]> git.proxmox.com Git - mirror_ubuntu-impish-kernel.git/blob - kernel/signal.c
UBUNTU: [Config] update configs and annotations after rebase to 5.13
[mirror_ubuntu-impish-kernel.git] / kernel / signal.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/kernel/signal.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 *
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
12 */
13
14 #include <linux/slab.h>
15 #include <linux/export.h>
16 #include <linux/init.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/sched/cputime.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/proc_fs.h>
26 #include <linux/tty.h>
27 #include <linux/binfmts.h>
28 #include <linux/coredump.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/ptrace.h>
32 #include <linux/signal.h>
33 #include <linux/signalfd.h>
34 #include <linux/ratelimit.h>
35 #include <linux/tracehook.h>
36 #include <linux/capability.h>
37 #include <linux/freezer.h>
38 #include <linux/pid_namespace.h>
39 #include <linux/nsproxy.h>
40 #include <linux/user_namespace.h>
41 #include <linux/uprobes.h>
42 #include <linux/compat.h>
43 #include <linux/cn_proc.h>
44 #include <linux/compiler.h>
45 #include <linux/posix-timers.h>
46 #include <linux/cgroup.h>
47 #include <linux/audit.h>
48
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/signal.h>
51
52 #include <asm/param.h>
53 #include <linux/uaccess.h>
54 #include <asm/unistd.h>
55 #include <asm/siginfo.h>
56 #include <asm/cacheflush.h>
57
58 /*
59 * SLAB caches for signal bits.
60 */
61
62 static struct kmem_cache *sigqueue_cachep;
63
64 int print_fatal_signals __read_mostly;
65
66 static void __user *sig_handler(struct task_struct *t, int sig)
67 {
68 return t->sighand->action[sig - 1].sa.sa_handler;
69 }
70
71 static inline bool sig_handler_ignored(void __user *handler, int sig)
72 {
73 /* Is it explicitly or implicitly ignored? */
74 return handler == SIG_IGN ||
75 (handler == SIG_DFL && sig_kernel_ignore(sig));
76 }
77
78 static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
79 {
80 void __user *handler;
81
82 handler = sig_handler(t, sig);
83
84 /* SIGKILL and SIGSTOP may not be sent to the global init */
85 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
86 return true;
87
88 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
89 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
90 return true;
91
92 /* Only allow kernel generated signals to this kthread */
93 if (unlikely((t->flags & PF_KTHREAD) &&
94 (handler == SIG_KTHREAD_KERNEL) && !force))
95 return true;
96
97 return sig_handler_ignored(handler, sig);
98 }
99
100 static bool sig_ignored(struct task_struct *t, int sig, bool force)
101 {
102 /*
103 * Blocked signals are never ignored, since the
104 * signal handler may change by the time it is
105 * unblocked.
106 */
107 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
108 return false;
109
110 /*
111 * Tracers may want to know about even ignored signal unless it
112 * is SIGKILL which can't be reported anyway but can be ignored
113 * by SIGNAL_UNKILLABLE task.
114 */
115 if (t->ptrace && sig != SIGKILL)
116 return false;
117
118 return sig_task_ignored(t, sig, force);
119 }
120
121 /*
122 * Re-calculate pending state from the set of locally pending
123 * signals, globally pending signals, and blocked signals.
124 */
125 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
126 {
127 unsigned long ready;
128 long i;
129
130 switch (_NSIG_WORDS) {
131 default:
132 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
133 ready |= signal->sig[i] &~ blocked->sig[i];
134 break;
135
136 case 4: ready = signal->sig[3] &~ blocked->sig[3];
137 ready |= signal->sig[2] &~ blocked->sig[2];
138 ready |= signal->sig[1] &~ blocked->sig[1];
139 ready |= signal->sig[0] &~ blocked->sig[0];
140 break;
141
142 case 2: ready = signal->sig[1] &~ blocked->sig[1];
143 ready |= signal->sig[0] &~ blocked->sig[0];
144 break;
145
146 case 1: ready = signal->sig[0] &~ blocked->sig[0];
147 }
148 return ready != 0;
149 }
150
151 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
152
153 static bool recalc_sigpending_tsk(struct task_struct *t)
154 {
155 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
156 PENDING(&t->pending, &t->blocked) ||
157 PENDING(&t->signal->shared_pending, &t->blocked) ||
158 cgroup_task_frozen(t)) {
159 set_tsk_thread_flag(t, TIF_SIGPENDING);
160 return true;
161 }
162
163 /*
164 * We must never clear the flag in another thread, or in current
165 * when it's possible the current syscall is returning -ERESTART*.
166 * So we don't clear it here, and only callers who know they should do.
167 */
168 return false;
169 }
170
171 /*
172 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
173 * This is superfluous when called on current, the wakeup is a harmless no-op.
174 */
175 void recalc_sigpending_and_wake(struct task_struct *t)
176 {
177 if (recalc_sigpending_tsk(t))
178 signal_wake_up(t, 0);
179 }
180
181 void recalc_sigpending(void)
182 {
183 if (!recalc_sigpending_tsk(current) && !freezing(current))
184 clear_thread_flag(TIF_SIGPENDING);
185
186 }
187 EXPORT_SYMBOL(recalc_sigpending);
188
189 void calculate_sigpending(void)
190 {
191 /* Have any signals or users of TIF_SIGPENDING been delayed
192 * until after fork?
193 */
194 spin_lock_irq(&current->sighand->siglock);
195 set_tsk_thread_flag(current, TIF_SIGPENDING);
196 recalc_sigpending();
197 spin_unlock_irq(&current->sighand->siglock);
198 }
199
200 /* Given the mask, find the first available signal that should be serviced. */
201
202 #define SYNCHRONOUS_MASK \
203 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
204 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
205
206 int next_signal(struct sigpending *pending, sigset_t *mask)
207 {
208 unsigned long i, *s, *m, x;
209 int sig = 0;
210
211 s = pending->signal.sig;
212 m = mask->sig;
213
214 /*
215 * Handle the first word specially: it contains the
216 * synchronous signals that need to be dequeued first.
217 */
218 x = *s &~ *m;
219 if (x) {
220 if (x & SYNCHRONOUS_MASK)
221 x &= SYNCHRONOUS_MASK;
222 sig = ffz(~x) + 1;
223 return sig;
224 }
225
226 switch (_NSIG_WORDS) {
227 default:
228 for (i = 1; i < _NSIG_WORDS; ++i) {
229 x = *++s &~ *++m;
230 if (!x)
231 continue;
232 sig = ffz(~x) + i*_NSIG_BPW + 1;
233 break;
234 }
235 break;
236
237 case 2:
238 x = s[1] &~ m[1];
239 if (!x)
240 break;
241 sig = ffz(~x) + _NSIG_BPW + 1;
242 break;
243
244 case 1:
245 /* Nothing to do */
246 break;
247 }
248
249 return sig;
250 }
251
252 static inline void print_dropped_signal(int sig)
253 {
254 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
255
256 if (!print_fatal_signals)
257 return;
258
259 if (!__ratelimit(&ratelimit_state))
260 return;
261
262 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
263 current->comm, current->pid, sig);
264 }
265
266 /**
267 * task_set_jobctl_pending - set jobctl pending bits
268 * @task: target task
269 * @mask: pending bits to set
270 *
271 * Clear @mask from @task->jobctl. @mask must be subset of
272 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
273 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
274 * cleared. If @task is already being killed or exiting, this function
275 * becomes noop.
276 *
277 * CONTEXT:
278 * Must be called with @task->sighand->siglock held.
279 *
280 * RETURNS:
281 * %true if @mask is set, %false if made noop because @task was dying.
282 */
283 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
284 {
285 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
286 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
287 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
288
289 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
290 return false;
291
292 if (mask & JOBCTL_STOP_SIGMASK)
293 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
294
295 task->jobctl |= mask;
296 return true;
297 }
298
299 /**
300 * task_clear_jobctl_trapping - clear jobctl trapping bit
301 * @task: target task
302 *
303 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
304 * Clear it and wake up the ptracer. Note that we don't need any further
305 * locking. @task->siglock guarantees that @task->parent points to the
306 * ptracer.
307 *
308 * CONTEXT:
309 * Must be called with @task->sighand->siglock held.
310 */
311 void task_clear_jobctl_trapping(struct task_struct *task)
312 {
313 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
314 task->jobctl &= ~JOBCTL_TRAPPING;
315 smp_mb(); /* advised by wake_up_bit() */
316 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
317 }
318 }
319
320 /**
321 * task_clear_jobctl_pending - clear jobctl pending bits
322 * @task: target task
323 * @mask: pending bits to clear
324 *
325 * Clear @mask from @task->jobctl. @mask must be subset of
326 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
327 * STOP bits are cleared together.
328 *
329 * If clearing of @mask leaves no stop or trap pending, this function calls
330 * task_clear_jobctl_trapping().
331 *
332 * CONTEXT:
333 * Must be called with @task->sighand->siglock held.
334 */
335 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
336 {
337 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
338
339 if (mask & JOBCTL_STOP_PENDING)
340 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
341
342 task->jobctl &= ~mask;
343
344 if (!(task->jobctl & JOBCTL_PENDING_MASK))
345 task_clear_jobctl_trapping(task);
346 }
347
348 /**
349 * task_participate_group_stop - participate in a group stop
350 * @task: task participating in a group stop
351 *
352 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
353 * Group stop states are cleared and the group stop count is consumed if
354 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
355 * stop, the appropriate `SIGNAL_*` flags are set.
356 *
357 * CONTEXT:
358 * Must be called with @task->sighand->siglock held.
359 *
360 * RETURNS:
361 * %true if group stop completion should be notified to the parent, %false
362 * otherwise.
363 */
364 static bool task_participate_group_stop(struct task_struct *task)
365 {
366 struct signal_struct *sig = task->signal;
367 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
368
369 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
370
371 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
372
373 if (!consume)
374 return false;
375
376 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
377 sig->group_stop_count--;
378
379 /*
380 * Tell the caller to notify completion iff we are entering into a
381 * fresh group stop. Read comment in do_signal_stop() for details.
382 */
383 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
384 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
385 return true;
386 }
387 return false;
388 }
389
390 void task_join_group_stop(struct task_struct *task)
391 {
392 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
393 struct signal_struct *sig = current->signal;
394
395 if (sig->group_stop_count) {
396 sig->group_stop_count++;
397 mask |= JOBCTL_STOP_CONSUME;
398 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
399 return;
400
401 /* Have the new thread join an on-going signal group stop */
402 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
403 }
404
405 /*
406 * allocate a new signal queue record
407 * - this may be called without locks if and only if t == current, otherwise an
408 * appropriate lock must be held to stop the target task from exiting
409 */
410 static struct sigqueue *
411 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
412 int override_rlimit, const unsigned int sigqueue_flags)
413 {
414 struct sigqueue *q = NULL;
415 struct user_struct *user;
416 int sigpending;
417
418 /*
419 * Protect access to @t credentials. This can go away when all
420 * callers hold rcu read lock.
421 *
422 * NOTE! A pending signal will hold on to the user refcount,
423 * and we get/put the refcount only when the sigpending count
424 * changes from/to zero.
425 */
426 rcu_read_lock();
427 user = __task_cred(t)->user;
428 sigpending = atomic_inc_return(&user->sigpending);
429 if (sigpending == 1)
430 get_uid(user);
431 rcu_read_unlock();
432
433 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
434 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
435 } else {
436 print_dropped_signal(sig);
437 }
438
439 if (unlikely(q == NULL)) {
440 if (atomic_dec_and_test(&user->sigpending))
441 free_uid(user);
442 } else {
443 INIT_LIST_HEAD(&q->list);
444 q->flags = sigqueue_flags;
445 q->user = user;
446 }
447
448 return q;
449 }
450
451 static void __sigqueue_free(struct sigqueue *q)
452 {
453 if (q->flags & SIGQUEUE_PREALLOC)
454 return;
455 if (atomic_dec_and_test(&q->user->sigpending))
456 free_uid(q->user);
457 kmem_cache_free(sigqueue_cachep, q);
458 }
459
460 void flush_sigqueue(struct sigpending *queue)
461 {
462 struct sigqueue *q;
463
464 sigemptyset(&queue->signal);
465 while (!list_empty(&queue->list)) {
466 q = list_entry(queue->list.next, struct sigqueue , list);
467 list_del_init(&q->list);
468 __sigqueue_free(q);
469 }
470 }
471
472 /*
473 * Flush all pending signals for this kthread.
474 */
475 void flush_signals(struct task_struct *t)
476 {
477 unsigned long flags;
478
479 spin_lock_irqsave(&t->sighand->siglock, flags);
480 clear_tsk_thread_flag(t, TIF_SIGPENDING);
481 flush_sigqueue(&t->pending);
482 flush_sigqueue(&t->signal->shared_pending);
483 spin_unlock_irqrestore(&t->sighand->siglock, flags);
484 }
485 EXPORT_SYMBOL(flush_signals);
486
487 #ifdef CONFIG_POSIX_TIMERS
488 static void __flush_itimer_signals(struct sigpending *pending)
489 {
490 sigset_t signal, retain;
491 struct sigqueue *q, *n;
492
493 signal = pending->signal;
494 sigemptyset(&retain);
495
496 list_for_each_entry_safe(q, n, &pending->list, list) {
497 int sig = q->info.si_signo;
498
499 if (likely(q->info.si_code != SI_TIMER)) {
500 sigaddset(&retain, sig);
501 } else {
502 sigdelset(&signal, sig);
503 list_del_init(&q->list);
504 __sigqueue_free(q);
505 }
506 }
507
508 sigorsets(&pending->signal, &signal, &retain);
509 }
510
511 void flush_itimer_signals(void)
512 {
513 struct task_struct *tsk = current;
514 unsigned long flags;
515
516 spin_lock_irqsave(&tsk->sighand->siglock, flags);
517 __flush_itimer_signals(&tsk->pending);
518 __flush_itimer_signals(&tsk->signal->shared_pending);
519 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
520 }
521 #endif
522
523 void ignore_signals(struct task_struct *t)
524 {
525 int i;
526
527 for (i = 0; i < _NSIG; ++i)
528 t->sighand->action[i].sa.sa_handler = SIG_IGN;
529
530 flush_signals(t);
531 }
532
533 /*
534 * Flush all handlers for a task.
535 */
536
537 void
538 flush_signal_handlers(struct task_struct *t, int force_default)
539 {
540 int i;
541 struct k_sigaction *ka = &t->sighand->action[0];
542 for (i = _NSIG ; i != 0 ; i--) {
543 if (force_default || ka->sa.sa_handler != SIG_IGN)
544 ka->sa.sa_handler = SIG_DFL;
545 ka->sa.sa_flags = 0;
546 #ifdef __ARCH_HAS_SA_RESTORER
547 ka->sa.sa_restorer = NULL;
548 #endif
549 sigemptyset(&ka->sa.sa_mask);
550 ka++;
551 }
552 }
553
554 bool unhandled_signal(struct task_struct *tsk, int sig)
555 {
556 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
557 if (is_global_init(tsk))
558 return true;
559
560 if (handler != SIG_IGN && handler != SIG_DFL)
561 return false;
562
563 /* if ptraced, let the tracer determine */
564 return !tsk->ptrace;
565 }
566
567 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
568 bool *resched_timer)
569 {
570 struct sigqueue *q, *first = NULL;
571
572 /*
573 * Collect the siginfo appropriate to this signal. Check if
574 * there is another siginfo for the same signal.
575 */
576 list_for_each_entry(q, &list->list, list) {
577 if (q->info.si_signo == sig) {
578 if (first)
579 goto still_pending;
580 first = q;
581 }
582 }
583
584 sigdelset(&list->signal, sig);
585
586 if (first) {
587 still_pending:
588 list_del_init(&first->list);
589 copy_siginfo(info, &first->info);
590
591 *resched_timer =
592 (first->flags & SIGQUEUE_PREALLOC) &&
593 (info->si_code == SI_TIMER) &&
594 (info->si_sys_private);
595
596 __sigqueue_free(first);
597 } else {
598 /*
599 * Ok, it wasn't in the queue. This must be
600 * a fast-pathed signal or we must have been
601 * out of queue space. So zero out the info.
602 */
603 clear_siginfo(info);
604 info->si_signo = sig;
605 info->si_errno = 0;
606 info->si_code = SI_USER;
607 info->si_pid = 0;
608 info->si_uid = 0;
609 }
610 }
611
612 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
613 kernel_siginfo_t *info, bool *resched_timer)
614 {
615 int sig = next_signal(pending, mask);
616
617 if (sig)
618 collect_signal(sig, pending, info, resched_timer);
619 return sig;
620 }
621
622 /*
623 * Dequeue a signal and return the element to the caller, which is
624 * expected to free it.
625 *
626 * All callers have to hold the siglock.
627 */
628 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
629 {
630 bool resched_timer = false;
631 int signr;
632
633 /* We only dequeue private signals from ourselves, we don't let
634 * signalfd steal them
635 */
636 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
637 if (!signr) {
638 signr = __dequeue_signal(&tsk->signal->shared_pending,
639 mask, info, &resched_timer);
640 #ifdef CONFIG_POSIX_TIMERS
641 /*
642 * itimer signal ?
643 *
644 * itimers are process shared and we restart periodic
645 * itimers in the signal delivery path to prevent DoS
646 * attacks in the high resolution timer case. This is
647 * compliant with the old way of self-restarting
648 * itimers, as the SIGALRM is a legacy signal and only
649 * queued once. Changing the restart behaviour to
650 * restart the timer in the signal dequeue path is
651 * reducing the timer noise on heavy loaded !highres
652 * systems too.
