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/task_work.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#include <asm/syscall.h> /* for syscall_get_* */
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 clear_thread_flag(TIF_SIGPENDING);
186
187}
188EXPORT_SYMBOL(recalc_sigpending);
189
190void calculate_sigpending(void)
191{
192 /* Have any signals or users of TIF_SIGPENDING been delayed
193 * until after fork?
194 */
195 spin_lock_irq(¤t->sighand->siglock);
196 set_tsk_thread_flag(current, TIF_SIGPENDING);
197 recalc_sigpending();
198 spin_unlock_irq(¤t->sighand->siglock);
199}
200
201/* Given the mask, find the first available signal that should be serviced. */
202
203#define SYNCHRONOUS_MASK \
204 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
205 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
206
207int next_signal(struct sigpending *pending, sigset_t *mask)
208{
209 unsigned long i, *s, *m, x;
210 int sig = 0;
211
212 s = pending->signal.sig;
213 m = mask->sig;
214
215 /*
216 * Handle the first word specially: it contains the
217 * synchronous signals that need to be dequeued first.
218 */
219 x = *s &~ *m;
220 if (x) {
221 if (x & SYNCHRONOUS_MASK)
222 x &= SYNCHRONOUS_MASK;
223 sig = ffz(~x) + 1;
224 return sig;
225 }
226
227 switch (_NSIG_WORDS) {
228 default:
229 for (i = 1; i < _NSIG_WORDS; ++i) {
230 x = *++s &~ *++m;
231 if (!x)
232 continue;
233 sig = ffz(~x) + i*_NSIG_BPW + 1;
234 break;
235 }
236 break;
237
238 case 2:
239 x = s[1] &~ m[1];
240 if (!x)
241 break;
242 sig = ffz(~x) + _NSIG_BPW + 1;
243 break;
244
245 case 1:
246 /* Nothing to do */
247 break;
248 }
249
250 return sig;
251}
252
253static inline void print_dropped_signal(int sig)
254{
255 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
256
257 if (!print_fatal_signals)
258 return;
259
260 if (!__ratelimit(&ratelimit_state))
261 return;
262
263 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
264 current->comm, current->pid, sig);
265}
266
267/**
268 * task_set_jobctl_pending - set jobctl pending bits
269 * @task: target task
270 * @mask: pending bits to set
271 *
272 * Clear @mask from @task->jobctl. @mask must be subset of
273 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
274 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
275 * cleared. If @task is already being killed or exiting, this function
276 * becomes noop.
277 *
278 * CONTEXT:
279 * Must be called with @task->sighand->siglock held.
280 *
281 * RETURNS:
282 * %true if @mask is set, %false if made noop because @task was dying.
283 */
284bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
285{
286 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
287 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
288 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
289
290 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
291 return false;
292
293 if (mask & JOBCTL_STOP_SIGMASK)
294 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
295
296 task->jobctl |= mask;
297 return true;
298}
299
300/**
301 * task_clear_jobctl_trapping - clear jobctl trapping bit
302 * @task: target task
303 *
304 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
305 * Clear it and wake up the ptracer. Note that we don't need any further
306 * locking. @task->siglock guarantees that @task->parent points to the
307 * ptracer.
308 *
309 * CONTEXT:
310 * Must be called with @task->sighand->siglock held.
311 */
312void task_clear_jobctl_trapping(struct task_struct *task)
313{
314 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
315 task->jobctl &= ~JOBCTL_TRAPPING;
316 smp_mb(); /* advised by wake_up_bit() */
317 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
318 }
319}
320
321/**
322 * task_clear_jobctl_pending - clear jobctl pending bits
323 * @task: target task
324 * @mask: pending bits to clear
325 *
326 * Clear @mask from @task->jobctl. @mask must be subset of
327 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
328 * STOP bits are cleared together.
329 *
330 * If clearing of @mask leaves no stop or trap pending, this function calls
331 * task_clear_jobctl_trapping().
332 *
333 * CONTEXT:
334 * Must be called with @task->sighand->siglock held.
335 */
336void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
337{
338 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
339
340 if (mask & JOBCTL_STOP_PENDING)
341 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
342
343 task->jobctl &= ~mask;
344
345 if (!(task->jobctl & JOBCTL_PENDING_MASK))
346 task_clear_jobctl_trapping(task);
347}
348
349/**
350 * task_participate_group_stop - participate in a group stop
351 * @task: task participating in a group stop
352 *
353 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
354 * Group stop states are cleared and the group stop count is consumed if
355 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
356 * stop, the appropriate `SIGNAL_*` flags are set.
357 *
358 * CONTEXT:
359 * Must be called with @task->sighand->siglock held.
360 *
361 * RETURNS:
362 * %true if group stop completion should be notified to the parent, %false
363 * otherwise.
364 */
365static bool task_participate_group_stop(struct task_struct *task)
366{
367 struct signal_struct *sig = task->signal;
368 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
369
370 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
371
372 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
373
374 if (!consume)
375 return false;
376
377 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
378 sig->group_stop_count--;
379
380 /*
381 * Tell the caller to notify completion iff we are entering into a
382 * fresh group stop. Read comment in do_signal_stop() for details.
383 */
384 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
385 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
386 return true;
387 }
388 return false;
389}
390
391void task_join_group_stop(struct task_struct *task)
392{
393 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
394 struct signal_struct *sig = current->signal;
395
396 if (sig->group_stop_count) {
397 sig->group_stop_count++;
398 mask |= JOBCTL_STOP_CONSUME;
399 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
400 return;
401
402 /* Have the new thread join an on-going signal group stop */
403 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
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 gfp_flags,
413 int override_rlimit, const unsigned int sigqueue_flags)
414{
415 struct sigqueue *q = NULL;
416 struct ucounts *ucounts = NULL;
417 long sigpending;
418
419 /*
420 * Protect access to @t credentials. This can go away when all
421 * callers hold rcu read lock.
422 *
423 * NOTE! A pending signal will hold on to the user refcount,
424 * and we get/put the refcount only when the sigpending count
425 * changes from/to zero.
426 */
427 rcu_read_lock();
428 ucounts = task_ucounts(t);
429 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
430 rcu_read_unlock();
431 if (!sigpending)
432 return NULL;
433
434 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
435 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
436 } else {
437 print_dropped_signal(sig);
438 }
439
440 if (unlikely(q == NULL)) {
441 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
442 } else {
443 INIT_LIST_HEAD(&q->list);
444 q->flags = sigqueue_flags;
445 q->ucounts = ucounts;
446 }
447 return q;
448}
449
450static void __sigqueue_free(struct sigqueue *q)
451{
452 if (q->flags & SIGQUEUE_PREALLOC)
453 return;
454 if (q->ucounts) {
455 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
456 q->ucounts = NULL;
457 }
458 kmem_cache_free(sigqueue_cachep, q);
459}
460
461void flush_sigqueue(struct sigpending *queue)
462{
463 struct sigqueue *q;
464
465 sigemptyset(&queue->signal);
466 while (!list_empty(&queue->list)) {
467 q = list_entry(queue->list.next, struct sigqueue , list);
468 list_del_init(&q->list);
469 __sigqueue_free(q);
470 }
471}
472
473/*
474 * Flush all pending signals for this kthread.
475 */
476void flush_signals(struct task_struct *t)
477{
478 unsigned long flags;
479
480 spin_lock_irqsave(&t->sighand->siglock, flags);
481 clear_tsk_thread_flag(t, TIF_SIGPENDING);
482 flush_sigqueue(&t->pending);
483 flush_sigqueue(&t->signal->shared_pending);
484 spin_unlock_irqrestore(&t->sighand->siglock, flags);
485}
486EXPORT_SYMBOL(flush_signals);
487
488#ifdef CONFIG_POSIX_TIMERS
489static void __flush_itimer_signals(struct sigpending *pending)
490{
491 sigset_t signal, retain;
492 struct sigqueue *q, *n;
493
494 signal = pending->signal;
495 sigemptyset(&retain);
496
497 list_for_each_entry_safe(q, n, &pending->list, list) {
498 int sig = q->info.si_signo;
499
500 if (likely(q->info.si_code != SI_TIMER)) {
501 sigaddset(&retain, sig);
502 } else {
503 sigdelset(&signal, sig);
504 list_del_init(&q->list);
505 __sigqueue_free(q);
506 }
507 }
508
509 sigorsets(&pending->signal, &signal, &retain);
510}
511
512void flush_itimer_signals(void)
513{
514 struct task_struct *tsk = current;
515 unsigned long flags;
516
517 spin_lock_irqsave(&tsk->sighand->siglock, flags);
518 __flush_itimer_signals(&tsk->pending);
519 __flush_itimer_signals(&tsk->signal->shared_pending);
520 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
521}
522#endif
523
524void ignore_signals(struct task_struct *t)
525{
526 int i;
527
528 for (i = 0; i < _NSIG; ++i)
529 t->sighand->action[i].sa.sa_handler = SIG_IGN;
530
531 flush_signals(t);
532}
533
534/*
535 * Flush all handlers for a task.
536 */
537
538void
539flush_signal_handlers(struct task_struct *t, int force_default)
540{
541 int i;
542 struct k_sigaction *ka = &t->sighand->action[0];
543 for (i = _NSIG ; i != 0 ; i--) {
544 if (force_default || ka->sa.sa_handler != SIG_IGN)
545 ka->sa.sa_handler = SIG_DFL;
546 ka->sa.sa_flags = 0;
547#ifdef __ARCH_HAS_SA_RESTORER
548 ka->sa.sa_restorer = NULL;
549#endif
550 sigemptyset(&ka->sa.sa_mask);
551 ka++;
552 }
553}
554
555bool unhandled_signal(struct task_struct *tsk, int sig)
556{
557 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
558 if (is_global_init(tsk))
559 return true;
560
561 if (handler != SIG_IGN && handler != SIG_DFL)
562 return false;
563
564 /* if ptraced, let the tracer determine */
565 return !tsk->ptrace;
566}
567
568static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
569 bool *resched_timer)
570{
571 struct sigqueue *q, *first = NULL;
572
573 /*
574 * Collect the siginfo appropriate to this signal. Check if
575 * there is another siginfo for the same signal.
576 */
577 list_for_each_entry(q, &list->list, list) {
578 if (q->info.si_signo == sig) {
579 if (first)
580 goto still_pending;
581 first = q;
582 }
583 }
584
585 sigdelset(&list->signal, sig);
586
587 if (first) {
588still_pending:
589 list_del_init(&first->list);
590 copy_siginfo(info, &first->info);
591
592 *resched_timer =
593 (first->flags & SIGQUEUE_PREALLOC) &&
594 (info->si_code == SI_TIMER) &&
595 (info->si_sys_private);
596
597 __sigqueue_free(first);
598 } else {
599 /*
600 * Ok, it wasn't in the queue. This must be
601 * a fast-pathed signal or we must have been
602 * out of queue space. So zero out the info.
603 */
604 clear_siginfo(info);
605 info->si_signo = sig;
606 info->si_errno = 0;
607 info->si_code = SI_USER;
608 info->si_pid = 0;
609 info->si_uid = 0;
610 }
611}
612
613static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
614 kernel_siginfo_t *info, bool *resched_timer)
615{
616 int sig = next_signal(pending, mask);
617
618 if (sig)
619 collect_signal(sig, pending, info, resched_timer);
620 return sig;
621}
622
623/*
624 * Dequeue a signal and return the element to the caller, which is
625 * expected to free it.
626 *
627 * All callers have to hold the siglock.
628 */
629int dequeue_signal(struct task_struct *tsk, sigset_t *mask,
630 kernel_siginfo_t *info, enum pid_type *type)
631{
632 bool resched_timer = false;
633 int signr;
634
635 /* We only dequeue private signals from ourselves, we don't let
636 * signalfd steal them
637 */
638 *type = PIDTYPE_PID;
639 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
640 if (!signr) {
641 *type = PIDTYPE_TGID;
642 signr = __dequeue_signal(&tsk->signal->shared_pending,
643 mask, info, &resched_timer);
644#ifdef CONFIG_POSIX_TIMERS
645 /*
646 * itimer signal ?
647 *
648 * itimers are process shared and we restart periodic
649 * itimers in the signal delivery path to prevent DoS
650 * attacks in the high resolution timer case. This is
651 * compliant with the old way of self-restarting
652 * itimers, as the SIGALRM is a legacy signal and only
653 * queued once. Changing the restart behaviour to
654 * restart the timer in the signal dequeue path is
655 * reducing the timer noise on heavy loaded !highres
656 * systems too.
657 */
658 if (unlikely(signr == SIGALRM)) {
659 struct hrtimer *tmr = &tsk->signal->real_timer;
660
661 if (!hrtimer_is_queued(tmr) &&
662 tsk->signal->it_real_incr != 0) {
663 hrtimer_forward(tmr, tmr->base->get_time(),
664 tsk->signal->it_real_incr);
665 hrtimer_restart(tmr);
666 }
667 }
668#endif
669 }
670
671 recalc_sigpending();
672 if (!signr)
673 return 0;
674
675 if (unlikely(sig_kernel_stop(signr))) {
676 /*
677 * Set a marker that we have dequeued a stop signal. Our
678 * caller might release the siglock and then the pending
679 * stop signal it is about to process is no longer in the
680 * pending bitmasks, but must still be cleared by a SIGCONT
681 * (and overruled by a SIGKILL). So those cases clear this
682 * shared flag after we've set it. Note that this flag may
683 * remain set after the signal we return is ignored or
684 * handled. That doesn't matter because its only purpose
685 * is to alert stop-signal processing code when another
686 * processor has come along and cleared the flag.
687 */
688 current->jobctl |= JOBCTL_STOP_DEQUEUED;
689 }
690#ifdef CONFIG_POSIX_TIMERS
691 if (resched_timer) {
692 /*
693 * Release the siglock to ensure proper locking order
694 * of timer locks outside of siglocks. Note, we leave
695 * irqs disabled here, since the posix-timers code is
696 * about to disable them again anyway.
697 */
698 spin_unlock(&tsk->sighand->siglock);
699 posixtimer_rearm(info);
700 spin_lock(&tsk->sighand->siglock);
701
702 /* Don't expose the si_sys_private value to userspace */
703 info->si_sys_private = 0;
704 }
705#endif
706 return signr;
707}
708EXPORT_SYMBOL_GPL(dequeue_signal);
709
710static int dequeue_synchronous_signal(kernel_siginfo_t *info)
711{
712 struct task_struct *tsk = current;
713 struct sigpending *pending = &tsk->pending;
714 struct sigqueue *q, *sync = NULL;
715
716 /*
717 * Might a synchronous signal be in the queue?
718 */
719 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
720 return 0;
721
722 /*
723 * Return the first synchronous signal in the queue.
724 */
725 list_for_each_entry(q, &pending->list, list) {
726 /* Synchronous signals have a positive si_code */
727 if ((q->info.si_code > SI_USER) &&
728 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
729 sync = q;
730 goto next;
731 }
732 }
733 return 0;
734next:
735 /*
736 * Check if there is another siginfo for the same signal.
737 */
738 list_for_each_entry_continue(q, &pending->list, list) {
739 if (q->info.si_signo == sync->info.si_signo)
740 goto still_pending;
741 }
742
743 sigdelset(&pending->signal, sync->info.si_signo);
744 recalc_sigpending();
745still_pending:
746 list_del_init(&sync->list);
747 copy_siginfo(info, &sync->info);
748 __sigqueue_free(sync);
749 return info->si_signo;
750}
751
752/*
753 * Tell a process that it has a new active signal..
754 *
755 * NOTE! we rely on the previous spin_lock to
756 * lock interrupts for us! We can only be called with
757 * "siglock" held, and the local interrupt must
758 * have been disabled when that got acquired!
759 *
760 * No need to set need_resched since signal event passing
761 * goes through ->blocked
762 */
763void signal_wake_up_state(struct task_struct *t, unsigned int state)
764{
765 lockdep_assert_held(&t->sighand->siglock);
766
767 set_tsk_thread_flag(t, TIF_SIGPENDING);
768
769 /*
770 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
771 * case. We don't check t->state here because there is a race with it
772 * executing another processor and just now entering stopped state.
773 * By using wake_up_state, we ensure the process will wake up and
774 * handle its death signal.
775 */
776 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
777 kick_process(t);
778}
779
780/*
781 * Remove signals in mask from the pending set and queue.
782 * Returns 1 if any signals were found.
783 *
784 * All callers must be holding the siglock.
785 */
786static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
787{
788 struct sigqueue *q, *n;
789 sigset_t m;
790
791 sigandsets(&m, mask, &s->signal);
792 if (sigisemptyset(&m))
793 return;
794
795 sigandnsets(&s->signal, &s->signal, mask);
796 list_for_each_entry_safe(q, n, &s->list, list) {
797 if (sigismember(mask, q->info.si_signo)) {
798 list_del_init(&q->list);
799 __sigqueue_free(q);
800 }
801 }
802}
803
804static inline int is_si_special(const struct kernel_siginfo *info)
805{
806 return info <= SEND_SIG_PRIV;
807}
808
809static inline bool si_fromuser(const struct kernel_siginfo *info)
810{
811 return info == SEND_SIG_NOINFO ||
812 (!is_si_special(info) && SI_FROMUSER(info));
813}
814
815/*
816 * called with RCU read lock from check_kill_permission()
817 */
818static bool kill_ok_by_cred(struct task_struct *t)
819{
820 const struct cred *cred = current_cred();
821 const struct cred *tcred = __task_cred(t);
822
823 return uid_eq(cred->euid, tcred->suid) ||
824 uid_eq(cred->euid, tcred->uid) ||
825 uid_eq(cred->uid, tcred->suid) ||
826 uid_eq(cred->uid, tcred->uid) ||
827 ns_capable(tcred->user_ns, CAP_KILL);
828}
829
830/*
831 * Bad permissions for sending the signal
832 * - the caller must hold the RCU read lock
833 */
834static int check_kill_permission(int sig, struct kernel_siginfo *info,
835 struct task_struct *t)
836{
837 struct pid *sid;
838 int error;
839
840 if (!valid_signal(sig))
841 return -EINVAL;
842
843 if (!si_fromuser(info))
844 return 0;
845
846 error = audit_signal_info(sig, t); /* Let audit system see the signal */
847 if (error)
848 return error;
849
850 if (!same_thread_group(current, t) &&
851 !kill_ok_by_cred(t)) {
852 switch (sig) {
853 case SIGCONT:
854 sid = task_session(t);
855 /*
856 * We don't return the error if sid == NULL. The
857 * task was unhashed, the caller must notice this.
858 */
859 if (!sid || sid == task_session(current))
860 break;
861 fallthrough;
862 default:
863 return -EPERM;
864 }
865 }
866
867 return security_task_kill(t, info, sig, NULL);
868}
869
870/**
871 * ptrace_trap_notify - schedule trap to notify ptracer
872 * @t: tracee wanting to notify tracer
873 *
874 * This function schedules sticky ptrace trap which is cleared on the next
875 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
876 * ptracer.
877 *
878 * If @t is running, STOP trap will be taken. If trapped for STOP and
879 * ptracer is listening for events, tracee is woken up so that it can
880 * re-trap for the new event. If trapped otherwise, STOP trap will be
881 * eventually taken without returning to userland after the existing traps
882 * are finished by PTRACE_CONT.
883 *
884 * CONTEXT:
885 * Must be called with @task->sighand->siglock held.
886 */
887static void ptrace_trap_notify(struct task_struct *t)
888{
889 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
890 lockdep_assert_held(&t->sighand->siglock);
891
892 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
893 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
894}
895
896/*
897 * Handle magic process-wide effects of stop/continue signals. Unlike
898 * the signal actions, these happen immediately at signal-generation
899 * time regardless of blocking, ignoring, or handling. This does the
900 * actual continuing for SIGCONT, but not the actual stopping for stop
901 * signals. The process stop is done as a signal action for SIG_DFL.
902 *
903 * Returns true if the signal should be actually delivered, otherwise
904 * it should be dropped.
905 */
906static bool prepare_signal(int sig, struct task_struct *p, bool force)
907{
908 struct signal_struct *signal = p->signal;
909 struct task_struct *t;
910 sigset_t flush;
911
912 if (signal->flags & SIGNAL_GROUP_EXIT) {
913 if (signal->core_state)
914 return sig == SIGKILL;
915 /*
916 * The process is in the middle of dying, drop the signal.
917 */
918 return false;
919 } else if (sig_kernel_stop(sig)) {
920 /*
921 * This is a stop signal. Remove SIGCONT from all queues.
922 */
923 siginitset(&flush, sigmask(SIGCONT));
924 flush_sigqueue_mask(&flush, &signal->shared_pending);
925 for_each_thread(p, t)
926 flush_sigqueue_mask(&flush, &t->pending);
927 } else if (sig == SIGCONT) {
928 unsigned int why;
929 /*
930 * Remove all stop signals from all queues, wake all threads.
931 */
932 siginitset(&flush, SIG_KERNEL_STOP_MASK);
933 flush_sigqueue_mask(&flush, &signal->shared_pending);
934 for_each_thread(p, t) {
935 flush_sigqueue_mask(&flush, &t->pending);
936 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
937 if (likely(!(t->ptrace & PT_SEIZED))) {
938 t->jobctl &= ~JOBCTL_STOPPED;
939 wake_up_state(t, __TASK_STOPPED);
940 } else
941 ptrace_trap_notify(t);
942 }
943
944 /*
945 * Notify the parent with CLD_CONTINUED if we were stopped.
946 *
947 * If we were in the middle of a group stop, we pretend it
948 * was already finished, and then continued. Since SIGCHLD
949 * doesn't queue we report only CLD_STOPPED, as if the next
950 * CLD_CONTINUED was dropped.
951 */
952 why = 0;
953 if (signal->flags & SIGNAL_STOP_STOPPED)
954 why |= SIGNAL_CLD_CONTINUED;
955 else if (signal->group_stop_count)
956 why |= SIGNAL_CLD_STOPPED;
957
958 if (why) {
959 /*
960 * The first thread which returns from do_signal_stop()
961 * will take ->siglock, notice SIGNAL_CLD_MASK, and
962 * notify its parent. See get_signal().
963 */
964 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
965 signal->group_stop_count = 0;
966 signal->group_exit_code = 0;
967 }
968 }
969
970 return !sig_ignored(p, sig, force);
971}
972
973/*
974 * Test if P wants to take SIG. After we've checked all threads with this,
975 * it's equivalent to finding no threads not blocking SIG. Any threads not
976 * blocking SIG were ruled out because they are not running and already
977 * have pending signals. Such threads will dequeue from the shared queue
978 * as soon as they're available, so putting the signal on the shared queue
979 * will be equivalent to sending it to one such thread.
980 */
981static inline bool wants_signal(int sig, struct task_struct *p)
982{
983 if (sigismember(&p->blocked, sig))
984 return false;
985
986 if (p->flags & PF_EXITING)
987 return false;
988
989 if (sig == SIGKILL)
990 return true;
991
992 if (task_is_stopped_or_traced(p))
993 return false;
994
995 return task_curr(p) || !task_sigpending(p);
996}
997
998static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
999{
1000 struct signal_struct *signal = p->signal;
1001 struct task_struct *t;
1002
1003 /*
1004 * Now find a thread we can wake up to take the signal off the queue.
1005 *
1006 * If the main thread wants the signal, it gets first crack.
1007 * Probably the least surprising to the average bear.
1008 */
1009 if (wants_signal(sig, p))
1010 t = p;
1011 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1012 /*
1013 * There is just one thread and it does not need to be woken.
1014 * It will dequeue unblocked signals before it runs again.
1015 */
1016 return;
1017 else {
1018 /*
1019 * Otherwise try to find a suitable thread.
1020 */
1021 t = signal->curr_target;
1022 while (!wants_signal(sig, t)) {
1023 t = next_thread(t);
1024 if (t == signal->curr_target)
1025 /*
1026 * No thread needs to be woken.
1027 * Any eligible threads will see
1028 * the signal in the queue soon.
1029 */
1030 return;
1031 }
1032 signal->curr_target = t;
1033 }
1034
1035 /*
1036 * Found a killable thread. If the signal will be fatal,
1037 * then start taking the whole group down immediately.
1038 */
1039 if (sig_fatal(p, sig) &&
1040 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1041 !sigismember(&t->real_blocked, sig) &&
1042 (sig == SIGKILL || !p->ptrace)) {
1043 /*
1044 * This signal will be fatal to the whole group.
1045 */
1046 if (!sig_kernel_coredump(sig)) {
1047 /*
1048 * Start a group exit and wake everybody up.
1049 * This way we don't have other threads
1050 * running and doing things after a slower
1051 * thread has the fatal signal pending.
1052 */
1053 signal->flags = SIGNAL_GROUP_EXIT;
1054 signal->group_exit_code = sig;
1055 signal->group_stop_count = 0;
1056 t = p;
1057 do {
1058 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1059 sigaddset(&t->pending.signal, SIGKILL);
1060 signal_wake_up(t, 1);
1061 } while_each_thread(p, t);
1062 return;
1063 }
1064 }
1065
1066 /*
1067 * The signal is already in the shared-pending queue.
1068 * Tell the chosen thread to wake up and dequeue it.
1069 */
1070 signal_wake_up(t, sig == SIGKILL);
1071 return;
1072}
1073
1074static inline bool legacy_queue(struct sigpending *signals, int sig)
1075{
1076 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1077}
1078
1079static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1080 struct task_struct *t, enum pid_type type, bool force)
1081{
1082 struct sigpending *pending;
1083 struct sigqueue *q;
1084 int override_rlimit;
1085 int ret = 0, result;
1086
1087 lockdep_assert_held(&t->sighand->siglock);
1088
1089 result = TRACE_SIGNAL_IGNORED;
1090 if (!prepare_signal(sig, t, force))
1091 goto ret;
1092
1093 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1094 /*
1095 * Short-circuit ignored signals and support queuing
1096 * exactly one non-rt signal, so that we can get more
1097 * detailed information about the cause of the signal.
1098 */
1099 result = TRACE_SIGNAL_ALREADY_PENDING;
1100 if (legacy_queue(pending, sig))
1101 goto ret;
1102
1103 result = TRACE_SIGNAL_DELIVERED;
1104 /*
1105 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1106 */
1107 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1108 goto out_set;
1109
1110 /*
1111 * Real-time signals must be queued if sent by sigqueue, or
1112 * some other real-time mechanism. It is implementation
1113 * defined whether kill() does so. We attempt to do so, on
1114 * the principle of least surprise, but since kill is not
1115 * allowed to fail with EAGAIN when low on memory we just
1116 * make sure at least one signal gets delivered and don't
1117 * pass on the info struct.
1118 */
1119 if (sig < SIGRTMIN)
1120 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1121 else
1122 override_rlimit = 0;
1123
1124 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1125
1126 if (q) {
1127 list_add_tail(&q->list, &pending->list);
1128 switch ((unsigned long) info) {
1129 case (unsigned long) SEND_SIG_NOINFO:
1130 clear_siginfo(&q->info);
1131 q->info.si_signo = sig;
1132 q->info.si_errno = 0;
1133 q->info.si_code = SI_USER;
1134 q->info.si_pid = task_tgid_nr_ns(current,
1135 task_active_pid_ns(t));
1136 rcu_read_lock();
1137 q->info.si_uid =
1138 from_kuid_munged(task_cred_xxx(t, user_ns),
1139 current_uid());
1140 rcu_read_unlock();
1141 break;
1142 case (unsigned long) SEND_SIG_PRIV:
1143 clear_siginfo(&q->info);
1144 q->info.si_signo = sig;
1145 q->info.si_errno = 0;
1146 q->info.si_code = SI_KERNEL;
1147 q->info.si_pid = 0;
1148 q->info.si_uid = 0;
1149 break;
1150 default:
1151 copy_siginfo(&q->info, info);
1152 break;
1153 }
1154 } else if (!is_si_special(info) &&
1155 sig >= SIGRTMIN && info->si_code != SI_USER) {
1156 /*
1157 * Queue overflow, abort. We may abort if the
1158 * signal was rt and sent by user using something
1159 * other than kill().
1160 */
1161 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1162 ret = -EAGAIN;
1163 goto ret;
1164 } else {
1165 /*
1166 * This is a silent loss of information. We still
1167 * send the signal, but the *info bits are lost.
1168 */
1169 result = TRACE_SIGNAL_LOSE_INFO;
1170 }
1171
1172out_set:
1173 signalfd_notify(t, sig);
1174 sigaddset(&pending->signal, sig);
1175
1176 /* Let multiprocess signals appear after on-going forks */
1177 if (type > PIDTYPE_TGID) {
1178 struct multiprocess_signals *delayed;
1179 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1180 sigset_t *signal = &delayed->signal;
1181 /* Can't queue both a stop and a continue signal */
1182 if (sig == SIGCONT)
1183 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1184 else if (sig_kernel_stop(sig))
1185 sigdelset(signal, SIGCONT);
1186 sigaddset(signal, sig);
1187 }
1188 }
1189
1190 complete_signal(sig, t, type);
1191ret:
1192 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1193 return ret;
1194}
1195
1196static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1197{
1198 bool ret = false;
1199 switch (siginfo_layout(info->si_signo, info->si_code)) {
1200 case SIL_KILL:
1201 case SIL_CHLD:
1202 case SIL_RT:
1203 ret = true;
1204 break;
1205 case SIL_TIMER:
1206 case SIL_POLL:
1207 case SIL_FAULT:
1208 case SIL_FAULT_TRAPNO:
1209 case SIL_FAULT_MCEERR:
1210 case SIL_FAULT_BNDERR:
1211 case SIL_FAULT_PKUERR:
1212 case SIL_FAULT_PERF_EVENT:
1213 case SIL_SYS:
1214 ret = false;
1215 break;
1216 }
1217 return ret;
1218}
1219
1220int send_signal_locked(int sig, struct kernel_siginfo *info,
1221 struct task_struct *t, enum pid_type type)
1222{
1223 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1224 bool force = false;
1225
1226 if (info == SEND_SIG_NOINFO) {
1227 /* Force if sent from an ancestor pid namespace */
1228 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1229 } else if (info == SEND_SIG_PRIV) {
1230 /* Don't ignore kernel generated signals */
1231 force = true;
1232 } else if (has_si_pid_and_uid(info)) {
1233 /* SIGKILL and SIGSTOP is special or has ids */
1234 struct user_namespace *t_user_ns;
1235
1236 rcu_read_lock();
1237 t_user_ns = task_cred_xxx(t, user_ns);
1238 if (current_user_ns() != t_user_ns) {
1239 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1240 info->si_uid = from_kuid_munged(t_user_ns, uid);
1241 }
1242 rcu_read_unlock();
1243
1244 /* A kernel generated signal? */
1245 force = (info->si_code == SI_KERNEL);
1246
1247 /* From an ancestor pid namespace? */
1248 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1249 info->si_pid = 0;
1250 force = true;
1251 }
1252 }
1253 return __send_signal_locked(sig, info, t, type, force);
1254}
1255
1256static void print_fatal_signal(int signr)
1257{
1258 struct pt_regs *regs = task_pt_regs(current);
1259 pr_info("potentially unexpected fatal signal %d.\n", signr);
1260
1261#if defined(__i386__) && !defined(__arch_um__)
1262 pr_info("code at %08lx: ", regs->ip);
1263 {
1264 int i;
1265 for (i = 0; i < 16; i++) {
1266 unsigned char insn;
1267
1268 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1269 break;
1270 pr_cont("%02x ", insn);
1271 }
1272 }
1273 pr_cont("\n");
1274#endif
1275 preempt_disable();
1276 show_regs(regs);
1277 preempt_enable();
1278}
1279
1280static int __init setup_print_fatal_signals(char *str)
1281{
1282 get_option (&str, &print_fatal_signals);
1283
1284 return 1;
1285}
1286
1287__setup("print-fatal-signals=", setup_print_fatal_signals);
1288
1289int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1290 enum pid_type type)
1291{
1292 unsigned long flags;
1293 int ret = -ESRCH;
1294
1295 if (lock_task_sighand(p, &flags)) {
1296 ret = send_signal_locked(sig, info, p, type);
1297 unlock_task_sighand(p, &flags);
1298 }
1299
1300 return ret;
1301}
1302
1303enum sig_handler {
1304 HANDLER_CURRENT, /* If reachable use the current handler */
1305 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1306 HANDLER_EXIT, /* Only visible as the process exit code */
1307};
1308
1309/*
1310 * Force a signal that the process can't ignore: if necessary
1311 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1312 *
1313 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1314 * since we do not want to have a signal handler that was blocked
1315 * be invoked when user space had explicitly blocked it.
1316 *
1317 * We don't want to have recursive SIGSEGV's etc, for example,
1318 * that is why we also clear SIGNAL_UNKILLABLE.
