdput(f);
3931 return ret;
3932}
3933
3934static int
3935do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3936{
3937 struct task_struct *p;
3938 int error = -ESRCH;
3939
3940 rcu_read_lock();
3941 p = find_task_by_vpid(pid);
3942 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3943 error = check_kill_permission(sig, info, p);
3944 /*
3945 * The null signal is a permissions and process existence
3946 * probe. No signal is actually delivered.
3947 */
3948 if (!error && sig) {
3949 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3950 /*
3951 * If lock_task_sighand() failed we pretend the task
3952 * dies after receiving the signal. The window is tiny,
3953 * and the signal is private anyway.
3954 */
3955 if (unlikely(error == -ESRCH))
3956 error = 0;
3957 }
3958 }
3959 rcu_read_unlock();
3960
3961 return error;
3962}
3963
3964static int do_tkill(pid_t tgid, pid_t pid, int sig)
3965{
3966 struct kernel_siginfo info;
3967
3968 clear_siginfo(&info);
3969 info.si_signo = sig;
3970 info.si_errno = 0;
3971 info.si_code = SI_TKILL;
3972 info.si_pid = task_tgid_vnr(current);
3973 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3974
3975 return do_send_specific(tgid, pid, sig, &info);
3976}
3977
3978/**
3979 * sys_tgkill - send signal to one specific thread
3980 * @tgid: the thread group ID of the thread
3981 * @pid: the PID of the thread
3982 * @sig: signal to be sent
3983 *
3984 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3985 * exists but it's not belonging to the target process anymore. This
3986 * method solves the problem of threads exiting and PIDs getting reused.
3987 */
3988SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3989{
3990 /* This is only valid for single tasks */
3991 if (pid <= 0 || tgid <= 0)
3992 return -EINVAL;
3993
3994 return do_tkill(tgid, pid, sig);
3995}
3996
3997/**
3998 * sys_tkill - send signal to one specific task
3999 * @pid: the PID of the task
4000 * @sig: signal to be sent
4001 *
4002 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4003 */
4004SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4005{
4006 /* This is only valid for single tasks */
4007 if (pid <= 0)
4008 return -EINVAL;
4009
4010 return do_tkill(0, pid, sig);
4011}
4012
4013static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4014{
4015 /* Not even root can pretend to send signals from the kernel.
4016 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4017 */
4018 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4019 (task_pid_vnr(current) != pid))
4020 return -EPERM;
4021
4022 /* POSIX.1b doesn't mention process groups. */
4023 return kill_proc_info(sig, info, pid);
4024}
4025
4026/**
4027 * sys_rt_sigqueueinfo - send signal information to a signal
4028 * @pid: the PID of the thread
4029 * @sig: signal to be sent
4030 * @uinfo: signal info to be sent
4031 */
4032SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4033 siginfo_t __user *, uinfo)
4034{
4035 kernel_siginfo_t info;
4036 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4037 if (unlikely(ret))
4038 return ret;
4039 return do_rt_sigqueueinfo(pid, sig, &info);
4040}
4041
4042#ifdef CONFIG_COMPAT
4043COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4044 compat_pid_t, pid,
4045 int, sig,
4046 struct compat_siginfo __user *, uinfo)
4047{
4048 kernel_siginfo_t info;
4049 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4050 if (unlikely(ret))
4051 return ret;
4052 return do_rt_sigqueueinfo(pid, sig, &info);
4053}
4054#endif
4055
4056static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4057{
4058 /* This is only valid for single tasks */
4059 if (pid <= 0 || tgid <= 0)
4060 return -EINVAL;
4061
4062 /* Not even root can pretend to send signals from the kernel.