653 */
654 if (unlikely(signr == SIGALRM)) {
655 struct hrtimer *tmr = &tsk->signal->real_timer;
656
657 if (!hrtimer_is_queued(tmr) &&
658 tsk->signal->it_real_incr != 0) {
659 hrtimer_forward(tmr, tmr->base->get_time(),
660 tsk->signal->it_real_incr);
661 hrtimer_restart(tmr);
662 }
663 }
664 #endif
665 }
666
667 recalc_sigpending();
668 if (!signr)
669 return 0;
670
671 if (unlikely(sig_kernel_stop(signr))) {
672 /*
673 * Set a marker that we have dequeued a stop signal. Our
674 * caller might release the siglock and then the pending
675 * stop signal it is about to process is no longer in the
676 * pending bitmasks, but must still be cleared by a SIGCONT
677 * (and overruled by a SIGKILL). So those cases clear this
678 * shared flag after we've set it. Note that this flag may
679 * remain set after the signal we return is ignored or
680 * handled. That doesn't matter because its only purpose
681 * is to alert stop-signal processing code when another
682 * processor has come along and cleared the flag.
683 */
684 current->jobctl |= JOBCTL_STOP_DEQUEUED;
685 }
686 #ifdef CONFIG_POSIX_TIMERS
687 if (resched_timer) {
688 /*
689 * Release the siglock to ensure proper locking order
690 * of timer locks outside of siglocks. Note, we leave
691 * irqs disabled here, since the posix-timers code is
692 * about to disable them again anyway.
693 */
694 spin_unlock(&tsk->sighand->siglock);
695 posixtimer_rearm(info);
696 spin_lock(&tsk->sighand->siglock);
697
698 /* Don't expose the si_sys_private value to userspace */
699 info->si_sys_private = 0;
700 }
701 #endif
702 return signr;
703 }
704 EXPORT_SYMBOL_GPL(dequeue_signal);
705
706 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
707 {
708 struct task_struct *tsk = current;
709 struct sigpending *pending = &tsk->pending;
710 struct sigqueue *q, *sync = NULL;
711
712 /*
713 * Might a synchronous signal be in the queue?
714 */
715 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
716 return 0;
717
718 /*
719 * Return the first synchronous signal in the queue.
720 */
721 list_for_each_entry(q, &pending->list, list) {
722 /* Synchronous signals have a positive si_code */
723 if ((q->info.si_code > SI_USER) &&
724 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
725 sync = q;
726 goto next;
727 }
728 }
729 return 0;
730 next:
731 /*
732 * Check if there is another siginfo for the same signal.
733 */
734 list_for_each_entry_continue(q, &pending->list, list) {
735 if (q->info.si_signo == sync->info.si_signo)
736 goto still_pending;
737 }
738
739 sigdelset(&pending->signal, sync->info.si_signo);
740 recalc_sigpending();
741 still_pending:
742 list_del_init(&sync->list);
743 copy_siginfo(info, &sync->info);
744 __sigqueue_free(sync);
745 return info->si_signo;
746 }
747
748 /*
749 * Tell a process that it has a new active signal..
750 *
751 * NOTE! we rely on the previous spin_lock to
752 * lock interrupts for us! We can only be called with
753 * "siglock" held, and the local interrupt must
754 * have been disabled when that got acquired!
755 *
756 * No need to set need_resched since signal event passing
757 * goes through ->blocked
758 */
759 void signal_wake_up_state(struct task_struct *t, unsigned int state)
760 {
761 set_tsk_thread_flag(t, TIF_SIGPENDING);
762 /*
763 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
764 * case. We don't check t->state here because there is a race with it
765 * executing another processor and just now entering stopped state.
766 * By using wake_up_state, we ensure the process will wake up and
767 * handle its death signal.
768 */
769 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
770 kick_process(t);
771 }
772
773 /*
774 * Remove signals in mask from the pending set and queue.
775 * Returns 1 if any signals were found.
776 *
777 * All callers must be holding the siglock.
778 */
779 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
780 {
781 struct sigqueue *q, *n;
782 sigset_t m;
783
784 sigandsets(&m, mask, &s->signal);
785 if (sigisemptyset(&m))
786 return;
787
788 sigandnsets(&s->signal, &s->signal, mask);
789 list_for_each_entry_safe(q, n, &s->list, list) {
790 if (sigismember(mask, q->info.si_signo)) {
791 list_del_init(&q->list);
792 __sigqueue_free(q);
793 }
794 }
795 }
796
797 static inline int is_si_special(const struct kernel_siginfo *info)
798 {
799 return info <= SEND_SIG_PRIV;
800 }
801
802 static inline bool si_fromuser(const struct kernel_siginfo *info)
803 {
804 return info == SEND_SIG_NOINFO ||
805 (!is_si_special(info) && SI_FROMUSER(info));
806 }
807
808 /*
809 * called with RCU read lock from check_kill_permission()
810 */
811 static bool kill_ok_by_cred(struct task_struct *t)
812 {
813 const struct cred *cred = current_cred();
814 const struct cred *tcred = __task_cred(t);
815
816 return uid_eq(cred->euid, tcred->suid) ||
817 uid_eq(cred->euid, tcred->uid) ||
818 uid_eq(cred->uid, tcred->suid) ||
819 uid_eq(cred->uid, tcred->uid) ||
820 ns_capable(tcred->user_ns, CAP_KILL);
821 }
822
823 /*
824 * Bad permissions for sending the signal
825 * - the caller must hold the RCU read lock
826 */
827 static int check_kill_permission(int sig, struct kernel_siginfo *info,
828 struct task_struct *t)
829 {
830 struct pid *sid;
831 int error;
832
833 if (!valid_signal(sig))
834 return -EINVAL;
835
836 if (!si_fromuser(info))
837 return 0;
838
839 error = audit_signal_info(sig, t); /* Let audit system see the signal */
840 if (error)
841 return error;
842
843 if (!same_thread_group(current, t) &&
844 !kill_ok_by_cred(t)) {
845 switch (sig) {
846 case SIGCONT:
847 sid = task_session(t);
848 /*
849 * We don't return the error if sid == NULL. The
850 * task was unhashed, the caller must notice this.
851 */
852 if (!sid || sid == task_session(current))
853 break;
854 fallthrough;
855 default:
856 return -EPERM;
857 }
858 }
859
860 return security_task_kill(t, info, sig, NULL);
861 }
862
863 /**
864 * ptrace_trap_notify - schedule trap to notify ptracer
865 * @t: tracee wanting to notify tracer
866 *
867 * This function schedules sticky ptrace trap which is cleared on the next
868 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
869 * ptracer.
870 *
871 * If @t is running, STOP trap will be taken. If trapped for STOP and
872 * ptracer is listening for events, tracee is woken up so that it can
873 * re-trap for the new event. If trapped otherwise, STOP trap will be
874 * eventually taken without returning to userland after the existing traps
875 * are finished by PTRACE_CONT.
876 *
877 * CONTEXT:
878 * Must be called with @task->sighand->siglock held.
879 */
880 static void ptrace_trap_notify(struct task_struct *t)
881 {
882 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
883 assert_spin_locked(&t->sighand->siglock);
884
885 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
886 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
887 }
888
889 /*
890 * Handle magic process-wide effects of stop/continue signals. Unlike
891 * the signal actions, these happen immediately at signal-generation
892 * time regardless of blocking, ignoring, or handling. This does the
893 * actual continuing for SIGCONT, but not the actual stopping for stop
894 * signals. The process stop is done as a signal action for SIG_DFL.
895 *
896 * Returns true if the signal should be actually delivered, otherwise
897 * it should be dropped.
898 */
899 static bool prepare_signal(int sig, struct task_struct *p, bool force)
900 {
901 struct signal_struct *signal = p->signal;
902 struct task_struct *t;
903 sigset_t flush;
904
905 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
906 if (!(signal->flags & SIGNAL_GROUP_EXIT))
907 return sig == SIGKILL;
908 /*
909 * The process is in the middle of dying, nothing to do.
910 */
911 } else if (sig_kernel_stop(sig)) {
912 /*
913 * This is a stop signal. Remove SIGCONT from all queues.
914 */
915 siginitset(&flush, sigmask(SIGCONT));
916 flush_sigqueue_mask(&flush, &signal->shared_pending);
917 for_each_thread(p, t)
918 flush_sigqueue_mask(&flush, &t->pending);
919 } else if (sig == SIGCONT) {
920 unsigned int why;
921 /*
922 * Remove all stop signals from all queues, wake all threads.
923 */
924 siginitset(&flush, SIG_KERNEL_STOP_MASK);
925 flush_sigqueue_mask(&flush, &signal->shared_pending);
926 for_each_thread(p, t) {
927 flush_sigqueue_mask(&flush, &t->pending);
928 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
929 if (likely(!(t->ptrace & PT_SEIZED)))
930 wake_up_state(t, __TASK_STOPPED);
931 else
932 ptrace_trap_notify(t);
933 }
934
935 /*
936 * Notify the parent with CLD_CONTINUED if we were stopped.
937 *
938 * If we were in the middle of a group stop, we pretend it
939 * was already finished, and then continued. Since SIGCHLD
940 * doesn't queue we report only CLD_STOPPED, as if the next
941 * CLD_CONTINUED was dropped.
942 */
943 why = 0;
944 if (signal->flags & SIGNAL_STOP_STOPPED)
945 why |= SIGNAL_CLD_CONTINUED;
946 else if (signal->group_stop_count)
947 why |= SIGNAL_CLD_STOPPED;
948
949 if (why) {
950 /*
951 * The first thread which returns from do_signal_stop()
952 * will take ->siglock, notice SIGNAL_CLD_MASK, and
953 * notify its parent. See get_signal().
954 */
955 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
956 signal->group_stop_count = 0;
957 signal->group_exit_code = 0;
958 }
959 }
960
961 return !sig_ignored(p, sig, force);
962 }
963
964 /*
965 * Test if P wants to take SIG. After we've checked all threads with this,
966 * it's equivalent to finding no threads not blocking SIG. Any threads not
967 * blocking SIG were ruled out because they are not running and already
968 * have pending signals. Such threads will dequeue from the shared queue
969 * as soon as they're available, so putting the signal on the shared queue
970 * will be equivalent to sending it to one such thread.
971 */
972 static inline bool wants_signal(int sig, struct task_struct *p)
973 {
974 if (sigismember(&p->blocked, sig))
975 return false;
976
977 if (p->flags & PF_EXITING)
978 return false;
979
980 if (sig == SIGKILL)
981 return true;
982
983 if (task_is_stopped_or_traced(p))
984 return false;
985
986 return task_curr(p) || !task_sigpending(p);
987 }
988
989 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
990 {
991 struct signal_struct *signal = p->signal;
992 struct task_struct *t;
993
994 /*
995 * Now find a thread we can wake up to take the signal off the queue.
996 *
997 * If the main thread wants the signal, it gets first crack.
998 * Probably the least surprising to the average bear.
999 */
1000 if (wants_signal(sig, p))
1001 t = p;
1002 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1003 /*
1004 * There is just one thread and it does not need to be woken.
1005 * It will dequeue unblocked signals before it runs again.
1006 */
1007 return;
1008 else {
1009 /*
1010 * Otherwise try to find a suitable thread.
1011 */
1012 t = signal->curr_target;
1013 while (!wants_signal(sig, t)) {
1014 t = next_thread(t);
1015 if (t == signal->curr_target)
1016 /*
1017 * No thread needs to be woken.
1018 * Any eligible threads will see
1019 * the signal in the queue soon.
1020 */
1021 return;
1022 }
1023 signal->curr_target = t;
1024 }
1025
1026 /*
1027 * Found a killable thread. If the signal will be fatal,
1028 * then start taking the whole group down immediately.
1029 */
1030 if (sig_fatal(p, sig) &&
1031 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1032 !sigismember(&t->real_blocked, sig) &&
1033 (sig == SIGKILL || !p->ptrace)) {
1034 /*
1035 * This signal will be fatal to the whole group.
1036 */
1037 if (!sig_kernel_coredump(sig)) {
1038 /*
1039 * Start a group exit and wake everybody up.
1040 * This way we don't have other threads
1041 * running and doing things after a slower
1042 * thread has the fatal signal pending.
1043 */
1044 signal->flags = SIGNAL_GROUP_EXIT;
1045 signal->group_exit_code = sig;
1046 signal->group_stop_count = 0;
1047 t = p;
1048 do {
1049 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1050 sigaddset(&t->pending.signal, SIGKILL);
1051 signal_wake_up(t, 1);
1052 } while_each_thread(p, t);
1053 return;
1054 }
1055 }
1056
1057 /*
1058 * The signal is already in the shared-pending queue.
1059 * Tell the chosen thread to wake up and dequeue it.
1060 */
1061 signal_wake_up(t, sig == SIGKILL);
1062 return;
1063 }
1064
1065 static inline bool legacy_queue(struct sigpending *signals, int sig)
1066 {
1067 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1068 }
1069
1070 static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1071 enum pid_type type, bool force)
1072 {
1073 struct sigpending *pending;
1074 struct sigqueue *q;
1075 int override_rlimit;
1076 int ret = 0, result;
1077
1078 assert_spin_locked(&t->sighand->siglock);
1079
1080 result = TRACE_SIGNAL_IGNORED;
1081 if (!prepare_signal(sig, t, force))
1082 goto ret;
1083
1084 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1085 /*
1086 * Short-circuit ignored signals and support queuing
1087 * exactly one non-rt signal, so that we can get more
1088 * detailed information about the cause of the signal.
1089 */
1090 result = TRACE_SIGNAL_ALREADY_PENDING;
1091 if (legacy_queue(pending, sig))
1092 goto ret;
1093
1094 result = TRACE_SIGNAL_DELIVERED;
1095 /*
1096 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1097 */
1098 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1099 goto out_set;
1100
1101 /*
1102 * Real-time signals must be queued if sent by sigqueue, or
1103 * some other real-time mechanism. It is implementation
1104 * defined whether kill() does so. We attempt to do so, on
1105 * the principle of least surprise, but since kill is not
1106 * allowed to fail with EAGAIN when low on memory we just
1107 * make sure at least one signal gets delivered and don't
1108 * pass on the info struct.
1109 */
1110 if (sig < SIGRTMIN)
1111 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1112 else
1113 override_rlimit = 0;
1114
1115 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1116
1117 if (q) {
1118 list_add_tail(&q->list, &pending->list);
1119 switch ((unsigned long) info) {
1120 case (unsigned long) SEND_SIG_NOINFO:
1121 clear_siginfo(&q->info);
1122 q->info.si_signo = sig;
1123 q->info.si_errno = 0;
1124 q->info.si_code = SI_USER;
1125 q->info.si_pid = task_tgid_nr_ns(current,
1126 task_active_pid_ns(t));
1127 rcu_read_lock();
1128 q->info.si_uid =
1129 from_kuid_munged(task_cred_xxx(t, user_ns),
1130 current_uid());
1131 rcu_read_unlock();
1132 break;
1133 case (unsigned long) SEND_SIG_PRIV:
1134 clear_siginfo(&q->info);
1135 q->info.si_signo = sig;
1136 q->info.si_errno = 0;
1137 q->info.si_code = SI_KERNEL;
1138 q->info.si_pid = 0;
1139 q->info.si_uid = 0;
1140 break;
1141 default:
1142 copy_siginfo(&q->info, info);
1143 break;
1144 }
1145 } else if (!is_si_special(info) &&
1146 sig >= SIGRTMIN && info->si_code != SI_USER) {
1147 /*
1148 * Queue overflow, abort. We may abort if the
1149 * signal was rt and sent by user using something
1150 * other than kill().
1151 */
1152 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1153 ret = -EAGAIN;
1154 goto ret;
1155 } else {
1156 /*
1157 * This is a silent loss of information. We still
1158 * send the signal, but the *info bits are lost.