1319 */
1320static int
1321force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1322 enum sig_handler handler)
1323{
1324 unsigned long int flags;
1325 int ret, blocked, ignored;
1326 struct k_sigaction *action;
1327 int sig = info->si_signo;
1328
1329 spin_lock_irqsave(&t->sighand->siglock, flags);
1330 action = &t->sighand->action[sig-1];
1331 ignored = action->sa.sa_handler == SIG_IGN;
1332 blocked = sigismember(&t->blocked, sig);
1333 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1334 action->sa.sa_handler = SIG_DFL;
1335 if (handler == HANDLER_EXIT)
1336 action->sa.sa_flags |= SA_IMMUTABLE;
1337 if (blocked) {
1338 sigdelset(&t->blocked, sig);
1339 recalc_sigpending_and_wake(t);
1340 }
1341 }
1342 /*
1343 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1344 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1345 */
1346 if (action->sa.sa_handler == SIG_DFL &&
1347 (!t->ptrace || (handler == HANDLER_EXIT)))
1348 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1349 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1350 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1351
1352 return ret;
1353}
1354
1355int force_sig_info(struct kernel_siginfo *info)
1356{
1357 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1358}
1359
1360/*
1361 * Nuke all other threads in the group.
1362 */
1363int zap_other_threads(struct task_struct *p)
1364{
1365 struct task_struct *t = p;
1366 int count = 0;
1367
1368 p->signal->group_stop_count = 0;
1369
1370 while_each_thread(p, t) {
1371 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1372 count++;
1373
1374 /* Don't bother with already dead threads */
1375 if (t->exit_state)
1376 continue;
1377 sigaddset(&t->pending.signal, SIGKILL);
1378 signal_wake_up(t, 1);
1379 }
1380
1381 return count;
1382}
1383
1384struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1385 unsigned long *flags)
1386{
1387 struct sighand_struct *sighand;
1388
1389 rcu_read_lock();
1390 for (;;) {
1391 sighand = rcu_dereference(tsk->sighand);
1392 if (unlikely(sighand == NULL))
1393 break;
1394
1395 /*
1396 * This sighand can be already freed and even reused, but
1397 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1398 * initializes ->siglock: this slab can't go away, it has
1399 * the same object type, ->siglock can't be reinitialized.
1400 *
1401 * We need to ensure that tsk->sighand is still the same
1402 * after we take the lock, we can race with de_thread() or
1403 * __exit_signal(). In the latter case the next iteration
1404 * must see ->sighand == NULL.
1405 */
1406 spin_lock_irqsave(&sighand->siglock, *flags);
1407 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1408 break;
1409 spin_unlock_irqrestore(&sighand->siglock, *flags);
1410 }
1411 rcu_read_unlock();
1412
1413 return sighand;
1414}
1415
1416#ifdef CONFIG_LOCKDEP
1417void lockdep_assert_task_sighand_held(struct task_struct *task)
1418{
1419 struct sighand_struct *sighand;
1420
1421 rcu_read_lock();
1422 sighand = rcu_dereference(task->sighand);
1423 if (sighand)
1424 lockdep_assert_held(&sighand->siglock);
1425 else
1426 WARN_ON_ONCE(1);
1427 rcu_read_unlock();
1428}
1429#endif
1430
1431/*
1432 * send signal info to all the members of a group
1433 */
1434int group_send_sig_info(int sig, struct kernel_siginfo *info,
1435 struct task_struct *p, enum pid_type type)
1436{
1437 int ret;
1438
1439 rcu_read_lock();
1440 ret = check_kill_permission(sig, info, p);
1441 rcu_read_unlock();
1442
1443 if (!ret && sig)
1444 ret = do_send_sig_info(sig, info, p, type);
1445
1446 return ret;
1447}
1448
1449/*
1450 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1451 * control characters do (^C, ^Z etc)
1452 * - the caller must hold at least a readlock on tasklist_lock
1453 */
1454int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1455{
1456 struct task_struct *p = NULL;
1457 int retval, success;
1458
1459 success = 0;
1460 retval = -ESRCH;
1461 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1462 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1463 success |= !err;
1464 retval = err;
1465 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1466 return success ? 0 : retval;
1467}
1468
1469int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1470{
1471 int error = -ESRCH;
1472 struct task_struct *p;
1473
1474 for (;;) {
1475 rcu_read_lock();
1476 p = pid_task(pid, PIDTYPE_PID);
1477 if (p)
1478 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1479 rcu_read_unlock();
1480 if (likely(!p || error != -ESRCH))
1481 return error;
1482
1483 /*
1484 * The task was unhashed in between, try again. If it
1485 * is dead, pid_task() will return NULL, if we race with
1486 * de_thread() it will find the new leader.
1487 */
1488 }
1489}
1490
1491static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1492{
1493 int error;
1494 rcu_read_lock();
1495 error = kill_pid_info(sig, info, find_vpid(pid));
1496 rcu_read_unlock();
1497 return error;
1498}
1499
1500static inline bool kill_as_cred_perm(const struct cred *cred,
1501 struct task_struct *target)
1502{
1503 const struct cred *pcred = __task_cred(target);
1504
1505 return uid_eq(cred->euid, pcred->suid) ||
1506 uid_eq(cred->euid, pcred->uid) ||
1507 uid_eq(cred->uid, pcred->suid) ||
1508 uid_eq(cred->uid, pcred->uid);
1509}
1510
1511/*
1512 * The usb asyncio usage of siginfo is wrong. The glibc support
1513 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1514 * AKA after the generic fields:
1515 * kernel_pid_t si_pid;
1516 * kernel_uid32_t si_uid;
1517 * sigval_t si_value;
1518 *
1519 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1520 * after the generic fields is:
1521 * void __user *si_addr;
1522 *
1523 * This is a practical problem when there is a 64bit big endian kernel
1524 * and a 32bit userspace. As the 32bit address will encoded in the low
1525 * 32bits of the pointer. Those low 32bits will be stored at higher
1526 * address than appear in a 32 bit pointer. So userspace will not
1527 * see the address it was expecting for it's completions.
1528 *
1529 * There is nothing in the encoding that can allow
1530 * copy_siginfo_to_user32 to detect this confusion of formats, so
1531 * handle this by requiring the caller of kill_pid_usb_asyncio to
1532 * notice when this situration takes place and to store the 32bit
1533 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1534 * parameter.
1535 */
1536int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1537 struct pid *pid, const struct cred *cred)
1538{
1539 struct kernel_siginfo info;
1540 struct task_struct *p;
1541 unsigned long flags;
1542 int ret = -EINVAL;
1543
1544 if (!valid_signal(sig))
1545 return ret;
1546
1547 clear_siginfo(&info);
1548 info.si_signo = sig;
1549 info.si_errno = errno;
1550 info.si_code = SI_ASYNCIO;
1551 *((sigval_t *)&info.si_pid) = addr;
1552
1553 rcu_read_lock();
1554 p = pid_task(pid, PIDTYPE_PID);
1555 if (!p) {
1556 ret = -ESRCH;
1557 goto out_unlock;
1558 }
1559 if (!kill_as_cred_perm(cred, p)) {
1560 ret = -EPERM;
1561 goto out_unlock;
1562 }
1563 ret = security_task_kill(p, &info, sig, cred);
1564 if (ret)
1565 goto out_unlock;
1566
1567 if (sig) {
1568 if (lock_task_sighand(p, &flags)) {
1569 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1570 unlock_task_sighand(p, &flags);
1571 } else
1572 ret = -ESRCH;
1573 }
1574out_unlock:
1575 rcu_read_unlock();
1576 return ret;
1577}
1578EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1579
1580/*
1581 * kill_something_info() interprets pid in interesting ways just like kill(2).
1582 *
1583 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1584 * is probably wrong. Should make it like BSD or SYSV.
1585 */
1586
1587static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1588{
1589 int ret;
1590
1591 if (pid > 0)
1592 return kill_proc_info(sig, info, pid);
1593
1594 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1595 if (pid == INT_MIN)
1596 return -ESRCH;
1597
1598 read_lock(&tasklist_lock);
1599 if (pid != -1) {
1600 ret = __kill_pgrp_info(sig, info,
1601 pid ? find_vpid(-pid) : task_pgrp(current));
1602 } else {
1603 int retval = 0, count = 0;
1604 struct task_struct * p;
1605
1606 for_each_process(p) {
1607 if (task_pid_vnr(p) > 1 &&
1608 !same_thread_group(p, current)) {
1609 int err = group_send_sig_info(sig, info, p,
1610 PIDTYPE_MAX);
1611 ++count;
1612 if (err != -EPERM)
1613 retval = err;
1614 }
1615 }
1616 ret = count ? retval : -ESRCH;
1617 }
1618 read_unlock(&tasklist_lock);
1619
1620 return ret;
1621}
1622
1623/*
1624 * These are for backward compatibility with the rest of the kernel source.
1625 */
1626
1627int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1628{
1629 /*
1630 * Make sure legacy kernel users don't send in bad values
1631 * (normal paths check this in check_kill_permission).
1632 */
1633 if (!valid_signal(sig))
1634 return -EINVAL;
1635
1636 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1637}
1638EXPORT_SYMBOL(send_sig_info);
1639
1640#define __si_special(priv) \
1641 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1642
1643int
1644send_sig(int sig, struct task_struct *p, int priv)
1645{
1646 return send_sig_info(sig, __si_special(priv), p);
1647}
1648EXPORT_SYMBOL(send_sig);
1649
1650void force_sig(int sig)
1651{
1652 struct kernel_siginfo info;
1653
1654 clear_siginfo(&info);
1655 info.si_signo = sig;
1656 info.si_errno = 0;
1657 info.si_code = SI_KERNEL;
1658 info.si_pid = 0;
1659 info.si_uid = 0;
1660 force_sig_info(&info);
1661}
1662EXPORT_SYMBOL(force_sig);
1663
1664void force_fatal_sig(int sig)
1665{
1666 struct kernel_siginfo info;
1667
1668 clear_siginfo(&info);
1669 info.si_signo = sig;
1670 info.si_errno = 0;
1671 info.si_code = SI_KERNEL;
1672 info.si_pid = 0;
1673 info.si_uid = 0;
1674 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1675}
1676
1677void force_exit_sig(int sig)
1678{
1679 struct kernel_siginfo info;
1680
1681 clear_siginfo(&info);
1682 info.si_signo = sig;
1683 info.si_errno = 0;
1684 info.si_code = SI_KERNEL;
1685 info.si_pid = 0;
1686 info.si_uid = 0;
1687 force_sig_info_to_task(&info, current, HANDLER_EXIT);
1688}
1689
1690/*
1691 * When things go south during signal handling, we
1692 * will force a SIGSEGV. And if the signal that caused
1693 * the problem was already a SIGSEGV, we'll want to
1694 * make sure we don't even try to deliver the signal..
1695 */
1696void force_sigsegv(int sig)
1697{
1698 if (sig == SIGSEGV)
1699 force_fatal_sig(SIGSEGV);
1700 else
1701 force_sig(SIGSEGV);
1702}
1703
1704int force_sig_fault_to_task(int sig, int code, void __user *addr
1705 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1706 , struct task_struct *t)
1707{
1708 struct kernel_siginfo info;
1709
1710 clear_siginfo(&info);
1711 info.si_signo = sig;
1712 info.si_errno = 0;
1713 info.si_code = code;
1714 info.si_addr = addr;
1715#ifdef __ia64__
1716 info.si_imm = imm;
1717 info.si_flags = flags;
1718 info.si_isr = isr;
1719#endif
1720 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1721}
1722
1723int force_sig_fault(int sig, int code, void __user *addr
1724 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1725{
1726 return force_sig_fault_to_task(sig, code, addr
1727 ___ARCH_SI_IA64(imm, flags, isr), current);
1728}
1729
1730int send_sig_fault(int sig, int code, void __user *addr
1731 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1732 , struct task_struct *t)
1733{
1734 struct kernel_siginfo info;
1735
1736 clear_siginfo(&info);
1737 info.si_signo = sig;
1738 info.si_errno = 0;
1739 info.si_code = code;
1740 info.si_addr = addr;
1741#ifdef __ia64__
1742 info.si_imm = imm;
1743 info.si_flags = flags;
1744 info.si_isr = isr;
1745#endif
1746 return send_sig_info(info.si_signo, &info, t);
1747}
1748
1749int force_sig_mceerr(int code, void __user *addr, short lsb)
1750{
1751 struct kernel_siginfo info;
1752
1753 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1754 clear_siginfo(&info);
1755 info.si_signo = SIGBUS;
1756 info.si_errno = 0;
1757 info.si_code = code;
1758 info.si_addr = addr;
1759 info.si_addr_lsb = lsb;
1760 return force_sig_info(&info);
1761}
1762
1763int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1764{
1765 struct kernel_siginfo info;
1766
1767 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1768 clear_siginfo(&info);
1769 info.si_signo = SIGBUS;
1770 info.si_errno = 0;
1771 info.si_code = code;
1772 info.si_addr = addr;
1773 info.si_addr_lsb = lsb;
1774 return send_sig_info(info.si_signo, &info, t);
1775}
1776EXPORT_SYMBOL(send_sig_mceerr);
1777
1778int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1779{
1780 struct kernel_siginfo info;
1781
1782 clear_siginfo(&info);
1783 info.si_signo = SIGSEGV;
1784 info.si_errno = 0;
1785 info.si_code = SEGV_BNDERR;
1786 info.si_addr = addr;
1787 info.si_lower = lower;
1788 info.si_upper = upper;
1789 return force_sig_info(&info);
1790}
1791
1792#ifdef SEGV_PKUERR
1793int force_sig_pkuerr(void __user *addr, u32 pkey)
1794{
1795 struct kernel_siginfo info;
1796
1797 clear_siginfo(&info);
1798 info.si_signo = SIGSEGV;
1799 info.si_errno = 0;
1800 info.si_code = SEGV_PKUERR;
1801 info.si_addr = addr;
1802 info.si_pkey = pkey;
1803 return force_sig_info(&info);
1804}
1805#endif
1806
1807int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1808{
1809 struct kernel_siginfo info;
1810
1811 clear_siginfo(&info);
1812 info.si_signo = SIGTRAP;
1813 info.si_errno = 0;
1814 info.si_code = TRAP_PERF;
1815 info.si_addr = addr;
1816 info.si_perf_data = sig_data;
1817 info.si_perf_type = type;
1818
1819 /*
1820 * Signals generated by perf events should not terminate the whole
1821 * process if SIGTRAP is blocked, however, delivering the signal
1822 * asynchronously is better than not delivering at all. But tell user
1823 * space if the signal was asynchronous, so it can clearly be
1824 * distinguished from normal synchronous ones.
1825 */
1826 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1827 TRAP_PERF_FLAG_ASYNC :
1828 0;
1829
1830 return send_sig_info(info.si_signo, &info, current);
1831}
1832
1833/**
1834 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1835 * @syscall: syscall number to send to userland
1836 * @reason: filter-supplied reason code to send to userland (via si_errno)
1837 * @force_coredump: true to trigger a coredump
1838 *
1839 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1840 */
1841int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1842{
1843 struct kernel_siginfo info;
1844
1845 clear_siginfo(&info);
1846 info.si_signo = SIGSYS;
1847 info.si_code = SYS_SECCOMP;
1848 info.si_call_addr = (void __user *)KSTK_EIP(current);
1849 info.si_errno = reason;
1850 info.si_arch = syscall_get_arch(current);
1851 info.si_syscall = syscall;
1852 return force_sig_info_to_task(&info, current,
1853 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1854}
1855
1856/* For the crazy architectures that include trap information in
1857 * the errno field, instead of an actual errno value.
1858 */
1859int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1860{
1861 struct kernel_siginfo info;
1862
1863 clear_siginfo(&info);
1864 info.si_signo = SIGTRAP;
1865 info.si_errno = errno;
1866 info.si_code = TRAP_HWBKPT;
1867 info.si_addr = addr;
1868 return force_sig_info(&info);
1869}
1870
1871/* For the rare architectures that include trap information using
1872 * si_trapno.
1873 */
1874int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1875{
1876 struct kernel_siginfo info;
1877
1878 clear_siginfo(&info);
1879 info.si_signo = sig;
1880 info.si_errno = 0;
1881 info.si_code = code;
1882 info.si_addr = addr;
1883 info.si_trapno = trapno;
1884 return force_sig_info(&info);
1885}
1886
1887/* For the rare architectures that include trap information using
1888 * si_trapno.
1889 */
1890int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1891 struct task_struct *t)
1892{
1893 struct kernel_siginfo info;
1894
1895 clear_siginfo(&info);
1896 info.si_signo = sig;
1897 info.si_errno = 0;
1898 info.si_code = code;
1899 info.si_addr = addr;
1900 info.si_trapno = trapno;
1901 return send_sig_info(info.si_signo, &info, t);
1902}
1903
1904int kill_pgrp(struct pid *pid, int sig, int priv)
1905{
1906 int ret;
1907
1908 read_lock(&tasklist_lock);
1909 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1910 read_unlock(&tasklist_lock);
1911
1912 return ret;
1913}
1914EXPORT_SYMBOL(kill_pgrp);
1915
1916int kill_pid(struct pid *pid, int sig, int priv)
1917{
1918 return kill_pid_info(sig, __si_special(priv), pid);
1919}
1920EXPORT_SYMBOL(kill_pid);
1921
1922/*
1923 * These functions support sending signals using preallocated sigqueue
1924 * structures. This is needed "because realtime applications cannot
1925 * afford to lose notifications of asynchronous events, like timer
1926 * expirations or I/O completions". In the case of POSIX Timers
1927 * we allocate the sigqueue structure from the timer_create. If this
1928 * allocation fails we are able to report the failure to the application
1929 * with an EAGAIN error.
1930 */
1931struct sigqueue *sigqueue_alloc(void)
1932{
1933 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1934}
1935
1936void sigqueue_free(struct sigqueue *q)
1937{
1938 unsigned long flags;
1939 spinlock_t *lock = ¤t->sighand->siglock;
1940
1941 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1942 /*
1943 * We must hold ->siglock while testing q->list
1944 * to serialize with collect_signal() or with
1945 * __exit_signal()->flush_sigqueue().
1946 */
1947 spin_lock_irqsave(lock, flags);
1948 q->flags &= ~SIGQUEUE_PREALLOC;
1949 /*
1950 * If it is queued it will be freed when dequeued,
1951 * like the "regular" sigqueue.
1952 */
1953 if (!list_empty(&q->list))
1954 q = NULL;
1955 spin_unlock_irqrestore(lock, flags);
1956
1957 if (q)
1958 __sigqueue_free(q);
1959}
1960
1961int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1962{
1963 int sig = q->info.si_signo;
1964 struct sigpending *pending;
1965 struct task_struct *t;
1966 unsigned long flags;
1967 int ret, result;
1968
1969 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1970
1971 ret = -1;
1972 rcu_read_lock();
1973 t = pid_task(pid, type);
1974 if (!t || !likely(lock_task_sighand(t, &flags)))
1975 goto ret;
1976
1977 ret = 1; /* the signal is ignored */
1978 result = TRACE_SIGNAL_IGNORED;
1979 if (!prepare_signal(sig, t, false))
1980 goto out;
1981
1982 ret = 0;
1983 if (unlikely(!list_empty(&q->list))) {
1984 /*
1985 * If an SI_TIMER entry is already queue just increment
1986 * the overrun count.
1987 */
1988 BUG_ON(q->info.si_code != SI_TIMER);
1989 q->info.si_overrun++;
1990 result = TRACE_SIGNAL_ALREADY_PENDING;
1991 goto out;
1992 }
1993 q->info.si_overrun = 0;
1994
1995 signalfd_notify(t, sig);
1996 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1997 list_add_tail(&q->list, &pending->list);
1998 sigaddset(&pending->signal, sig);
1999 complete_signal(sig, t, type);
2000 result = TRACE_SIGNAL_DELIVERED;
2001out:
2002 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
2003 unlock_task_sighand(t, &flags);
2004ret:
2005 rcu_read_unlock();
2006 return ret;
2007}
2008
2009static void do_notify_pidfd(struct task_struct *task)
2010{
2011 struct pid *pid;
2012
2013 WARN_ON(task->exit_state == 0);
2014 pid = task_pid(task);
2015 wake_up_all(&pid->wait_pidfd);
2016}
2017
2018/*
2019 * Let a parent know about the death of a child.
2020 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2021 *
2022 * Returns true if our parent ignored us and so we've switched to
2023 * self-reaping.
2024 */
2025bool do_notify_parent(struct task_struct *tsk, int sig)
2026{
2027 struct kernel_siginfo info;
2028 unsigned long flags;
2029 struct sighand_struct *psig;
2030 bool autoreap = false;
2031 u64 utime, stime;
2032
2033 WARN_ON_ONCE(sig == -1);
2034
2035 /* do_notify_parent_cldstop should have been called instead. */
2036 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2037
2038 WARN_ON_ONCE(!tsk->ptrace &&
2039 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2040
2041 /* Wake up all pidfd waiters */
2042 do_notify_pidfd(tsk);
2043
2044 if (sig != SIGCHLD) {
2045 /*
2046 * This is only possible if parent == real_parent.
2047 * Check if it has changed security domain.
2048 */
2049 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2050 sig = SIGCHLD;
2051 }
2052
2053 clear_siginfo(&info);
2054 info.si_signo = sig;
2055 info.si_errno = 0;
2056 /*
2057 * We are under tasklist_lock here so our parent is tied to
2058 * us and cannot change.
2059 *
2060 * task_active_pid_ns will always return the same pid namespace
2061 * until a task passes through release_task.
2062 *
2063 * write_lock() currently calls preempt_disable() which is the
2064 * same as rcu_read_lock(), but according to Oleg, this is not
2065 * correct to rely on this
2066 */
2067 rcu_read_lock();
2068 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2069 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2070 task_uid(tsk));
2071 rcu_read_unlock();
2072
2073 task_cputime(tsk, &utime, &stime);
2074 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2075 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2076
2077 info.si_status = tsk->exit_code & 0x7f;
2078 if (tsk->exit_code & 0x80)
2079 info.si_code = CLD_DUMPED;
2080 else if (tsk->exit_code & 0x7f)
2081 info.si_code = CLD_KILLED;
2082 else {
2083 info.si_code = CLD_EXITED;
2084 info.si_status = tsk->exit_code >> 8;
2085 }
2086
2087 psig = tsk->parent->sighand;
2088 spin_lock_irqsave(&psig->siglock, flags);
2089 if (!tsk->ptrace && sig == SIGCHLD &&
2090 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2091 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2092 /*
2093 * We are exiting and our parent doesn't care. POSIX.1
2094 * defines special semantics for setting SIGCHLD to SIG_IGN
2095 * or setting the SA_NOCLDWAIT flag: we should be reaped
2096 * automatically and not left for our parent's wait4 call.
2097 * Rather than having the parent do it as a magic kind of
2098 * signal handler, we just set this to tell do_exit that we
2099 * can be cleaned up without becoming a zombie. Note that
2100 * we still call __wake_up_parent in this case, because a
2101 * blocked sys_wait4 might now return -ECHILD.
2102 *
2103 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2104 * is implementation-defined: we do (if you don't want
2105 * it, just use SIG_IGN instead).
2106 */
2107 autoreap = true;
2108 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2109 sig = 0;
2110 }
2111 /*
2112 * Send with __send_signal as si_pid and si_uid are in the
2113 * parent's namespaces.
2114 */
2115 if (valid_signal(sig) && sig)
2116 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2117 __wake_up_parent(tsk, tsk->parent);
2118 spin_unlock_irqrestore(&psig->siglock, flags);
2119
2120 return autoreap;
2121}
2122
2123/**
2124 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2125 * @tsk: task reporting the state change
2126 * @for_ptracer: the notification is for ptracer
2127 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2128 *
2129 * Notify @tsk's parent that the stopped/continued state has changed. If
2130 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2131 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2132 *
2133 * CONTEXT:
2134 * Must be called with tasklist_lock at least read locked.
2135 */
2136static void do_notify_parent_cldstop(struct task_struct *tsk,
2137 bool for_ptracer, int why)
2138{
2139 struct kernel_siginfo info;
2140 unsigned long flags;
2141 struct task_struct *parent;
2142 struct sighand_struct *sighand;
2143 u64 utime, stime;
2144
2145 if (for_ptracer) {
2146 parent = tsk->parent;
2147 } else {
2148 tsk = tsk->group_leader;
2149 parent = tsk->real_parent;
2150 }
2151
2152 clear_siginfo(&info);
2153 info.si_signo = SIGCHLD;
2154 info.si_errno = 0;
2155 /*
2156 * see comment in do_notify_parent() about the following 4 lines
2157 */
2158 rcu_read_lock();
2159 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2160 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2161 rcu_read_unlock();
2162
2163 task_cputime(tsk, &utime, &stime);
2164 info.si_utime = nsec_to_clock_t(utime);
2165 info.si_stime = nsec_to_clock_t(stime);
2166
2167 info.si_code = why;
2168 switch (why) {
2169 case CLD_CONTINUED:
2170 info.si_status = SIGCONT;
2171 break;
2172 case CLD_STOPPED:
2173 info.si_status = tsk->signal->group_exit_code & 0x7f;
2174 break;
2175 case CLD_TRAPPED:
2176 info.si_status = tsk->exit_code & 0x7f;
2177 break;
2178 default:
2179 BUG();
2180 }
2181
2182 sighand = parent->sighand;
2183 spin_lock_irqsave(&sighand->siglock, flags);
2184 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2185 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2186 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2187 /*
2188 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2189 */
2190 __wake_up_parent(tsk, parent);
2191 spin_unlock_irqrestore(&sighand->siglock, flags);
2192}
2193
2194/*
2195 * This must be called with current->sighand->siglock held.
2196 *
2197 * This should be the path for all ptrace stops.
2198 * We always set current->last_siginfo while stopped here.
2199 * That makes it a way to test a stopped process for
2200 * being ptrace-stopped vs being job-control-stopped.
2201 *
2202 * Returns the signal the ptracer requested the code resume
2203 * with. If the code did not stop because the tracer is gone,
2204 * the stop signal remains unchanged unless clear_code.
2205 */
2206static int ptrace_stop(int exit_code, int why, unsigned long message,
2207 kernel_siginfo_t *info)
2208 __releases(¤t->sighand->siglock)
2209 __acquires(¤t->sighand->siglock)
2210{
2211 bool gstop_done = false;
2212
2213 if (arch_ptrace_stop_needed()) {
2214 /*
2215 * The arch code has something special to do before a
2216 * ptrace stop. This is allowed to block, e.g. for faults
2217 * on user stack pages. We can't keep the siglock while
2218 * calling arch_ptrace_stop, so we must release it now.
2219 * To preserve proper semantics, we must do this before
2220 * any signal bookkeeping like checking group_stop_count.
2221 */
2222 spin_unlock_irq(¤t->sighand->siglock);
2223 arch_ptrace_stop();
2224 spin_lock_irq(¤t->sighand->siglock);
2225 }
2226
2227 /*
2228 * After this point ptrace_signal_wake_up or signal_wake_up
2229 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2230 * signal comes in. Handle previous ptrace_unlinks and fatal
2231 * signals here to prevent ptrace_stop sleeping in schedule.
2232 */
2233 if (!current->ptrace || __fatal_signal_pending(current))
2234 return exit_code;
2235
2236 set_special_state(TASK_TRACED);
2237 current->jobctl |= JOBCTL_TRACED;
2238
2239 /*
2240 * We're committing to trapping. TRACED should be visible before
2241 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2242 * Also, transition to TRACED and updates to ->jobctl should be
2243 * atomic with respect to siglock and should be done after the arch
2244 * hook as siglock is released and regrabbed across it.
2245 *
2246 * TRACER TRACEE
2247 *
2248 * ptrace_attach()
2249 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2250 * do_wait()
2251 * set_current_state() smp_wmb();
2252 * ptrace_do_wait()
2253 * wait_task_stopped()
2254 * task_stopped_code()
2255 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2256 */
2257 smp_wmb();
2258
2259 current->ptrace_message = message;
2260 current->last_siginfo = info;
2261 current->exit_code = exit_code;
2262
2263 /*
2264 * If @why is CLD_STOPPED, we're trapping to participate in a group
2265 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2266 * across siglock relocks since INTERRUPT was scheduled, PENDING
2267 * could be clear now. We act as if SIGCONT is received after
2268 * TASK_TRACED is entered - ignore it.
2269 */
2270 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2271 gstop_done = task_participate_group_stop(current);
2272
2273 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2274 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2275 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2276 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2277
2278 /* entering a trap, clear TRAPPING */
2279 task_clear_jobctl_trapping(current);
2280
2281 spin_unlock_irq(¤t->sighand->siglock);
2282 read_lock(&tasklist_lock);
2283 /*
2284 * Notify parents of the stop.
2285 *
2286 * While ptraced, there are two parents - the ptracer and
2287 * the real_parent of the group_leader. The ptracer should
2288 * know about every stop while the real parent is only
2289 * interested in the completion of group stop. The states
2290 * for the two don't interact with each other. Notify
2291 * separately unless they're gonna be duplicates.
2292 */
2293 if (current->ptrace)
2294 do_notify_parent_cldstop(current, true, why);
2295 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2296 do_notify_parent_cldstop(current, false, why);
2297
2298 /*
2299 * Don't want to allow preemption here, because
2300 * sys_ptrace() needs this task to be inactive.
2301 *
2302 * XXX: implement read_unlock_no_resched().
2303 */
2304 preempt_disable();
2305 read_unlock(&tasklist_lock);
2306 cgroup_enter_frozen();
2307 preempt_enable_no_resched();
2308 schedule();
2309 cgroup_leave_frozen(true);
2310
2311 /*
2312 * We are back. Now reacquire the siglock before touching
2313 * last_siginfo, so that we are sure to have synchronized with
2314 * any signal-sending on another CPU that wants to examine it.
2315 */
2316 spin_lock_irq(¤t->sighand->siglock);
2317 exit_code = current->exit_code;
2318 current->last_siginfo = NULL;
2319 current->ptrace_message = 0;
2320 current->exit_code = 0;
2321
2322 /* LISTENING can be set only during STOP traps, clear it */
2323 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2324
2325 /*
2326 * Queued signals ignored us while we were stopped for tracing.
2327 * So check for any that we should take before resuming user mode.
2328 * This sets TIF_SIGPENDING, but never clears it.
2329 */
2330 recalc_sigpending_tsk(current);
2331 return exit_code;
2332}
2333
2334static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2335{
2336 kernel_siginfo_t info;
2337
2338 clear_siginfo(&info);
2339 info.si_signo = signr;
2340 info.si_code = exit_code;
2341 info.si_pid = task_pid_vnr(current);
2342 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2343
2344 /* Let the debugger run. */
2345 return ptrace_stop(exit_code, why, message, &info);
2346}
2347
2348int ptrace_notify(int exit_code, unsigned long message)
2349{
2350 int signr;
2351
2352 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2353 if (unlikely(task_work_pending(current)))
2354 task_work_run();
2355
2356 spin_lock_irq(¤t->sighand->siglock);
2357 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2358 spin_unlock_irq(¤t->sighand->siglock);
2359 return signr;
2360}
2361
2362/**
2363 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2364 * @signr: signr causing group stop if initiating
2365 *
2366 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2367 * and participate in it. If already set, participate in the existing
2368 * group stop. If participated in a group stop (and thus slept), %true is
2369 * returned with siglock released.
2370 *
2371 * If ptraced, this function doesn't handle stop itself. Instead,
2372 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2373 * untouched. The caller must ensure that INTERRUPT trap handling takes
2374 * places afterwards.
2375 *
2376 * CONTEXT:
2377 * Must be called with @current->sighand->siglock held, which is released
2378 * on %true return.
2379 *
2380 * RETURNS:
2381 * %false if group stop is already cancelled or ptrace trap is scheduled.
2382 * %true if participated in group stop.
2383 */
2384static bool do_signal_stop(int signr)
2385 __releases(¤t->sighand->siglock)
2386{
2387 struct signal_struct *sig = current->signal;
2388
2389 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2390 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2391 struct task_struct *t;
2392
2393 /* signr will be recorded in task->jobctl for retries */
2394 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2395
2396 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2397 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2398 unlikely(sig->group_exec_task))
2399 return false;
2400 /*
2401 * There is no group stop already in progress. We must
2402 * initiate one now.
2403 *
2404 * While ptraced, a task may be resumed while group stop is
2405 * still in effect and then receive a stop signal and
2406 * initiate another group stop. This deviates from the
2407 * usual behavior as two consecutive stop signals can't
2408 * cause two group stops when !ptraced. That is why we
2409 * also check !task_is_stopped(t) below.
