Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/signal.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 *
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
12 */
13
14#include <linux/slab.h>
15#include <linux/export.h>
16#include <linux/init.h>
17#include <linux/sched/mm.h>
18#include <linux/sched/user.h>
19#include <linux/sched/debug.h>
20#include <linux/sched/task.h>
21#include <linux/sched/task_stack.h>
22#include <linux/sched/cputime.h>
23#include <linux/file.h>
24#include <linux/fs.h>
25#include <linux/proc_fs.h>
26#include <linux/tty.h>
27#include <linux/binfmts.h>
28#include <linux/coredump.h>
29#include <linux/security.h>
30#include <linux/syscalls.h>
31#include <linux/ptrace.h>
32#include <linux/signal.h>
33#include <linux/signalfd.h>
34#include <linux/ratelimit.h>
35#include <linux/tracehook.h>
36#include <linux/capability.h>
37#include <linux/freezer.h>
38#include <linux/pid_namespace.h>
39#include <linux/nsproxy.h>
40#include <linux/user_namespace.h>
41#include <linux/uprobes.h>
42#include <linux/compat.h>
43#include <linux/cn_proc.h>
44#include <linux/compiler.h>
45#include <linux/posix-timers.h>
46#include <linux/livepatch.h>
47#include <linux/cgroup.h>
48#include <linux/audit.h>
49
50#define CREATE_TRACE_POINTS
51#include <trace/events/signal.h>
52
53#include <asm/param.h>
54#include <linux/uaccess.h>
55#include <asm/unistd.h>
56#include <asm/siginfo.h>
57#include <asm/cacheflush.h>
58
59/*
60 * SLAB caches for signal bits.
61 */
62
63static struct kmem_cache *sigqueue_cachep;
64
65int print_fatal_signals __read_mostly;
66
67static void __user *sig_handler(struct task_struct *t, int sig)
68{
69 return t->sighand->action[sig - 1].sa.sa_handler;
70}
71
72static inline bool sig_handler_ignored(void __user *handler, int sig)
73{
74 /* Is it explicitly or implicitly ignored? */
75 return handler == SIG_IGN ||
76 (handler == SIG_DFL && sig_kernel_ignore(sig));
77}
78
79static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
80{
81 void __user *handler;
82
83 handler = sig_handler(t, sig);
84
85 /* SIGKILL and SIGSTOP may not be sent to the global init */
86 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
87 return true;
88
89 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
90 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
91 return true;
92
93 /* Only allow kernel generated signals to this kthread */
94 if (unlikely((t->flags & PF_KTHREAD) &&
95 (handler == SIG_KTHREAD_KERNEL) && !force))
96 return true;
97
98 return sig_handler_ignored(handler, sig);
99}
100
101static bool sig_ignored(struct task_struct *t, int sig, bool force)
102{
103 /*
104 * Blocked signals are never ignored, since the
105 * signal handler may change by the time it is
106 * unblocked.
107 */
108 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
109 return false;
110
111 /*
112 * Tracers may want to know about even ignored signal unless it
113 * is SIGKILL which can't be reported anyway but can be ignored
114 * by SIGNAL_UNKILLABLE task.
115 */
116 if (t->ptrace && sig != SIGKILL)
117 return false;
118
119 return sig_task_ignored(t, sig, force);
120}
121
122/*
123 * Re-calculate pending state from the set of locally pending
124 * signals, globally pending signals, and blocked signals.
125 */
126static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
127{
128 unsigned long ready;
129 long i;
130
131 switch (_NSIG_WORDS) {
132 default:
133 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
134 ready |= signal->sig[i] &~ blocked->sig[i];
135 break;
136
137 case 4: ready = signal->sig[3] &~ blocked->sig[3];
138 ready |= signal->sig[2] &~ blocked->sig[2];
139 ready |= signal->sig[1] &~ blocked->sig[1];
140 ready |= signal->sig[0] &~ blocked->sig[0];
141 break;
142
143 case 2: ready = signal->sig[1] &~ blocked->sig[1];
144 ready |= signal->sig[0] &~ blocked->sig[0];
145 break;
146
147 case 1: ready = signal->sig[0] &~ blocked->sig[0];
148 }
149 return ready != 0;
150}
151
152#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
153
154static bool recalc_sigpending_tsk(struct task_struct *t)
155{
156 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
157 PENDING(&t->pending, &t->blocked) ||
158 PENDING(&t->signal->shared_pending, &t->blocked) ||
159 cgroup_task_frozen(t)) {
160 set_tsk_thread_flag(t, TIF_SIGPENDING);
161 return true;
162 }
163
164 /*
165 * We must never clear the flag in another thread, or in current
166 * when it's possible the current syscall is returning -ERESTART*.
167 * So we don't clear it here, and only callers who know they should do.
168 */
169 return false;
170}
171
172/*
173 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
174 * This is superfluous when called on current, the wakeup is a harmless no-op.
175 */
176void recalc_sigpending_and_wake(struct task_struct *t)
177{
178 if (recalc_sigpending_tsk(t))
179 signal_wake_up(t, 0);
180}
181
182void recalc_sigpending(void)
183{
184 if (!recalc_sigpending_tsk(current) && !freezing(current) &&
185 !klp_patch_pending(current))
186 clear_thread_flag(TIF_SIGPENDING);
187
188}
189EXPORT_SYMBOL(recalc_sigpending);
190
191void calculate_sigpending(void)
192{
193 /* Have any signals or users of TIF_SIGPENDING been delayed
194 * until after fork?
195 */
196 spin_lock_irq(¤t->sighand->siglock);
197 set_tsk_thread_flag(current, TIF_SIGPENDING);
198 recalc_sigpending();
199 spin_unlock_irq(¤t->sighand->siglock);
200}
201
202/* Given the mask, find the first available signal that should be serviced. */
203
204#define SYNCHRONOUS_MASK \
205 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
206 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
207
208int next_signal(struct sigpending *pending, sigset_t *mask)
209{
210 unsigned long i, *s, *m, x;
211 int sig = 0;
212
213 s = pending->signal.sig;
214 m = mask->sig;
215
216 /*
217 * Handle the first word specially: it contains the
218 * synchronous signals that need to be dequeued first.
219 */
220 x = *s &~ *m;
221 if (x) {
222 if (x & SYNCHRONOUS_MASK)
223 x &= SYNCHRONOUS_MASK;
224 sig = ffz(~x) + 1;
225 return sig;
226 }
227
228 switch (_NSIG_WORDS) {
229 default:
230 for (i = 1; i < _NSIG_WORDS; ++i) {
231 x = *++s &~ *++m;
232 if (!x)
233 continue;
234 sig = ffz(~x) + i*_NSIG_BPW + 1;
235 break;
236 }
237 break;
238
239 case 2:
240 x = s[1] &~ m[1];
241 if (!x)
242 break;
243 sig = ffz(~x) + _NSIG_BPW + 1;
244 break;
245
246 case 1:
247 /* Nothing to do */
248 break;
249 }
250
251 return sig;
252}
253
254static inline void print_dropped_signal(int sig)
255{
256 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
257
258 if (!print_fatal_signals)
259 return;
260
261 if (!__ratelimit(&ratelimit_state))
262 return;
263
264 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
265 current->comm, current->pid, sig);
266}
267
268/**
269 * task_set_jobctl_pending - set jobctl pending bits
270 * @task: target task
271 * @mask: pending bits to set
272 *
273 * Clear @mask from @task->jobctl. @mask must be subset of
274 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
275 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
276 * cleared. If @task is already being killed or exiting, this function
277 * becomes noop.
278 *
279 * CONTEXT:
280 * Must be called with @task->sighand->siglock held.
281 *
282 * RETURNS:
283 * %true if @mask is set, %false if made noop because @task was dying.
284 */
285bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
286{
287 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
288 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
289 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
290
291 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
292 return false;
293
294 if (mask & JOBCTL_STOP_SIGMASK)
295 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
296
297 task->jobctl |= mask;
298 return true;
299}
300
301/**
302 * task_clear_jobctl_trapping - clear jobctl trapping bit
303 * @task: target task
304 *
305 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
306 * Clear it and wake up the ptracer. Note that we don't need any further
307 * locking. @task->siglock guarantees that @task->parent points to the
308 * ptracer.
309 *
310 * CONTEXT:
311 * Must be called with @task->sighand->siglock held.
312 */
313void task_clear_jobctl_trapping(struct task_struct *task)
314{
315 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
316 task->jobctl &= ~JOBCTL_TRAPPING;
317 smp_mb(); /* advised by wake_up_bit() */
318 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
319 }
320}
321
322/**
323 * task_clear_jobctl_pending - clear jobctl pending bits
324 * @task: target task
325 * @mask: pending bits to clear
326 *
327 * Clear @mask from @task->jobctl. @mask must be subset of
328 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
329 * STOP bits are cleared together.
330 *
331 * If clearing of @mask leaves no stop or trap pending, this function calls
332 * task_clear_jobctl_trapping().
333 *
334 * CONTEXT:
335 * Must be called with @task->sighand->siglock held.
336 */
337void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
338{
339 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
340
341 if (mask & JOBCTL_STOP_PENDING)
342 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
343
344 task->jobctl &= ~mask;
345
346 if (!(task->jobctl & JOBCTL_PENDING_MASK))
347 task_clear_jobctl_trapping(task);
348}
349
350/**
351 * task_participate_group_stop - participate in a group stop
352 * @task: task participating in a group stop
353 *
354 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
355 * Group stop states are cleared and the group stop count is consumed if
356 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
357 * stop, the appropriate `SIGNAL_*` flags are set.
358 *
359 * CONTEXT:
360 * Must be called with @task->sighand->siglock held.
361 *
362 * RETURNS:
363 * %true if group stop completion should be notified to the parent, %false
364 * otherwise.
365 */
366static bool task_participate_group_stop(struct task_struct *task)
367{
368 struct signal_struct *sig = task->signal;
369 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
370
371 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
372
373 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
374
375 if (!consume)
376 return false;
377
378 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
379 sig->group_stop_count--;
380
381 /*
382 * Tell the caller to notify completion iff we are entering into a
383 * fresh group stop. Read comment in do_signal_stop() for details.
384 */
385 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
386 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
387 return true;
388 }
389 return false;
390}
391
392void task_join_group_stop(struct task_struct *task)
393{
394 /* Have the new thread join an on-going signal group stop */
395 unsigned long jobctl = current->jobctl;
396 if (jobctl & JOBCTL_STOP_PENDING) {
397 struct signal_struct *sig = current->signal;
398 unsigned long signr = jobctl & JOBCTL_STOP_SIGMASK;
399 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
400 if (task_set_jobctl_pending(task, signr | gstop)) {
401 sig->group_stop_count++;
402 }
403 }
404}
405
406/*
407 * allocate a new signal queue record
408 * - this may be called without locks if and only if t == current, otherwise an
409 * appropriate lock must be held to stop the target task from exiting
410 */
411static struct sigqueue *
412__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
413{
414 struct sigqueue *q = NULL;
415 struct user_struct *user;
416
417 /*
418 * Protect access to @t credentials. This can go away when all
419 * callers hold rcu read lock.
420 */
421 rcu_read_lock();
422 user = get_uid(__task_cred(t)->user);
423 atomic_inc(&user->sigpending);
424 rcu_read_unlock();
425
426 if (override_rlimit ||
427 atomic_read(&user->sigpending) <=
428 task_rlimit(t, RLIMIT_SIGPENDING)) {
429 q = kmem_cache_alloc(sigqueue_cachep, flags);
430 } else {
431 print_dropped_signal(sig);
432 }
433
434 if (unlikely(q == NULL)) {
435 atomic_dec(&user->sigpending);
436 free_uid(user);
437 } else {
438 INIT_LIST_HEAD(&q->list);
439 q->flags = 0;
440 q->user = user;
441 }
442
443 return q;
444}
445
446static void __sigqueue_free(struct sigqueue *q)
447{
448 if (q->flags & SIGQUEUE_PREALLOC)
449 return;
450 atomic_dec(&q->user->sigpending);
451 free_uid(q->user);
452 kmem_cache_free(sigqueue_cachep, q);
453}
454
455void flush_sigqueue(struct sigpending *queue)
456{
457 struct sigqueue *q;
458
459 sigemptyset(&queue->signal);
460 while (!list_empty(&queue->list)) {
461 q = list_entry(queue->list.next, struct sigqueue , list);
462 list_del_init(&q->list);
463 __sigqueue_free(q);
464 }
465}
466
467/*
468 * Flush all pending signals for this kthread.
469 */
470void flush_signals(struct task_struct *t)
471{
472 unsigned long flags;
473
474 spin_lock_irqsave(&t->sighand->siglock, flags);
475 clear_tsk_thread_flag(t, TIF_SIGPENDING);
476 flush_sigqueue(&t->pending);
477 flush_sigqueue(&t->signal->shared_pending);
478 spin_unlock_irqrestore(&t->sighand->siglock, flags);
479}
480EXPORT_SYMBOL(flush_signals);
481
482#ifdef CONFIG_POSIX_TIMERS
483static void __flush_itimer_signals(struct sigpending *pending)
484{
485 sigset_t signal, retain;
486 struct sigqueue *q, *n;
487
488 signal = pending->signal;
489 sigemptyset(&retain);
490
491 list_for_each_entry_safe(q, n, &pending->list, list) {
492 int sig = q->info.si_signo;
493
494 if (likely(q->info.si_code != SI_TIMER)) {
495 sigaddset(&retain, sig);
496 } else {
497 sigdelset(&signal, sig);
498 list_del_init(&q->list);
499 __sigqueue_free(q);
500 }
501 }
502
503 sigorsets(&pending->signal, &signal, &retain);
504}
505
506void flush_itimer_signals(void)
507{
508 struct task_struct *tsk = current;
509 unsigned long flags;
510
511 spin_lock_irqsave(&tsk->sighand->siglock, flags);
512 __flush_itimer_signals(&tsk->pending);
513 __flush_itimer_signals(&tsk->signal->shared_pending);
514 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
515}
516#endif
517
518void ignore_signals(struct task_struct *t)
519{
520 int i;
521
522 for (i = 0; i < _NSIG; ++i)
523 t->sighand->action[i].sa.sa_handler = SIG_IGN;
524
525 flush_signals(t);
526}
527
528/*
529 * Flush all handlers for a task.
530 */
531
532void
533flush_signal_handlers(struct task_struct *t, int force_default)
534{
535 int i;
536 struct k_sigaction *ka = &t->sighand->action[0];
537 for (i = _NSIG ; i != 0 ; i--) {
538 if (force_default || ka->sa.sa_handler != SIG_IGN)
539 ka->sa.sa_handler = SIG_DFL;
540 ka->sa.sa_flags = 0;
541#ifdef __ARCH_HAS_SA_RESTORER
542 ka->sa.sa_restorer = NULL;
543#endif
544 sigemptyset(&ka->sa.sa_mask);
545 ka++;
546 }
547}
548
549bool unhandled_signal(struct task_struct *tsk, int sig)
550{
551 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
552 if (is_global_init(tsk))
553 return true;
554
555 if (handler != SIG_IGN && handler != SIG_DFL)
556 return false;
557
558 /* if ptraced, let the tracer determine */
559 return !tsk->ptrace;
560}
561
562static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
563 bool *resched_timer)
564{
565 struct sigqueue *q, *first = NULL;
566
567 /*
568 * Collect the siginfo appropriate to this signal. Check if
569 * there is another siginfo for the same signal.
570 */
571 list_for_each_entry(q, &list->list, list) {
572 if (q->info.si_signo == sig) {
573 if (first)
574 goto still_pending;
575 first = q;
576 }
577 }
578
579 sigdelset(&list->signal, sig);
580
581 if (first) {
582still_pending:
583 list_del_init(&first->list);
584 copy_siginfo(info, &first->info);
585
586 *resched_timer =
587 (first->flags & SIGQUEUE_PREALLOC) &&
588 (info->si_code == SI_TIMER) &&
589 (info->si_sys_private);
590
591 __sigqueue_free(first);
592 } else {
593 /*
594 * Ok, it wasn't in the queue. This must be
595 * a fast-pathed signal or we must have been
596 * out of queue space. So zero out the info.
597 */
598 clear_siginfo(info);
599 info->si_signo = sig;
600 info->si_errno = 0;
601 info->si_code = SI_USER;
602 info->si_pid = 0;
603 info->si_uid = 0;
604 }
605}
606
607static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
608 kernel_siginfo_t *info, bool *resched_timer)
609{
610 int sig = next_signal(pending, mask);
611
612 if (sig)
613 collect_signal(sig, pending, info, resched_timer);
614 return sig;
615}
616
617/*
618 * Dequeue a signal and return the element to the caller, which is
619 * expected to free it.
620 *
621 * All callers have to hold the siglock.
622 */
623int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
624{
625 bool resched_timer = false;
626 int signr;
627
628 /* We only dequeue private signals from ourselves, we don't let
629 * signalfd steal them
630 */
631 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
632 if (!signr) {
633 signr = __dequeue_signal(&tsk->signal->shared_pending,
634 mask, info, &resched_timer);
635#ifdef CONFIG_POSIX_TIMERS
636 /*
637 * itimer signal ?
638 *
639 * itimers are process shared and we restart periodic
640 * itimers in the signal delivery path to prevent DoS
641 * attacks in the high resolution timer case. This is
642 * compliant with the old way of self-restarting
643 * itimers, as the SIGALRM is a legacy signal and only
644 * queued once. Changing the restart behaviour to
645 * restart the timer in the signal dequeue path is
646 * reducing the timer noise on heavy loaded !highres
647 * systems too.
648 */
649 if (unlikely(signr == SIGALRM)) {
650 struct hrtimer *tmr = &tsk->signal->real_timer;
651
652 if (!hrtimer_is_queued(tmr) &&
653 tsk->signal->it_real_incr != 0) {
654 hrtimer_forward(tmr, tmr->base->get_time(),
655 tsk->signal->it_real_incr);
656 hrtimer_restart(tmr);
657 }
658 }
659#endif
660 }
661
662 recalc_sigpending();
663 if (!signr)
664 return 0;
665
666 if (unlikely(sig_kernel_stop(signr))) {
667 /*
668 * Set a marker that we have dequeued a stop signal. Our
669 * caller might release the siglock and then the pending
670 * stop signal it is about to process is no longer in the
671 * pending bitmasks, but must still be cleared by a SIGCONT
672 * (and overruled by a SIGKILL). So those cases clear this
673 * shared flag after we've set it. Note that this flag may
674 * remain set after the signal we return is ignored or
675 * handled. That doesn't matter because its only purpose
676 * is to alert stop-signal processing code when another
677 * processor has come along and cleared the flag.
678 */
679 current->jobctl |= JOBCTL_STOP_DEQUEUED;
680 }
681#ifdef CONFIG_POSIX_TIMERS
682 if (resched_timer) {
683 /*
684 * Release the siglock to ensure proper locking order
685 * of timer locks outside of siglocks. Note, we leave
686 * irqs disabled here, since the posix-timers code is
687 * about to disable them again anyway.
688 */
689 spin_unlock(&tsk->sighand->siglock);
690 posixtimer_rearm(info);
691 spin_lock(&tsk->sighand->siglock);
692
693 /* Don't expose the si_sys_private value to userspace */
694 info->si_sys_private = 0;
695 }
696#endif
697 return signr;
698}
699EXPORT_SYMBOL_GPL(dequeue_signal);
700
701static int dequeue_synchronous_signal(kernel_siginfo_t *info)
702{
703 struct task_struct *tsk = current;
704 struct sigpending *pending = &tsk->pending;
705 struct sigqueue *q, *sync = NULL;
706
707 /*
708 * Might a synchronous signal be in the queue?
709 */
710 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
711 return 0;
712
713 /*
714 * Return the first synchronous signal in the queue.
715 */
716 list_for_each_entry(q, &pending->list, list) {
717 /* Synchronous signals have a postive si_code */
718 if ((q->info.si_code > SI_USER) &&
719 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
720 sync = q;
721 goto next;
722 }
723 }
724 return 0;
725next:
726 /*
727 * Check if there is another siginfo for the same signal.
728 */
729 list_for_each_entry_continue(q, &pending->list, list) {
730 if (q->info.si_signo == sync->info.si_signo)
731 goto still_pending;
732 }
733
734 sigdelset(&pending->signal, sync->info.si_signo);
735 recalc_sigpending();
736still_pending:
737 list_del_init(&sync->list);
738 copy_siginfo(info, &sync->info);
739 __sigqueue_free(sync);
740 return info->si_signo;
741}
742
743/*
744 * Tell a process that it has a new active signal..
745 *
746 * NOTE! we rely on the previous spin_lock to
747 * lock interrupts for us! We can only be called with
748 * "siglock" held, and the local interrupt must
749 * have been disabled when that got acquired!
750 *
751 * No need to set need_resched since signal event passing
752 * goes through ->blocked
753 */
754void signal_wake_up_state(struct task_struct *t, unsigned int state)
755{
756 set_tsk_thread_flag(t, TIF_SIGPENDING);
757 /*
758 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
759 * case. We don't check t->state here because there is a race with it
760 * executing another processor and just now entering stopped state.
761 * By using wake_up_state, we ensure the process will wake up and
762 * handle its death signal.
763 */
764 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
765 kick_process(t);
766}
767
768/*
769 * Remove signals in mask from the pending set and queue.
770 * Returns 1 if any signals were found.
771 *
772 * All callers must be holding the siglock.
773 */
774static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
775{
776 struct sigqueue *q, *n;
777 sigset_t m;
778
779 sigandsets(&m, mask, &s->signal);
780 if (sigisemptyset(&m))
781 return;
782
783 sigandnsets(&s->signal, &s->signal, mask);
784 list_for_each_entry_safe(q, n, &s->list, list) {
785 if (sigismember(mask, q->info.si_signo)) {
786 list_del_init(&q->list);
787 __sigqueue_free(q);
788 }
789 }
790}
791
792static inline int is_si_special(const struct kernel_siginfo *info)
793{
794 return info <= SEND_SIG_PRIV;
795}
796
797static inline bool si_fromuser(const struct kernel_siginfo *info)
798{
799 return info == SEND_SIG_NOINFO ||
800 (!is_si_special(info) && SI_FROMUSER(info));
801}
802
803/*
804 * called with RCU read lock from check_kill_permission()
805 */
806static bool kill_ok_by_cred(struct task_struct *t)
807{
808 const struct cred *cred = current_cred();
809 const struct cred *tcred = __task_cred(t);
810
811 return uid_eq(cred->euid, tcred->suid) ||
812 uid_eq(cred->euid, tcred->uid) ||
813 uid_eq(cred->uid, tcred->suid) ||
814 uid_eq(cred->uid, tcred->uid) ||
815 ns_capable(tcred->user_ns, CAP_KILL);
816}
817
818/*
819 * Bad permissions for sending the signal
820 * - the caller must hold the RCU read lock
821 */
822static int check_kill_permission(int sig, struct kernel_siginfo *info,
823 struct task_struct *t)
824{
825 struct pid *sid;
826 int error;
827
828 if (!valid_signal(sig))
829 return -EINVAL;
830
831 if (!si_fromuser(info))
832 return 0;
833
834 error = audit_signal_info(sig, t); /* Let audit system see the signal */
835 if (error)
836 return error;
837
838 if (!same_thread_group(current, t) &&
839 !kill_ok_by_cred(t)) {
840 switch (sig) {
841 case SIGCONT:
842 sid = task_session(t);
843 /*
844 * We don't return the error if sid == NULL. The
845 * task was unhashed, the caller must notice this.
846 */
847 if (!sid || sid == task_session(current))
848 break;
849 /* fall through */
850 default:
851 return -EPERM;
852 }
853 }
854
855 return security_task_kill(t, info, sig, NULL);
856}
857
858/**
859 * ptrace_trap_notify - schedule trap to notify ptracer
860 * @t: tracee wanting to notify tracer
861 *
862 * This function schedules sticky ptrace trap which is cleared on the next
863 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
864 * ptracer.
865 *
866 * If @t is running, STOP trap will be taken. If trapped for STOP and
867 * ptracer is listening for events, tracee is woken up so that it can
868 * re-trap for the new event. If trapped otherwise, STOP trap will be
869 * eventually taken without returning to userland after the existing traps
870 * are finished by PTRACE_CONT.
871 *
872 * CONTEXT:
873 * Must be called with @task->sighand->siglock held.
874 */
875static void ptrace_trap_notify(struct task_struct *t)
876{
877 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
878 assert_spin_locked(&t->sighand->siglock);
879
880 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
881 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
882}
883
884/*
885 * Handle magic process-wide effects of stop/continue signals. Unlike
886 * the signal actions, these happen immediately at signal-generation
887 * time regardless of blocking, ignoring, or handling. This does the
888 * actual continuing for SIGCONT, but not the actual stopping for stop
889 * signals. The process stop is done as a signal action for SIG_DFL.
890 *
891 * Returns true if the signal should be actually delivered, otherwise
892 * it should be dropped.
893 */
894static bool prepare_signal(int sig, struct task_struct *p, bool force)
895{
896 struct signal_struct *signal = p->signal;
897 struct task_struct *t;
898 sigset_t flush;
899
900 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
901 if (!(signal->flags & SIGNAL_GROUP_EXIT))
902 return sig == SIGKILL;
903 /*
904 * The process is in the middle of dying, nothing to do.
905 */
906 } else if (sig_kernel_stop(sig)) {
907 /*
908 * This is a stop signal. Remove SIGCONT from all queues.
909 */
910 siginitset(&flush, sigmask(SIGCONT));
911 flush_sigqueue_mask(&flush, &signal->shared_pending);
912 for_each_thread(p, t)
913 flush_sigqueue_mask(&flush, &t->pending);
914 } else if (sig == SIGCONT) {
915 unsigned int why;
916 /*
917 * Remove all stop signals from all queues, wake all threads.
918 */
919 siginitset(&flush, SIG_KERNEL_STOP_MASK);
920 flush_sigqueue_mask(&flush, &signal->shared_pending);
921 for_each_thread(p, t) {
922 flush_sigqueue_mask(&flush, &t->pending);
923 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
924 if (likely(!(t->ptrace & PT_SEIZED)))
925 wake_up_state(t, __TASK_STOPPED);
926 else
927 ptrace_trap_notify(t);
928 }
929
930 /*
931 * Notify the parent with CLD_CONTINUED if we were stopped.
932 *
933 * If we were in the middle of a group stop, we pretend it
934 * was already finished, and then continued. Since SIGCHLD
935 * doesn't queue we report only CLD_STOPPED, as if the next
936 * CLD_CONTINUED was dropped.
937 */
938 why = 0;
939 if (signal->flags & SIGNAL_STOP_STOPPED)
940 why |= SIGNAL_CLD_CONTINUED;
941 else if (signal->group_stop_count)
942 why |= SIGNAL_CLD_STOPPED;
943
944 if (why) {
945 /*
946 * The first thread which returns from do_signal_stop()
947 * will take ->siglock, notice SIGNAL_CLD_MASK, and
948 * notify its parent. See get_signal().
949 */
950 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
951 signal->group_stop_count = 0;
952 signal->group_exit_code = 0;
953 }
954 }
955
956 return !sig_ignored(p, sig, force);
957}
958
959/*
960 * Test if P wants to take SIG. After we've checked all threads with this,
961 * it's equivalent to finding no threads not blocking SIG. Any threads not
962 * blocking SIG were ruled out because they are not running and already
963 * have pending signals. Such threads will dequeue from the shared queue
964 * as soon as they're available, so putting the signal on the shared queue
965 * will be equivalent to sending it to one such thread.
966 */
967static inline bool wants_signal(int sig, struct task_struct *p)
968{
969 if (sigismember(&p->blocked, sig))
970 return false;
971
972 if (p->flags & PF_EXITING)
973 return false;
974
975 if (sig == SIGKILL)
976 return true;
977
978 if (task_is_stopped_or_traced(p))
979 return false;
980
981 return task_curr(p) || !signal_pending(p);
982}
983
984static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
985{
986 struct signal_struct *signal = p->signal;
987 struct task_struct *t;
988
989 /*
990 * Now find a thread we can wake up to take the signal off the queue.
991 *
992 * If the main thread wants the signal, it gets first crack.
993 * Probably the least surprising to the average bear.
994 */
995 if (wants_signal(sig, p))
996 t = p;
997 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
998 /*
999 * There is just one thread and it does not need to be woken.
1000 * It will dequeue unblocked signals before it runs again.
1001 */
1002 return;
1003 else {
1004 /*
1005 * Otherwise try to find a suitable thread.
1006 */
1007 t = signal->curr_target;
1008 while (!wants_signal(sig, t)) {
1009 t = next_thread(t);
1010 if (t == signal->curr_target)
1011 /*
1012 * No thread needs to be woken.
1013 * Any eligible threads will see
1014 * the signal in the queue soon.
1015 */
1016 return;
1017 }
1018 signal->curr_target = t;
1019 }
1020
1021 /*
1022 * Found a killable thread. If the signal will be fatal,
1023 * then start taking the whole group down immediately.
1024 */
1025 if (sig_fatal(p, sig) &&
1026 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1027 !sigismember(&t->real_blocked, sig) &&
1028 (sig == SIGKILL || !p->ptrace)) {
1029 /*
1030 * This signal will be fatal to the whole group.
1031 */
1032 if (!sig_kernel_coredump(sig)) {
1033 /*
1034 * Start a group exit and wake everybody up.
1035 * This way we don't have other threads
1036 * running and doing things after a slower
1037 * thread has the fatal signal pending.
1038 */
1039 signal->flags = SIGNAL_GROUP_EXIT;
1040 signal->group_exit_code = sig;
1041 signal->group_stop_count = 0;
1042 t = p;
1043 do {
1044 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1045 sigaddset(&t->pending.signal, SIGKILL);
1046 signal_wake_up(t, 1);
1047 } while_each_thread(p, t);
1048 return;
1049 }
1050 }
1051
1052 /*
1053 * The signal is already in the shared-pending queue.
1054 * Tell the chosen thread to wake up and dequeue it.
1055 */
1056 signal_wake_up(t, sig == SIGKILL);
1057 return;
1058}
1059
1060static inline bool legacy_queue(struct sigpending *signals, int sig)
1061{
1062 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1063}
1064
1065static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1066 enum pid_type type, bool force)
1067{
1068 struct sigpending *pending;
1069 struct sigqueue *q;
1070 int override_rlimit;
1071 int ret = 0, result;
1072
1073 assert_spin_locked(&t->sighand->siglock);
1074
1075 result = TRACE_SIGNAL_IGNORED;
1076 if (!prepare_signal(sig, t, force))
1077 goto ret;
1078
1079 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1080 /*
1081 * Short-circuit ignored signals and support queuing
1082 * exactly one non-rt signal, so that we can get more
1083 * detailed information about the cause of the signal.
1084 */
1085 result = TRACE_SIGNAL_ALREADY_PENDING;
1086 if (legacy_queue(pending, sig))
1087 goto ret;
1088
1089 result = TRACE_SIGNAL_DELIVERED;
1090 /*
1091 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1092 */
1093 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1094 goto out_set;
1095
1096 /*
1097 * Real-time signals must be queued if sent by sigqueue, or
1098 * some other real-time mechanism. It is implementation
1099 * defined whether kill() does so. We attempt to do so, on
1100 * the principle of least surprise, but since kill is not
1101 * allowed to fail with EAGAIN when low on memory we just
1102 * make sure at least one signal gets delivered and don't
1103 * pass on the info struct.