4063 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4064 */
4065 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4066 (task_pid_vnr(current) != pid))
4067 return -EPERM;
4068
4069 return do_send_specific(tgid, pid, sig, info);
4070}
4071
4072SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4073 siginfo_t __user *, uinfo)
4074{
4075 kernel_siginfo_t info;
4076 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4077 if (unlikely(ret))
4078 return ret;
4079 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4080}
4081
4082#ifdef CONFIG_COMPAT
4083COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4084 compat_pid_t, tgid,
4085 compat_pid_t, pid,
4086 int, sig,
4087 struct compat_siginfo __user *, uinfo)
4088{
4089 kernel_siginfo_t info;
4090 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4091 if (unlikely(ret))
4092 return ret;
4093 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4094}
4095#endif
4096
4097/*
4098 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4099 */
4100void kernel_sigaction(int sig, __sighandler_t action)
4101{
4102 spin_lock_irq(¤t->sighand->siglock);
4103 current->sighand->action[sig - 1].sa.sa_handler = action;
4104 if (action == SIG_IGN) {
4105 sigset_t mask;
4106
4107 sigemptyset(&mask);
4108 sigaddset(&mask, sig);
4109
4110 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
4111 flush_sigqueue_mask(&mask, ¤t->pending);
4112 recalc_sigpending();
4113 }
4114 spin_unlock_irq(¤t->sighand->siglock);
4115}
4116EXPORT_SYMBOL(kernel_sigaction);
4117
4118void __weak sigaction_compat_abi(struct k_sigaction *act,
4119 struct k_sigaction *oact)
4120{
4121}
4122
4123int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4124{
4125 struct task_struct *p = current, *t;
4126 struct k_sigaction *k;
4127 sigset_t mask;
4128
4129 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4130 return -EINVAL;
4131
4132 k = &p->sighand->action[sig-1];
4133
4134 spin_lock_irq(&p->sighand->siglock);
4135 if (k->sa.sa_flags & SA_IMMUTABLE) {
4136 spin_unlock_irq(&p->sighand->siglock);
4137 return -EINVAL;
4138 }
4139 if (oact)
4140 *oact = *k;
4141
4142 /*
4143 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4144 * e.g. by having an architecture use the bit in their uapi.
4145 */
4146 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4147
4148 /*
4149 * Clear unknown flag bits in order to allow userspace to detect missing
4150 * support for flag bits and to allow the kernel to use non-uapi bits
4151 * internally.
4152 */
4153 if (act)
4154 act->sa.sa_flags &= UAPI_SA_FLAGS;
4155 if (oact)
4156 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4157
4158 sigaction_compat_abi(act, oact);
4159
4160 if (act) {
4161 sigdelsetmask(&act->sa.sa_mask,
4162 sigmask(SIGKILL) | sigmask(SIGSTOP));
4163 *k = *act;
4164 /*
4165 * POSIX 3.3.1.3:
4166 * "Setting a signal action to SIG_IGN for a signal that is
4167 * pending shall cause the pending signal to be discarded,
4168 * whether or not it is blocked."
4169 *
4170 * "Setting a signal action to SIG_DFL for a signal that is
4171 * pending and whose default action is to ignore the signal
4172 * (for example, SIGCHLD), shall cause the pending signal to
4173 * be discarded, whether or not it is blocked"
4174 */
4175 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4176 sigemptyset(&mask);
4177 sigaddset(&mask, sig);
4178 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4179 for_each_thread(p, t)
4180 flush_sigqueue_mask(&mask, &t->pending);
4181 }
4182 }
4183
4184 spin_unlock_irq(&p->sighand->siglock);
4185 return 0;
4186}
4187
4188#ifdef CONFIG_DYNAMIC_SIGFRAME
4189static inline void sigaltstack_lock(void)
4190 __acquires(¤t->sighand->siglock)
4191{
4192 spin_lock_irq(¤t->sighand->siglock);
4193}
4194
4195static inline void sigaltstack_unlock(void)
4196 __releases(¤t->sighand->siglock)
4197{
4198 spin_unlock_irq(¤t->sighand->siglock);
4199}
4200#else
4201static inline void sigaltstack_lock(void) { }
4202static inline void sigaltstack_unlock(void) { }
4203#endif
4204
4205static int
4206do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4207 size_t min_ss_size)
4208{
4209 struct task_struct *t = current;
4210 int ret = 0;
4211
4212 if (oss) {
4213 memset(oss, 0, sizeof(stack_t));
4214 oss->ss_sp = (void __user *) t->sas_ss_sp;
4215 oss->ss_size = t->sas_ss_size;
4216 oss->ss_flags = sas_ss_flags(sp) |
4217 (current->sas_ss_flags & SS_FLAG_BITS);
4218 }
4219
4220 if (ss) {
4221 void __user *ss_sp = ss->ss_sp;
4222 size_t ss_size = ss->ss_size;
4223 unsigned ss_flags = ss->ss_flags;
4224 int ss_mode;
4225
4226 if (unlikely(on_sig_stack(sp)))
4227 return -EPERM;
4228
4229 ss_mode = ss_flags & ~SS_FLAG_BITS;
4230 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4231 ss_mode != 0))
4232 return -EINVAL;
4233
4234 /*
4235 * Return before taking any locks if no actual
4236 * sigaltstack changes were requested.