2410 *
2411 * The condition can be distinguished by testing whether
2412 * SIGNAL_STOP_STOPPED is already set. Don't generate
2413 * group_exit_code in such case.
2414 *
2415 * This is not necessary for SIGNAL_STOP_CONTINUED because
2416 * an intervening stop signal is required to cause two
2417 * continued events regardless of ptrace.
2418 */
2419 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2420 sig->group_exit_code = signr;
2421
2422 sig->group_stop_count = 0;
2423
2424 if (task_set_jobctl_pending(current, signr | gstop))
2425 sig->group_stop_count++;
2426
2427 t = current;
2428 while_each_thread(current, t) {
2429 /*
2430 * Setting state to TASK_STOPPED for a group
2431 * stop is always done with the siglock held,
2432 * so this check has no races.
2433 */
2434 if (!task_is_stopped(t) &&
2435 task_set_jobctl_pending(t, signr | gstop)) {
2436 sig->group_stop_count++;
2437 if (likely(!(t->ptrace & PT_SEIZED)))
2438 signal_wake_up(t, 0);
2439 else
2440 ptrace_trap_notify(t);
2441 }
2442 }
2443 }
2444
2445 if (likely(!current->ptrace)) {
2446 int notify = 0;
2447
2448 /*
2449 * If there are no other threads in the group, or if there
2450 * is a group stop in progress and we are the last to stop,
2451 * report to the parent.
2452 */
2453 if (task_participate_group_stop(current))
2454 notify = CLD_STOPPED;
2455
2456 current->jobctl |= JOBCTL_STOPPED;
2457 set_special_state(TASK_STOPPED);
2458 spin_unlock_irq(¤t->sighand->siglock);
2459
2460 /*
2461 * Notify the parent of the group stop completion. Because
2462 * we're not holding either the siglock or tasklist_lock
2463 * here, ptracer may attach inbetween; however, this is for
2464 * group stop and should always be delivered to the real
2465 * parent of the group leader. The new ptracer will get
2466 * its notification when this task transitions into
2467 * TASK_TRACED.
2468 */
2469 if (notify) {
2470 read_lock(&tasklist_lock);
2471 do_notify_parent_cldstop(current, false, notify);
2472 read_unlock(&tasklist_lock);
2473 }
2474
2475 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2476 cgroup_enter_frozen();
2477 schedule();
2478 return true;
2479 } else {
2480 /*
2481 * While ptraced, group stop is handled by STOP trap.
2482 * Schedule it and let the caller deal with it.
2483 */
2484 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2485 return false;
2486 }
2487}
2488
2489/**
2490 * do_jobctl_trap - take care of ptrace jobctl traps
2491 *
2492 * When PT_SEIZED, it's used for both group stop and explicit
2493 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2494 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2495 * the stop signal; otherwise, %SIGTRAP.
2496 *
2497 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2498 * number as exit_code and no siginfo.
2499 *
2500 * CONTEXT:
2501 * Must be called with @current->sighand->siglock held, which may be
2502 * released and re-acquired before returning with intervening sleep.
2503 */
2504static void do_jobctl_trap(void)
2505{
2506 struct signal_struct *signal = current->signal;
2507 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2508
2509 if (current->ptrace & PT_SEIZED) {
2510 if (!signal->group_stop_count &&
2511 !(signal->flags & SIGNAL_STOP_STOPPED))
2512 signr = SIGTRAP;
2513 WARN_ON_ONCE(!signr);
2514 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2515 CLD_STOPPED, 0);
2516 } else {
2517 WARN_ON_ONCE(!signr);
2518 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2519 }
2520}
2521
2522/**
2523 * do_freezer_trap - handle the freezer jobctl trap
2524 *
2525 * Puts the task into frozen state, if only the task is not about to quit.
2526 * In this case it drops JOBCTL_TRAP_FREEZE.
2527 *
2528 * CONTEXT:
2529 * Must be called with @current->sighand->siglock held,
2530 * which is always released before returning.
2531 */
2532static void do_freezer_trap(void)
2533 __releases(¤t->sighand->siglock)
2534{
2535 /*
2536 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2537 * let's make another loop to give it a chance to be handled.
2538 * In any case, we'll return back.
2539 */
2540 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2541 JOBCTL_TRAP_FREEZE) {
2542 spin_unlock_irq(¤t->sighand->siglock);
2543 return;
2544 }
2545
2546 /*
2547 * Now we're sure that there is no pending fatal signal and no
2548 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2549 * immediately (if there is a non-fatal signal pending), and
2550 * put the task into sleep.
2551 */
2552 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2553 clear_thread_flag(TIF_SIGPENDING);
2554 spin_unlock_irq(¤t->sighand->siglock);
2555 cgroup_enter_frozen();
2556 schedule();
2557}
2558
2559static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2560{
2561 /*
2562 * We do not check sig_kernel_stop(signr) but set this marker
2563 * unconditionally because we do not know whether debugger will
2564 * change signr. This flag has no meaning unless we are going
2565 * to stop after return from ptrace_stop(). In this case it will
2566 * be checked in do_signal_stop(), we should only stop if it was
2567 * not cleared by SIGCONT while we were sleeping. See also the
2568 * comment in dequeue_signal().
2569 */
2570 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2571 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2572
2573 /* We're back. Did the debugger cancel the sig? */
2574 if (signr == 0)
2575 return signr;
2576
2577 /*
2578 * Update the siginfo structure if the signal has
2579 * changed. If the debugger wanted something
2580 * specific in the siginfo structure then it should
2581 * have updated *info via PTRACE_SETSIGINFO.
2582 */
2583 if (signr != info->si_signo) {
2584 clear_siginfo(info);
2585 info->si_signo = signr;
2586 info->si_errno = 0;
2587 info->si_code = SI_USER;
2588 rcu_read_lock();
2589 info->si_pid = task_pid_vnr(current->parent);
2590 info->si_uid = from_kuid_munged(current_user_ns(),
2591 task_uid(current->parent));
2592 rcu_read_unlock();
2593 }
2594
2595 /* If the (new) signal is now blocked, requeue it. */
2596 if (sigismember(¤t->blocked, signr) ||
2597 fatal_signal_pending(current)) {
2598 send_signal_locked(signr, info, current, type);
2599 signr = 0;
2600 }
2601
2602 return signr;
2603}
2604
2605static void hide_si_addr_tag_bits(struct ksignal *ksig)
2606{
2607 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2608 case SIL_FAULT:
2609 case SIL_FAULT_TRAPNO:
2610 case SIL_FAULT_MCEERR:
2611 case SIL_FAULT_BNDERR:
2612 case SIL_FAULT_PKUERR:
2613 case SIL_FAULT_PERF_EVENT:
2614 ksig->info.si_addr = arch_untagged_si_addr(
2615 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2616 break;
2617 case SIL_KILL:
2618 case SIL_TIMER:
2619 case SIL_POLL:
2620 case SIL_CHLD:
2621 case SIL_RT:
2622 case SIL_SYS:
2623 break;
2624 }
2625}
2626
2627bool get_signal(struct ksignal *ksig)
2628{
2629 struct sighand_struct *sighand = current->sighand;
2630 struct signal_struct *signal = current->signal;
2631 int signr;
2632
2633 clear_notify_signal();
2634 if (unlikely(task_work_pending(current)))
2635 task_work_run();
2636
2637 if (!task_sigpending(current))
2638 return false;
2639
2640 if (unlikely(uprobe_deny_signal()))
2641 return false;
2642
2643 /*
2644 * Do this once, we can't return to user-mode if freezing() == T.
2645 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2646 * thus do not need another check after return.
2647 */
2648 try_to_freeze();
2649
2650relock:
2651 spin_lock_irq(&sighand->siglock);
2652
2653 /*
2654 * Every stopped thread goes here after wakeup. Check to see if
2655 * we should notify the parent, prepare_signal(SIGCONT) encodes
2656 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2657 */
2658 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2659 int why;
2660
2661 if (signal->flags & SIGNAL_CLD_CONTINUED)
2662 why = CLD_CONTINUED;
2663 else
2664 why = CLD_STOPPED;
2665
2666 signal->flags &= ~SIGNAL_CLD_MASK;
2667
2668 spin_unlock_irq(&sighand->siglock);
2669
2670 /*
2671 * Notify the parent that we're continuing. This event is
2672 * always per-process and doesn't make whole lot of sense
2673 * for ptracers, who shouldn't consume the state via
2674 * wait(2) either, but, for backward compatibility, notify
2675 * the ptracer of the group leader too unless it's gonna be
2676 * a duplicate.
2677 */
2678 read_lock(&tasklist_lock);
2679 do_notify_parent_cldstop(current, false, why);
2680
2681 if (ptrace_reparented(current->group_leader))
2682 do_notify_parent_cldstop(current->group_leader,
2683 true, why);
2684 read_unlock(&tasklist_lock);
2685
2686 goto relock;
2687 }
2688
2689 for (;;) {
2690 struct k_sigaction *ka;
2691 enum pid_type type;
2692
2693 /* Has this task already been marked for death? */
2694 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2695 signal->group_exec_task) {
2696 clear_siginfo(&ksig->info);
2697 ksig->info.si_signo = signr = SIGKILL;
2698 sigdelset(¤t->pending.signal, SIGKILL);
2699 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2700 &sighand->action[SIGKILL - 1]);
2701 recalc_sigpending();
2702 goto fatal;
2703 }
2704
2705 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2706 do_signal_stop(0))
2707 goto relock;
2708
2709 if (unlikely(current->jobctl &
2710 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2711 if (current->jobctl & JOBCTL_TRAP_MASK) {
2712 do_jobctl_trap();
2713 spin_unlock_irq(&sighand->siglock);
2714 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2715 do_freezer_trap();
2716
2717 goto relock;
2718 }
2719
2720 /*
2721 * If the task is leaving the frozen state, let's update
2722 * cgroup counters and reset the frozen bit.
2723 */
2724 if (unlikely(cgroup_task_frozen(current))) {
2725 spin_unlock_irq(&sighand->siglock);
2726 cgroup_leave_frozen(false);
2727 goto relock;
2728 }
2729
2730 /*
2731 * Signals generated by the execution of an instruction
2732 * need to be delivered before any other pending signals
2733 * so that the instruction pointer in the signal stack
2734 * frame points to the faulting instruction.
2735 */
2736 type = PIDTYPE_PID;
2737 signr = dequeue_synchronous_signal(&ksig->info);
2738 if (!signr)
2739 signr = dequeue_signal(current, ¤t->blocked,
2740 &ksig->info, &type);
2741
2742 if (!signr)
2743 break; /* will return 0 */
2744
2745 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2746 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2747 signr = ptrace_signal(signr, &ksig->info, type);
2748 if (!signr)
2749 continue;
2750 }
2751
2752 ka = &sighand->action[signr-1];
2753
2754 /* Trace actually delivered signals. */
2755 trace_signal_deliver(signr, &ksig->info, ka);
2756
2757 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2758 continue;
2759 if (ka->sa.sa_handler != SIG_DFL) {
2760 /* Run the handler. */
2761 ksig->ka = *ka;
2762
2763 if (ka->sa.sa_flags & SA_ONESHOT)
2764 ka->sa.sa_handler = SIG_DFL;
2765
2766 break; /* will return non-zero "signr" value */
2767 }
2768
2769 /*
2770 * Now we are doing the default action for this signal.
2771 */
2772 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2773 continue;
2774
2775 /*
2776 * Global init gets no signals it doesn't want.
2777 * Container-init gets no signals it doesn't want from same
2778 * container.
2779 *
2780 * Note that if global/container-init sees a sig_kernel_only()
2781 * signal here, the signal must have been generated internally
2782 * or must have come from an ancestor namespace. In either
2783 * case, the signal cannot be dropped.
2784 */
2785 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2786 !sig_kernel_only(signr))
2787 continue;
2788
2789 if (sig_kernel_stop(signr)) {
2790 /*
2791 * The default action is to stop all threads in
2792 * the thread group. The job control signals
2793 * do nothing in an orphaned pgrp, but SIGSTOP
2794 * always works. Note that siglock needs to be
2795 * dropped during the call to is_orphaned_pgrp()
2796 * because of lock ordering with tasklist_lock.
2797 * This allows an intervening SIGCONT to be posted.
2798 * We need to check for that and bail out if necessary.
2799 */
2800 if (signr != SIGSTOP) {
2801 spin_unlock_irq(&sighand->siglock);
2802
2803 /* signals can be posted during this window */
2804
2805 if (is_current_pgrp_orphaned())
2806 goto relock;
2807
2808 spin_lock_irq(&sighand->siglock);
2809 }
2810
2811 if (likely(do_signal_stop(ksig->info.si_signo))) {
2812 /* It released the siglock. */
2813 goto relock;
2814 }
2815
2816 /*
2817 * We didn't actually stop, due to a race
2818 * with SIGCONT or something like that.
2819 */
2820 continue;
2821 }
2822
2823 fatal:
2824 spin_unlock_irq(&sighand->siglock);
2825 if (unlikely(cgroup_task_frozen(current)))
2826 cgroup_leave_frozen(true);
2827
2828 /*
2829 * Anything else is fatal, maybe with a core dump.
2830 */
2831 current->flags |= PF_SIGNALED;
2832
2833 if (sig_kernel_coredump(signr)) {
2834 if (print_fatal_signals)
2835 print_fatal_signal(ksig->info.si_signo);
2836 proc_coredump_connector(current);
2837 /*
2838 * If it was able to dump core, this kills all
2839 * other threads in the group and synchronizes with
2840 * their demise. If we lost the race with another
2841 * thread getting here, it set group_exit_code
2842 * first and our do_group_exit call below will use
2843 * that value and ignore the one we pass it.
2844 */
2845 do_coredump(&ksig->info);
2846 }
2847
2848 /*
2849 * PF_IO_WORKER threads will catch and exit on fatal signals
2850 * themselves. They have cleanup that must be performed, so
2851 * we cannot call do_exit() on their behalf.
2852 */
2853 if (current->flags & PF_IO_WORKER)
2854 goto out;
2855
2856 /*
2857 * Death signals, no core dump.
2858 */
2859 do_group_exit(ksig->info.si_signo);
2860 /* NOTREACHED */
2861 }
2862 spin_unlock_irq(&sighand->siglock);
2863out:
2864 ksig->sig = signr;
2865
2866 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2867 hide_si_addr_tag_bits(ksig);
2868
2869 return ksig->sig > 0;
2870}
2871
2872/**
2873 * signal_delivered - called after signal delivery to update blocked signals
2874 * @ksig: kernel signal struct
2875 * @stepping: nonzero if debugger single-step or block-step in use
2876 *
2877 * This function should be called when a signal has successfully been
2878 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2879 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2880 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2881 */
2882static void signal_delivered(struct ksignal *ksig, int stepping)
2883{
2884 sigset_t blocked;
2885
2886 /* A signal was successfully delivered, and the
2887 saved sigmask was stored on the signal frame,
2888 and will be restored by sigreturn. So we can
2889 simply clear the restore sigmask flag. */
2890 clear_restore_sigmask();
2891
2892 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2893 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2894 sigaddset(&blocked, ksig->sig);
2895 set_current_blocked(&blocked);
2896 if (current->sas_ss_flags & SS_AUTODISARM)
2897 sas_ss_reset(current);
2898 if (stepping)
2899 ptrace_notify(SIGTRAP, 0);
2900}
2901
2902void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2903{
2904 if (failed)
2905 force_sigsegv(ksig->sig);
2906 else
2907 signal_delivered(ksig, stepping);
2908}
2909
2910/*
2911 * It could be that complete_signal() picked us to notify about the
2912 * group-wide signal. Other threads should be notified now to take
2913 * the shared signals in @which since we will not.
2914 */
2915static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2916{
2917 sigset_t retarget;
2918 struct task_struct *t;
2919
2920 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2921 if (sigisemptyset(&retarget))
2922 return;
2923
2924 t = tsk;
2925 while_each_thread(tsk, t) {
2926 if (t->flags & PF_EXITING)
2927 continue;
2928
2929 if (!has_pending_signals(&retarget, &t->blocked))
2930 continue;
2931 /* Remove the signals this thread can handle. */
2932 sigandsets(&retarget, &retarget, &t->blocked);
2933
2934 if (!task_sigpending(t))
2935 signal_wake_up(t, 0);
2936
2937 if (sigisemptyset(&retarget))
2938 break;
2939 }
2940}
2941
2942void exit_signals(struct task_struct *tsk)
2943{
2944 int group_stop = 0;
2945 sigset_t unblocked;
2946
2947 /*
2948 * @tsk is about to have PF_EXITING set - lock out users which
2949 * expect stable threadgroup.
2950 */
2951 cgroup_threadgroup_change_begin(tsk);
2952
2953 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
2954 tsk->flags |= PF_EXITING;
2955 cgroup_threadgroup_change_end(tsk);
2956 return;
2957 }
2958
2959 spin_lock_irq(&tsk->sighand->siglock);
2960 /*
2961 * From now this task is not visible for group-wide signals,
2962 * see wants_signal(), do_signal_stop().
2963 */
2964 tsk->flags |= PF_EXITING;
2965
2966 cgroup_threadgroup_change_end(tsk);
2967
2968 if (!task_sigpending(tsk))
2969 goto out;
2970
2971 unblocked = tsk->blocked;
2972 signotset(&unblocked);
2973 retarget_shared_pending(tsk, &unblocked);
2974
2975 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2976 task_participate_group_stop(tsk))
2977 group_stop = CLD_STOPPED;
2978out:
2979 spin_unlock_irq(&tsk->sighand->siglock);
2980
2981 /*
2982 * If group stop has completed, deliver the notification. This
2983 * should always go to the real parent of the group leader.
2984 */
2985 if (unlikely(group_stop)) {
2986 read_lock(&tasklist_lock);
2987 do_notify_parent_cldstop(tsk, false, group_stop);
2988 read_unlock(&tasklist_lock);
2989 }
2990}
2991
2992/*
2993 * System call entry points.
2994 */
2995
2996/**
2997 * sys_restart_syscall - restart a system call
2998 */
2999SYSCALL_DEFINE0(restart_syscall)
3000{
3001 struct restart_block *restart = ¤t->restart_block;
3002 return restart->fn(restart);
3003}
3004
3005long do_no_restart_syscall(struct restart_block *param)
3006{
3007 return -EINTR;
3008}
3009
3010static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3011{
3012 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3013 sigset_t newblocked;
3014 /* A set of now blocked but previously unblocked signals. */
3015 sigandnsets(&newblocked, newset, ¤t->blocked);
3016 retarget_shared_pending(tsk, &newblocked);
3017 }
3018 tsk->blocked = *newset;
3019 recalc_sigpending();
3020}
3021
3022/**
3023 * set_current_blocked - change current->blocked mask
3024 * @newset: new mask
3025 *
3026 * It is wrong to change ->blocked directly, this helper should be used
3027 * to ensure the process can't miss a shared signal we are going to block.
3028 */
3029void set_current_blocked(sigset_t *newset)
3030{
3031 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3032 __set_current_blocked(newset);
3033}
3034
3035void __set_current_blocked(const sigset_t *newset)
3036{
3037 struct task_struct *tsk = current;
3038
3039 /*
3040 * In case the signal mask hasn't changed, there is nothing we need
3041 * to do. The current->blocked shouldn't be modified by other task.
3042 */
3043 if (sigequalsets(&tsk->blocked, newset))
3044 return;
3045
3046 spin_lock_irq(&tsk->sighand->siglock);
3047 __set_task_blocked(tsk, newset);
3048 spin_unlock_irq(&tsk->sighand->siglock);
3049}
3050
3051/*
3052 * This is also useful for kernel threads that want to temporarily
3053 * (or permanently) block certain signals.
3054 *
3055 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3056 * interface happily blocks "unblockable" signals like SIGKILL
3057 * and friends.
3058 */
3059int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3060{
3061 struct task_struct *tsk = current;
3062 sigset_t newset;
3063
3064 /* Lockless, only current can change ->blocked, never from irq */
3065 if (oldset)
3066 *oldset = tsk->blocked;
3067
3068 switch (how) {
3069 case SIG_BLOCK:
3070 sigorsets(&newset, &tsk->blocked, set);
3071 break;
3072 case SIG_UNBLOCK:
3073 sigandnsets(&newset, &tsk->blocked, set);
3074 break;
3075 case SIG_SETMASK:
3076 newset = *set;
3077 break;
3078 default:
3079 return -EINVAL;
3080 }
3081
3082 __set_current_blocked(&newset);
3083 return 0;
3084}
3085EXPORT_SYMBOL(sigprocmask);
3086
3087/*
3088 * The api helps set app-provided sigmasks.
3089 *
3090 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3091 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3092 *
3093 * Note that it does set_restore_sigmask() in advance, so it must be always
3094 * paired with restore_saved_sigmask_unless() before return from syscall.
3095 */
3096int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3097{
3098 sigset_t kmask;
3099
3100 if (!umask)
3101 return 0;
3102 if (sigsetsize != sizeof(sigset_t))
3103 return -EINVAL;
3104 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3105 return -EFAULT;
3106
3107 set_restore_sigmask();
3108 current->saved_sigmask = current->blocked;
3109 set_current_blocked(&kmask);
3110
3111 return 0;
3112}
3113
3114#ifdef CONFIG_COMPAT
3115int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3116 size_t sigsetsize)
3117{
3118 sigset_t kmask;
3119
3120 if (!umask)
3121 return 0;
3122 if (sigsetsize != sizeof(compat_sigset_t))
3123 return -EINVAL;
3124 if (get_compat_sigset(&kmask, umask))
3125 return -EFAULT;
3126
3127 set_restore_sigmask();
3128 current->saved_sigmask = current->blocked;
3129 set_current_blocked(&kmask);
3130
3131 return 0;
3132}
3133#endif
3134
3135/**
3136 * sys_rt_sigprocmask - change the list of currently blocked signals
3137 * @how: whether to add, remove, or set signals
3138 * @nset: stores pending signals
3139 * @oset: previous value of signal mask if non-null
3140 * @sigsetsize: size of sigset_t type
3141 */
3142SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3143 sigset_t __user *, oset, size_t, sigsetsize)
3144{
3145 sigset_t old_set, new_set;
3146 int error;
3147
3148 /* XXX: Don't preclude handling different sized sigset_t's. */
3149 if (sigsetsize != sizeof(sigset_t))
3150 return -EINVAL;
3151
3152 old_set = current->blocked;
3153
3154 if (nset) {
3155 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3156 return -EFAULT;
3157 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3158
3159 error = sigprocmask(how, &new_set, NULL);
3160 if (error)
3161 return error;
3162 }
3163
3164 if (oset) {
3165 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3166 return -EFAULT;
3167 }
3168
3169 return 0;
3170}
3171
3172#ifdef CONFIG_COMPAT
3173COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3174 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3175{
3176 sigset_t old_set = current->blocked;
3177
3178 /* XXX: Don't preclude handling different sized sigset_t's. */
3179 if (sigsetsize != sizeof(sigset_t))
3180 return -EINVAL;
3181
3182 if (nset) {
3183 sigset_t new_set;
3184 int error;
3185 if (get_compat_sigset(&new_set, nset))
3186 return -EFAULT;
3187 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3188
3189 error = sigprocmask(how, &new_set, NULL);
3190 if (error)
3191 return error;
3192 }
3193 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3194}
3195#endif
3196
3197static void do_sigpending(sigset_t *set)
3198{
3199 spin_lock_irq(¤t->sighand->siglock);
3200 sigorsets(set, ¤t->pending.signal,
3201 ¤t->signal->shared_pending.signal);
3202 spin_unlock_irq(¤t->sighand->siglock);
3203
3204 /* Outside the lock because only this thread touches it. */
3205 sigandsets(set, ¤t->blocked, set);
3206}
3207
3208/**
3209 * sys_rt_sigpending - examine a pending signal that has been raised
3210 * while blocked
3211 * @uset: stores pending signals
3212 * @sigsetsize: size of sigset_t type or larger
3213 */
3214SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3215{
3216 sigset_t set;
3217
3218 if (sigsetsize > sizeof(*uset))
3219 return -EINVAL;
3220
3221 do_sigpending(&set);
3222
3223 if (copy_to_user(uset, &set, sigsetsize))
3224 return -EFAULT;
3225
3226 return 0;
3227}
3228
3229#ifdef CONFIG_COMPAT
3230COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3231 compat_size_t, sigsetsize)
3232{
3233 sigset_t set;
3234
3235 if (sigsetsize > sizeof(*uset))
3236 return -EINVAL;
3237
3238 do_sigpending(&set);
3239
3240 return put_compat_sigset(uset, &set, sigsetsize);
3241}
3242#endif
3243
3244static const struct {
3245 unsigned char limit, layout;
3246} sig_sicodes[] = {
3247 [SIGILL] = { NSIGILL, SIL_FAULT },
3248 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3249 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3250 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3251 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3252#if defined(SIGEMT)
3253 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3254#endif
3255 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3256 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3257 [SIGSYS] = { NSIGSYS, SIL_SYS },
3258};
3259
3260static bool known_siginfo_layout(unsigned sig, int si_code)
3261{
3262 if (si_code == SI_KERNEL)
3263 return true;
3264 else if ((si_code > SI_USER)) {
3265 if (sig_specific_sicodes(sig)) {
3266 if (si_code <= sig_sicodes[sig].limit)
3267 return true;
3268 }
3269 else if (si_code <= NSIGPOLL)
3270 return true;
3271 }
3272 else if (si_code >= SI_DETHREAD)
3273 return true;
3274 else if (si_code == SI_ASYNCNL)
3275 return true;
3276 return false;
3277}
3278
3279enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3280{
3281 enum siginfo_layout layout = SIL_KILL;
3282 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3283 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3284 (si_code <= sig_sicodes[sig].limit)) {
3285 layout = sig_sicodes[sig].layout;
3286 /* Handle the exceptions */
3287 if ((sig == SIGBUS) &&
3288 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3289 layout = SIL_FAULT_MCEERR;
3290 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3291 layout = SIL_FAULT_BNDERR;
3292#ifdef SEGV_PKUERR
3293 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3294 layout = SIL_FAULT_PKUERR;
3295#endif
3296 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3297 layout = SIL_FAULT_PERF_EVENT;
3298 else if (IS_ENABLED(CONFIG_SPARC) &&
3299 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3300 layout = SIL_FAULT_TRAPNO;
3301 else if (IS_ENABLED(CONFIG_ALPHA) &&
3302 ((sig == SIGFPE) ||
3303 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3304 layout = SIL_FAULT_TRAPNO;
3305 }
3306 else if (si_code <= NSIGPOLL)
3307 layout = SIL_POLL;
3308 } else {
3309 if (si_code == SI_TIMER)
3310 layout = SIL_TIMER;
3311 else if (si_code == SI_SIGIO)
3312 layout = SIL_POLL;
3313 else if (si_code < 0)
3314 layout = SIL_RT;
3315 }
3316 return layout;
3317}
3318
3319static inline char __user *si_expansion(const siginfo_t __user *info)
3320{
3321 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3322}
3323
3324int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3325{
3326 char __user *expansion = si_expansion(to);
3327 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3328 return -EFAULT;
3329 if (clear_user(expansion, SI_EXPANSION_SIZE))
3330 return -EFAULT;
3331 return 0;
3332}
3333
3334static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3335 const siginfo_t __user *from)
3336{
3337 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3338 char __user *expansion = si_expansion(from);
3339 char buf[SI_EXPANSION_SIZE];
3340 int i;
3341 /*
3342 * An unknown si_code might need more than
3343 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3344 * extra bytes are 0. This guarantees copy_siginfo_to_user
3345 * will return this data to userspace exactly.
3346 */
3347 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3348 return -EFAULT;
3349 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3350 if (buf[i] != 0)
3351 return -E2BIG;
3352 }
3353 }
3354 return 0;
3355}
3356
3357static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3358 const siginfo_t __user *from)
3359{
3360 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3361 return -EFAULT;
3362 to->si_signo = signo;
3363 return post_copy_siginfo_from_user(to, from);
3364}
3365
3366int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3367{
3368 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3369 return -EFAULT;
3370 return post_copy_siginfo_from_user(to, from);
3371}
3372
3373#ifdef CONFIG_COMPAT
3374/**
3375 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3376 * @to: compat siginfo destination
3377 * @from: kernel siginfo source
3378 *
3379 * Note: This function does not work properly for the SIGCHLD on x32, but
3380 * fortunately it doesn't have to. The only valid callers for this function are
3381 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3382 * The latter does not care because SIGCHLD will never cause a coredump.
3383 */
3384void copy_siginfo_to_external32(struct compat_siginfo *to,
3385 const struct kernel_siginfo *from)
3386{
3387 memset(to, 0, sizeof(*to));
3388
3389 to->si_signo = from->si_signo;
3390 to->si_errno = from->si_errno;
3391 to->si_code = from->si_code;
3392 switch(siginfo_layout(from->si_signo, from->si_code)) {
3393 case SIL_KILL:
3394 to->si_pid = from->si_pid;
3395 to->si_uid = from->si_uid;
3396 break;
3397 case SIL_TIMER:
3398 to->si_tid = from->si_tid;
3399 to->si_overrun = from->si_overrun;
3400 to->si_int = from->si_int;
3401 break;
3402 case SIL_POLL:
3403 to->si_band = from->si_band;
3404 to->si_fd = from->si_fd;
3405 break;
3406 case SIL_FAULT:
3407 to->si_addr = ptr_to_compat(from->si_addr);
3408 break;
3409 case SIL_FAULT_TRAPNO:
3410 to->si_addr = ptr_to_compat(from->si_addr);
3411 to->si_trapno = from->si_trapno;
3412 break;
3413 case SIL_FAULT_MCEERR:
3414 to->si_addr = ptr_to_compat(from->si_addr);
3415 to->si_addr_lsb = from->si_addr_lsb;
3416 break;
3417 case SIL_FAULT_BNDERR:
3418 to->si_addr = ptr_to_compat(from->si_addr);
3419 to->si_lower = ptr_to_compat(from->si_lower);
3420 to->si_upper = ptr_to_compat(from->si_upper);
3421 break;
3422 case SIL_FAULT_PKUERR:
3423 to->si_addr = ptr_to_compat(from->si_addr);
3424 to->si_pkey = from->si_pkey;
3425 break;
3426 case SIL_FAULT_PERF_EVENT:
3427 to->si_addr = ptr_to_compat(from->si_addr);
3428 to->si_perf_data = from->si_perf_data;
3429 to->si_perf_type = from->si_perf_type;
3430 to->si_perf_flags = from->si_perf_flags;
3431 break;
3432 case SIL_CHLD:
3433 to->si_pid = from->si_pid;
3434 to->si_uid = from->si_uid;
3435 to->si_status = from->si_status;
3436 to->si_utime = from->si_utime;
3437 to->si_stime = from->si_stime;
3438 break;
3439 case SIL_RT:
3440 to->si_pid = from->si_pid;
3441 to->si_uid = from->si_uid;
3442 to->si_int = from->si_int;
3443 break;
3444 case SIL_SYS:
3445 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3446 to->si_syscall = from->si_syscall;
3447 to->si_arch = from->si_arch;
3448 break;
3449 }
3450}
3451
3452int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3453 const struct kernel_siginfo *from)
3454{
3455 struct compat_siginfo new;
3456
3457 copy_siginfo_to_external32(&new, from);
3458 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3459 return -EFAULT;
3460 return 0;
3461}
3462
3463static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3464 const struct compat_siginfo *from)
3465{
3466 clear_siginfo(to);
3467 to->si_signo = from->si_signo;
3468 to->si_errno = from->si_errno;
3469 to->si_code = from->si_code;
3470 switch(siginfo_layout(from->si_signo, from->si_code)) {
3471 case SIL_KILL:
3472 to->si_pid = from->si_pid;
3473 to->si_uid = from->si_uid;
3474 break;
3475 case SIL_TIMER:
3476 to->si_tid = from->si_tid;
3477 to->si_overrun = from->si_overrun;
3478 to->si_int = from->si_int;
3479 break;
3480 case SIL_POLL:
3481 to->si_band = from->si_band;
3482 to->si_fd = from->si_fd;
3483 break;
3484 case SIL_FAULT:
3485 to->si_addr = compat_ptr(from->si_addr);
3486 break;
3487 case SIL_FAULT_TRAPNO:
3488 to->si_addr = compat_ptr(from->si_addr);
3489 to->si_trapno = from->si_trapno;
3490 break;
3491 case SIL_FAULT_MCEERR:
3492 to->si_addr = compat_ptr(from->si_addr);
3493 to->si_addr_lsb = from->si_addr_lsb;
3494 break;
3495 case SIL_FAULT_BNDERR:
3496 to->si_addr = compat_ptr(from->si_addr);
3497 to->si_lower = compat_ptr(from->si_lower);
3498 to->si_upper = compat_ptr(from->si_upper);
3499 break;
3500 case SIL_FAULT_PKUERR:
3501 to->si_addr = compat_ptr(from->si_addr);
3502 to->si_pkey = from->si_pkey;
3503 break;
3504 case SIL_FAULT_PERF_EVENT:
3505 to->si_addr = compat_ptr(from->si_addr);
3506 to->si_perf_data = from->si_perf_data;
3507 to->si_perf_type = from->si_perf_type;
3508 to->si_perf_flags = from->si_perf_flags;
3509 break;
3510 case SIL_CHLD:
3511 to->si_pid = from->si_pid;
3512 to->si_uid = from->si_uid;
3513 to->si_status = from->si_status;
3514#ifdef CONFIG_X86_X32_ABI
3515 if (in_x32_syscall()) {
3516 to->si_utime = from->_sifields._sigchld_x32._utime;
3517 to->si_stime = from->_sifields._sigchld_x32._stime;
3518 } else
3519#endif
3520 {
3521 to->si_utime = from->si_utime;
3522 to->si_stime = from->si_stime;
3523 }
3524 break;
3525 case SIL_RT:
3526 to->si_pid = from->si_pid;
3527 to->si_uid = from->si_uid;
3528 to->si_int = from->si_int;
3529 break;
3530 case SIL_SYS:
3531 to->si_call_addr = compat_ptr(from->si_call_addr);
3532 to->si_syscall = from->si_syscall;
3533 to->si_arch = from->si_arch;
3534 break;
3535 }
3536 return 0;
3537}
3538
3539static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3540 const struct compat_siginfo __user *ufrom)
3541{
3542 struct compat_siginfo from;
3543
3544 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3545 return -EFAULT;
3546
3547 from.si_signo = signo;
3548 return post_copy_siginfo_from_user32(to, &from);
3549}
3550
3551int copy_siginfo_from_user32(struct kernel_siginfo *to,
3552 const struct compat_siginfo __user *ufrom)
3553{
3554 struct compat_siginfo from;
3555
3556 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3557 return -EFAULT;
3558
3559 return post_copy_siginfo_from_user32(to, &from);
3560}
3561#endif /* CONFIG_COMPAT */
3562
3563/**
3564 * do_sigtimedwait - wait for queued signals specified in @which
3565 * @which: queued signals to wait for
3566 * @info: if non-null, the signal's siginfo is returned here
3567 * @ts: upper bound on process time suspension
3568 */
3569static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3570 const struct timespec64 *ts)
3571{
3572 ktime_t *to = NULL, timeout = KTIME_MAX;
3573 struct task_struct *tsk = current;
3574 sigset_t mask = *which;
3575 enum pid_type type;
3576 int sig, ret = 0;
3577
3578 if (ts) {
3579 if (!timespec64_valid(ts))
3580 return -EINVAL;
3581 timeout = timespec64_to_ktime(*ts);
3582 to = &timeout;
3583 }
3584
3585 /*
3586 * Invert the set of allowed signals to get those we want to block.