1159 */
1160 result = TRACE_SIGNAL_LOSE_INFO;
1161 }
1162
1163 out_set:
1164 signalfd_notify(t, sig);
1165 sigaddset(&pending->signal, sig);
1166
1167 /* Let multiprocess signals appear after on-going forks */
1168 if (type > PIDTYPE_TGID) {
1169 struct multiprocess_signals *delayed;
1170 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1171 sigset_t *signal = &delayed->signal;
1172 /* Can't queue both a stop and a continue signal */
1173 if (sig == SIGCONT)
1174 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1175 else if (sig_kernel_stop(sig))
1176 sigdelset(signal, SIGCONT);
1177 sigaddset(signal, sig);
1178 }
1179 }
1180
1181 complete_signal(sig, t, type);
1182 ret:
1183 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1184 return ret;
1185 }
1186
1187 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1188 {
1189 bool ret = false;
1190 switch (siginfo_layout(info->si_signo, info->si_code)) {
1191 case SIL_KILL:
1192 case SIL_CHLD:
1193 case SIL_RT:
1194 ret = true;
1195 break;
1196 case SIL_TIMER:
1197 case SIL_POLL:
1198 case SIL_FAULT:
1199 case SIL_FAULT_TRAPNO:
1200 case SIL_FAULT_MCEERR:
1201 case SIL_FAULT_BNDERR:
1202 case SIL_FAULT_PKUERR:
1203 case SIL_PERF_EVENT:
1204 case SIL_SYS:
1205 ret = false;
1206 break;
1207 }
1208 return ret;
1209 }
1210
1211 static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1212 enum pid_type type)
1213 {
1214 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1215 bool force = false;
1216
1217 if (info == SEND_SIG_NOINFO) {
1218 /* Force if sent from an ancestor pid namespace */
1219 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1220 } else if (info == SEND_SIG_PRIV) {
1221 /* Don't ignore kernel generated signals */
1222 force = true;
1223 } else if (has_si_pid_and_uid(info)) {
1224 /* SIGKILL and SIGSTOP is special or has ids */
1225 struct user_namespace *t_user_ns;
1226
1227 rcu_read_lock();
1228 t_user_ns = task_cred_xxx(t, user_ns);
1229 if (current_user_ns() != t_user_ns) {
1230 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1231 info->si_uid = from_kuid_munged(t_user_ns, uid);
1232 }
1233 rcu_read_unlock();
1234
1235 /* A kernel generated signal? */
1236 force = (info->si_code == SI_KERNEL);
1237
1238 /* From an ancestor pid namespace? */
1239 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1240 info->si_pid = 0;
1241 force = true;
1242 }
1243 }
1244 return __send_signal(sig, info, t, type, force);
1245 }
1246
1247 static void print_fatal_signal(int signr)
1248 {
1249 struct pt_regs *regs = signal_pt_regs();
1250 pr_info("potentially unexpected fatal signal %d.\n", signr);
1251
1252 #if defined(__i386__) && !defined(__arch_um__)
1253 pr_info("code at %08lx: ", regs->ip);
1254 {
1255 int i;
1256 for (i = 0; i < 16; i++) {
1257 unsigned char insn;
1258
1259 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1260 break;
1261 pr_cont("%02x ", insn);
1262 }
1263 }
1264 pr_cont("\n");
1265 #endif
1266 preempt_disable();
1267 show_regs(regs);
1268 preempt_enable();
1269 }
1270
1271 static int __init setup_print_fatal_signals(char *str)
1272 {
1273 get_option (&str, &print_fatal_signals);
1274
1275 return 1;
1276 }
1277
1278 __setup("print-fatal-signals=", setup_print_fatal_signals);
1279
1280 int
1281 __group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1282 {
1283 return send_signal(sig, info, p, PIDTYPE_TGID);
1284 }
1285
1286 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1287 enum pid_type type)
1288 {
1289 unsigned long flags;
1290 int ret = -ESRCH;
1291
1292 if (lock_task_sighand(p, &flags)) {
1293 ret = send_signal(sig, info, p, type);
1294 unlock_task_sighand(p, &flags);
1295 }
1296
1297 return ret;
1298 }
1299
1300 /*
1301 * Force a signal that the process can't ignore: if necessary
1302 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1303 *
1304 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1305 * since we do not want to have a signal handler that was blocked
1306 * be invoked when user space had explicitly blocked it.
1307 *
1308 * We don't want to have recursive SIGSEGV's etc, for example,
1309 * that is why we also clear SIGNAL_UNKILLABLE.
1310 */
1311 static int
1312 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1313 {
1314 unsigned long int flags;
1315 int ret, blocked, ignored;
1316 struct k_sigaction *action;
1317 int sig = info->si_signo;
1318
1319 spin_lock_irqsave(&t->sighand->siglock, flags);
1320 action = &t->sighand->action[sig-1];
1321 ignored = action->sa.sa_handler == SIG_IGN;
1322 blocked = sigismember(&t->blocked, sig);
1323 if (blocked || ignored) {
1324 action->sa.sa_handler = SIG_DFL;
1325 if (blocked) {
1326 sigdelset(&t->blocked, sig);
1327 recalc_sigpending_and_wake(t);
1328 }
1329 }
1330 /*
1331 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1332 * debugging to leave init killable.
1333 */
1334 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1335 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1336 ret = send_signal(sig, info, t, PIDTYPE_PID);
1337 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1338
1339 return ret;
1340 }
1341
1342 int force_sig_info(struct kernel_siginfo *info)
1343 {
1344 return force_sig_info_to_task(info, current);
1345 }
1346
1347 /*
1348 * Nuke all other threads in the group.
1349 */
1350 int zap_other_threads(struct task_struct *p)
1351 {
1352 struct task_struct *t = p;
1353 int count = 0;
1354
1355 p->signal->group_stop_count = 0;
1356
1357 while_each_thread(p, t) {
1358 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1359 count++;
1360
1361 /* Don't bother with already dead threads */
1362 if (t->exit_state)
1363 continue;
1364 sigaddset(&t->pending.signal, SIGKILL);
1365 signal_wake_up(t, 1);
1366 }
1367
1368 return count;
1369 }
1370
1371 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1372 unsigned long *flags)
1373 {
1374 struct sighand_struct *sighand;
1375
1376 rcu_read_lock();
1377 for (;;) {
1378 sighand = rcu_dereference(tsk->sighand);
1379 if (unlikely(sighand == NULL))
1380 break;
1381
1382 /*
1383 * This sighand can be already freed and even reused, but
1384 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1385 * initializes ->siglock: this slab can't go away, it has
1386 * the same object type, ->siglock can't be reinitialized.
1387 *
1388 * We need to ensure that tsk->sighand is still the same
1389 * after we take the lock, we can race with de_thread() or
1390 * __exit_signal(). In the latter case the next iteration
1391 * must see ->sighand == NULL.
1392 */
1393 spin_lock_irqsave(&sighand->siglock, *flags);
1394 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1395 break;
1396 spin_unlock_irqrestore(&sighand->siglock, *flags);
1397 }
1398 rcu_read_unlock();
1399
1400 return sighand;
1401 }
1402
1403 /*
1404 * send signal info to all the members of a group
1405 */
1406 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1407 struct task_struct *p, enum pid_type type)
1408 {
1409 int ret;
1410
1411 rcu_read_lock();
1412 ret = check_kill_permission(sig, info, p);
1413 rcu_read_unlock();
1414
1415 if (!ret && sig)
1416 ret = do_send_sig_info(sig, info, p, type);
1417
1418 return ret;
1419 }
1420
1421 /*
1422 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1423 * control characters do (^C, ^Z etc)
1424 * - the caller must hold at least a readlock on tasklist_lock
1425 */
1426 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1427 {
1428 struct task_struct *p = NULL;
1429 int retval, success;
1430
1431 success = 0;
1432 retval = -ESRCH;
1433 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1434 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1435 success |= !err;
1436 retval = err;
1437 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1438 return success ? 0 : retval;
1439 }
1440
1441 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1442 {
1443 int error = -ESRCH;
1444 struct task_struct *p;
1445
1446 for (;;) {
1447 rcu_read_lock();
1448 p = pid_task(pid, PIDTYPE_PID);
1449 if (p)
1450 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1451 rcu_read_unlock();
1452 if (likely(!p || error != -ESRCH))
1453 return error;
1454
1455 /*
1456 * The task was unhashed in between, try again. If it
1457 * is dead, pid_task() will return NULL, if we race with
1458 * de_thread() it will find the new leader.
1459 */
1460 }
1461 }
1462
1463 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1464 {
1465 int error;
1466 rcu_read_lock();
1467 error = kill_pid_info(sig, info, find_vpid(pid));
1468 rcu_read_unlock();
1469 return error;
1470 }
1471
1472 static inline bool kill_as_cred_perm(const struct cred *cred,
1473 struct task_struct *target)
1474 {
1475 const struct cred *pcred = __task_cred(target);
1476
1477 return uid_eq(cred->euid, pcred->suid) ||
1478 uid_eq(cred->euid, pcred->uid) ||
1479 uid_eq(cred->uid, pcred->suid) ||
1480 uid_eq(cred->uid, pcred->uid);
1481 }
1482
1483 /*
1484 * The usb asyncio usage of siginfo is wrong. The glibc support
1485 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1486 * AKA after the generic fields:
1487 * kernel_pid_t si_pid;
1488 * kernel_uid32_t si_uid;
1489 * sigval_t si_value;
1490 *
1491 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1492 * after the generic fields is:
1493 * void __user *si_addr;
1494 *
1495 * This is a practical problem when there is a 64bit big endian kernel
1496 * and a 32bit userspace. As the 32bit address will encoded in the low
1497 * 32bits of the pointer. Those low 32bits will be stored at higher
1498 * address than appear in a 32 bit pointer. So userspace will not
1499 * see the address it was expecting for it's completions.
1500 *
1501 * There is nothing in the encoding that can allow
1502 * copy_siginfo_to_user32 to detect this confusion of formats, so
1503 * handle this by requiring the caller of kill_pid_usb_asyncio to
1504 * notice when this situration takes place and to store the 32bit
1505 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1506 * parameter.
1507 */
1508 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1509 struct pid *pid, const struct cred *cred)
1510 {
1511 struct kernel_siginfo info;
1512 struct task_struct *p;
1513 unsigned long flags;
1514 int ret = -EINVAL;
1515
1516 if (!valid_signal(sig))
1517 return ret;
1518
1519 clear_siginfo(&info);
1520 info.si_signo = sig;
1521 info.si_errno = errno;
1522 info.si_code = SI_ASYNCIO;
1523 *((sigval_t *)&info.si_pid) = addr;
1524
1525 rcu_read_lock();
1526 p = pid_task(pid, PIDTYPE_PID);
1527 if (!p) {
1528 ret = -ESRCH;
1529 goto out_unlock;
1530 }
1531 if (!kill_as_cred_perm(cred, p)) {
1532 ret = -EPERM;
1533 goto out_unlock;
1534 }
1535 ret = security_task_kill(p, &info, sig, cred);
1536 if (ret)
1537 goto out_unlock;
1538
1539 if (sig) {
1540 if (lock_task_sighand(p, &flags)) {
1541 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1542 unlock_task_sighand(p, &flags);
1543 } else
1544 ret = -ESRCH;
1545 }
1546 out_unlock:
1547 rcu_read_unlock();
1548 return ret;
1549 }
1550 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1551
1552 /*
1553 * kill_something_info() interprets pid in interesting ways just like kill(2).
1554 *
1555 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1556 * is probably wrong. Should make it like BSD or SYSV.
1557 */
1558
1559 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1560 {
1561 int ret;
1562
1563 if (pid > 0)
1564 return kill_proc_info(sig, info, pid);
1565
1566 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1567 if (pid == INT_MIN)
1568 return -ESRCH;
1569
1570 read_lock(&tasklist_lock);
1571 if (pid != -1) {
1572 ret = __kill_pgrp_info(sig, info,
1573 pid ? find_vpid(-pid) : task_pgrp(current));
1574 } else {
1575 int retval = 0, count = 0;
1576 struct task_struct * p;
1577
1578 for_each_process(p) {
1579 if (task_pid_vnr(p) > 1 &&
1580 !same_thread_group(p, current)) {
1581 int err = group_send_sig_info(sig, info, p,
1582 PIDTYPE_MAX);
1583 ++count;
1584 if (err != -EPERM)
1585 retval = err;
1586 }
1587 }
1588 ret = count ? retval : -ESRCH;
1589 }
1590 read_unlock(&tasklist_lock);
1591
1592 return ret;
1593 }
1594
1595 /*
1596 * These are for backward compatibility with the rest of the kernel source.
1597 */
1598
1599 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1600 {
1601 /*
1602 * Make sure legacy kernel users don't send in bad values
1603 * (normal paths check this in check_kill_permission).
1604 */
1605 if (!valid_signal(sig))
1606 return -EINVAL;
1607
1608 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1609 }
1610 EXPORT_SYMBOL(send_sig_info);
1611
1612 #define __si_special(priv) \
1613 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1614
1615 int
1616 send_sig(int sig, struct task_struct *p, int priv)
1617 {
1618 return send_sig_info(sig, __si_special(priv), p);
1619 }
1620 EXPORT_SYMBOL(send_sig);
1621
1622 void force_sig(int sig)
1623 {
1624 struct kernel_siginfo info;
1625
1626 clear_siginfo(&info);
1627 info.si_signo = sig;
1628 info.si_errno = 0;
1629 info.si_code = SI_KERNEL;
1630 info.si_pid = 0;
1631 info.si_uid = 0;
1632 force_sig_info(&info);
1633 }
1634 EXPORT_SYMBOL(force_sig);
1635
1636 /*
1637 * When things go south during signal handling, we
1638 * will force a SIGSEGV. And if the signal that caused
1639 * the problem was already a SIGSEGV, we'll want to
1640 * make sure we don't even try to deliver the signal..
1641 */
1642 void force_sigsegv(int sig)
1643 {
1644 struct task_struct *p = current;
1645
1646 if (sig == SIGSEGV) {
1647 unsigned long flags;
1648 spin_lock_irqsave(&p->sighand->siglock, flags);
1649 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1650 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1651 }
1652 force_sig(SIGSEGV);
1653 }
1654
1655 int force_sig_fault_to_task(int sig, int code, void __user *addr
1656 ___ARCH_SI_TRAPNO(int trapno)
1657 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1658 , struct task_struct *t)
1659 {
1660 struct kernel_siginfo info;
1661
1662 clear_siginfo(&info);
1663 info.si_signo = sig;
1664 info.si_errno = 0;
1665 info.si_code = code;
1666 info.si_addr = addr;
1667 #ifdef __ARCH_SI_TRAPNO
1668 info.si_trapno = trapno;
1669 #endif
1670 #ifdef __ia64__
1671 info.si_imm = imm;
1672 info.si_flags = flags;
1673 info.si_isr = isr;
1674 #endif
1675 return force_sig_info_to_task(&info, t);
1676 }
1677
1678 int force_sig_fault(int sig, int code, void __user *addr
1679 ___ARCH_SI_TRAPNO(int trapno)
1680 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1681 {
1682 return force_sig_fault_to_task(sig, code, addr
1683 ___ARCH_SI_TRAPNO(trapno)
1684 ___ARCH_SI_IA64(imm, flags, isr), current);
1685 }
1686
1687 int send_sig_fault(int sig, int code, void __user *addr
1688 ___ARCH_SI_TRAPNO(int trapno)
1689 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1690 , struct task_struct *t)
1691 {
1692 struct kernel_siginfo info;
1693
1694 clear_siginfo(&info);
1695 info.si_signo = sig;
1696 info.si_errno = 0;
1697 info.si_code = code;
1698 info.si_addr = addr;
1699 #ifdef __ARCH_SI_TRAPNO
1700 info.si_trapno = trapno;
1701 #endif
1702 #ifdef __ia64__
1703 info.si_imm = imm;
1704 info.si_flags = flags;
1705 info.si_isr = isr;
1706 #endif
1707 return send_sig_info(info.si_signo, &info, t);
1708 }
1709
1710 int force_sig_mceerr(int code, void __user *addr, short lsb)
1711 {
1712 struct kernel_siginfo info;
1713
1714 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1715 clear_siginfo(&info);
1716 info.si_signo = SIGBUS;
1717 info.si_errno = 0;
1718 info.si_code = code;
1719 info.si_addr = addr;
1720 info.si_addr_lsb = lsb;
1721 return force_sig_info(&info);
1722 }
1723
1724 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1725 {
1726 struct kernel_siginfo info;
1727
1728 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1729 clear_siginfo(&info);
1730 info.si_signo = SIGBUS;
1731 info.si_errno = 0;
1732 info.si_code = code;
1733 info.si_addr = addr;
1734 info.si_addr_lsb = lsb;
1735 return send_sig_info(info.si_signo, &info, t);
1736 }
1737 EXPORT_SYMBOL(send_sig_mceerr);
1738
1739 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1740 {
1741 struct kernel_siginfo info;
1742
1743 clear_siginfo(&info);
1744 info.si_signo = SIGSEGV;
1745 info.si_errno = 0;
1746 info.si_code = SEGV_BNDERR;
1747 info.si_addr = addr;
1748 info.si_lower = lower;
1749 info.si_upper = upper;
1750 return force_sig_info(&info);
1751 }
1752
1753 #ifdef SEGV_PKUERR
1754 int force_sig_pkuerr(void __user *addr, u32 pkey)
1755 {
1756 struct kernel_siginfo info;
1757
1758 clear_siginfo(&info);
1759 info.si_signo = SIGSEGV;
1760 info.si_errno = 0;
1761 info.si_code = SEGV_PKUERR;
1762 info.si_addr = addr;
1763 info.si_pkey = pkey;
1764 return force_sig_info(&info);
1765 }
1766 #endif
1767
1768 int force_sig_perf(void __user *addr, u32 type, u64 sig_data)
1769 {
1770 struct kernel_siginfo info;
1771
1772 clear_siginfo(&info);
1773 info.si_signo = SIGTRAP;
1774 info.si_errno = 0;
1775 info.si_code = TRAP_PERF;
1776 info.si_addr = addr;
1777 info.si_perf_data = sig_data;
1778 info.si_perf_type = type;
1779
1780 return force_sig_info(&info);
1781 }
1782
1783 /* For the crazy architectures that include trap information in
1784 * the errno field, instead of an actual errno value.
1785 */
1786 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1787 {
1788 struct kernel_siginfo info;
1789
1790 clear_siginfo(&info);
1791 info.si_signo = SIGTRAP;
1792 info.si_errno = errno;
1793 info.si_code = TRAP_HWBKPT;
1794 info.si_addr = addr;
1795 return force_sig_info(&info);
1796 }
1797
1798 int kill_pgrp(struct pid *pid, int sig, int priv)
1799 {
1800 int ret;
1801
1802 read_lock(&tasklist_lock);
1803 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1804 read_unlock(&tasklist_lock);
1805
1806 return ret;
1807 }
1808 EXPORT_SYMBOL(kill_pgrp);
1809
1810 int kill_pid(struct pid *pid, int sig, int priv)
1811 {
1812 return kill_pid_info(sig, __si_special(priv), pid);
1813 }
1814 EXPORT_SYMBOL(kill_pid);
1815
1816 /*
1817 * These functions support sending signals using preallocated sigqueue
1818 * structures. This is needed "because realtime applications cannot
1819 * afford to lose notifications of asynchronous events, like timer
1820 * expirations or I/O completions". In the case of POSIX Timers
1821 * we allocate the sigqueue structure from the timer_create. If this
1822 * allocation fails we are able to report the failure to the application
1823 * with an EAGAIN error.