1104 */
1105 if (sig < SIGRTMIN)
1106 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1107 else
1108 override_rlimit = 0;
1109
1110 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit);
1111 if (q) {
1112 list_add_tail(&q->list, &pending->list);
1113 switch ((unsigned long) info) {
1114 case (unsigned long) SEND_SIG_NOINFO:
1115 clear_siginfo(&q->info);
1116 q->info.si_signo = sig;
1117 q->info.si_errno = 0;
1118 q->info.si_code = SI_USER;
1119 q->info.si_pid = task_tgid_nr_ns(current,
1120 task_active_pid_ns(t));
1121 rcu_read_lock();
1122 q->info.si_uid =
1123 from_kuid_munged(task_cred_xxx(t, user_ns),
1124 current_uid());
1125 rcu_read_unlock();
1126 break;
1127 case (unsigned long) SEND_SIG_PRIV:
1128 clear_siginfo(&q->info);
1129 q->info.si_signo = sig;
1130 q->info.si_errno = 0;
1131 q->info.si_code = SI_KERNEL;
1132 q->info.si_pid = 0;
1133 q->info.si_uid = 0;
1134 break;
1135 default:
1136 copy_siginfo(&q->info, info);
1137 break;
1138 }
1139 } else if (!is_si_special(info) &&
1140 sig >= SIGRTMIN && info->si_code != SI_USER) {
1141 /*
1142 * Queue overflow, abort. We may abort if the
1143 * signal was rt and sent by user using something
1144 * other than kill().
1145 */
1146 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1147 ret = -EAGAIN;
1148 goto ret;
1149 } else {
1150 /*
1151 * This is a silent loss of information. We still
1152 * send the signal, but the *info bits are lost.
1153 */
1154 result = TRACE_SIGNAL_LOSE_INFO;
1155 }
1156
1157out_set:
1158 signalfd_notify(t, sig);
1159 sigaddset(&pending->signal, sig);
1160
1161 /* Let multiprocess signals appear after on-going forks */
1162 if (type > PIDTYPE_TGID) {
1163 struct multiprocess_signals *delayed;
1164 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1165 sigset_t *signal = &delayed->signal;
1166 /* Can't queue both a stop and a continue signal */
1167 if (sig == SIGCONT)
1168 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1169 else if (sig_kernel_stop(sig))
1170 sigdelset(signal, SIGCONT);
1171 sigaddset(signal, sig);
1172 }
1173 }
1174
1175 complete_signal(sig, t, type);
1176ret:
1177 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1178 return ret;
1179}
1180
1181static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1182{
1183 bool ret = false;
1184 switch (siginfo_layout(info->si_signo, info->si_code)) {
1185 case SIL_KILL:
1186 case SIL_CHLD:
1187 case SIL_RT:
1188 ret = true;
1189 break;
1190 case SIL_TIMER:
1191 case SIL_POLL:
1192 case SIL_FAULT:
1193 case SIL_FAULT_MCEERR:
1194 case SIL_FAULT_BNDERR:
1195 case SIL_FAULT_PKUERR:
1196 case SIL_SYS:
1197 ret = false;
1198 break;
1199 }
1200 return ret;
1201}
1202
1203static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1204 enum pid_type type)
1205{
1206 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1207 bool force = false;
1208
1209 if (info == SEND_SIG_NOINFO) {
1210 /* Force if sent from an ancestor pid namespace */
1211 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1212 } else if (info == SEND_SIG_PRIV) {
1213 /* Don't ignore kernel generated signals */
1214 force = true;
1215 } else if (has_si_pid_and_uid(info)) {
1216 /* SIGKILL and SIGSTOP is special or has ids */
1217 struct user_namespace *t_user_ns;
1218
1219 rcu_read_lock();
1220 t_user_ns = task_cred_xxx(t, user_ns);
1221 if (current_user_ns() != t_user_ns) {
1222 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1223 info->si_uid = from_kuid_munged(t_user_ns, uid);
1224 }
1225 rcu_read_unlock();
1226
1227 /* A kernel generated signal? */
1228 force = (info->si_code == SI_KERNEL);
1229
1230 /* From an ancestor pid namespace? */
1231 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1232 info->si_pid = 0;
1233 force = true;
1234 }
1235 }
1236 return __send_signal(sig, info, t, type, force);
1237}
1238
1239static void print_fatal_signal(int signr)
1240{
1241 struct pt_regs *regs = signal_pt_regs();
1242 pr_info("potentially unexpected fatal signal %d.\n", signr);
1243
1244#if defined(__i386__) && !defined(__arch_um__)
1245 pr_info("code at %08lx: ", regs->ip);
1246 {
1247 int i;
1248 for (i = 0; i < 16; i++) {
1249 unsigned char insn;
1250
1251 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1252 break;
1253 pr_cont("%02x ", insn);
1254 }
1255 }
1256 pr_cont("\n");
1257#endif
1258 preempt_disable();
1259 show_regs(regs);
1260 preempt_enable();
1261}
1262
1263static int __init setup_print_fatal_signals(char *str)
1264{
1265 get_option (&str, &print_fatal_signals);
1266
1267 return 1;
1268}
1269
1270__setup("print-fatal-signals=", setup_print_fatal_signals);
1271
1272int
1273__group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1274{
1275 return send_signal(sig, info, p, PIDTYPE_TGID);
1276}
1277
1278int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1279 enum pid_type type)
1280{
1281 unsigned long flags;
1282 int ret = -ESRCH;
1283
1284 if (lock_task_sighand(p, &flags)) {
1285 ret = send_signal(sig, info, p, type);
1286 unlock_task_sighand(p, &flags);
1287 }
1288
1289 return ret;
1290}
1291
1292/*
1293 * Force a signal that the process can't ignore: if necessary
1294 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1295 *
1296 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1297 * since we do not want to have a signal handler that was blocked
1298 * be invoked when user space had explicitly blocked it.
1299 *
1300 * We don't want to have recursive SIGSEGV's etc, for example,
1301 * that is why we also clear SIGNAL_UNKILLABLE.
1302 */
1303static int
1304force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1305{
1306 unsigned long int flags;
1307 int ret, blocked, ignored;
1308 struct k_sigaction *action;
1309 int sig = info->si_signo;
1310
1311 spin_lock_irqsave(&t->sighand->siglock, flags);
1312 action = &t->sighand->action[sig-1];
1313 ignored = action->sa.sa_handler == SIG_IGN;
1314 blocked = sigismember(&t->blocked, sig);
1315 if (blocked || ignored) {
1316 action->sa.sa_handler = SIG_DFL;
1317 if (blocked) {
1318 sigdelset(&t->blocked, sig);
1319 recalc_sigpending_and_wake(t);
1320 }
1321 }
1322 /*
1323 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1324 * debugging to leave init killable.
1325 */
1326 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1327 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1328 ret = send_signal(sig, info, t, PIDTYPE_PID);
1329 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1330
1331 return ret;
1332}
1333
1334int force_sig_info(struct kernel_siginfo *info)
1335{
1336 return force_sig_info_to_task(info, current);
1337}
1338
1339/*
1340 * Nuke all other threads in the group.
1341 */
1342int zap_other_threads(struct task_struct *p)
1343{
1344 struct task_struct *t = p;
1345 int count = 0;
1346
1347 p->signal->group_stop_count = 0;
1348
1349 while_each_thread(p, t) {
1350 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1351 count++;
1352
1353 /* Don't bother with already dead threads */
1354 if (t->exit_state)
1355 continue;
1356 sigaddset(&t->pending.signal, SIGKILL);
1357 signal_wake_up(t, 1);
1358 }
1359
1360 return count;
1361}
1362
1363struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1364 unsigned long *flags)
1365{
1366 struct sighand_struct *sighand;
1367
1368 rcu_read_lock();
1369 for (;;) {
1370 sighand = rcu_dereference(tsk->sighand);
1371 if (unlikely(sighand == NULL))
1372 break;
1373
1374 /*
1375 * This sighand can be already freed and even reused, but
1376 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1377 * initializes ->siglock: this slab can't go away, it has
1378 * the same object type, ->siglock can't be reinitialized.
1379 *
1380 * We need to ensure that tsk->sighand is still the same
1381 * after we take the lock, we can race with de_thread() or
1382 * __exit_signal(). In the latter case the next iteration
1383 * must see ->sighand == NULL.
1384 */
1385 spin_lock_irqsave(&sighand->siglock, *flags);
1386 if (likely(sighand == tsk->sighand))
1387 break;
1388 spin_unlock_irqrestore(&sighand->siglock, *flags);
1389 }
1390 rcu_read_unlock();
1391
1392 return sighand;
1393}
1394
1395/*
1396 * send signal info to all the members of a group
1397 */
1398int group_send_sig_info(int sig, struct kernel_siginfo *info,
1399 struct task_struct *p, enum pid_type type)
1400{
1401 int ret;
1402
1403 rcu_read_lock();
1404 ret = check_kill_permission(sig, info, p);
1405 rcu_read_unlock();
1406
1407 if (!ret && sig)
1408 ret = do_send_sig_info(sig, info, p, type);
1409
1410 return ret;
1411}
1412
1413/*
1414 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1415 * control characters do (^C, ^Z etc)
1416 * - the caller must hold at least a readlock on tasklist_lock
1417 */
1418int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1419{
1420 struct task_struct *p = NULL;
1421 int retval, success;
1422
1423 success = 0;
1424 retval = -ESRCH;
1425 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1426 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1427 success |= !err;
1428 retval = err;
1429 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1430 return success ? 0 : retval;
1431}
1432
1433int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1434{
1435 int error = -ESRCH;
1436 struct task_struct *p;
1437
1438 for (;;) {
1439 rcu_read_lock();
1440 p = pid_task(pid, PIDTYPE_PID);
1441 if (p)
1442 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1443 rcu_read_unlock();
1444 if (likely(!p || error != -ESRCH))
1445 return error;
1446
1447 /*
1448 * The task was unhashed in between, try again. If it
1449 * is dead, pid_task() will return NULL, if we race with
1450 * de_thread() it will find the new leader.
1451 */
1452 }
1453}
1454
1455static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1456{
1457 int error;
1458 rcu_read_lock();
1459 error = kill_pid_info(sig, info, find_vpid(pid));
1460 rcu_read_unlock();
1461 return error;
1462}
1463
1464static inline bool kill_as_cred_perm(const struct cred *cred,
1465 struct task_struct *target)
1466{
1467 const struct cred *pcred = __task_cred(target);
1468
1469 return uid_eq(cred->euid, pcred->suid) ||
1470 uid_eq(cred->euid, pcred->uid) ||
1471 uid_eq(cred->uid, pcred->suid) ||
1472 uid_eq(cred->uid, pcred->uid);
1473}
1474
1475/*
1476 * The usb asyncio usage of siginfo is wrong. The glibc support
1477 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1478 * AKA after the generic fields:
1479 * kernel_pid_t si_pid;
1480 * kernel_uid32_t si_uid;
1481 * sigval_t si_value;
1482 *
1483 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1484 * after the generic fields is:
1485 * void __user *si_addr;
1486 *
1487 * This is a practical problem when there is a 64bit big endian kernel
1488 * and a 32bit userspace. As the 32bit address will encoded in the low
1489 * 32bits of the pointer. Those low 32bits will be stored at higher
1490 * address than appear in a 32 bit pointer. So userspace will not
1491 * see the address it was expecting for it's completions.
1492 *
1493 * There is nothing in the encoding that can allow
1494 * copy_siginfo_to_user32 to detect this confusion of formats, so
1495 * handle this by requiring the caller of kill_pid_usb_asyncio to
1496 * notice when this situration takes place and to store the 32bit
1497 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1498 * parameter.
1499 */
1500int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1501 struct pid *pid, const struct cred *cred)
1502{
1503 struct kernel_siginfo info;
1504 struct task_struct *p;
1505 unsigned long flags;
1506 int ret = -EINVAL;
1507
1508 clear_siginfo(&info);
1509 info.si_signo = sig;
1510 info.si_errno = errno;
1511 info.si_code = SI_ASYNCIO;
1512 *((sigval_t *)&info.si_pid) = addr;
1513
1514 if (!valid_signal(sig))
1515 return ret;
1516
1517 rcu_read_lock();
1518 p = pid_task(pid, PIDTYPE_PID);
1519 if (!p) {
1520 ret = -ESRCH;
1521 goto out_unlock;
1522 }
1523 if (!kill_as_cred_perm(cred, p)) {
1524 ret = -EPERM;
1525 goto out_unlock;
1526 }
1527 ret = security_task_kill(p, &info, sig, cred);
1528 if (ret)
1529 goto out_unlock;
1530
1531 if (sig) {
1532 if (lock_task_sighand(p, &flags)) {
1533 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1534 unlock_task_sighand(p, &flags);
1535 } else
1536 ret = -ESRCH;
1537 }
1538out_unlock:
1539 rcu_read_unlock();
1540 return ret;
1541}
1542EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1543
1544/*
1545 * kill_something_info() interprets pid in interesting ways just like kill(2).
1546 *
1547 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1548 * is probably wrong. Should make it like BSD or SYSV.
1549 */
1550
1551static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1552{
1553 int ret;
1554
1555 if (pid > 0) {
1556 rcu_read_lock();
1557 ret = kill_pid_info(sig, info, find_vpid(pid));
1558 rcu_read_unlock();
1559 return ret;
1560 }
1561
1562 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1563 if (pid == INT_MIN)
1564 return -ESRCH;
1565
1566 read_lock(&tasklist_lock);
1567 if (pid != -1) {
1568 ret = __kill_pgrp_info(sig, info,
1569 pid ? find_vpid(-pid) : task_pgrp(current));
1570 } else {
1571 int retval = 0, count = 0;
1572 struct task_struct * p;
1573
1574 for_each_process(p) {
1575 if (task_pid_vnr(p) > 1 &&
1576 !same_thread_group(p, current)) {
1577 int err = group_send_sig_info(sig, info, p,
1578 PIDTYPE_MAX);
1579 ++count;
1580 if (err != -EPERM)
1581 retval = err;
1582 }
1583 }
1584 ret = count ? retval : -ESRCH;
1585 }
1586 read_unlock(&tasklist_lock);
1587
1588 return ret;
1589}
1590
1591/*
1592 * These are for backward compatibility with the rest of the kernel source.
1593 */
1594
1595int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1596{
1597 /*
1598 * Make sure legacy kernel users don't send in bad values
1599 * (normal paths check this in check_kill_permission).
1600 */
1601 if (!valid_signal(sig))
1602 return -EINVAL;
1603
1604 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1605}
1606EXPORT_SYMBOL(send_sig_info);
1607
1608#define __si_special(priv) \
1609 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1610
1611int
1612send_sig(int sig, struct task_struct *p, int priv)
1613{
1614 return send_sig_info(sig, __si_special(priv), p);
1615}
1616EXPORT_SYMBOL(send_sig);
1617
1618void force_sig(int sig)
1619{
1620 struct kernel_siginfo info;
1621
1622 clear_siginfo(&info);
1623 info.si_signo = sig;
1624 info.si_errno = 0;
1625 info.si_code = SI_KERNEL;
1626 info.si_pid = 0;
1627 info.si_uid = 0;
1628 force_sig_info(&info);
1629}
1630EXPORT_SYMBOL(force_sig);
1631
1632/*
1633 * When things go south during signal handling, we
1634 * will force a SIGSEGV. And if the signal that caused
1635 * the problem was already a SIGSEGV, we'll want to
1636 * make sure we don't even try to deliver the signal..
1637 */
1638void force_sigsegv(int sig)
1639{
1640 struct task_struct *p = current;
1641
1642 if (sig == SIGSEGV) {
1643 unsigned long flags;
1644 spin_lock_irqsave(&p->sighand->siglock, flags);
1645 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1646 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1647 }
1648 force_sig(SIGSEGV);
1649}
1650
1651int force_sig_fault_to_task(int sig, int code, void __user *addr
1652 ___ARCH_SI_TRAPNO(int trapno)
1653 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1654 , struct task_struct *t)
1655{
1656 struct kernel_siginfo info;
1657
1658 clear_siginfo(&info);
1659 info.si_signo = sig;
1660 info.si_errno = 0;
1661 info.si_code = code;
1662 info.si_addr = addr;
1663#ifdef __ARCH_SI_TRAPNO
1664 info.si_trapno = trapno;
1665#endif
1666#ifdef __ia64__
1667 info.si_imm = imm;
1668 info.si_flags = flags;
1669 info.si_isr = isr;
1670#endif
1671 return force_sig_info_to_task(&info, t);
1672}
1673
1674int force_sig_fault(int sig, int code, void __user *addr
1675 ___ARCH_SI_TRAPNO(int trapno)
1676 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1677{
1678 return force_sig_fault_to_task(sig, code, addr
1679 ___ARCH_SI_TRAPNO(trapno)
1680 ___ARCH_SI_IA64(imm, flags, isr), current);
1681}
1682
1683int send_sig_fault(int sig, int code, void __user *addr
1684 ___ARCH_SI_TRAPNO(int trapno)
1685 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1686 , struct task_struct *t)
1687{
1688 struct kernel_siginfo info;
1689
1690 clear_siginfo(&info);
1691 info.si_signo = sig;
1692 info.si_errno = 0;
1693 info.si_code = code;
1694 info.si_addr = addr;
1695#ifdef __ARCH_SI_TRAPNO
1696 info.si_trapno = trapno;
1697#endif
1698#ifdef __ia64__
1699 info.si_imm = imm;
1700 info.si_flags = flags;
1701 info.si_isr = isr;
1702#endif
1703 return send_sig_info(info.si_signo, &info, t);
1704}
1705
1706int force_sig_mceerr(int code, void __user *addr, short lsb)
1707{
1708 struct kernel_siginfo info;
1709
1710 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1711 clear_siginfo(&info);
1712 info.si_signo = SIGBUS;
1713 info.si_errno = 0;
1714 info.si_code = code;
1715 info.si_addr = addr;
1716 info.si_addr_lsb = lsb;
1717 return force_sig_info(&info);
1718}
1719
1720int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1721{
1722 struct kernel_siginfo info;
1723
1724 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1725 clear_siginfo(&info);
1726 info.si_signo = SIGBUS;
1727 info.si_errno = 0;
1728 info.si_code = code;
1729 info.si_addr = addr;
1730 info.si_addr_lsb = lsb;
1731 return send_sig_info(info.si_signo, &info, t);
1732}
1733EXPORT_SYMBOL(send_sig_mceerr);
1734
1735int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1736{
1737 struct kernel_siginfo info;
1738
1739 clear_siginfo(&info);
1740 info.si_signo = SIGSEGV;
1741 info.si_errno = 0;
1742 info.si_code = SEGV_BNDERR;
1743 info.si_addr = addr;
1744 info.si_lower = lower;
1745 info.si_upper = upper;
1746 return force_sig_info(&info);
1747}
1748
1749#ifdef SEGV_PKUERR
1750int force_sig_pkuerr(void __user *addr, u32 pkey)
1751{
1752 struct kernel_siginfo info;
1753
1754 clear_siginfo(&info);
1755 info.si_signo = SIGSEGV;
1756 info.si_errno = 0;
1757 info.si_code = SEGV_PKUERR;
1758 info.si_addr = addr;
1759 info.si_pkey = pkey;
1760 return force_sig_info(&info);
1761}
1762#endif
1763
1764/* For the crazy architectures that include trap information in
1765 * the errno field, instead of an actual errno value.
1766 */
1767int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1768{
1769 struct kernel_siginfo info;
1770
1771 clear_siginfo(&info);
1772 info.si_signo = SIGTRAP;
1773 info.si_errno = errno;
1774 info.si_code = TRAP_HWBKPT;
1775 info.si_addr = addr;
1776 return force_sig_info(&info);
1777}
1778
1779int kill_pgrp(struct pid *pid, int sig, int priv)
1780{
1781 int ret;
1782
1783 read_lock(&tasklist_lock);
1784 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1785 read_unlock(&tasklist_lock);
1786
1787 return ret;
1788}
1789EXPORT_SYMBOL(kill_pgrp);
1790
1791int kill_pid(struct pid *pid, int sig, int priv)
1792{
1793 return kill_pid_info(sig, __si_special(priv), pid);
1794}
1795EXPORT_SYMBOL(kill_pid);
1796
1797/*
1798 * These functions support sending signals using preallocated sigqueue
1799 * structures. This is needed "because realtime applications cannot
1800 * afford to lose notifications of asynchronous events, like timer
1801 * expirations or I/O completions". In the case of POSIX Timers
1802 * we allocate the sigqueue structure from the timer_create. If this
1803 * allocation fails we are able to report the failure to the application
1804 * with an EAGAIN error.
1805 */
1806struct sigqueue *sigqueue_alloc(void)
1807{
1808 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1809
1810 if (q)
1811 q->flags |= SIGQUEUE_PREALLOC;
1812
1813 return q;
1814}
1815
1816void sigqueue_free(struct sigqueue *q)
1817{
1818 unsigned long flags;
1819 spinlock_t *lock = ¤t->sighand->siglock;
1820
1821 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1822 /*
1823 * We must hold ->siglock while testing q->list
1824 * to serialize with collect_signal() or with
1825 * __exit_signal()->flush_sigqueue().
1826 */
1827 spin_lock_irqsave(lock, flags);
1828 q->flags &= ~SIGQUEUE_PREALLOC;
1829 /*
1830 * If it is queued it will be freed when dequeued,
1831 * like the "regular" sigqueue.
1832 */
1833 if (!list_empty(&q->list))
1834 q = NULL;
1835 spin_unlock_irqrestore(lock, flags);
1836
1837 if (q)
1838 __sigqueue_free(q);
1839}
1840
1841int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1842{
1843 int sig = q->info.si_signo;
1844 struct sigpending *pending;
1845 struct task_struct *t;
1846 unsigned long flags;
1847 int ret, result;
1848
1849 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1850
1851 ret = -1;
1852 rcu_read_lock();
1853 t = pid_task(pid, type);
1854 if (!t || !likely(lock_task_sighand(t, &flags)))
1855 goto ret;
1856
1857 ret = 1; /* the signal is ignored */
1858 result = TRACE_SIGNAL_IGNORED;
1859 if (!prepare_signal(sig, t, false))
1860 goto out;
1861
1862 ret = 0;
1863 if (unlikely(!list_empty(&q->list))) {
1864 /*
1865 * If an SI_TIMER entry is already queue just increment
1866 * the overrun count.
1867 */
1868 BUG_ON(q->info.si_code != SI_TIMER);
1869 q->info.si_overrun++;
1870 result = TRACE_SIGNAL_ALREADY_PENDING;
1871 goto out;
1872 }
1873 q->info.si_overrun = 0;
1874
1875 signalfd_notify(t, sig);
1876 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1877 list_add_tail(&q->list, &pending->list);
1878 sigaddset(&pending->signal, sig);
1879 complete_signal(sig, t, type);
1880 result = TRACE_SIGNAL_DELIVERED;
1881out:
1882 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1883 unlock_task_sighand(t, &flags);
1884ret:
1885 rcu_read_unlock();
1886 return ret;
1887}
1888
1889static void do_notify_pidfd(struct task_struct *task)
1890{
1891 struct pid *pid;
1892
1893 WARN_ON(task->exit_state == 0);
1894 pid = task_pid(task);
1895 wake_up_all(&pid->wait_pidfd);
1896}
1897
1898/*
1899 * Let a parent know about the death of a child.
1900 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1901 *
1902 * Returns true if our parent ignored us and so we've switched to
1903 * self-reaping.
1904 */
1905bool do_notify_parent(struct task_struct *tsk, int sig)
1906{
1907 struct kernel_siginfo info;
1908 unsigned long flags;
1909 struct sighand_struct *psig;
1910 bool autoreap = false;
1911 u64 utime, stime;
1912
1913 BUG_ON(sig == -1);
1914
1915 /* do_notify_parent_cldstop should have been called instead. */
1916 BUG_ON(task_is_stopped_or_traced(tsk));
1917
1918 BUG_ON(!tsk->ptrace &&
1919 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1920
1921 /* Wake up all pidfd waiters */
1922 do_notify_pidfd(tsk);
1923
1924 if (sig != SIGCHLD) {
1925 /*
1926 * This is only possible if parent == real_parent.
1927 * Check if it has changed security domain.
1928 */
1929 if (tsk->parent_exec_id != tsk->parent->self_exec_id)
1930 sig = SIGCHLD;
1931 }
1932
1933 clear_siginfo(&info);
1934 info.si_signo = sig;
1935 info.si_errno = 0;
1936 /*
1937 * We are under tasklist_lock here so our parent is tied to
1938 * us and cannot change.
1939 *
1940 * task_active_pid_ns will always return the same pid namespace
1941 * until a task passes through release_task.
1942 *
1943 * write_lock() currently calls preempt_disable() which is the
1944 * same as rcu_read_lock(), but according to Oleg, this is not
1945 * correct to rely on this
1946 */
1947 rcu_read_lock();
1948 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1949 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1950 task_uid(tsk));
1951 rcu_read_unlock();
1952
1953 task_cputime(tsk, &utime, &stime);
1954 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1955 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1956
1957 info.si_status = tsk->exit_code & 0x7f;
1958 if (tsk->exit_code & 0x80)
1959 info.si_code = CLD_DUMPED;
1960 else if (tsk->exit_code & 0x7f)
1961 info.si_code = CLD_KILLED;
1962 else {
1963 info.si_code = CLD_EXITED;
1964 info.si_status = tsk->exit_code >> 8;
1965 }
1966
1967 psig = tsk->parent->sighand;
1968 spin_lock_irqsave(&psig->siglock, flags);
1969 if (!tsk->ptrace && sig == SIGCHLD &&
1970 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1971 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1972 /*
1973 * We are exiting and our parent doesn't care. POSIX.1
1974 * defines special semantics for setting SIGCHLD to SIG_IGN
1975 * or setting the SA_NOCLDWAIT flag: we should be reaped
1976 * automatically and not left for our parent's wait4 call.
1977 * Rather than having the parent do it as a magic kind of
1978 * signal handler, we just set this to tell do_exit that we
1979 * can be cleaned up without becoming a zombie. Note that
1980 * we still call __wake_up_parent in this case, because a
1981 * blocked sys_wait4 might now return -ECHILD.
1982 *
1983 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1984 * is implementation-defined: we do (if you don't want
1985 * it, just use SIG_IGN instead).
1986 */
1987 autoreap = true;
1988 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1989 sig = 0;
1990 }
1991 if (valid_signal(sig) && sig)
1992 __group_send_sig_info(sig, &info, tsk->parent);
1993 __wake_up_parent(tsk, tsk->parent);
1994 spin_unlock_irqrestore(&psig->siglock, flags);
1995
1996 return autoreap;
1997}
1998
1999/**
2000 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2001 * @tsk: task reporting the state change
2002 * @for_ptracer: the notification is for ptracer
2003 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2004 *
2005 * Notify @tsk's parent that the stopped/continued state has changed. If
2006 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2007 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2008 *
2009 * CONTEXT:
2010 * Must be called with tasklist_lock at least read locked.
2011 */
2012static void do_notify_parent_cldstop(struct task_struct *tsk,
2013 bool for_ptracer, int why)
2014{
2015 struct kernel_siginfo info;
2016 unsigned long flags;
2017 struct task_struct *parent;
2018 struct sighand_struct *sighand;
2019 u64 utime, stime;
2020
2021 if (for_ptracer) {
2022 parent = tsk->parent;
2023 } else {
2024 tsk = tsk->group_leader;
2025 parent = tsk->real_parent;
2026 }
2027
2028 clear_siginfo(&info);
2029 info.si_signo = SIGCHLD;
2030 info.si_errno = 0;
2031 /*
2032 * see comment in do_notify_parent() about the following 4 lines
2033 */
2034 rcu_read_lock();
2035 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2036 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2037 rcu_read_unlock();
2038
2039 task_cputime(tsk, &utime, &stime);
2040 info.si_utime = nsec_to_clock_t(utime);
2041 info.si_stime = nsec_to_clock_t(stime);
2042
2043 info.si_code = why;
2044 switch (why) {
2045 case CLD_CONTINUED:
2046 info.si_status = SIGCONT;
2047 break;
2048 case CLD_STOPPED:
2049 info.si_status = tsk->signal->group_exit_code & 0x7f;
2050 break;
2051 case CLD_TRAPPED:
2052 info.si_status = tsk->exit_code & 0x7f;
2053 break;
2054 default:
2055 BUG();
2056 }
2057
2058 sighand = parent->sighand;
2059 spin_lock_irqsave(&sighand->siglock, flags);
2060 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2061 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2062 __group_send_sig_info(SIGCHLD, &info, parent);
2063 /*
2064 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2065 */
2066 __wake_up_parent(tsk, parent);
2067 spin_unlock_irqrestore(&sighand->siglock, flags);
2068}
2069
2070static inline bool may_ptrace_stop(void)
2071{
2072 if (!likely(current->ptrace))
2073 return false;
2074 /*
2075 * Are we in the middle of do_coredump?
2076 * If so and our tracer is also part of the coredump stopping
2077 * is a deadlock situation, and pointless because our tracer
2078 * is dead so don't allow us to stop.
2079 * If SIGKILL was already sent before the caller unlocked
2080 * ->siglock we must see ->core_state != NULL. Otherwise it
2081 * is safe to enter schedule().
2082 *
2083 * This is almost outdated, a task with the pending SIGKILL can't
2084 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2085 * after SIGKILL was already dequeued.
2086 */
2087 if (unlikely(current->mm->core_state) &&
2088 unlikely(current->mm == current->parent->mm))
2089 return false;
2090
2091 return true;
2092}
2093
2094/*
2095 * Return non-zero if there is a SIGKILL that should be waking us up.
2096 * Called with the siglock held.
2097 */
2098static bool sigkill_pending(struct task_struct *tsk)
2099{
2100 return sigismember(&tsk->pending.signal, SIGKILL) ||
2101 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2102}
2103
2104/*
2105 * This must be called with current->sighand->siglock held.
2106 *
2107 * This should be the path for all ptrace stops.
2108 * We always set current->last_siginfo while stopped here.
2109 * That makes it a way to test a stopped process for
2110 * being ptrace-stopped vs being job-control-stopped.
2111 *
2112 * If we actually decide not to stop at all because the tracer
2113 * is gone, we keep current->exit_code unless clear_code.
2114 */
2115static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2116 __releases(¤t->sighand->siglock)
2117 __acquires(¤t->sighand->siglock)
2118{
2119 bool gstop_done = false;
2120
2121 if (arch_ptrace_stop_needed(exit_code, info)) {
2122 /*
2123 * The arch code has something special to do before a
2124 * ptrace stop. This is allowed to block, e.g. for faults
2125 * on user stack pages. We can't keep the siglock while
2126 * calling arch_ptrace_stop, so we must release it now.
2127 * To preserve proper semantics, we must do this before
2128 * any signal bookkeeping like checking group_stop_count.
