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