4237 */
4238 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4239 t->sas_ss_size == ss_size &&
4240 t->sas_ss_flags == ss_flags)
4241 return 0;
4242
4243 sigaltstack_lock();
4244 if (ss_mode == SS_DISABLE) {
4245 ss_size = 0;
4246 ss_sp = NULL;
4247 } else {
4248 if (unlikely(ss_size < min_ss_size))
4249 ret = -ENOMEM;
4250 if (!sigaltstack_size_valid(ss_size))
4251 ret = -ENOMEM;
4252 }
4253 if (!ret) {
4254 t->sas_ss_sp = (unsigned long) ss_sp;
4255 t->sas_ss_size = ss_size;
4256 t->sas_ss_flags = ss_flags;
4257 }
4258 sigaltstack_unlock();
4259 }
4260 return ret;
4261}
4262
4263SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4264{
4265 stack_t new, old;
4266 int err;
4267 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4268 return -EFAULT;
4269 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4270 current_user_stack_pointer(),
4271 MINSIGSTKSZ);
4272 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4273 err = -EFAULT;
4274 return err;
4275}
4276
4277int restore_altstack(const stack_t __user *uss)
4278{
4279 stack_t new;
4280 if (copy_from_user(&new, uss, sizeof(stack_t)))
4281 return -EFAULT;
4282 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4283 MINSIGSTKSZ);
4284 /* squash all but EFAULT for now */
4285 return 0;
4286}
4287
4288int __save_altstack(stack_t __user *uss, unsigned long sp)
4289{
4290 struct task_struct *t = current;
4291 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4292 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4293 __put_user(t->sas_ss_size, &uss->ss_size);
4294 return err;
4295}
4296
4297#ifdef CONFIG_COMPAT
4298static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4299 compat_stack_t __user *uoss_ptr)
4300{
4301 stack_t uss, uoss;
4302 int ret;
4303
4304 if (uss_ptr) {
4305 compat_stack_t uss32;
4306 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4307 return -EFAULT;
4308 uss.ss_sp = compat_ptr(uss32.ss_sp);
4309 uss.ss_flags = uss32.ss_flags;
4310 uss.ss_size = uss32.ss_size;
4311 }
4312 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4313 compat_user_stack_pointer(),
4314 COMPAT_MINSIGSTKSZ);
4315 if (ret >= 0 && uoss_ptr) {
4316 compat_stack_t old;
4317 memset(&old, 0, sizeof(old));
4318 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4319 old.ss_flags = uoss.ss_flags;
4320 old.ss_size = uoss.ss_size;
4321 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4322 ret = -EFAULT;
4323 }
4324 return ret;
4325}
4326
4327COMPAT_SYSCALL_DEFINE2(sigaltstack,
4328 const compat_stack_t __user *, uss_ptr,
4329 compat_stack_t __user *, uoss_ptr)
4330{
4331 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4332}
4333
4334int compat_restore_altstack(const compat_stack_t __user *uss)
4335{
4336 int err = do_compat_sigaltstack(uss, NULL);
4337 /* squash all but -EFAULT for now */
4338 return err == -EFAULT ? err : 0;
4339}
4340
4341int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4342{
4343 int err;
4344 struct task_struct *t = current;
4345 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4346 &uss->ss_sp) |
4347 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4348 __put_user(t->sas_ss_size, &uss->ss_size);
4349 return err;
4350}
4351#endif
4352
4353#ifdef __ARCH_WANT_SYS_SIGPENDING
4354
4355/**
4356 * sys_sigpending - examine pending signals
4357 * @uset: where mask of pending signal is returned
4358 */
4359SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4360{
4361 sigset_t set;
4362
4363 if (sizeof(old_sigset_t) > sizeof(*uset))
4364 return -EINVAL;
4365