2129 * Meanwhile, a SIGKILL could come in before we retake the
2130 * siglock. That must prevent us from sleeping in TASK_TRACED.
2131 * So after regaining the lock, we must check for SIGKILL.
2132 */
2133 spin_unlock_irq(¤t->sighand->siglock);
2134 arch_ptrace_stop(exit_code, info);
2135 spin_lock_irq(¤t->sighand->siglock);
2136 if (sigkill_pending(current))
2137 return;
2138 }
2139
2140 set_special_state(TASK_TRACED);
2141
2142 /*
2143 * We're committing to trapping. TRACED should be visible before
2144 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2145 * Also, transition to TRACED and updates to ->jobctl should be
2146 * atomic with respect to siglock and should be done after the arch
2147 * hook as siglock is released and regrabbed across it.
2148 *
2149 * TRACER TRACEE
2150 *
2151 * ptrace_attach()
2152 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2153 * do_wait()
2154 * set_current_state() smp_wmb();
2155 * ptrace_do_wait()
2156 * wait_task_stopped()
2157 * task_stopped_code()
2158 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2159 */
2160 smp_wmb();
2161
2162 current->last_siginfo = info;
2163 current->exit_code = exit_code;
2164
2165 /*
2166 * If @why is CLD_STOPPED, we're trapping to participate in a group
2167 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2168 * across siglock relocks since INTERRUPT was scheduled, PENDING
2169 * could be clear now. We act as if SIGCONT is received after
2170 * TASK_TRACED is entered - ignore it.
2171 */
2172 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2173 gstop_done = task_participate_group_stop(current);
2174
2175 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2176 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2177 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2178 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2179
2180 /* entering a trap, clear TRAPPING */
2181 task_clear_jobctl_trapping(current);
2182
2183 spin_unlock_irq(¤t->sighand->siglock);
2184 read_lock(&tasklist_lock);
2185 if (may_ptrace_stop()) {
2186 /*
2187 * Notify parents of the stop.
2188 *
2189 * While ptraced, there are two parents - the ptracer and
2190 * the real_parent of the group_leader. The ptracer should
2191 * know about every stop while the real parent is only
2192 * interested in the completion of group stop. The states
2193 * for the two don't interact with each other. Notify
2194 * separately unless they're gonna be duplicates.
2195 */
2196 do_notify_parent_cldstop(current, true, why);
2197 if (gstop_done && ptrace_reparented(current))
2198 do_notify_parent_cldstop(current, false, why);
2199
2200 /*
2201 * Don't want to allow preemption here, because
2202 * sys_ptrace() needs this task to be inactive.
2203 *
2204 * XXX: implement read_unlock_no_resched().
2205 */
2206 preempt_disable();
2207 read_unlock(&tasklist_lock);
2208 cgroup_enter_frozen();
2209 preempt_enable_no_resched();
2210 freezable_schedule();
2211 cgroup_leave_frozen(true);
2212 } else {
2213 /*
2214 * By the time we got the lock, our tracer went away.
2215 * Don't drop the lock yet, another tracer may come.
2216 *
2217 * If @gstop_done, the ptracer went away between group stop
2218 * completion and here. During detach, it would have set
2219 * JOBCTL_STOP_PENDING on us and we'll re-enter
2220 * TASK_STOPPED in do_signal_stop() on return, so notifying
2221 * the real parent of the group stop completion is enough.
2222 */
2223 if (gstop_done)
2224 do_notify_parent_cldstop(current, false, why);
2225
2226 /* tasklist protects us from ptrace_freeze_traced() */
2227 __set_current_state(TASK_RUNNING);
2228 if (clear_code)
2229 current->exit_code = 0;
2230 read_unlock(&tasklist_lock);
2231 }
2232
2233 /*
2234 * We are back. Now reacquire the siglock before touching
2235 * last_siginfo, so that we are sure to have synchronized with
2236 * any signal-sending on another CPU that wants to examine it.
2237 */
2238 spin_lock_irq(¤t->sighand->siglock);
2239 current->last_siginfo = NULL;
2240
2241 /* LISTENING can be set only during STOP traps, clear it */
2242 current->jobctl &= ~JOBCTL_LISTENING;
2243
2244 /*
2245 * Queued signals ignored us while we were stopped for tracing.
2246 * So check for any that we should take before resuming user mode.
2247 * This sets TIF_SIGPENDING, but never clears it.
2248 */
2249 recalc_sigpending_tsk(current);
2250}
2251
2252static void ptrace_do_notify(int signr, int exit_code, int why)
2253{
2254 kernel_siginfo_t info;
2255
2256 clear_siginfo(&info);
2257 info.si_signo = signr;
2258 info.si_code = exit_code;
2259 info.si_pid = task_pid_vnr(current);
2260 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2261
2262 /* Let the debugger run. */
2263 ptrace_stop(exit_code, why, 1, &info);
2264}
2265
2266void ptrace_notify(int exit_code)
2267{
2268 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2269 if (unlikely(current->task_works))
2270 task_work_run();
2271
2272 spin_lock_irq(¤t->sighand->siglock);
2273 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2274 spin_unlock_irq(¤t->sighand->siglock);
2275}
2276
2277/**
2278 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2279 * @signr: signr causing group stop if initiating
2280 *
2281 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2282 * and participate in it. If already set, participate in the existing
2283 * group stop. If participated in a group stop (and thus slept), %true is
2284 * returned with siglock released.
2285 *
2286 * If ptraced, this function doesn't handle stop itself. Instead,
2287 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2288 * untouched. The caller must ensure that INTERRUPT trap handling takes
2289 * places afterwards.
2290 *
2291 * CONTEXT:
2292 * Must be called with @current->sighand->siglock held, which is released
2293 * on %true return.
2294 *
2295 * RETURNS:
2296 * %false if group stop is already cancelled or ptrace trap is scheduled.
2297 * %true if participated in group stop.
2298 */
2299static bool do_signal_stop(int signr)
2300 __releases(¤t->sighand->siglock)
2301{
2302 struct signal_struct *sig = current->signal;
2303
2304 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2305 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2306 struct task_struct *t;
2307
2308 /* signr will be recorded in task->jobctl for retries */
2309 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2310
2311 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2312 unlikely(signal_group_exit(sig)))
2313 return false;
2314 /*
2315 * There is no group stop already in progress. We must
2316 * initiate one now.
2317 *
2318 * While ptraced, a task may be resumed while group stop is
2319 * still in effect and then receive a stop signal and
2320 * initiate another group stop. This deviates from the
2321 * usual behavior as two consecutive stop signals can't
2322 * cause two group stops when !ptraced. That is why we
2323 * also check !task_is_stopped(t) below.
2324 *
2325 * The condition can be distinguished by testing whether
2326 * SIGNAL_STOP_STOPPED is already set. Don't generate
2327 * group_exit_code in such case.
2328 *
2329 * This is not necessary for SIGNAL_STOP_CONTINUED because
2330 * an intervening stop signal is required to cause two
2331 * continued events regardless of ptrace.
2332 */
2333 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2334 sig->group_exit_code = signr;
2335
2336 sig->group_stop_count = 0;
2337
2338 if (task_set_jobctl_pending(current, signr | gstop))
2339 sig->group_stop_count++;
2340
2341 t = current;
2342 while_each_thread(current, t) {
2343 /*
2344 * Setting state to TASK_STOPPED for a group
2345 * stop is always done with the siglock held,
2346 * so this check has no races.
2347 */
2348 if (!task_is_stopped(t) &&
2349 task_set_jobctl_pending(t, signr | gstop)) {
2350 sig->group_stop_count++;
2351 if (likely(!(t->ptrace & PT_SEIZED)))
2352 signal_wake_up(t, 0);
2353 else
2354 ptrace_trap_notify(t);
2355 }
2356 }
2357 }
2358
2359 if (likely(!current->ptrace)) {
2360 int notify = 0;
2361
2362 /*
2363 * If there are no other threads in the group, or if there
2364 * is a group stop in progress and we are the last to stop,
2365 * report to the parent.
2366 */
2367 if (task_participate_group_stop(current))
2368 notify = CLD_STOPPED;
2369
2370 set_special_state(TASK_STOPPED);
2371 spin_unlock_irq(¤t->sighand->siglock);
2372
2373 /*
2374 * Notify the parent of the group stop completion. Because
2375 * we're not holding either the siglock or tasklist_lock
2376 * here, ptracer may attach inbetween; however, this is for
2377 * group stop and should always be delivered to the real
2378 * parent of the group leader. The new ptracer will get
2379 * its notification when this task transitions into
2380 * TASK_TRACED.
2381 */
2382 if (notify) {
2383 read_lock(&tasklist_lock);
2384 do_notify_parent_cldstop(current, false, notify);
2385 read_unlock(&tasklist_lock);
2386 }
2387
2388 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2389 cgroup_enter_frozen();
2390 freezable_schedule();
2391 return true;
2392 } else {
2393 /*
2394 * While ptraced, group stop is handled by STOP trap.
2395 * Schedule it and let the caller deal with it.
2396 */
2397 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2398 return false;
2399 }
2400}
2401
2402/**
2403 * do_jobctl_trap - take care of ptrace jobctl traps
2404 *
2405 * When PT_SEIZED, it's used for both group stop and explicit
2406 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2407 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2408 * the stop signal; otherwise, %SIGTRAP.
2409 *
2410 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2411 * number as exit_code and no siginfo.
2412 *
2413 * CONTEXT:
2414 * Must be called with @current->sighand->siglock held, which may be
2415 * released and re-acquired before returning with intervening sleep.
2416 */
2417static void do_jobctl_trap(void)
2418{
2419 struct signal_struct *signal = current->signal;
2420 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2421
2422 if (current->ptrace & PT_SEIZED) {
2423 if (!signal->group_stop_count &&
2424 !(signal->flags & SIGNAL_STOP_STOPPED))
2425 signr = SIGTRAP;
2426 WARN_ON_ONCE(!signr);
2427 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2428 CLD_STOPPED);
2429 } else {
2430 WARN_ON_ONCE(!signr);
2431 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2432 current->exit_code = 0;
2433 }
2434}
2435
2436/**
2437 * do_freezer_trap - handle the freezer jobctl trap
2438 *
2439 * Puts the task into frozen state, if only the task is not about to quit.
2440 * In this case it drops JOBCTL_TRAP_FREEZE.
2441 *
2442 * CONTEXT:
2443 * Must be called with @current->sighand->siglock held,
2444 * which is always released before returning.
2445 */
2446static void do_freezer_trap(void)
2447 __releases(¤t->sighand->siglock)
2448{
2449 /*
2450 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2451 * let's make another loop to give it a chance to be handled.
2452 * In any case, we'll return back.
2453 */
2454 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2455 JOBCTL_TRAP_FREEZE) {
2456 spin_unlock_irq(¤t->sighand->siglock);
2457 return;
2458 }
2459
2460 /*
2461 * Now we're sure that there is no pending fatal signal and no
2462 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2463 * immediately (if there is a non-fatal signal pending), and
2464 * put the task into sleep.
2465 */
2466 __set_current_state(TASK_INTERRUPTIBLE);
2467 clear_thread_flag(TIF_SIGPENDING);
2468 spin_unlock_irq(¤t->sighand->siglock);
2469 cgroup_enter_frozen();
2470 freezable_schedule();
2471}
2472
2473static int ptrace_signal(int signr, kernel_siginfo_t *info)
2474{
2475 /*
2476 * We do not check sig_kernel_stop(signr) but set this marker
2477 * unconditionally because we do not know whether debugger will
2478 * change signr. This flag has no meaning unless we are going
2479 * to stop after return from ptrace_stop(). In this case it will
2480 * be checked in do_signal_stop(), we should only stop if it was
2481 * not cleared by SIGCONT while we were sleeping. See also the
2482 * comment in dequeue_signal().
2483 */
2484 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2485 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2486
2487 /* We're back. Did the debugger cancel the sig? */
2488 signr = current->exit_code;
2489 if (signr == 0)
2490 return signr;
2491
2492 current->exit_code = 0;
2493
2494 /*
2495 * Update the siginfo structure if the signal has
2496 * changed. If the debugger wanted something
2497 * specific in the siginfo structure then it should
2498 * have updated *info via PTRACE_SETSIGINFO.
2499 */
2500 if (signr != info->si_signo) {
2501 clear_siginfo(info);
2502 info->si_signo = signr;
2503 info->si_errno = 0;
2504 info->si_code = SI_USER;
2505 rcu_read_lock();
2506 info->si_pid = task_pid_vnr(current->parent);
2507 info->si_uid = from_kuid_munged(current_user_ns(),
2508 task_uid(current->parent));
2509 rcu_read_unlock();
2510 }
2511
2512 /* If the (new) signal is now blocked, requeue it. */
2513 if (sigismember(¤t->blocked, signr)) {
2514 send_signal(signr, info, current, PIDTYPE_PID);
2515 signr = 0;
2516 }
2517
2518 return signr;
2519}
2520
2521bool get_signal(struct ksignal *ksig)
2522{
2523 struct sighand_struct *sighand = current->sighand;
2524 struct signal_struct *signal = current->signal;
2525 int signr;
2526
2527 if (unlikely(current->task_works))
2528 task_work_run();
2529
2530 if (unlikely(uprobe_deny_signal()))
2531 return false;
2532
2533 /*
2534 * Do this once, we can't return to user-mode if freezing() == T.
2535 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2536 * thus do not need another check after return.
2537 */
2538 try_to_freeze();
2539
2540relock:
2541 spin_lock_irq(&sighand->siglock);
2542 /*
2543 * Every stopped thread goes here after wakeup. Check to see if
2544 * we should notify the parent, prepare_signal(SIGCONT) encodes
2545 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2546 */
2547 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2548 int why;
2549
2550 if (signal->flags & SIGNAL_CLD_CONTINUED)
2551 why = CLD_CONTINUED;
2552 else
2553 why = CLD_STOPPED;
2554
2555 signal->flags &= ~SIGNAL_CLD_MASK;
2556
2557 spin_unlock_irq(&sighand->siglock);
2558
2559 /*
2560 * Notify the parent that we're continuing. This event is
2561 * always per-process and doesn't make whole lot of sense
2562 * for ptracers, who shouldn't consume the state via
2563 * wait(2) either, but, for backward compatibility, notify
2564 * the ptracer of the group leader too unless it's gonna be
2565 * a duplicate.
2566 */
2567 read_lock(&tasklist_lock);
2568 do_notify_parent_cldstop(current, false, why);
2569
2570 if (ptrace_reparented(current->group_leader))
2571 do_notify_parent_cldstop(current->group_leader,
2572 true, why);
2573 read_unlock(&tasklist_lock);
2574
2575 goto relock;
2576 }
2577
2578 /* Has this task already been marked for death? */
2579 if (signal_group_exit(signal)) {
2580 ksig->info.si_signo = signr = SIGKILL;
2581 sigdelset(¤t->pending.signal, SIGKILL);
2582 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2583 &sighand->action[SIGKILL - 1]);
2584 recalc_sigpending();
2585 goto fatal;
2586 }
2587
2588 for (;;) {
2589 struct k_sigaction *ka;
2590
2591 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2592 do_signal_stop(0))
2593 goto relock;
2594
2595 if (unlikely(current->jobctl &
2596 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2597 if (current->jobctl & JOBCTL_TRAP_MASK) {
2598 do_jobctl_trap();
2599 spin_unlock_irq(&sighand->siglock);
2600 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2601 do_freezer_trap();
2602
2603 goto relock;
2604 }
2605
2606 /*
2607 * If the task is leaving the frozen state, let's update
2608 * cgroup counters and reset the frozen bit.
2609 */
2610 if (unlikely(cgroup_task_frozen(current))) {
2611 spin_unlock_irq(&sighand->siglock);
2612 cgroup_leave_frozen(false);
2613 goto relock;
2614 }
2615
2616 /*
2617 * Signals generated by the execution of an instruction
2618 * need to be delivered before any other pending signals
2619 * so that the instruction pointer in the signal stack
2620 * frame points to the faulting instruction.
2621 */
2622 signr = dequeue_synchronous_signal(&ksig->info);
2623 if (!signr)
2624 signr = dequeue_signal(current, ¤t->blocked, &ksig->info);
2625
2626 if (!signr)
2627 break; /* will return 0 */
2628
2629 if (unlikely(current->ptrace) && signr != SIGKILL) {
2630 signr = ptrace_signal(signr, &ksig->info);
2631 if (!signr)
2632 continue;
2633 }
2634
2635 ka = &sighand->action[signr-1];
2636
2637 /* Trace actually delivered signals. */
2638 trace_signal_deliver(signr, &ksig->info, ka);
2639
2640 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2641 continue;
2642 if (ka->sa.sa_handler != SIG_DFL) {
2643 /* Run the handler. */
2644 ksig->ka = *ka;
2645
2646 if (ka->sa.sa_flags & SA_ONESHOT)
2647 ka->sa.sa_handler = SIG_DFL;
2648
2649 break; /* will return non-zero "signr" value */
2650 }
2651
2652 /*
2653 * Now we are doing the default action for this signal.
2654 */
2655 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2656 continue;
2657
2658 /*
2659 * Global init gets no signals it doesn't want.
2660 * Container-init gets no signals it doesn't want from same
2661 * container.
2662 *
2663 * Note that if global/container-init sees a sig_kernel_only()
2664 * signal here, the signal must have been generated internally
2665 * or must have come from an ancestor namespace. In either
2666 * case, the signal cannot be dropped.
2667 */
2668 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2669 !sig_kernel_only(signr))
2670 continue;
2671
2672 if (sig_kernel_stop(signr)) {
2673 /*
2674 * The default action is to stop all threads in
2675 * the thread group. The job control signals
2676 * do nothing in an orphaned pgrp, but SIGSTOP
2677 * always works. Note that siglock needs to be
2678 * dropped during the call to is_orphaned_pgrp()
2679 * because of lock ordering with tasklist_lock.
2680 * This allows an intervening SIGCONT to be posted.
2681 * We need to check for that and bail out if necessary.
2682 */
2683 if (signr != SIGSTOP) {
2684 spin_unlock_irq(&sighand->siglock);
2685
2686 /* signals can be posted during this window */
2687
2688 if (is_current_pgrp_orphaned())
2689 goto relock;
2690
2691 spin_lock_irq(&sighand->siglock);
2692 }
2693
2694 if (likely(do_signal_stop(ksig->info.si_signo))) {
2695 /* It released the siglock. */
2696 goto relock;
2697 }
2698
2699 /*
2700 * We didn't actually stop, due to a race
2701 * with SIGCONT or something like that.
2702 */
2703 continue;
2704 }
2705
2706 fatal:
2707 spin_unlock_irq(&sighand->siglock);
2708 if (unlikely(cgroup_task_frozen(current)))
2709 cgroup_leave_frozen(true);
2710
2711 /*
2712 * Anything else is fatal, maybe with a core dump.
2713 */
2714 current->flags |= PF_SIGNALED;
2715
2716 if (sig_kernel_coredump(signr)) {
2717 if (print_fatal_signals)
2718 print_fatal_signal(ksig->info.si_signo);
2719 proc_coredump_connector(current);
2720 /*
2721 * If it was able to dump core, this kills all
2722 * other threads in the group and synchronizes with
2723 * their demise. If we lost the race with another
2724 * thread getting here, it set group_exit_code
2725 * first and our do_group_exit call below will use
2726 * that value and ignore the one we pass it.
2727 */
2728 do_coredump(&ksig->info);
2729 }
2730
2731 /*
2732 * Death signals, no core dump.
2733 */
2734 do_group_exit(ksig->info.si_signo);
2735 /* NOTREACHED */
2736 }
2737 spin_unlock_irq(&sighand->siglock);
2738
2739 ksig->sig = signr;
2740 return ksig->sig > 0;
2741}
2742
2743/**
2744 * signal_delivered -
2745 * @ksig: kernel signal struct
2746 * @stepping: nonzero if debugger single-step or block-step in use
2747 *
2748 * This function should be called when a signal has successfully been
2749 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2750 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2751 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2752 */
2753static void signal_delivered(struct ksignal *ksig, int stepping)
2754{
2755 sigset_t blocked;
2756
2757 /* A signal was successfully delivered, and the
2758 saved sigmask was stored on the signal frame,
2759 and will be restored by sigreturn. So we can
2760 simply clear the restore sigmask flag. */
2761 clear_restore_sigmask();
2762
2763 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2764 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2765 sigaddset(&blocked, ksig->sig);
2766 set_current_blocked(&blocked);
2767 tracehook_signal_handler(stepping);
2768}
2769
2770void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2771{
2772 if (failed)
2773 force_sigsegv(ksig->sig);
2774 else
2775 signal_delivered(ksig, stepping);
2776}
2777
2778/*
2779 * It could be that complete_signal() picked us to notify about the
2780 * group-wide signal. Other threads should be notified now to take
2781 * the shared signals in @which since we will not.
2782 */
2783static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2784{
2785 sigset_t retarget;
2786 struct task_struct *t;
2787
2788 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2789 if (sigisemptyset(&retarget))
2790 return;
2791
2792 t = tsk;
2793 while_each_thread(tsk, t) {
2794 if (t->flags & PF_EXITING)
2795 continue;
2796
2797 if (!has_pending_signals(&retarget, &t->blocked))
2798 continue;
2799 /* Remove the signals this thread can handle. */
2800 sigandsets(&retarget, &retarget, &t->blocked);
2801
2802 if (!signal_pending(t))
2803 signal_wake_up(t, 0);
2804
2805 if (sigisemptyset(&retarget))
2806 break;
2807 }
2808}
2809
2810void exit_signals(struct task_struct *tsk)
2811{
2812 int group_stop = 0;
2813 sigset_t unblocked;
2814
2815 /*
2816 * @tsk is about to have PF_EXITING set - lock out users which
2817 * expect stable threadgroup.
2818 */
2819 cgroup_threadgroup_change_begin(tsk);
2820
2821 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2822 tsk->flags |= PF_EXITING;
2823 cgroup_threadgroup_change_end(tsk);
2824 return;
2825 }
2826
2827 spin_lock_irq(&tsk->sighand->siglock);
2828 /*
2829 * From now this task is not visible for group-wide signals,
2830 * see wants_signal(), do_signal_stop().
2831 */
2832 tsk->flags |= PF_EXITING;
2833
2834 cgroup_threadgroup_change_end(tsk);
2835
2836 if (!signal_pending(tsk))
2837 goto out;
2838
2839 unblocked = tsk->blocked;
2840 signotset(&unblocked);
2841 retarget_shared_pending(tsk, &unblocked);
2842
2843 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2844 task_participate_group_stop(tsk))
2845 group_stop = CLD_STOPPED;
2846out:
2847 spin_unlock_irq(&tsk->sighand->siglock);
2848
2849 /*
2850 * If group stop has completed, deliver the notification. This
2851 * should always go to the real parent of the group leader.
2852 */
2853 if (unlikely(group_stop)) {
2854 read_lock(&tasklist_lock);
2855 do_notify_parent_cldstop(tsk, false, group_stop);
2856 read_unlock(&tasklist_lock);
2857 }
2858}
2859
2860/*
2861 * System call entry points.
2862 */
2863
2864/**
2865 * sys_restart_syscall - restart a system call
2866 */
2867SYSCALL_DEFINE0(restart_syscall)
2868{
2869 struct restart_block *restart = ¤t->restart_block;
2870 return restart->fn(restart);
2871}
2872
2873long do_no_restart_syscall(struct restart_block *param)
2874{
2875 return -EINTR;
2876}
2877
2878static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2879{
2880 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2881 sigset_t newblocked;
2882 /* A set of now blocked but previously unblocked signals. */
2883 sigandnsets(&newblocked, newset, ¤t->blocked);
2884 retarget_shared_pending(tsk, &newblocked);
2885 }
2886 tsk->blocked = *newset;
2887 recalc_sigpending();
2888}
2889
2890/**
2891 * set_current_blocked - change current->blocked mask
2892 * @newset: new mask
2893 *
2894 * It is wrong to change ->blocked directly, this helper should be used
2895 * to ensure the process can't miss a shared signal we are going to block.
2896 */
2897void set_current_blocked(sigset_t *newset)
2898{
2899 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2900 __set_current_blocked(newset);
2901}
2902
2903void __set_current_blocked(const sigset_t *newset)
2904{
2905 struct task_struct *tsk = current;
2906
2907 /*
2908 * In case the signal mask hasn't changed, there is nothing we need
2909 * to do. The current->blocked shouldn't be modified by other task.
2910 */
2911 if (sigequalsets(&tsk->blocked, newset))
2912 return;
2913
2914 spin_lock_irq(&tsk->sighand->siglock);
2915 __set_task_blocked(tsk, newset);
2916 spin_unlock_irq(&tsk->sighand->siglock);
2917}
2918
2919/*
2920 * This is also useful for kernel threads that want to temporarily
2921 * (or permanently) block certain signals.
2922 *
2923 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2924 * interface happily blocks "unblockable" signals like SIGKILL
2925 * and friends.
2926 */
2927int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2928{
2929 struct task_struct *tsk = current;
2930 sigset_t newset;
2931
2932 /* Lockless, only current can change ->blocked, never from irq */
2933 if (oldset)
2934 *oldset = tsk->blocked;
2935
2936 switch (how) {
2937 case SIG_BLOCK:
2938 sigorsets(&newset, &tsk->blocked, set);
2939 break;
2940 case SIG_UNBLOCK:
2941 sigandnsets(&newset, &tsk->blocked, set);
2942 break;
2943 case SIG_SETMASK:
2944 newset = *set;
2945 break;
2946 default:
2947 return -EINVAL;
2948 }
2949
2950 __set_current_blocked(&newset);
2951 return 0;
2952}
2953EXPORT_SYMBOL(sigprocmask);
2954
2955/*
2956 * The api helps set app-provided sigmasks.
2957 *
2958 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
2959 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
2960 *
2961 * Note that it does set_restore_sigmask() in advance, so it must be always
2962 * paired with restore_saved_sigmask_unless() before return from syscall.
2963 */
2964int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
2965{
2966 sigset_t kmask;
2967
2968 if (!umask)
2969 return 0;
2970 if (sigsetsize != sizeof(sigset_t))
2971 return -EINVAL;
2972 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
2973 return -EFAULT;
2974
2975 set_restore_sigmask();
2976 current->saved_sigmask = current->blocked;
2977 set_current_blocked(&kmask);
2978
2979 return 0;
2980}
2981
2982#ifdef CONFIG_COMPAT
2983int set_compat_user_sigmask(const compat_sigset_t __user *umask,
2984 size_t sigsetsize)
2985{
2986 sigset_t kmask;
2987
2988 if (!umask)
2989 return 0;
2990 if (sigsetsize != sizeof(compat_sigset_t))
2991 return -EINVAL;
2992 if (get_compat_sigset(&kmask, umask))
2993 return -EFAULT;
2994
2995 set_restore_sigmask();
2996 current->saved_sigmask = current->blocked;
2997 set_current_blocked(&kmask);
2998
2999 return 0;
3000}
3001#endif
3002
3003/**
3004 * sys_rt_sigprocmask - change the list of currently blocked signals
3005 * @how: whether to add, remove, or set signals
3006 * @nset: stores pending signals
3007 * @oset: previous value of signal mask if non-null
3008 * @sigsetsize: size of sigset_t type
3009 */
3010SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3011 sigset_t __user *, oset, size_t, sigsetsize)
3012{
3013 sigset_t old_set, new_set;
3014 int error;
3015
3016 /* XXX: Don't preclude handling different sized sigset_t's. */
3017 if (sigsetsize != sizeof(sigset_t))
3018 return -EINVAL;
3019
3020 old_set = current->blocked;
3021
3022 if (nset) {
3023 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3024 return -EFAULT;
3025 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3026
3027 error = sigprocmask(how, &new_set, NULL);
3028 if (error)
3029 return error;
3030 }
3031
3032 if (oset) {
3033 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3034 return -EFAULT;
3035 }
3036
3037 return 0;
3038}
3039
3040#ifdef CONFIG_COMPAT
3041COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3042 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3043{
3044 sigset_t old_set = current->blocked;
3045
3046 /* XXX: Don't preclude handling different sized sigset_t's. */
3047 if (sigsetsize != sizeof(sigset_t))
3048 return -EINVAL;
3049
3050 if (nset) {
3051 sigset_t new_set;
3052 int error;
3053 if (get_compat_sigset(&new_set, nset))
3054 return -EFAULT;
3055 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3056
3057 error = sigprocmask(how, &new_set, NULL);
3058 if (error)
3059 return error;
3060 }
3061 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3062}
3063#endif
3064
3065static void do_sigpending(sigset_t *set)
3066{
3067 spin_lock_irq(¤t->sighand->siglock);
3068 sigorsets(set, ¤t->pending.signal,
3069 ¤t->signal->shared_pending.signal);
3070 spin_unlock_irq(¤t->sighand->siglock);
3071
3072 /* Outside the lock because only this thread touches it. */
3073 sigandsets(set, ¤t->blocked, set);
3074}
3075
3076/**
3077 * sys_rt_sigpending - examine a pending signal that has been raised
3078 * while blocked
3079 * @uset: stores pending signals
3080 * @sigsetsize: size of sigset_t type or larger
3081 */
3082SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3083{
3084 sigset_t set;
3085
3086 if (sigsetsize > sizeof(*uset))
3087 return -EINVAL;
3088
3089 do_sigpending(&set);
3090
3091 if (copy_to_user(uset, &set, sigsetsize))
3092 return -EFAULT;
3093
3094 return 0;
3095}
3096
3097#ifdef CONFIG_COMPAT
3098COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3099 compat_size_t, sigsetsize)
3100{
3101 sigset_t set;
3102
3103 if (sigsetsize > sizeof(*uset))
3104 return -EINVAL;
3105
3106 do_sigpending(&set);
3107
3108 return put_compat_sigset(uset, &set, sigsetsize);
3109}
3110#endif
3111
3112static const struct {
3113 unsigned char limit, layout;
3114} sig_sicodes[] = {
3115 [SIGILL] = { NSIGILL, SIL_FAULT },
3116 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3117 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3118 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3119 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3120#if defined(SIGEMT)
3121 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3122#endif
3123 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3124 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3125 [SIGSYS] = { NSIGSYS, SIL_SYS },
3126};
3127
3128static bool known_siginfo_layout(unsigned sig, int si_code)
3129{
3130 if (si_code == SI_KERNEL)
3131 return true;
3132 else if ((si_code > SI_USER)) {
3133 if (sig_specific_sicodes(sig)) {
3134 if (si_code <= sig_sicodes[sig].limit)
3135 return true;
3136 }
3137 else if (si_code <= NSIGPOLL)
3138 return true;
3139 }
3140 else if (si_code >= SI_DETHREAD)
3141 return true;
3142 else if (si_code == SI_ASYNCNL)
3143 return true;
3144 return false;
3145}
3146
3147enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3148{
3149 enum siginfo_layout layout = SIL_KILL;
3150 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3151 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3152 (si_code <= sig_sicodes[sig].limit)) {
3153 layout = sig_sicodes[sig].layout;
3154 /* Handle the exceptions */
3155 if ((sig == SIGBUS) &&
3156 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3157 layout = SIL_FAULT_MCEERR;
3158 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3159 layout = SIL_FAULT_BNDERR;
3160#ifdef SEGV_PKUERR
3161 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3162 layout = SIL_FAULT_PKUERR;
3163#endif
3164 }
3165 else if (si_code <= NSIGPOLL)
3166 layout = SIL_POLL;
3167 } else {
3168 if (si_code == SI_TIMER)
3169 layout = SIL_TIMER;
3170 else if (si_code == SI_SIGIO)
3171 layout = SIL_POLL;
3172 else if (si_code < 0)
3173 layout = SIL_RT;
3174 }
3175 return layout;
3176}
3177
3178static inline char __user *si_expansion(const siginfo_t __user *info)
3179{
3180 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3181}
3182
3183int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3184{
3185 char __user *expansion = si_expansion(to);
3186 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3187 return -EFAULT;
3188 if (clear_user(expansion, SI_EXPANSION_SIZE))
3189 return -EFAULT;
3190 return 0;
3191}
3192
3193static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3194 const siginfo_t __user *from)
3195{
3196 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3197 char __user *expansion = si_expansion(from);
3198 char buf[SI_EXPANSION_SIZE];
3199 int i;
3200 /*
3201 * An unknown si_code might need more than
3202 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3203 * extra bytes are 0. This guarantees copy_siginfo_to_user
3204 * will return this data to userspace exactly.
