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