1824 */
1825 struct sigqueue *sigqueue_alloc(void)
1826 {
1827 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1828 }
1829
1830 void sigqueue_free(struct sigqueue *q)
1831 {
1832 unsigned long flags;
1833 spinlock_t *lock = &current->sighand->siglock;
1834
1835 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1836 /*
1837 * We must hold ->siglock while testing q->list
1838 * to serialize with collect_signal() or with
1839 * __exit_signal()->flush_sigqueue().
1840 */
1841 spin_lock_irqsave(lock, flags);
1842 q->flags &= ~SIGQUEUE_PREALLOC;
1843 /*
1844 * If it is queued it will be freed when dequeued,
1845 * like the "regular" sigqueue.
1846 */
1847 if (!list_empty(&q->list))
1848 q = NULL;
1849 spin_unlock_irqrestore(lock, flags);
1850
1851 if (q)
1852 __sigqueue_free(q);
1853 }
1854
1855 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1856 {
1857 int sig = q->info.si_signo;
1858 struct sigpending *pending;
1859 struct task_struct *t;
1860 unsigned long flags;
1861 int ret, result;
1862
1863 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1864
1865 ret = -1;
1866 rcu_read_lock();
1867 t = pid_task(pid, type);
1868 if (!t || !likely(lock_task_sighand(t, &flags)))
1869 goto ret;
1870
1871 ret = 1; /* the signal is ignored */
1872 result = TRACE_SIGNAL_IGNORED;
1873 if (!prepare_signal(sig, t, false))
1874 goto out;
1875
1876 ret = 0;
1877 if (unlikely(!list_empty(&q->list))) {
1878 /*
1879 * If an SI_TIMER entry is already queue just increment
1880 * the overrun count.
1881 */
1882 BUG_ON(q->info.si_code != SI_TIMER);
1883 q->info.si_overrun++;
1884 result = TRACE_SIGNAL_ALREADY_PENDING;
1885 goto out;
1886 }
1887 q->info.si_overrun = 0;
1888
1889 signalfd_notify(t, sig);
1890 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1891 list_add_tail(&q->list, &pending->list);
1892 sigaddset(&pending->signal, sig);
1893 complete_signal(sig, t, type);
1894 result = TRACE_SIGNAL_DELIVERED;
1895 out:
1896 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1897 unlock_task_sighand(t, &flags);
1898 ret:
1899 rcu_read_unlock();
1900 return ret;
1901 }
1902
1903 static void do_notify_pidfd(struct task_struct *task)
1904 {
1905 struct pid *pid;
1906
1907 WARN_ON(task->exit_state == 0);
1908 pid = task_pid(task);
1909 wake_up_all(&pid->wait_pidfd);
1910 }
1911
1912 /*
1913 * Let a parent know about the death of a child.
1914 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1915 *
1916 * Returns true if our parent ignored us and so we've switched to
1917 * self-reaping.
1918 */
1919 bool do_notify_parent(struct task_struct *tsk, int sig)
1920 {
1921 struct kernel_siginfo info;
1922 unsigned long flags;
1923 struct sighand_struct *psig;
1924 bool autoreap = false;
1925 u64 utime, stime;
1926
1927 BUG_ON(sig == -1);
1928
1929 /* do_notify_parent_cldstop should have been called instead. */
1930 BUG_ON(task_is_stopped_or_traced(tsk));
1931
1932 BUG_ON(!tsk->ptrace &&
1933 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1934
1935 /* Wake up all pidfd waiters */
1936 do_notify_pidfd(tsk);
1937
1938 if (sig != SIGCHLD) {
1939 /*
1940 * This is only possible if parent == real_parent.
1941 * Check if it has changed security domain.
1942 */
1943 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
1944 sig = SIGCHLD;
1945 }
1946
1947 clear_siginfo(&info);
1948 info.si_signo = sig;
1949 info.si_errno = 0;
1950 /*
1951 * We are under tasklist_lock here so our parent is tied to
1952 * us and cannot change.
1953 *
1954 * task_active_pid_ns will always return the same pid namespace
1955 * until a task passes through release_task.
1956 *
1957 * write_lock() currently calls preempt_disable() which is the
1958 * same as rcu_read_lock(), but according to Oleg, this is not
1959 * correct to rely on this
1960 */
1961 rcu_read_lock();
1962 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1963 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1964 task_uid(tsk));
1965 rcu_read_unlock();
1966
1967 task_cputime(tsk, &utime, &stime);
1968 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1969 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1970
1971 info.si_status = tsk->exit_code & 0x7f;
1972 if (tsk->exit_code & 0x80)
1973 info.si_code = CLD_DUMPED;
1974 else if (tsk->exit_code & 0x7f)
1975 info.si_code = CLD_KILLED;
1976 else {
1977 info.si_code = CLD_EXITED;
1978 info.si_status = tsk->exit_code >> 8;
1979 }
1980
1981 psig = tsk->parent->sighand;
1982 spin_lock_irqsave(&psig->siglock, flags);
1983 if (!tsk->ptrace && sig == SIGCHLD &&
1984 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1985 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1986 /*
1987 * We are exiting and our parent doesn't care. POSIX.1
1988 * defines special semantics for setting SIGCHLD to SIG_IGN
1989 * or setting the SA_NOCLDWAIT flag: we should be reaped
1990 * automatically and not left for our parent's wait4 call.
1991 * Rather than having the parent do it as a magic kind of
1992 * signal handler, we just set this to tell do_exit that we
1993 * can be cleaned up without becoming a zombie. Note that
1994 * we still call __wake_up_parent in this case, because a
1995 * blocked sys_wait4 might now return -ECHILD.
1996 *
1997 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1998 * is implementation-defined: we do (if you don't want
1999 * it, just use SIG_IGN instead).
2000 */
2001 autoreap = true;
2002 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2003 sig = 0;
2004 }
2005 /*
2006 * Send with __send_signal as si_pid and si_uid are in the
2007 * parent's namespaces.
2008 */
2009 if (valid_signal(sig) && sig)
2010 __send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2011 __wake_up_parent(tsk, tsk->parent);
2012 spin_unlock_irqrestore(&psig->siglock, flags);
2013
2014 return autoreap;
2015 }
2016
2017 /**
2018 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2019 * @tsk: task reporting the state change
2020 * @for_ptracer: the notification is for ptracer
2021 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2022 *
2023 * Notify @tsk's parent that the stopped/continued state has changed. If
2024 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2025 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2026 *
2027 * CONTEXT:
2028 * Must be called with tasklist_lock at least read locked.
2029 */
2030 static void do_notify_parent_cldstop(struct task_struct *tsk,
2031 bool for_ptracer, int why)
2032 {
2033 struct kernel_siginfo info;
2034 unsigned long flags;
2035 struct task_struct *parent;
2036 struct sighand_struct *sighand;
2037 u64 utime, stime;
2038
2039 if (for_ptracer) {
2040 parent = tsk->parent;
2041 } else {
2042 tsk = tsk->group_leader;
2043 parent = tsk->real_parent;
2044 }
2045
2046 clear_siginfo(&info);
2047 info.si_signo = SIGCHLD;
2048 info.si_errno = 0;
2049 /*
2050 * see comment in do_notify_parent() about the following 4 lines
2051 */
2052 rcu_read_lock();
2053 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2054 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2055 rcu_read_unlock();
2056
2057 task_cputime(tsk, &utime, &stime);
2058 info.si_utime = nsec_to_clock_t(utime);
2059 info.si_stime = nsec_to_clock_t(stime);
2060
2061 info.si_code = why;
2062 switch (why) {
2063 case CLD_CONTINUED:
2064 info.si_status = SIGCONT;
2065 break;
2066 case CLD_STOPPED:
2067 info.si_status = tsk->signal->group_exit_code & 0x7f;
2068 break;
2069 case CLD_TRAPPED:
2070 info.si_status = tsk->exit_code & 0x7f;
2071 break;
2072 default:
2073 BUG();
2074 }
2075
2076 sighand = parent->sighand;
2077 spin_lock_irqsave(&sighand->siglock, flags);
2078 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2079 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2080 __group_send_sig_info(SIGCHLD, &info, parent);
2081 /*
2082 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2083 */
2084 __wake_up_parent(tsk, parent);
2085 spin_unlock_irqrestore(&sighand->siglock, flags);
2086 }
2087
2088 static inline bool may_ptrace_stop(void)
2089 {
2090 if (!likely(current->ptrace))
2091 return false;
2092 /*
2093 * Are we in the middle of do_coredump?
2094 * If so and our tracer is also part of the coredump stopping
2095 * is a deadlock situation, and pointless because our tracer
2096 * is dead so don't allow us to stop.
2097 * If SIGKILL was already sent before the caller unlocked
2098 * ->siglock we must see ->core_state != NULL. Otherwise it
2099 * is safe to enter schedule().
2100 *
2101 * This is almost outdated, a task with the pending SIGKILL can't
2102 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2103 * after SIGKILL was already dequeued.
2104 */
2105 if (unlikely(current->mm->core_state) &&
2106 unlikely(current->mm == current->parent->mm))
2107 return false;
2108
2109 return true;
2110 }
2111
2112 /*
2113 * Return non-zero if there is a SIGKILL that should be waking us up.
2114 * Called with the siglock held.
2115 */
2116 static bool sigkill_pending(struct task_struct *tsk)
2117 {
2118 return sigismember(&tsk->pending.signal, SIGKILL) ||
2119 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2120 }
2121
2122 /*
2123 * This must be called with current->sighand->siglock held.
2124 *
2125 * This should be the path for all ptrace stops.
2126 * We always set current->last_siginfo while stopped here.
2127 * That makes it a way to test a stopped process for
2128 * being ptrace-stopped vs being job-control-stopped.
2129 *
2130 * If we actually decide not to stop at all because the tracer
2131 * is gone, we keep current->exit_code unless clear_code.
2132 */
2133 static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2134 __releases(&current->sighand->siglock)
2135 __acquires(&current->sighand->siglock)
2136 {
2137 bool gstop_done = false;
2138
2139 if (arch_ptrace_stop_needed(exit_code, info)) {
2140 /*
2141 * The arch code has something special to do before a
2142 * ptrace stop. This is allowed to block, e.g. for faults
2143 * on user stack pages. We can't keep the siglock while
2144 * calling arch_ptrace_stop, so we must release it now.
2145 * To preserve proper semantics, we must do this before
2146 * any signal bookkeeping like checking group_stop_count.
2147 * Meanwhile, a SIGKILL could come in before we retake the
2148 * siglock. That must prevent us from sleeping in TASK_TRACED.
2149 * So after regaining the lock, we must check for SIGKILL.
2150 */
2151 spin_unlock_irq(&current->sighand->siglock);
2152 arch_ptrace_stop(exit_code, info);
2153 spin_lock_irq(&current->sighand->siglock);
2154 if (sigkill_pending(current))
2155 return;
2156 }
2157
2158 set_special_state(TASK_TRACED);
2159
2160 /*
2161 * We're committing to trapping. TRACED should be visible before
2162 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2163 * Also, transition to TRACED and updates to ->jobctl should be
2164 * atomic with respect to siglock and should be done after the arch
2165 * hook as siglock is released and regrabbed across it.
2166 *
2167 * TRACER TRACEE
2168 *
2169 * ptrace_attach()
2170 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2171 * do_wait()
2172 * set_current_state() smp_wmb();
2173 * ptrace_do_wait()
2174 * wait_task_stopped()
2175 * task_stopped_code()
2176 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2177 */
2178 smp_wmb();
2179
2180 current->last_siginfo = info;
2181 current->exit_code = exit_code;
2182
2183 /*
2184 * If @why is CLD_STOPPED, we're trapping to participate in a group
2185 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2186 * across siglock relocks since INTERRUPT was scheduled, PENDING
2187 * could be clear now. We act as if SIGCONT is received after
2188 * TASK_TRACED is entered - ignore it.
2189 */
2190 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2191 gstop_done = task_participate_group_stop(current);
2192
2193 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2194 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2195 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2196 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2197
2198 /* entering a trap, clear TRAPPING */
2199 task_clear_jobctl_trapping(current);
2200
2201 spin_unlock_irq(&current->sighand->siglock);
2202 read_lock(&tasklist_lock);
2203 if (may_ptrace_stop()) {
2204 /*
2205 * Notify parents of the stop.
2206 *
2207 * While ptraced, there are two parents - the ptracer and
2208 * the real_parent of the group_leader. The ptracer should
2209 * know about every stop while the real parent is only
2210 * interested in the completion of group stop. The states
2211 * for the two don't interact with each other. Notify
2212 * separately unless they're gonna be duplicates.
2213 */
2214 do_notify_parent_cldstop(current, true, why);
2215 if (gstop_done && ptrace_reparented(current))
2216 do_notify_parent_cldstop(current, false, why);
2217
2218 /*
2219 * Don't want to allow preemption here, because
2220 * sys_ptrace() needs this task to be inactive.
2221 *
2222 * XXX: implement read_unlock_no_resched().
2223 */
2224 preempt_disable();
2225 read_unlock(&tasklist_lock);
2226 cgroup_enter_frozen();
2227 preempt_enable_no_resched();
2228 freezable_schedule();
2229 cgroup_leave_frozen(true);
2230 } else {
2231 /*
2232 * By the time we got the lock, our tracer went away.
2233 * Don't drop the lock yet, another tracer may come.
2234 *
2235 * If @gstop_done, the ptracer went away between group stop
2236 * completion and here. During detach, it would have set
2237 * JOBCTL_STOP_PENDING on us and we'll re-enter
2238 * TASK_STOPPED in do_signal_stop() on return, so notifying
2239 * the real parent of the group stop completion is enough.
2240 */
2241 if (gstop_done)
2242 do_notify_parent_cldstop(current, false, why);
2243
2244 /* tasklist protects us from ptrace_freeze_traced() */
2245 __set_current_state(TASK_RUNNING);
2246 if (clear_code)
2247 current->exit_code = 0;
2248 read_unlock(&tasklist_lock);
2249 }
2250
2251 /*
2252 * We are back. Now reacquire the siglock before touching
2253 * last_siginfo, so that we are sure to have synchronized with
2254 * any signal-sending on another CPU that wants to examine it.
2255 */
2256 spin_lock_irq(&current->sighand->siglock);
2257 current->last_siginfo = NULL;
2258
2259 /* LISTENING can be set only during STOP traps, clear it */
2260 current->jobctl &= ~JOBCTL_LISTENING;
2261
2262 /*
2263 * Queued signals ignored us while we were stopped for tracing.
2264 * So check for any that we should take before resuming user mode.
2265 * This sets TIF_SIGPENDING, but never clears it.
2266 */
2267 recalc_sigpending_tsk(current);
2268 }
2269
2270 static void ptrace_do_notify(int signr, int exit_code, int why)
2271 {
2272 kernel_siginfo_t info;
2273
2274 clear_siginfo(&info);
2275 info.si_signo = signr;
2276 info.si_code = exit_code;
2277 info.si_pid = task_pid_vnr(current);
2278 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2279
2280 /* Let the debugger run. */
2281 ptrace_stop(exit_code, why, 1, &info);
2282 }
2283
2284 void ptrace_notify(int exit_code)
2285 {
2286 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2287 if (unlikely(current->task_works))
2288 task_work_run();
2289
2290 spin_lock_irq(&current->sighand->siglock);
2291 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2292 spin_unlock_irq(&current->sighand->siglock);
2293 }
2294
2295 /**
2296 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2297 * @signr: signr causing group stop if initiating
2298 *
2299 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2300 * and participate in it. If already set, participate in the existing
2301 * group stop. If participated in a group stop (and thus slept), %true is
2302 * returned with siglock released.
2303 *
2304 * If ptraced, this function doesn't handle stop itself. Instead,
2305 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2306 * untouched. The caller must ensure that INTERRUPT trap handling takes
2307 * places afterwards.
2308 *
2309 * CONTEXT:
2310 * Must be called with @current->sighand->siglock held, which is released
2311 * on %true return.
2312 *
2313 * RETURNS:
2314 * %false if group stop is already cancelled or ptrace trap is scheduled.
2315 * %true if participated in group stop.
2316 */
2317 static bool do_signal_stop(int signr)
2318 __releases(&current->sighand->siglock)
2319 {
2320 struct signal_struct *sig = current->signal;
2321
2322 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2323 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2324 struct task_struct *t;
2325
2326 /* signr will be recorded in task->jobctl for retries */
2327 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2328
2329 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2330 unlikely(signal_group_exit(sig)))
2331 return false;
2332 /*
2333 * There is no group stop already in progress. We must
2334 * initiate one now.
2335 *
2336 * While ptraced, a task may be resumed while group stop is
2337 * still in effect and then receive a stop signal and
2338 * initiate another group stop. This deviates from the
2339 * usual behavior as two consecutive stop signals can't
2340 * cause two group stops when !ptraced. That is why we
2341 * also check !task_is_stopped(t) below.
2342 *
2343 * The condition can be distinguished by testing whether
2344 * SIGNAL_STOP_STOPPED is already set. Don't generate
2345 * group_exit_code in such case.
2346 *
2347 * This is not necessary for SIGNAL_STOP_CONTINUED because
2348 * an intervening stop signal is required to cause two
2349 * continued events regardless of ptrace.