3205 */
3206 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3207 return -EFAULT;
3208 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3209 if (buf[i] != 0)
3210 return -E2BIG;
3211 }
3212 }
3213 return 0;
3214}
3215
3216static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3217 const siginfo_t __user *from)
3218{
3219 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3220 return -EFAULT;
3221 to->si_signo = signo;
3222 return post_copy_siginfo_from_user(to, from);
3223}
3224
3225int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3226{
3227 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3228 return -EFAULT;
3229 return post_copy_siginfo_from_user(to, from);
3230}
3231
3232#ifdef CONFIG_COMPAT
3233int copy_siginfo_to_user32(struct compat_siginfo __user *to,
3234 const struct kernel_siginfo *from)
3235#if defined(CONFIG_X86_X32_ABI) || defined(CONFIG_IA32_EMULATION)
3236{
3237 return __copy_siginfo_to_user32(to, from, in_x32_syscall());
3238}
3239int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3240 const struct kernel_siginfo *from, bool x32_ABI)
3241#endif
3242{
3243 struct compat_siginfo new;
3244 memset(&new, 0, sizeof(new));
3245
3246 new.si_signo = from->si_signo;
3247 new.si_errno = from->si_errno;
3248 new.si_code = from->si_code;
3249 switch(siginfo_layout(from->si_signo, from->si_code)) {
3250 case SIL_KILL:
3251 new.si_pid = from->si_pid;
3252 new.si_uid = from->si_uid;
3253 break;
3254 case SIL_TIMER:
3255 new.si_tid = from->si_tid;
3256 new.si_overrun = from->si_overrun;
3257 new.si_int = from->si_int;
3258 break;
3259 case SIL_POLL:
3260 new.si_band = from->si_band;
3261 new.si_fd = from->si_fd;
3262 break;
3263 case SIL_FAULT:
3264 new.si_addr = ptr_to_compat(from->si_addr);
3265#ifdef __ARCH_SI_TRAPNO
3266 new.si_trapno = from->si_trapno;
3267#endif
3268 break;
3269 case SIL_FAULT_MCEERR:
3270 new.si_addr = ptr_to_compat(from->si_addr);
3271#ifdef __ARCH_SI_TRAPNO
3272 new.si_trapno = from->si_trapno;
3273#endif
3274 new.si_addr_lsb = from->si_addr_lsb;
3275 break;
3276 case SIL_FAULT_BNDERR:
3277 new.si_addr = ptr_to_compat(from->si_addr);
3278#ifdef __ARCH_SI_TRAPNO
3279 new.si_trapno = from->si_trapno;
3280#endif
3281 new.si_lower = ptr_to_compat(from->si_lower);
3282 new.si_upper = ptr_to_compat(from->si_upper);
3283 break;
3284 case SIL_FAULT_PKUERR:
3285 new.si_addr = ptr_to_compat(from->si_addr);
3286#ifdef __ARCH_SI_TRAPNO
3287 new.si_trapno = from->si_trapno;
3288#endif
3289 new.si_pkey = from->si_pkey;
3290 break;
3291 case SIL_CHLD:
3292 new.si_pid = from->si_pid;
3293 new.si_uid = from->si_uid;
3294 new.si_status = from->si_status;
3295#ifdef CONFIG_X86_X32_ABI
3296 if (x32_ABI) {
3297 new._sifields._sigchld_x32._utime = from->si_utime;
3298 new._sifields._sigchld_x32._stime = from->si_stime;
3299 } else
3300#endif
3301 {
3302 new.si_utime = from->si_utime;
3303 new.si_stime = from->si_stime;
3304 }
3305 break;
3306 case SIL_RT:
3307 new.si_pid = from->si_pid;
3308 new.si_uid = from->si_uid;
3309 new.si_int = from->si_int;
3310 break;
3311 case SIL_SYS:
3312 new.si_call_addr = ptr_to_compat(from->si_call_addr);
3313 new.si_syscall = from->si_syscall;
3314 new.si_arch = from->si_arch;
3315 break;
3316 }
3317
3318 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3319 return -EFAULT;
3320
3321 return 0;
3322}
3323
3324static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3325 const struct compat_siginfo *from)
3326{
3327 clear_siginfo(to);
3328 to->si_signo = from->si_signo;
3329 to->si_errno = from->si_errno;
3330 to->si_code = from->si_code;
3331 switch(siginfo_layout(from->si_signo, from->si_code)) {
3332 case SIL_KILL:
3333 to->si_pid = from->si_pid;
3334 to->si_uid = from->si_uid;
3335 break;
3336 case SIL_TIMER:
3337 to->si_tid = from->si_tid;
3338 to->si_overrun = from->si_overrun;
3339 to->si_int = from->si_int;
3340 break;
3341 case SIL_POLL:
3342 to->si_band = from->si_band;
3343 to->si_fd = from->si_fd;
3344 break;
3345 case SIL_FAULT:
3346 to->si_addr = compat_ptr(from->si_addr);
3347#ifdef __ARCH_SI_TRAPNO
3348 to->si_trapno = from->si_trapno;
3349#endif
3350 break;
3351 case SIL_FAULT_MCEERR:
3352 to->si_addr = compat_ptr(from->si_addr);
3353#ifdef __ARCH_SI_TRAPNO
3354 to->si_trapno = from->si_trapno;
3355#endif
3356 to->si_addr_lsb = from->si_addr_lsb;
3357 break;
3358 case SIL_FAULT_BNDERR:
3359 to->si_addr = compat_ptr(from->si_addr);
3360#ifdef __ARCH_SI_TRAPNO
3361 to->si_trapno = from->si_trapno;
3362#endif
3363 to->si_lower = compat_ptr(from->si_lower);
3364 to->si_upper = compat_ptr(from->si_upper);
3365 break;
3366 case SIL_FAULT_PKUERR:
3367 to->si_addr = compat_ptr(from->si_addr);
3368#ifdef __ARCH_SI_TRAPNO
3369 to->si_trapno = from->si_trapno;
3370#endif
3371 to->si_pkey = from->si_pkey;
3372 break;
3373 case SIL_CHLD:
3374 to->si_pid = from->si_pid;
3375 to->si_uid = from->si_uid;
3376 to->si_status = from->si_status;
3377#ifdef CONFIG_X86_X32_ABI
3378 if (in_x32_syscall()) {
3379 to->si_utime = from->_sifields._sigchld_x32._utime;
3380 to->si_stime = from->_sifields._sigchld_x32._stime;
3381 } else
3382#endif
3383 {
3384 to->si_utime = from->si_utime;
3385 to->si_stime = from->si_stime;
3386 }
3387 break;
3388 case SIL_RT:
3389 to->si_pid = from->si_pid;
3390 to->si_uid = from->si_uid;
3391 to->si_int = from->si_int;
3392 break;
3393 case SIL_SYS:
3394 to->si_call_addr = compat_ptr(from->si_call_addr);
3395 to->si_syscall = from->si_syscall;
3396 to->si_arch = from->si_arch;
3397 break;
3398 }
3399 return 0;
3400}
3401
3402static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3403 const struct compat_siginfo __user *ufrom)
3404{
3405 struct compat_siginfo from;
3406
3407 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3408 return -EFAULT;
3409
3410 from.si_signo = signo;
3411 return post_copy_siginfo_from_user32(to, &from);
3412}
3413
3414int copy_siginfo_from_user32(struct kernel_siginfo *to,
3415 const struct compat_siginfo __user *ufrom)
3416{
3417 struct compat_siginfo from;
3418
3419 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3420 return -EFAULT;
3421
3422 return post_copy_siginfo_from_user32(to, &from);
3423}
3424#endif /* CONFIG_COMPAT */
3425
3426/**
3427 * do_sigtimedwait - wait for queued signals specified in @which
3428 * @which: queued signals to wait for
3429 * @info: if non-null, the signal's siginfo is returned here
3430 * @ts: upper bound on process time suspension
3431 */
3432static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3433 const struct timespec64 *ts)
3434{
3435 ktime_t *to = NULL, timeout = KTIME_MAX;
3436 struct task_struct *tsk = current;
3437 sigset_t mask = *which;
3438 int sig, ret = 0;
3439
3440 if (ts) {
3441 if (!timespec64_valid(ts))
3442 return -EINVAL;
3443 timeout = timespec64_to_ktime(*ts);
3444 to = &timeout;
3445 }
3446
3447 /*
3448 * Invert the set of allowed signals to get those we want to block.
3449 */
3450 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3451 signotset(&mask);
3452
3453 spin_lock_irq(&tsk->sighand->siglock);
3454 sig = dequeue_signal(tsk, &mask, info);
3455 if (!sig && timeout) {
3456 /*
3457 * None ready, temporarily unblock those we're interested
3458 * while we are sleeping in so that we'll be awakened when
3459 * they arrive. Unblocking is always fine, we can avoid
3460 * set_current_blocked().
3461 */
3462 tsk->real_blocked = tsk->blocked;
3463 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3464 recalc_sigpending();
3465 spin_unlock_irq(&tsk->sighand->siglock);
3466
3467 __set_current_state(TASK_INTERRUPTIBLE);
3468 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3469 HRTIMER_MODE_REL);
3470 spin_lock_irq(&tsk->sighand->siglock);
3471 __set_task_blocked(tsk, &tsk->real_blocked);
3472 sigemptyset(&tsk->real_blocked);
3473 sig = dequeue_signal(tsk, &mask, info);
3474 }
3475 spin_unlock_irq(&tsk->sighand->siglock);
3476
3477 if (sig)
3478 return sig;
3479 return ret ? -EINTR : -EAGAIN;
3480}
3481
3482/**
3483 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3484 * in @uthese
3485 * @uthese: queued signals to wait for
3486 * @uinfo: if non-null, the signal's siginfo is returned here
3487 * @uts: upper bound on process time suspension
3488 * @sigsetsize: size of sigset_t type
3489 */
3490SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3491 siginfo_t __user *, uinfo,
3492 const struct __kernel_timespec __user *, uts,
3493 size_t, sigsetsize)
3494{
3495 sigset_t these;
3496 struct timespec64 ts;
3497 kernel_siginfo_t info;
3498 int ret;
3499
3500 /* XXX: Don't preclude handling different sized sigset_t's. */
3501 if (sigsetsize != sizeof(sigset_t))
3502 return -EINVAL;
3503
3504 if (copy_from_user(&these, uthese, sizeof(these)))
3505 return -EFAULT;
3506
3507 if (uts) {
3508 if (get_timespec64(&ts, uts))
3509 return -EFAULT;
3510 }
3511
3512 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3513
3514 if (ret > 0 && uinfo) {
3515 if (copy_siginfo_to_user(uinfo, &info))
3516 ret = -EFAULT;
3517 }
3518
3519 return ret;
3520}
3521
3522#ifdef CONFIG_COMPAT_32BIT_TIME
3523SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3524 siginfo_t __user *, uinfo,
3525 const struct old_timespec32 __user *, uts,
3526 size_t, sigsetsize)
3527{
3528 sigset_t these;
3529 struct timespec64 ts;
3530 kernel_siginfo_t info;
3531 int ret;
3532
3533 if (sigsetsize != sizeof(sigset_t))
3534 return -EINVAL;
3535
3536 if (copy_from_user(&these, uthese, sizeof(these)))
3537 return -EFAULT;
3538
3539 if (uts) {
3540 if (get_old_timespec32(&ts, uts))
3541 return -EFAULT;
3542 }
3543
3544 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3545
3546 if (ret > 0 && uinfo) {
3547 if (copy_siginfo_to_user(uinfo, &info))
3548 ret = -EFAULT;
3549 }
3550
3551 return ret;
3552}
3553#endif
3554
3555#ifdef CONFIG_COMPAT
3556COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3557 struct compat_siginfo __user *, uinfo,
3558 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3559{
3560 sigset_t s;
3561 struct timespec64 t;
3562 kernel_siginfo_t info;
3563 long ret;
3564
3565 if (sigsetsize != sizeof(sigset_t))
3566 return -EINVAL;
3567
3568 if (get_compat_sigset(&s, uthese))
3569 return -EFAULT;
3570
3571 if (uts) {
3572 if (get_timespec64(&t, uts))
3573 return -EFAULT;
3574 }
3575
3576 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3577
3578 if (ret > 0 && uinfo) {
3579 if (copy_siginfo_to_user32(uinfo, &info))
3580 ret = -EFAULT;
3581 }
3582
3583 return ret;
3584}
3585
3586#ifdef CONFIG_COMPAT_32BIT_TIME
3587COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3588 struct compat_siginfo __user *, uinfo,
3589 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3590{
3591 sigset_t s;
3592 struct timespec64 t;
3593 kernel_siginfo_t info;
3594 long ret;
3595
3596 if (sigsetsize != sizeof(sigset_t))
3597 return -EINVAL;
3598
3599 if (get_compat_sigset(&s, uthese))
3600 return -EFAULT;
3601
3602 if (uts) {
3603 if (get_old_timespec32(&t, uts))
3604 return -EFAULT;
3605 }
3606
3607 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3608
3609 if (ret > 0 && uinfo) {
3610 if (copy_siginfo_to_user32(uinfo, &info))
3611 ret = -EFAULT;
3612 }
3613
3614 return ret;
3615}
3616#endif
3617#endif
3618
3619static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3620{
3621 clear_siginfo(info);
3622 info->si_signo = sig;
3623 info->si_errno = 0;
3624 info->si_code = SI_USER;
3625 info->si_pid = task_tgid_vnr(current);
3626 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3627}
3628
3629/**
3630 * sys_kill - send a signal to a process
3631 * @pid: the PID of the process
3632 * @sig: signal to be sent
3633 */
3634SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3635{
3636 struct kernel_siginfo info;
3637
3638 prepare_kill_siginfo(sig, &info);
3639
3640 return kill_something_info(sig, &info, pid);
3641}
3642
3643/*
3644 * Verify that the signaler and signalee either are in the same pid namespace
3645 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3646 * namespace.
3647 */
3648static bool access_pidfd_pidns(struct pid *pid)
3649{
3650 struct pid_namespace *active = task_active_pid_ns(current);
3651 struct pid_namespace *p = ns_of_pid(pid);
3652
3653 for (;;) {
3654 if (!p)
3655 return false;
3656 if (p == active)
3657 break;
3658 p = p->parent;
3659 }
3660
3661 return true;
3662}
3663
3664static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo, siginfo_t *info)
3665{
3666#ifdef CONFIG_COMPAT
3667 /*
3668 * Avoid hooking up compat syscalls and instead handle necessary
3669 * conversions here. Note, this is a stop-gap measure and should not be
3670 * considered a generic solution.
3671 */
3672 if (in_compat_syscall())
3673 return copy_siginfo_from_user32(
3674 kinfo, (struct compat_siginfo __user *)info);
3675#endif
3676 return copy_siginfo_from_user(kinfo, info);
3677}
3678
3679static struct pid *pidfd_to_pid(const struct file *file)
3680{
3681 struct pid *pid;
3682
3683 pid = pidfd_pid(file);
3684 if (!IS_ERR(pid))
3685 return pid;
3686
3687 return tgid_pidfd_to_pid(file);
3688}
3689
3690/**
3691 * sys_pidfd_send_signal - Signal a process through a pidfd
3692 * @pidfd: file descriptor of the process
3693 * @sig: signal to send
3694 * @info: signal info
3695 * @flags: future flags
3696 *
3697 * The syscall currently only signals via PIDTYPE_PID which covers
3698 * kill(<positive-pid>, <signal>. It does not signal threads or process
3699 * groups.
3700 * In order to extend the syscall to threads and process groups the @flags
3701 * argument should be used. In essence, the @flags argument will determine
3702 * what is signaled and not the file descriptor itself. Put in other words,
3703 * grouping is a property of the flags argument not a property of the file
3704 * descriptor.
3705 *
3706 * Return: 0 on success, negative errno on failure
3707 */
3708SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3709 siginfo_t __user *, info, unsigned int, flags)
3710{
3711 int ret;
3712 struct fd f;
3713 struct pid *pid;
3714 kernel_siginfo_t kinfo;
3715
3716 /* Enforce flags be set to 0 until we add an extension. */
3717 if (flags)
3718 return -EINVAL;
3719
3720 f = fdget(pidfd);
3721 if (!f.file)
3722 return -EBADF;
3723
3724 /* Is this a pidfd? */
3725 pid = pidfd_to_pid(f.file);
3726 if (IS_ERR(pid)) {
3727 ret = PTR_ERR(pid);
3728 goto err;
3729 }
3730
3731 ret = -EINVAL;
3732 if (!access_pidfd_pidns(pid))
3733 goto err;
3734
3735 if (info) {
3736 ret = copy_siginfo_from_user_any(&kinfo, info);
3737 if (unlikely(ret))
3738 goto err;
3739
3740 ret = -EINVAL;
3741 if (unlikely(sig != kinfo.si_signo))
3742 goto err;
3743
3744 /* Only allow sending arbitrary signals to yourself. */
3745 ret = -EPERM;
3746 if ((task_pid(current) != pid) &&
3747 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3748 goto err;
3749 } else {
3750 prepare_kill_siginfo(sig, &kinfo);
3751 }
3752
3753 ret = kill_pid_info(sig, &kinfo, pid);
3754
3755err:
3756 fdput(f);
3757 return ret;
3758}
3759
3760static int
3761do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3762{
3763 struct task_struct *p;
3764 int error = -ESRCH;
3765
3766 rcu_read_lock();
3767 p = find_task_by_vpid(pid);
3768 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3769 error = check_kill_permission(sig, info, p);
3770 /*
3771 * The null signal is a permissions and process existence
3772 * probe. No signal is actually delivered.
3773 */
3774 if (!error && sig) {
3775 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3776 /*
3777 * If lock_task_sighand() failed we pretend the task
3778 * dies after receiving the signal. The window is tiny,
3779 * and the signal is private anyway.
3780 */
3781 if (unlikely(error == -ESRCH))
3782 error = 0;
3783 }
3784 }
3785 rcu_read_unlock();
3786
3787 return error;
3788}
3789
3790static int do_tkill(pid_t tgid, pid_t pid, int sig)
3791{
3792 struct kernel_siginfo info;
3793
3794 clear_siginfo(&info);
3795 info.si_signo = sig;
3796 info.si_errno = 0;
3797 info.si_code = SI_TKILL;
3798 info.si_pid = task_tgid_vnr(current);
3799 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3800
3801 return do_send_specific(tgid, pid, sig, &info);
3802}
3803
3804/**
3805 * sys_tgkill - send signal to one specific thread
3806 * @tgid: the thread group ID of the thread
3807 * @pid: the PID of the thread
3808 * @sig: signal to be sent
3809 *
3810 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3811 * exists but it's not belonging to the target process anymore. This
3812 * method solves the problem of threads exiting and PIDs getting reused.
3813 */
3814SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3815{
3816 /* This is only valid for single tasks */
3817 if (pid <= 0 || tgid <= 0)
3818 return -EINVAL;
3819
3820 return do_tkill(tgid, pid, sig);
3821}
3822
3823/**
3824 * sys_tkill - send signal to one specific task
3825 * @pid: the PID of the task
3826 * @sig: signal to be sent
3827 *
3828 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3829 */
3830SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3831{
3832 /* This is only valid for single tasks */
3833 if (pid <= 0)
3834 return -EINVAL;
3835
3836 return do_tkill(0, pid, sig);
3837}
3838
3839static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3840{
3841 /* Not even root can pretend to send signals from the kernel.
3842 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3843 */
3844 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3845 (task_pid_vnr(current) != pid))
3846 return -EPERM;
3847
3848 /* POSIX.1b doesn't mention process groups. */
3849 return kill_proc_info(sig, info, pid);
3850}
3851
3852/**
3853 * sys_rt_sigqueueinfo - send signal information to a signal
3854 * @pid: the PID of the thread
3855 * @sig: signal to be sent
3856 * @uinfo: signal info to be sent
3857 */
3858SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3859 siginfo_t __user *, uinfo)
3860{
3861 kernel_siginfo_t info;
3862 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3863 if (unlikely(ret))
3864 return ret;
3865 return do_rt_sigqueueinfo(pid, sig, &info);
3866}
3867
3868#ifdef CONFIG_COMPAT
3869COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3870 compat_pid_t, pid,
3871 int, sig,
3872 struct compat_siginfo __user *, uinfo)
3873{
3874 kernel_siginfo_t info;
3875 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3876 if (unlikely(ret))
3877 return ret;
3878 return do_rt_sigqueueinfo(pid, sig, &info);
3879}
3880#endif
3881
3882static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3883{
3884 /* This is only valid for single tasks */
3885 if (pid <= 0 || tgid <= 0)
3886 return -EINVAL;
3887
3888 /* Not even root can pretend to send signals from the kernel.
3889 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3890 */
3891 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3892 (task_pid_vnr(current) != pid))
3893 return -EPERM;
3894
3895 return do_send_specific(tgid, pid, sig, info);
3896}
3897
3898SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3899 siginfo_t __user *, uinfo)
3900{
3901 kernel_siginfo_t info;
3902 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3903 if (unlikely(ret))
3904 return ret;
3905 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3906}
3907
3908#ifdef CONFIG_COMPAT
3909COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3910 compat_pid_t, tgid,
3911 compat_pid_t, pid,
3912 int, sig,
3913 struct compat_siginfo __user *, uinfo)
3914{
3915 kernel_siginfo_t info;
3916 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3917 if (unlikely(ret))
3918 return ret;
3919 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3920}
3921#endif
3922
3923/*
3924 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
3925 */
3926void kernel_sigaction(int sig, __sighandler_t action)
3927{
3928 spin_lock_irq(¤t->sighand->siglock);
3929 current->sighand->action[sig - 1].sa.sa_handler = action;
3930 if (action == SIG_IGN) {
3931 sigset_t mask;
3932
3933 sigemptyset(&mask);
3934 sigaddset(&mask, sig);
3935
3936 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
3937 flush_sigqueue_mask(&mask, ¤t->pending);
3938 recalc_sigpending();
3939 }
3940 spin_unlock_irq(¤t->sighand->siglock);
3941}
3942EXPORT_SYMBOL(kernel_sigaction);
3943
3944void __weak sigaction_compat_abi(struct k_sigaction *act,
3945 struct k_sigaction *oact)
3946{
3947}
3948
3949int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
3950{
3951 struct task_struct *p = current, *t;
3952 struct k_sigaction *k;
3953 sigset_t mask;
3954
3955 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
3956 return -EINVAL;
3957
3958 k = &p->sighand->action[sig-1];
3959
3960 spin_lock_irq(&p->sighand->siglock);
3961 if (oact)
3962 *oact = *k;
3963
3964 sigaction_compat_abi(act, oact);
3965
3966 if (act) {
3967 sigdelsetmask(&act->sa.sa_mask,
3968 sigmask(SIGKILL) | sigmask(SIGSTOP));
3969 *k = *act;
3970 /*
3971 * POSIX 3.3.1.3:
3972 * "Setting a signal action to SIG_IGN for a signal that is
3973 * pending shall cause the pending signal to be discarded,
3974 * whether or not it is blocked."
3975 *
3976 * "Setting a signal action to SIG_DFL for a signal that is
3977 * pending and whose default action is to ignore the signal
3978 * (for example, SIGCHLD), shall cause the pending signal to
3979 * be discarded, whether or not it is blocked"
3980 */
3981 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
3982 sigemptyset(&mask);
3983 sigaddset(&mask, sig);
3984 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
3985 for_each_thread(p, t)
3986 flush_sigqueue_mask(&mask, &t->pending);
3987 }
3988 }
3989
3990 spin_unlock_irq(&p->sighand->siglock);
3991 return 0;
3992}
3993
3994static int
3995do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
3996 size_t min_ss_size)
3997{
3998 struct task_struct *t = current;
3999
4000 if (oss) {
4001 memset(oss, 0, sizeof(stack_t));
4002 oss->ss_sp = (void __user *) t->sas_ss_sp;
4003 oss->ss_size = t->sas_ss_size;
4004 oss->ss_flags = sas_ss_flags(sp) |
4005 (current->sas_ss_flags & SS_FLAG_BITS);
4006 }
4007
4008 if (ss) {
4009 void __user *ss_sp = ss->ss_sp;
4010 size_t ss_size = ss->ss_size;
4011 unsigned ss_flags = ss->ss_flags;
4012 int ss_mode;
4013
4014 if (unlikely(on_sig_stack(sp)))
4015 return -EPERM;
4016
4017 ss_mode = ss_flags & ~SS_FLAG_BITS;
4018 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4019 ss_mode != 0))
4020 return -EINVAL;
4021
4022 if (ss_mode == SS_DISABLE) {
4023 ss_size = 0;
4024 ss_sp = NULL;
4025 } else {
4026 if (unlikely(ss_size < min_ss_size))
4027 return -ENOMEM;
4028 }
4029
4030 t->sas_ss_sp = (unsigned long) ss_sp;
4031 t->sas_ss_size = ss_size;
4032 t->sas_ss_flags = ss_flags;
4033 }
4034 return 0;
4035}
4036
4037SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4038{
4039 stack_t new, old;
4040 int err;
4041 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4042 return -EFAULT;
4043 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4044 current_user_stack_pointer(),
4045 MINSIGSTKSZ);
4046 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4047 err = -EFAULT;
4048 return err;
4049}
4050
4051int restore_altstack(const stack_t __user *uss)
4052{
4053 stack_t new;
4054 if (copy_from_user(&new, uss, sizeof(stack_t)))
4055 return -EFAULT;
4056 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4057 MINSIGSTKSZ);
4058 /* squash all but EFAULT for now */
4059 return 0;
4060}
4061
4062int __save_altstack(stack_t __user *uss, unsigned long sp)
4063{
4064 struct task_struct *t = current;
4065 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4066 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4067 __put_user(t->sas_ss_size, &uss->ss_size);
4068 if (err)
4069 return err;
4070 if (t->sas_ss_flags & SS_AUTODISARM)
4071 sas_ss_reset(t);
4072 return 0;
4073}
4074
4075#ifdef CONFIG_COMPAT
4076static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4077 compat_stack_t __user *uoss_ptr)
4078{
4079 stack_t uss, uoss;
4080 int ret;
4081
4082 if (uss_ptr) {
4083 compat_stack_t uss32;
4084 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4085 return -EFAULT;
4086 uss.ss_sp = compat_ptr(uss32.ss_sp);
4087 uss.ss_flags = uss32.ss_flags;
4088 uss.ss_size = uss32.ss_size;
4089 }
4090 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4091 compat_user_stack_pointer(),
4092 COMPAT_MINSIGSTKSZ);
4093 if (ret >= 0 && uoss_ptr) {
4094 compat_stack_t old;
4095 memset(&old, 0, sizeof(old));
4096 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4097 old.ss_flags = uoss.ss_flags;
4098 old.ss_size = uoss.ss_size;
4099 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4100 ret = -EFAULT;
4101 }
4102 return ret;
4103}
4104
4105COMPAT_SYSCALL_DEFINE2(sigaltstack,
4106 const compat_stack_t __user *, uss_ptr,
4107 compat_stack_t __user *, uoss_ptr)
4108{
4109 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4110}
4111
4112int compat_restore_altstack(const compat_stack_t __user *uss)
4113{
4114 int err = do_compat_sigaltstack(uss, NULL);
4115 /* squash all but -EFAULT for now */
4116 return err == -EFAULT ? err : 0;
4117}
4118
4119int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4120{
4121 int err;
4122 struct task_struct *t = current;
4123 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4124 &uss->ss_sp) |
4125 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4126 __put_user(t->sas_ss_size, &uss->ss_size);
4127 if (err)
4128 return err;
4129 if (t->sas_ss_flags & SS_AUTODISARM)
4130 sas_ss_reset(t);
4131 return 0;
4132}
4133#endif
4134
4135#ifdef __ARCH_WANT_SYS_SIGPENDING
4136
4137/**
4138 * sys_sigpending - examine pending signals
4139 * @uset: where mask of pending signal is returned
4140 */
4141SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4142{
4143 sigset_t set;
4144
4145 if (sizeof(old_sigset_t) > sizeof(*uset))
4146 return -EINVAL;
4147
4148 do_sigpending(&set);
4149
4150 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4151 return -EFAULT;
4152
4153 return 0;
4154}
4155
4156#ifdef CONFIG_COMPAT
4157COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4158{
4159 sigset_t set;
4160
4161 do_sigpending(&set);
4162
4163 return put_user(set.sig[0], set32);
4164}
4165#endif
4166
4167#endif
4168
4169#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4170/**
4171 * sys_sigprocmask - examine and change blocked signals
4172 * @how: whether to add, remove, or set signals
4173 * @nset: signals to add or remove (if non-null)
4174 * @oset: previous value of signal mask if non-null
4175 *
4176 * Some platforms have their own version with special arguments;
4177 * others support only sys_rt_sigprocmask.