3587 */
3588 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3589 signotset(&mask);
3590
3591 spin_lock_irq(&tsk->sighand->siglock);
3592 sig = dequeue_signal(tsk, &mask, info, &type);
3593 if (!sig && timeout) {
3594 /*
3595 * None ready, temporarily unblock those we're interested
3596 * while we are sleeping in so that we'll be awakened when
3597 * they arrive. Unblocking is always fine, we can avoid
3598 * set_current_blocked().
3599 */
3600 tsk->real_blocked = tsk->blocked;
3601 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3602 recalc_sigpending();
3603 spin_unlock_irq(&tsk->sighand->siglock);
3604
3605 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3606 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3607 HRTIMER_MODE_REL);
3608 spin_lock_irq(&tsk->sighand->siglock);
3609 __set_task_blocked(tsk, &tsk->real_blocked);
3610 sigemptyset(&tsk->real_blocked);
3611 sig = dequeue_signal(tsk, &mask, info, &type);
3612 }
3613 spin_unlock_irq(&tsk->sighand->siglock);
3614
3615 if (sig)
3616 return sig;
3617 return ret ? -EINTR : -EAGAIN;
3618}
3619
3620/**
3621 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3622 * in @uthese
3623 * @uthese: queued signals to wait for
3624 * @uinfo: if non-null, the signal's siginfo is returned here
3625 * @uts: upper bound on process time suspension
3626 * @sigsetsize: size of sigset_t type
3627 */
3628SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3629 siginfo_t __user *, uinfo,
3630 const struct __kernel_timespec __user *, uts,
3631 size_t, sigsetsize)
3632{
3633 sigset_t these;
3634 struct timespec64 ts;
3635 kernel_siginfo_t info;
3636 int ret;
3637
3638 /* XXX: Don't preclude handling different sized sigset_t's. */
3639 if (sigsetsize != sizeof(sigset_t))
3640 return -EINVAL;
3641
3642 if (copy_from_user(&these, uthese, sizeof(these)))
3643 return -EFAULT;
3644
3645 if (uts) {
3646 if (get_timespec64(&ts, uts))
3647 return -EFAULT;
3648 }
3649
3650 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3651
3652 if (ret > 0 && uinfo) {
3653 if (copy_siginfo_to_user(uinfo, &info))
3654 ret = -EFAULT;
3655 }
3656
3657 return ret;
3658}
3659
3660#ifdef CONFIG_COMPAT_32BIT_TIME
3661SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3662 siginfo_t __user *, uinfo,
3663 const struct old_timespec32 __user *, uts,
3664 size_t, sigsetsize)
3665{
3666 sigset_t these;
3667 struct timespec64 ts;
3668 kernel_siginfo_t info;
3669 int ret;
3670
3671 if (sigsetsize != sizeof(sigset_t))
3672 return -EINVAL;
3673
3674 if (copy_from_user(&these, uthese, sizeof(these)))
3675 return -EFAULT;
3676
3677 if (uts) {
3678 if (get_old_timespec32(&ts, uts))
3679 return -EFAULT;
3680 }
3681
3682 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3683
3684 if (ret > 0 && uinfo) {
3685 if (copy_siginfo_to_user(uinfo, &info))
3686 ret = -EFAULT;
3687 }
3688
3689 return ret;
3690}
3691#endif
3692
3693#ifdef CONFIG_COMPAT
3694COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3695 struct compat_siginfo __user *, uinfo,
3696 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3697{
3698 sigset_t s;
3699 struct timespec64 t;
3700 kernel_siginfo_t info;
3701 long ret;
3702
3703 if (sigsetsize != sizeof(sigset_t))
3704 return -EINVAL;
3705
3706 if (get_compat_sigset(&s, uthese))
3707 return -EFAULT;
3708
3709 if (uts) {
3710 if (get_timespec64(&t, uts))
3711 return -EFAULT;
3712 }
3713
3714 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3715
3716 if (ret > 0 && uinfo) {
3717 if (copy_siginfo_to_user32(uinfo, &info))
3718 ret = -EFAULT;
3719 }
3720
3721 return ret;
3722}
3723
3724#ifdef CONFIG_COMPAT_32BIT_TIME
3725COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3726 struct compat_siginfo __user *, uinfo,
3727 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3728{
3729 sigset_t s;
3730 struct timespec64 t;
3731 kernel_siginfo_t info;
3732 long ret;
3733
3734 if (sigsetsize != sizeof(sigset_t))
3735 return -EINVAL;
3736
3737 if (get_compat_sigset(&s, uthese))
3738 return -EFAULT;
3739
3740 if (uts) {
3741 if (get_old_timespec32(&t, uts))
3742 return -EFAULT;
3743 }
3744
3745 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3746
3747 if (ret > 0 && uinfo) {
3748 if (copy_siginfo_to_user32(uinfo, &info))
3749 ret = -EFAULT;
3750 }
3751
3752 return ret;
3753}
3754#endif
3755#endif
3756
3757static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3758{
3759 clear_siginfo(info);
3760 info->si_signo = sig;
3761 info->si_errno = 0;
3762 info->si_code = SI_USER;
3763 info->si_pid = task_tgid_vnr(current);
3764 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3765}
3766
3767/**
3768 * sys_kill - send a signal to a process
3769 * @pid: the PID of the process
3770 * @sig: signal to be sent
3771 */
3772SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3773{
3774 struct kernel_siginfo info;
3775
3776 prepare_kill_siginfo(sig, &info);
3777
3778 return kill_something_info(sig, &info, pid);
3779}
3780
3781/*
3782 * Verify that the signaler and signalee either are in the same pid namespace
3783 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3784 * namespace.
3785 */
3786static bool access_pidfd_pidns(struct pid *pid)
3787{
3788 struct pid_namespace *active = task_active_pid_ns(current);
3789 struct pid_namespace *p = ns_of_pid(pid);
3790
3791 for (;;) {
3792 if (!p)
3793 return false;
3794 if (p == active)
3795 break;
3796 p = p->parent;
3797 }
3798
3799 return true;
3800}
3801
3802static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3803 siginfo_t __user *info)
3804{
3805#ifdef CONFIG_COMPAT
3806 /*
3807 * Avoid hooking up compat syscalls and instead handle necessary
3808 * conversions here. Note, this is a stop-gap measure and should not be
3809 * considered a generic solution.
3810 */
3811 if (in_compat_syscall())
3812 return copy_siginfo_from_user32(
3813 kinfo, (struct compat_siginfo __user *)info);
3814#endif
3815 return copy_siginfo_from_user(kinfo, info);
3816}
3817
3818static struct pid *pidfd_to_pid(const struct file *file)
3819{
3820 struct pid *pid;
3821
3822 pid = pidfd_pid(file);
3823 if (!IS_ERR(pid))
3824 return pid;
3825
3826 return tgid_pidfd_to_pid(file);
3827}
3828
3829/**
3830 * sys_pidfd_send_signal - Signal a process through a pidfd
3831 * @pidfd: file descriptor of the process
3832 * @sig: signal to send
3833 * @info: signal info
3834 * @flags: future flags
3835 *
3836 * The syscall currently only signals via PIDTYPE_PID which covers
3837 * kill(<positive-pid>, <signal>. It does not signal threads or process
3838 * groups.
3839 * In order to extend the syscall to threads and process groups the @flags
3840 * argument should be used. In essence, the @flags argument will determine
3841 * what is signaled and not the file descriptor itself. Put in other words,
3842 * grouping is a property of the flags argument not a property of the file
3843 * descriptor.
3844 *
3845 * Return: 0 on success, negative errno on failure
3846 */
3847SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3848 siginfo_t __user *, info, unsigned int, flags)
3849{
3850 int ret;
3851 struct fd f;
3852 struct pid *pid;
3853 kernel_siginfo_t kinfo;
3854
3855 /* Enforce flags be set to 0 until we add an extension. */
3856 if (flags)
3857 return -EINVAL;
3858
3859 f = fdget(pidfd);
3860 if (!f.file)
3861 return -EBADF;
3862
3863 /* Is this a pidfd? */
3864 pid = pidfd_to_pid(f.file);
3865 if (IS_ERR(pid)) {
3866 ret = PTR_ERR(pid);
3867 goto err;
3868 }
3869
3870 ret = -EINVAL;
3871 if (!access_pidfd_pidns(pid))
3872 goto err;
3873
3874 if (info) {
3875 ret = copy_siginfo_from_user_any(&kinfo, info);
3876 if (unlikely(ret))
3877 goto err;
3878
3879 ret = -EINVAL;
3880 if (unlikely(sig != kinfo.si_signo))
3881 goto err;
3882
3883 /* Only allow sending arbitrary signals to yourself. */
3884 ret = -EPERM;
3885 if ((task_pid(current) != pid) &&
3886 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3887 goto err;
3888 } else {
3889 prepare_kill_siginfo(sig, &kinfo);
3890 }
3891
3892 ret = kill_pid_info(sig, &kinfo, pid);
3893
3894err:
3895 fdput(f);
3896 return ret;
3897}
3898
3899static int
3900do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3901{
3902 struct task_struct *p;
3903 int error = -ESRCH;
3904
3905 rcu_read_lock();
3906 p = find_task_by_vpid(pid);
3907 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3908 error = check_kill_permission(sig, info, p);
3909 /*
3910 * The null signal is a permissions and process existence
3911 * probe. No signal is actually delivered.
3912 */
3913 if (!error && sig) {
3914 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3915 /*
3916 * If lock_task_sighand() failed we pretend the task
3917 * dies after receiving the signal. The window is tiny,
3918 * and the signal is private anyway.
3919 */
3920 if (unlikely(error == -ESRCH))
3921 error = 0;
3922 }
3923 }
3924 rcu_read_unlock();
3925
3926 return error;
3927}
3928
3929static int do_tkill(pid_t tgid, pid_t pid, int sig)
3930{
3931 struct kernel_siginfo info;
3932
3933 clear_siginfo(&info);
3934 info.si_signo = sig;
3935 info.si_errno = 0;
3936 info.si_code = SI_TKILL;
3937 info.si_pid = task_tgid_vnr(current);
3938 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3939
3940 return do_send_specific(tgid, pid, sig, &info);
3941}
3942
3943/**
3944 * sys_tgkill - send signal to one specific thread
3945 * @tgid: the thread group ID of the thread
3946 * @pid: the PID of the thread
3947 * @sig: signal to be sent
3948 *
3949 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3950 * exists but it's not belonging to the target process anymore. This
3951 * method solves the problem of threads exiting and PIDs getting reused.
3952 */
3953SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3954{
3955 /* This is only valid for single tasks */
3956 if (pid <= 0 || tgid <= 0)
3957 return -EINVAL;
3958
3959 return do_tkill(tgid, pid, sig);
3960}
3961
3962/**
3963 * sys_tkill - send signal to one specific task
3964 * @pid: the PID of the task
3965 * @sig: signal to be sent
3966 *
3967 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3968 */
3969SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3970{
3971 /* This is only valid for single tasks */
3972 if (pid <= 0)
3973 return -EINVAL;
3974
3975 return do_tkill(0, pid, sig);
3976}
3977
3978static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3979{
3980 /* Not even root can pretend to send signals from the kernel.
3981 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3982 */
3983 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3984 (task_pid_vnr(current) != pid))
3985 return -EPERM;
3986
3987 /* POSIX.1b doesn't mention process groups. */
3988 return kill_proc_info(sig, info, pid);
3989}
3990
3991/**
3992 * sys_rt_sigqueueinfo - send signal information to a signal
3993 * @pid: the PID of the thread
3994 * @sig: signal to be sent
3995 * @uinfo: signal info to be sent
3996 */
3997SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3998 siginfo_t __user *, uinfo)
3999{
4000 kernel_siginfo_t info;
4001 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4002 if (unlikely(ret))
4003 return ret;
4004 return do_rt_sigqueueinfo(pid, sig, &info);
4005}
4006
4007#ifdef CONFIG_COMPAT
4008COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4009 compat_pid_t, pid,
4010 int, sig,
4011 struct compat_siginfo __user *, uinfo)
4012{
4013 kernel_siginfo_t info;
4014 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4015 if (unlikely(ret))
4016 return ret;
4017 return do_rt_sigqueueinfo(pid, sig, &info);
4018}
4019#endif
4020
4021static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4022{
4023 /* This is only valid for single tasks */
4024 if (pid <= 0 || tgid <= 0)
4025 return -EINVAL;
4026
4027 /* Not even root can pretend to send signals from the kernel.
4028 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4029 */
4030 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4031 (task_pid_vnr(current) != pid))
4032 return -EPERM;
4033
4034 return do_send_specific(tgid, pid, sig, info);
4035}
4036
4037SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4038 siginfo_t __user *, uinfo)
4039{
4040 kernel_siginfo_t info;
4041 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4042 if (unlikely(ret))
4043 return ret;
4044 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4045}
4046
4047#ifdef CONFIG_COMPAT
4048COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4049 compat_pid_t, tgid,
4050 compat_pid_t, pid,
4051 int, sig,
4052 struct compat_siginfo __user *, uinfo)
4053{
4054 kernel_siginfo_t info;
4055 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4056 if (unlikely(ret))
4057 return ret;
4058 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4059}
4060#endif
4061
4062/*
4063 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4064 */
4065void kernel_sigaction(int sig, __sighandler_t action)
4066{
4067 spin_lock_irq(¤t->sighand->siglock);
4068 current->sighand->action[sig - 1].sa.sa_handler = action;
4069 if (action == SIG_IGN) {
4070 sigset_t mask;
4071
4072 sigemptyset(&mask);
4073 sigaddset(&mask, sig);
4074
4075 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
4076 flush_sigqueue_mask(&mask, ¤t->pending);
4077 recalc_sigpending();
4078 }
4079 spin_unlock_irq(¤t->sighand->siglock);
4080}
4081EXPORT_SYMBOL(kernel_sigaction);
4082
4083void __weak sigaction_compat_abi(struct k_sigaction *act,
4084 struct k_sigaction *oact)
4085{
4086}
4087
4088int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4089{
4090 struct task_struct *p = current, *t;
4091 struct k_sigaction *k;
4092 sigset_t mask;
4093
4094 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4095 return -EINVAL;
4096
4097 k = &p->sighand->action[sig-1];
4098
4099 spin_lock_irq(&p->sighand->siglock);
4100 if (k->sa.sa_flags & SA_IMMUTABLE) {
4101 spin_unlock_irq(&p->sighand->siglock);
4102 return -EINVAL;
4103 }
4104 if (oact)
4105 *oact = *k;
4106
4107 /*
4108 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4109 * e.g. by having an architecture use the bit in their uapi.
4110 */
4111 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4112
4113 /*
4114 * Clear unknown flag bits in order to allow userspace to detect missing
4115 * support for flag bits and to allow the kernel to use non-uapi bits
4116 * internally.
4117 */
4118 if (act)
4119 act->sa.sa_flags &= UAPI_SA_FLAGS;
4120 if (oact)
4121 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4122
4123 sigaction_compat_abi(act, oact);
4124
4125 if (act) {
4126 sigdelsetmask(&act->sa.sa_mask,
4127 sigmask(SIGKILL) | sigmask(SIGSTOP));
4128 *k = *act;
4129 /*
4130 * POSIX 3.3.1.3:
4131 * "Setting a signal action to SIG_IGN for a signal that is
4132 * pending shall cause the pending signal to be discarded,
4133 * whether or not it is blocked."
4134 *
4135 * "Setting a signal action to SIG_DFL for a signal that is
4136 * pending and whose default action is to ignore the signal
4137 * (for example, SIGCHLD), shall cause the pending signal to
4138 * be discarded, whether or not it is blocked"
4139 */
4140 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4141 sigemptyset(&mask);
4142 sigaddset(&mask, sig);
4143 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4144 for_each_thread(p, t)
4145 flush_sigqueue_mask(&mask, &t->pending);
4146 }
4147 }
4148
4149 spin_unlock_irq(&p->sighand->siglock);
4150 return 0;
4151}
4152
4153#ifdef CONFIG_DYNAMIC_SIGFRAME
4154static inline void sigaltstack_lock(void)
4155 __acquires(¤t->sighand->siglock)
4156{
4157 spin_lock_irq(¤t->sighand->siglock);
4158}
4159
4160static inline void sigaltstack_unlock(void)
4161 __releases(¤t->sighand->siglock)
4162{
4163 spin_unlock_irq(¤t->sighand->siglock);
4164}
4165#else
4166static inline void sigaltstack_lock(void) { }
4167static inline void sigaltstack_unlock(void) { }
4168#endif
4169
4170static int
4171do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4172 size_t min_ss_size)
4173{
4174 struct task_struct *t = current;
4175 int ret = 0;
4176
4177 if (oss) {
4178 memset(oss, 0, sizeof(stack_t));
4179 oss->ss_sp = (void __user *) t->sas_ss_sp;
4180 oss->ss_size = t->sas_ss_size;
4181 oss->ss_flags = sas_ss_flags(sp) |
4182 (current->sas_ss_flags & SS_FLAG_BITS);
4183 }
4184
4185 if (ss) {
4186 void __user *ss_sp = ss->ss_sp;
4187 size_t ss_size = ss->ss_size;
4188 unsigned ss_flags = ss->ss_flags;
4189 int ss_mode;
4190
4191 if (unlikely(on_sig_stack(sp)))
4192 return -EPERM;
4193
4194 ss_mode = ss_flags & ~SS_FLAG_BITS;
4195 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4196 ss_mode != 0))
4197 return -EINVAL;
4198
4199 /*
4200 * Return before taking any locks if no actual
4201 * sigaltstack changes were requested.
4202 */
4203 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4204 t->sas_ss_size == ss_size &&
4205 t->sas_ss_flags == ss_flags)
4206 return 0;
4207
4208 sigaltstack_lock();
4209 if (ss_mode == SS_DISABLE) {
4210 ss_size = 0;
4211 ss_sp = NULL;
4212 } else {
4213 if (unlikely(ss_size < min_ss_size))
4214 ret = -ENOMEM;
4215 if (!sigaltstack_size_valid(ss_size))
4216 ret = -ENOMEM;
4217 }
4218 if (!ret) {
4219 t->sas_ss_sp = (unsigned long) ss_sp;
4220 t->sas_ss_size = ss_size;
4221 t->sas_ss_flags = ss_flags;
4222 }
4223 sigaltstack_unlock();
4224 }
4225 return ret;
4226}
4227
4228SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4229{
4230 stack_t new, old;
4231 int err;
4232 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4233 return -EFAULT;
4234 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4235 current_user_stack_pointer(),
4236 MINSIGSTKSZ);
4237 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4238 err = -EFAULT;
4239 return err;
4240}
4241
4242int restore_altstack(const stack_t __user *uss)
4243{
4244 stack_t new;
4245 if (copy_from_user(&new, uss, sizeof(stack_t)))
4246 return -EFAULT;
4247 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4248 MINSIGSTKSZ);
4249 /* squash all but EFAULT for now */
4250 return 0;
4251}
4252
4253int __save_altstack(stack_t __user *uss, unsigned long sp)
4254{
4255 struct task_struct *t = current;
4256 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4257 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4258 __put_user(t->sas_ss_size, &uss->ss_size);
4259 return err;
4260}
4261
4262#ifdef CONFIG_COMPAT
4263static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4264 compat_stack_t __user *uoss_ptr)
4265{
4266 stack_t uss, uoss;
4267 int ret;
4268
4269 if (uss_ptr) {
4270 compat_stack_t uss32;
4271 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4272 return -EFAULT;
4273 uss.ss_sp = compat_ptr(uss32.ss_sp);
4274 uss.ss_flags = uss32.ss_flags;
4275 uss.ss_size = uss32.ss_size;
4276 }
4277 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4278 compat_user_stack_pointer(),
4279 COMPAT_MINSIGSTKSZ);
4280 if (ret >= 0 && uoss_ptr) {
4281 compat_stack_t old;
4282 memset(&old, 0, sizeof(old));
4283 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4284 old.ss_flags = uoss.ss_flags;
4285 old.ss_size = uoss.ss_size;
4286 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4287 ret = -EFAULT;
4288 }
4289 return ret;
4290}
4291
4292COMPAT_SYSCALL_DEFINE2(sigaltstack,
4293 const compat_stack_t __user *, uss_ptr,
4294 compat_stack_t __user *, uoss_ptr)
4295{
4296 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4297}
4298
4299int compat_restore_altstack(const compat_stack_t __user *uss)
4300{
4301 int err = do_compat_sigaltstack(uss, NULL);
4302 /* squash all but -EFAULT for now */
4303 return err == -EFAULT ? err : 0;
4304}
4305
4306int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4307{
4308 int err;
4309 struct task_struct *t = current;
4310 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4311 &uss->ss_sp) |
4312 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4313 __put_user(t->sas_ss_size, &uss->ss_size);
4314 return err;
4315}
4316#endif
4317
4318#ifdef __ARCH_WANT_SYS_SIGPENDING
4319
4320/**
4321 * sys_sigpending - examine pending signals
4322 * @uset: where mask of pending signal is returned
4323 */
4324SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4325{
4326 sigset_t set;
4327
4328 if (sizeof(old_sigset_t) > sizeof(*uset))
4329 return -EINVAL;
4330
4331 do_sigpending(&set);
4332
4333 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4334 return -EFAULT;
4335
4336 return 0;
4337}
4338
4339#ifdef CONFIG_COMPAT
4340COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4341{
4342 sigset_t set;
4343
4344 do_sigpending(&set);
4345
4346 return put_user(set.sig[0], set32);
4347}
4348#endif
4349
4350#endif
4351
4352#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4353/**
4354 * sys_sigprocmask - examine and change blocked signals
4355 * @how: whether to add, remove, or set signals
4356 * @nset: signals to add or remove (if non-null)
4357 * @oset: previous value of signal mask if non-null
4358 *
4359 * Some platforms have their own version with special arguments;
4360 * others support only sys_rt_sigprocmask.
4361 */
4362
4363SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4364 old_sigset_t __user *, oset)
4365{
4366 old_sigset_t old_set, new_set;
4367 sigset_t new_blocked;
4368
4369 old_set = current->blocked.sig[0];
4370
4371 if (nset) {
4372 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4373 return -EFAULT;
4374
4375 new_blocked = current->blocked;
4376
4377 switch (how) {
4378 case SIG_BLOCK:
4379 sigaddsetmask(&new_blocked, new_set);
4380 break;
4381 case SIG_UNBLOCK:
4382 sigdelsetmask(&new_blocked, new_set);
4383 break;
4384 case SIG_SETMASK:
4385 new_blocked.sig[0] = new_set;
4386 break;
4387 default:
4388 return -EINVAL;
4389 }
4390
4391 set_current_blocked(&new_blocked);
4392 }
4393
4394 if (oset) {
4395 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4396 return -EFAULT;
4397 }
4398
4399 return 0;
4400}
4401#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4402
4403#ifndef CONFIG_ODD_RT_SIGACTION
4404/**
4405 * sys_rt_sigaction - alter an action taken by a process
4406 * @sig: signal to be sent
4407 * @act: new sigaction
4408 * @oact: used to save the previous sigaction
4409 * @sigsetsize: size of sigset_t type
4410 */
4411SYSCALL_DEFINE4(rt_sigaction, int, sig,
4412 const struct sigaction __user *, act,
4413 struct sigaction __user *, oact,
4414 size_t, sigsetsize)
4415{
4416 struct k_sigaction new_sa, old_sa;
4417 int ret;
4418
4419 /* XXX: Don't preclude handling different sized sigset_t's. */
4420 if (sigsetsize != sizeof(sigset_t))
4421 return -EINVAL;
4422
4423 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4424 return -EFAULT;
4425
4426 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4427 if (ret)
4428 return ret;
4429
4430 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4431 return -EFAULT;
4432
4433 return 0;
4434}
4435#ifdef CONFIG_COMPAT
4436COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4437 const struct compat_sigaction __user *, act,
4438 struct compat_sigaction __user *, oact,
4439 compat_size_t, sigsetsize)
4440{
4441 struct k_sigaction new_ka, old_ka;
4442#ifdef __ARCH_HAS_SA_RESTORER
4443 compat_uptr_t restorer;
4444#endif
4445 int ret;
4446
4447 /* XXX: Don't preclude handling different sized sigset_t's. */
4448 if (sigsetsize != sizeof(compat_sigset_t))
4449 return -EINVAL;
4450
4451 if (act) {
4452 compat_uptr_t handler;
4453 ret = get_user(handler, &act->sa_handler);
4454 new_ka.sa.sa_handler = compat_ptr(handler);
4455#ifdef __ARCH_HAS_SA_RESTORER
4456 ret |= get_user(restorer, &act->sa_restorer);
4457 new_ka.sa.sa_restorer = compat_ptr(restorer);
4458#endif
4459 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4460 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4461 if (ret)
4462 return -EFAULT;
4463 }
4464
4465 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4466 if (!ret && oact) {
4467 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4468 &oact->sa_handler);
4469 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4470 sizeof(oact->sa_mask));
4471 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4472#ifdef __ARCH_HAS_SA_RESTORER
4473 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4474 &oact->sa_restorer);
4475#endif
4476 }
4477 return ret;
4478}
4479#endif
4480#endif /* !CONFIG_ODD_RT_SIGACTION */
4481
4482#ifdef CONFIG_OLD_SIGACTION
4483SYSCALL_DEFINE3(sigaction, int, sig,
4484 const struct old_sigaction __user *, act,
4485 struct old_sigaction __user *, oact)
4486{
4487 struct k_sigaction new_ka, old_ka;
4488 int ret;
4489
4490 if (act) {
4491 old_sigset_t mask;
4492 if (!access_ok(act, sizeof(*act)) ||
4493 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4494 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4495 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4496 __get_user(mask, &act->sa_mask))
4497 return -EFAULT;
4498#ifdef __ARCH_HAS_KA_RESTORER
4499 new_ka.ka_restorer = NULL;
4500#endif
4501 siginitset(&new_ka.sa.sa_mask, mask);
4502 }
4503
4504 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4505
4506 if (!ret && oact) {
4507 if (!access_ok(oact, sizeof(*oact)) ||
4508 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4509 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4510 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4511 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4512 return -EFAULT;
4513 }
4514
4515 return ret;
4516}
4517#endif
4518#ifdef CONFIG_COMPAT_OLD_SIGACTION
4519COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4520 const struct compat_old_sigaction __user *, act,
4521 struct compat_old_sigaction __user *, oact)
4522{
4523 struct k_sigaction new_ka, old_ka;
4524 int ret;
4525 compat_old_sigset_t mask;
4526 compat_uptr_t handler, restorer;
4527
4528 if (act) {
4529 if (!access_ok(act, sizeof(*act)) ||
4530 __get_user(handler, &act->sa_handler) ||
4531 __get_user(restorer, &act->sa_restorer) ||
4532 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4533 __get_user(mask, &act->sa_mask))
4534 return -EFAULT;
4535
4536#ifdef __ARCH_HAS_KA_RESTORER
4537 new_ka.ka_restorer = NULL;
4538#endif
4539 new_ka.sa.sa_handler = compat_ptr(handler);
4540 new_ka.sa.sa_restorer = compat_ptr(restorer);
4541 siginitset(&new_ka.sa.sa_mask, mask);
4542 }
4543
4544 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4545
4546 if (!ret && oact) {
4547 if (!access_ok(oact, sizeof(*oact)) ||
4548 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4549 &oact->sa_handler) ||
4550 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4551 &oact->sa_restorer) ||
4552 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4553 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4554 return -EFAULT;
4555 }
4556 return ret;
4557}
4558#endif
4559
4560#ifdef CONFIG_SGETMASK_SYSCALL
4561
4562/*
4563 * For backwards compatibility. Functionality superseded by sigprocmask.
4564 */
4565SYSCALL_DEFINE0(sgetmask)
4566{
4567 /* SMP safe */
4568 return current->blocked.sig[0];
4569}
4570
4571SYSCALL_DEFINE1(ssetmask, int, newmask)
4572{
4573 int old = current->blocked.sig[0];
4574 sigset_t newset;
4575
4576 siginitset(&newset, newmask);
4577 set_current_blocked(&newset);
4578
4579 return old;
4580}
4581#endif /* CONFIG_SGETMASK_SYSCALL */
4582
4583#ifdef __ARCH_WANT_SYS_SIGNAL
4584/*
4585 * For backwards compatibility. Functionality superseded by sigaction.