4366 do_sigpending(&set);
4367
4368 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4369 return -EFAULT;
4370
4371 return 0;
4372}
4373
4374#ifdef CONFIG_COMPAT
4375COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4376{
4377 sigset_t set;
4378
4379 do_sigpending(&set);
4380
4381 return put_user(set.sig[0], set32);
4382}
4383#endif
4384
4385#endif
4386
4387#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4388/**
4389 * sys_sigprocmask - examine and change blocked signals
4390 * @how: whether to add, remove, or set signals
4391 * @nset: signals to add or remove (if non-null)
4392 * @oset: previous value of signal mask if non-null
4393 *
4394 * Some platforms have their own version with special arguments;
4395 * others support only sys_rt_sigprocmask.
4396 */
4397
4398SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4399 old_sigset_t __user *, oset)
4400{
4401 old_sigset_t old_set, new_set;
4402 sigset_t new_blocked;
4403
4404 old_set = current->blocked.sig[0];
4405
4406 if (nset) {
4407 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4408 return -EFAULT;
4409
4410 new_blocked = current->blocked;
4411
4412 switch (how) {
4413 case SIG_BLOCK:
4414 sigaddsetmask(&new_blocked, new_set);
4415 break;
4416 case SIG_UNBLOCK:
4417 sigdelsetmask(&new_blocked, new_set);
4418 break;
4419 case SIG_SETMASK:
4420 new_blocked.sig[0] = new_set;
4421 break;
4422 default:
4423 return -EINVAL;
4424 }
4425
4426 set_current_blocked(&new_blocked);
4427 }
4428
4429 if (oset) {
4430 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4431 return -EFAULT;
4432 }
4433
4434 return 0;
4435}
4436#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4437
4438#ifndef CONFIG_ODD_RT_SIGACTION
4439/**
4440 * sys_rt_sigaction - alter an action taken by a process
4441 * @sig: signal to be sent
4442 * @act: new sigaction
4443 * @oact: used to save the previous sigaction
4444 * @sigsetsize: size of sigset_t type
4445 */
4446SYSCALL_DEFINE4(rt_sigaction, int, sig,
4447 const struct sigaction __user *, act,
4448 struct sigaction __user *, oact,
4449 size_t, sigsetsize)
4450{
4451 struct k_sigaction new_sa, old_sa;
4452 int ret;
4453
4454 /* XXX: Don't preclude handling different sized sigset_t's. */
4455 if (sigsetsize != sizeof(sigset_t))
4456 return -EINVAL;
4457
4458 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4459 return -EFAULT;
4460
4461 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4462 if (ret)
4463 return ret;
4464
4465 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4466 return -EFAULT;
4467
4468 return 0;
4469}
4470#ifdef CONFIG_COMPAT
4471COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4472 const struct compat_sigaction __user *, act,
4473 struct compat_sigaction __user *, oact,
4474 compat_size_t, sigsetsize)
4475{
4476 struct k_sigaction new_ka, old_ka;
4477#ifdef __ARCH_HAS_SA_RESTORER
4478 compat_uptr_t restorer;
4479#endif
4480 int ret;
4481
4482 /* XXX: Don't preclude handling different sized sigset_t's. */
4483 if (sigsetsize != sizeof(compat_sigset_t))
4484 return -EINVAL;
4485
4486 if (act) {
4487 compat_uptr_t handler;
4488 ret = get_user(handler, &act->sa_handler);
4489 new_ka.sa.sa_handler = compat_ptr(handler);
4490#ifdef __ARCH_HAS_SA_RESTORER
4491 ret |= get_user(restorer, &act->sa_restorer);
4492 new_ka.sa.sa_restorer = compat_ptr(restorer);
4493#endif
4494 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4495 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4496 if (ret)
4497 return -EFAULT;
4498 }
4499