4178 */
4179
4180SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4181 old_sigset_t __user *, oset)
4182{
4183 old_sigset_t old_set, new_set;
4184 sigset_t new_blocked;
4185
4186 old_set = current->blocked.sig[0];
4187
4188 if (nset) {
4189 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4190 return -EFAULT;
4191
4192 new_blocked = current->blocked;
4193
4194 switch (how) {
4195 case SIG_BLOCK:
4196 sigaddsetmask(&new_blocked, new_set);
4197 break;
4198 case SIG_UNBLOCK:
4199 sigdelsetmask(&new_blocked, new_set);
4200 break;
4201 case SIG_SETMASK:
4202 new_blocked.sig[0] = new_set;
4203 break;
4204 default:
4205 return -EINVAL;
4206 }
4207
4208 set_current_blocked(&new_blocked);
4209 }
4210
4211 if (oset) {
4212 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4213 return -EFAULT;
4214 }
4215
4216 return 0;
4217}
4218#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4219
4220#ifndef CONFIG_ODD_RT_SIGACTION
4221/**
4222 * sys_rt_sigaction - alter an action taken by a process
4223 * @sig: signal to be sent
4224 * @act: new sigaction
4225 * @oact: used to save the previous sigaction
4226 * @sigsetsize: size of sigset_t type
4227 */
4228SYSCALL_DEFINE4(rt_sigaction, int, sig,
4229 const struct sigaction __user *, act,
4230 struct sigaction __user *, oact,
4231 size_t, sigsetsize)
4232{
4233 struct k_sigaction new_sa, old_sa;
4234 int ret;
4235
4236 /* XXX: Don't preclude handling different sized sigset_t's. */
4237 if (sigsetsize != sizeof(sigset_t))
4238 return -EINVAL;
4239
4240 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4241 return -EFAULT;
4242
4243 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4244 if (ret)
4245 return ret;
4246
4247 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4248 return -EFAULT;
4249
4250 return 0;
4251}
4252#ifdef CONFIG_COMPAT
4253COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4254 const struct compat_sigaction __user *, act,
4255 struct compat_sigaction __user *, oact,
4256 compat_size_t, sigsetsize)
4257{
4258 struct k_sigaction new_ka, old_ka;
4259#ifdef __ARCH_HAS_SA_RESTORER
4260 compat_uptr_t restorer;
4261#endif
4262 int ret;
4263
4264 /* XXX: Don't preclude handling different sized sigset_t's. */
4265 if (sigsetsize != sizeof(compat_sigset_t))
4266 return -EINVAL;
4267
4268 if (act) {
4269 compat_uptr_t handler;
4270 ret = get_user(handler, &act->sa_handler);
4271 new_ka.sa.sa_handler = compat_ptr(handler);
4272#ifdef __ARCH_HAS_SA_RESTORER
4273 ret |= get_user(restorer, &act->sa_restorer);
4274 new_ka.sa.sa_restorer = compat_ptr(restorer);
4275#endif
4276 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4277 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4278 if (ret)
4279 return -EFAULT;
4280 }
4281
4282 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4283 if (!ret && oact) {
4284 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4285 &oact->sa_handler);
4286 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4287 sizeof(oact->sa_mask));
4288 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4289#ifdef __ARCH_HAS_SA_RESTORER
4290 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4291 &oact->sa_restorer);
4292#endif
4293 }
4294 return ret;
4295}
4296#endif
4297#endif /* !CONFIG_ODD_RT_SIGACTION */
4298
4299#ifdef CONFIG_OLD_SIGACTION
4300SYSCALL_DEFINE3(sigaction, int, sig,
4301 const struct old_sigaction __user *, act,
4302 struct old_sigaction __user *, oact)
4303{
4304 struct k_sigaction new_ka, old_ka;
4305 int ret;
4306
4307 if (act) {
4308 old_sigset_t mask;
4309 if (!access_ok(act, sizeof(*act)) ||
4310 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4311 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4312 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4313 __get_user(mask, &act->sa_mask))
4314 return -EFAULT;
4315#ifdef __ARCH_HAS_KA_RESTORER
4316 new_ka.ka_restorer = NULL;
4317#endif
4318 siginitset(&new_ka.sa.sa_mask, mask);
4319 }
4320
4321 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4322
4323 if (!ret && oact) {
4324 if (!access_ok(oact, sizeof(*oact)) ||
4325 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4326 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4327 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4328 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4329 return -EFAULT;
4330 }
4331
4332 return ret;
4333}
4334#endif
4335#ifdef CONFIG_COMPAT_OLD_SIGACTION
4336COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4337 const struct compat_old_sigaction __user *, act,
4338 struct compat_old_sigaction __user *, oact)
4339{
4340 struct k_sigaction new_ka, old_ka;
4341 int ret;
4342 compat_old_sigset_t mask;
4343 compat_uptr_t handler, restorer;
4344
4345 if (act) {
4346 if (!access_ok(act, sizeof(*act)) ||
4347 __get_user(handler, &act->sa_handler) ||
4348 __get_user(restorer, &act->sa_restorer) ||
4349 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4350 __get_user(mask, &act->sa_mask))
4351 return -EFAULT;
4352
4353#ifdef __ARCH_HAS_KA_RESTORER
4354 new_ka.ka_restorer = NULL;
4355#endif
4356 new_ka.sa.sa_handler = compat_ptr(handler);
4357 new_ka.sa.sa_restorer = compat_ptr(restorer);
4358 siginitset(&new_ka.sa.sa_mask, mask);
4359 }
4360
4361 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4362
4363 if (!ret && oact) {
4364 if (!access_ok(oact, sizeof(*oact)) ||
4365 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4366 &oact->sa_handler) ||
4367 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4368 &oact->sa_restorer) ||
4369 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4370 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4371 return -EFAULT;
4372 }
4373 return ret;
4374}
4375#endif
4376
4377#ifdef CONFIG_SGETMASK_SYSCALL
4378
4379/*
4380 * For backwards compatibility. Functionality superseded by sigprocmask.
4381 */
4382SYSCALL_DEFINE0(sgetmask)
4383{
4384 /* SMP safe */
4385 return current->blocked.sig[0];
4386}
4387
4388SYSCALL_DEFINE1(ssetmask, int, newmask)
4389{
4390 int old = current->blocked.sig[0];
4391 sigset_t newset;
4392
4393 siginitset(&newset, newmask);
4394 set_current_blocked(&newset);
4395
4396 return old;
4397}
4398#endif /* CONFIG_SGETMASK_SYSCALL */
4399
4400#ifdef __ARCH_WANT_SYS_SIGNAL
4401/*
4402 * For backwards compatibility. Functionality superseded by sigaction.
4403 */
4404SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4405{
4406 struct k_sigaction new_sa, old_sa;
4407 int ret;
4408
4409 new_sa.sa.sa_handler = handler;
4410 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4411 sigemptyset(&new_sa.sa.sa_mask);
4412
4413 ret = do_sigaction(sig, &new_sa, &old_sa);
4414
4415 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4416}
4417#endif /* __ARCH_WANT_SYS_SIGNAL */
4418
4419#ifdef __ARCH_WANT_SYS_PAUSE
4420
4421SYSCALL_DEFINE0(pause)
4422{
4423 while (!signal_pending(current)) {
4424 __set_current_state(TASK_INTERRUPTIBLE);
4425 schedule();
4426 }
4427 return -ERESTARTNOHAND;
4428}
4429
4430#endif
4431
4432static int sigsuspend(sigset_t *set)
4433{
4434 current->saved_sigmask = current->blocked;
4435 set_current_blocked(set);
4436
4437 while (!signal_pending(current)) {
4438 __set_current_state(TASK_INTERRUPTIBLE);
4439 schedule();
4440 }
4441 set_restore_sigmask();
4442 return -ERESTARTNOHAND;
4443}
4444
4445/**
4446 * sys_rt_sigsuspend - replace the signal mask for a value with the
4447 * @unewset value until a signal is received
4448 * @unewset: new signal mask value
4449 * @sigsetsize: size of sigset_t type
4450 */
4451SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4452{
4453 sigset_t newset;
4454
4455 /* XXX: Don't preclude handling different sized sigset_t's. */
4456 if (sigsetsize != sizeof(sigset_t))
4457 return -EINVAL;
4458
4459 if (copy_from_user(&newset, unewset, sizeof(newset)))
4460 return -EFAULT;
4461 return sigsuspend(&newset);
4462}
4463
4464#ifdef CONFIG_COMPAT
4465COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4466{
4467 sigset_t newset;
4468
4469 /* XXX: Don't preclude handling different sized sigset_t's. */
4470 if (sigsetsize != sizeof(sigset_t))
4471 return -EINVAL;
4472
4473 if (get_compat_sigset(&newset, unewset))
4474 return -EFAULT;
4475 return sigsuspend(&newset);
4476}
4477#endif
4478
4479#ifdef CONFIG_OLD_SIGSUSPEND
4480SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4481{
4482 sigset_t blocked;
4483 siginitset(&blocked, mask);
4484 return sigsuspend(&blocked);
4485}
4486#endif
4487#ifdef CONFIG_OLD_SIGSUSPEND3
4488SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4489{
4490 sigset_t blocked;
4491 siginitset(&blocked, mask);
4492 return sigsuspend(&blocked);
4493}
4494#endif
4495
4496__weak const char *arch_vma_name(struct vm_area_struct *vma)
4497{
4498 return NULL;
4499}
4500
4501static inline void siginfo_buildtime_checks(void)
4502{
4503 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4504
4505 /* Verify the offsets in the two siginfos match */
4506#define CHECK_OFFSET(field) \
4507 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4508
4509 /* kill */
4510 CHECK_OFFSET(si_pid);
4511 CHECK_OFFSET(si_uid);
4512
4513 /* timer */
4514 CHECK_OFFSET(si_tid);
4515 CHECK_OFFSET(si_overrun);
4516 CHECK_OFFSET(si_value);
4517
4518 /* rt */
4519 CHECK_OFFSET(si_pid);
4520 CHECK_OFFSET(si_uid);
4521 CHECK_OFFSET(si_value);
4522
4523 /* sigchld */
4524 CHECK_OFFSET(si_pid);
4525 CHECK_OFFSET(si_uid);
4526 CHECK_OFFSET(si_status);
4527 CHECK_OFFSET(si_utime);
4528 CHECK_OFFSET(si_stime);
4529
4530 /* sigfault */
4531 CHECK_OFFSET(si_addr);
4532 CHECK_OFFSET(si_addr_lsb);
4533 CHECK_OFFSET(si_lower);
4534 CHECK_OFFSET(si_upper);
4535 CHECK_OFFSET(si_pkey);
4536
4537 /* sigpoll */
4538 CHECK_OFFSET(si_band);
4539 CHECK_OFFSET(si_fd);
4540
4541 /* sigsys */
4542 CHECK_OFFSET(si_call_addr);
4543 CHECK_OFFSET(si_syscall);
4544 CHECK_OFFSET(si_arch);
4545#undef CHECK_OFFSET
4546
4547 /* usb asyncio */
4548 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4549 offsetof(struct siginfo, si_addr));
4550 if (sizeof(int) == sizeof(void __user *)) {
4551 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4552 sizeof(void __user *));
4553 } else {
4554 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4555 sizeof_field(struct siginfo, si_uid)) !=
4556 sizeof(void __user *));
4557 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4558 offsetof(struct siginfo, si_uid));
4559 }
4560#ifdef CONFIG_COMPAT
4561 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4562 offsetof(struct compat_siginfo, si_addr));
4563 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4564 sizeof(compat_uptr_t));
4565 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4566 sizeof_field(struct siginfo, si_pid));
4567#endif
4568}
4569
4570void __init signals_init(void)
4571{
4572 siginfo_buildtime_checks();
4573
4574 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4575}
4576
4577#ifdef CONFIG_KGDB_KDB
4578#include <linux/kdb.h>
4579/*
4580 * kdb_send_sig - Allows kdb to send signals without exposing
4581 * signal internals. This function checks if the required locks are
4582 * available before calling the main signal code, to avoid kdb
4583 * deadlocks.
4584 */
4585void kdb_send_sig(struct task_struct *t, int sig)
4586{
4587 static struct task_struct *kdb_prev_t;
4588 int new_t, ret;
4589 if (!spin_trylock(&t->sighand->siglock)) {
4590 kdb_printf("Can't do kill command now.\n"
4591 "The sigmask lock is held somewhere else in "
4592 "kernel, try again later\n");
4593 return;
4594 }
4595 new_t = kdb_prev_t != t;
4596 kdb_prev_t = t;
4597 if (t->state != TASK_RUNNING && new_t) {
4598 spin_unlock(&t->sighand->siglock);
4599 kdb_printf("Process is not RUNNING, sending a signal from "
4600 "kdb risks deadlock\n"
4601 "on the run queue locks. "
4602 "The signal has _not_ been sent.\n"
4603 "Reissue the kill command if you want to risk "
4604 "the deadlock.\n");
4605 return;
4606 }
4607 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4608 spin_unlock(&t->sighand->siglock);
4609 if (ret)
4610 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4611 sig, t->pid);
4612 else
4613 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4614}
4615#endif /* CONFIG_KGDB_KDB */
1/*
2 * linux/kernel/signal.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
7 *
8 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
9 * Changes to use preallocated sigqueue structures
10 * to allow signals to be sent reliably.
11 */
12
13#include <linux/slab.h>
14#include <linux/export.h>
15#include <linux/init.h>
16#include <linux/sched.h>
17#include <linux/fs.h>
18#include <linux/tty.h>
19#include <linux/binfmts.h>
20#include <linux/coredump.h>
21#include <linux/security.h>
22#include <linux/syscalls.h>
23#include <linux/ptrace.h>
24#include <linux/signal.h>
25#include <linux/signalfd.h>
26#include <linux/ratelimit.h>
27#include <linux/tracehook.h>
28#include <linux/capability.h>
29#include <linux/freezer.h>
30#include <linux/pid_namespace.h>
31#include <linux/nsproxy.h>
32#include <linux/user_namespace.h>
33#include <linux/uprobes.h>
34#include <linux/compat.h>
35#include <linux/cn_proc.h>
36#include <linux/compiler.h>
37
38#define CREATE_TRACE_POINTS
39#include <trace/events/signal.h>
40
41#include <asm/param.h>
42#include <asm/uaccess.h>
43#include <asm/unistd.h>
44#include <asm/siginfo.h>
45#include <asm/cacheflush.h>
46#include "audit.h" /* audit_signal_info() */
47
48/*
49 * SLAB caches for signal bits.
50 */
51
52static struct kmem_cache *sigqueue_cachep;
53
54int print_fatal_signals __read_mostly;
55
56static void __user *sig_handler(struct task_struct *t, int sig)
57{
58 return t->sighand->action[sig - 1].sa.sa_handler;
59}
60
61static int sig_handler_ignored(void __user *handler, int sig)
62{
63 /* Is it explicitly or implicitly ignored? */
64 return handler == SIG_IGN ||
65 (handler == SIG_DFL && sig_kernel_ignore(sig));
66}
67
68static int sig_task_ignored(struct task_struct *t, int sig, bool force)
69{
70 void __user *handler;
71
72 handler = sig_handler(t, sig);
73
74 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
75 handler == SIG_DFL && !force)
76 return 1;
77
78 return sig_handler_ignored(handler, sig);
79}
80
81static int sig_ignored(struct task_struct *t, int sig, bool force)
82{
83 /*
84 * Blocked signals are never ignored, since the
85 * signal handler may change by the time it is
86 * unblocked.
87 */
88 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
89 return 0;
90
91 if (!sig_task_ignored(t, sig, force))
92 return 0;
93
94 /*
95 * Tracers may want to know about even ignored signals.
96 */
97 return !t->ptrace;
98}
99
100/*
101 * Re-calculate pending state from the set of locally pending
102 * signals, globally pending signals, and blocked signals.
103 */
104static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
105{
106 unsigned long ready;
107 long i;
108
109 switch (_NSIG_WORDS) {
110 default:
111 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
112 ready |= signal->sig[i] &~ blocked->sig[i];
113 break;
114
115 case 4: ready = signal->sig[3] &~ blocked->sig[3];
116 ready |= signal->sig[2] &~ blocked->sig[2];
117 ready |= signal->sig[1] &~ blocked->sig[1];
118 ready |= signal->sig[0] &~ blocked->sig[0];
119 break;
120
121 case 2: ready = signal->sig[1] &~ blocked->sig[1];
122 ready |= signal->sig[0] &~ blocked->sig[0];
123 break;
124
125 case 1: ready = signal->sig[0] &~ blocked->sig[0];
126 }
127 return ready != 0;
128}
129
130#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
131
132static int recalc_sigpending_tsk(struct task_struct *t)
133{
134 if ((t->jobctl & JOBCTL_PENDING_MASK) ||
135 PENDING(&t->pending, &t->blocked) ||
136 PENDING(&t->signal->shared_pending, &t->blocked)) {
137 set_tsk_thread_flag(t, TIF_SIGPENDING);
138 return 1;
139 }
140 /*
141 * We must never clear the flag in another thread, or in current
142 * when it's possible the current syscall is returning -ERESTART*.
143 * So we don't clear it here, and only callers who know they should do.
144 */
145 return 0;
146}
147
148/*
149 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
150 * This is superfluous when called on current, the wakeup is a harmless no-op.
151 */
152void recalc_sigpending_and_wake(struct task_struct *t)
153{
154 if (recalc_sigpending_tsk(t))
155 signal_wake_up(t, 0);
156}
157
158void recalc_sigpending(void)
159{
160 if (!recalc_sigpending_tsk(current) && !freezing(current))
161 clear_thread_flag(TIF_SIGPENDING);
162
163}
164
165/* Given the mask, find the first available signal that should be serviced. */
166
167#define SYNCHRONOUS_MASK \
168 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
169 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
170
171int next_signal(struct sigpending *pending, sigset_t *mask)
172{
173 unsigned long i, *s, *m, x;
174 int sig = 0;
175
176 s = pending->signal.sig;
177 m = mask->sig;
178
179 /*
180 * Handle the first word specially: it contains the
181 * synchronous signals that need to be dequeued first.
182 */
183 x = *s &~ *m;
184 if (x) {
185 if (x & SYNCHRONOUS_MASK)
186 x &= SYNCHRONOUS_MASK;
187 sig = ffz(~x) + 1;
188 return sig;
189 }
190
191 switch (_NSIG_WORDS) {
192 default:
193 for (i = 1; i < _NSIG_WORDS; ++i) {
194 x = *++s &~ *++m;
195 if (!x)
196 continue;
197 sig = ffz(~x) + i*_NSIG_BPW + 1;
198 break;
199 }
200 break;
201
202 case 2:
203 x = s[1] &~ m[1];
204 if (!x)
205 break;
206 sig = ffz(~x) + _NSIG_BPW + 1;
207 break;
208
209 case 1:
210 /* Nothing to do */
211 break;
212 }
213
214 return sig;
215}
216
217static inline void print_dropped_signal(int sig)
218{
219 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
220
221 if (!print_fatal_signals)
222 return;
223
224 if (!__ratelimit(&ratelimit_state))
225 return;
226
227 printk(KERN_INFO "%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
228 current->comm, current->pid, sig);
229}
230
231/**
232 * task_set_jobctl_pending - set jobctl pending bits
233 * @task: target task
234 * @mask: pending bits to set
235 *
236 * Clear @mask from @task->jobctl. @mask must be subset of
237 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
238 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
239 * cleared. If @task is already being killed or exiting, this function
240 * becomes noop.
241 *
242 * CONTEXT:
243 * Must be called with @task->sighand->siglock held.
244 *
245 * RETURNS:
246 * %true if @mask is set, %false if made noop because @task was dying.
247 */
248bool task_set_jobctl_pending(struct task_struct *task, unsigned int mask)
249{
250 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
251 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
252 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
253
254 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
255 return false;
256
257 if (mask & JOBCTL_STOP_SIGMASK)
258 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
259
260 task->jobctl |= mask;
261 return true;
262}
263
264/**
265 * task_clear_jobctl_trapping - clear jobctl trapping bit
266 * @task: target task
267 *
268 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
269 * Clear it and wake up the ptracer. Note that we don't need any further
270 * locking. @task->siglock guarantees that @task->parent points to the
271 * ptracer.
272 *
273 * CONTEXT:
274 * Must be called with @task->sighand->siglock held.
275 */
276void task_clear_jobctl_trapping(struct task_struct *task)
277{
278 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
279 task->jobctl &= ~JOBCTL_TRAPPING;
280 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
281 }
282}
283
284/**
285 * task_clear_jobctl_pending - clear jobctl pending bits
286 * @task: target task
287 * @mask: pending bits to clear
288 *
289 * Clear @mask from @task->jobctl. @mask must be subset of
290 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
291 * STOP bits are cleared together.
292 *
293 * If clearing of @mask leaves no stop or trap pending, this function calls
294 * task_clear_jobctl_trapping().
295 *
296 * CONTEXT:
297 * Must be called with @task->sighand->siglock held.
298 */
299void task_clear_jobctl_pending(struct task_struct *task, unsigned int mask)
300{
301 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
302
303 if (mask & JOBCTL_STOP_PENDING)
304 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
305
306 task->jobctl &= ~mask;
307
308 if (!(task->jobctl & JOBCTL_PENDING_MASK))
309 task_clear_jobctl_trapping(task);
310}
311
312/**
313 * task_participate_group_stop - participate in a group stop
314 * @task: task participating in a group stop
315 *
316 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
317 * Group stop states are cleared and the group stop count is consumed if
318 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
319 * stop, the appropriate %SIGNAL_* flags are set.
320 *
321 * CONTEXT:
322 * Must be called with @task->sighand->siglock held.
323 *
324 * RETURNS:
325 * %true if group stop completion should be notified to the parent, %false
326 * otherwise.
327 */
328static bool task_participate_group_stop(struct task_struct *task)
329{
330 struct signal_struct *sig = task->signal;
331 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
332
333 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
334
335 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
336
337 if (!consume)
338 return false;
339
340 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
341 sig->group_stop_count--;
342
343 /*
344 * Tell the caller to notify completion iff we are entering into a
345 * fresh group stop. Read comment in do_signal_stop() for details.
346 */
347 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
348 sig->flags = SIGNAL_STOP_STOPPED;
349 return true;
350 }
351 return false;
352}
353
354/*
355 * allocate a new signal queue record
356 * - this may be called without locks if and only if t == current, otherwise an
357 * appropriate lock must be held to stop the target task from exiting
358 */
359static struct sigqueue *
360__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
361{
362 struct sigqueue *q = NULL;
363 struct user_struct *user;
364
365 /*
366 * Protect access to @t credentials. This can go away when all
367 * callers hold rcu read lock.
368 */
369 rcu_read_lock();
370 user = get_uid(__task_cred(t)->user);
371 atomic_inc(&user->sigpending);
372 rcu_read_unlock();
373
374 if (override_rlimit ||
375 atomic_read(&user->sigpending) <=
376 task_rlimit(t, RLIMIT_SIGPENDING)) {
377 q = kmem_cache_alloc(sigqueue_cachep, flags);
378 } else {
379 print_dropped_signal(sig);
380 }
381
382 if (unlikely(q == NULL)) {
383 atomic_dec(&user->sigpending);
384 free_uid(user);
385 } else {
386 INIT_LIST_HEAD(&q->list);
387 q->flags = 0;
388 q->user = user;
389 }
390
391 return q;
392}
393
394static void __sigqueue_free(struct sigqueue *q)
395{
396 if (q->flags & SIGQUEUE_PREALLOC)
397 return;
398 atomic_dec(&q->user->sigpending);
399 free_uid(q->user);
400 kmem_cache_free(sigqueue_cachep, q);
401}
402
403void flush_sigqueue(struct sigpending *queue)
404{
405 struct sigqueue *q;
406
407 sigemptyset(&queue->signal);
408 while (!list_empty(&queue->list)) {
409 q = list_entry(queue->list.next, struct sigqueue , list);
410 list_del_init(&q->list);
411 __sigqueue_free(q);
412 }
413}
414
415/*
416 * Flush all pending signals for a task.
417 */
418void __flush_signals(struct task_struct *t)
419{
420 clear_tsk_thread_flag(t, TIF_SIGPENDING);
421 flush_sigqueue(&t->pending);
422 flush_sigqueue(&t->signal->shared_pending);
423}
424
425void flush_signals(struct task_struct *t)
426{
427 unsigned long flags;
428
429 spin_lock_irqsave(&t->sighand->siglock, flags);
430 __flush_signals(t);
431 spin_unlock_irqrestore(&t->sighand->siglock, flags);
432}
433
434static void __flush_itimer_signals(struct sigpending *pending)
435{
436 sigset_t signal, retain;
437 struct sigqueue *q, *n;
438
439 signal = pending->signal;
440 sigemptyset(&retain);
441
442 list_for_each_entry_safe(q, n, &pending->list, list) {
443 int sig = q->info.si_signo;
444
445 if (likely(q->info.si_code != SI_TIMER)) {
446 sigaddset(&retain, sig);
447 } else {
448 sigdelset(&signal, sig);
449 list_del_init(&q->list);
450 __sigqueue_free(q);
451 }
452 }
453
454 sigorsets(&pending->signal, &signal, &retain);
455}
456
457void flush_itimer_signals(void)
458{
459 struct task_struct *tsk = current;
460 unsigned long flags;
461
462 spin_lock_irqsave(&tsk->sighand->siglock, flags);
463 __flush_itimer_signals(&tsk->pending);
464 __flush_itimer_signals(&tsk->signal->shared_pending);
465 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
466}
467
468void ignore_signals(struct task_struct *t)
469{
470 int i;
471
472 for (i = 0; i < _NSIG; ++i)
473 t->sighand->action[i].sa.sa_handler = SIG_IGN;
474
475 flush_signals(t);
476}
477
478/*
479 * Flush all handlers for a task.
480 */
481
482void
483flush_signal_handlers(struct task_struct *t, int force_default)
484{
485 int i;
486 struct k_sigaction *ka = &t->sighand->action[0];
487 for (i = _NSIG ; i != 0 ; i--) {
488 if (force_default || ka->sa.sa_handler != SIG_IGN)
489 ka->sa.sa_handler = SIG_DFL;
490 ka->sa.sa_flags = 0;
491#ifdef __ARCH_HAS_SA_RESTORER
492 ka->sa.sa_restorer = NULL;
493#endif
494 sigemptyset(&ka->sa.sa_mask);
495 ka++;
496 }
497}
498
499int unhandled_signal(struct task_struct *tsk, int sig)
500{
501 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
502 if (is_global_init(tsk))
503 return 1;
504 if (handler != SIG_IGN && handler != SIG_DFL)
505 return 0;
506 /* if ptraced, let the tracer determine */
507 return !tsk->ptrace;
508}
509
510/*
511 * Notify the system that a driver wants to block all signals for this
512 * process, and wants to be notified if any signals at all were to be
513 * sent/acted upon. If the notifier routine returns non-zero, then the
514 * signal will be acted upon after all. If the notifier routine returns 0,
515 * then then signal will be blocked. Only one block per process is
516 * allowed. priv is a pointer to private data that the notifier routine
517 * can use to determine if the signal should be blocked or not.
518 */
519void
520block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
521{
522 unsigned long flags;
523
524 spin_lock_irqsave(¤t->sighand->siglock, flags);
525 current->notifier_mask = mask;
526 current->notifier_data = priv;
527 current->notifier = notifier;
528 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
529}
530
531/* Notify the system that blocking has ended. */
532
533void
534unblock_all_signals(void)
535{
536 unsigned long flags;
537
538 spin_lock_irqsave(¤t->sighand->siglock, flags);
539 current->notifier = NULL;
540 current->notifier_data = NULL;
541 recalc_sigpending();
542 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
543}
544
545static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
546{
547 struct sigqueue *q, *first = NULL;
548
549 /*
550 * Collect the siginfo appropriate to this signal. Check if
551 * there is another siginfo for the same signal.
552 */
553 list_for_each_entry(q, &list->list, list) {
554 if (q->info.si_signo == sig) {
555 if (first)
556 goto still_pending;
557 first = q;
558 }
559 }
560
561 sigdelset(&list->signal, sig);
562
563 if (first) {
564still_pending:
565 list_del_init(&first->list);
566 copy_siginfo(info, &first->info);
567 __sigqueue_free(first);
568 } else {
569 /*
570 * Ok, it wasn't in the queue. This must be
571 * a fast-pathed signal or we must have been
572 * out of queue space. So zero out the info.
573 */
574 info->si_signo = sig;
575 info->si_errno = 0;
576 info->si_code = SI_USER;
577 info->si_pid = 0;
578 info->si_uid = 0;
579 }
580}
581
582static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
583 siginfo_t *info)
584{
585 int sig = next_signal(pending, mask);
586
587 if (sig) {
588 if (current->notifier) {
589 if (sigismember(current->notifier_mask, sig)) {
590 if (!(current->notifier)(current->notifier_data)) {
591 clear_thread_flag(TIF_SIGPENDING);
592 return 0;
593 }
594 }
595 }
596
597 collect_signal(sig, pending, info);
598 }
599
600 return sig;
601}
602
603/*
604 * Dequeue a signal and return the element to the caller, which is
605 * expected to free it.
606 *
607 * All callers have to hold the siglock.
608 */
609int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
610{
611 int signr;
612
613 /* We only dequeue private signals from ourselves, we don't let
614 * signalfd steal them
615 */
616 signr = __dequeue_signal(&tsk->pending, mask, info);
617 if (!signr) {
618 signr = __dequeue_signal(&tsk->signal->shared_pending,
619 mask, info);
620 /*
621 * itimer signal ?
622 *
623 * itimers are process shared and we restart periodic
624 * itimers in the signal delivery path to prevent DoS
625 * attacks in the high resolution timer case. This is
626 * compliant with the old way of self-restarting
627 * itimers, as the SIGALRM is a legacy signal and only
628 * queued once. Changing the restart behaviour to
629 * restart the timer in the signal dequeue path is
630 * reducing the timer noise on heavy loaded !highres
631 * systems too.
632 */
633 if (unlikely(signr == SIGALRM)) {
634 struct hrtimer *tmr = &tsk->signal->real_timer;
635
636 if (!hrtimer_is_queued(tmr) &&
637 tsk->signal->it_real_incr.tv64 != 0) {
638 hrtimer_forward(tmr, tmr->base->get_time(),
639 tsk->signal->it_real_incr);
640 hrtimer_restart(tmr);
641 }
642 }
643 }
644
645 recalc_sigpending();
646 if (!signr)
647 return 0;
648
649 if (unlikely(sig_kernel_stop(signr))) {
650 /*
651 * Set a marker that we have dequeued a stop signal. Our
652 * caller might release the siglock and then the pending
653 * stop signal it is about to process is no longer in the
654 * pending bitmasks, but must still be cleared by a SIGCONT
655 * (and overruled by a SIGKILL). So those cases clear this
656 * shared flag after we've set it. Note that this flag may
657 * remain set after the signal we return is ignored or
658 * handled. That doesn't matter because its only purpose
659 * is to alert stop-signal processing code when another
660 * processor has come along and cleared the flag.
661 */
662 current->jobctl |= JOBCTL_STOP_DEQUEUED;
663 }
664 if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
665 /*
666 * Release the siglock to ensure proper locking order
667 * of timer locks outside of siglocks. Note, we leave
668 * irqs disabled here, since the posix-timers code is
669 * about to disable them again anyway.
670 */
671 spin_unlock(&tsk->sighand->siglock);
672 do_schedule_next_timer(info);
673 spin_lock(&tsk->sighand->siglock);
674 }
675 return signr;
676}
677
678/*
679 * Tell a process that it has a new active signal..
680 *
681 * NOTE! we rely on the previous spin_lock to
682 * lock interrupts for us! We can only be called with
683 * "siglock" held, and the local interrupt must
684 * have been disabled when that got acquired!
685 *
686 * No need to set need_resched since signal event passing
687 * goes through ->blocked
688 */
689void signal_wake_up_state(struct task_struct *t, unsigned int state)
690{
691 set_tsk_thread_flag(t, TIF_SIGPENDING);
692 /*
693 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
694 * case. We don't check t->state here because there is a race with it
695 * executing another processor and just now entering stopped state.
