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