2350 */
2351 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2352 sig->group_exit_code = signr;
2353
2354 sig->group_stop_count = 0;
2355
2356 if (task_set_jobctl_pending(current, signr | gstop))
2357 sig->group_stop_count++;
2358
2359 t = current;
2360 while_each_thread(current, t) {
2361 /*
2362 * Setting state to TASK_STOPPED for a group
2363 * stop is always done with the siglock held,
2364 * so this check has no races.
2365 */
2366 if (!task_is_stopped(t) &&
2367 task_set_jobctl_pending(t, signr | gstop)) {
2368 sig->group_stop_count++;
2369 if (likely(!(t->ptrace & PT_SEIZED)))
2370 signal_wake_up(t, 0);
2371 else
2372 ptrace_trap_notify(t);
2373 }
2374 }
2375 }
2376
2377 if (likely(!current->ptrace)) {
2378 int notify = 0;
2379
2380 /*
2381 * If there are no other threads in the group, or if there
2382 * is a group stop in progress and we are the last to stop,
2383 * report to the parent.
2384 */
2385 if (task_participate_group_stop(current))
2386 notify = CLD_STOPPED;
2387
2388 set_special_state(TASK_STOPPED);
2389 spin_unlock_irq(&current->sighand->siglock);
2390
2391 /*
2392 * Notify the parent of the group stop completion. Because
2393 * we're not holding either the siglock or tasklist_lock
2394 * here, ptracer may attach inbetween; however, this is for
2395 * group stop and should always be delivered to the real
2396 * parent of the group leader. The new ptracer will get
2397 * its notification when this task transitions into
2398 * TASK_TRACED.
2399 */
2400 if (notify) {
2401 read_lock(&tasklist_lock);
2402 do_notify_parent_cldstop(current, false, notify);
2403 read_unlock(&tasklist_lock);
2404 }
2405
2406 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2407 cgroup_enter_frozen();
2408 freezable_schedule();
2409 return true;
2410 } else {
2411 /*
2412 * While ptraced, group stop is handled by STOP trap.
2413 * Schedule it and let the caller deal with it.
2414 */
2415 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2416 return false;
2417 }
2418 }
2419
2420 /**
2421 * do_jobctl_trap - take care of ptrace jobctl traps
2422 *
2423 * When PT_SEIZED, it's used for both group stop and explicit
2424 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2425 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2426 * the stop signal; otherwise, %SIGTRAP.
2427 *
2428 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2429 * number as exit_code and no siginfo.
2430 *
2431 * CONTEXT:
2432 * Must be called with @current->sighand->siglock held, which may be
2433 * released and re-acquired before returning with intervening sleep.
2434 */
2435 static void do_jobctl_trap(void)
2436 {
2437 struct signal_struct *signal = current->signal;
2438 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2439
2440 if (current->ptrace & PT_SEIZED) {
2441 if (!signal->group_stop_count &&
2442 !(signal->flags & SIGNAL_STOP_STOPPED))
2443 signr = SIGTRAP;
2444 WARN_ON_ONCE(!signr);
2445 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2446 CLD_STOPPED);
2447 } else {
2448 WARN_ON_ONCE(!signr);
2449 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2450 current->exit_code = 0;
2451 }
2452 }
2453
2454 /**
2455 * do_freezer_trap - handle the freezer jobctl trap
2456 *
2457 * Puts the task into frozen state, if only the task is not about to quit.
2458 * In this case it drops JOBCTL_TRAP_FREEZE.
2459 *
2460 * CONTEXT:
2461 * Must be called with @current->sighand->siglock held,
2462 * which is always released before returning.
2463 */
2464 static void do_freezer_trap(void)
2465 __releases(&current->sighand->siglock)
2466 {
2467 /*
2468 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2469 * let's make another loop to give it a chance to be handled.
2470 * In any case, we'll return back.
2471 */
2472 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2473 JOBCTL_TRAP_FREEZE) {
2474 spin_unlock_irq(&current->sighand->siglock);
2475 return;
2476 }
2477
2478 /*
2479 * Now we're sure that there is no pending fatal signal and no
2480 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2481 * immediately (if there is a non-fatal signal pending), and
2482 * put the task into sleep.
2483 */
2484 __set_current_state(TASK_INTERRUPTIBLE);
2485 clear_thread_flag(TIF_SIGPENDING);
2486 spin_unlock_irq(&current->sighand->siglock);
2487 cgroup_enter_frozen();
2488 freezable_schedule();
2489 }
2490
2491 static int ptrace_signal(int signr, kernel_siginfo_t *info)
2492 {
2493 /*
2494 * We do not check sig_kernel_stop(signr) but set this marker
2495 * unconditionally because we do not know whether debugger will
2496 * change signr. This flag has no meaning unless we are going
2497 * to stop after return from ptrace_stop(). In this case it will
2498 * be checked in do_signal_stop(), we should only stop if it was
2499 * not cleared by SIGCONT while we were sleeping. See also the
2500 * comment in dequeue_signal().
2501 */
2502 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2503 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2504
2505 /* We're back. Did the debugger cancel the sig? */
2506 signr = current->exit_code;
2507 if (signr == 0)
2508 return signr;
2509
2510 current->exit_code = 0;
2511
2512 /*
2513 * Update the siginfo structure if the signal has
2514 * changed. If the debugger wanted something
2515 * specific in the siginfo structure then it should
2516 * have updated *info via PTRACE_SETSIGINFO.
2517 */
2518 if (signr != info->si_signo) {
2519 clear_siginfo(info);
2520 info->si_signo = signr;
2521 info->si_errno = 0;
2522 info->si_code = SI_USER;
2523 rcu_read_lock();
2524 info->si_pid = task_pid_vnr(current->parent);
2525 info->si_uid = from_kuid_munged(current_user_ns(),
2526 task_uid(current->parent));
2527 rcu_read_unlock();
2528 }
2529
2530 /* If the (new) signal is now blocked, requeue it. */
2531 if (sigismember(&current->blocked, signr)) {
2532 send_signal(signr, info, current, PIDTYPE_PID);
2533 signr = 0;
2534 }
2535
2536 return signr;
2537 }
2538
2539 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2540 {
2541 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2542 case SIL_FAULT:
2543 case SIL_FAULT_TRAPNO:
2544 case SIL_FAULT_MCEERR:
2545 case SIL_FAULT_BNDERR:
2546 case SIL_FAULT_PKUERR:
2547 case SIL_PERF_EVENT:
2548 ksig->info.si_addr = arch_untagged_si_addr(
2549 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2550 break;
2551 case SIL_KILL:
2552 case SIL_TIMER:
2553 case SIL_POLL:
2554 case SIL_CHLD:
2555 case SIL_RT:
2556 case SIL_SYS:
2557 break;
2558 }
2559 }
2560
2561 bool get_signal(struct ksignal *ksig)
2562 {
2563 struct sighand_struct *sighand = current->sighand;
2564 struct signal_struct *signal = current->signal;
2565 int signr;
2566
2567 if (unlikely(current->task_works))
2568 task_work_run();
2569
2570 /*
2571 * For non-generic architectures, check for TIF_NOTIFY_SIGNAL so
2572 * that the arch handlers don't all have to do it. If we get here
2573 * without TIF_SIGPENDING, just exit after running signal work.
2574 */
2575 if (!IS_ENABLED(CONFIG_GENERIC_ENTRY)) {
2576 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
2577 tracehook_notify_signal();
2578 if (!task_sigpending(current))
2579 return false;
2580 }
2581
2582 if (unlikely(uprobe_deny_signal()))
2583 return false;
2584
2585 /*
2586 * Do this once, we can't return to user-mode if freezing() == T.
2587 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2588 * thus do not need another check after return.
2589 */
2590 try_to_freeze();
2591
2592 relock:
2593 spin_lock_irq(&sighand->siglock);
2594
2595 /*
2596 * Every stopped thread goes here after wakeup. Check to see if
2597 * we should notify the parent, prepare_signal(SIGCONT) encodes
2598 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2599 */
2600 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2601 int why;
2602
2603 if (signal->flags & SIGNAL_CLD_CONTINUED)
2604 why = CLD_CONTINUED;
2605 else
2606 why = CLD_STOPPED;
2607
2608 signal->flags &= ~SIGNAL_CLD_MASK;
2609
2610 spin_unlock_irq(&sighand->siglock);
2611
2612 /*
2613 * Notify the parent that we're continuing. This event is
2614 * always per-process and doesn't make whole lot of sense
2615 * for ptracers, who shouldn't consume the state via
2616 * wait(2) either, but, for backward compatibility, notify
2617 * the ptracer of the group leader too unless it's gonna be
2618 * a duplicate.
2619 */
2620 read_lock(&tasklist_lock);
2621 do_notify_parent_cldstop(current, false, why);
2622
2623 if (ptrace_reparented(current->group_leader))
2624 do_notify_parent_cldstop(current->group_leader,
2625 true, why);
2626 read_unlock(&tasklist_lock);
2627
2628 goto relock;
2629 }
2630
2631 /* Has this task already been marked for death? */
2632 if (signal_group_exit(signal)) {
2633 ksig->info.si_signo = signr = SIGKILL;
2634 sigdelset(&current->pending.signal, SIGKILL);
2635 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2636 &sighand->action[SIGKILL - 1]);
2637 recalc_sigpending();
2638 goto fatal;
2639 }
2640
2641 for (;;) {
2642 struct k_sigaction *ka;
2643
2644 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2645 do_signal_stop(0))
2646 goto relock;
2647
2648 if (unlikely(current->jobctl &
2649 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2650 if (current->jobctl & JOBCTL_TRAP_MASK) {
2651 do_jobctl_trap();
2652 spin_unlock_irq(&sighand->siglock);
2653 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2654 do_freezer_trap();
2655
2656 goto relock;
2657 }
2658
2659 /*
2660 * If the task is leaving the frozen state, let's update
2661 * cgroup counters and reset the frozen bit.
2662 */
2663 if (unlikely(cgroup_task_frozen(current))) {
2664 spin_unlock_irq(&sighand->siglock);
2665 cgroup_leave_frozen(false);
2666 goto relock;
2667 }
2668
2669 /*
2670 * Signals generated by the execution of an instruction
2671 * need to be delivered before any other pending signals
2672 * so that the instruction pointer in the signal stack
2673 * frame points to the faulting instruction.
2674 */
2675 signr = dequeue_synchronous_signal(&ksig->info);
2676 if (!signr)
2677 signr = dequeue_signal(current, &current->blocked, &ksig->info);
2678
2679 if (!signr)
2680 break; /* will return 0 */
2681
2682 if (unlikely(current->ptrace) && signr != SIGKILL) {
2683 signr = ptrace_signal(signr, &ksig->info);
2684 if (!signr)
2685 continue;
2686 }
2687
2688 ka = &sighand->action[signr-1];
2689
2690 /* Trace actually delivered signals. */
2691 trace_signal_deliver(signr, &ksig->info, ka);
2692
2693 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2694 continue;
2695 if (ka->sa.sa_handler != SIG_DFL) {
2696 /* Run the handler. */
2697 ksig->ka = *ka;
2698
2699 if (ka->sa.sa_flags & SA_ONESHOT)
2700 ka->sa.sa_handler = SIG_DFL;
2701
2702 break; /* will return non-zero "signr" value */
2703 }
2704
2705 /*
2706 * Now we are doing the default action for this signal.
2707 */
2708 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2709 continue;
2710
2711 /*
2712 * Global init gets no signals it doesn't want.
2713 * Container-init gets no signals it doesn't want from same
2714 * container.
2715 *
2716 * Note that if global/container-init sees a sig_kernel_only()
2717 * signal here, the signal must have been generated internally
2718 * or must have come from an ancestor namespace. In either
2719 * case, the signal cannot be dropped.
2720 */
2721 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2722 !sig_kernel_only(signr))
2723 continue;
2724
2725 if (sig_kernel_stop(signr)) {
2726 /*
2727 * The default action is to stop all threads in
2728 * the thread group. The job control signals
2729 * do nothing in an orphaned pgrp, but SIGSTOP
2730 * always works. Note that siglock needs to be
2731 * dropped during the call to is_orphaned_pgrp()
2732 * because of lock ordering with tasklist_lock.
2733 * This allows an intervening SIGCONT to be posted.
2734 * We need to check for that and bail out if necessary.
2735 */
2736 if (signr != SIGSTOP) {
2737 spin_unlock_irq(&sighand->siglock);
2738
2739 /* signals can be posted during this window */
2740
2741 if (is_current_pgrp_orphaned())
2742 goto relock;
2743
2744 spin_lock_irq(&sighand->siglock);
2745 }
2746
2747 if (likely(do_signal_stop(ksig->info.si_signo))) {
2748 /* It released the siglock. */
2749 goto relock;
2750 }
2751
2752 /*
2753 * We didn't actually stop, due to a race
2754 * with SIGCONT or something like that.
2755 */
2756 continue;
2757 }
2758
2759 fatal:
2760 spin_unlock_irq(&sighand->siglock);
2761 if (unlikely(cgroup_task_frozen(current)))
2762 cgroup_leave_frozen(true);
2763
2764 /*
2765 * Anything else is fatal, maybe with a core dump.
2766 */
2767 current->flags |= PF_SIGNALED;
2768
2769 if (sig_kernel_coredump(signr)) {
2770 if (print_fatal_signals)
2771 print_fatal_signal(ksig->info.si_signo);
2772 proc_coredump_connector(current);
2773 /*
2774 * If it was able to dump core, this kills all
2775 * other threads in the group and synchronizes with
2776 * their demise. If we lost the race with another
2777 * thread getting here, it set group_exit_code
2778 * first and our do_group_exit call below will use
2779 * that value and ignore the one we pass it.
2780 */
2781 do_coredump(&ksig->info);
2782 }
2783
2784 /*
2785 * PF_IO_WORKER threads will catch and exit on fatal signals
2786 * themselves. They have cleanup that must be performed, so
2787 * we cannot call do_exit() on their behalf.
2788 */
2789 if (current->flags & PF_IO_WORKER)
2790 goto out;
2791
2792 /*
2793 * Death signals, no core dump.
2794 */
2795 do_group_exit(ksig->info.si_signo);
2796 /* NOTREACHED */
2797 }
2798 spin_unlock_irq(&sighand->siglock);
2799 out:
2800 ksig->sig = signr;
2801
2802 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2803 hide_si_addr_tag_bits(ksig);
2804
2805 return ksig->sig > 0;
2806 }
2807
2808 /**
2809 * signal_delivered -
2810 * @ksig: kernel signal struct
2811 * @stepping: nonzero if debugger single-step or block-step in use
2812 *
2813 * This function should be called when a signal has successfully been
2814 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2815 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2816 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2817 */
2818 static void signal_delivered(struct ksignal *ksig, int stepping)
2819 {
2820 sigset_t blocked;
2821
2822 /* A signal was successfully delivered, and the
2823 saved sigmask was stored on the signal frame,
2824 and will be restored by sigreturn. So we can
2825 simply clear the restore sigmask flag. */
2826 clear_restore_sigmask();
2827
2828 sigorsets(&blocked, &current->blocked, &ksig->ka.sa.sa_mask);
2829 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2830 sigaddset(&blocked, ksig->sig);
2831 set_current_blocked(&blocked);
2832 tracehook_signal_handler(stepping);
2833 }
2834
2835 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2836 {
2837 if (failed)
2838 force_sigsegv(ksig->sig);
2839 else
2840 signal_delivered(ksig, stepping);
2841 }
2842
2843 /*
2844 * It could be that complete_signal() picked us to notify about the
2845 * group-wide signal. Other threads should be notified now to take
2846 * the shared signals in @which since we will not.
2847 */
2848 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2849 {
2850 sigset_t retarget;
2851 struct task_struct *t;
2852
2853 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2854 if (sigisemptyset(&retarget))
2855 return;
2856
2857 t = tsk;
2858 while_each_thread(tsk, t) {
2859 if (t->flags & PF_EXITING)
2860 continue;
2861
2862 if (!has_pending_signals(&retarget, &t->blocked))
2863 continue;
2864 /* Remove the signals this thread can handle. */
2865 sigandsets(&retarget, &retarget, &t->blocked);
2866
2867 if (!task_sigpending(t))
2868 signal_wake_up(t, 0);
2869
2870 if (sigisemptyset(&retarget))
2871 break;
2872 }
2873 }
2874
2875 void exit_signals(struct task_struct *tsk)
2876 {
2877 int group_stop = 0;
2878 sigset_t unblocked;
2879
2880 /*
2881 * @tsk is about to have PF_EXITING set - lock out users which
2882 * expect stable threadgroup.
2883 */
2884 cgroup_threadgroup_change_begin(tsk);
2885
2886 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2887 tsk->flags |= PF_EXITING;
2888 cgroup_threadgroup_change_end(tsk);
2889 return;
2890 }
2891
2892 spin_lock_irq(&tsk->sighand->siglock);
2893 /*
2894 * From now this task is not visible for group-wide signals,
2895 * see wants_signal(), do_signal_stop().
2896 */
2897 tsk->flags |= PF_EXITING;
2898
2899 cgroup_threadgroup_change_end(tsk);
2900
2901 if (!task_sigpending(tsk))
2902 goto out;
2903
2904 unblocked = tsk->blocked;
2905 signotset(&unblocked);
2906 retarget_shared_pending(tsk, &unblocked);
2907
2908 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2909 task_participate_group_stop(tsk))
2910 group_stop = CLD_STOPPED;
2911 out:
2912 spin_unlock_irq(&tsk->sighand->siglock);
2913
2914 /*
2915 * If group stop has completed, deliver the notification. This
2916 * should always go to the real parent of the group leader.