696 * By using wake_up_state, we ensure the process will wake up and
697 * handle its death signal.
698 */
699 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
700 kick_process(t);
701}
702
703/*
704 * Remove signals in mask from the pending set and queue.
705 * Returns 1 if any signals were found.
706 *
707 * All callers must be holding the siglock.
708 *
709 * This version takes a sigset mask and looks at all signals,
710 * not just those in the first mask word.
711 */
712static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
713{
714 struct sigqueue *q, *n;
715 sigset_t m;
716
717 sigandsets(&m, mask, &s->signal);
718 if (sigisemptyset(&m))
719 return 0;
720
721 sigandnsets(&s->signal, &s->signal, mask);
722 list_for_each_entry_safe(q, n, &s->list, list) {
723 if (sigismember(mask, q->info.si_signo)) {
724 list_del_init(&q->list);
725 __sigqueue_free(q);
726 }
727 }
728 return 1;
729}
730/*
731 * Remove signals in mask from the pending set and queue.
732 * Returns 1 if any signals were found.
733 *
734 * All callers must be holding the siglock.
735 */
736static int rm_from_queue(unsigned long mask, struct sigpending *s)
737{
738 struct sigqueue *q, *n;
739
740 if (!sigtestsetmask(&s->signal, mask))
741 return 0;
742
743 sigdelsetmask(&s->signal, mask);
744 list_for_each_entry_safe(q, n, &s->list, list) {
745 if (q->info.si_signo < SIGRTMIN &&
746 (mask & sigmask(q->info.si_signo))) {
747 list_del_init(&q->list);
748 __sigqueue_free(q);
749 }
750 }
751 return 1;
752}
753
754static inline int is_si_special(const struct siginfo *info)
755{
756 return info <= SEND_SIG_FORCED;
757}
758
759static inline bool si_fromuser(const struct siginfo *info)
760{
761 return info == SEND_SIG_NOINFO ||
762 (!is_si_special(info) && SI_FROMUSER(info));
763}
764
765/*
766 * called with RCU read lock from check_kill_permission()
767 */
768static int kill_ok_by_cred(struct task_struct *t)
769{
770 const struct cred *cred = current_cred();
771 const struct cred *tcred = __task_cred(t);
772
773 if (uid_eq(cred->euid, tcred->suid) ||
774 uid_eq(cred->euid, tcred->uid) ||
775 uid_eq(cred->uid, tcred->suid) ||
776 uid_eq(cred->uid, tcred->uid))
777 return 1;
778
779 if (ns_capable(tcred->user_ns, CAP_KILL))
780 return 1;
781
782 return 0;
783}
784
785/*
786 * Bad permissions for sending the signal
787 * - the caller must hold the RCU read lock
788 */
789static int check_kill_permission(int sig, struct siginfo *info,
790 struct task_struct *t)
791{
792 struct pid *sid;
793 int error;
794
795 if (!valid_signal(sig))
796 return -EINVAL;
797
798 if (!si_fromuser(info))
799 return 0;
800
801 error = audit_signal_info(sig, t); /* Let audit system see the signal */
802 if (error)
803 return error;
804
805 if (!same_thread_group(current, t) &&
806 !kill_ok_by_cred(t)) {
807 switch (sig) {
808 case SIGCONT:
809 sid = task_session(t);
810 /*
811 * We don't return the error if sid == NULL. The
812 * task was unhashed, the caller must notice this.
813 */
814 if (!sid || sid == task_session(current))
815 break;
816 default:
817 return -EPERM;
818 }
819 }
820
821 return security_task_kill(t, info, sig, 0);
822}
823
824/**
825 * ptrace_trap_notify - schedule trap to notify ptracer
826 * @t: tracee wanting to notify tracer
827 *
828 * This function schedules sticky ptrace trap which is cleared on the next
829 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
830 * ptracer.
831 *
832 * If @t is running, STOP trap will be taken. If trapped for STOP and
833 * ptracer is listening for events, tracee is woken up so that it can
834 * re-trap for the new event. If trapped otherwise, STOP trap will be
835 * eventually taken without returning to userland after the existing traps
836 * are finished by PTRACE_CONT.
837 *
838 * CONTEXT:
839 * Must be called with @task->sighand->siglock held.
840 */
841static void ptrace_trap_notify(struct task_struct *t)
842{
843 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
844 assert_spin_locked(&t->sighand->siglock);
845
846 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
847 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
848}
849
850/*
851 * Handle magic process-wide effects of stop/continue signals. Unlike
852 * the signal actions, these happen immediately at signal-generation
853 * time regardless of blocking, ignoring, or handling. This does the
854 * actual continuing for SIGCONT, but not the actual stopping for stop
855 * signals. The process stop is done as a signal action for SIG_DFL.
856 *
857 * Returns true if the signal should be actually delivered, otherwise
858 * it should be dropped.
859 */
860static bool prepare_signal(int sig, struct task_struct *p, bool force)
861{
862 struct signal_struct *signal = p->signal;
863 struct task_struct *t;
864
865 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
866 if (signal->flags & SIGNAL_GROUP_COREDUMP)
867 return sig == SIGKILL;
868 /*
869 * The process is in the middle of dying, nothing to do.
870 */
871 } else if (sig_kernel_stop(sig)) {
872 /*
873 * This is a stop signal. Remove SIGCONT from all queues.
874 */
875 rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
876 t = p;
877 do {
878 rm_from_queue(sigmask(SIGCONT), &t->pending);
879 } while_each_thread(p, t);
880 } else if (sig == SIGCONT) {
881 unsigned int why;
882 /*
883 * Remove all stop signals from all queues, wake all threads.
884 */
885 rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
886 t = p;
887 do {
888 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
889 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
890 if (likely(!(t->ptrace & PT_SEIZED)))
891 wake_up_state(t, __TASK_STOPPED);
892 else
893 ptrace_trap_notify(t);
894 } while_each_thread(p, t);
895
896 /*
897 * Notify the parent with CLD_CONTINUED if we were stopped.
898 *
899 * If we were in the middle of a group stop, we pretend it
900 * was already finished, and then continued. Since SIGCHLD
901 * doesn't queue we report only CLD_STOPPED, as if the next
902 * CLD_CONTINUED was dropped.
903 */
904 why = 0;
905 if (signal->flags & SIGNAL_STOP_STOPPED)
906 why |= SIGNAL_CLD_CONTINUED;
907 else if (signal->group_stop_count)
908 why |= SIGNAL_CLD_STOPPED;
909
910 if (why) {
911 /*
912 * The first thread which returns from do_signal_stop()
913 * will take ->siglock, notice SIGNAL_CLD_MASK, and
914 * notify its parent. See get_signal_to_deliver().
915 */
916 signal->flags = why | SIGNAL_STOP_CONTINUED;
917 signal->group_stop_count = 0;
918 signal->group_exit_code = 0;
919 }
920 }
921
922 return !sig_ignored(p, sig, force);
923}
924
925/*
926 * Test if P wants to take SIG. After we've checked all threads with this,
927 * it's equivalent to finding no threads not blocking SIG. Any threads not
928 * blocking SIG were ruled out because they are not running and already
929 * have pending signals. Such threads will dequeue from the shared queue
930 * as soon as they're available, so putting the signal on the shared queue
931 * will be equivalent to sending it to one such thread.
932 */
933static inline int wants_signal(int sig, struct task_struct *p)
934{
935 if (sigismember(&p->blocked, sig))
936 return 0;
937 if (p->flags & PF_EXITING)
938 return 0;
939 if (sig == SIGKILL)
940 return 1;
941 if (task_is_stopped_or_traced(p))
942 return 0;
943 return task_curr(p) || !signal_pending(p);
944}
945
946static void complete_signal(int sig, struct task_struct *p, int group)
947{
948 struct signal_struct *signal = p->signal;
949 struct task_struct *t;
950
951 /*
952 * Now find a thread we can wake up to take the signal off the queue.
953 *
954 * If the main thread wants the signal, it gets first crack.
955 * Probably the least surprising to the average bear.
956 */
957 if (wants_signal(sig, p))
958 t = p;
959 else if (!group || thread_group_empty(p))
960 /*
961 * There is just one thread and it does not need to be woken.
962 * It will dequeue unblocked signals before it runs again.
963 */
964 return;
965 else {
966 /*
967 * Otherwise try to find a suitable thread.
968 */
969 t = signal->curr_target;
970 while (!wants_signal(sig, t)) {
971 t = next_thread(t);
972 if (t == signal->curr_target)
973 /*
974 * No thread needs to be woken.
975 * Any eligible threads will see
976 * the signal in the queue soon.
977 */
978 return;
979 }
980 signal->curr_target = t;
981 }
982
983 /*
984 * Found a killable thread. If the signal will be fatal,
985 * then start taking the whole group down immediately.
986 */
987 if (sig_fatal(p, sig) &&
988 !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
989 !sigismember(&t->real_blocked, sig) &&
990 (sig == SIGKILL || !t->ptrace)) {
991 /*
992 * This signal will be fatal to the whole group.
993 */
994 if (!sig_kernel_coredump(sig)) {
995 /*
996 * Start a group exit and wake everybody up.
997 * This way we don't have other threads
998 * running and doing things after a slower
999 * thread has the fatal signal pending.
1000 */
1001 signal->flags = SIGNAL_GROUP_EXIT;
1002 signal->group_exit_code = sig;
1003 signal->group_stop_count = 0;
1004 t = p;
1005 do {
1006 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1007 sigaddset(&t->pending.signal, SIGKILL);
1008 signal_wake_up(t, 1);
1009 } while_each_thread(p, t);
1010 return;
1011 }
1012 }
1013
1014 /*
1015 * The signal is already in the shared-pending queue.
1016 * Tell the chosen thread to wake up and dequeue it.
1017 */
1018 signal_wake_up(t, sig == SIGKILL);
1019 return;
1020}
1021
1022static inline int legacy_queue(struct sigpending *signals, int sig)
1023{
1024 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1025}
1026
1027#ifdef CONFIG_USER_NS
1028static inline void userns_fixup_signal_uid(struct siginfo *info, struct task_struct *t)
1029{
1030 if (current_user_ns() == task_cred_xxx(t, user_ns))
1031 return;
1032
1033 if (SI_FROMKERNEL(info))
1034 return;
1035
1036 rcu_read_lock();
1037 info->si_uid = from_kuid_munged(task_cred_xxx(t, user_ns),
1038 make_kuid(current_user_ns(), info->si_uid));
1039 rcu_read_unlock();
1040}
1041#else
1042static inline void userns_fixup_signal_uid(struct siginfo *info, struct task_struct *t)
1043{
1044 return;
1045}
1046#endif
1047
1048static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
1049 int group, int from_ancestor_ns)
1050{
1051 struct sigpending *pending;
1052 struct sigqueue *q;
1053 int override_rlimit;
1054 int ret = 0, result;
1055
1056 assert_spin_locked(&t->sighand->siglock);
1057
1058 result = TRACE_SIGNAL_IGNORED;
1059 if (!prepare_signal(sig, t,
1060 from_ancestor_ns || (info == SEND_SIG_FORCED)))
1061 goto ret;
1062
1063 pending = group ? &t->signal->shared_pending : &t->pending;
1064 /*
1065 * Short-circuit ignored signals and support queuing
1066 * exactly one non-rt signal, so that we can get more
1067 * detailed information about the cause of the signal.
1068 */
1069 result = TRACE_SIGNAL_ALREADY_PENDING;
1070 if (legacy_queue(pending, sig))
1071 goto ret;
1072
1073 result = TRACE_SIGNAL_DELIVERED;
1074 /*
1075 * fast-pathed signals for kernel-internal things like SIGSTOP
1076 * or SIGKILL.
1077 */
1078 if (info == SEND_SIG_FORCED)
1079 goto out_set;
1080
1081 /*
1082 * Real-time signals must be queued if sent by sigqueue, or
1083 * some other real-time mechanism. It is implementation
1084 * defined whether kill() does so. We attempt to do so, on
1085 * the principle of least surprise, but since kill is not
1086 * allowed to fail with EAGAIN when low on memory we just
1087 * make sure at least one signal gets delivered and don't
1088 * pass on the info struct.
1089 */
1090 if (sig < SIGRTMIN)
1091 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1092 else
1093 override_rlimit = 0;
1094
1095 q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
1096 override_rlimit);
1097 if (q) {
1098 list_add_tail(&q->list, &pending->list);
1099 switch ((unsigned long) info) {
1100 case (unsigned long) SEND_SIG_NOINFO:
1101 q->info.si_signo = sig;
1102 q->info.si_errno = 0;
1103 q->info.si_code = SI_USER;
1104 q->info.si_pid = task_tgid_nr_ns(current,
1105 task_active_pid_ns(t));
1106 q->info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
1107 break;
1108 case (unsigned long) SEND_SIG_PRIV:
1109 q->info.si_signo = sig;
1110 q->info.si_errno = 0;
1111 q->info.si_code = SI_KERNEL;
1112 q->info.si_pid = 0;
1113 q->info.si_uid = 0;
1114 break;
1115 default:
1116 copy_siginfo(&q->info, info);
1117 if (from_ancestor_ns)
1118 q->info.si_pid = 0;
1119 break;
1120 }
1121
1122 userns_fixup_signal_uid(&q->info, t);
1123
1124 } else if (!is_si_special(info)) {
1125 if (sig >= SIGRTMIN && info->si_code != SI_USER) {
1126 /*
1127 * Queue overflow, abort. We may abort if the
1128 * signal was rt and sent by user using something
1129 * other than kill().
1130 */
1131 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1132 ret = -EAGAIN;
1133 goto ret;
1134 } else {
1135 /*
1136 * This is a silent loss of information. We still
1137 * send the signal, but the *info bits are lost.
1138 */
1139 result = TRACE_SIGNAL_LOSE_INFO;
1140 }
1141 }
1142
1143out_set:
1144 signalfd_notify(t, sig);
1145 sigaddset(&pending->signal, sig);
1146 complete_signal(sig, t, group);
1147ret:
1148 trace_signal_generate(sig, info, t, group, result);
1149 return ret;
1150}
1151
1152static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
1153 int group)
1154{
1155 int from_ancestor_ns = 0;
1156
1157#ifdef CONFIG_PID_NS
1158 from_ancestor_ns = si_fromuser(info) &&
1159 !task_pid_nr_ns(current, task_active_pid_ns(t));
1160#endif
1161
1162 return __send_signal(sig, info, t, group, from_ancestor_ns);
1163}
1164
1165static void print_fatal_signal(int signr)
1166{
1167 struct pt_regs *regs = signal_pt_regs();
1168 printk(KERN_INFO "potentially unexpected fatal signal %d.\n", signr);
1169
1170#if defined(__i386__) && !defined(__arch_um__)
1171 printk(KERN_INFO "code at %08lx: ", regs->ip);
1172 {
1173 int i;
1174 for (i = 0; i < 16; i++) {
1175 unsigned char insn;
1176
1177 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1178 break;
1179 printk(KERN_CONT "%02x ", insn);
1180 }
1181 }
1182 printk(KERN_CONT "\n");
1183#endif
1184 preempt_disable();
1185 show_regs(regs);
1186 preempt_enable();
1187}
1188
1189static int __init setup_print_fatal_signals(char *str)
1190{
1191 get_option (&str, &print_fatal_signals);
1192
1193 return 1;
1194}
1195
1196__setup("print-fatal-signals=", setup_print_fatal_signals);
1197
1198int
1199__group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1200{
1201 return send_signal(sig, info, p, 1);
1202}
1203
1204static int
1205specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1206{
1207 return send_signal(sig, info, t, 0);
1208}
1209
1210int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
1211 bool group)
1212{
1213 unsigned long flags;
1214 int ret = -ESRCH;
1215
1216 if (lock_task_sighand(p, &flags)) {
1217 ret = send_signal(sig, info, p, group);
1218 unlock_task_sighand(p, &flags);
1219 }
1220
1221 return ret;
1222}
1223
1224/*
1225 * Force a signal that the process can't ignore: if necessary
1226 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1227 *
1228 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1229 * since we do not want to have a signal handler that was blocked
1230 * be invoked when user space had explicitly blocked it.
1231 *
1232 * We don't want to have recursive SIGSEGV's etc, for example,
1233 * that is why we also clear SIGNAL_UNKILLABLE.
1234 */
1235int
1236force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1237{
1238 unsigned long int flags;
1239 int ret, blocked, ignored;
1240 struct k_sigaction *action;
1241
1242 spin_lock_irqsave(&t->sighand->siglock, flags);
1243 action = &t->sighand->action[sig-1];
1244 ignored = action->sa.sa_handler == SIG_IGN;
1245 blocked = sigismember(&t->blocked, sig);
1246 if (blocked || ignored) {
1247 action->sa.sa_handler = SIG_DFL;
1248 if (blocked) {
1249 sigdelset(&t->blocked, sig);
1250 recalc_sigpending_and_wake(t);
1251 }
1252 }
1253 if (action->sa.sa_handler == SIG_DFL)
1254 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1255 ret = specific_send_sig_info(sig, info, t);
1256 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1257
1258 return ret;
1259}
1260
1261/*
1262 * Nuke all other threads in the group.
1263 */
1264int zap_other_threads(struct task_struct *p)
1265{
1266 struct task_struct *t = p;
1267 int count = 0;
1268
1269 p->signal->group_stop_count = 0;
1270
1271 while_each_thread(p, t) {
1272 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1273 count++;
1274
1275 /* Don't bother with already dead threads */
1276 if (t->exit_state)
1277 continue;
1278 sigaddset(&t->pending.signal, SIGKILL);
1279 signal_wake_up(t, 1);
1280 }
1281
1282 return count;
1283}
1284
1285struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1286 unsigned long *flags)
1287{
1288 struct sighand_struct *sighand;
1289
1290 for (;;) {
1291 local_irq_save(*flags);
1292 rcu_read_lock();
1293 sighand = rcu_dereference(tsk->sighand);
1294 if (unlikely(sighand == NULL)) {
1295 rcu_read_unlock();
1296 local_irq_restore(*flags);
1297 break;
1298 }
1299
1300 spin_lock(&sighand->siglock);
1301 if (likely(sighand == tsk->sighand)) {
1302 rcu_read_unlock();
1303 break;
1304 }
1305 spin_unlock(&sighand->siglock);
1306 rcu_read_unlock();
1307 local_irq_restore(*flags);
1308 }
1309
1310 return sighand;
1311}
1312
1313/*
1314 * send signal info to all the members of a group
1315 */
1316int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1317{
1318 int ret;
1319
1320 rcu_read_lock();
1321 ret = check_kill_permission(sig, info, p);
1322 rcu_read_unlock();
1323
1324 if (!ret && sig)
1325 ret = do_send_sig_info(sig, info, p, true);
1326
1327 return ret;
1328}
1329
1330/*
1331 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1332 * control characters do (^C, ^Z etc)
1333 * - the caller must hold at least a readlock on tasklist_lock
1334 */
1335int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1336{
1337 struct task_struct *p = NULL;
1338 int retval, success;
1339
1340 success = 0;
1341 retval = -ESRCH;
1342 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1343 int err = group_send_sig_info(sig, info, p);
1344 success |= !err;
1345 retval = err;
1346 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1347 return success ? 0 : retval;
1348}
1349
1350int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1351{
1352 int error = -ESRCH;
1353 struct task_struct *p;
1354
1355 rcu_read_lock();
1356retry:
1357 p = pid_task(pid, PIDTYPE_PID);
1358 if (p) {
1359 error = group_send_sig_info(sig, info, p);
1360 if (unlikely(error == -ESRCH))
1361 /*
1362 * The task was unhashed in between, try again.
1363 * If it is dead, pid_task() will return NULL,
1364 * if we race with de_thread() it will find the
1365 * new leader.
1366 */
1367 goto retry;
1368 }
1369 rcu_read_unlock();
1370
1371 return error;
1372}
1373
1374int kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1375{
1376 int error;
1377 rcu_read_lock();
1378 error = kill_pid_info(sig, info, find_vpid(pid));
1379 rcu_read_unlock();
1380 return error;
1381}
1382
1383static int kill_as_cred_perm(const struct cred *cred,
1384 struct task_struct *target)
1385{
1386 const struct cred *pcred = __task_cred(target);
1387 if (!uid_eq(cred->euid, pcred->suid) && !uid_eq(cred->euid, pcred->uid) &&
1388 !uid_eq(cred->uid, pcred->suid) && !uid_eq(cred->uid, pcred->uid))
1389 return 0;
1390 return 1;
1391}
1392
1393/* like kill_pid_info(), but doesn't use uid/euid of "current" */
1394int kill_pid_info_as_cred(int sig, struct siginfo *info, struct pid *pid,
1395 const struct cred *cred, u32 secid)
1396{
1397 int ret = -EINVAL;
1398 struct task_struct *p;
1399 unsigned long flags;
1400
1401 if (!valid_signal(sig))
1402 return ret;
1403
1404 rcu_read_lock();
1405 p = pid_task(pid, PIDTYPE_PID);
1406 if (!p) {
1407 ret = -ESRCH;
1408 goto out_unlock;
1409 }
1410 if (si_fromuser(info) && !kill_as_cred_perm(cred, p)) {
1411 ret = -EPERM;
1412 goto out_unlock;
1413 }
1414 ret = security_task_kill(p, info, sig, secid);
1415 if (ret)
1416 goto out_unlock;
1417
1418 if (sig) {
1419 if (lock_task_sighand(p, &flags)) {
1420 ret = __send_signal(sig, info, p, 1, 0);
1421 unlock_task_sighand(p, &flags);
1422 } else
1423 ret = -ESRCH;
1424 }
1425out_unlock:
1426 rcu_read_unlock();
1427 return ret;
1428}
1429EXPORT_SYMBOL_GPL(kill_pid_info_as_cred);
1430
1431/*
1432 * kill_something_info() interprets pid in interesting ways just like kill(2).
1433 *
1434 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1435 * is probably wrong. Should make it like BSD or SYSV.
1436 */
1437
1438static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1439{
1440 int ret;
1441
1442 if (pid > 0) {
1443 rcu_read_lock();
1444 ret = kill_pid_info(sig, info, find_vpid(pid));
1445 rcu_read_unlock();
1446 return ret;
1447 }
1448
1449 read_lock(&tasklist_lock);
1450 if (pid != -1) {
1451 ret = __kill_pgrp_info(sig, info,
1452 pid ? find_vpid(-pid) : task_pgrp(current));
1453 } else {
1454 int retval = 0, count = 0;
1455 struct task_struct * p;
1456
1457 for_each_process(p) {
1458 if (task_pid_vnr(p) > 1 &&
1459 !same_thread_group(p, current)) {
1460 int err = group_send_sig_info(sig, info, p);
1461 ++count;
1462 if (err != -EPERM)
1463 retval = err;
1464 }
1465 }
1466 ret = count ? retval : -ESRCH;
1467 }
1468 read_unlock(&tasklist_lock);
1469
1470 return ret;
1471}
1472
1473/*
1474 * These are for backward compatibility with the rest of the kernel source.
1475 */
1476
1477int send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1478{
1479 /*
1480 * Make sure legacy kernel users don't send in bad values
1481 * (normal paths check this in check_kill_permission).
1482 */
1483 if (!valid_signal(sig))
1484 return -EINVAL;
1485
1486 return do_send_sig_info(sig, info, p, false);
1487}
1488
1489#define __si_special(priv) \
1490 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1491
1492int
1493send_sig(int sig, struct task_struct *p, int priv)
1494{
1495 return send_sig_info(sig, __si_special(priv), p);
1496}
1497
1498void
1499force_sig(int sig, struct task_struct *p)
1500{
1501 force_sig_info(sig, SEND_SIG_PRIV, p);
1502}
1503
1504/*
1505 * When things go south during signal handling, we
1506 * will force a SIGSEGV. And if the signal that caused
1507 * the problem was already a SIGSEGV, we'll want to
1508 * make sure we don't even try to deliver the signal..
1509 */
1510int
1511force_sigsegv(int sig, struct task_struct *p)
1512{
1513 if (sig == SIGSEGV) {
1514 unsigned long flags;
1515 spin_lock_irqsave(&p->sighand->siglock, flags);
1516 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1517 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1518 }
1519 force_sig(SIGSEGV, p);
1520 return 0;
1521}
1522
1523int kill_pgrp(struct pid *pid, int sig, int priv)
1524{
1525 int ret;
1526
1527 read_lock(&tasklist_lock);
1528 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1529 read_unlock(&tasklist_lock);
1530
1531 return ret;
1532}
1533EXPORT_SYMBOL(kill_pgrp);
1534
1535int kill_pid(struct pid *pid, int sig, int priv)
1536{
1537 return kill_pid_info(sig, __si_special(priv), pid);
1538}
1539EXPORT_SYMBOL(kill_pid);
1540
1541/*
1542 * These functions support sending signals using preallocated sigqueue
1543 * structures. This is needed "because realtime applications cannot
1544 * afford to lose notifications of asynchronous events, like timer
1545 * expirations or I/O completions". In the case of POSIX Timers
1546 * we allocate the sigqueue structure from the timer_create. If this
1547 * allocation fails we are able to report the failure to the application
1548 * with an EAGAIN error.
1549 */
1550struct sigqueue *sigqueue_alloc(void)
1551{
1552 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1553
1554 if (q)
1555 q->flags |= SIGQUEUE_PREALLOC;
1556
1557 return q;
1558}
1559
1560void sigqueue_free(struct sigqueue *q)
1561{
1562 unsigned long flags;
1563 spinlock_t *lock = ¤t->sighand->siglock;
1564
1565 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1566 /*
1567 * We must hold ->siglock while testing q->list
1568 * to serialize with collect_signal() or with
1569 * __exit_signal()->flush_sigqueue().
1570 */
1571 spin_lock_irqsave(lock, flags);
1572 q->flags &= ~SIGQUEUE_PREALLOC;
1573 /*
1574 * If it is queued it will be freed when dequeued,
1575 * like the "regular" sigqueue.
1576 */
1577 if (!list_empty(&q->list))
1578 q = NULL;
1579 spin_unlock_irqrestore(lock, flags);
1580
1581 if (q)
1582 __sigqueue_free(q);
1583}
1584
1585int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1586{
1587 int sig = q->info.si_signo;
1588 struct sigpending *pending;
1589 unsigned long flags;
1590 int ret, result;
1591
1592 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1593
1594 ret = -1;
1595 if (!likely(lock_task_sighand(t, &flags)))
1596 goto ret;
1597
1598 ret = 1; /* the signal is ignored */
1599 result = TRACE_SIGNAL_IGNORED;
1600 if (!prepare_signal(sig, t, false))
1601 goto out;
1602
1603 ret = 0;
1604 if (unlikely(!list_empty(&q->list))) {
1605 /*
1606 * If an SI_TIMER entry is already queue just increment
1607 * the overrun count.
1608 */
1609 BUG_ON(q->info.si_code != SI_TIMER);
1610 q->info.si_overrun++;
1611 result = TRACE_SIGNAL_ALREADY_PENDING;
1612 goto out;
1613 }
1614 q->info.si_overrun = 0;
1615
1616 signalfd_notify(t, sig);
1617 pending = group ? &t->signal->shared_pending : &t->pending;
1618 list_add_tail(&q->list, &pending->list);
1619 sigaddset(&pending->signal, sig);
1620 complete_signal(sig, t, group);
1621 result = TRACE_SIGNAL_DELIVERED;
1622out:
1623 trace_signal_generate(sig, &q->info, t, group, result);
1624 unlock_task_sighand(t, &flags);
1625ret:
1626 return ret;
1627}
1628
1629/*
1630 * Let a parent know about the death of a child.
1631 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1632 *
1633 * Returns true if our parent ignored us and so we've switched to
1634 * self-reaping.
1635 */
1636bool do_notify_parent(struct task_struct *tsk, int sig)
1637{
1638 struct siginfo info;
1639 unsigned long flags;
1640 struct sighand_struct *psig;
1641 bool autoreap = false;
1642 cputime_t utime, stime;
1643
1644 BUG_ON(sig == -1);
1645
1646 /* do_notify_parent_cldstop should have been called instead. */
1647 BUG_ON(task_is_stopped_or_traced(tsk));
1648
1649 BUG_ON(!tsk->ptrace &&
1650 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1651
1652 if (sig != SIGCHLD) {
1653 /*
1654 * This is only possible if parent == real_parent.
1655 * Check if it has changed security domain.
1656 */
1657 if (tsk->parent_exec_id != tsk->parent->self_exec_id)
1658 sig = SIGCHLD;
1659 }
1660
1661 info.si_signo = sig;
1662 info.si_errno = 0;
1663 /*
1664 * We are under tasklist_lock here so our parent is tied to
1665 * us and cannot change.
1666 *
1667 * task_active_pid_ns will always return the same pid namespace
1668 * until a task passes through release_task.
1669 *
1670 * write_lock() currently calls preempt_disable() which is the
1671 * same as rcu_read_lock(), but according to Oleg, this is not
1672 * correct to rely on this
1673 */
1674 rcu_read_lock();
1675 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1676 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1677 task_uid(tsk));
1678 rcu_read_unlock();
1679
1680 task_cputime(tsk, &utime, &stime);
1681 info.si_utime = cputime_to_clock_t(utime + tsk->signal->utime);
1682 info.si_stime = cputime_to_clock_t(stime + tsk->signal->stime);
1683
1684 info.si_status = tsk->exit_code & 0x7f;
1685 if (tsk->exit_code & 0x80)
1686 info.si_code = CLD_DUMPED;
1687 else if (tsk->exit_code & 0x7f)
1688 info.si_code = CLD_KILLED;
1689 else {
1690 info.si_code = CLD_EXITED;
1691 info.si_status = tsk->exit_code >> 8;
1692 }
1693
1694 psig = tsk->parent->sighand;
1695 spin_lock_irqsave(&psig->siglock, flags);
1696 if (!tsk->ptrace && sig == SIGCHLD &&
1697 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1698 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1699 /*
1700 * We are exiting and our parent doesn't care. POSIX.1
1701 * defines special semantics for setting SIGCHLD to SIG_IGN
1702 * or setting the SA_NOCLDWAIT flag: we should be reaped
1703 * automatically and not left for our parent's wait4 call.
1704 * Rather than having the parent do it as a magic kind of
1705 * signal handler, we just set this to tell do_exit that we
1706 * can be cleaned up without becoming a zombie. Note that
1707 * we still call __wake_up_parent in this case, because a
1708 * blocked sys_wait4 might now return -ECHILD.
1709 *
1710 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1711 * is implementation-defined: we do (if you don't want
1712 * it, just use SIG_IGN instead).
1713 */
1714 autoreap = true;
1715 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1716 sig = 0;
1717 }
1718 if (valid_signal(sig) && sig)
1719 __group_send_sig_info(sig, &info, tsk->parent);
1720 __wake_up_parent(tsk, tsk->parent);
1721 spin_unlock_irqrestore(&psig->siglock, flags);
1722
1723 return autoreap;
1724}
1725
1726/**
1727 * do_notify_parent_cldstop - notify parent of stopped/continued state change
1728 * @tsk: task reporting the state change
1729 * @for_ptracer: the notification is for ptracer
1730 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
1731 *
1732 * Notify @tsk's parent that the stopped/continued state has changed. If
1733 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
1734 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
1735 *
1736 * CONTEXT:
1737 * Must be called with tasklist_lock at least read locked.