4586 */
4587SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4588{
4589 struct k_sigaction new_sa, old_sa;
4590 int ret;
4591
4592 new_sa.sa.sa_handler = handler;
4593 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4594 sigemptyset(&new_sa.sa.sa_mask);
4595
4596 ret = do_sigaction(sig, &new_sa, &old_sa);
4597
4598 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4599}
4600#endif /* __ARCH_WANT_SYS_SIGNAL */
4601
4602#ifdef __ARCH_WANT_SYS_PAUSE
4603
4604SYSCALL_DEFINE0(pause)
4605{
4606 while (!signal_pending(current)) {
4607 __set_current_state(TASK_INTERRUPTIBLE);
4608 schedule();
4609 }
4610 return -ERESTARTNOHAND;
4611}
4612
4613#endif
4614
4615static int sigsuspend(sigset_t *set)
4616{
4617 current->saved_sigmask = current->blocked;
4618 set_current_blocked(set);
4619
4620 while (!signal_pending(current)) {
4621 __set_current_state(TASK_INTERRUPTIBLE);
4622 schedule();
4623 }
4624 set_restore_sigmask();
4625 return -ERESTARTNOHAND;
4626}
4627
4628/**
4629 * sys_rt_sigsuspend - replace the signal mask for a value with the
4630 * @unewset value until a signal is received
4631 * @unewset: new signal mask value
4632 * @sigsetsize: size of sigset_t type
4633 */
4634SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4635{
4636 sigset_t newset;
4637
4638 /* XXX: Don't preclude handling different sized sigset_t's. */
4639 if (sigsetsize != sizeof(sigset_t))
4640 return -EINVAL;
4641
4642 if (copy_from_user(&newset, unewset, sizeof(newset)))
4643 return -EFAULT;
4644 return sigsuspend(&newset);
4645}
4646
4647#ifdef CONFIG_COMPAT
4648COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4649{
4650 sigset_t newset;
4651
4652 /* XXX: Don't preclude handling different sized sigset_t's. */
4653 if (sigsetsize != sizeof(sigset_t))
4654 return -EINVAL;
4655
4656 if (get_compat_sigset(&newset, unewset))
4657 return -EFAULT;
4658 return sigsuspend(&newset);
4659}
4660#endif
4661
4662#ifdef CONFIG_OLD_SIGSUSPEND
4663SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4664{
4665 sigset_t blocked;
4666 siginitset(&blocked, mask);
4667 return sigsuspend(&blocked);
4668}
4669#endif
4670#ifdef CONFIG_OLD_SIGSUSPEND3
4671SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4672{
4673 sigset_t blocked;
4674 siginitset(&blocked, mask);
4675 return sigsuspend(&blocked);
4676}
4677#endif
4678
4679__weak const char *arch_vma_name(struct vm_area_struct *vma)
4680{
4681 return NULL;
4682}
4683
4684static inline void siginfo_buildtime_checks(void)
4685{
4686 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4687
4688 /* Verify the offsets in the two siginfos match */
4689#define CHECK_OFFSET(field) \
4690 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4691
4692 /* kill */
4693 CHECK_OFFSET(si_pid);
4694 CHECK_OFFSET(si_uid);
4695
4696 /* timer */
4697 CHECK_OFFSET(si_tid);
4698 CHECK_OFFSET(si_overrun);
4699 CHECK_OFFSET(si_value);
4700
4701 /* rt */
4702 CHECK_OFFSET(si_pid);
4703 CHECK_OFFSET(si_uid);
4704 CHECK_OFFSET(si_value);
4705
4706 /* sigchld */
4707 CHECK_OFFSET(si_pid);
4708 CHECK_OFFSET(si_uid);
4709 CHECK_OFFSET(si_status);
4710 CHECK_OFFSET(si_utime);
4711 CHECK_OFFSET(si_stime);
4712
4713 /* sigfault */
4714 CHECK_OFFSET(si_addr);
4715 CHECK_OFFSET(si_trapno);
4716 CHECK_OFFSET(si_addr_lsb);
4717 CHECK_OFFSET(si_lower);
4718 CHECK_OFFSET(si_upper);
4719 CHECK_OFFSET(si_pkey);
4720 CHECK_OFFSET(si_perf_data);
4721 CHECK_OFFSET(si_perf_type);
4722 CHECK_OFFSET(si_perf_flags);
4723
4724 /* sigpoll */
4725 CHECK_OFFSET(si_band);
4726 CHECK_OFFSET(si_fd);
4727
4728 /* sigsys */
4729 CHECK_OFFSET(si_call_addr);
4730 CHECK_OFFSET(si_syscall);
4731 CHECK_OFFSET(si_arch);
4732#undef CHECK_OFFSET
4733
4734 /* usb asyncio */
4735 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4736 offsetof(struct siginfo, si_addr));
4737 if (sizeof(int) == sizeof(void __user *)) {
4738 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4739 sizeof(void __user *));
4740 } else {
4741 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4742 sizeof_field(struct siginfo, si_uid)) !=
4743 sizeof(void __user *));
4744 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4745 offsetof(struct siginfo, si_uid));
4746 }
4747#ifdef CONFIG_COMPAT
4748 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4749 offsetof(struct compat_siginfo, si_addr));
4750 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4751 sizeof(compat_uptr_t));
4752 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4753 sizeof_field(struct siginfo, si_pid));
4754#endif
4755}
4756
4757void __init signals_init(void)
4758{
4759 siginfo_buildtime_checks();
4760
4761 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4762}
4763
4764#ifdef CONFIG_KGDB_KDB
4765#include <linux/kdb.h>
4766/*
4767 * kdb_send_sig - Allows kdb to send signals without exposing
4768 * signal internals. This function checks if the required locks are
4769 * available before calling the main signal code, to avoid kdb
4770 * deadlocks.
4771 */
4772void kdb_send_sig(struct task_struct *t, int sig)
4773{
4774 static struct task_struct *kdb_prev_t;
4775 int new_t, ret;
4776 if (!spin_trylock(&t->sighand->siglock)) {
4777 kdb_printf("Can't do kill command now.\n"
4778 "The sigmask lock is held somewhere else in "
4779 "kernel, try again later\n");
4780 return;
4781 }
4782 new_t = kdb_prev_t != t;
4783 kdb_prev_t = t;
4784 if (!task_is_running(t) && new_t) {
4785 spin_unlock(&t->sighand->siglock);
4786 kdb_printf("Process is not RUNNING, sending a signal from "
4787 "kdb risks deadlock\n"
4788 "on the run queue locks. "
4789 "The signal has _not_ been sent.\n"
4790 "Reissue the kill command if you want to risk "
4791 "the deadlock.\n");
4792 return;
4793 }
4794 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4795 spin_unlock(&t->sighand->siglock);
4796 if (ret)
4797 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4798 sig, t->pid);
4799 else
4800 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4801}
4802#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
417 /*
418 * Protect access to @t credentials. This can go away when all
419 * callers hold rcu read lock.
420 */
421 rcu_read_lock();
422 user = get_uid(__task_cred(t)->user);
423 atomic_inc(&user->sigpending);
424 rcu_read_unlock();
425
426 if (override_rlimit ||
427 atomic_read(&user->sigpending) <=
428 task_rlimit(t, RLIMIT_SIGPENDING)) {
429 q = kmem_cache_alloc(sigqueue_cachep, flags);
430 } else {
431 print_dropped_signal(sig);
432 }
433
434 if (unlikely(q == NULL)) {
435 atomic_dec(&user->sigpending);
436 free_uid(user);
437 } else {
438 INIT_LIST_HEAD(&q->list);
439 q->flags = 0;
440 q->user = user;
441 }
442
443 return q;
444}
445
446static void __sigqueue_free(struct sigqueue *q)
447{
448 if (q->flags & SIGQUEUE_PREALLOC)
449 return;
450 atomic_dec(&q->user->sigpending);
451 free_uid(q->user);
452 kmem_cache_free(sigqueue_cachep, q);
453}
454
455void flush_sigqueue(struct sigpending *queue)
456{
457 struct sigqueue *q;
458
459 sigemptyset(&queue->signal);
460 while (!list_empty(&queue->list)) {
461 q = list_entry(queue->list.next, struct sigqueue , list);
462 list_del_init(&q->list);
463 __sigqueue_free(q);
464 }
465}
466
467/*
468 * Flush all pending signals for this kthread.
469 */
470void flush_signals(struct task_struct *t)
471{
472 unsigned long flags;
473
474 spin_lock_irqsave(&t->sighand->siglock, flags);
475 clear_tsk_thread_flag(t, TIF_SIGPENDING);
476 flush_sigqueue(&t->pending);
477 flush_sigqueue(&t->signal->shared_pending);
478 spin_unlock_irqrestore(&t->sighand->siglock, flags);
479}
480EXPORT_SYMBOL(flush_signals);
481
482#ifdef CONFIG_POSIX_TIMERS
483static void __flush_itimer_signals(struct sigpending *pending)
484{
485 sigset_t signal, retain;
486 struct sigqueue *q, *n;
487
488 signal = pending->signal;
489 sigemptyset(&retain);
490
491 list_for_each_entry_safe(q, n, &pending->list, list) {
492 int sig = q->info.si_signo;
493
494 if (likely(q->info.si_code != SI_TIMER)) {
495 sigaddset(&retain, sig);
496 } else {
497 sigdelset(&signal, sig);
498 list_del_init(&q->list);
499 __sigqueue_free(q);
500 }
501 }
502
503 sigorsets(&pending->signal, &signal, &retain);
504}
505
506void flush_itimer_signals(void)
507{
508 struct task_struct *tsk = current;
509 unsigned long flags;
510
511 spin_lock_irqsave(&tsk->sighand->siglock, flags);
512 __flush_itimer_signals(&tsk->pending);
513 __flush_itimer_signals(&tsk->signal->shared_pending);
514 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
515}
516#endif
517
518void ignore_signals(struct task_struct *t)
519{
520 int i;
521
522 for (i = 0; i < _NSIG; ++i)
523 t->sighand->action[i].sa.sa_handler = SIG_IGN;
524
525 flush_signals(t);
526}
527
528/*
529 * Flush all handlers for a task.
530 */
531
532void
533flush_signal_handlers(struct task_struct *t, int force_default)
534{
535 int i;
536 struct k_sigaction *ka = &t->sighand->action[0];
537 for (i = _NSIG ; i != 0 ; i--) {
538 if (force_default || ka->sa.sa_handler != SIG_IGN)
539 ka->sa.sa_handler = SIG_DFL;
540 ka->sa.sa_flags = 0;
541#ifdef __ARCH_HAS_SA_RESTORER
542 ka->sa.sa_restorer = NULL;
543#endif
544 sigemptyset(&ka->sa.sa_mask);
545 ka++;
546 }
547}
548
549bool unhandled_signal(struct task_struct *tsk, int sig)
550{
551 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
552 if (is_global_init(tsk))
553 return true;
554
555 if (handler != SIG_IGN && handler != SIG_DFL)
556 return false;
557
558 /* if ptraced, let the tracer determine */
559 return !tsk->ptrace;
560}
561
562static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
563 bool *resched_timer)
564{
565 struct sigqueue *q, *first = NULL;
566
567 /*
568 * Collect the siginfo appropriate to this signal. Check if
569 * there is another siginfo for the same signal.
570 */
571 list_for_each_entry(q, &list->list, list) {
572 if (q->info.si_signo == sig) {
573 if (first)
574 goto still_pending;
575 first = q;
576 }
577 }
578
579 sigdelset(&list->signal, sig);
580
581 if (first) {
582still_pending:
583 list_del_init(&first->list);
584 copy_siginfo(info, &first->info);
585
586 *resched_timer =
587 (first->flags & SIGQUEUE_PREALLOC) &&
588 (info->si_code == SI_TIMER) &&
589 (info->si_sys_private);
590
591 __sigqueue_free(first);
592 } else {
593 /*
594 * Ok, it wasn't in the queue. This must be
595 * a fast-pathed signal or we must have been
596 * out of queue space. So zero out the info.
597 */
598 clear_siginfo(info);
599 info->si_signo = sig;
600 info->si_errno = 0;
601 info->si_code = SI_USER;
602 info->si_pid = 0;
603 info->si_uid = 0;
604 }
605}
606
607static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
608 kernel_siginfo_t *info, bool *resched_timer)
609{
610 int sig = next_signal(pending, mask);
611
612 if (sig)
613 collect_signal(sig, pending, info, resched_timer);
614 return sig;
615}
616
617/*
618 * Dequeue a signal and return the element to the caller, which is
619 * expected to free it.
620 *
621 * All callers have to hold the siglock.
622 */
623int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
624{
625 bool resched_timer = false;
626 int signr;
627
628 /* We only dequeue private signals from ourselves, we don't let
629 * signalfd steal them
630 */
631 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
632 if (!signr) {
633 signr = __dequeue_signal(&tsk->signal->shared_pending,
634 mask, info, &resched_timer);
635#ifdef CONFIG_POSIX_TIMERS
636 /*
637 * itimer signal ?
638 *
639 * itimers are process shared and we restart periodic
640 * itimers in the signal delivery path to prevent DoS
641 * attacks in the high resolution timer case. This is
642 * compliant with the old way of self-restarting
643 * itimers, as the SIGALRM is a legacy signal and only
644 * queued once. Changing the restart behaviour to
645 * restart the timer in the signal dequeue path is
646 * reducing the timer noise on heavy loaded !highres
647 * systems too.
648 */
649 if (unlikely(signr == SIGALRM)) {
650 struct hrtimer *tmr = &tsk->signal->real_timer;
651
652 if (!hrtimer_is_queued(tmr) &&
653 tsk->signal->it_real_incr != 0) {
654 hrtimer_forward(tmr, tmr->base->get_time(),
655 tsk->signal->it_real_incr);
656 hrtimer_restart(tmr);
657 }
658 }
659#endif
660 }
661
662 recalc_sigpending();
663 if (!signr)
664 return 0;
665
666 if (unlikely(sig_kernel_stop(signr))) {
667 /*
668 * Set a marker that we have dequeued a stop signal. Our
669 * caller might release the siglock and then the pending
670 * stop signal it is about to process is no longer in the
671 * pending bitmasks, but must still be cleared by a SIGCONT
672 * (and overruled by a SIGKILL). So those cases clear this
673 * shared flag after we've set it. Note that this flag may
674 * remain set after the signal we return is ignored or
675 * handled. That doesn't matter because its only purpose
676 * is to alert stop-signal processing code when another
677 * processor has come along and cleared the flag.
678 */
679 current->jobctl |= JOBCTL_STOP_DEQUEUED;
680 }
681#ifdef CONFIG_POSIX_TIMERS
682 if (resched_timer) {
683 /*
684 * Release the siglock to ensure proper locking order
685 * of timer locks outside of siglocks. Note, we leave
686 * irqs disabled here, since the posix-timers code is
687 * about to disable them again anyway.
688 */
689 spin_unlock(&tsk->sighand->siglock);
690 posixtimer_rearm(info);
691 spin_lock(&tsk->sighand->siglock);
692
693 /* Don't expose the si_sys_private value to userspace */
694 info->si_sys_private = 0;
695 }
696#endif
697 return signr;
698}
699EXPORT_SYMBOL_GPL(dequeue_signal);
700
701static int dequeue_synchronous_signal(kernel_siginfo_t *info)
702{
703 struct task_struct *tsk = current;
704 struct sigpending *pending = &tsk->pending;
705 struct sigqueue *q, *sync = NULL;
706
707 /*
708 * Might a synchronous signal be in the queue?
709 */
710 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
711 return 0;
712
713 /*
714 * Return the first synchronous signal in the queue.
715 */
716 list_for_each_entry(q, &pending->list, list) {
717 /* Synchronous signals have a postive si_code */
718 if ((q->info.si_code > SI_USER) &&
719 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
720 sync = q;
721 goto next;
722 }
723 }
724 return 0;
725next:
726 /*
727 * Check if there is another siginfo for the same signal.
728 */
729 list_for_each_entry_continue(q, &pending->list, list) {
730 if (q->info.si_signo == sync->info.si_signo)
731 goto still_pending;
732 }
733
734 sigdelset(&pending->signal, sync->info.si_signo);
735 recalc_sigpending();
736still_pending:
737 list_del_init(&sync->list);
738 copy_siginfo(info, &sync->info);
739 __sigqueue_free(sync);
740 return info->si_signo;
741}
742
743/*
744 * Tell a process that it has a new active signal..
745 *
746 * NOTE! we rely on the previous spin_lock to
747 * lock interrupts for us! We can only be called with
748 * "siglock" held, and the local interrupt must
749 * have been disabled when that got acquired!
750 *
751 * No need to set need_resched since signal event passing
752 * goes through ->blocked
753 */
754void signal_wake_up_state(struct task_struct *t, unsigned int state)
755{
756 set_tsk_thread_flag(t, TIF_SIGPENDING);
757 /*
758 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
759 * case. We don't check t->state here because there is a race with it
760 * executing another processor and just now entering stopped state.
761 * By using wake_up_state, we ensure the process will wake up and
762 * handle its death signal.
763 */
764 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
765 kick_process(t);
766}
767
768/*
769 * Remove signals in mask from the pending set and queue.
770 * Returns 1 if any signals were found.
771 *
772 * All callers must be holding the siglock.
773 */
774static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
775{
776 struct sigqueue *q, *n;
777 sigset_t m;
778
779 sigandsets(&m, mask, &s->signal);
780 if (sigisemptyset(&m))
781 return;
782
783 sigandnsets(&s->signal, &s->signal, mask);
784 list_for_each_entry_safe(q, n, &s->list, list) {
785 if (sigismember(mask, q->info.si_signo)) {
786 list_del_init(&q->list);
787 __sigqueue_free(q);
788 }
789 }
790}
791
792static inline int is_si_special(const struct kernel_siginfo *info)
793{
794 return info <= SEND_SIG_PRIV;
795}
796
797static inline bool si_fromuser(const struct kernel_siginfo *info)
798{
799 return info == SEND_SIG_NOINFO ||
800 (!is_si_special(info) && SI_FROMUSER(info));
801}
802
803/*
804 * called with RCU read lock from check_kill_permission()
805 */
806static bool kill_ok_by_cred(struct task_struct *t)
807{
808 const struct cred *cred = current_cred();
809 const struct cred *tcred = __task_cred(t);
810
811 return uid_eq(cred->euid, tcred->suid) ||
812 uid_eq(cred->euid, tcred->uid) ||
813 uid_eq(cred->uid, tcred->suid) ||
814 uid_eq(cred->uid, tcred->uid) ||
815 ns_capable(tcred->user_ns, CAP_KILL);
816}
817
818/*
819 * Bad permissions for sending the signal
820 * - the caller must hold the RCU read lock
821 */
822static int check_kill_permission(int sig, struct kernel_siginfo *info,
823 struct task_struct *t)
824{
825 struct pid *sid;
826 int error;
827
828 if (!valid_signal(sig))
829 return -EINVAL;
830
831 if (!si_fromuser(info))
832 return 0;
833
834 error = audit_signal_info(sig, t); /* Let audit system see the signal */
835 if (error)
836 return error;
837
838 if (!same_thread_group(current, t) &&
839 !kill_ok_by_cred(t)) {
840 switch (sig) {
841 case SIGCONT:
842 sid = task_session(t);
843 /*
844 * We don't return the error if sid == NULL. The
845 * task was unhashed, the caller must notice this.
846 */
847 if (!sid || sid == task_session(current))
848 break;
849 /* fall through */
850 default:
851 return -EPERM;
852 }
853 }
854
855 return security_task_kill(t, info, sig, NULL);
856}
857
858/**
859 * ptrace_trap_notify - schedule trap to notify ptracer
860 * @t: tracee wanting to notify tracer
861 *
862 * This function schedules sticky ptrace trap which is cleared on the next
863 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
864 * ptracer.
865 *
866 * If @t is running, STOP trap will be taken. If trapped for STOP and
867 * ptracer is listening for events, tracee is woken up so that it can
868 * re-trap for the new event. If trapped otherwise, STOP trap will be
869 * eventually taken without returning to userland after the existing traps
870 * are finished by PTRACE_CONT.
871 *
872 * CONTEXT:
873 * Must be called with @task->sighand->siglock held.
874 */
875static void ptrace_trap_notify(struct task_struct *t)
876{
877 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
878 assert_spin_locked(&t->sighand->siglock);
879
880 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
881 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
882}
883
884/*
885 * Handle magic process-wide effects of stop/continue signals. Unlike
886 * the signal actions, these happen immediately at signal-generation
887 * time regardless of blocking, ignoring, or handling. This does the
888 * actual continuing for SIGCONT, but not the actual stopping for stop
889 * signals. The process stop is done as a signal action for SIG_DFL.
890 *
891 * Returns true if the signal should be actually delivered, otherwise
892 * it should be dropped.
893 */
894static bool prepare_signal(int sig, struct task_struct *p, bool force)
895{
896 struct signal_struct *signal = p->signal;
897 struct task_struct *t;
898 sigset_t flush;
899
900 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
901 if (!(signal->flags & SIGNAL_GROUP_EXIT))
902 return sig == SIGKILL;
903 /*
904 * The process is in the middle of dying, nothing to do.
905 */
906 } else if (sig_kernel_stop(sig)) {
907 /*
908 * This is a stop signal. Remove SIGCONT from all queues.
909 */
910 siginitset(&flush, sigmask(SIGCONT));
911 flush_sigqueue_mask(&flush, &signal->shared_pending);
912 for_each_thread(p, t)
913 flush_sigqueue_mask(&flush, &t->pending);
914 } else if (sig == SIGCONT) {
915 unsigned int why;
916 /*
917 * Remove all stop signals from all queues, wake all threads.
918 */
919 siginitset(&flush, SIG_KERNEL_STOP_MASK);
920 flush_sigqueue_mask(&flush, &signal->shared_pending);
921 for_each_thread(p, t) {
922 flush_sigqueue_mask(&flush, &t->pending);
923 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
924 if (likely(!(t->ptrace & PT_SEIZED)))
925 wake_up_state(t, __TASK_STOPPED);
926 else
927 ptrace_trap_notify(t);
928 }
929
930 /*
931 * Notify the parent with CLD_CONTINUED if we were stopped.
932 *
933 * If we were in the middle of a group stop, we pretend it
934 * was already finished, and then continued. Since SIGCHLD
935 * doesn't queue we report only CLD_STOPPED, as if the next
936 * CLD_CONTINUED was dropped.
937 */
938 why = 0;
939 if (signal->flags & SIGNAL_STOP_STOPPED)
940 why |= SIGNAL_CLD_CONTINUED;
941 else if (signal->group_stop_count)
942 why |= SIGNAL_CLD_STOPPED;
943
944 if (why) {
945 /*
946 * The first thread which returns from do_signal_stop()
947 * will take ->siglock, notice SIGNAL_CLD_MASK, and
948 * notify its parent. See get_signal().
949 */
950 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
951 signal->group_stop_count = 0;
952 signal->group_exit_code = 0;
953 }
954 }
955
956 return !sig_ignored(p, sig, force);
957}
958
959/*
960 * Test if P wants to take SIG. After we've checked all threads with this,
961 * it's equivalent to finding no threads not blocking SIG. Any threads not
962 * blocking SIG were ruled out because they are not running and already
963 * have pending signals. Such threads will dequeue from the shared queue
964 * as soon as they're available, so putting the signal on the shared queue
965 * will be equivalent to sending it to one such thread.
966 */
967static inline bool wants_signal(int sig, struct task_struct *p)
968{
969 if (sigismember(&p->blocked, sig))
970 return false;
971
972 if (p->flags & PF_EXITING)
973 return false;
974
975 if (sig == SIGKILL)
976 return true;
977
978 if (task_is_stopped_or_traced(p))
979 return false;
980
981 return task_curr(p) || !signal_pending(p);
982}
983
984static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
985{
986 struct signal_struct *signal = p->signal;
987 struct task_struct *t;
988
989 /*
990 * Now find a thread we can wake up to take the signal off the queue.
991 *
992 * If the main thread wants the signal, it gets first crack.
993 * Probably the least surprising to the average bear.
994 */
995 if (wants_signal(sig, p))
996 t = p;
997 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
998 /*
999 * There is just one thread and it does not need to be woken.
1000 * It will dequeue unblocked signals before it runs again.
1001 */
1002 return;
1003 else {
1004 /*
1005 * Otherwise try to find a suitable thread.
1006 */
1007 t = signal->curr_target;
1008 while (!wants_signal(sig, t)) {
1009 t = next_thread(t);
1010 if (t == signal->curr_target)
1011 /*
1012 * No thread needs to be woken.
1013 * Any eligible threads will see
1014 * the signal in the queue soon.
1015 */
1016 return;
1017 }
1018 signal->curr_target = t;
1019 }
1020
1021 /*
1022 * Found a killable thread. If the signal will be fatal,
1023 * then start taking the whole group down immediately.
1024 */
1025 if (sig_fatal(p, sig) &&
1026 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1027 !sigismember(&t->real_blocked, sig) &&
1028 (sig == SIGKILL || !p->ptrace)) {
1029 /*
1030 * This signal will be fatal to the whole group.
1031 */
1032 if (!sig_kernel_coredump(sig)) {
1033 /*
1034 * Start a group exit and wake everybody up.
1035 * This way we don't have other threads
1036 * running and doing things after a slower
1037 * thread has the fatal signal pending.
1038 */
1039 signal->flags = SIGNAL_GROUP_EXIT;
1040 signal->group_exit_code = sig;
1041 signal->group_stop_count = 0;
1042 t = p;
1043 do {
1044 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1045 sigaddset(&t->pending.signal, SIGKILL);
1046 signal_wake_up(t, 1);
1047 } while_each_thread(p, t);
1048 return;
1049 }
1050 }
1051
1052 /*
1053 * The signal is already in the shared-pending queue.
1054 * Tell the chosen thread to wake up and dequeue it.
1055 */
1056 signal_wake_up(t, sig == SIGKILL);
1057 return;
1058}
1059
1060static inline bool legacy_queue(struct sigpending *signals, int sig)
1061{
1062 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1063}
1064
1065static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1066 enum pid_type type, bool force)
1067{
1068 struct sigpending *pending;
1069 struct sigqueue *q;
1070 int override_rlimit;
1071 int ret = 0, result;
1072
1073 assert_spin_locked(&t->sighand->siglock);
1074
1075 result = TRACE_SIGNAL_IGNORED;
1076 if (!prepare_signal(sig, t, force))
1077 goto ret;
1078
1079 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1080 /*
1081 * Short-circuit ignored signals and support queuing
1082 * exactly one non-rt signal, so that we can get more
1083 * detailed information about the cause of the signal.
1084 */
1085 result = TRACE_SIGNAL_ALREADY_PENDING;
1086 if (legacy_queue(pending, sig))
1087 goto ret;
1088
1089 result = TRACE_SIGNAL_DELIVERED;
1090 /*
1091 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1092 */
1093 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1094 goto out_set;
1095
1096 /*
1097 * Real-time signals must be queued if sent by sigqueue, or
1098 * some other real-time mechanism. It is implementation
1099 * defined whether kill() does so. We attempt to do so, on
1100 * the principle of least surprise, but since kill is not
1101 * allowed to fail with EAGAIN when low on memory we just
1102 * make sure at least one signal gets delivered and don't
1103 * pass on the info struct.
1104 */
1105 if (sig < SIGRTMIN)
1106 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1107 else
1108 override_rlimit = 0;
1109
1110 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit);
1111 if (q) {
1112 list_add_tail(&q->list, &pending->list);
1113 switch ((unsigned long) info) {
1114 case (unsigned long) SEND_SIG_NOINFO:
1115 clear_siginfo(&q->info);
1116 q->info.si_signo = sig;
1117 q->info.si_errno = 0;
1118 q->info.si_code = SI_USER;
1119 q->info.si_pid = task_tgid_nr_ns(current,
1120 task_active_pid_ns(t));
1121 rcu_read_lock();
1122 q->info.si_uid =
1123 from_kuid_munged(task_cred_xxx(t, user_ns),
1124 current_uid());
1125 rcu_read_unlock();
1126 break;
1127 case (unsigned long) SEND_SIG_PRIV:
1128 clear_siginfo(&q->info);
1129 q->info.si_signo = sig;
1130 q->info.si_errno = 0;
1131 q->info.si_code = SI_KERNEL;
1132 q->info.si_pid = 0;
1133 q->info.si_uid = 0;
1134 break;
1135 default:
1136 copy_siginfo(&q->info, info);
1137 break;
1138 }
1139 } else if (!is_si_special(info) &&
1140 sig >= SIGRTMIN && info->si_code != SI_USER) {
1141 /*
1142 * Queue overflow, abort. We may abort if the
1143 * signal was rt and sent by user using something
1144 * other than kill().
1145 */
1146 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1147 ret = -EAGAIN;
1148 goto ret;
1149 } else {
1150 /*
1151 * This is a silent loss of information. We still
1152 * send the signal, but the *info bits are lost.
1153 */
1154 result = TRACE_SIGNAL_LOSE_INFO;
1155 }
1156
1157out_set:
1158 signalfd_notify(t, sig);
1159 sigaddset(&pending->signal, sig);
1160
1161 /* Let multiprocess signals appear after on-going forks */
1162 if (type > PIDTYPE_TGID) {
1163 struct multiprocess_signals *delayed;
1164 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1165 sigset_t *signal = &delayed->signal;
1166 /* Can't queue both a stop and a continue signal */
1167 if (sig == SIGCONT)
1168 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1169 else if (sig_kernel_stop(sig))
1170 sigdelset(signal, SIGCONT);
1171 sigaddset(signal, sig);
1172 }
1173 }
1174
1175 complete_signal(sig, t, type);
1176ret:
1177 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1178 return ret;
1179}
1180
1181static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1182{
1183 bool ret = false;
1184 switch (siginfo_layout(info->si_signo, info->si_code)) {
1185 case SIL_KILL:
1186 case SIL_CHLD:
1187 case SIL_RT:
1188 ret = true;
1189 break;
1190 case SIL_TIMER:
1191 case SIL_POLL:
1192 case SIL_FAULT:
1193 case SIL_FAULT_MCEERR:
1194 case SIL_FAULT_BNDERR:
1195 case SIL_FAULT_PKUERR:
1196 case SIL_SYS:
1197 ret = false;
1198 break;
1199 }
1200 return ret;
1201}
1202
1203static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1204 enum pid_type type)
1205{
1206 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1207 bool force = false;
1208
1209 if (info == SEND_SIG_NOINFO) {
1210 /* Force if sent from an ancestor pid namespace */
1211 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1212 } else if (info == SEND_SIG_PRIV) {
1213 /* Don't ignore kernel generated signals */
1214 force = true;
1215 } else if (has_si_pid_and_uid(info)) {
1216 /* SIGKILL and SIGSTOP is special or has ids */
1217 struct user_namespace *t_user_ns;
1218
1219 rcu_read_lock();
1220 t_user_ns = task_cred_xxx(t, user_ns);
1221 if (current_user_ns() != t_user_ns) {
1222 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1223 info->si_uid = from_kuid_munged(t_user_ns, uid);
1224 }
1225 rcu_read_unlock();
1226
1227 /* A kernel generated signal? */
1228 force = (info->si_code == SI_KERNEL);
1229
1230 /* From an ancestor pid namespace? */
1231 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1232 info->si_pid = 0;
1233 force = true;
1234 }
1235 }
1236 return __send_signal(sig, info, t, type, force);
1237}
1238
1239static void print_fatal_signal(int signr)
1240{
1241 struct pt_regs *regs = signal_pt_regs();
1242 pr_info("potentially unexpected fatal signal %d.\n", signr);
1243
1244#if defined(__i386__) && !defined(__arch_um__)
1245 pr_info("code at %08lx: ", regs->ip);
1246 {
1247 int i;
1248 for (i = 0; i < 16; i++) {
1249 unsigned char insn;
1250
1251 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1252 break;
1253 pr_cont("%02x ", insn);
1254 }
1255 }
1256 pr_cont("\n");
1257#endif
1258 preempt_disable();
1259 show_regs(regs);
1260 preempt_enable();
1261}
1262
1263static int __init setup_print_fatal_signals(char *str)
1264{
1265 get_option (&str, &print_fatal_signals);
1266
1267 return 1;
1268}
1269
1270__setup("print-fatal-signals=", setup_print_fatal_signals);
1271
1272int
1273__group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1274{
1275 return send_signal(sig, info, p, PIDTYPE_TGID);
1276}
1277
1278int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1279 enum pid_type type)
1280{
1281 unsigned long flags;
1282 int ret = -ESRCH;
1283
1284 if (lock_task_sighand(p, &flags)) {
1285 ret = send_signal(sig, info, p, type);
1286 unlock_task_sighand(p, &flags);
1287 }
1288
1289 return ret;
1290}
1291
1292/*
1293 * Force a signal that the process can't ignore: if necessary
1294 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1295 *
1296 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1297 * since we do not want to have a signal handler that was blocked
1298 * be invoked when user space had explicitly blocked it.
1299 *
1300 * We don't want to have recursive SIGSEGV's etc, for example,
1301 * that is why we also clear SIGNAL_UNKILLABLE.
1302 */
1303static int
1304force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1305{
1306 unsigned long int flags;
1307 int ret, blocked, ignored;
1308 struct k_sigaction *action;
1309 int sig = info->si_signo;
1310
1311 spin_lock_irqsave(&t->sighand->siglock, flags);
1312 action = &t->sighand->action[sig-1];
1313 ignored = action->sa.sa_handler == SIG_IGN;
1314 blocked = sigismember(&t->blocked, sig);
1315 if (blocked || ignored) {
1316 action->sa.sa_handler = SIG_DFL;
1317 if (blocked) {
1318 sigdelset(&t->blocked, sig);
1319 recalc_sigpending_and_wake(t);
1320 }
1321 }
1322 /*
1323 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1324 * debugging to leave init killable.
1325 */
1326 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1327 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1328 ret = send_signal(sig, info, t, PIDTYPE_PID);
1329 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1330
1331 return ret;
1332}
1333
1334int force_sig_info(struct kernel_siginfo *info)
1335{
1336 return force_sig_info_to_task(info, current);
1337}
1338
1339/*
1340 * Nuke all other threads in the group.