2917 */
2918 if (unlikely(group_stop)) {
2919 read_lock(&tasklist_lock);
2920 do_notify_parent_cldstop(tsk, false, group_stop);
2921 read_unlock(&tasklist_lock);
2922 }
2923 }
2924
2925 /*
2926 * System call entry points.
2927 */
2928
2929 /**
2930 * sys_restart_syscall - restart a system call
2931 */
2932 SYSCALL_DEFINE0(restart_syscall)
2933 {
2934 struct restart_block *restart = &current->restart_block;
2935 return restart->fn(restart);
2936 }
2937
2938 long do_no_restart_syscall(struct restart_block *param)
2939 {
2940 return -EINTR;
2941 }
2942
2943 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2944 {
2945 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
2946 sigset_t newblocked;
2947 /* A set of now blocked but previously unblocked signals. */
2948 sigandnsets(&newblocked, newset, &current->blocked);
2949 retarget_shared_pending(tsk, &newblocked);
2950 }
2951 tsk->blocked = *newset;
2952 recalc_sigpending();
2953 }
2954
2955 /**
2956 * set_current_blocked - change current->blocked mask
2957 * @newset: new mask
2958 *
2959 * It is wrong to change ->blocked directly, this helper should be used
2960 * to ensure the process can't miss a shared signal we are going to block.
2961 */
2962 void set_current_blocked(sigset_t *newset)
2963 {
2964 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2965 __set_current_blocked(newset);
2966 }
2967
2968 void __set_current_blocked(const sigset_t *newset)
2969 {
2970 struct task_struct *tsk = current;
2971
2972 /*
2973 * In case the signal mask hasn't changed, there is nothing we need
2974 * to do. The current->blocked shouldn't be modified by other task.
2975 */
2976 if (sigequalsets(&tsk->blocked, newset))
2977 return;
2978
2979 spin_lock_irq(&tsk->sighand->siglock);
2980 __set_task_blocked(tsk, newset);
2981 spin_unlock_irq(&tsk->sighand->siglock);
2982 }
2983
2984 /*
2985 * This is also useful for kernel threads that want to temporarily
2986 * (or permanently) block certain signals.
2987 *
2988 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2989 * interface happily blocks "unblockable" signals like SIGKILL
2990 * and friends.
2991 */
2992 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2993 {
2994 struct task_struct *tsk = current;
2995 sigset_t newset;
2996
2997 /* Lockless, only current can change ->blocked, never from irq */
2998 if (oldset)
2999 *oldset = tsk->blocked;
3000
3001 switch (how) {
3002 case SIG_BLOCK:
3003 sigorsets(&newset, &tsk->blocked, set);
3004 break;
3005 case SIG_UNBLOCK:
3006 sigandnsets(&newset, &tsk->blocked, set);
3007 break;
3008 case SIG_SETMASK:
3009 newset = *set;
3010 break;
3011 default:
3012 return -EINVAL;
3013 }
3014
3015 __set_current_blocked(&newset);
3016 return 0;
3017 }
3018 EXPORT_SYMBOL(sigprocmask);
3019
3020 /*
3021 * The api helps set app-provided sigmasks.
3022 *
3023 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3024 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3025 *
3026 * Note that it does set_restore_sigmask() in advance, so it must be always
3027 * paired with restore_saved_sigmask_unless() before return from syscall.
3028 */
3029 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3030 {
3031 sigset_t kmask;
3032
3033 if (!umask)
3034 return 0;
3035 if (sigsetsize != sizeof(sigset_t))
3036 return -EINVAL;
3037 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3038 return -EFAULT;
3039
3040 set_restore_sigmask();
3041 current->saved_sigmask = current->blocked;
3042 set_current_blocked(&kmask);
3043
3044 return 0;
3045 }
3046
3047 #ifdef CONFIG_COMPAT
3048 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3049 size_t sigsetsize)
3050 {
3051 sigset_t kmask;
3052
3053 if (!umask)
3054 return 0;
3055 if (sigsetsize != sizeof(compat_sigset_t))
3056 return -EINVAL;
3057 if (get_compat_sigset(&kmask, umask))
3058 return -EFAULT;
3059
3060 set_restore_sigmask();
3061 current->saved_sigmask = current->blocked;
3062 set_current_blocked(&kmask);
3063
3064 return 0;
3065 }
3066 #endif
3067
3068 /**
3069 * sys_rt_sigprocmask - change the list of currently blocked signals
3070 * @how: whether to add, remove, or set signals
3071 * @nset: stores pending signals
3072 * @oset: previous value of signal mask if non-null
3073 * @sigsetsize: size of sigset_t type
3074 */
3075 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3076 sigset_t __user *, oset, size_t, sigsetsize)
3077 {
3078 sigset_t old_set, new_set;
3079 int error;
3080
3081 /* XXX: Don't preclude handling different sized sigset_t's. */
3082 if (sigsetsize != sizeof(sigset_t))
3083 return -EINVAL;
3084
3085 old_set = current->blocked;
3086
3087 if (nset) {
3088 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3089 return -EFAULT;
3090 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3091
3092 error = sigprocmask(how, &new_set, NULL);
3093 if (error)
3094 return error;
3095 }
3096
3097 if (oset) {
3098 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3099 return -EFAULT;
3100 }
3101
3102 return 0;
3103 }
3104
3105 #ifdef CONFIG_COMPAT
3106 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3107 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3108 {
3109 sigset_t old_set = current->blocked;
3110
3111 /* XXX: Don't preclude handling different sized sigset_t's. */
3112 if (sigsetsize != sizeof(sigset_t))
3113 return -EINVAL;
3114
3115 if (nset) {
3116 sigset_t new_set;
3117 int error;
3118 if (get_compat_sigset(&new_set, nset))
3119 return -EFAULT;
3120 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3121
3122 error = sigprocmask(how, &new_set, NULL);
3123 if (error)
3124 return error;
3125 }
3126 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3127 }
3128 #endif
3129
3130 static void do_sigpending(sigset_t *set)
3131 {
3132 spin_lock_irq(&current->sighand->siglock);
3133 sigorsets(set, &current->pending.signal,
3134 &current->signal->shared_pending.signal);
3135 spin_unlock_irq(&current->sighand->siglock);
3136
3137 /* Outside the lock because only this thread touches it. */
3138 sigandsets(set, &current->blocked, set);
3139 }
3140
3141 /**
3142 * sys_rt_sigpending - examine a pending signal that has been raised
3143 * while blocked
3144 * @uset: stores pending signals
3145 * @sigsetsize: size of sigset_t type or larger
3146 */
3147 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3148 {
3149 sigset_t set;
3150
3151 if (sigsetsize > sizeof(*uset))
3152 return -EINVAL;
3153
3154 do_sigpending(&set);
3155
3156 if (copy_to_user(uset, &set, sigsetsize))
3157 return -EFAULT;
3158
3159 return 0;
3160 }
3161
3162 #ifdef CONFIG_COMPAT
3163 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3164 compat_size_t, sigsetsize)
3165 {
3166 sigset_t set;
3167
3168 if (sigsetsize > sizeof(*uset))
3169 return -EINVAL;
3170
3171 do_sigpending(&set);
3172
3173 return put_compat_sigset(uset, &set, sigsetsize);
3174 }
3175 #endif
3176
3177 static const struct {
3178 unsigned char limit, layout;
3179 } sig_sicodes[] = {
3180 [SIGILL] = { NSIGILL, SIL_FAULT },
3181 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3182 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3183 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3184 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3185 #if defined(SIGEMT)
3186 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3187 #endif
3188 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3189 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3190 [SIGSYS] = { NSIGSYS, SIL_SYS },
3191 };
3192
3193 static bool known_siginfo_layout(unsigned sig, int si_code)
3194 {
3195 if (si_code == SI_KERNEL)
3196 return true;
3197 else if ((si_code > SI_USER)) {
3198 if (sig_specific_sicodes(sig)) {
3199 if (si_code <= sig_sicodes[sig].limit)
3200 return true;
3201 }
3202 else if (si_code <= NSIGPOLL)
3203 return true;
3204 }
3205 else if (si_code >= SI_DETHREAD)
3206 return true;
3207 else if (si_code == SI_ASYNCNL)
3208 return true;
3209 return false;
3210 }
3211
3212 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3213 {
3214 enum siginfo_layout layout = SIL_KILL;
3215 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3216 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3217 (si_code <= sig_sicodes[sig].limit)) {
3218 layout = sig_sicodes[sig].layout;
3219 /* Handle the exceptions */
3220 if ((sig == SIGBUS) &&
3221 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3222 layout = SIL_FAULT_MCEERR;
3223 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3224 layout = SIL_FAULT_BNDERR;
3225 #ifdef SEGV_PKUERR
3226 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3227 layout = SIL_FAULT_PKUERR;
3228 #endif
3229 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3230 layout = SIL_PERF_EVENT;
3231 #ifdef __ARCH_SI_TRAPNO
3232 else if (layout == SIL_FAULT)
3233 layout = SIL_FAULT_TRAPNO;
3234 #endif
3235 }
3236 else if (si_code <= NSIGPOLL)
3237 layout = SIL_POLL;
3238 } else {
3239 if (si_code == SI_TIMER)
3240 layout = SIL_TIMER;
3241 else if (si_code == SI_SIGIO)
3242 layout = SIL_POLL;
3243 else if (si_code < 0)
3244 layout = SIL_RT;
3245 }
3246 return layout;
3247 }
3248
3249 static inline char __user *si_expansion(const siginfo_t __user *info)
3250 {
3251 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3252 }
3253
3254 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3255 {
3256 char __user *expansion = si_expansion(to);
3257 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3258 return -EFAULT;
3259 if (clear_user(expansion, SI_EXPANSION_SIZE))
3260 return -EFAULT;
3261 return 0;
3262 }
3263
3264 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3265 const siginfo_t __user *from)
3266 {
3267 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3268 char __user *expansion = si_expansion(from);
3269 char buf[SI_EXPANSION_SIZE];
3270 int i;
3271 /*
3272 * An unknown si_code might need more than
3273 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3274 * extra bytes are 0. This guarantees copy_siginfo_to_user
3275 * will return this data to userspace exactly.
3276 */
3277 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3278 return -EFAULT;
3279 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3280 if (buf[i] != 0)
3281 return -E2BIG;
3282 }
3283 }
3284 return 0;
3285 }
3286
3287 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3288 const siginfo_t __user *from)
3289 {
3290 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3291 return -EFAULT;
3292 to->si_signo = signo;
3293 return post_copy_siginfo_from_user(to, from);
3294 }
3295
3296 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3297 {
3298 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3299 return -EFAULT;
3300 return post_copy_siginfo_from_user(to, from);
3301 }
3302
3303 #ifdef CONFIG_COMPAT
3304 /**
3305 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3306 * @to: compat siginfo destination
3307 * @from: kernel siginfo source
3308 *
3309 * Note: This function does not work properly for the SIGCHLD on x32, but
3310 * fortunately it doesn't have to. The only valid callers for this function are
3311 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3312 * The latter does not care because SIGCHLD will never cause a coredump.
3313 */
3314 void copy_siginfo_to_external32(struct compat_siginfo *to,
3315 const struct kernel_siginfo *from)
3316 {
3317 memset(to, 0, sizeof(*to));
3318
3319 to->si_signo = from->si_signo;
3320 to->si_errno = from->si_errno;
3321 to->si_code = from->si_code;
3322 switch(siginfo_layout(from->si_signo, from->si_code)) {
3323 case SIL_KILL:
3324 to->si_pid = from->si_pid;
3325 to->si_uid = from->si_uid;
3326 break;
3327 case SIL_TIMER:
3328 to->si_tid = from->si_tid;
3329 to->si_overrun = from->si_overrun;
3330 to->si_int = from->si_int;
3331 break;
3332 case SIL_POLL:
3333 to->si_band = from->si_band;
3334 to->si_fd = from->si_fd;
3335 break;
3336 case SIL_FAULT:
3337 to->si_addr = ptr_to_compat(from->si_addr);
3338 break;
3339 case SIL_FAULT_TRAPNO:
3340 to->si_addr = ptr_to_compat(from->si_addr);
3341 to->si_trapno = from->si_trapno;
3342 break;
3343 case SIL_FAULT_MCEERR:
3344 to->si_addr = ptr_to_compat(from->si_addr);
3345 to->si_addr_lsb = from->si_addr_lsb;
3346 break;
3347 case SIL_FAULT_BNDERR:
3348 to->si_addr = ptr_to_compat(from->si_addr);
3349 to->si_lower = ptr_to_compat(from->si_lower);
3350 to->si_upper = ptr_to_compat(from->si_upper);
3351 break;
3352 case SIL_FAULT_PKUERR:
3353 to->si_addr = ptr_to_compat(from->si_addr);
3354 to->si_pkey = from->si_pkey;
3355 break;
3356 case SIL_PERF_EVENT:
3357 to->si_addr = ptr_to_compat(from->si_addr);
3358 to->si_perf_data = from->si_perf_data;
3359 to->si_perf_type = from->si_perf_type;
3360 break;
3361 case SIL_CHLD:
3362 to->si_pid = from->si_pid;
3363 to->si_uid = from->si_uid;
3364 to->si_status = from->si_status;
3365 to->si_utime = from->si_utime;
3366 to->si_stime = from->si_stime;
3367 break;
3368 case SIL_RT:
3369 to->si_pid = from->si_pid;
3370 to->si_uid = from->si_uid;
3371 to->si_int = from->si_int;
3372 break;
3373 case SIL_SYS:
3374 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3375 to->si_syscall = from->si_syscall;
3376 to->si_arch = from->si_arch;
3377 break;
3378 }
3379 }
3380
3381 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3382 const struct kernel_siginfo *from)
3383 {
3384 struct compat_siginfo new;
3385
3386 copy_siginfo_to_external32(&new, from);
3387 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3388 return -EFAULT;
3389 return 0;
3390 }
3391
3392 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3393 const struct compat_siginfo *from)
3394 {
3395 clear_siginfo(to);
3396 to->si_signo = from->si_signo;
3397 to->si_errno = from->si_errno;
3398 to->si_code = from->si_code;
3399 switch(siginfo_layout(from->si_signo, from->si_code)) {
3400 case SIL_KILL:
3401 to->si_pid = from->si_pid;
3402 to->si_uid = from->si_uid;
3403 break;
3404 case SIL_TIMER:
3405 to->si_tid = from->si_tid;
3406 to->si_overrun = from->si_overrun;
3407 to->si_int = from->si_int;
3408 break;
3409 case SIL_POLL:
3410 to->si_band = from->si_band;
3411 to->si_fd = from->si_fd;
3412 break;
3413 case SIL_FAULT:
3414 to->si_addr = compat_ptr(from->si_addr);
3415 break;
3416 case SIL_FAULT_TRAPNO:
3417 to->si_addr = compat_ptr(from->si_addr);
3418 to->si_trapno = from->si_trapno;
3419 break;
3420 case SIL_FAULT_MCEERR:
3421 to->si_addr = compat_ptr(from->si_addr);
3422 to->si_addr_lsb = from->si_addr_lsb;
3423 break;
3424 case SIL_FAULT_BNDERR:
3425 to->si_addr = compat_ptr(from->si_addr);
3426 to->si_lower = compat_ptr(from->si_lower);
3427 to->si_upper = compat_ptr(from->si_upper);
3428 break;
3429 case SIL_FAULT_PKUERR:
3430 to->si_addr = compat_ptr(from->si_addr);
3431 to->si_pkey = from->si_pkey;
3432 break;
3433 case SIL_PERF_EVENT:
3434 to->si_addr = compat_ptr(from->si_addr);
3435 to->si_perf_data = from->si_perf_data;
3436 to->si_perf_type = from->si_perf_type;
3437 break;
3438 case SIL_CHLD:
3439 to->si_pid = from->si_pid;
3440 to->si_uid = from->si_uid;
3441 to->si_status = from->si_status;
3442 #ifdef CONFIG_X86_X32_ABI
3443 if (in_x32_syscall()) {
3444 to->si_utime = from->_sifields._sigchld_x32._utime;
3445 to->si_stime = from->_sifields._sigchld_x32._stime;
3446 } else
3447 #endif
3448 {
3449 to->si_utime = from->si_utime;
3450 to->si_stime = from->si_stime;
3451 }
3452 break;
3453 case SIL_RT:
3454 to->si_pid = from->si_pid;
3455 to->si_uid = from->si_uid;
3456 to->si_int = from->si_int;
3457 break;
3458 case SIL_SYS:
3459 to->si_call_addr = compat_ptr(from->si_call_addr);
3460 to->si_syscall = from->si_syscall;
3461 to->si_arch = from->si_arch;
3462 break;
3463 }
3464 return 0;
3465 }
3466
3467 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3468 const struct compat_siginfo __user *ufrom)
3469 {
3470 struct compat_siginfo from;
3471
3472 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3473 return -EFAULT;
3474
3475 from.si_signo = signo;
3476 return post_copy_siginfo_from_user32(to, &from);
3477 }
3478
3479 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3480 const struct compat_siginfo __user *ufrom)
3481 {
3482 struct compat_siginfo from;
3483
3484 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3485 return -EFAULT;
3486
3487 return post_copy_siginfo_from_user32(to, &from);
3488 }
3489 #endif /* CONFIG_COMPAT */
3490
3491 /**
3492 * do_sigtimedwait - wait for queued signals specified in @which
3493 * @which: queued signals to wait for
3494 * @info: if non-null, the signal's siginfo is returned here
3495 * @ts: upper bound on process time suspension
3496 */
3497 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3498 const struct timespec64 *ts)
3499 {
3500 ktime_t *to = NULL, timeout = KTIME_MAX;
3501 struct task_struct *tsk = current;
3502 sigset_t mask = *which;
3503 int sig, ret = 0;
3504
3505 if (ts) {
3506 if (!timespec64_valid(ts))
3507 return -EINVAL;
3508 timeout = timespec64_to_ktime(*ts);
3509 to = &timeout;
3510 }
3511
3512 /*
3513 * Invert the set of allowed signals to get those we want to block.