1738 */
1739static void do_notify_parent_cldstop(struct task_struct *tsk,
1740 bool for_ptracer, int why)
1741{
1742 struct siginfo info;
1743 unsigned long flags;
1744 struct task_struct *parent;
1745 struct sighand_struct *sighand;
1746 cputime_t utime, stime;
1747
1748 if (for_ptracer) {
1749 parent = tsk->parent;
1750 } else {
1751 tsk = tsk->group_leader;
1752 parent = tsk->real_parent;
1753 }
1754
1755 info.si_signo = SIGCHLD;
1756 info.si_errno = 0;
1757 /*
1758 * see comment in do_notify_parent() about the following 4 lines
1759 */
1760 rcu_read_lock();
1761 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
1762 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
1763 rcu_read_unlock();
1764
1765 task_cputime(tsk, &utime, &stime);
1766 info.si_utime = cputime_to_clock_t(utime);
1767 info.si_stime = cputime_to_clock_t(stime);
1768
1769 info.si_code = why;
1770 switch (why) {
1771 case CLD_CONTINUED:
1772 info.si_status = SIGCONT;
1773 break;
1774 case CLD_STOPPED:
1775 info.si_status = tsk->signal->group_exit_code & 0x7f;
1776 break;
1777 case CLD_TRAPPED:
1778 info.si_status = tsk->exit_code & 0x7f;
1779 break;
1780 default:
1781 BUG();
1782 }
1783
1784 sighand = parent->sighand;
1785 spin_lock_irqsave(&sighand->siglock, flags);
1786 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1787 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1788 __group_send_sig_info(SIGCHLD, &info, parent);
1789 /*
1790 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1791 */
1792 __wake_up_parent(tsk, parent);
1793 spin_unlock_irqrestore(&sighand->siglock, flags);
1794}
1795
1796static inline int may_ptrace_stop(void)
1797{
1798 if (!likely(current->ptrace))
1799 return 0;
1800 /*
1801 * Are we in the middle of do_coredump?
1802 * If so and our tracer is also part of the coredump stopping
1803 * is a deadlock situation, and pointless because our tracer
1804 * is dead so don't allow us to stop.
1805 * If SIGKILL was already sent before the caller unlocked
1806 * ->siglock we must see ->core_state != NULL. Otherwise it
1807 * is safe to enter schedule().
1808 *
1809 * This is almost outdated, a task with the pending SIGKILL can't
1810 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
1811 * after SIGKILL was already dequeued.
1812 */
1813 if (unlikely(current->mm->core_state) &&
1814 unlikely(current->mm == current->parent->mm))
1815 return 0;
1816
1817 return 1;
1818}
1819
1820/*
1821 * Return non-zero if there is a SIGKILL that should be waking us up.
1822 * Called with the siglock held.
1823 */
1824static int sigkill_pending(struct task_struct *tsk)
1825{
1826 return sigismember(&tsk->pending.signal, SIGKILL) ||
1827 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1828}
1829
1830/*
1831 * This must be called with current->sighand->siglock held.
1832 *
1833 * This should be the path for all ptrace stops.
1834 * We always set current->last_siginfo while stopped here.
1835 * That makes it a way to test a stopped process for
1836 * being ptrace-stopped vs being job-control-stopped.
1837 *
1838 * If we actually decide not to stop at all because the tracer
1839 * is gone, we keep current->exit_code unless clear_code.
1840 */
1841static void ptrace_stop(int exit_code, int why, int clear_code, siginfo_t *info)
1842 __releases(¤t->sighand->siglock)
1843 __acquires(¤t->sighand->siglock)
1844{
1845 bool gstop_done = false;
1846
1847 if (arch_ptrace_stop_needed(exit_code, info)) {
1848 /*
1849 * The arch code has something special to do before a
1850 * ptrace stop. This is allowed to block, e.g. for faults
1851 * on user stack pages. We can't keep the siglock while
1852 * calling arch_ptrace_stop, so we must release it now.
1853 * To preserve proper semantics, we must do this before
1854 * any signal bookkeeping like checking group_stop_count.
1855 * Meanwhile, a SIGKILL could come in before we retake the
1856 * siglock. That must prevent us from sleeping in TASK_TRACED.
1857 * So after regaining the lock, we must check for SIGKILL.
1858 */
1859 spin_unlock_irq(¤t->sighand->siglock);
1860 arch_ptrace_stop(exit_code, info);
1861 spin_lock_irq(¤t->sighand->siglock);
1862 if (sigkill_pending(current))
1863 return;
1864 }
1865
1866 /*
1867 * We're committing to trapping. TRACED should be visible before
1868 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
1869 * Also, transition to TRACED and updates to ->jobctl should be
1870 * atomic with respect to siglock and should be done after the arch
1871 * hook as siglock is released and regrabbed across it.
1872 */
1873 set_current_state(TASK_TRACED);
1874
1875 current->last_siginfo = info;
1876 current->exit_code = exit_code;
1877
1878 /*
1879 * If @why is CLD_STOPPED, we're trapping to participate in a group
1880 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
1881 * across siglock relocks since INTERRUPT was scheduled, PENDING
1882 * could be clear now. We act as if SIGCONT is received after
1883 * TASK_TRACED is entered - ignore it.
1884 */
1885 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
1886 gstop_done = task_participate_group_stop(current);
1887
1888 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
1889 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
1890 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
1891 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
1892
1893 /* entering a trap, clear TRAPPING */
1894 task_clear_jobctl_trapping(current);
1895
1896 spin_unlock_irq(¤t->sighand->siglock);
1897 read_lock(&tasklist_lock);
1898 if (may_ptrace_stop()) {
1899 /*
1900 * Notify parents of the stop.
1901 *
1902 * While ptraced, there are two parents - the ptracer and
1903 * the real_parent of the group_leader. The ptracer should
1904 * know about every stop while the real parent is only
1905 * interested in the completion of group stop. The states
1906 * for the two don't interact with each other. Notify
1907 * separately unless they're gonna be duplicates.
1908 */
1909 do_notify_parent_cldstop(current, true, why);
1910 if (gstop_done && ptrace_reparented(current))
1911 do_notify_parent_cldstop(current, false, why);
1912
1913 /*
1914 * Don't want to allow preemption here, because
1915 * sys_ptrace() needs this task to be inactive.
1916 *
1917 * XXX: implement read_unlock_no_resched().
1918 */
1919 preempt_disable();
1920 read_unlock(&tasklist_lock);
1921 preempt_enable_no_resched();
1922 freezable_schedule();
1923 } else {
1924 /*
1925 * By the time we got the lock, our tracer went away.
1926 * Don't drop the lock yet, another tracer may come.
1927 *
1928 * If @gstop_done, the ptracer went away between group stop
1929 * completion and here. During detach, it would have set
1930 * JOBCTL_STOP_PENDING on us and we'll re-enter
1931 * TASK_STOPPED in do_signal_stop() on return, so notifying
1932 * the real parent of the group stop completion is enough.
1933 */
1934 if (gstop_done)
1935 do_notify_parent_cldstop(current, false, why);
1936
1937 /* tasklist protects us from ptrace_freeze_traced() */
1938 __set_current_state(TASK_RUNNING);
1939 if (clear_code)
1940 current->exit_code = 0;
1941 read_unlock(&tasklist_lock);
1942 }
1943
1944 /*
1945 * We are back. Now reacquire the siglock before touching
1946 * last_siginfo, so that we are sure to have synchronized with
1947 * any signal-sending on another CPU that wants to examine it.
1948 */
1949 spin_lock_irq(¤t->sighand->siglock);
1950 current->last_siginfo = NULL;
1951
1952 /* LISTENING can be set only during STOP traps, clear it */
1953 current->jobctl &= ~JOBCTL_LISTENING;
1954
1955 /*
1956 * Queued signals ignored us while we were stopped for tracing.
1957 * So check for any that we should take before resuming user mode.
1958 * This sets TIF_SIGPENDING, but never clears it.
1959 */
1960 recalc_sigpending_tsk(current);
1961}
1962
1963static void ptrace_do_notify(int signr, int exit_code, int why)
1964{
1965 siginfo_t info;
1966
1967 memset(&info, 0, sizeof info);
1968 info.si_signo = signr;
1969 info.si_code = exit_code;
1970 info.si_pid = task_pid_vnr(current);
1971 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
1972
1973 /* Let the debugger run. */
1974 ptrace_stop(exit_code, why, 1, &info);
1975}
1976
1977void ptrace_notify(int exit_code)
1978{
1979 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1980 if (unlikely(current->task_works))
1981 task_work_run();
1982
1983 spin_lock_irq(¤t->sighand->siglock);
1984 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
1985 spin_unlock_irq(¤t->sighand->siglock);
1986}
1987
1988/**
1989 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
1990 * @signr: signr causing group stop if initiating
1991 *
1992 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
1993 * and participate in it. If already set, participate in the existing
1994 * group stop. If participated in a group stop (and thus slept), %true is
1995 * returned with siglock released.
1996 *
1997 * If ptraced, this function doesn't handle stop itself. Instead,
1998 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
1999 * untouched. The caller must ensure that INTERRUPT trap handling takes
2000 * places afterwards.
2001 *
2002 * CONTEXT:
2003 * Must be called with @current->sighand->siglock held, which is released
2004 * on %true return.
2005 *
2006 * RETURNS:
2007 * %false if group stop is already cancelled or ptrace trap is scheduled.
2008 * %true if participated in group stop.
2009 */
2010static bool do_signal_stop(int signr)
2011 __releases(¤t->sighand->siglock)
2012{
2013 struct signal_struct *sig = current->signal;
2014
2015 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2016 unsigned int gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2017 struct task_struct *t;
2018
2019 /* signr will be recorded in task->jobctl for retries */
2020 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2021
2022 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2023 unlikely(signal_group_exit(sig)))
2024 return false;
2025 /*
2026 * There is no group stop already in progress. We must
2027 * initiate one now.
2028 *
2029 * While ptraced, a task may be resumed while group stop is
2030 * still in effect and then receive a stop signal and
2031 * initiate another group stop. This deviates from the
2032 * usual behavior as two consecutive stop signals can't
2033 * cause two group stops when !ptraced. That is why we
2034 * also check !task_is_stopped(t) below.
2035 *
2036 * The condition can be distinguished by testing whether
2037 * SIGNAL_STOP_STOPPED is already set. Don't generate
2038 * group_exit_code in such case.
2039 *
2040 * This is not necessary for SIGNAL_STOP_CONTINUED because
2041 * an intervening stop signal is required to cause two
2042 * continued events regardless of ptrace.
2043 */
2044 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2045 sig->group_exit_code = signr;
2046
2047 sig->group_stop_count = 0;
2048
2049 if (task_set_jobctl_pending(current, signr | gstop))
2050 sig->group_stop_count++;
2051
2052 t = current;
2053 while_each_thread(current, t) {
2054 /*
2055 * Setting state to TASK_STOPPED for a group
2056 * stop is always done with the siglock held,
2057 * so this check has no races.
2058 */
2059 if (!task_is_stopped(t) &&
2060 task_set_jobctl_pending(t, signr | gstop)) {
2061 sig->group_stop_count++;
2062 if (likely(!(t->ptrace & PT_SEIZED)))
2063 signal_wake_up(t, 0);
2064 else
2065 ptrace_trap_notify(t);
2066 }
2067 }
2068 }
2069
2070 if (likely(!current->ptrace)) {
2071 int notify = 0;
2072
2073 /*
2074 * If there are no other threads in the group, or if there
2075 * is a group stop in progress and we are the last to stop,
2076 * report to the parent.
2077 */
2078 if (task_participate_group_stop(current))
2079 notify = CLD_STOPPED;
2080
2081 __set_current_state(TASK_STOPPED);
2082 spin_unlock_irq(¤t->sighand->siglock);
2083
2084 /*
2085 * Notify the parent of the group stop completion. Because
2086 * we're not holding either the siglock or tasklist_lock
2087 * here, ptracer may attach inbetween; however, this is for
2088 * group stop and should always be delivered to the real
2089 * parent of the group leader. The new ptracer will get
2090 * its notification when this task transitions into
2091 * TASK_TRACED.
2092 */
2093 if (notify) {
2094 read_lock(&tasklist_lock);
2095 do_notify_parent_cldstop(current, false, notify);
2096 read_unlock(&tasklist_lock);
2097 }
2098
2099 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2100 freezable_schedule();
2101 return true;
2102 } else {
2103 /*
2104 * While ptraced, group stop is handled by STOP trap.
2105 * Schedule it and let the caller deal with it.
2106 */
2107 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2108 return false;
2109 }
2110}
2111
2112/**
2113 * do_jobctl_trap - take care of ptrace jobctl traps
2114 *
2115 * When PT_SEIZED, it's used for both group stop and explicit
2116 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2117 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2118 * the stop signal; otherwise, %SIGTRAP.
2119 *
2120 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2121 * number as exit_code and no siginfo.
2122 *
2123 * CONTEXT:
2124 * Must be called with @current->sighand->siglock held, which may be
2125 * released and re-acquired before returning with intervening sleep.
2126 */
2127static void do_jobctl_trap(void)
2128{
2129 struct signal_struct *signal = current->signal;
2130 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2131
2132 if (current->ptrace & PT_SEIZED) {
2133 if (!signal->group_stop_count &&
2134 !(signal->flags & SIGNAL_STOP_STOPPED))
2135 signr = SIGTRAP;
2136 WARN_ON_ONCE(!signr);
2137 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2138 CLD_STOPPED);
2139 } else {
2140 WARN_ON_ONCE(!signr);
2141 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2142 current->exit_code = 0;
2143 }
2144}
2145
2146static int ptrace_signal(int signr, siginfo_t *info)
2147{
2148 ptrace_signal_deliver();
2149 /*
2150 * We do not check sig_kernel_stop(signr) but set this marker
2151 * unconditionally because we do not know whether debugger will
2152 * change signr. This flag has no meaning unless we are going
2153 * to stop after return from ptrace_stop(). In this case it will
2154 * be checked in do_signal_stop(), we should only stop if it was
2155 * not cleared by SIGCONT while we were sleeping. See also the
2156 * comment in dequeue_signal().
2157 */
2158 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2159 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2160
2161 /* We're back. Did the debugger cancel the sig? */
2162 signr = current->exit_code;
2163 if (signr == 0)
2164 return signr;
2165
2166 current->exit_code = 0;
2167
2168 /*
2169 * Update the siginfo structure if the signal has
2170 * changed. If the debugger wanted something
2171 * specific in the siginfo structure then it should
2172 * have updated *info via PTRACE_SETSIGINFO.
2173 */
2174 if (signr != info->si_signo) {
2175 info->si_signo = signr;
2176 info->si_errno = 0;
2177 info->si_code = SI_USER;
2178 rcu_read_lock();
2179 info->si_pid = task_pid_vnr(current->parent);
2180 info->si_uid = from_kuid_munged(current_user_ns(),
2181 task_uid(current->parent));
2182 rcu_read_unlock();
2183 }
2184
2185 /* If the (new) signal is now blocked, requeue it. */
2186 if (sigismember(¤t->blocked, signr)) {
2187 specific_send_sig_info(signr, info, current);
2188 signr = 0;
2189 }
2190
2191 return signr;
2192}
2193
2194int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
2195 struct pt_regs *regs, void *cookie)
2196{
2197 struct sighand_struct *sighand = current->sighand;
2198 struct signal_struct *signal = current->signal;
2199 int signr;
2200
2201 if (unlikely(current->task_works))
2202 task_work_run();
2203
2204 if (unlikely(uprobe_deny_signal()))
2205 return 0;
2206
2207 /*
2208 * Do this once, we can't return to user-mode if freezing() == T.
2209 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2210 * thus do not need another check after return.
2211 */
2212 try_to_freeze();
2213
2214relock:
2215 spin_lock_irq(&sighand->siglock);
2216 /*
2217 * Every stopped thread goes here after wakeup. Check to see if
2218 * we should notify the parent, prepare_signal(SIGCONT) encodes
2219 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2220 */
2221 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2222 int why;
2223
2224 if (signal->flags & SIGNAL_CLD_CONTINUED)
2225 why = CLD_CONTINUED;
2226 else
2227 why = CLD_STOPPED;
2228
2229 signal->flags &= ~SIGNAL_CLD_MASK;
2230
2231 spin_unlock_irq(&sighand->siglock);
2232
2233 /*
2234 * Notify the parent that we're continuing. This event is
2235 * always per-process and doesn't make whole lot of sense
2236 * for ptracers, who shouldn't consume the state via
2237 * wait(2) either, but, for backward compatibility, notify
2238 * the ptracer of the group leader too unless it's gonna be
2239 * a duplicate.
2240 */
2241 read_lock(&tasklist_lock);
2242 do_notify_parent_cldstop(current, false, why);
2243
2244 if (ptrace_reparented(current->group_leader))
2245 do_notify_parent_cldstop(current->group_leader,
2246 true, why);
2247 read_unlock(&tasklist_lock);
2248
2249 goto relock;
2250 }
2251
2252 for (;;) {
2253 struct k_sigaction *ka;
2254
2255 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2256 do_signal_stop(0))
2257 goto relock;
2258
2259 if (unlikely(current->jobctl & JOBCTL_TRAP_MASK)) {
2260 do_jobctl_trap();
2261 spin_unlock_irq(&sighand->siglock);
2262 goto relock;
2263 }
2264
2265 signr = dequeue_signal(current, ¤t->blocked, info);
2266
2267 if (!signr)
2268 break; /* will return 0 */
2269
2270 if (unlikely(current->ptrace) && signr != SIGKILL) {
2271 signr = ptrace_signal(signr, info);
2272 if (!signr)
2273 continue;
2274 }
2275
2276 ka = &sighand->action[signr-1];
2277
2278 /* Trace actually delivered signals. */
2279 trace_signal_deliver(signr, info, ka);
2280
2281 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2282 continue;
2283 if (ka->sa.sa_handler != SIG_DFL) {
2284 /* Run the handler. */
2285 *return_ka = *ka;
2286
2287 if (ka->sa.sa_flags & SA_ONESHOT)
2288 ka->sa.sa_handler = SIG_DFL;
2289
2290 break; /* will return non-zero "signr" value */
2291 }
2292
2293 /*
2294 * Now we are doing the default action for this signal.
2295 */
2296 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2297 continue;
2298
2299 /*
2300 * Global init gets no signals it doesn't want.
2301 * Container-init gets no signals it doesn't want from same
2302 * container.
2303 *
2304 * Note that if global/container-init sees a sig_kernel_only()
2305 * signal here, the signal must have been generated internally
2306 * or must have come from an ancestor namespace. In either
2307 * case, the signal cannot be dropped.
2308 */
2309 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2310 !sig_kernel_only(signr))
2311 continue;
2312
2313 if (sig_kernel_stop(signr)) {
2314 /*
2315 * The default action is to stop all threads in
2316 * the thread group. The job control signals
2317 * do nothing in an orphaned pgrp, but SIGSTOP
2318 * always works. Note that siglock needs to be
2319 * dropped during the call to is_orphaned_pgrp()
2320 * because of lock ordering with tasklist_lock.
2321 * This allows an intervening SIGCONT to be posted.
2322 * We need to check for that and bail out if necessary.
2323 */
2324 if (signr != SIGSTOP) {
2325 spin_unlock_irq(&sighand->siglock);
2326
2327 /* signals can be posted during this window */
2328
2329 if (is_current_pgrp_orphaned())
2330 goto relock;
2331
2332 spin_lock_irq(&sighand->siglock);
2333 }
2334
2335 if (likely(do_signal_stop(info->si_signo))) {
2336 /* It released the siglock. */
2337 goto relock;
2338 }
2339
2340 /*
2341 * We didn't actually stop, due to a race
2342 * with SIGCONT or something like that.
2343 */
2344 continue;
2345 }
2346
2347 spin_unlock_irq(&sighand->siglock);
2348
2349 /*
2350 * Anything else is fatal, maybe with a core dump.
2351 */
2352 current->flags |= PF_SIGNALED;
2353
2354 if (sig_kernel_coredump(signr)) {
2355 if (print_fatal_signals)
2356 print_fatal_signal(info->si_signo);
2357 proc_coredump_connector(current);
2358 /*
2359 * If it was able to dump core, this kills all
2360 * other threads in the group and synchronizes with
2361 * their demise. If we lost the race with another
2362 * thread getting here, it set group_exit_code
2363 * first and our do_group_exit call below will use
2364 * that value and ignore the one we pass it.
2365 */
2366 do_coredump(info);
2367 }
2368
2369 /*
2370 * Death signals, no core dump.
2371 */
2372 do_group_exit(info->si_signo);
2373 /* NOTREACHED */
2374 }
2375 spin_unlock_irq(&sighand->siglock);
2376 return signr;
2377}
2378
2379/**
2380 * signal_delivered -
2381 * @sig: number of signal being delivered
2382 * @info: siginfo_t of signal being delivered
2383 * @ka: sigaction setting that chose the handler
2384 * @regs: user register state
2385 * @stepping: nonzero if debugger single-step or block-step in use
2386 *
2387 * This function should be called when a signal has successfully been
2388 * delivered. It updates the blocked signals accordingly (@ka->sa.sa_mask
2389 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2390 * is set in @ka->sa.sa_flags. Tracing is notified.
2391 */
2392void signal_delivered(int sig, siginfo_t *info, struct k_sigaction *ka,
2393 struct pt_regs *regs, int stepping)
2394{
2395 sigset_t blocked;
2396
2397 /* A signal was successfully delivered, and the
2398 saved sigmask was stored on the signal frame,
2399 and will be restored by sigreturn. So we can
2400 simply clear the restore sigmask flag. */
2401 clear_restore_sigmask();
2402
2403 sigorsets(&blocked, ¤t->blocked, &ka->sa.sa_mask);
2404 if (!(ka->sa.sa_flags & SA_NODEFER))
2405 sigaddset(&blocked, sig);
2406 set_current_blocked(&blocked);
2407 tracehook_signal_handler(sig, info, ka, regs, stepping);
2408}
2409
2410void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2411{
2412 if (failed)
2413 force_sigsegv(ksig->sig, current);
2414 else
2415 signal_delivered(ksig->sig, &ksig->info, &ksig->ka,
2416 signal_pt_regs(), stepping);
2417}
2418
2419/*
2420 * It could be that complete_signal() picked us to notify about the
2421 * group-wide signal. Other threads should be notified now to take
2422 * the shared signals in @which since we will not.
2423 */
2424static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2425{
2426 sigset_t retarget;
2427 struct task_struct *t;
2428
2429 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2430 if (sigisemptyset(&retarget))
2431 return;
2432
2433 t = tsk;
2434 while_each_thread(tsk, t) {
2435 if (t->flags & PF_EXITING)
2436 continue;
2437
2438 if (!has_pending_signals(&retarget, &t->blocked))
2439 continue;
2440 /* Remove the signals this thread can handle. */
2441 sigandsets(&retarget, &retarget, &t->blocked);
2442
2443 if (!signal_pending(t))
2444 signal_wake_up(t, 0);
2445
2446 if (sigisemptyset(&retarget))
2447 break;
2448 }
2449}
2450
2451void exit_signals(struct task_struct *tsk)
2452{
2453 int group_stop = 0;
2454 sigset_t unblocked;
2455
2456 /*
2457 * @tsk is about to have PF_EXITING set - lock out users which
2458 * expect stable threadgroup.
2459 */
2460 threadgroup_change_begin(tsk);
2461
2462 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2463 tsk->flags |= PF_EXITING;
2464 threadgroup_change_end(tsk);
2465 return;
2466 }
2467
2468 spin_lock_irq(&tsk->sighand->siglock);
2469 /*
2470 * From now this task is not visible for group-wide signals,
2471 * see wants_signal(), do_signal_stop().
2472 */
2473 tsk->flags |= PF_EXITING;
2474
2475 threadgroup_change_end(tsk);
2476
2477 if (!signal_pending(tsk))
2478 goto out;
2479
2480 unblocked = tsk->blocked;
2481 signotset(&unblocked);
2482 retarget_shared_pending(tsk, &unblocked);
2483
2484 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2485 task_participate_group_stop(tsk))
2486 group_stop = CLD_STOPPED;
2487out:
2488 spin_unlock_irq(&tsk->sighand->siglock);
2489
2490 /*
2491 * If group stop has completed, deliver the notification. This
2492 * should always go to the real parent of the group leader.
2493 */
2494 if (unlikely(group_stop)) {
2495 read_lock(&tasklist_lock);
2496 do_notify_parent_cldstop(tsk, false, group_stop);
2497 read_unlock(&tasklist_lock);
2498 }
2499}
2500
2501EXPORT_SYMBOL(recalc_sigpending);
2502EXPORT_SYMBOL_GPL(dequeue_signal);
2503EXPORT_SYMBOL(flush_signals);
2504EXPORT_SYMBOL(force_sig);
2505EXPORT_SYMBOL(send_sig);
2506EXPORT_SYMBOL(send_sig_info);
2507EXPORT_SYMBOL(sigprocmask);
2508EXPORT_SYMBOL(block_all_signals);
2509EXPORT_SYMBOL(unblock_all_signals);
2510
2511
2512/*
2513 * System call entry points.
2514 */
2515
2516/**
2517 * sys_restart_syscall - restart a system call
2518 */
2519SYSCALL_DEFINE0(restart_syscall)
2520{
2521 struct restart_block *restart = ¤t_thread_info()->restart_block;
2522 return restart->fn(restart);
2523}
2524
2525long do_no_restart_syscall(struct restart_block *param)
2526{
2527 return -EINTR;
2528}
2529
2530static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2531{
2532 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2533 sigset_t newblocked;
2534 /* A set of now blocked but previously unblocked signals. */
2535 sigandnsets(&newblocked, newset, ¤t->blocked);
2536 retarget_shared_pending(tsk, &newblocked);
2537 }
2538 tsk->blocked = *newset;
2539 recalc_sigpending();
2540}
2541
2542/**
2543 * set_current_blocked - change current->blocked mask
2544 * @newset: new mask
2545 *
2546 * It is wrong to change ->blocked directly, this helper should be used
2547 * to ensure the process can't miss a shared signal we are going to block.
2548 */
2549void set_current_blocked(sigset_t *newset)
2550{
2551 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2552 __set_current_blocked(newset);
2553}
2554
2555void __set_current_blocked(const sigset_t *newset)
2556{
2557 struct task_struct *tsk = current;
2558
2559 spin_lock_irq(&tsk->sighand->siglock);
2560 __set_task_blocked(tsk, newset);
2561 spin_unlock_irq(&tsk->sighand->siglock);
2562}
2563
2564/*
2565 * This is also useful for kernel threads that want to temporarily
2566 * (or permanently) block certain signals.
2567 *
2568 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2569 * interface happily blocks "unblockable" signals like SIGKILL
2570 * and friends.
2571 */
2572int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2573{
2574 struct task_struct *tsk = current;
2575 sigset_t newset;
2576
2577 /* Lockless, only current can change ->blocked, never from irq */
2578 if (oldset)
2579 *oldset = tsk->blocked;
2580
2581 switch (how) {
2582 case SIG_BLOCK:
2583 sigorsets(&newset, &tsk->blocked, set);
2584 break;
2585 case SIG_UNBLOCK:
2586 sigandnsets(&newset, &tsk->blocked, set);
2587 break;
2588 case SIG_SETMASK:
2589 newset = *set;
2590 break;
2591 default:
2592 return -EINVAL;
2593 }
2594
2595 __set_current_blocked(&newset);
2596 return 0;
2597}
2598
2599/**
2600 * sys_rt_sigprocmask - change the list of currently blocked signals
2601 * @how: whether to add, remove, or set signals
2602 * @nset: stores pending signals
2603 * @oset: previous value of signal mask if non-null
2604 * @sigsetsize: size of sigset_t type
2605 */
2606SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
2607 sigset_t __user *, oset, size_t, sigsetsize)
2608{
2609 sigset_t old_set, new_set;
2610 int error;
2611
2612 /* XXX: Don't preclude handling different sized sigset_t's. */
2613 if (sigsetsize != sizeof(sigset_t))
2614 return -EINVAL;
2615
2616 old_set = current->blocked;
2617
2618 if (nset) {
2619 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
2620 return -EFAULT;
2621 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2622
2623 error = sigprocmask(how, &new_set, NULL);
2624 if (error)
2625 return error;
2626 }
2627
2628 if (oset) {
2629 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
2630 return -EFAULT;
2631 }
2632
2633 return 0;
2634}
2635
2636#ifdef CONFIG_COMPAT
2637COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
2638 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
2639{
2640#ifdef __BIG_ENDIAN
2641 sigset_t old_set = current->blocked;
2642
2643 /* XXX: Don't preclude handling different sized sigset_t's. */
2644 if (sigsetsize != sizeof(sigset_t))
2645 return -EINVAL;
2646
2647 if (nset) {
2648 compat_sigset_t new32;
2649 sigset_t new_set;
2650 int error;
2651 if (copy_from_user(&new32, nset, sizeof(compat_sigset_t)))
2652 return -EFAULT;
2653
2654 sigset_from_compat(&new_set, &new32);
2655 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2656
2657 error = sigprocmask(how, &new_set, NULL);
2658 if (error)
2659 return error;
2660 }
2661 if (oset) {
2662 compat_sigset_t old32;
2663 sigset_to_compat(&old32, &old_set);
2664 if (copy_to_user(oset, &old32, sizeof(compat_sigset_t)))
2665 return -EFAULT;
2666 }
2667 return 0;
2668#else
2669 return sys_rt_sigprocmask(how, (sigset_t __user *)nset,
2670 (sigset_t __user *)oset, sigsetsize);
2671#endif
2672}
2673#endif
2674
2675static int do_sigpending(void *set, unsigned long sigsetsize)
2676{
2677 if (sigsetsize > sizeof(sigset_t))
2678 return -EINVAL;
2679
2680 spin_lock_irq(¤t->sighand->siglock);
2681 sigorsets(set, ¤t->pending.signal,
2682 ¤t->signal->shared_pending.signal);
2683 spin_unlock_irq(¤t->sighand->siglock);
2684
2685 /* Outside the lock because only this thread touches it. */
2686 sigandsets(set, ¤t->blocked, set);
2687 return 0;
2688}
2689
2690/**
2691 * sys_rt_sigpending - examine a pending signal that has been raised
2692 * while blocked
2693 * @uset: stores pending signals
2694 * @sigsetsize: size of sigset_t type or larger
2695 */
2696SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
2697{
2698 sigset_t set;
2699 int err = do_sigpending(&set, sigsetsize);
2700 if (!err && copy_to_user(uset, &set, sigsetsize))
2701 err = -EFAULT;
2702 return err;
2703}
2704
2705#ifdef CONFIG_COMPAT
2706COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
2707 compat_size_t, sigsetsize)
2708{
2709#ifdef __BIG_ENDIAN
2710 sigset_t set;
2711 int err = do_sigpending(&set, sigsetsize);
2712 if (!err) {
2713 compat_sigset_t set32;
2714 sigset_to_compat(&set32, &set);
2715 /* we can get here only if sigsetsize <= sizeof(set) */
2716 if (copy_to_user(uset, &set32, sigsetsize))
2717 err = -EFAULT;
2718 }
2719 return err;
2720#else
2721 return sys_rt_sigpending((sigset_t __user *)uset, sigsetsize);
2722#endif
2723}
2724#endif
2725
2726#ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2727
2728int copy_siginfo_to_user(siginfo_t __user *to, const siginfo_t *from)
2729{
2730 int err;
2731
2732 if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2733 return -EFAULT;
2734 if (from->si_code < 0)
2735 return __copy_to_user(to, from, sizeof(siginfo_t))
2736 ? -EFAULT : 0;
2737 /*
2738 * If you change siginfo_t structure, please be sure
2739 * this code is fixed accordingly.