4500 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4501 if (!ret && oact) {
4502 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4503 &oact->sa_handler);
4504 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4505 sizeof(oact->sa_mask));
4506 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4507#ifdef __ARCH_HAS_SA_RESTORER
4508 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4509 &oact->sa_restorer);
4510#endif
4511 }
4512 return ret;
4513}
4514#endif
4515#endif /* !CONFIG_ODD_RT_SIGACTION */
4516
4517#ifdef CONFIG_OLD_SIGACTION
4518SYSCALL_DEFINE3(sigaction, int, sig,
4519 const struct old_sigaction __user *, act,
4520 struct old_sigaction __user *, oact)
4521{
4522 struct k_sigaction new_ka, old_ka;
4523 int ret;
4524
4525 if (act) {
4526 old_sigset_t mask;
4527 if (!access_ok(act, sizeof(*act)) ||
4528 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4529 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4530 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4531 __get_user(mask, &act->sa_mask))
4532 return -EFAULT;
4533#ifdef __ARCH_HAS_KA_RESTORER
4534 new_ka.ka_restorer = NULL;
4535#endif
4536 siginitset(&new_ka.sa.sa_mask, mask);
4537 }
4538
4539 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4540
4541 if (!ret && oact) {
4542 if (!access_ok(oact, sizeof(*oact)) ||
4543 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4544 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4545 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4546 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4547 return -EFAULT;
4548 }
4549
4550 return ret;
4551}
4552#endif
4553#ifdef CONFIG_COMPAT_OLD_SIGACTION
4554COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4555 const struct compat_old_sigaction __user *, act,
4556 struct compat_old_sigaction __user *, oact)
4557{
4558 struct k_sigaction new_ka, old_ka;
4559 int ret;
4560 compat_old_sigset_t mask;
4561 compat_uptr_t handler, restorer;
4562
4563 if (act) {
4564 if (!access_ok(act, sizeof(*act)) ||
4565 __get_user(handler, &act->sa_handler) ||
4566 __get_user(restorer, &act->sa_restorer) ||
4567 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4568 __get_user(mask, &act->sa_mask))
4569 return -EFAULT;
4570
4571#ifdef __ARCH_HAS_KA_RESTORER
4572 new_ka.ka_restorer = NULL;
4573#endif
4574 new_ka.sa.sa_handler = compat_ptr(handler);
4575 new_ka.sa.sa_restorer = compat_ptr(restorer);
4576 siginitset(&new_ka.sa.sa_mask, mask);
4577 }
4578
4579 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4580
4581 if (!ret && oact) {
4582 if (!access_ok(oact, sizeof(*oact)) ||
4583 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4584 &oact->sa_handler) ||
4585 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4586 &oact->sa_restorer) ||
4587 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4588 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4589 return -EFAULT;
4590 }
4591 return ret;
4592}
4593#endif
4594
4595#ifdef CONFIG_SGETMASK_SYSCALL
4596
4597/*
4598 * For backwards compatibility. Functionality superseded by sigprocmask.
4599 */
4600SYSCALL_DEFINE0(sgetmask)
4601{
4602 /* SMP safe */
4603 return current->blocked.sig[0];
4604}
4605
4606SYSCALL_DEFINE1(ssetmask, int, newmask)
4607{
4608 int old = current->blocked.sig[0];
4609 sigset_t newset;
4610
4611 siginitset(&newset, newmask);
4612 set_current_blocked(&newset);
4613
4614 return old;
4615}
4616#endif /* CONFIG_SGETMASK_SYSCALL */
4617
4618#ifdef __ARCH_WANT_SYS_SIGNAL
4619/*
4620 * For backwards compatibility. Functionality superseded by sigaction.