3514 */
3515 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3516 signotset(&mask);
3517
3518 spin_lock_irq(&tsk->sighand->siglock);
3519 sig = dequeue_signal(tsk, &mask, info);
3520 if (!sig && timeout) {
3521 /*
3522 * None ready, temporarily unblock those we're interested
3523 * while we are sleeping in so that we'll be awakened when
3524 * they arrive. Unblocking is always fine, we can avoid
3525 * set_current_blocked().
3526 */
3527 tsk->real_blocked = tsk->blocked;
3528 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3529 recalc_sigpending();
3530 spin_unlock_irq(&tsk->sighand->siglock);
3531
3532 __set_current_state(TASK_INTERRUPTIBLE);
3533 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3534 HRTIMER_MODE_REL);
3535 spin_lock_irq(&tsk->sighand->siglock);
3536 __set_task_blocked(tsk, &tsk->real_blocked);
3537 sigemptyset(&tsk->real_blocked);
3538 sig = dequeue_signal(tsk, &mask, info);
3539 }
3540 spin_unlock_irq(&tsk->sighand->siglock);
3541
3542 if (sig)
3543 return sig;
3544 return ret ? -EINTR : -EAGAIN;
3545 }
3546
3547 /**
3548 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3549 * in @uthese
3550 * @uthese: queued signals to wait for
3551 * @uinfo: if non-null, the signal's siginfo is returned here
3552 * @uts: upper bound on process time suspension
3553 * @sigsetsize: size of sigset_t type
3554 */
3555 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3556 siginfo_t __user *, uinfo,
3557 const struct __kernel_timespec __user *, uts,
3558 size_t, sigsetsize)
3559 {
3560 sigset_t these;
3561 struct timespec64 ts;
3562 kernel_siginfo_t info;
3563 int ret;
3564
3565 /* XXX: Don't preclude handling different sized sigset_t's. */
3566 if (sigsetsize != sizeof(sigset_t))
3567 return -EINVAL;
3568
3569 if (copy_from_user(&these, uthese, sizeof(these)))
3570 return -EFAULT;
3571
3572 if (uts) {
3573 if (get_timespec64(&ts, uts))
3574 return -EFAULT;
3575 }
3576
3577 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3578
3579 if (ret > 0 && uinfo) {
3580 if (copy_siginfo_to_user(uinfo, &info))
3581 ret = -EFAULT;
3582 }
3583
3584 return ret;
3585 }
3586
3587 #ifdef CONFIG_COMPAT_32BIT_TIME
3588 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3589 siginfo_t __user *, uinfo,
3590 const struct old_timespec32 __user *, uts,
3591 size_t, sigsetsize)
3592 {
3593 sigset_t these;
3594 struct timespec64 ts;
3595 kernel_siginfo_t info;
3596 int ret;
3597
3598 if (sigsetsize != sizeof(sigset_t))
3599 return -EINVAL;
3600
3601 if (copy_from_user(&these, uthese, sizeof(these)))
3602 return -EFAULT;
3603
3604 if (uts) {
3605 if (get_old_timespec32(&ts, uts))
3606 return -EFAULT;
3607 }
3608
3609 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3610
3611 if (ret > 0 && uinfo) {
3612 if (copy_siginfo_to_user(uinfo, &info))
3613 ret = -EFAULT;
3614 }
3615
3616 return ret;
3617 }
3618 #endif
3619
3620 #ifdef CONFIG_COMPAT
3621 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3622 struct compat_siginfo __user *, uinfo,
3623 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3624 {
3625 sigset_t s;
3626 struct timespec64 t;
3627 kernel_siginfo_t info;
3628 long ret;
3629
3630 if (sigsetsize != sizeof(sigset_t))
3631 return -EINVAL;
3632
3633 if (get_compat_sigset(&s, uthese))
3634 return -EFAULT;
3635
3636 if (uts) {
3637 if (get_timespec64(&t, uts))
3638 return -EFAULT;
3639 }
3640
3641 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3642
3643 if (ret > 0 && uinfo) {
3644 if (copy_siginfo_to_user32(uinfo, &info))
3645 ret = -EFAULT;
3646 }
3647
3648 return ret;
3649 }
3650
3651 #ifdef CONFIG_COMPAT_32BIT_TIME
3652 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3653 struct compat_siginfo __user *, uinfo,
3654 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3655 {
3656 sigset_t s;
3657 struct timespec64 t;
3658 kernel_siginfo_t info;
3659 long ret;
3660
3661 if (sigsetsize != sizeof(sigset_t))
3662 return -EINVAL;
3663
3664 if (get_compat_sigset(&s, uthese))
3665 return -EFAULT;
3666
3667 if (uts) {
3668 if (get_old_timespec32(&t, uts))
3669 return -EFAULT;
3670 }
3671
3672 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3673
3674 if (ret > 0 && uinfo) {
3675 if (copy_siginfo_to_user32(uinfo, &info))
3676 ret = -EFAULT;
3677 }
3678
3679 return ret;
3680 }
3681 #endif
3682 #endif
3683
3684 static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3685 {
3686 clear_siginfo(info);
3687 info->si_signo = sig;
3688 info->si_errno = 0;
3689 info->si_code = SI_USER;
3690 info->si_pid = task_tgid_vnr(current);
3691 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3692 }
3693
3694 /**
3695 * sys_kill - send a signal to a process
3696 * @pid: the PID of the process
3697 * @sig: signal to be sent
3698 */
3699 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3700 {
3701 struct kernel_siginfo info;
3702
3703 prepare_kill_siginfo(sig, &info);
3704
3705 return kill_something_info(sig, &info, pid);
3706 }
3707
3708 /*
3709 * Verify that the signaler and signalee either are in the same pid namespace
3710 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3711 * namespace.
3712 */
3713 static bool access_pidfd_pidns(struct pid *pid)
3714 {
3715 struct pid_namespace *active = task_active_pid_ns(current);
3716 struct pid_namespace *p = ns_of_pid(pid);
3717
3718 for (;;) {
3719 if (!p)
3720 return false;
3721 if (p == active)
3722 break;
3723 p = p->parent;
3724 }
3725
3726 return true;
3727 }
3728
3729 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3730 siginfo_t __user *info)
3731 {
3732 #ifdef CONFIG_COMPAT
3733 /*
3734 * Avoid hooking up compat syscalls and instead handle necessary
3735 * conversions here. Note, this is a stop-gap measure and should not be
3736 * considered a generic solution.
3737 */
3738 if (in_compat_syscall())
3739 return copy_siginfo_from_user32(
3740 kinfo, (struct compat_siginfo __user *)info);
3741 #endif
3742 return copy_siginfo_from_user(kinfo, info);
3743 }
3744
3745 static struct pid *pidfd_to_pid(const struct file *file)
3746 {
3747 struct pid *pid;
3748
3749 pid = pidfd_pid(file);
3750 if (!IS_ERR(pid))
3751 return pid;
3752
3753 return tgid_pidfd_to_pid(file);
3754 }
3755
3756 /**
3757 * sys_pidfd_send_signal - Signal a process through a pidfd
3758 * @pidfd: file descriptor of the process
3759 * @sig: signal to send
3760 * @info: signal info
3761 * @flags: future flags
3762 *
3763 * The syscall currently only signals via PIDTYPE_PID which covers
3764 * kill(<positive-pid>, <signal>. It does not signal threads or process
3765 * groups.
3766 * In order to extend the syscall to threads and process groups the @flags
3767 * argument should be used. In essence, the @flags argument will determine
3768 * what is signaled and not the file descriptor itself. Put in other words,
3769 * grouping is a property of the flags argument not a property of the file
3770 * descriptor.
3771 *
3772 * Return: 0 on success, negative errno on failure
3773 */
3774 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3775 siginfo_t __user *, info, unsigned int, flags)
3776 {
3777 int ret;
3778 struct fd f;
3779 struct pid *pid;
3780 kernel_siginfo_t kinfo;
3781
3782 /* Enforce flags be set to 0 until we add an extension. */
3783 if (flags)
3784 return -EINVAL;
3785
3786 f = fdget(pidfd);
3787 if (!f.file)
3788 return -EBADF;
3789
3790 /* Is this a pidfd? */
3791 pid = pidfd_to_pid(f.file);
3792 if (IS_ERR(pid)) {
3793 ret = PTR_ERR(pid);
3794 goto err;
3795 }
3796
3797 ret = -EINVAL;
3798 if (!access_pidfd_pidns(pid))
3799 goto err;
3800
3801 if (info) {
3802 ret = copy_siginfo_from_user_any(&kinfo, info);
3803 if (unlikely(ret))
3804 goto err;
3805
3806 ret = -EINVAL;
3807 if (unlikely(sig != kinfo.si_signo))
3808 goto err;
3809
3810 /* Only allow sending arbitrary signals to yourself. */
3811 ret = -EPERM;
3812 if ((task_pid(current) != pid) &&
3813 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3814 goto err;
3815 } else {
3816 prepare_kill_siginfo(sig, &kinfo);
3817 }
3818
3819 ret = kill_pid_info(sig, &kinfo, pid);
3820
3821 err:
3822 fdput(f);
3823 return ret;
3824 }
3825
3826 static int
3827 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3828 {
3829 struct task_struct *p;
3830 int error = -ESRCH;
3831
3832 rcu_read_lock();
3833 p = find_task_by_vpid(pid);
3834 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3835 error = check_kill_permission(sig, info, p);
3836 /*
3837 * The null signal is a permissions and process existence
3838 * probe. No signal is actually delivered.
3839 */
3840 if (!error && sig) {
3841 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3842 /*
3843 * If lock_task_sighand() failed we pretend the task
3844 * dies after receiving the signal. The window is tiny,
3845 * and the signal is private anyway.
3846 */
3847 if (unlikely(error == -ESRCH))
3848 error = 0;
3849 }
3850 }
3851 rcu_read_unlock();
3852
3853 return error;
3854 }
3855
3856 static int do_tkill(pid_t tgid, pid_t pid, int sig)
3857 {
3858 struct kernel_siginfo info;
3859
3860 clear_siginfo(&info);
3861 info.si_signo = sig;
3862 info.si_errno = 0;
3863 info.si_code = SI_TKILL;
3864 info.si_pid = task_tgid_vnr(current);
3865 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3866
3867 return do_send_specific(tgid, pid, sig, &info);
3868 }
3869
3870 /**
3871 * sys_tgkill - send signal to one specific thread
3872 * @tgid: the thread group ID of the thread
3873 * @pid: the PID of the thread
3874 * @sig: signal to be sent
3875 *
3876 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3877 * exists but it's not belonging to the target process anymore. This
3878 * method solves the problem of threads exiting and PIDs getting reused.
3879 */
3880 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3881 {
3882 /* This is only valid for single tasks */
3883 if (pid <= 0 || tgid <= 0)
3884 return -EINVAL;
3885
3886 return do_tkill(tgid, pid, sig);
3887 }
3888
3889 /**
3890 * sys_tkill - send signal to one specific task
3891 * @pid: the PID of the task
3892 * @sig: signal to be sent
3893 *
3894 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3895 */
3896 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3897 {
3898 /* This is only valid for single tasks */
3899 if (pid <= 0)
3900 return -EINVAL;
3901
3902 return do_tkill(0, pid, sig);
3903 }
3904
3905 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3906 {
3907 /* Not even root can pretend to send signals from the kernel.
3908 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3909 */
3910 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3911 (task_pid_vnr(current) != pid))
3912 return -EPERM;
3913
3914 /* POSIX.1b doesn't mention process groups. */
3915 return kill_proc_info(sig, info, pid);
3916 }
3917
3918 /**
3919 * sys_rt_sigqueueinfo - send signal information to a signal
3920 * @pid: the PID of the thread
3921 * @sig: signal to be sent
3922 * @uinfo: signal info to be sent
3923 */
3924 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3925 siginfo_t __user *, uinfo)
3926 {
3927 kernel_siginfo_t info;
3928 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3929 if (unlikely(ret))
3930 return ret;
3931 return do_rt_sigqueueinfo(pid, sig, &info);
3932 }
3933
3934 #ifdef CONFIG_COMPAT
3935 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3936 compat_pid_t, pid,
3937 int, sig,
3938 struct compat_siginfo __user *, uinfo)
3939 {
3940 kernel_siginfo_t info;
3941 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3942 if (unlikely(ret))
3943 return ret;
3944 return do_rt_sigqueueinfo(pid, sig, &info);
3945 }
3946 #endif
3947
3948 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3949 {
3950 /* This is only valid for single tasks */
3951 if (pid <= 0 || tgid <= 0)
3952 return -EINVAL;
3953
3954 /* Not even root can pretend to send signals from the kernel.
3955 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3956 */
3957 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3958 (task_pid_vnr(current) != pid))
3959 return -EPERM;
3960
3961 return do_send_specific(tgid, pid, sig, info);
3962 }
3963
3964 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3965 siginfo_t __user *, uinfo)
3966 {
3967 kernel_siginfo_t info;
3968 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3969 if (unlikely(ret))
3970 return ret;
3971 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3972 }
3973
3974 #ifdef CONFIG_COMPAT
3975 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3976 compat_pid_t, tgid,
3977 compat_pid_t, pid,
3978 int, sig,
3979 struct compat_siginfo __user *, uinfo)
3980 {
3981 kernel_siginfo_t info;
3982 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3983 if (unlikely(ret))
3984 return ret;
3985 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3986 }
3987 #endif
3988
3989 /*
3990 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
3991 */
3992 void kernel_sigaction(int sig, __sighandler_t action)
3993 {
3994 spin_lock_irq(&current->sighand->siglock);
3995 current->sighand->action[sig - 1].sa.sa_handler = action;
3996 if (action == SIG_IGN) {
3997 sigset_t mask;
3998
3999 sigemptyset(&mask);
4000 sigaddset(&mask, sig);
4001
4002 flush_sigqueue_mask(&mask, &current->signal->shared_pending);
4003 flush_sigqueue_mask(&mask, &current->pending);
4004 recalc_sigpending();
4005 }
4006 spin_unlock_irq(&current->sighand->siglock);
4007 }
4008 EXPORT_SYMBOL(kernel_sigaction);
4009
4010 void __weak sigaction_compat_abi(struct k_sigaction *act,
4011 struct k_sigaction *oact)
4012 {
4013 }
4014
4015 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4016 {
4017 struct task_struct *p = current, *t;
4018 struct k_sigaction *k;
4019 sigset_t mask;
4020
4021 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4022 return -EINVAL;
4023
4024 k = &p->sighand->action[sig-1];
4025
4026 spin_lock_irq(&p->sighand->siglock);
4027 if (oact)
4028 *oact = *k;
4029
4030 /*
4031 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4032 * e.g. by having an architecture use the bit in their uapi.
4033 */
4034 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4035
4036 /*
4037 * Clear unknown flag bits in order to allow userspace to detect missing
4038 * support for flag bits and to allow the kernel to use non-uapi bits
4039 * internally.
4040 */
4041 if (act)
4042 act->sa.sa_flags &= UAPI_SA_FLAGS;
4043 if (oact)
4044 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4045
4046 sigaction_compat_abi(act, oact);
4047
4048 if (act) {
4049 sigdelsetmask(&act->sa.sa_mask,
4050 sigmask(SIGKILL) | sigmask(SIGSTOP));
4051 *k = *act;
4052 /*
4053 * POSIX 3.3.1.3:
4054 * "Setting a signal action to SIG_IGN for a signal that is
4055 * pending shall cause the pending signal to be discarded,
4056 * whether or not it is blocked."