1341 */
1342int zap_other_threads(struct task_struct *p)
1343{
1344 struct task_struct *t = p;
1345 int count = 0;
1346
1347 p->signal->group_stop_count = 0;
1348
1349 while_each_thread(p, t) {
1350 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1351 count++;
1352
1353 /* Don't bother with already dead threads */
1354 if (t->exit_state)
1355 continue;
1356 sigaddset(&t->pending.signal, SIGKILL);
1357 signal_wake_up(t, 1);
1358 }
1359
1360 return count;
1361}
1362
1363struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1364 unsigned long *flags)
1365{
1366 struct sighand_struct *sighand;
1367
1368 rcu_read_lock();
1369 for (;;) {
1370 sighand = rcu_dereference(tsk->sighand);
1371 if (unlikely(sighand == NULL))
1372 break;
1373
1374 /*
1375 * This sighand can be already freed and even reused, but
1376 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1377 * initializes ->siglock: this slab can't go away, it has
1378 * the same object type, ->siglock can't be reinitialized.
1379 *
1380 * We need to ensure that tsk->sighand is still the same
1381 * after we take the lock, we can race with de_thread() or
1382 * __exit_signal(). In the latter case the next iteration
1383 * must see ->sighand == NULL.
1384 */
1385 spin_lock_irqsave(&sighand->siglock, *flags);
1386 if (likely(sighand == tsk->sighand))
1387 break;
1388 spin_unlock_irqrestore(&sighand->siglock, *flags);
1389 }
1390 rcu_read_unlock();
1391
1392 return sighand;
1393}
1394
1395/*
1396 * send signal info to all the members of a group
1397 */
1398int group_send_sig_info(int sig, struct kernel_siginfo *info,
1399 struct task_struct *p, enum pid_type type)
1400{
1401 int ret;
1402
1403 rcu_read_lock();
1404 ret = check_kill_permission(sig, info, p);
1405 rcu_read_unlock();
1406
1407 if (!ret && sig)
1408 ret = do_send_sig_info(sig, info, p, type);
1409
1410 return ret;
1411}
1412
1413/*
1414 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1415 * control characters do (^C, ^Z etc)
1416 * - the caller must hold at least a readlock on tasklist_lock
1417 */
1418int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1419{
1420 struct task_struct *p = NULL;
1421 int retval, success;
1422
1423 success = 0;
1424 retval = -ESRCH;
1425 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1426 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1427 success |= !err;
1428 retval = err;
1429 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1430 return success ? 0 : retval;
1431}
1432
1433int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1434{
1435 int error = -ESRCH;
1436 struct task_struct *p;
1437
1438 for (;;) {
1439 rcu_read_lock();
1440 p = pid_task(pid, PIDTYPE_PID);
1441 if (p)
1442 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1443 rcu_read_unlock();
1444 if (likely(!p || error != -ESRCH))
1445 return error;
1446
1447 /*
1448 * The task was unhashed in between, try again. If it
1449 * is dead, pid_task() will return NULL, if we race with
1450 * de_thread() it will find the new leader.
1451 */
1452 }
1453}
1454
1455static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1456{
1457 int error;
1458 rcu_read_lock();
1459 error = kill_pid_info(sig, info, find_vpid(pid));
1460 rcu_read_unlock();
1461 return error;
1462}
1463
1464static inline bool kill_as_cred_perm(const struct cred *cred,
1465 struct task_struct *target)
1466{
1467 const struct cred *pcred = __task_cred(target);
1468
1469 return uid_eq(cred->euid, pcred->suid) ||
1470 uid_eq(cred->euid, pcred->uid) ||
1471 uid_eq(cred->uid, pcred->suid) ||
1472 uid_eq(cred->uid, pcred->uid);
1473}
1474
1475/*
1476 * The usb asyncio usage of siginfo is wrong. The glibc support
1477 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1478 * AKA after the generic fields:
1479 * kernel_pid_t si_pid;
1480 * kernel_uid32_t si_uid;
1481 * sigval_t si_value;
1482 *
1483 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1484 * after the generic fields is:
1485 * void __user *si_addr;
1486 *
1487 * This is a practical problem when there is a 64bit big endian kernel
1488 * and a 32bit userspace. As the 32bit address will encoded in the low
1489 * 32bits of the pointer. Those low 32bits will be stored at higher
1490 * address than appear in a 32 bit pointer. So userspace will not
1491 * see the address it was expecting for it's completions.
1492 *
1493 * There is nothing in the encoding that can allow
1494 * copy_siginfo_to_user32 to detect this confusion of formats, so
1495 * handle this by requiring the caller of kill_pid_usb_asyncio to
1496 * notice when this situration takes place and to store the 32bit
1497 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1498 * parameter.
1499 */
1500int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1501 struct pid *pid, const struct cred *cred)
1502{
1503 struct kernel_siginfo info;
1504 struct task_struct *p;
1505 unsigned long flags;
1506 int ret = -EINVAL;
1507
1508 clear_siginfo(&info);
1509 info.si_signo = sig;
1510 info.si_errno = errno;
1511 info.si_code = SI_ASYNCIO;
1512 *((sigval_t *)&info.si_pid) = addr;
1513
1514 if (!valid_signal(sig))
1515 return ret;
1516
1517 rcu_read_lock();
1518 p = pid_task(pid, PIDTYPE_PID);
1519 if (!p) {
1520 ret = -ESRCH;
1521 goto out_unlock;
1522 }
1523 if (!kill_as_cred_perm(cred, p)) {
1524 ret = -EPERM;
1525 goto out_unlock;
1526 }
1527 ret = security_task_kill(p, &info, sig, cred);
1528 if (ret)
1529 goto out_unlock;
1530
1531 if (sig) {
1532 if (lock_task_sighand(p, &flags)) {
1533 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1534 unlock_task_sighand(p, &flags);
1535 } else
1536 ret = -ESRCH;
1537 }
1538out_unlock:
1539 rcu_read_unlock();
1540 return ret;
1541}
1542EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1543
1544/*
1545 * kill_something_info() interprets pid in interesting ways just like kill(2).
1546 *
1547 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1548 * is probably wrong. Should make it like BSD or SYSV.
1549 */
1550
1551static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1552{
1553 int ret;
1554
1555 if (pid > 0) {
1556 rcu_read_lock();
1557 ret = kill_pid_info(sig, info, find_vpid(pid));
1558 rcu_read_unlock();
1559 return ret;
1560 }
1561
1562 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1563 if (pid == INT_MIN)
1564 return -ESRCH;
1565
1566 read_lock(&tasklist_lock);
1567 if (pid != -1) {
1568 ret = __kill_pgrp_info(sig, info,
1569 pid ? find_vpid(-pid) : task_pgrp(current));
1570 } else {
1571 int retval = 0, count = 0;
1572 struct task_struct * p;
1573
1574 for_each_process(p) {
1575 if (task_pid_vnr(p) > 1 &&
1576 !same_thread_group(p, current)) {
1577 int err = group_send_sig_info(sig, info, p,
1578 PIDTYPE_MAX);
1579 ++count;
1580 if (err != -EPERM)
1581 retval = err;
1582 }
1583 }
1584 ret = count ? retval : -ESRCH;
1585 }
1586 read_unlock(&tasklist_lock);
1587
1588 return ret;
1589}
1590
1591/*
1592 * These are for backward compatibility with the rest of the kernel source.
1593 */
1594
1595int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1596{
1597 /*
1598 * Make sure legacy kernel users don't send in bad values
1599 * (normal paths check this in check_kill_permission).
1600 */
1601 if (!valid_signal(sig))
1602 return -EINVAL;
1603
1604 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1605}
1606EXPORT_SYMBOL(send_sig_info);
1607
1608#define __si_special(priv) \
1609 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1610
1611int
1612send_sig(int sig, struct task_struct *p, int priv)
1613{
1614 return send_sig_info(sig, __si_special(priv), p);
1615}
1616EXPORT_SYMBOL(send_sig);
1617
1618void force_sig(int sig)
1619{
1620 struct kernel_siginfo info;
1621
1622 clear_siginfo(&info);
1623 info.si_signo = sig;
1624 info.si_errno = 0;
1625 info.si_code = SI_KERNEL;
1626 info.si_pid = 0;
1627 info.si_uid = 0;
1628 force_sig_info(&info);
1629}
1630EXPORT_SYMBOL(force_sig);
1631
1632/*
1633 * When things go south during signal handling, we
1634 * will force a SIGSEGV. And if the signal that caused
1635 * the problem was already a SIGSEGV, we'll want to
1636 * make sure we don't even try to deliver the signal..
1637 */
1638void force_sigsegv(int sig)
1639{
1640 struct task_struct *p = current;
1641
1642 if (sig == SIGSEGV) {
1643 unsigned long flags;
1644 spin_lock_irqsave(&p->sighand->siglock, flags);
1645 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1646 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1647 }
1648 force_sig(SIGSEGV);
1649}
1650
1651int force_sig_fault_to_task(int sig, int code, void __user *addr
1652 ___ARCH_SI_TRAPNO(int trapno)
1653 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1654 , struct task_struct *t)
1655{
1656 struct kernel_siginfo info;
1657
1658 clear_siginfo(&info);
1659 info.si_signo = sig;
1660 info.si_errno = 0;
1661 info.si_code = code;
1662 info.si_addr = addr;
1663#ifdef __ARCH_SI_TRAPNO
1664 info.si_trapno = trapno;
1665#endif
1666#ifdef __ia64__
1667 info.si_imm = imm;
1668 info.si_flags = flags;
1669 info.si_isr = isr;
1670#endif
1671 return force_sig_info_to_task(&info, t);
1672}
1673
1674int force_sig_fault(int sig, int code, void __user *addr
1675 ___ARCH_SI_TRAPNO(int trapno)
1676 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1677{
1678 return force_sig_fault_to_task(sig, code, addr
1679 ___ARCH_SI_TRAPNO(trapno)
1680 ___ARCH_SI_IA64(imm, flags, isr), current);
1681}
1682
1683int send_sig_fault(int sig, int code, void __user *addr
1684 ___ARCH_SI_TRAPNO(int trapno)
1685 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1686 , struct task_struct *t)
1687{
1688 struct kernel_siginfo info;
1689
1690 clear_siginfo(&info);
1691 info.si_signo = sig;
1692 info.si_errno = 0;
1693 info.si_code = code;
1694 info.si_addr = addr;
1695#ifdef __ARCH_SI_TRAPNO
1696 info.si_trapno = trapno;
1697#endif
1698#ifdef __ia64__
1699 info.si_imm = imm;
1700 info.si_flags = flags;
1701 info.si_isr = isr;
1702#endif
1703 return send_sig_info(info.si_signo, &info, t);
1704}
1705
1706int force_sig_mceerr(int code, void __user *addr, short lsb)
1707{
1708 struct kernel_siginfo info;
1709
1710 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1711 clear_siginfo(&info);
1712 info.si_signo = SIGBUS;
1713 info.si_errno = 0;
1714 info.si_code = code;
1715 info.si_addr = addr;
1716 info.si_addr_lsb = lsb;
1717 return force_sig_info(&info);
1718}
1719
1720int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1721{
1722 struct kernel_siginfo info;
1723
1724 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1725 clear_siginfo(&info);
1726 info.si_signo = SIGBUS;
1727 info.si_errno = 0;
1728 info.si_code = code;
1729 info.si_addr = addr;
1730 info.si_addr_lsb = lsb;
1731 return send_sig_info(info.si_signo, &info, t);
1732}
1733EXPORT_SYMBOL(send_sig_mceerr);
1734
1735int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1736{
1737 struct kernel_siginfo info;
1738
1739 clear_siginfo(&info);
1740 info.si_signo = SIGSEGV;
1741 info.si_errno = 0;
1742 info.si_code = SEGV_BNDERR;
1743 info.si_addr = addr;
1744 info.si_lower = lower;
1745 info.si_upper = upper;
1746 return force_sig_info(&info);
1747}
1748
1749#ifdef SEGV_PKUERR
1750int force_sig_pkuerr(void __user *addr, u32 pkey)
1751{
1752 struct kernel_siginfo info;
1753
1754 clear_siginfo(&info);
1755 info.si_signo = SIGSEGV;
1756 info.si_errno = 0;
1757 info.si_code = SEGV_PKUERR;
1758 info.si_addr = addr;
1759 info.si_pkey = pkey;
1760 return force_sig_info(&info);
1761}
1762#endif
1763
1764/* For the crazy architectures that include trap information in
1765 * the errno field, instead of an actual errno value.
1766 */
1767int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1768{
1769 struct kernel_siginfo info;
1770
1771 clear_siginfo(&info);
1772 info.si_signo = SIGTRAP;
1773 info.si_errno = errno;
1774 info.si_code = TRAP_HWBKPT;
1775 info.si_addr = addr;
1776 return force_sig_info(&info);
1777}
1778
1779int kill_pgrp(struct pid *pid, int sig, int priv)
1780{
1781 int ret;
1782
1783 read_lock(&tasklist_lock);
1784 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1785 read_unlock(&tasklist_lock);
1786
1787 return ret;
1788}
1789EXPORT_SYMBOL(kill_pgrp);
1790
1791int kill_pid(struct pid *pid, int sig, int priv)
1792{
1793 return kill_pid_info(sig, __si_special(priv), pid);
1794}
1795EXPORT_SYMBOL(kill_pid);
1796
1797/*
1798 * These functions support sending signals using preallocated sigqueue
1799 * structures. This is needed "because realtime applications cannot
1800 * afford to lose notifications of asynchronous events, like timer
1801 * expirations or I/O completions". In the case of POSIX Timers
1802 * we allocate the sigqueue structure from the timer_create. If this
1803 * allocation fails we are able to report the failure to the application
1804 * with an EAGAIN error.
1805 */
1806struct sigqueue *sigqueue_alloc(void)
1807{
1808 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1809
1810 if (q)
1811 q->flags |= SIGQUEUE_PREALLOC;
1812
1813 return q;
1814}
1815
1816void sigqueue_free(struct sigqueue *q)
1817{
1818 unsigned long flags;
1819 spinlock_t *lock = ¤t->sighand->siglock;
1820
1821 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1822 /*
1823 * We must hold ->siglock while testing q->list
1824 * to serialize with collect_signal() or with
1825 * __exit_signal()->flush_sigqueue().
1826 */
1827 spin_lock_irqsave(lock, flags);
1828 q->flags &= ~SIGQUEUE_PREALLOC;
1829 /*
1830 * If it is queued it will be freed when dequeued,
1831 * like the "regular" sigqueue.
1832 */
1833 if (!list_empty(&q->list))
1834 q = NULL;
1835 spin_unlock_irqrestore(lock, flags);
1836
1837 if (q)
1838 __sigqueue_free(q);
1839}
1840
1841int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1842{
1843 int sig = q->info.si_signo;
1844 struct sigpending *pending;
1845 struct task_struct *t;
1846 unsigned long flags;
1847 int ret, result;
1848
1849 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1850
1851 ret = -1;
1852 rcu_read_lock();
1853 t = pid_task(pid, type);
1854 if (!t || !likely(lock_task_sighand(t, &flags)))
1855 goto ret;
1856
1857 ret = 1; /* the signal is ignored */
1858 result = TRACE_SIGNAL_IGNORED;
1859 if (!prepare_signal(sig, t, false))
1860 goto out;
1861
1862 ret = 0;
1863 if (unlikely(!list_empty(&q->list))) {
1864 /*
1865 * If an SI_TIMER entry is already queue just increment
1866 * the overrun count.
1867 */
1868 BUG_ON(q->info.si_code != SI_TIMER);
1869 q->info.si_overrun++;
1870 result = TRACE_SIGNAL_ALREADY_PENDING;
1871 goto out;
1872 }
1873 q->info.si_overrun = 0;
1874
1875 signalfd_notify(t, sig);
1876 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1877 list_add_tail(&q->list, &pending->list);
1878 sigaddset(&pending->signal, sig);
1879 complete_signal(sig, t, type);
1880 result = TRACE_SIGNAL_DELIVERED;
1881out:
1882 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1883 unlock_task_sighand(t, &flags);
1884ret:
1885 rcu_read_unlock();
1886 return ret;
1887}
1888
1889static void do_notify_pidfd(struct task_struct *task)
1890{
1891 struct pid *pid;
1892
1893 WARN_ON(task->exit_state == 0);
1894 pid = task_pid(task);
1895 wake_up_all(&pid->wait_pidfd);
1896}
1897
1898/*
1899 * Let a parent know about the death of a child.
1900 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1901 *
1902 * Returns true if our parent ignored us and so we've switched to
1903 * self-reaping.
1904 */
1905bool do_notify_parent(struct task_struct *tsk, int sig)
1906{
1907 struct kernel_siginfo info;
1908 unsigned long flags;
1909 struct sighand_struct *psig;
1910 bool autoreap = false;
1911 u64 utime, stime;
1912
1913 BUG_ON(sig == -1);
1914
1915 /* do_notify_parent_cldstop should have been called instead. */
1916 BUG_ON(task_is_stopped_or_traced(tsk));
1917
1918 BUG_ON(!tsk->ptrace &&
1919 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1920
1921 /* Wake up all pidfd waiters */
1922 do_notify_pidfd(tsk);
1923
1924 if (sig != SIGCHLD) {
1925 /*
1926 * This is only possible if parent == real_parent.
1927 * Check if it has changed security domain.
1928 */
1929 if (tsk->parent_exec_id != tsk->parent->self_exec_id)
1930 sig = SIGCHLD;
1931 }
1932
1933 clear_siginfo(&info);
1934 info.si_signo = sig;
1935 info.si_errno = 0;
1936 /*
1937 * We are under tasklist_lock here so our parent is tied to
1938 * us and cannot change.
1939 *
1940 * task_active_pid_ns will always return the same pid namespace
1941 * until a task passes through release_task.
1942 *
1943 * write_lock() currently calls preempt_disable() which is the
1944 * same as rcu_read_lock(), but according to Oleg, this is not
1945 * correct to rely on this
1946 */
1947 rcu_read_lock();
1948 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1949 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1950 task_uid(tsk));
1951 rcu_read_unlock();
1952
1953 task_cputime(tsk, &utime, &stime);
1954 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1955 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1956
1957 info.si_status = tsk->exit_code & 0x7f;
1958 if (tsk->exit_code & 0x80)
1959 info.si_code = CLD_DUMPED;
1960 else if (tsk->exit_code & 0x7f)
1961 info.si_code = CLD_KILLED;
1962 else {
1963 info.si_code = CLD_EXITED;
1964 info.si_status = tsk->exit_code >> 8;
1965 }
1966
1967 psig = tsk->parent->sighand;
1968 spin_lock_irqsave(&psig->siglock, flags);
1969 if (!tsk->ptrace && sig == SIGCHLD &&
1970 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1971 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1972 /*
1973 * We are exiting and our parent doesn't care. POSIX.1
1974 * defines special semantics for setting SIGCHLD to SIG_IGN
1975 * or setting the SA_NOCLDWAIT flag: we should be reaped
1976 * automatically and not left for our parent's wait4 call.
1977 * Rather than having the parent do it as a magic kind of
1978 * signal handler, we just set this to tell do_exit that we
1979 * can be cleaned up without becoming a zombie. Note that
1980 * we still call __wake_up_parent in this case, because a
1981 * blocked sys_wait4 might now return -ECHILD.
1982 *
1983 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1984 * is implementation-defined: we do (if you don't want
1985 * it, just use SIG_IGN instead).
1986 */
1987 autoreap = true;
1988 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1989 sig = 0;
1990 }
1991 if (valid_signal(sig) && sig)
1992 __group_send_sig_info(sig, &info, tsk->parent);
1993 __wake_up_parent(tsk, tsk->parent);
1994 spin_unlock_irqrestore(&psig->siglock, flags);
1995
1996 return autoreap;
1997}
1998
1999/**
2000 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2001 * @tsk: task reporting the state change
2002 * @for_ptracer: the notification is for ptracer
2003 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2004 *
2005 * Notify @tsk's parent that the stopped/continued state has changed. If
2006 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2007 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2008 *
2009 * CONTEXT:
2010 * Must be called with tasklist_lock at least read locked.
2011 */
2012static void do_notify_parent_cldstop(struct task_struct *tsk,
2013 bool for_ptracer, int why)
2014{
2015 struct kernel_siginfo info;
2016 unsigned long flags;
2017 struct task_struct *parent;
2018 struct sighand_struct *sighand;
2019 u64 utime, stime;
2020
2021 if (for_ptracer) {
2022 parent = tsk->parent;
2023 } else {
2024 tsk = tsk->group_leader;
2025 parent = tsk->real_parent;
2026 }
2027
2028 clear_siginfo(&info);
2029 info.si_signo = SIGCHLD;
2030 info.si_errno = 0;
2031 /*
2032 * see comment in do_notify_parent() about the following 4 lines
2033 */
2034 rcu_read_lock();
2035 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2036 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2037 rcu_read_unlock();
2038
2039 task_cputime(tsk, &utime, &stime);
2040 info.si_utime = nsec_to_clock_t(utime);
2041 info.si_stime = nsec_to_clock_t(stime);
2042
2043 info.si_code = why;
2044 switch (why) {
2045 case CLD_CONTINUED:
2046 info.si_status = SIGCONT;
2047 break;
2048 case CLD_STOPPED:
2049 info.si_status = tsk->signal->group_exit_code & 0x7f;
2050 break;
2051 case CLD_TRAPPED:
2052 info.si_status = tsk->exit_code & 0x7f;
2053 break;
2054 default:
2055 BUG();
2056 }
2057
2058 sighand = parent->sighand;
2059 spin_lock_irqsave(&sighand->siglock, flags);
2060 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2061 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2062 __group_send_sig_info(SIGCHLD, &info, parent);
2063 /*
2064 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2065 */
2066 __wake_up_parent(tsk, parent);
2067 spin_unlock_irqrestore(&sighand->siglock, flags);
2068}
2069
2070static inline bool may_ptrace_stop(void)
2071{
2072 if (!likely(current->ptrace))
2073 return false;
2074 /*
2075 * Are we in the middle of do_coredump?
2076 * If so and our tracer is also part of the coredump stopping
2077 * is a deadlock situation, and pointless because our tracer
2078 * is dead so don't allow us to stop.
2079 * If SIGKILL was already sent before the caller unlocked
2080 * ->siglock we must see ->core_state != NULL. Otherwise it
2081 * is safe to enter schedule().
2082 *
2083 * This is almost outdated, a task with the pending SIGKILL can't
2084 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2085 * after SIGKILL was already dequeued.
2086 */
2087 if (unlikely(current->mm->core_state) &&
2088 unlikely(current->mm == current->parent->mm))
2089 return false;
2090
2091 return true;
2092}
2093
2094/*
2095 * Return non-zero if there is a SIGKILL that should be waking us up.
2096 * Called with the siglock held.
2097 */
2098static bool sigkill_pending(struct task_struct *tsk)
2099{
2100 return sigismember(&tsk->pending.signal, SIGKILL) ||
2101 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2102}
2103
2104/*
2105 * This must be called with current->sighand->siglock held.
2106 *
2107 * This should be the path for all ptrace stops.
2108 * We always set current->last_siginfo while stopped here.
2109 * That makes it a way to test a stopped process for
2110 * being ptrace-stopped vs being job-control-stopped.
2111 *
2112 * If we actually decide not to stop at all because the tracer
2113 * is gone, we keep current->exit_code unless clear_code.
2114 */
2115static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2116 __releases(¤t->sighand->siglock)
2117 __acquires(¤t->sighand->siglock)
2118{
2119 bool gstop_done = false;
2120
2121 if (arch_ptrace_stop_needed(exit_code, info)) {
2122 /*
2123 * The arch code has something special to do before a
2124 * ptrace stop. This is allowed to block, e.g. for faults
2125 * on user stack pages. We can't keep the siglock while
2126 * calling arch_ptrace_stop, so we must release it now.
2127 * To preserve proper semantics, we must do this before
2128 * any signal bookkeeping like checking group_stop_count.
2129 * Meanwhile, a SIGKILL could come in before we retake the
2130 * siglock. That must prevent us from sleeping in TASK_TRACED.
2131 * So after regaining the lock, we must check for SIGKILL.
2132 */
2133 spin_unlock_irq(¤t->sighand->siglock);
2134 arch_ptrace_stop(exit_code, info);
2135 spin_lock_irq(¤t->sighand->siglock);
2136 if (sigkill_pending(current))
2137 return;
2138 }
2139
2140 set_special_state(TASK_TRACED);
2141
2142 /*
2143 * We're committing to trapping. TRACED should be visible before
2144 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2145 * Also, transition to TRACED and updates to ->jobctl should be
2146 * atomic with respect to siglock and should be done after the arch
2147 * hook as siglock is released and regrabbed across it.
2148 *
2149 * TRACER TRACEE
2150 *
2151 * ptrace_attach()
2152 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2153 * do_wait()
2154 * set_current_state() smp_wmb();
2155 * ptrace_do_wait()
2156 * wait_task_stopped()
2157 * task_stopped_code()
2158 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2159 */
2160 smp_wmb();
2161
2162 current->last_siginfo = info;
2163 current->exit_code = exit_code;
2164
2165 /*
2166 * If @why is CLD_STOPPED, we're trapping to participate in a group
2167 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2168 * across siglock relocks since INTERRUPT was scheduled, PENDING
2169 * could be clear now. We act as if SIGCONT is received after
2170 * TASK_TRACED is entered - ignore it.
2171 */
2172 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2173 gstop_done = task_participate_group_stop(current);
2174
2175 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2176 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2177 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2178 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2179
2180 /* entering a trap, clear TRAPPING */
2181 task_clear_jobctl_trapping(current);
2182
2183 spin_unlock_irq(¤t->sighand->siglock);
2184 read_lock(&tasklist_lock);
2185 if (may_ptrace_stop()) {
2186 /*
2187 * Notify parents of the stop.
2188 *
2189 * While ptraced, there are two parents - the ptracer and
2190 * the real_parent of the group_leader. The ptracer should
2191 * know about every stop while the real parent is only
2192 * interested in the completion of group stop. The states
2193 * for the two don't interact with each other. Notify
2194 * separately unless they're gonna be duplicates.
2195 */
2196 do_notify_parent_cldstop(current, true, why);
2197 if (gstop_done && ptrace_reparented(current))
2198 do_notify_parent_cldstop(current, false, why);
2199
2200 /*
2201 * Don't want to allow preemption here, because
2202 * sys_ptrace() needs this task to be inactive.
2203 *
2204 * XXX: implement read_unlock_no_resched().
2205 */
2206 preempt_disable();
2207 read_unlock(&tasklist_lock);
2208 cgroup_enter_frozen();
2209 preempt_enable_no_resched();
2210 freezable_schedule();
2211 cgroup_leave_frozen(true);
2212 } else {
2213 /*
2214 * By the time we got the lock, our tracer went away.
2215 * Don't drop the lock yet, another tracer may come.
2216 *
2217 * If @gstop_done, the ptracer went away between group stop
2218 * completion and here. During detach, it would have set
2219 * JOBCTL_STOP_PENDING on us and we'll re-enter
2220 * TASK_STOPPED in do_signal_stop() on return, so notifying
2221 * the real parent of the group stop completion is enough.
2222 */
2223 if (gstop_done)
2224 do_notify_parent_cldstop(current, false, why);
2225
2226 /* tasklist protects us from ptrace_freeze_traced() */
2227 __set_current_state(TASK_RUNNING);
2228 if (clear_code)
2229 current->exit_code = 0;
2230 read_unlock(&tasklist_lock);
2231 }
2232
2233 /*
2234 * We are back. Now reacquire the siglock before touching
2235 * last_siginfo, so that we are sure to have synchronized with
2236 * any signal-sending on another CPU that wants to examine it.
2237 */
2238 spin_lock_irq(¤t->sighand->siglock);
2239 current->last_siginfo = NULL;
2240
2241 /* LISTENING can be set only during STOP traps, clear it */
2242 current->jobctl &= ~JOBCTL_LISTENING;
2243
2244 /*
2245 * Queued signals ignored us while we were stopped for tracing.
2246 * So check for any that we should take before resuming user mode.
2247 * This sets TIF_SIGPENDING, but never clears it.
2248 */
2249 recalc_sigpending_tsk(current);
2250}
2251
2252static void ptrace_do_notify(int signr, int exit_code, int why)
2253{
2254 kernel_siginfo_t info;
2255
2256 clear_siginfo(&info);
2257 info.si_signo = signr;
2258 info.si_code = exit_code;
2259 info.si_pid = task_pid_vnr(current);
2260 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2261
2262 /* Let the debugger run. */
2263 ptrace_stop(exit_code, why, 1, &info);
2264}
2265
2266void ptrace_notify(int exit_code)
2267{
2268 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2269 if (unlikely(current->task_works))
2270 task_work_run();
2271
2272 spin_lock_irq(¤t->sighand->siglock);
2273 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2274 spin_unlock_irq(¤t->sighand->siglock);
2275}
2276
2277/**
2278 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2279 * @signr: signr causing group stop if initiating
2280 *
2281 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2282 * and participate in it. If already set, participate in the existing
2283 * group stop. If participated in a group stop (and thus slept), %true is
2284 * returned with siglock released.
2285 *
2286 * If ptraced, this function doesn't handle stop itself. Instead,
2287 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2288 * untouched. The caller must ensure that INTERRUPT trap handling takes
2289 * places afterwards.
2290 *
2291 * CONTEXT:
2292 * Must be called with @current->sighand->siglock held, which is released
2293 * on %true return.
2294 *
2295 * RETURNS:
2296 * %false if group stop is already cancelled or ptrace trap is scheduled.
2297 * %true if participated in group stop.
2298 */
2299static bool do_signal_stop(int signr)
2300 __releases(¤t->sighand->siglock)
2301{
2302 struct signal_struct *sig = current->signal;
2303
2304 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2305 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2306 struct task_struct *t;
2307
2308 /* signr will be recorded in task->jobctl for retries */
2309 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2310
2311 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2312 unlikely(signal_group_exit(sig)))
2313 return false;
2314 /*
2315 * There is no group stop already in progress. We must
2316 * initiate one now.
2317 *
2318 * While ptraced, a task may be resumed while group stop is
2319 * still in effect and then receive a stop signal and
2320 * initiate another group stop. This deviates from the
2321 * usual behavior as two consecutive stop signals can't
2322 * cause two group stops when !ptraced. That is why we
2323 * also check !task_is_stopped(t) below.
2324 *
2325 * The condition can be distinguished by testing whether
2326 * SIGNAL_STOP_STOPPED is already set. Don't generate
2327 * group_exit_code in such case.
2328 *
2329 * This is not necessary for SIGNAL_STOP_CONTINUED because
2330 * an intervening stop signal is required to cause two
2331 * continued events regardless of ptrace.
2332 */
2333 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2334 sig->group_exit_code = signr;
2335
2336 sig->group_stop_count = 0;
2337
2338 if (task_set_jobctl_pending(current, signr | gstop))
2339 sig->group_stop_count++;
2340
2341 t = current;
2342 while_each_thread(current, t) {
2343 /*
2344 * Setting state to TASK_STOPPED for a group
2345 * stop is always done with the siglock held,
2346 * so this check has no races.
2347 */
2348 if (!task_is_stopped(t) &&
2349 task_set_jobctl_pending(t, signr | gstop)) {
2350 sig->group_stop_count++;
2351 if (likely(!(t->ptrace & PT_SEIZED)))
2352 signal_wake_up(t, 0);
2353 else
2354 ptrace_trap_notify(t);
2355 }
2356 }
2357 }
2358
2359 if (likely(!current->ptrace)) {
2360 int notify = 0;
2361
2362 /*
2363 * If there are no other threads in the group, or if there
2364 * is a group stop in progress and we are the last to stop,
2365 * report to the parent.
2366 */
2367 if (task_participate_group_stop(current))
2368 notify = CLD_STOPPED;
2369
2370 set_special_state(TASK_STOPPED);
2371 spin_unlock_irq(¤t->sighand->siglock);
2372
2373 /*
2374 * Notify the parent of the group stop completion. Because
2375 * we're not holding either the siglock or tasklist_lock
2376 * here, ptracer may attach inbetween; however, this is for
2377 * group stop and should always be delivered to the real
2378 * parent of the group leader. The new ptracer will get
2379 * its notification when this task transitions into
2380 * TASK_TRACED.
2381 */
2382 if (notify) {
2383 read_lock(&tasklist_lock);
2384 do_notify_parent_cldstop(current, false, notify);
2385 read_unlock(&tasklist_lock);
2386 }
2387
2388 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2389 cgroup_enter_frozen();
2390 freezable_schedule();
2391 return true;
2392 } else {
2393 /*
2394 * While ptraced, group stop is handled by STOP trap.