4621 */
4622SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4623{
4624 struct k_sigaction new_sa, old_sa;
4625 int ret;
4626
4627 new_sa.sa.sa_handler = handler;
4628 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4629 sigemptyset(&new_sa.sa.sa_mask);
4630
4631 ret = do_sigaction(sig, &new_sa, &old_sa);
4632
4633 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4634}
4635#endif /* __ARCH_WANT_SYS_SIGNAL */
4636
4637#ifdef __ARCH_WANT_SYS_PAUSE
4638
4639SYSCALL_DEFINE0(pause)
4640{
4641 while (!signal_pending(current)) {
4642 __set_current_state(TASK_INTERRUPTIBLE);
4643 schedule();
4644 }
4645 return -ERESTARTNOHAND;
4646}
4647
4648#endif
4649
4650static int sigsuspend(sigset_t *set)
4651{
4652 current->saved_sigmask = current->blocked;
4653 set_current_blocked(set);
4654
4655 while (!signal_pending(current)) {
4656 __set_current_state(TASK_INTERRUPTIBLE);
4657 schedule();
4658 }
4659 set_restore_sigmask();
4660 return -ERESTARTNOHAND;
4661}
4662
4663/**
4664 * sys_rt_sigsuspend - replace the signal mask for a value with the
4665 * @unewset value until a signal is received
4666 * @unewset: new signal mask value
4667 * @sigsetsize: size of sigset_t type
4668 */
4669SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4670{
4671 sigset_t newset;
4672
4673 /* XXX: Don't preclude handling different sized sigset_t's. */
4674 if (sigsetsize != sizeof(sigset_t))
4675 return -EINVAL;
4676
4677 if (copy_from_user(&newset, unewset, sizeof(newset)))
4678 return -EFAULT;
4679 return sigsuspend(&newset);
4680}
4681
4682#ifdef CONFIG_COMPAT
4683COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4684{
4685 sigset_t newset;
4686
4687 /* XXX: Don't preclude handling different sized sigset_t's. */
4688 if (sigsetsize != sizeof(sigset_t))
4689 return -EINVAL;
4690
4691 if (get_compat_sigset(&newset, unewset))
4692 return -EFAULT;
4693 return sigsuspend(&newset);
4694}
4695#endif
4696
4697#ifdef CONFIG_OLD_SIGSUSPEND
4698SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4699{
4700 sigset_t blocked;
4701 siginitset(&blocked, mask);
4702 return sigsuspend(&blocked);
4703}
4704#endif
4705#ifdef CONFIG_OLD_SIGSUSPEND3
4706SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4707{
4708 sigset_t blocked;
4709 siginitset(&blocked, mask);
4710 return sigsuspend(&blocked);
4711}
4712#endif
4713
4714__weak const char *arch_vma_name(struct vm_area_struct *vma)
4715{
4716 return NULL;
4717}
4718
4719static inline void siginfo_buildtime_checks(void)
4720{
4721 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4722
4723 /* Verify the offsets in the two siginfos match */
4724#define CHECK_OFFSET(field) \
4725 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4726
4727 /* kill */
4728 CHECK_OFFSET(si_pid);
4729 CHECK_OFFSET(si_uid);
4730
4731 /* timer */
4732 CHECK_OFFSET(si_tid);
4733 CHECK_OFFSET(si_overrun);
4734 CHECK_OFFSET(si_value);
4735
4736 /* rt */
4737 CHECK_OFFSET(si_pid);
4738 CHECK_OFFSET(si_uid);
4739 CHECK_OFFSET(si_value);
4740
4741 /* sigchld */
4742 CHECK_OFFSET(si_pid);
4743 CHECK_OFFSET(si_uid);
4744 CHECK_OFFSET(si_status);
4745 CHECK_OFFSET(si_utime);
4746 CHECK_OFFSET(si_stime);
4747
4748 /* sigfault */
4749 CHECK_OFFSET(si_addr);
4750 CHECK_OFFSET(si_trapno);
4751 CHECK_OFFSET(si_addr_lsb);