2740 * Please remember to update the signalfd_copyinfo() function
2741 * inside fs/signalfd.c too, in case siginfo_t changes.
2742 * It should never copy any pad contained in the structure
2743 * to avoid security leaks, but must copy the generic
2744 * 3 ints plus the relevant union member.
2745 */
2746 err = __put_user(from->si_signo, &to->si_signo);
2747 err |= __put_user(from->si_errno, &to->si_errno);
2748 err |= __put_user((short)from->si_code, &to->si_code);
2749 switch (from->si_code & __SI_MASK) {
2750 case __SI_KILL:
2751 err |= __put_user(from->si_pid, &to->si_pid);
2752 err |= __put_user(from->si_uid, &to->si_uid);
2753 break;
2754 case __SI_TIMER:
2755 err |= __put_user(from->si_tid, &to->si_tid);
2756 err |= __put_user(from->si_overrun, &to->si_overrun);
2757 err |= __put_user(from->si_ptr, &to->si_ptr);
2758 break;
2759 case __SI_POLL:
2760 err |= __put_user(from->si_band, &to->si_band);
2761 err |= __put_user(from->si_fd, &to->si_fd);
2762 break;
2763 case __SI_FAULT:
2764 err |= __put_user(from->si_addr, &to->si_addr);
2765#ifdef __ARCH_SI_TRAPNO
2766 err |= __put_user(from->si_trapno, &to->si_trapno);
2767#endif
2768#ifdef BUS_MCEERR_AO
2769 /*
2770 * Other callers might not initialize the si_lsb field,
2771 * so check explicitly for the right codes here.
2772 */
2773 if (from->si_code == BUS_MCEERR_AR || from->si_code == BUS_MCEERR_AO)
2774 err |= __put_user(from->si_addr_lsb, &to->si_addr_lsb);
2775#endif
2776 break;
2777 case __SI_CHLD:
2778 err |= __put_user(from->si_pid, &to->si_pid);
2779 err |= __put_user(from->si_uid, &to->si_uid);
2780 err |= __put_user(from->si_status, &to->si_status);
2781 err |= __put_user(from->si_utime, &to->si_utime);
2782 err |= __put_user(from->si_stime, &to->si_stime);
2783 break;
2784 case __SI_RT: /* This is not generated by the kernel as of now. */
2785 case __SI_MESGQ: /* But this is */
2786 err |= __put_user(from->si_pid, &to->si_pid);
2787 err |= __put_user(from->si_uid, &to->si_uid);
2788 err |= __put_user(from->si_ptr, &to->si_ptr);
2789 break;
2790#ifdef __ARCH_SIGSYS
2791 case __SI_SYS:
2792 err |= __put_user(from->si_call_addr, &to->si_call_addr);
2793 err |= __put_user(from->si_syscall, &to->si_syscall);
2794 err |= __put_user(from->si_arch, &to->si_arch);
2795 break;
2796#endif
2797 default: /* this is just in case for now ... */
2798 err |= __put_user(from->si_pid, &to->si_pid);
2799 err |= __put_user(from->si_uid, &to->si_uid);
2800 break;
2801 }
2802 return err;
2803}
2804
2805#endif
2806
2807/**
2808 * do_sigtimedwait - wait for queued signals specified in @which
2809 * @which: queued signals to wait for
2810 * @info: if non-null, the signal's siginfo is returned here
2811 * @ts: upper bound on process time suspension
2812 */
2813int do_sigtimedwait(const sigset_t *which, siginfo_t *info,
2814 const struct timespec *ts)
2815{
2816 struct task_struct *tsk = current;
2817 long timeout = MAX_SCHEDULE_TIMEOUT;
2818 sigset_t mask = *which;
2819 int sig;
2820
2821 if (ts) {
2822 if (!timespec_valid(ts))
2823 return -EINVAL;
2824 timeout = timespec_to_jiffies(ts);
2825 /*
2826 * We can be close to the next tick, add another one
2827 * to ensure we will wait at least the time asked for.
2828 */
2829 if (ts->tv_sec || ts->tv_nsec)
2830 timeout++;
2831 }
2832
2833 /*
2834 * Invert the set of allowed signals to get those we want to block.
2835 */
2836 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
2837 signotset(&mask);
2838
2839 spin_lock_irq(&tsk->sighand->siglock);
2840 sig = dequeue_signal(tsk, &mask, info);
2841 if (!sig && timeout) {
2842 /*
2843 * None ready, temporarily unblock those we're interested
2844 * while we are sleeping in so that we'll be awakened when
2845 * they arrive. Unblocking is always fine, we can avoid
2846 * set_current_blocked().
2847 */
2848 tsk->real_blocked = tsk->blocked;
2849 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
2850 recalc_sigpending();
2851 spin_unlock_irq(&tsk->sighand->siglock);
2852
2853 timeout = freezable_schedule_timeout_interruptible(timeout);
2854
2855 spin_lock_irq(&tsk->sighand->siglock);
2856 __set_task_blocked(tsk, &tsk->real_blocked);
2857 siginitset(&tsk->real_blocked, 0);
2858 sig = dequeue_signal(tsk, &mask, info);
2859 }
2860 spin_unlock_irq(&tsk->sighand->siglock);
2861
2862 if (sig)
2863 return sig;
2864 return timeout ? -EINTR : -EAGAIN;
2865}
2866
2867/**
2868 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
2869 * in @uthese
2870 * @uthese: queued signals to wait for
2871 * @uinfo: if non-null, the signal's siginfo is returned here
2872 * @uts: upper bound on process time suspension
2873 * @sigsetsize: size of sigset_t type
2874 */
2875SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2876 siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2877 size_t, sigsetsize)
2878{
2879 sigset_t these;
2880 struct timespec ts;
2881 siginfo_t info;
2882 int ret;
2883
2884 /* XXX: Don't preclude handling different sized sigset_t's. */
2885 if (sigsetsize != sizeof(sigset_t))
2886 return -EINVAL;
2887
2888 if (copy_from_user(&these, uthese, sizeof(these)))
2889 return -EFAULT;
2890
2891 if (uts) {
2892 if (copy_from_user(&ts, uts, sizeof(ts)))
2893 return -EFAULT;
2894 }
2895
2896 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
2897
2898 if (ret > 0 && uinfo) {
2899 if (copy_siginfo_to_user(uinfo, &info))
2900 ret = -EFAULT;
2901 }
2902
2903 return ret;
2904}
2905
2906/**
2907 * sys_kill - send a signal to a process
2908 * @pid: the PID of the process
2909 * @sig: signal to be sent
2910 */
2911SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2912{
2913 struct siginfo info;
2914
2915 info.si_signo = sig;
2916 info.si_errno = 0;
2917 info.si_code = SI_USER;
2918 info.si_pid = task_tgid_vnr(current);
2919 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2920
2921 return kill_something_info(sig, &info, pid);
2922}
2923
2924static int
2925do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
2926{
2927 struct task_struct *p;
2928 int error = -ESRCH;
2929
2930 rcu_read_lock();
2931 p = find_task_by_vpid(pid);
2932 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2933 error = check_kill_permission(sig, info, p);
2934 /*
2935 * The null signal is a permissions and process existence
2936 * probe. No signal is actually delivered.
2937 */
2938 if (!error && sig) {
2939 error = do_send_sig_info(sig, info, p, false);
2940 /*
2941 * If lock_task_sighand() failed we pretend the task
2942 * dies after receiving the signal. The window is tiny,
2943 * and the signal is private anyway.
2944 */
2945 if (unlikely(error == -ESRCH))
2946 error = 0;
2947 }
2948 }
2949 rcu_read_unlock();
2950
2951 return error;
2952}
2953
2954static int do_tkill(pid_t tgid, pid_t pid, int sig)
2955{
2956 struct siginfo info = {};
2957
2958 info.si_signo = sig;
2959 info.si_errno = 0;
2960 info.si_code = SI_TKILL;
2961 info.si_pid = task_tgid_vnr(current);
2962 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2963
2964 return do_send_specific(tgid, pid, sig, &info);
2965}
2966
2967/**
2968 * sys_tgkill - send signal to one specific thread
2969 * @tgid: the thread group ID of the thread
2970 * @pid: the PID of the thread
2971 * @sig: signal to be sent
2972 *
2973 * This syscall also checks the @tgid and returns -ESRCH even if the PID
2974 * exists but it's not belonging to the target process anymore. This
2975 * method solves the problem of threads exiting and PIDs getting reused.
2976 */
2977SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2978{
2979 /* This is only valid for single tasks */
2980 if (pid <= 0 || tgid <= 0)
2981 return -EINVAL;
2982
2983 return do_tkill(tgid, pid, sig);
2984}
2985
2986/**
2987 * sys_tkill - send signal to one specific task
2988 * @pid: the PID of the task
2989 * @sig: signal to be sent
2990 *
2991 * Send a signal to only one task, even if it's a CLONE_THREAD task.
2992 */
2993SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2994{
2995 /* This is only valid for single tasks */
2996 if (pid <= 0)
2997 return -EINVAL;
2998
2999 return do_tkill(0, pid, sig);
3000}
3001
3002static int do_rt_sigqueueinfo(pid_t pid, int sig, siginfo_t *info)
3003{
3004 /* Not even root can pretend to send signals from the kernel.
3005 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3006 */
3007 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3008 (task_pid_vnr(current) != pid)) {
3009 /* We used to allow any < 0 si_code */
3010 WARN_ON_ONCE(info->si_code < 0);
3011 return -EPERM;
3012 }
3013 info->si_signo = sig;
3014
3015 /* POSIX.1b doesn't mention process groups. */
3016 return kill_proc_info(sig, info, pid);
3017}
3018
3019/**
3020 * sys_rt_sigqueueinfo - send signal information to a signal
3021 * @pid: the PID of the thread
3022 * @sig: signal to be sent
3023 * @uinfo: signal info to be sent
3024 */
3025SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3026 siginfo_t __user *, uinfo)
3027{
3028 siginfo_t info;
3029 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
3030 return -EFAULT;
3031 return do_rt_sigqueueinfo(pid, sig, &info);
3032}
3033
3034#ifdef CONFIG_COMPAT
3035COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3036 compat_pid_t, pid,
3037 int, sig,
3038 struct compat_siginfo __user *, uinfo)
3039{
3040 siginfo_t info;
3041 int ret = copy_siginfo_from_user32(&info, uinfo);
3042 if (unlikely(ret))
3043 return ret;
3044 return do_rt_sigqueueinfo(pid, sig, &info);
3045}
3046#endif
3047
3048static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
3049{
3050 /* This is only valid for single tasks */
3051 if (pid <= 0 || tgid <= 0)
3052 return -EINVAL;
3053
3054 /* Not even root can pretend to send signals from the kernel.
3055 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3056 */
3057 if (((info->si_code >= 0 || info->si_code == SI_TKILL)) &&
3058 (task_pid_vnr(current) != pid)) {
3059 /* We used to allow any < 0 si_code */
3060 WARN_ON_ONCE(info->si_code < 0);
3061 return -EPERM;
3062 }
3063 info->si_signo = sig;
3064
3065 return do_send_specific(tgid, pid, sig, info);
3066}
3067
3068SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3069 siginfo_t __user *, uinfo)
3070{
3071 siginfo_t info;
3072
3073 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
3074 return -EFAULT;
3075
3076 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3077}
3078
3079#ifdef CONFIG_COMPAT
3080COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3081 compat_pid_t, tgid,
3082 compat_pid_t, pid,
3083 int, sig,
3084 struct compat_siginfo __user *, uinfo)
3085{
3086 siginfo_t info;
3087
3088 if (copy_siginfo_from_user32(&info, uinfo))
3089 return -EFAULT;
3090 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3091}
3092#endif
3093
3094int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
3095{
3096 struct task_struct *t = current;
3097 struct k_sigaction *k;
3098 sigset_t mask;
3099
3100 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
3101 return -EINVAL;
3102
3103 k = &t->sighand->action[sig-1];
3104
3105 spin_lock_irq(¤t->sighand->siglock);
3106 if (oact)
3107 *oact = *k;
3108
3109 if (act) {
3110 sigdelsetmask(&act->sa.sa_mask,
3111 sigmask(SIGKILL) | sigmask(SIGSTOP));
3112 *k = *act;
3113 /*
3114 * POSIX 3.3.1.3:
3115 * "Setting a signal action to SIG_IGN for a signal that is
3116 * pending shall cause the pending signal to be discarded,
3117 * whether or not it is blocked."
3118 *
3119 * "Setting a signal action to SIG_DFL for a signal that is
3120 * pending and whose default action is to ignore the signal
3121 * (for example, SIGCHLD), shall cause the pending signal to
3122 * be discarded, whether or not it is blocked"
3123 */
3124 if (sig_handler_ignored(sig_handler(t, sig), sig)) {
3125 sigemptyset(&mask);
3126 sigaddset(&mask, sig);
3127 rm_from_queue_full(&mask, &t->signal->shared_pending);
3128 do {
3129 rm_from_queue_full(&mask, &t->pending);
3130 } while_each_thread(current, t);
3131 }
3132 }
3133
3134 spin_unlock_irq(¤t->sighand->siglock);
3135 return 0;
3136}
3137
3138static int
3139do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
3140{
3141 stack_t oss;
3142 int error;
3143
3144 oss.ss_sp = (void __user *) current->sas_ss_sp;
3145 oss.ss_size = current->sas_ss_size;
3146 oss.ss_flags = sas_ss_flags(sp);
3147
3148 if (uss) {
3149 void __user *ss_sp;
3150 size_t ss_size;
3151 int ss_flags;
3152
3153 error = -EFAULT;
3154 if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
3155 goto out;
3156 error = __get_user(ss_sp, &uss->ss_sp) |
3157 __get_user(ss_flags, &uss->ss_flags) |
3158 __get_user(ss_size, &uss->ss_size);
3159 if (error)
3160 goto out;
3161
3162 error = -EPERM;
3163 if (on_sig_stack(sp))
3164 goto out;
3165
3166 error = -EINVAL;
3167 /*
3168 * Note - this code used to test ss_flags incorrectly:
3169 * old code may have been written using ss_flags==0
3170 * to mean ss_flags==SS_ONSTACK (as this was the only
3171 * way that worked) - this fix preserves that older
3172 * mechanism.
3173 */
3174 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
3175 goto out;
3176
3177 if (ss_flags == SS_DISABLE) {
3178 ss_size = 0;
3179 ss_sp = NULL;
3180 } else {
3181 error = -ENOMEM;
3182 if (ss_size < MINSIGSTKSZ)
3183 goto out;
3184 }
3185
3186 current->sas_ss_sp = (unsigned long) ss_sp;
3187 current->sas_ss_size = ss_size;
3188 }
3189
3190 error = 0;
3191 if (uoss) {
3192 error = -EFAULT;
3193 if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
3194 goto out;
3195 error = __put_user(oss.ss_sp, &uoss->ss_sp) |
3196 __put_user(oss.ss_size, &uoss->ss_size) |
3197 __put_user(oss.ss_flags, &uoss->ss_flags);
3198 }
3199
3200out:
3201 return error;
3202}
3203SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
3204{
3205 return do_sigaltstack(uss, uoss, current_user_stack_pointer());
3206}
3207
3208int restore_altstack(const stack_t __user *uss)
3209{
3210 int err = do_sigaltstack(uss, NULL, current_user_stack_pointer());
3211 /* squash all but EFAULT for now */
3212 return err == -EFAULT ? err : 0;
3213}
3214
3215int __save_altstack(stack_t __user *uss, unsigned long sp)
3216{
3217 struct task_struct *t = current;
3218 return __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
3219 __put_user(sas_ss_flags(sp), &uss->ss_flags) |
3220 __put_user(t->sas_ss_size, &uss->ss_size);
3221}
3222
3223#ifdef CONFIG_COMPAT
3224COMPAT_SYSCALL_DEFINE2(sigaltstack,
3225 const compat_stack_t __user *, uss_ptr,
3226 compat_stack_t __user *, uoss_ptr)
3227{
3228 stack_t uss, uoss;
3229 int ret;
3230 mm_segment_t seg;
3231
3232 if (uss_ptr) {
3233 compat_stack_t uss32;
3234
3235 memset(&uss, 0, sizeof(stack_t));
3236 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
3237 return -EFAULT;
3238 uss.ss_sp = compat_ptr(uss32.ss_sp);
3239 uss.ss_flags = uss32.ss_flags;
3240 uss.ss_size = uss32.ss_size;
3241 }
3242 seg = get_fs();
3243 set_fs(KERNEL_DS);
3244 ret = do_sigaltstack((stack_t __force __user *) (uss_ptr ? &uss : NULL),
3245 (stack_t __force __user *) &uoss,
3246 compat_user_stack_pointer());
3247 set_fs(seg);
3248 if (ret >= 0 && uoss_ptr) {
3249 if (!access_ok(VERIFY_WRITE, uoss_ptr, sizeof(compat_stack_t)) ||
3250 __put_user(ptr_to_compat(uoss.ss_sp), &uoss_ptr->ss_sp) ||
3251 __put_user(uoss.ss_flags, &uoss_ptr->ss_flags) ||
3252 __put_user(uoss.ss_size, &uoss_ptr->ss_size))
3253 ret = -EFAULT;
3254 }
3255 return ret;
3256}
3257
3258int compat_restore_altstack(const compat_stack_t __user *uss)
3259{
3260 int err = compat_sys_sigaltstack(uss, NULL);
3261 /* squash all but -EFAULT for now */
3262 return err == -EFAULT ? err : 0;
3263}
3264
3265int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
3266{
3267 struct task_struct *t = current;
3268 return __put_user(ptr_to_compat((void __user *)t->sas_ss_sp), &uss->ss_sp) |
3269 __put_user(sas_ss_flags(sp), &uss->ss_flags) |
3270 __put_user(t->sas_ss_size, &uss->ss_size);
3271}
3272#endif
3273
3274#ifdef __ARCH_WANT_SYS_SIGPENDING
3275
3276/**
3277 * sys_sigpending - examine pending signals
3278 * @set: where mask of pending signal is returned
3279 */
3280SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
3281{
3282 return sys_rt_sigpending((sigset_t __user *)set, sizeof(old_sigset_t));
3283}
3284
3285#endif
3286
3287#ifdef __ARCH_WANT_SYS_SIGPROCMASK
3288/**
3289 * sys_sigprocmask - examine and change blocked signals
3290 * @how: whether to add, remove, or set signals
3291 * @nset: signals to add or remove (if non-null)
3292 * @oset: previous value of signal mask if non-null
3293 *
3294 * Some platforms have their own version with special arguments;
3295 * others support only sys_rt_sigprocmask.
3296 */
3297
3298SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
3299 old_sigset_t __user *, oset)
3300{
3301 old_sigset_t old_set, new_set;
3302 sigset_t new_blocked;
3303
3304 old_set = current->blocked.sig[0];
3305
3306 if (nset) {
3307 if (copy_from_user(&new_set, nset, sizeof(*nset)))
3308 return -EFAULT;
3309
3310 new_blocked = current->blocked;
3311
3312 switch (how) {
3313 case SIG_BLOCK:
3314 sigaddsetmask(&new_blocked, new_set);
3315 break;
3316 case SIG_UNBLOCK:
3317 sigdelsetmask(&new_blocked, new_set);
3318 break;
3319 case SIG_SETMASK:
3320 new_blocked.sig[0] = new_set;
3321 break;
3322 default:
3323 return -EINVAL;
3324 }
3325
3326 set_current_blocked(&new_blocked);
3327 }
3328
3329 if (oset) {
3330 if (copy_to_user(oset, &old_set, sizeof(*oset)))
3331 return -EFAULT;
3332 }
3333
3334 return 0;
3335}
3336#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
3337
3338#ifndef CONFIG_ODD_RT_SIGACTION
3339/**
3340 * sys_rt_sigaction - alter an action taken by a process
3341 * @sig: signal to be sent
3342 * @act: new sigaction
3343 * @oact: used to save the previous sigaction
3344 * @sigsetsize: size of sigset_t type
3345 */
3346SYSCALL_DEFINE4(rt_sigaction, int, sig,
3347 const struct sigaction __user *, act,
3348 struct sigaction __user *, oact,
3349 size_t, sigsetsize)
3350{
3351 struct k_sigaction new_sa, old_sa;
3352 int ret = -EINVAL;
3353
3354 /* XXX: Don't preclude handling different sized sigset_t's. */
3355 if (sigsetsize != sizeof(sigset_t))
3356 goto out;
3357
3358 if (act) {
3359 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
3360 return -EFAULT;
3361 }
3362
3363 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
3364
3365 if (!ret && oact) {
3366 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
3367 return -EFAULT;
3368 }
3369out:
3370 return ret;
3371}
3372#ifdef CONFIG_COMPAT
3373COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
3374 const struct compat_sigaction __user *, act,
3375 struct compat_sigaction __user *, oact,
3376 compat_size_t, sigsetsize)
3377{
3378 struct k_sigaction new_ka, old_ka;
3379 compat_sigset_t mask;
3380#ifdef __ARCH_HAS_SA_RESTORER
3381 compat_uptr_t restorer;
3382#endif
3383 int ret;
3384
3385 /* XXX: Don't preclude handling different sized sigset_t's. */
3386 if (sigsetsize != sizeof(compat_sigset_t))
3387 return -EINVAL;
3388
3389 if (act) {
3390 compat_uptr_t handler;
3391 ret = get_user(handler, &act->sa_handler);
3392 new_ka.sa.sa_handler = compat_ptr(handler);
3393#ifdef __ARCH_HAS_SA_RESTORER
3394 ret |= get_user(restorer, &act->sa_restorer);
3395 new_ka.sa.sa_restorer = compat_ptr(restorer);
3396#endif
3397 ret |= copy_from_user(&mask, &act->sa_mask, sizeof(mask));
3398 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
3399 if (ret)
3400 return -EFAULT;
3401 sigset_from_compat(&new_ka.sa.sa_mask, &mask);
3402 }
3403
3404 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
3405 if (!ret && oact) {
3406 sigset_to_compat(&mask, &old_ka.sa.sa_mask);
3407 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
3408 &oact->sa_handler);
3409 ret |= copy_to_user(&oact->sa_mask, &mask, sizeof(mask));
3410 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
3411#ifdef __ARCH_HAS_SA_RESTORER
3412 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
3413 &oact->sa_restorer);
3414#endif
3415 }
3416 return ret;
3417}
3418#endif
3419#endif /* !CONFIG_ODD_RT_SIGACTION */
3420
3421#ifdef CONFIG_OLD_SIGACTION
3422SYSCALL_DEFINE3(sigaction, int, sig,
3423 const struct old_sigaction __user *, act,
3424 struct old_sigaction __user *, oact)
3425{
3426 struct k_sigaction new_ka, old_ka;
3427 int ret;
3428
3429 if (act) {
3430 old_sigset_t mask;
3431 if (!access_ok(VERIFY_READ, act, sizeof(*act)) ||
3432 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
3433 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
3434 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
3435 __get_user(mask, &act->sa_mask))
3436 return -EFAULT;
3437#ifdef __ARCH_HAS_KA_RESTORER
3438 new_ka.ka_restorer = NULL;
3439#endif
3440 siginitset(&new_ka.sa.sa_mask, mask);
3441 }
3442
3443 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
3444
3445 if (!ret && oact) {
3446 if (!access_ok(VERIFY_WRITE, oact, sizeof(*oact)) ||
3447 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
3448 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
3449 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
3450 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
3451 return -EFAULT;
3452 }
3453
3454 return ret;
3455}
3456#endif
3457#ifdef CONFIG_COMPAT_OLD_SIGACTION
3458COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
3459 const struct compat_old_sigaction __user *, act,
3460 struct compat_old_sigaction __user *, oact)
3461{
3462 struct k_sigaction new_ka, old_ka;
3463 int ret;
3464 compat_old_sigset_t mask;
3465 compat_uptr_t handler, restorer;
3466
3467 if (act) {
3468 if (!access_ok(VERIFY_READ, act, sizeof(*act)) ||
3469 __get_user(handler, &act->sa_handler) ||
3470 __get_user(restorer, &act->sa_restorer) ||
3471 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
3472 __get_user(mask, &act->sa_mask))
3473 return -EFAULT;
3474
3475#ifdef __ARCH_HAS_KA_RESTORER
3476 new_ka.ka_restorer = NULL;
3477#endif
3478 new_ka.sa.sa_handler = compat_ptr(handler);
3479 new_ka.sa.sa_restorer = compat_ptr(restorer);
3480 siginitset(&new_ka.sa.sa_mask, mask);
3481 }
3482
3483 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
3484
3485 if (!ret && oact) {
3486 if (!access_ok(VERIFY_WRITE, oact, sizeof(*oact)) ||
3487 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
3488 &oact->sa_handler) ||
3489 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
3490 &oact->sa_restorer) ||
3491 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
3492 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
3493 return -EFAULT;
3494 }
3495 return ret;
3496}
3497#endif
3498
3499#ifdef __ARCH_WANT_SYS_SGETMASK
3500
3501/*
3502 * For backwards compatibility. Functionality superseded by sigprocmask.
3503 */
3504SYSCALL_DEFINE0(sgetmask)
3505{
3506 /* SMP safe */
3507 return current->blocked.sig[0];
3508}
3509
3510SYSCALL_DEFINE1(ssetmask, int, newmask)
3511{
3512 int old = current->blocked.sig[0];
3513 sigset_t newset;
3514
3515 siginitset(&newset, newmask);
3516 set_current_blocked(&newset);
3517
3518 return old;
3519}
3520#endif /* __ARCH_WANT_SGETMASK */
3521
3522#ifdef __ARCH_WANT_SYS_SIGNAL
3523/*
3524 * For backwards compatibility. Functionality superseded by sigaction.
3525 */
3526SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
3527{
3528 struct k_sigaction new_sa, old_sa;
3529 int ret;
3530
3531 new_sa.sa.sa_handler = handler;
3532 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
3533 sigemptyset(&new_sa.sa.sa_mask);
3534
3535 ret = do_sigaction(sig, &new_sa, &old_sa);
3536
3537 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
3538}
3539#endif /* __ARCH_WANT_SYS_SIGNAL */
3540
3541#ifdef __ARCH_WANT_SYS_PAUSE
3542
3543SYSCALL_DEFINE0(pause)
3544{
3545 while (!signal_pending(current)) {
3546 current->state = TASK_INTERRUPTIBLE;
3547 schedule();
3548 }
3549 return -ERESTARTNOHAND;
3550}
3551
3552#endif
3553
3554int sigsuspend(sigset_t *set)
3555{
3556 current->saved_sigmask = current->blocked;
3557 set_current_blocked(set);
3558
3559 current->state = TASK_INTERRUPTIBLE;
3560 schedule();
3561 set_restore_sigmask();
3562 return -ERESTARTNOHAND;
3563}
3564
3565/**
3566 * sys_rt_sigsuspend - replace the signal mask for a value with the
3567 * @unewset value until a signal is received
3568 * @unewset: new signal mask value
3569 * @sigsetsize: size of sigset_t type
3570 */
3571SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
3572{
3573 sigset_t newset;
3574
3575 /* XXX: Don't preclude handling different sized sigset_t's. */
3576 if (sigsetsize != sizeof(sigset_t))
3577 return -EINVAL;
3578
3579 if (copy_from_user(&newset, unewset, sizeof(newset)))
3580 return -EFAULT;
3581 return sigsuspend(&newset);
3582}
3583
3584#ifdef CONFIG_COMPAT
3585COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
3586{
3587#ifdef __BIG_ENDIAN
3588 sigset_t newset;
3589 compat_sigset_t newset32;
3590
3591 /* XXX: Don't preclude handling different sized sigset_t's. */
3592 if (sigsetsize != sizeof(sigset_t))
3593 return -EINVAL;
3594
3595 if (copy_from_user(&newset32, unewset, sizeof(compat_sigset_t)))
3596 return -EFAULT;
3597 sigset_from_compat(&newset, &newset32);
3598 return sigsuspend(&newset);
3599#else
3600 /* on little-endian bitmaps don't care about granularity */
3601 return sys_rt_sigsuspend((sigset_t __user *)unewset, sigsetsize);
3602#endif
3603}
3604#endif
3605
3606#ifdef CONFIG_OLD_SIGSUSPEND
3607SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
3608{
3609 sigset_t blocked;
3610 siginitset(&blocked, mask);
3611 return sigsuspend(&blocked);
3612}
3613#endif
3614#ifdef CONFIG_OLD_SIGSUSPEND3
3615SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
3616{
3617 sigset_t blocked;
3618 siginitset(&blocked, mask);
3619 return sigsuspend(&blocked);
3620}
3621#endif
3622
3623__weak const char *arch_vma_name(struct vm_area_struct *vma)
3624{
3625 return NULL;
3626}
3627
3628void __init signals_init(void)
3629{
3630 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
3631}
3632
3633#ifdef CONFIG_KGDB_KDB
3634#include <linux/kdb.h>
3635/*
3636 * kdb_send_sig_info - Allows kdb to send signals without exposing
3637 * signal internals. This function checks if the required locks are
3638 * available before calling the main signal code, to avoid kdb
3639 * deadlocks.
3640 */
3641void
3642kdb_send_sig_info(struct task_struct *t, struct siginfo *info)
3643{
3644 static struct task_struct *kdb_prev_t;
3645 int sig, new_t;
3646 if (!spin_trylock(&t->sighand->siglock)) {
3647 kdb_printf("Can't do kill command now.\n"
3648 "The sigmask lock is held somewhere else in "
3649 "kernel, try again later\n");
3650 return;
3651 }
3652 spin_unlock(&t->sighand->siglock);
3653 new_t = kdb_prev_t != t;
3654 kdb_prev_t = t;
3655 if (t->state != TASK_RUNNING && new_t) {
3656 kdb_printf("Process is not RUNNING, sending a signal from "
3657 "kdb risks deadlock\n"
3658 "on the run queue locks. "
3659 "The signal has _not_ been sent.\n"
3660 "Reissue the kill command if you want to risk "
3661 "the deadlock.\n");
3662 return;
3663 }
3664 sig = info->si_signo;
3665 if (send_sig_info(sig, info, t))
3666 kdb_printf("Fail to deliver Signal %d to process %d.\n",
3667 sig, t->pid);
3668 else
3669 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
3670}
3671#endif /* CONFIG_KGDB_KDB */