4057 *
4058 * "Setting a signal action to SIG_DFL for a signal that is
4059 * pending and whose default action is to ignore the signal
4060 * (for example, SIGCHLD), shall cause the pending signal to
4061 * be discarded, whether or not it is blocked"
4062 */
4063 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4064 sigemptyset(&mask);
4065 sigaddset(&mask, sig);
4066 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4067 for_each_thread(p, t)
4068 flush_sigqueue_mask(&mask, &t->pending);
4069 }
4070 }
4071
4072 spin_unlock_irq(&p->sighand->siglock);
4073 return 0;
4074 }
4075
4076 static int
4077 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4078 size_t min_ss_size)
4079 {
4080 struct task_struct *t = current;
4081
4082 if (oss) {
4083 memset(oss, 0, sizeof(stack_t));
4084 oss->ss_sp = (void __user *) t->sas_ss_sp;
4085 oss->ss_size = t->sas_ss_size;
4086 oss->ss_flags = sas_ss_flags(sp) |
4087 (current->sas_ss_flags & SS_FLAG_BITS);
4088 }
4089
4090 if (ss) {
4091 void __user *ss_sp = ss->ss_sp;
4092 size_t ss_size = ss->ss_size;
4093 unsigned ss_flags = ss->ss_flags;
4094 int ss_mode;
4095
4096 if (unlikely(on_sig_stack(sp)))
4097 return -EPERM;
4098
4099 ss_mode = ss_flags & ~SS_FLAG_BITS;
4100 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4101 ss_mode != 0))
4102 return -EINVAL;
4103
4104 if (ss_mode == SS_DISABLE) {
4105 ss_size = 0;
4106 ss_sp = NULL;
4107 } else {
4108 if (unlikely(ss_size < min_ss_size))
4109 return -ENOMEM;
4110 }
4111
4112 t->sas_ss_sp = (unsigned long) ss_sp;
4113 t->sas_ss_size = ss_size;
4114 t->sas_ss_flags = ss_flags;
4115 }
4116 return 0;
4117 }
4118
4119 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4120 {
4121 stack_t new, old;
4122 int err;
4123 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4124 return -EFAULT;
4125 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4126 current_user_stack_pointer(),
4127 MINSIGSTKSZ);
4128 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4129 err = -EFAULT;
4130 return err;
4131 }
4132
4133 int restore_altstack(const stack_t __user *uss)
4134 {
4135 stack_t new;
4136 if (copy_from_user(&new, uss, sizeof(stack_t)))
4137 return -EFAULT;
4138 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4139 MINSIGSTKSZ);
4140 /* squash all but EFAULT for now */
4141 return 0;
4142 }
4143
4144 int __save_altstack(stack_t __user *uss, unsigned long sp)
4145 {
4146 struct task_struct *t = current;
4147 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4148 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4149 __put_user(t->sas_ss_size, &uss->ss_size);
4150 if (err)
4151 return err;
4152 if (t->sas_ss_flags & SS_AUTODISARM)
4153 sas_ss_reset(t);
4154 return 0;
4155 }
4156
4157 #ifdef CONFIG_COMPAT
4158 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4159 compat_stack_t __user *uoss_ptr)
4160 {
4161 stack_t uss, uoss;
4162 int ret;
4163
4164 if (uss_ptr) {
4165 compat_stack_t uss32;
4166 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4167 return -EFAULT;
4168 uss.ss_sp = compat_ptr(uss32.ss_sp);
4169 uss.ss_flags = uss32.ss_flags;
4170 uss.ss_size = uss32.ss_size;
4171 }
4172 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4173 compat_user_stack_pointer(),
4174 COMPAT_MINSIGSTKSZ);
4175 if (ret >= 0 && uoss_ptr) {
4176 compat_stack_t old;
4177 memset(&old, 0, sizeof(old));
4178 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4179 old.ss_flags = uoss.ss_flags;
4180 old.ss_size = uoss.ss_size;
4181 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4182 ret = -EFAULT;
4183 }
4184 return ret;
4185 }
4186
4187 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4188 const compat_stack_t __user *, uss_ptr,
4189 compat_stack_t __user *, uoss_ptr)
4190 {
4191 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4192 }
4193
4194 int compat_restore_altstack(const compat_stack_t __user *uss)
4195 {
4196 int err = do_compat_sigaltstack(uss, NULL);
4197 /* squash all but -EFAULT for now */
4198 return err == -EFAULT ? err : 0;
4199 }
4200
4201 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4202 {
4203 int err;
4204 struct task_struct *t = current;
4205 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4206 &uss->ss_sp) |
4207 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4208 __put_user(t->sas_ss_size, &uss->ss_size);
4209 if (err)
4210 return err;
4211 if (t->sas_ss_flags & SS_AUTODISARM)
4212 sas_ss_reset(t);
4213 return 0;
4214 }
4215 #endif
4216
4217 #ifdef __ARCH_WANT_SYS_SIGPENDING
4218
4219 /**
4220 * sys_sigpending - examine pending signals
4221 * @uset: where mask of pending signal is returned
4222 */
4223 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4224 {
4225 sigset_t set;
4226
4227 if (sizeof(old_sigset_t) > sizeof(*uset))
4228 return -EINVAL;
4229
4230 do_sigpending(&set);
4231
4232 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4233 return -EFAULT;
4234
4235 return 0;
4236 }
4237
4238 #ifdef CONFIG_COMPAT
4239 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4240 {
4241 sigset_t set;
4242
4243 do_sigpending(&set);
4244
4245 return put_user(set.sig[0], set32);
4246 }
4247 #endif
4248
4249 #endif
4250
4251 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4252 /**
4253 * sys_sigprocmask - examine and change blocked signals
4254 * @how: whether to add, remove, or set signals
4255 * @nset: signals to add or remove (if non-null)
4256 * @oset: previous value of signal mask if non-null
4257 *
4258 * Some platforms have their own version with special arguments;
4259 * others support only sys_rt_sigprocmask.
4260 */
4261
4262 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4263 old_sigset_t __user *, oset)
4264 {
4265 old_sigset_t old_set, new_set;
4266 sigset_t new_blocked;
4267
4268 old_set = current->blocked.sig[0];
4269
4270 if (nset) {
4271 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4272 return -EFAULT;
4273
4274 new_blocked = current->blocked;
4275
4276 switch (how) {
4277 case SIG_BLOCK:
4278 sigaddsetmask(&new_blocked, new_set);
4279 break;
4280 case SIG_UNBLOCK:
4281 sigdelsetmask(&new_blocked, new_set);
4282 break;
4283 case SIG_SETMASK:
4284 new_blocked.sig[0] = new_set;
4285 break;
4286 default:
4287 return -EINVAL;
4288 }
4289
4290 set_current_blocked(&new_blocked);
4291 }
4292
4293 if (oset) {
4294 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4295 return -EFAULT;
4296 }
4297
4298 return 0;
4299 }
4300 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4301
4302 #ifndef CONFIG_ODD_RT_SIGACTION
4303 /**
4304 * sys_rt_sigaction - alter an action taken by a process
4305 * @sig: signal to be sent
4306 * @act: new sigaction
4307 * @oact: used to save the previous sigaction
4308 * @sigsetsize: size of sigset_t type
4309 */
4310 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4311 const struct sigaction __user *, act,
4312 struct sigaction __user *, oact,
4313 size_t, sigsetsize)
4314 {
4315 struct k_sigaction new_sa, old_sa;
4316 int ret;
4317
4318 /* XXX: Don't preclude handling different sized sigset_t's. */
4319 if (sigsetsize != sizeof(sigset_t))
4320 return -EINVAL;
4321
4322 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4323 return -EFAULT;
4324
4325 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4326 if (ret)
4327 return ret;
4328
4329 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4330 return -EFAULT;
4331
4332 return 0;
4333 }
4334 #ifdef CONFIG_COMPAT
4335 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4336 const struct compat_sigaction __user *, act,
4337 struct compat_sigaction __user *, oact,
4338 compat_size_t, sigsetsize)
4339 {
4340 struct k_sigaction new_ka, old_ka;
4341 #ifdef __ARCH_HAS_SA_RESTORER
4342 compat_uptr_t restorer;
4343 #endif
4344 int ret;
4345
4346 /* XXX: Don't preclude handling different sized sigset_t's. */
4347 if (sigsetsize != sizeof(compat_sigset_t))
4348 return -EINVAL;
4349
4350 if (act) {
4351 compat_uptr_t handler;
4352 ret = get_user(handler, &act->sa_handler);
4353 new_ka.sa.sa_handler = compat_ptr(handler);
4354 #ifdef __ARCH_HAS_SA_RESTORER
4355 ret |= get_user(restorer, &act->sa_restorer);
4356 new_ka.sa.sa_restorer = compat_ptr(restorer);
4357 #endif
4358 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4359 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4360 if (ret)
4361 return -EFAULT;
4362 }
4363
4364 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4365 if (!ret && oact) {
4366 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4367 &oact->sa_handler);
4368 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4369 sizeof(oact->sa_mask));
4370 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4371 #ifdef __ARCH_HAS_SA_RESTORER
4372 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4373 &oact->sa_restorer);
4374 #endif
4375 }
4376 return ret;
4377 }
4378 #endif
4379 #endif /* !CONFIG_ODD_RT_SIGACTION */
4380
4381 #ifdef CONFIG_OLD_SIGACTION
4382 SYSCALL_DEFINE3(sigaction, int, sig,
4383 const struct old_sigaction __user *, act,
4384 struct old_sigaction __user *, oact)
4385 {
4386 struct k_sigaction new_ka, old_ka;
4387 int ret;
4388
4389 if (act) {
4390 old_sigset_t mask;
4391 if (!access_ok(act, sizeof(*act)) ||
4392 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4393 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4394 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4395 __get_user(mask, &act->sa_mask))
4396 return -EFAULT;
4397 #ifdef __ARCH_HAS_KA_RESTORER
4398 new_ka.ka_restorer = NULL;
4399 #endif
4400 siginitset(&new_ka.sa.sa_mask, mask);
4401 }
4402
4403 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4404
4405 if (!ret && oact) {
4406 if (!access_ok(oact, sizeof(*oact)) ||
4407 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4408 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4409 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4410 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4411 return -EFAULT;
4412 }
4413
4414 return ret;
4415 }
4416 #endif
4417 #ifdef CONFIG_COMPAT_OLD_SIGACTION
4418 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4419 const struct compat_old_sigaction __user *, act,
4420 struct compat_old_sigaction __user *, oact)
4421 {
4422 struct k_sigaction new_ka, old_ka;
4423 int ret;
4424 compat_old_sigset_t mask;
4425 compat_uptr_t handler, restorer;
4426
4427 if (act) {
4428 if (!access_ok(act, sizeof(*act)) ||
4429 __get_user(handler, &act->sa_handler) ||
4430 __get_user(restorer, &act->sa_restorer) ||
4431 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4432 __get_user(mask, &act->sa_mask))
4433 return -EFAULT;
4434
4435 #ifdef __ARCH_HAS_KA_RESTORER
4436 new_ka.ka_restorer = NULL;
4437 #endif
4438 new_ka.sa.sa_handler = compat_ptr(handler);
4439 new_ka.sa.sa_restorer = compat_ptr(restorer);
4440 siginitset(&new_ka.sa.sa_mask, mask);
4441 }
4442
4443 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4444
4445 if (!ret && oact) {
4446 if (!access_ok(oact, sizeof(*oact)) ||
4447 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4448 &oact->sa_handler) ||
4449 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4450 &oact->sa_restorer) ||
4451 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4452 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4453 return -EFAULT;
4454 }
4455 return ret;
4456 }
4457 #endif
4458
4459 #ifdef CONFIG_SGETMASK_SYSCALL
4460
4461 /*
4462 * For backwards compatibility. Functionality superseded by sigprocmask.
4463 */
4464 SYSCALL_DEFINE0(sgetmask)
4465 {
4466 /* SMP safe */
4467 return current->blocked.sig[0];
4468 }
4469
4470 SYSCALL_DEFINE1(ssetmask, int, newmask)
4471 {
4472 int old = current->blocked.sig[0];
4473 sigset_t newset;
4474
4475 siginitset(&newset, newmask);
4476 set_current_blocked(&newset);
4477
4478 return old;
4479 }
4480 #endif /* CONFIG_SGETMASK_SYSCALL */
4481
4482 #ifdef __ARCH_WANT_SYS_SIGNAL
4483 /*
4484 * For backwards compatibility. Functionality superseded by sigaction.
4485 */
4486 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4487 {
4488 struct k_sigaction new_sa, old_sa;
4489 int ret;
4490
4491 new_sa.sa.sa_handler = handler;
4492 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4493 sigemptyset(&new_sa.sa.sa_mask);
4494
4495 ret = do_sigaction(sig, &new_sa, &old_sa);
4496
4497 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4498 }
4499 #endif /* __ARCH_WANT_SYS_SIGNAL */
4500
4501 #ifdef __ARCH_WANT_SYS_PAUSE
4502
4503 SYSCALL_DEFINE0(pause)
4504 {
4505 while (!signal_pending(current)) {
4506 __set_current_state(TASK_INTERRUPTIBLE);
4507 schedule();
4508 }
4509 return -ERESTARTNOHAND;
4510 }
4511
4512 #endif
4513
4514 static int sigsuspend(sigset_t *set)
4515 {
4516 current->saved_sigmask = current->blocked;
4517 set_current_blocked(set);
4518
4519 while (!signal_pending(current)) {
4520 __set_current_state(TASK_INTERRUPTIBLE);
4521 schedule();
4522 }
4523 set_restore_sigmask();
4524 return -ERESTARTNOHAND;
4525 }
4526
4527 /**
4528 * sys_rt_sigsuspend - replace the signal mask for a value with the
4529 * @unewset value until a signal is received
4530 * @unewset: new signal mask value
4531 * @sigsetsize: size of sigset_t type
4532 */
4533 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4534 {
4535 sigset_t newset;
4536
4537 /* XXX: Don't preclude handling different sized sigset_t's. */
4538 if (sigsetsize != sizeof(sigset_t))
4539 return -EINVAL;
4540
4541 if (copy_from_user(&newset, unewset, sizeof(newset)))
4542 return -EFAULT;
4543 return sigsuspend(&newset);
4544 }
4545
4546 #ifdef CONFIG_COMPAT
4547 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4548 {
4549 sigset_t newset;
4550
4551 /* XXX: Don't preclude handling different sized sigset_t's. */
4552 if (sigsetsize != sizeof(sigset_t))
4553 return -EINVAL;
4554
4555 if (get_compat_sigset(&newset, unewset))
4556 return -EFAULT;
4557 return sigsuspend(&newset);
4558 }
4559 #endif
4560
4561 #ifdef CONFIG_OLD_SIGSUSPEND
4562 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4563 {
4564 sigset_t blocked;
4565 siginitset(&blocked, mask);
4566 return sigsuspend(&blocked);
4567 }
4568 #endif
4569 #ifdef CONFIG_OLD_SIGSUSPEND3
4570 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4571 {
4572 sigset_t blocked;
4573 siginitset(&blocked, mask);
4574 return sigsuspend(&blocked);
4575 }
4576 #endif
4577
4578 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4579 {
4580 return NULL;
4581 }
4582
4583 static inline void siginfo_buildtime_checks(void)
4584 {
4585 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4586
4587 /* Verify the offsets in the two siginfos match */
4588 #define CHECK_OFFSET(field) \
4589 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4590
4591 /* kill */
4592 CHECK_OFFSET(si_pid);
4593 CHECK_OFFSET(si_uid);
4594
4595 /* timer */
4596 CHECK_OFFSET(si_tid);
4597 CHECK_OFFSET(si_overrun);
4598 CHECK_OFFSET(si_value);
4599
4600 /* rt */
4601 CHECK_OFFSET(si_pid);
4602 CHECK_OFFSET(si_uid);
4603 CHECK_OFFSET(si_value);
4604
4605 /* sigchld */
4606 CHECK_OFFSET(si_pid);
4607 CHECK_OFFSET(si_uid);
4608 CHECK_OFFSET(si_status);
4609 CHECK_OFFSET(si_utime);
4610 CHECK_OFFSET(si_stime);
4611
4612 /* sigfault */
4613 CHECK_OFFSET(si_addr);
4614 CHECK_OFFSET(si_trapno);
4615 CHECK_OFFSET(si_addr_lsb);
4616 CHECK_OFFSET(si_lower);
4617 CHECK_OFFSET(si_upper);
4618 CHECK_OFFSET(si_pkey);
4619 CHECK_OFFSET(si_perf_data);
4620 CHECK_OFFSET(si_perf_type);
4621
4622 /* sigpoll */
4623 CHECK_OFFSET(si_band);
4624 CHECK_OFFSET(si_fd);
4625
4626 /* sigsys */
4627 CHECK_OFFSET(si_call_addr);
4628 CHECK_OFFSET(si_syscall);
4629 CHECK_OFFSET(si_arch);
4630 #undef CHECK_OFFSET
4631
4632 /* usb asyncio */
4633 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4634 offsetof(struct siginfo, si_addr));
4635 if (sizeof(int) == sizeof(void __user *)) {
4636 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4637 sizeof(void __user *));
4638 } else {
4639 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4640 sizeof_field(struct siginfo, si_uid)) !=
4641 sizeof(void __user *));
4642 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4643 offsetof(struct siginfo, si_uid));
4644 }
4645 #ifdef CONFIG_COMPAT
4646 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4647 offsetof(struct compat_siginfo, si_addr));
4648 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4649 sizeof(compat_uptr_t));
4650 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4651 sizeof_field(struct siginfo, si_pid));
4652 #endif
4653 }
4654
4655 void __init signals_init(void)
4656 {
4657 siginfo_buildtime_checks();
4658
4659 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4660 }
4661
4662 #ifdef CONFIG_KGDB_KDB
4663 #include <linux/kdb.h>
4664 /*
4665 * kdb_send_sig - Allows kdb to send signals without exposing
4666 * signal internals. This function checks if the required locks are
4667 * available before calling the main signal code, to avoid kdb
4668 * deadlocks.
4669 */
4670 void kdb_send_sig(struct task_struct *t, int sig)
4671 {
4672 static struct task_struct *kdb_prev_t;
4673 int new_t, ret;
4674 if (!spin_trylock(&t->sighand->siglock)) {
4675 kdb_printf("Can't do kill command now.\n"
4676 "The sigmask lock is held somewhere else in "
4677 "kernel, try again later\n");
4678 return;
4679 }
4680 new_t = kdb_prev_t != t;
4681 kdb_prev_t = t;
4682 if (t->state != TASK_RUNNING && new_t) {
4683 spin_unlock(&t->sighand->siglock);
4684 kdb_printf("Process is not RUNNING, sending a signal from "
4685 "kdb risks deadlock\n"
4686 "on the run queue locks. "
4687 "The signal has _not_ been sent.\n"
4688 "Reissue the kill command if you want to risk "
4689 "the deadlock.\n");
4690 return;
4691 }
4692 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4693 spin_unlock(&t->sighand->siglock);
4694 if (ret)
4695 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4696 sig, t->pid);
4697 else
4698 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4699 }
4700 #endif /* CONFIG_KGDB_KDB */