4752 CHECK_OFFSET(si_lower);
4753 CHECK_OFFSET(si_upper);
4754 CHECK_OFFSET(si_pkey);
4755 CHECK_OFFSET(si_perf_data);
4756 CHECK_OFFSET(si_perf_type);
4757 CHECK_OFFSET(si_perf_flags);
4758
4759 /* sigpoll */
4760 CHECK_OFFSET(si_band);
4761 CHECK_OFFSET(si_fd);
4762
4763 /* sigsys */
4764 CHECK_OFFSET(si_call_addr);
4765 CHECK_OFFSET(si_syscall);
4766 CHECK_OFFSET(si_arch);
4767#undef CHECK_OFFSET
4768
4769 /* usb asyncio */
4770 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4771 offsetof(struct siginfo, si_addr));
4772 if (sizeof(int) == sizeof(void __user *)) {
4773 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4774 sizeof(void __user *));
4775 } else {
4776 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4777 sizeof_field(struct siginfo, si_uid)) !=
4778 sizeof(void __user *));
4779 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4780 offsetof(struct siginfo, si_uid));
4781 }
4782#ifdef CONFIG_COMPAT
4783 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4784 offsetof(struct compat_siginfo, si_addr));
4785 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4786 sizeof(compat_uptr_t));
4787 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4788 sizeof_field(struct siginfo, si_pid));
4789#endif
4790}
4791
4792#if defined(CONFIG_SYSCTL)
4793static struct ctl_table signal_debug_table[] = {
4794#ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4795 {
4796 .procname = "exception-trace",
4797 .data = &show_unhandled_signals,
4798 .maxlen = sizeof(int),
4799 .mode = 0644,
4800 .proc_handler = proc_dointvec
4801 },
4802#endif
4803 { }
4804};
4805
4806static int __init init_signal_sysctls(void)
4807{
4808 register_sysctl_init("debug", signal_debug_table);
4809 return 0;
4810}
4811early_initcall(init_signal_sysctls);
4812#endif /* CONFIG_SYSCTL */
4813
4814void __init signals_init(void)
4815{
4816 siginfo_buildtime_checks();
4817
4818 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4819}
4820
4821#ifdef CONFIG_KGDB_KDB
4822#include <linux/kdb.h>
4823/*
4824 * kdb_send_sig - Allows kdb to send signals without exposing
4825 * signal internals. This function checks if the required locks are
4826 * available before calling the main signal code, to avoid kdb
4827 * deadlocks.
4828 */
4829void kdb_send_sig(struct task_struct *t, int sig)
4830{
4831 static struct task_struct *kdb_prev_t;
4832 int new_t, ret;
4833 if (!spin_trylock(&t->sighand->siglock)) {
4834 kdb_printf("Can't do kill command now.\n"
4835 "The sigmask lock is held somewhere else in "
4836 "kernel, try again later\n");
4837 return;
4838 }
4839 new_t = kdb_prev_t != t;
4840 kdb_prev_t = t;
4841 if (!task_is_running(t) && new_t) {
4842 spin_unlock(&t->sighand->siglock);
4843 kdb_printf("Process is not RUNNING, sending a signal from "
4844 "kdb risks deadlock\n"
4845 "on the run queue locks. "
4846 "The signal has _not_ been sent.\n"
4847 "Reissue the kill command if you want to risk "
4848 "the deadlock.\n");
4849 return;
4850 }
4851 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4852 spin_unlock(&t->sighand->siglock);
4853 if (ret)
4854 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4855 sig, t->pid);
4856 else
4857 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4858}
4859#endif /* CONFIG_KGDB_KDB */