2395 * Schedule it and let the caller deal with it.
2396 */
2397 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2398 return false;
2399 }
2400}
2401
2402/**
2403 * do_jobctl_trap - take care of ptrace jobctl traps
2404 *
2405 * When PT_SEIZED, it's used for both group stop and explicit
2406 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2407 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2408 * the stop signal; otherwise, %SIGTRAP.
2409 *
2410 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2411 * number as exit_code and no siginfo.
2412 *
2413 * CONTEXT:
2414 * Must be called with @current->sighand->siglock held, which may be
2415 * released and re-acquired before returning with intervening sleep.
2416 */
2417static void do_jobctl_trap(void)
2418{
2419 struct signal_struct *signal = current->signal;
2420 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2421
2422 if (current->ptrace & PT_SEIZED) {
2423 if (!signal->group_stop_count &&
2424 !(signal->flags & SIGNAL_STOP_STOPPED))
2425 signr = SIGTRAP;
2426 WARN_ON_ONCE(!signr);
2427 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2428 CLD_STOPPED);
2429 } else {
2430 WARN_ON_ONCE(!signr);
2431 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2432 current->exit_code = 0;
2433 }
2434}
2435
2436/**
2437 * do_freezer_trap - handle the freezer jobctl trap
2438 *
2439 * Puts the task into frozen state, if only the task is not about to quit.
2440 * In this case it drops JOBCTL_TRAP_FREEZE.
2441 *
2442 * CONTEXT:
2443 * Must be called with @current->sighand->siglock held,
2444 * which is always released before returning.
2445 */
2446static void do_freezer_trap(void)
2447 __releases(¤t->sighand->siglock)
2448{
2449 /*
2450 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2451 * let's make another loop to give it a chance to be handled.
2452 * In any case, we'll return back.
2453 */
2454 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2455 JOBCTL_TRAP_FREEZE) {
2456 spin_unlock_irq(¤t->sighand->siglock);
2457 return;
2458 }
2459
2460 /*
2461 * Now we're sure that there is no pending fatal signal and no
2462 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2463 * immediately (if there is a non-fatal signal pending), and
2464 * put the task into sleep.
2465 */
2466 __set_current_state(TASK_INTERRUPTIBLE);
2467 clear_thread_flag(TIF_SIGPENDING);
2468 spin_unlock_irq(¤t->sighand->siglock);
2469 cgroup_enter_frozen();
2470 freezable_schedule();
2471}
2472
2473static int ptrace_signal(int signr, kernel_siginfo_t *info)
2474{
2475 /*
2476 * We do not check sig_kernel_stop(signr) but set this marker
2477 * unconditionally because we do not know whether debugger will
2478 * change signr. This flag has no meaning unless we are going
2479 * to stop after return from ptrace_stop(). In this case it will
2480 * be checked in do_signal_stop(), we should only stop if it was
2481 * not cleared by SIGCONT while we were sleeping. See also the
2482 * comment in dequeue_signal().
2483 */
2484 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2485 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2486
2487 /* We're back. Did the debugger cancel the sig? */
2488 signr = current->exit_code;
2489 if (signr == 0)
2490 return signr;
2491
2492 current->exit_code = 0;
2493
2494 /*
2495 * Update the siginfo structure if the signal has
2496 * changed. If the debugger wanted something
2497 * specific in the siginfo structure then it should
2498 * have updated *info via PTRACE_SETSIGINFO.
2499 */
2500 if (signr != info->si_signo) {
2501 clear_siginfo(info);
2502 info->si_signo = signr;
2503 info->si_errno = 0;
2504 info->si_code = SI_USER;
2505 rcu_read_lock();
2506 info->si_pid = task_pid_vnr(current->parent);
2507 info->si_uid = from_kuid_munged(current_user_ns(),
2508 task_uid(current->parent));
2509 rcu_read_unlock();
2510 }
2511
2512 /* If the (new) signal is now blocked, requeue it. */
2513 if (sigismember(¤t->blocked, signr)) {
2514 send_signal(signr, info, current, PIDTYPE_PID);
2515 signr = 0;
2516 }
2517
2518 return signr;
2519}
2520
2521bool get_signal(struct ksignal *ksig)
2522{
2523 struct sighand_struct *sighand = current->sighand;
2524 struct signal_struct *signal = current->signal;
2525 int signr;
2526
2527 if (unlikely(current->task_works))
2528 task_work_run();
2529
2530 if (unlikely(uprobe_deny_signal()))
2531 return false;
2532
2533 /*
2534 * Do this once, we can't return to user-mode if freezing() == T.
2535 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2536 * thus do not need another check after return.
2537 */
2538 try_to_freeze();
2539
2540relock:
2541 spin_lock_irq(&sighand->siglock);
2542 /*
2543 * Every stopped thread goes here after wakeup. Check to see if
2544 * we should notify the parent, prepare_signal(SIGCONT) encodes
2545 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2546 */
2547 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2548 int why;
2549
2550 if (signal->flags & SIGNAL_CLD_CONTINUED)
2551 why = CLD_CONTINUED;
2552 else
2553 why = CLD_STOPPED;
2554
2555 signal->flags &= ~SIGNAL_CLD_MASK;
2556
2557 spin_unlock_irq(&sighand->siglock);
2558
2559 /*
2560 * Notify the parent that we're continuing. This event is
2561 * always per-process and doesn't make whole lot of sense
2562 * for ptracers, who shouldn't consume the state via
2563 * wait(2) either, but, for backward compatibility, notify
2564 * the ptracer of the group leader too unless it's gonna be
2565 * a duplicate.
2566 */
2567 read_lock(&tasklist_lock);
2568 do_notify_parent_cldstop(current, false, why);
2569
2570 if (ptrace_reparented(current->group_leader))
2571 do_notify_parent_cldstop(current->group_leader,
2572 true, why);
2573 read_unlock(&tasklist_lock);
2574
2575 goto relock;
2576 }
2577
2578 /* Has this task already been marked for death? */
2579 if (signal_group_exit(signal)) {
2580 ksig->info.si_signo = signr = SIGKILL;
2581 sigdelset(¤t->pending.signal, SIGKILL);
2582 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2583 &sighand->action[SIGKILL - 1]);
2584 recalc_sigpending();
2585 goto fatal;
2586 }
2587
2588 for (;;) {
2589 struct k_sigaction *ka;
2590
2591 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2592 do_signal_stop(0))
2593 goto relock;
2594
2595 if (unlikely(current->jobctl &
2596 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2597 if (current->jobctl & JOBCTL_TRAP_MASK) {
2598 do_jobctl_trap();
2599 spin_unlock_irq(&sighand->siglock);
2600 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2601 do_freezer_trap();
2602
2603 goto relock;
2604 }
2605
2606 /*
2607 * If the task is leaving the frozen state, let's update
2608 * cgroup counters and reset the frozen bit.
2609 */
2610 if (unlikely(cgroup_task_frozen(current))) {
2611 spin_unlock_irq(&sighand->siglock);
2612 cgroup_leave_frozen(false);
2613 goto relock;
2614 }
2615
2616 /*
2617 * Signals generated by the execution of an instruction
2618 * need to be delivered before any other pending signals
2619 * so that the instruction pointer in the signal stack
2620 * frame points to the faulting instruction.
2621 */
2622 signr = dequeue_synchronous_signal(&ksig->info);
2623 if (!signr)
2624 signr = dequeue_signal(current, ¤t->blocked, &ksig->info);
2625
2626 if (!signr)
2627 break; /* will return 0 */
2628
2629 if (unlikely(current->ptrace) && signr != SIGKILL) {
2630 signr = ptrace_signal(signr, &ksig->info);
2631 if (!signr)
2632 continue;
2633 }
2634
2635 ka = &sighand->action[signr-1];
2636
2637 /* Trace actually delivered signals. */
2638 trace_signal_deliver(signr, &ksig->info, ka);
2639
2640 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2641 continue;
2642 if (ka->sa.sa_handler != SIG_DFL) {
2643 /* Run the handler. */
2644 ksig->ka = *ka;
2645
2646 if (ka->sa.sa_flags & SA_ONESHOT)
2647 ka->sa.sa_handler = SIG_DFL;
2648
2649 break; /* will return non-zero "signr" value */
2650 }
2651
2652 /*
2653 * Now we are doing the default action for this signal.
2654 */
2655 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2656 continue;
2657
2658 /*
2659 * Global init gets no signals it doesn't want.
2660 * Container-init gets no signals it doesn't want from same
2661 * container.
2662 *
2663 * Note that if global/container-init sees a sig_kernel_only()
2664 * signal here, the signal must have been generated internally
2665 * or must have come from an ancestor namespace. In either
2666 * case, the signal cannot be dropped.
2667 */
2668 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2669 !sig_kernel_only(signr))
2670 continue;
2671
2672 if (sig_kernel_stop(signr)) {
2673 /*
2674 * The default action is to stop all threads in
2675 * the thread group. The job control signals
2676 * do nothing in an orphaned pgrp, but SIGSTOP
2677 * always works. Note that siglock needs to be
2678 * dropped during the call to is_orphaned_pgrp()
2679 * because of lock ordering with tasklist_lock.
2680 * This allows an intervening SIGCONT to be posted.
2681 * We need to check for that and bail out if necessary.
2682 */
2683 if (signr != SIGSTOP) {
2684 spin_unlock_irq(&sighand->siglock);
2685
2686 /* signals can be posted during this window */
2687
2688 if (is_current_pgrp_orphaned())
2689 goto relock;
2690
2691 spin_lock_irq(&sighand->siglock);
2692 }
2693
2694 if (likely(do_signal_stop(ksig->info.si_signo))) {
2695 /* It released the siglock. */
2696 goto relock;
2697 }
2698
2699 /*
2700 * We didn't actually stop, due to a race
2701 * with SIGCONT or something like that.
2702 */
2703 continue;
2704 }
2705
2706 fatal:
2707 spin_unlock_irq(&sighand->siglock);
2708 if (unlikely(cgroup_task_frozen(current)))
2709 cgroup_leave_frozen(true);
2710
2711 /*
2712 * Anything else is fatal, maybe with a core dump.
2713 */
2714 current->flags |= PF_SIGNALED;
2715
2716 if (sig_kernel_coredump(signr)) {
2717 if (print_fatal_signals)
2718 print_fatal_signal(ksig->info.si_signo);
2719 proc_coredump_connector(current);
2720 /*
2721 * If it was able to dump core, this kills all
2722 * other threads in the group and synchronizes with
2723 * their demise. If we lost the race with another
2724 * thread getting here, it set group_exit_code
2725 * first and our do_group_exit call below will use
2726 * that value and ignore the one we pass it.
2727 */
2728 do_coredump(&ksig->info);
2729 }
2730
2731 /*
2732 * Death signals, no core dump.
2733 */
2734 do_group_exit(ksig->info.si_signo);
2735 /* NOTREACHED */
2736 }
2737 spin_unlock_irq(&sighand->siglock);
2738
2739 ksig->sig = signr;
2740 return ksig->sig > 0;
2741}
2742
2743/**
2744 * signal_delivered -
2745 * @ksig: kernel signal struct
2746 * @stepping: nonzero if debugger single-step or block-step in use
2747 *
2748 * This function should be called when a signal has successfully been
2749 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2750 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2751 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2752 */
2753static void signal_delivered(struct ksignal *ksig, int stepping)
2754{
2755 sigset_t blocked;
2756
2757 /* A signal was successfully delivered, and the
2758 saved sigmask was stored on the signal frame,
2759 and will be restored by sigreturn. So we can
2760 simply clear the restore sigmask flag. */
2761 clear_restore_sigmask();
2762
2763 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2764 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2765 sigaddset(&blocked, ksig->sig);
2766 set_current_blocked(&blocked);
2767 tracehook_signal_handler(stepping);
2768}
2769
2770void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2771{
2772 if (failed)
2773 force_sigsegv(ksig->sig);
2774 else
2775 signal_delivered(ksig, stepping);
2776}
2777
2778/*
2779 * It could be that complete_signal() picked us to notify about the
2780 * group-wide signal. Other threads should be notified now to take
2781 * the shared signals in @which since we will not.
2782 */
2783static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2784{
2785 sigset_t retarget;
2786 struct task_struct *t;
2787
2788 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2789 if (sigisemptyset(&retarget))
2790 return;
2791
2792 t = tsk;
2793 while_each_thread(tsk, t) {
2794 if (t->flags & PF_EXITING)
2795 continue;
2796
2797 if (!has_pending_signals(&retarget, &t->blocked))
2798 continue;
2799 /* Remove the signals this thread can handle. */
2800 sigandsets(&retarget, &retarget, &t->blocked);
2801
2802 if (!signal_pending(t))
2803 signal_wake_up(t, 0);
2804
2805 if (sigisemptyset(&retarget))
2806 break;
2807 }
2808}
2809
2810void exit_signals(struct task_struct *tsk)
2811{
2812 int group_stop = 0;
2813 sigset_t unblocked;
2814
2815 /*
2816 * @tsk is about to have PF_EXITING set - lock out users which
2817 * expect stable threadgroup.
2818 */
2819 cgroup_threadgroup_change_begin(tsk);
2820
2821 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2822 tsk->flags |= PF_EXITING;
2823 cgroup_threadgroup_change_end(tsk);
2824 return;
2825 }
2826
2827 spin_lock_irq(&tsk->sighand->siglock);
2828 /*
2829 * From now this task is not visible for group-wide signals,
2830 * see wants_signal(), do_signal_stop().
2831 */
2832 tsk->flags |= PF_EXITING;
2833
2834 cgroup_threadgroup_change_end(tsk);
2835
2836 if (!signal_pending(tsk))
2837 goto out;
2838
2839 unblocked = tsk->blocked;
2840 signotset(&unblocked);
2841 retarget_shared_pending(tsk, &unblocked);
2842
2843 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2844 task_participate_group_stop(tsk))
2845 group_stop = CLD_STOPPED;
2846out:
2847 spin_unlock_irq(&tsk->sighand->siglock);
2848
2849 /*
2850 * If group stop has completed, deliver the notification. This
2851 * should always go to the real parent of the group leader.
2852 */
2853 if (unlikely(group_stop)) {
2854 read_lock(&tasklist_lock);
2855 do_notify_parent_cldstop(tsk, false, group_stop);
2856 read_unlock(&tasklist_lock);
2857 }
2858}
2859
2860/*
2861 * System call entry points.
2862 */
2863
2864/**
2865 * sys_restart_syscall - restart a system call
2866 */
2867SYSCALL_DEFINE0(restart_syscall)
2868{
2869 struct restart_block *restart = ¤t->restart_block;
2870 return restart->fn(restart);
2871}
2872
2873long do_no_restart_syscall(struct restart_block *param)
2874{
2875 return -EINTR;
2876}
2877
2878static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2879{
2880 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2881 sigset_t newblocked;
2882 /* A set of now blocked but previously unblocked signals. */
2883 sigandnsets(&newblocked, newset, ¤t->blocked);
2884 retarget_shared_pending(tsk, &newblocked);
2885 }
2886 tsk->blocked = *newset;
2887 recalc_sigpending();
2888}
2889
2890/**
2891 * set_current_blocked - change current->blocked mask
2892 * @newset: new mask
2893 *
2894 * It is wrong to change ->blocked directly, this helper should be used
2895 * to ensure the process can't miss a shared signal we are going to block.
2896 */
2897void set_current_blocked(sigset_t *newset)
2898{
2899 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2900 __set_current_blocked(newset);
2901}
2902
2903void __set_current_blocked(const sigset_t *newset)
2904{
2905 struct task_struct *tsk = current;
2906
2907 /*
2908 * In case the signal mask hasn't changed, there is nothing we need
2909 * to do. The current->blocked shouldn't be modified by other task.
2910 */
2911 if (sigequalsets(&tsk->blocked, newset))
2912 return;
2913
2914 spin_lock_irq(&tsk->sighand->siglock);
2915 __set_task_blocked(tsk, newset);
2916 spin_unlock_irq(&tsk->sighand->siglock);
2917}
2918
2919/*
2920 * This is also useful for kernel threads that want to temporarily
2921 * (or permanently) block certain signals.
2922 *
2923 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2924 * interface happily blocks "unblockable" signals like SIGKILL
2925 * and friends.
2926 */
2927int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2928{
2929 struct task_struct *tsk = current;
2930 sigset_t newset;
2931
2932 /* Lockless, only current can change ->blocked, never from irq */
2933 if (oldset)
2934 *oldset = tsk->blocked;
2935
2936 switch (how) {
2937 case SIG_BLOCK:
2938 sigorsets(&newset, &tsk->blocked, set);
2939 break;
2940 case SIG_UNBLOCK:
2941 sigandnsets(&newset, &tsk->blocked, set);
2942 break;
2943 case SIG_SETMASK:
2944 newset = *set;
2945 break;
2946 default:
2947 return -EINVAL;
2948 }
2949
2950 __set_current_blocked(&newset);
2951 return 0;
2952}
2953EXPORT_SYMBOL(sigprocmask);
2954
2955/*
2956 * The api helps set app-provided sigmasks.
2957 *
2958 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
2959 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
2960 *
2961 * Note that it does set_restore_sigmask() in advance, so it must be always
2962 * paired with restore_saved_sigmask_unless() before return from syscall.
2963 */
2964int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
2965{
2966 sigset_t kmask;
2967
2968 if (!umask)
2969 return 0;
2970 if (sigsetsize != sizeof(sigset_t))
2971 return -EINVAL;
2972 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
2973 return -EFAULT;
2974
2975 set_restore_sigmask();
2976 current->saved_sigmask = current->blocked;
2977 set_current_blocked(&kmask);
2978
2979 return 0;
2980}
2981
2982#ifdef CONFIG_COMPAT
2983int set_compat_user_sigmask(const compat_sigset_t __user *umask,
2984 size_t sigsetsize)
2985{
2986 sigset_t kmask;
2987
2988 if (!umask)
2989 return 0;
2990 if (sigsetsize != sizeof(compat_sigset_t))
2991 return -EINVAL;
2992 if (get_compat_sigset(&kmask, umask))
2993 return -EFAULT;
2994
2995 set_restore_sigmask();
2996 current->saved_sigmask = current->blocked;
2997 set_current_blocked(&kmask);
2998
2999 return 0;
3000}
3001#endif
3002
3003/**
3004 * sys_rt_sigprocmask - change the list of currently blocked signals
3005 * @how: whether to add, remove, or set signals
3006 * @nset: stores pending signals
3007 * @oset: previous value of signal mask if non-null
3008 * @sigsetsize: size of sigset_t type
3009 */
3010SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3011 sigset_t __user *, oset, size_t, sigsetsize)
3012{
3013 sigset_t old_set, new_set;
3014 int error;
3015
3016 /* XXX: Don't preclude handling different sized sigset_t's. */
3017 if (sigsetsize != sizeof(sigset_t))
3018 return -EINVAL;
3019
3020 old_set = current->blocked;
3021
3022 if (nset) {
3023 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3024 return -EFAULT;
3025 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3026
3027 error = sigprocmask(how, &new_set, NULL);
3028 if (error)
3029 return error;
3030 }
3031
3032 if (oset) {
3033 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3034 return -EFAULT;
3035 }
3036
3037 return 0;
3038}
3039
3040#ifdef CONFIG_COMPAT
3041COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3042 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3043{
3044 sigset_t old_set = current->blocked;
3045
3046 /* XXX: Don't preclude handling different sized sigset_t's. */
3047 if (sigsetsize != sizeof(sigset_t))
3048 return -EINVAL;
3049
3050 if (nset) {
3051 sigset_t new_set;
3052 int error;
3053 if (get_compat_sigset(&new_set, nset))
3054 return -EFAULT;
3055 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3056
3057 error = sigprocmask(how, &new_set, NULL);
3058 if (error)
3059 return error;
3060 }
3061 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3062}
3063#endif
3064
3065static void do_sigpending(sigset_t *set)
3066{
3067 spin_lock_irq(¤t->sighand->siglock);
3068 sigorsets(set, ¤t->pending.signal,
3069 ¤t->signal->shared_pending.signal);
3070 spin_unlock_irq(¤t->sighand->siglock);
3071
3072 /* Outside the lock because only this thread touches it. */
3073 sigandsets(set, ¤t->blocked, set);
3074}
3075
3076/**
3077 * sys_rt_sigpending - examine a pending signal that has been raised
3078 * while blocked
3079 * @uset: stores pending signals
3080 * @sigsetsize: size of sigset_t type or larger
3081 */
3082SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3083{
3084 sigset_t set;
3085
3086 if (sigsetsize > sizeof(*uset))
3087 return -EINVAL;
3088
3089 do_sigpending(&set);
3090
3091 if (copy_to_user(uset, &set, sigsetsize))
3092 return -EFAULT;
3093
3094 return 0;
3095}
3096
3097#ifdef CONFIG_COMPAT
3098COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3099 compat_size_t, sigsetsize)
3100{
3101 sigset_t set;
3102
3103 if (sigsetsize > sizeof(*uset))
3104 return -EINVAL;
3105
3106 do_sigpending(&set);
3107
3108 return put_compat_sigset(uset, &set, sigsetsize);
3109}
3110#endif
3111
3112static const struct {
3113 unsigned char limit, layout;
3114} sig_sicodes[] = {
3115 [SIGILL] = { NSIGILL, SIL_FAULT },
3116 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3117 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3118 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3119 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3120#if defined(SIGEMT)
3121 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3122#endif
3123 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3124 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3125 [SIGSYS] = { NSIGSYS, SIL_SYS },
3126};
3127
3128static bool known_siginfo_layout(unsigned sig, int si_code)
3129{
3130 if (si_code == SI_KERNEL)
3131 return true;
3132 else if ((si_code > SI_USER)) {
3133 if (sig_specific_sicodes(sig)) {
3134 if (si_code <= sig_sicodes[sig].limit)
3135 return true;
3136 }
3137 else if (si_code <= NSIGPOLL)
3138 return true;
3139 }
3140 else if (si_code >= SI_DETHREAD)
3141 return true;
3142 else if (si_code == SI_ASYNCNL)
3143 return true;
3144 return false;
3145}
3146
3147enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3148{
3149 enum siginfo_layout layout = SIL_KILL;
3150 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3151 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3152 (si_code <= sig_sicodes[sig].limit)) {
3153 layout = sig_sicodes[sig].layout;
3154 /* Handle the exceptions */
3155 if ((sig == SIGBUS) &&
3156 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3157 layout = SIL_FAULT_MCEERR;
3158 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3159 layout = SIL_FAULT_BNDERR;
3160#ifdef SEGV_PKUERR
3161 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3162 layout = SIL_FAULT_PKUERR;
3163#endif
3164 }
3165 else if (si_code <= NSIGPOLL)
3166 layout = SIL_POLL;
3167 } else {
3168 if (si_code == SI_TIMER)
3169 layout = SIL_TIMER;
3170 else if (si_code == SI_SIGIO)
3171 layout = SIL_POLL;
3172 else if (si_code < 0)
3173 layout = SIL_RT;
3174 }
3175 return layout;
3176}
3177
3178static inline char __user *si_expansion(const siginfo_t __user *info)
3179{
3180 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3181}
3182
3183int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3184{
3185 char __user *expansion = si_expansion(to);
3186 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3187 return -EFAULT;
3188 if (clear_user(expansion, SI_EXPANSION_SIZE))
3189 return -EFAULT;
3190 return 0;
3191}
3192
3193static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3194 const siginfo_t __user *from)
3195{
3196 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3197 char __user *expansion = si_expansion(from);
3198 char buf[SI_EXPANSION_SIZE];
3199 int i;
3200 /*
3201 * An unknown si_code might need more than
3202 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3203 * extra bytes are 0. This guarantees copy_siginfo_to_user
3204 * will return this data to userspace exactly.
3205 */
3206 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3207 return -EFAULT;
3208 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3209 if (buf[i] != 0)
3210 return -E2BIG;
3211 }
3212 }
3213 return 0;
3214}
3215
3216static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3217 const siginfo_t __user *from)
3218{
3219 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3220 return -EFAULT;
3221 to->si_signo = signo;
3222 return post_copy_siginfo_from_user(to, from);
3223}
3224
3225int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3226{
3227 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3228 return -EFAULT;
3229 return post_copy_siginfo_from_user(to, from);
3230}
3231
3232#ifdef CONFIG_COMPAT
3233int copy_siginfo_to_user32(struct compat_siginfo __user *to,
3234 const struct kernel_siginfo *from)
3235#if defined(CONFIG_X86_X32_ABI) || defined(CONFIG_IA32_EMULATION)
3236{
3237 return __copy_siginfo_to_user32(to, from, in_x32_syscall());
3238}
3239int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3240 const struct kernel_siginfo *from, bool x32_ABI)
3241#endif
3242{
3243 struct compat_siginfo new;
3244 memset(&new, 0, sizeof(new));
3245
3246 new.si_signo = from->si_signo;
3247 new.si_errno = from->si_errno;
3248 new.si_code = from->si_code;
3249 switch(siginfo_layout(from->si_signo, from->si_code)) {
3250 case SIL_KILL:
3251 new.si_pid = from->si_pid;
3252 new.si_uid = from->si_uid;
3253 break;
3254 case SIL_TIMER:
3255 new.si_tid = from->si_tid;
3256 new.si_overrun = from->si_overrun;
3257 new.si_int = from->si_int;
3258 break;
3259 case SIL_POLL:
3260 new.si_band = from->si_band;
3261 new.si_fd = from->si_fd;
3262 break;
3263 case SIL_FAULT:
3264 new.si_addr = ptr_to_compat(from->si_addr);
3265#ifdef __ARCH_SI_TRAPNO
3266 new.si_trapno = from->si_trapno;
3267#endif
3268 break;
3269 case SIL_FAULT_MCEERR:
3270 new.si_addr = ptr_to_compat(from->si_addr);
3271#ifdef __ARCH_SI_TRAPNO
3272 new.si_trapno = from->si_trapno;
3273#endif
3274 new.si_addr_lsb = from->si_addr_lsb;
3275 break;
3276 case SIL_FAULT_BNDERR:
3277 new.si_addr = ptr_to_compat(from->si_addr);
3278#ifdef __ARCH_SI_TRAPNO
3279 new.si_trapno = from->si_trapno;
3280#endif
3281 new.si_lower = ptr_to_compat(from->si_lower);
3282 new.si_upper = ptr_to_compat(from->si_upper);
3283 break;
3284 case SIL_FAULT_PKUERR:
3285 new.si_addr = ptr_to_compat(from->si_addr);
3286#ifdef __ARCH_SI_TRAPNO
3287 new.si_trapno = from->si_trapno;
3288#endif
3289 new.si_pkey = from->si_pkey;
3290 break;
3291 case SIL_CHLD:
3292 new.si_pid = from->si_pid;
3293 new.si_uid = from->si_uid;
3294 new.si_status = from->si_status;
3295#ifdef CONFIG_X86_X32_ABI
3296 if (x32_ABI) {
3297 new._sifields._sigchld_x32._utime = from->si_utime;
3298 new._sifields._sigchld_x32._stime = from->si_stime;
3299 } else
3300#endif
3301 {
3302 new.si_utime = from->si_utime;
3303 new.si_stime = from->si_stime;
3304 }
3305 break;
3306 case SIL_RT:
3307 new.si_pid = from->si_pid;
3308 new.si_uid = from->si_uid;
3309 new.si_int = from->si_int;
3310 break;
3311 case SIL_SYS:
3312 new.si_call_addr = ptr_to_compat(from->si_call_addr);
3313 new.si_syscall = from->si_syscall;
3314 new.si_arch = from->si_arch;
3315 break;
3316 }
3317
3318 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3319 return -EFAULT;
3320
3321 return 0;
3322}
3323
3324static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3325 const struct compat_siginfo *from)
3326{
3327 clear_siginfo(to);
3328 to->si_signo = from->si_signo;
3329 to->si_errno = from->si_errno;
3330 to->si_code = from->si_code;
3331 switch(siginfo_layout(from->si_signo, from->si_code)) {
3332 case SIL_KILL:
3333 to->si_pid = from->si_pid;
3334 to->si_uid = from->si_uid;
3335 break;
3336 case SIL_TIMER:
3337 to->si_tid = from->si_tid;
3338 to->si_overrun = from->si_overrun;
3339 to->si_int = from->si_int;
3340 break;
3341 case SIL_POLL:
3342 to->si_band = from->si_band;
3343 to->si_fd = from->si_fd;
3344 break;
3345 case SIL_FAULT:
3346 to->si_addr = compat_ptr(from->si_addr);
3347#ifdef __ARCH_SI_TRAPNO
3348 to->si_trapno = from->si_trapno;
3349#endif
3350 break;
3351 case SIL_FAULT_MCEERR:
3352 to->si_addr = compat_ptr(from->si_addr);
3353#ifdef __ARCH_SI_TRAPNO
3354 to->si_trapno = from->si_trapno;
3355#endif
3356 to->si_addr_lsb = from->si_addr_lsb;
3357 break;
3358 case SIL_FAULT_BNDERR:
3359 to->si_addr = compat_ptr(from->si_addr);
3360#ifdef __ARCH_SI_TRAPNO
3361 to->si_trapno = from->si_trapno;
3362#endif
3363 to->si_lower = compat_ptr(from->si_lower);
3364 to->si_upper = compat_ptr(from->si_upper);
3365 break;
3366 case SIL_FAULT_PKUERR:
3367 to->si_addr = compat_ptr(from->si_addr);
3368#ifdef __ARCH_SI_TRAPNO
3369 to->si_trapno = from->si_trapno;
3370#endif
3371 to->si_pkey = from->si_pkey;
3372 break;
3373 case SIL_CHLD:
3374 to->si_pid = from->si_pid;
3375 to->si_uid = from->si_uid;
3376 to->si_status = from->si_status;
3377#ifdef CONFIG_X86_X32_ABI
3378 if (in_x32_syscall()) {
3379 to->si_utime = from->_sifields._sigchld_x32._utime;
3380 to->si_stime = from->_sifields._sigchld_x32._stime;
3381 } else
3382#endif
3383 {
3384 to->si_utime = from->si_utime;
3385 to->si_stime = from->si_stime;
3386 }
3387 break;
3388 case SIL_RT:
3389 to->si_pid = from->si_pid;
3390 to->si_uid = from->si_uid;
3391 to->si_int = from->si_int;
3392 break;
3393 case SIL_SYS:
3394 to->si_call_addr = compat_ptr(from->si_call_addr);
3395 to->si_syscall = from->si_syscall;
3396 to->si_arch = from->si_arch;
3397 break;
3398 }
3399 return 0;
3400}
3401
3402static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3403 const struct compat_siginfo __user *ufrom)
3404{
3405 struct compat_siginfo from;
3406
3407 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3408 return -EFAULT;
3409
3410 from.si_signo = signo;
3411 return post_copy_siginfo_from_user32(to, &from);
3412}
3413
3414int copy_siginfo_from_user32(struct kernel_siginfo *to,
3415 const struct compat_siginfo __user *ufrom)
3416{
3417 struct compat_siginfo from;
3418
3419 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3420 return -EFAULT;
3421
3422 return post_copy_siginfo_from_user32(to, &from);
3423}
3424#endif /* CONFIG_COMPAT */
3425
3426/**
3427 * do_sigtimedwait - wait for queued signals specified in @which
3428 * @which: queued signals to wait for
3429 * @info: if non-null, the signal's siginfo is returned here
3430 * @ts: upper bound on process time suspension
3431 */
3432static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3433 const struct timespec64 *ts)
3434{
3435 ktime_t *to = NULL, timeout = KTIME_MAX;
3436 struct task_struct *tsk = current;
3437 sigset_t mask = *which;
3438 int sig, ret = 0;
3439
3440 if (ts) {
3441 if (!timespec64_valid(ts))
3442 return -EINVAL;
3443 timeout = timespec64_to_ktime(*ts);
3444 to = &timeout;
3445 }
3446
3447 /*
3448 * Invert the set of allowed signals to get those we want to block.
3449 */
3450 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3451 signotset(&mask);
3452
3453 spin_lock_irq(&tsk->sighand->siglock);
3454 sig = dequeue_signal(tsk, &mask, info);
3455 if (!sig && timeout) {
3456 /*
3457 * None ready, temporarily unblock those we're interested
3458 * while we are sleeping in so that we'll be awakened when
3459 * they arrive. Unblocking is always fine, we can avoid
3460 * set_current_blocked().
3461 */
3462 tsk->real_blocked = tsk->blocked;
3463 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3464 recalc_sigpending();
3465 spin_unlock_irq(&tsk->sighand->siglock);
3466
3467 __set_current_state(TASK_INTERRUPTIBLE);
3468 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3469 HRTIMER_MODE_REL);
3470 spin_lock_irq(&tsk->sighand->siglock);
3471 __set_task_blocked(tsk, &tsk->real_blocked);
3472 sigemptyset(&tsk->real_blocked);
3473 sig = dequeue_signal(tsk, &mask, info);
3474 }
3475 spin_unlock_irq(&tsk->sighand->siglock);
3476
3477 if (sig)
3478 return sig;
3479 return ret ? -EINTR : -EAGAIN;
3480}
3481
3482/**
3483 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3484 * in @uthese
3485 * @uthese: queued signals to wait for
3486 * @uinfo: if non-null, the signal's siginfo is returned here
3487 * @uts: upper bound on process time suspension
3488 * @sigsetsize: size of sigset_t type
3489 */
3490SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3491 siginfo_t __user *, uinfo,
3492 const struct __kernel_timespec __user *, uts,
3493 size_t, sigsetsize)
3494{
3495 sigset_t these;
3496 struct timespec64 ts;
3497 kernel_siginfo_t info;
3498 int ret;
3499
3500 /* XXX: Don't preclude handling different sized sigset_t's. */
3501 if (sigsetsize != sizeof(sigset_t))
3502 return -EINVAL;
3503
3504 if (copy_from_user(&these, uthese, sizeof(these)))
3505 return -EFAULT;
3506
3507 if (uts) {
3508 if (get_timespec64(&ts, uts))
3509 return -EFAULT;
3510 }
3511
3512 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3513
3514 if (ret > 0 && uinfo) {
3515 if (copy_siginfo_to_user(uinfo, &info))
3516 ret = -EFAULT;
3517 }
3518
3519 return ret;
3520}
3521
3522#ifdef CONFIG_COMPAT_32BIT_TIME
3523SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3524 siginfo_t __user *, uinfo,
3525 const struct old_timespec32 __user *, uts,
3526 size_t, sigsetsize)
3527{
3528 sigset_t these;
3529 struct timespec64 ts;
3530 kernel_siginfo_t info;
3531 int ret;
3532
3533 if (sigsetsize != sizeof(sigset_t))
3534 return -EINVAL;
3535
3536 if (copy_from_user(&these, uthese, sizeof(these)))
3537 return -EFAULT;
3538
3539 if (uts) {
3540 if (get_old_timespec32(&ts, uts))
3541 return -EFAULT;
3542 }
3543
3544 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3545
3546 if (ret > 0 && uinfo) {
3547 if (copy_siginfo_to_user(uinfo, &info))
3548 ret = -EFAULT;
3549 }
3550
3551 return ret;
3552}
3553#endif
3554
3555#ifdef CONFIG_COMPAT
3556COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3557 struct compat_siginfo __user *, uinfo,
3558 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3559{
3560 sigset_t s;
3561 struct timespec64 t;
3562 kernel_siginfo_t info;
3563 long ret;
3564
3565 if (sigsetsize != sizeof(sigset_t))
3566 return -EINVAL;
3567
3568 if (get_compat_sigset(&s, uthese))
3569 return -EFAULT;
3570
3571 if (uts) {
3572 if (get_timespec64(&t, uts))
3573 return -EFAULT;
3574 }
3575
3576 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3577
3578 if (ret > 0 && uinfo) {
3579 if (copy_siginfo_to_user32(uinfo, &info))
3580 ret = -EFAULT;
3581 }
3582
3583 return ret;
3584}
3585
3586#ifdef CONFIG_COMPAT_32BIT_TIME
3587COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3588 struct compat_siginfo __user *, uinfo,
3589 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3590{
3591 sigset_t s;
3592 struct timespec64 t;
3593 kernel_siginfo_t info;
3594 long ret;
3595
3596 if (sigsetsize != sizeof(sigset_t))
3597 return -EINVAL;
3598
3599 if (get_compat_sigset(&s, uthese))
3600 return -EFAULT;
3601
3602 if (uts) {
3603 if (get_old_timespec32(&t, uts))
3604 return -EFAULT;
3605 }
3606
3607 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3608
3609 if (ret > 0 && uinfo) {
3610 if (copy_siginfo_to_user32(uinfo, &info))
3611 ret = -EFAULT;
3612 }
3613
3614 return ret;
3615}
3616#endif
3617#endif
3618
3619static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3620{
3621 clear_siginfo(info);
3622 info->si_signo = sig;
3623 info->si_errno = 0;
3624 info->si_code = SI_USER;
3625 info->si_pid = task_tgid_vnr(current);
3626 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3627}
3628
3629/**
3630 * sys_kill - send a signal to a process
3631 * @pid: the PID of the process
3632 * @sig: signal to be sent
3633 */
3634SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3635{
3636 struct kernel_siginfo info;
3637
3638 prepare_kill_siginfo(sig, &info);
3639
3640 return kill_something_info(sig, &info, pid);
3641}
3642
3643/*
3644 * Verify that the signaler and signalee either are in the same pid namespace
3645 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3646 * namespace.
3647 */
3648static bool access_pidfd_pidns(struct pid *pid)
3649{
3650 struct pid_namespace *active = task_active_pid_ns(current);
3651 struct pid_namespace *p = ns_of_pid(pid);
3652
3653 for (;;) {
3654 if (!p)
3655 return false;
3656 if (p == active)
3657 break;
3658 p = p->parent;
3659 }
3660
3661 return true;
3662}
3663
3664static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo, siginfo_t *info)
3665{
3666#ifdef CONFIG_COMPAT
3667 /*
3668 * Avoid hooking up compat syscalls and instead handle necessary
3669 * conversions here. Note, this is a stop-gap measure and should not be
3670 * considered a generic solution.
3671 */
3672 if (in_compat_syscall())
3673 return copy_siginfo_from_user32(
3674 kinfo, (struct compat_siginfo __user *)info);
3675#endif
3676 return copy_siginfo_from_user(kinfo, info);
3677}
3678
3679static struct pid *pidfd_to_pid(const struct file *file)
3680{
3681 struct pid *pid;
3682
3683 pid = pidfd_pid(file);
3684 if (!IS_ERR(pid))
3685 return pid;
3686
3687 return tgid_pidfd_to_pid(file);
3688}
3689
3690/**
3691 * sys_pidfd_send_signal - Signal a process through a pidfd
3692 * @pidfd: file descriptor of the process
3693 * @sig: signal to send
3694 * @info: signal info
3695 * @flags: future flags
3696 *
3697 * The syscall currently only signals via PIDTYPE_PID which covers
3698 * kill(<positive-pid>, <signal>. It does not signal threads or process
3699 * groups.
3700 * In order to extend the syscall to threads and process groups the @flags
3701 * argument should be used. In essence, the @flags argument will determine
3702 * what is signaled and not the file descriptor itself. Put in other words,
3703 * grouping is a property of the flags argument not a property of the file
3704 * descriptor.
3705 *
3706 * Return: 0 on success, negative errno on failure
3707 */
3708SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3709 siginfo_t __user *, info, unsigned int, flags)
3710{
3711 int ret;
3712 struct fd f;
3713 struct pid *pid;
3714 kernel_siginfo_t kinfo;
3715
3716 /* Enforce flags be set to 0 until we add an extension. */
3717 if (flags)
3718 return -EINVAL;
3719
3720 f = fdget(pidfd);
3721 if (!f.file)
3722 return -EBADF;
3723
3724 /* Is this a pidfd? */
3725 pid = pidfd_to_pid(f.file);
3726 if (IS_ERR(pid)) {
3727 ret = PTR_ERR(pid);
3728 goto err;
3729 }
3730
3731 ret = -EINVAL;
3732 if (!access_pidfd_pidns(pid))
3733 goto err;
3734
3735 if (info) {
3736 ret = copy_siginfo_from_user_any(&kinfo, info);
3737 if (unlikely(ret))
3738 goto err;
3739
3740 ret = -EINVAL;
3741 if (unlikely(sig != kinfo.si_signo))
3742 goto err;
3743
3744 /* Only allow sending arbitrary signals to yourself. */
3745 ret = -EPERM;
3746 if ((task_pid(current) != pid) &&
3747 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3748 goto err;
3749 } else {
3750 prepare_kill_siginfo(sig, &kinfo);
3751 }
3752
3753 ret = kill_pid_info(sig, &kinfo, pid);
3754
3755err:
3756 fdput(f);
3757 return ret;
3758}
3759
3760static int
3761do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3762{
3763 struct task_struct *p;
3764 int error = -ESRCH;
3765
3766 rcu_read_lock();
3767 p = find_task_by_vpid(pid);
3768 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3769 error = check_kill_permission(sig, info, p);
3770 /*
3771 * The null signal is a permissions and process existence
3772 * probe. No signal is actually delivered.
3773 */
3774 if (!error && sig) {
3775 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3776 /*
3777 * If lock_task_sighand() failed we pretend the task
3778 * dies after receiving the signal. The window is tiny,
3779 * and the signal is private anyway.
3780 */
3781 if (unlikely(error == -ESRCH))
3782 error = 0;
3783 }
3784 }
3785 rcu_read_unlock();
3786
3787 return error;
3788}
3789
3790static int do_tkill(pid_t tgid, pid_t pid, int sig)
3791{
3792 struct kernel_siginfo info;
3793
3794 clear_siginfo(&info);
3795 info.si_signo = sig;
3796 info.si_errno = 0;
3797 info.si_code = SI_TKILL;
3798 info.si_pid = task_tgid_vnr(current);
3799 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3800
3801 return do_send_specific(tgid, pid, sig, &info);
3802}
3803
3804/**
3805 * sys_tgkill - send signal to one specific thread
3806 * @tgid: the thread group ID of the thread
3807 * @pid: the PID of the thread
3808 * @sig: signal to be sent
3809 *
3810 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3811 * exists but it's not belonging to the target process anymore. This
3812 * method solves the problem of threads exiting and PIDs getting reused.
3813 */
3814SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3815{
3816 /* This is only valid for single tasks */
3817 if (pid <= 0 || tgid <= 0)
3818 return -EINVAL;
3819
3820 return do_tkill(tgid, pid, sig);
3821}
3822
3823/**
3824 * sys_tkill - send signal to one specific task
3825 * @pid: the PID of the task
3826 * @sig: signal to be sent
3827 *
3828 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3829 */
3830SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3831{
3832 /* This is only valid for single tasks */
3833 if (pid <= 0)
3834 return -EINVAL;
3835
3836 return do_tkill(0, pid, sig);
3837}
3838
3839static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3840{
3841 /* Not even root can pretend to send signals from the kernel.
3842 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3843 */
3844 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3845 (task_pid_vnr(current) != pid))
3846 return -EPERM;
3847
3848 /* POSIX.1b doesn't mention process groups. */
3849 return kill_proc_info(sig, info, pid);
3850}
3851
3852/**
3853 * sys_rt_sigqueueinfo - send signal information to a signal
3854 * @pid: the PID of the thread
3855 * @sig: signal to be sent
3856 * @uinfo: signal info to be sent
3857 */
3858SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3859 siginfo_t __user *, uinfo)
3860{
3861 kernel_siginfo_t info;
3862 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3863 if (unlikely(ret))
3864 return ret;
3865 return do_rt_sigqueueinfo(pid, sig, &info);
3866}
3867
3868#ifdef CONFIG_COMPAT
3869COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3870 compat_pid_t, pid,
3871 int, sig,
3872 struct compat_siginfo __user *, uinfo)
3873{
3874 kernel_siginfo_t info;
3875 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3876 if (unlikely(ret))
3877 return ret;
3878 return do_rt_sigqueueinfo(pid, sig, &info);
3879}
3880#endif
3881
3882static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3883{
3884 /* This is only valid for single tasks */
3885 if (pid <= 0 || tgid <= 0)
3886 return -EINVAL;
3887
3888 /* Not even root can pretend to send signals from the kernel.
3889 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3890 */
3891 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3892 (task_pid_vnr(current) != pid))
3893 return -EPERM;
3894
3895 return do_send_specific(tgid, pid, sig, info);
3896}
3897
3898SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3899 siginfo_t __user *, uinfo)
3900{
3901 kernel_siginfo_t info;
3902 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3903 if (unlikely(ret))
3904 return ret;
3905 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3906}
3907
3908#ifdef CONFIG_COMPAT
3909COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3910 compat_pid_t, tgid,
3911 compat_pid_t, pid,
3912 int, sig,
3913 struct compat_siginfo __user *, uinfo)
3914{
3915 kernel_siginfo_t info;
3916 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3917 if (unlikely(ret))
3918 return ret;
3919 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3920}
3921#endif
3922
3923/*
3924 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
3925 */
3926void kernel_sigaction(int sig, __sighandler_t action)
3927{
3928 spin_lock_irq(¤t->sighand->siglock);
3929 current->sighand->action[sig - 1].sa.sa_handler = action;
3930 if (action == SIG_IGN) {
3931 sigset_t mask;
3932
3933 sigemptyset(&mask);
3934 sigaddset(&mask, sig);
3935
3936 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
3937 flush_sigqueue_mask(&mask, ¤t->pending);
3938 recalc_sigpending();
3939 }
3940 spin_unlock_irq(¤t->sighand->siglock);
3941}
3942EXPORT_SYMBOL(kernel_sigaction);
3943
3944void __weak sigaction_compat_abi(struct k_sigaction *act,
3945 struct k_sigaction *oact)
3946{
3947}
3948
3949int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
3950{
3951 struct task_struct *p = current, *t;
3952 struct k_sigaction *k;
3953 sigset_t mask;
3954
3955 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
3956 return -EINVAL;
3957
3958 k = &p->sighand->action[sig-1];
3959
3960 spin_lock_irq(&p->sighand->siglock);
3961 if (oact)
3962 *oact = *k;
3963
3964 sigaction_compat_abi(act, oact);
3965
3966 if (act) {
3967 sigdelsetmask(&act->sa.sa_mask,
3968 sigmask(SIGKILL) | sigmask(SIGSTOP));
3969 *k = *act;
3970 /*
3971 * POSIX 3.3.1.3:
3972 * "Setting a signal action to SIG_IGN for a signal that is
3973 * pending shall cause the pending signal to be discarded,
3974 * whether or not it is blocked."
3975 *
3976 * "Setting a signal action to SIG_DFL for a signal that is
3977 * pending and whose default action is to ignore the signal
3978 * (for example, SIGCHLD), shall cause the pending signal to
3979 * be discarded, whether or not it is blocked"
3980 */
3981 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
3982 sigemptyset(&mask);
3983 sigaddset(&mask, sig);
3984 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
3985 for_each_thread(p, t)
3986 flush_sigqueue_mask(&mask, &t->pending);
3987 }
3988 }
3989
3990 spin_unlock_irq(&p->sighand->siglock);
3991 return 0;
3992}
3993
3994static int
3995do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
3996 size_t min_ss_size)
3997{
3998 struct task_struct *t = current;
3999
4000 if (oss) {
4001 memset(oss, 0, sizeof(stack_t));
4002 oss->ss_sp = (void __user *) t->sas_ss_sp;
4003 oss->ss_size = t->sas_ss_size;
4004 oss->ss_flags = sas_ss_flags(sp) |
4005 (current->sas_ss_flags & SS_FLAG_BITS);
4006 }
4007
4008 if (ss) {
4009 void __user *ss_sp = ss->ss_sp;
4010 size_t ss_size = ss->ss_size;
4011 unsigned ss_flags = ss->ss_flags;
4012 int ss_mode;
4013
4014 if (unlikely(on_sig_stack(sp)))
4015 return -EPERM;
4016
4017 ss_mode = ss_flags & ~SS_FLAG_BITS;
4018 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4019 ss_mode != 0))
4020 return -EINVAL;
4021
4022 if (ss_mode == SS_DISABLE) {
4023 ss_size = 0;
4024 ss_sp = NULL;
4025 } else {
4026 if (unlikely(ss_size < min_ss_size))
4027 return -ENOMEM;
4028 }
4029
4030 t->sas_ss_sp = (unsigned long) ss_sp;
4031 t->sas_ss_size = ss_size;
4032 t->sas_ss_flags = ss_flags;
4033 }
4034 return 0;
4035}
4036
4037SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4038{
4039 stack_t new, old;
4040 int err;
4041 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4042 return -EFAULT;
4043 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4044 current_user_stack_pointer(),
4045 MINSIGSTKSZ);
4046 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4047 err = -EFAULT;
4048 return err;
4049}
4050
4051int restore_altstack(const stack_t __user *uss)
4052{
4053 stack_t new;
4054 if (copy_from_user(&new, uss, sizeof(stack_t)))
4055 return -EFAULT;
4056 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4057 MINSIGSTKSZ);
4058 /* squash all but EFAULT for now */
4059 return 0;
4060}
4061
4062int __save_altstack(stack_t __user *uss, unsigned long sp)
4063{
4064 struct task_struct *t = current;
4065 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4066 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4067 __put_user(t->sas_ss_size, &uss->ss_size);
4068 if (err)
4069 return err;
4070 if (t->sas_ss_flags & SS_AUTODISARM)
4071 sas_ss_reset(t);
4072 return 0;
4073}
4074
4075#ifdef CONFIG_COMPAT
4076static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4077 compat_stack_t __user *uoss_ptr)
4078{
4079 stack_t uss, uoss;
4080 int ret;
4081
4082 if (uss_ptr) {
4083 compat_stack_t uss32;
4084 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4085 return -EFAULT;
4086 uss.ss_sp = compat_ptr(uss32.ss_sp);
4087 uss.ss_flags = uss32.ss_flags;
4088 uss.ss_size = uss32.ss_size;
4089 }
4090 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4091 compat_user_stack_pointer(),
4092 COMPAT_MINSIGSTKSZ);
4093 if (ret >= 0 && uoss_ptr) {
4094 compat_stack_t old;
4095 memset(&old, 0, sizeof(old));
4096 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4097 old.ss_flags = uoss.ss_flags;
4098 old.ss_size = uoss.ss_size;
4099 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4100 ret = -EFAULT;
4101 }
4102 return ret;
4103}
4104
4105COMPAT_SYSCALL_DEFINE2(sigaltstack,
4106 const compat_stack_t __user *, uss_ptr,
4107 compat_stack_t __user *, uoss_ptr)
4108{
4109 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4110}
4111
4112int compat_restore_altstack(const compat_stack_t __user *uss)
4113{
4114 int err = do_compat_sigaltstack(uss, NULL);
4115 /* squash all but -EFAULT for now */
4116 return err == -EFAULT ? err : 0;
4117}
4118
4119int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4120{
4121 int err;
4122 struct task_struct *t = current;
4123 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4124 &uss->ss_sp) |
4125 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4126 __put_user(t->sas_ss_size, &uss->ss_size);
4127 if (err)
4128 return err;
4129 if (t->sas_ss_flags & SS_AUTODISARM)
4130 sas_ss_reset(t);
4131 return 0;
4132}
4133#endif
4134
4135#ifdef __ARCH_WANT_SYS_SIGPENDING
4136
4137/**
4138 * sys_sigpending - examine pending signals
4139 * @uset: where mask of pending signal is returned
4140 */
4141SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4142{
4143 sigset_t set;
4144
4145 if (sizeof(old_sigset_t) > sizeof(*uset))
4146 return -EINVAL;
4147
4148 do_sigpending(&set);
4149
4150 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4151 return -EFAULT;
4152
4153 return 0;
4154}
4155
4156#ifdef CONFIG_COMPAT
4157COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4158{
4159 sigset_t set;
4160
4161 do_sigpending(&set);
4162
4163 return put_user(set.sig[0], set32);
4164}
4165#endif
4166
4167#endif
4168
4169#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4170/**
4171 * sys_sigprocmask - examine and change blocked signals
4172 * @how: whether to add, remove, or set signals
4173 * @nset: signals to add or remove (if non-null)
4174 * @oset: previous value of signal mask if non-null
4175 *
4176 * Some platforms have their own version with special arguments;
4177 * others support only sys_rt_sigprocmask.
4178 */
4179
4180SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4181 old_sigset_t __user *, oset)
4182{
4183 old_sigset_t old_set, new_set;
4184 sigset_t new_blocked;
4185
4186 old_set = current->blocked.sig[0];
4187
4188 if (nset) {
4189 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4190 return -EFAULT;
4191
4192 new_blocked = current->blocked;
4193
4194 switch (how) {
4195 case SIG_BLOCK:
4196 sigaddsetmask(&new_blocked, new_set);
4197 break;
4198 case SIG_UNBLOCK:
4199 sigdelsetmask(&new_blocked, new_set);
4200 break;
4201 case SIG_SETMASK:
4202 new_blocked.sig[0] = new_set;
4203 break;
4204 default:
4205 return -EINVAL;
4206 }
4207
4208 set_current_blocked(&new_blocked);
4209 }
4210
4211 if (oset) {
4212 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4213 return -EFAULT;
4214 }
4215
4216 return 0;
4217}
4218#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4219
4220#ifndef CONFIG_ODD_RT_SIGACTION
4221/**
4222 * sys_rt_sigaction - alter an action taken by a process
4223 * @sig: signal to be sent
4224 * @act: new sigaction
4225 * @oact: used to save the previous sigaction
4226 * @sigsetsize: size of sigset_t type
4227 */
4228SYSCALL_DEFINE4(rt_sigaction, int, sig,
4229 const struct sigaction __user *, act,
4230 struct sigaction __user *, oact,
4231 size_t, sigsetsize)
4232{
4233 struct k_sigaction new_sa, old_sa;
4234 int ret;
4235
4236 /* XXX: Don't preclude handling different sized sigset_t's. */
4237 if (sigsetsize != sizeof(sigset_t))
4238 return -EINVAL;
4239
4240 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4241 return -EFAULT;
4242
4243 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4244 if (ret)
4245 return ret;
4246
4247 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4248 return -EFAULT;
4249
4250 return 0;
4251}
4252#ifdef CONFIG_COMPAT
4253COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4254 const struct compat_sigaction __user *, act,
4255 struct compat_sigaction __user *, oact,
4256 compat_size_t, sigsetsize)
4257{
4258 struct k_sigaction new_ka, old_ka;
4259#ifdef __ARCH_HAS_SA_RESTORER
4260 compat_uptr_t restorer;
4261#endif
4262 int ret;
4263
4264 /* XXX: Don't preclude handling different sized sigset_t's. */
4265 if (sigsetsize != sizeof(compat_sigset_t))
4266 return -EINVAL;
4267
4268 if (act) {
4269 compat_uptr_t handler;
4270 ret = get_user(handler, &act->sa_handler);
4271 new_ka.sa.sa_handler = compat_ptr(handler);
4272#ifdef __ARCH_HAS_SA_RESTORER
4273 ret |= get_user(restorer, &act->sa_restorer);
4274 new_ka.sa.sa_restorer = compat_ptr(restorer);
4275#endif
4276 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4277 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4278 if (ret)
4279 return -EFAULT;
4280 }
4281
4282 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4283 if (!ret && oact) {
4284 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4285 &oact->sa_handler);
4286 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4287 sizeof(oact->sa_mask));
4288 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4289#ifdef __ARCH_HAS_SA_RESTORER
4290 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4291 &oact->sa_restorer);
4292#endif
4293 }
4294 return ret;
4295}
4296#endif
4297#endif /* !CONFIG_ODD_RT_SIGACTION */
4298
4299#ifdef CONFIG_OLD_SIGACTION
4300SYSCALL_DEFINE3(sigaction, int, sig,
4301 const struct old_sigaction __user *, act,
4302 struct old_sigaction __user *, oact)
4303{
4304 struct k_sigaction new_ka, old_ka;
4305 int ret;
4306
4307 if (act) {
4308 old_sigset_t mask;
4309 if (!access_ok(act, sizeof(*act)) ||
4310 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4311 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4312 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4313 __get_user(mask, &act->sa_mask))
4314 return -EFAULT;
4315#ifdef __ARCH_HAS_KA_RESTORER
4316 new_ka.ka_restorer = NULL;
4317#endif
4318 siginitset(&new_ka.sa.sa_mask, mask);
4319 }
4320
4321 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4322
4323 if (!ret && oact) {
4324 if (!access_ok(oact, sizeof(*oact)) ||
4325 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4326 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4327 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4328 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4329 return -EFAULT;
4330 }
4331
4332 return ret;
4333}
4334#endif
4335#ifdef CONFIG_COMPAT_OLD_SIGACTION
4336COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4337 const struct compat_old_sigaction __user *, act,
4338 struct compat_old_sigaction __user *, oact)
4339{
4340 struct k_sigaction new_ka, old_ka;
4341 int ret;
4342 compat_old_sigset_t mask;
4343 compat_uptr_t handler, restorer;
4344
4345 if (act) {
4346 if (!access_ok(act, sizeof(*act)) ||
4347 __get_user(handler, &act->sa_handler) ||
4348 __get_user(restorer, &act->sa_restorer) ||
4349 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4350 __get_user(mask, &act->sa_mask))
4351 return -EFAULT;
4352
4353#ifdef __ARCH_HAS_KA_RESTORER
4354 new_ka.ka_restorer = NULL;
4355#endif
4356 new_ka.sa.sa_handler = compat_ptr(handler);
4357 new_ka.sa.sa_restorer = compat_ptr(restorer);
4358 siginitset(&new_ka.sa.sa_mask, mask);
4359 }
4360
4361 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4362
4363 if (!ret && oact) {
4364 if (!access_ok(oact, sizeof(*oact)) ||
4365 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4366 &oact->sa_handler) ||
4367 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4368 &oact->sa_restorer) ||
4369 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4370 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4371 return -EFAULT;
4372 }
4373 return ret;
4374}
4375#endif
4376
4377#ifdef CONFIG_SGETMASK_SYSCALL
4378
4379/*
4380 * For backwards compatibility. Functionality superseded by sigprocmask.
4381 */
4382SYSCALL_DEFINE0(sgetmask)
4383{
4384 /* SMP safe */
4385 return current->blocked.sig[0];
4386}
4387
4388SYSCALL_DEFINE1(ssetmask, int, newmask)
4389{
4390 int old = current->blocked.sig[0];
4391 sigset_t newset;
4392
4393 siginitset(&newset, newmask);
4394 set_current_blocked(&newset);
4395
4396 return old;
4397}
4398#endif /* CONFIG_SGETMASK_SYSCALL */
4399
4400#ifdef __ARCH_WANT_SYS_SIGNAL
4401/*
4402 * For backwards compatibility. Functionality superseded by sigaction.
4403 */
4404SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4405{
4406 struct k_sigaction new_sa, old_sa;
4407 int ret;
4408
4409 new_sa.sa.sa_handler = handler;
4410 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4411 sigemptyset(&new_sa.sa.sa_mask);
4412
4413 ret = do_sigaction(sig, &new_sa, &old_sa);
4414
4415 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4416}
4417#endif /* __ARCH_WANT_SYS_SIGNAL */
4418
4419#ifdef __ARCH_WANT_SYS_PAUSE
4420
4421SYSCALL_DEFINE0(pause)
4422{
4423 while (!signal_pending(current)) {
4424 __set_current_state(TASK_INTERRUPTIBLE);
4425 schedule();
4426 }
4427 return -ERESTARTNOHAND;
4428}
4429
4430#endif
4431
4432static int sigsuspend(sigset_t *set)
4433{
4434 current->saved_sigmask = current->blocked;
4435 set_current_blocked(set);
4436
4437 while (!signal_pending(current)) {
4438 __set_current_state(TASK_INTERRUPTIBLE);
4439 schedule();
4440 }
4441 set_restore_sigmask();
4442 return -ERESTARTNOHAND;
4443}
4444
4445/**
4446 * sys_rt_sigsuspend - replace the signal mask for a value with the
4447 * @unewset value until a signal is received
4448 * @unewset: new signal mask value
4449 * @sigsetsize: size of sigset_t type
4450 */
4451SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4452{
4453 sigset_t newset;
4454
4455 /* XXX: Don't preclude handling different sized sigset_t's. */
4456 if (sigsetsize != sizeof(sigset_t))
4457 return -EINVAL;
4458
4459 if (copy_from_user(&newset, unewset, sizeof(newset)))
4460 return -EFAULT;
4461 return sigsuspend(&newset);
4462}
4463
4464#ifdef CONFIG_COMPAT
4465COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4466{
4467 sigset_t newset;
4468
4469 /* XXX: Don't preclude handling different sized sigset_t's. */
4470 if (sigsetsize != sizeof(sigset_t))
4471 return -EINVAL;
4472
4473 if (get_compat_sigset(&newset, unewset))
4474 return -EFAULT;
4475 return sigsuspend(&newset);
4476}
4477#endif
4478
4479#ifdef CONFIG_OLD_SIGSUSPEND
4480SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4481{
4482 sigset_t blocked;
4483 siginitset(&blocked, mask);
4484 return sigsuspend(&blocked);
4485}
4486#endif
4487#ifdef CONFIG_OLD_SIGSUSPEND3
4488SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4489{
4490 sigset_t blocked;
4491 siginitset(&blocked, mask);
4492 return sigsuspend(&blocked);
4493}
4494#endif
4495
4496__weak const char *arch_vma_name(struct vm_area_struct *vma)
4497{
4498 return NULL;
4499}
4500
4501static inline void siginfo_buildtime_checks(void)
4502{
4503 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4504
4505 /* Verify the offsets in the two siginfos match */
4506#define CHECK_OFFSET(field) \
4507 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4508
4509 /* kill */
4510 CHECK_OFFSET(si_pid);
4511 CHECK_OFFSET(si_uid);
4512
4513 /* timer */
4514 CHECK_OFFSET(si_tid);
4515 CHECK_OFFSET(si_overrun);
4516 CHECK_OFFSET(si_value);
4517
4518 /* rt */
4519 CHECK_OFFSET(si_pid);
4520 CHECK_OFFSET(si_uid);
4521 CHECK_OFFSET(si_value);
4522
4523 /* sigchld */
4524 CHECK_OFFSET(si_pid);
4525 CHECK_OFFSET(si_uid);
4526 CHECK_OFFSET(si_status);
4527 CHECK_OFFSET(si_utime);
4528 CHECK_OFFSET(si_stime);
4529
4530 /* sigfault */
4531 CHECK_OFFSET(si_addr);
4532 CHECK_OFFSET(si_addr_lsb);
4533 CHECK_OFFSET(si_lower);
4534 CHECK_OFFSET(si_upper);
4535 CHECK_OFFSET(si_pkey);
4536
4537 /* sigpoll */
4538 CHECK_OFFSET(si_band);
4539 CHECK_OFFSET(si_fd);
4540
4541 /* sigsys */
4542 CHECK_OFFSET(si_call_addr);
4543 CHECK_OFFSET(si_syscall);
4544 CHECK_OFFSET(si_arch);
4545#undef CHECK_OFFSET
4546
4547 /* usb asyncio */
4548 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4549 offsetof(struct siginfo, si_addr));
4550 if (sizeof(int) == sizeof(void __user *)) {
4551 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4552 sizeof(void __user *));
4553 } else {
4554 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4555 sizeof_field(struct siginfo, si_uid)) !=
4556 sizeof(void __user *));
4557 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4558 offsetof(struct siginfo, si_uid));
4559 }
4560#ifdef CONFIG_COMPAT
4561 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4562 offsetof(struct compat_siginfo, si_addr));
4563 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4564 sizeof(compat_uptr_t));
4565 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4566 sizeof_field(struct siginfo, si_pid));
4567#endif
4568}
4569
4570void __init signals_init(void)
4571{
4572 siginfo_buildtime_checks();
4573
4574 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4575}
4576
4577#ifdef CONFIG_KGDB_KDB
4578#include <linux/kdb.h>
4579/*
4580 * kdb_send_sig - Allows kdb to send signals without exposing
4581 * signal internals. This function checks if the required locks are
4582 * available before calling the main signal code, to avoid kdb
4583 * deadlocks.
4584 */
4585void kdb_send_sig(struct task_struct *t, int sig)
4586{
4587 static struct task_struct *kdb_prev_t;
4588 int new_t, ret;
4589 if (!spin_trylock(&t->sighand->siglock)) {
4590 kdb_printf("Can't do kill command now.\n"
4591 "The sigmask lock is held somewhere else in "
4592 "kernel, try again later\n");
4593 return;
4594 }
4595 new_t = kdb_prev_t != t;
4596 kdb_prev_t = t;
4597 if (t->state != TASK_RUNNING && new_t) {
4598 spin_unlock(&t->sighand->siglock);
4599 kdb_printf("Process is not RUNNING, sending a signal from "
4600 "kdb risks deadlock\n"
4601 "on the run queue locks. "
4602 "The signal has _not_ been sent.\n"
4603 "Reissue the kill command if you want to risk "
4604 "the deadlock.\n");
4605 return;
4606 }
4607 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4608 spin_unlock(&t->sighand->siglock);
4609 if (ret)
4610 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4611 sig, t->pid);
4612 else
4613 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4614}
4615#endif /* CONFIG_KGDB_KDB */